BULLETIN

BULLETIN OFOF THE POLISH ACADEMY

THE POLISH ACADEMY OF OF SCIENCES
SCIENCES SPECIAL SECTION
TECHNICAL SCIENCES, Vol. 68, No. 4,
TECHNICAL SCIENCES, Vol. 68, No. 4, 2020 2020
DOI:
DOI: 10.24425/bpasts.2020.134178
10.24425/bpasts.2020.XXXXXX

Effective approach to distributed optimal operation control

Effective approach to distributed optimal operation control in rural
in rural low voltage
low voltage microgrids
microgrids
M. PAROL ∗, P. *KAPLER,
M. PAROL , P. KAPLER,
J. J. MARZECKI, R.
MARZECKI, R. PAROL,
PAROL,M.M.
POŁECKI, and Ł.
POŁECKI, andROKICKI
Ł. ROKICKI
Warsaw University of Technology, Faculty of Electrical Engineering, Pl. Politechniki 1, 00-661 Warsaw, Poland
Warsaw University of Technology, Faculty of Electrical Engineering, Pl. Politechniki 1, 00-661 Warsaw, Poland

Abstract. The paper raises the issue of optimizing the control of the rural low voltage microgrids. Microgrids can operate in a synchronous
Abstract.
mode with The
gridspaper raises the system
of distribution issue ofoperators
optimizingandthe control
in an islandofmode.
the rural
We low
can voltage microgrids.
distinguish Microgrids
two control strategiescan operate in aone
in microgrids: synchronous
approach
mode with grids of distribution system operators and in an island mode. We can distinguish two control strategies in microgrids:
based on centralized control logic, which is usually used, and another on decentralized control logic. In this paper we decided to present one approach
the
based on centralized control logic, which is usually used, and another on decentralized control logic. In this paper we decided
approach based on the distributed control, combining the efforts of the distributed cooperative control and modified Monte Carlo optimization to present the
approach based on the distributed control, combining the efforts of the distributed cooperative control and modified Monte Carlo
method. Special attention has been paid to the impact of the order of processing particular devices’ groups on results of optimization calculations. optimization
method. Special
Moreover, attention
different has been
scenarios paid to of
of behavior thethe
impact of the control
microgrid order ofsystem
processing
with particular
respect to devices’ groups on results
the communication of optimization
loss have calculations.
been also presented. The
Moreover, different scenarios of behavior of the microgrid control system with respect to the communication loss have been
influence of the issue of continuity of communication between particular devices’ groups on the possibility of carrying out the optimization also presented. The
influence of the issue of continuity of communication between particular devices’ groups on the possibility of carrying out
process has been investigated. Additionally, characteristics of power loads and generation of electricity from small renewable energy sources the optimization
process hasinbeen
appearing ruralinvestigated.
areas have beenAdditionally,
described characteristics of power
and the sensitivity of theloads and generation
optimization of electricity
algorithm from small
to the changes renewable
of demanded energy
power sources
values and
changes of values of power generated by renewable energy sources has been studied. We analyzed different objective functions which canand
appearing in rural areas have been described and the sensitivity of the optimization algorithm to the changes of demanded power values be
changes of values of power generated by renewable energy sources has been studied. We analyzed different objective functions which can be
used as an optimization goal both in synchronous and island operation modes of microgrid. We decided to intensively test our approach on
used as an optimization goal both in synchronous and island operation modes of microgrid. We decided to intensively test our approach on a
a sample rural LV microgrid, which is typical in the countryside. The observed results of the tests have been presented and analyzed in detail.
sample rural LV microgrid, which is typical in the countryside. The observed results of the tests have been presented and analyzed in detail.
Generally, results achieved with the use of proposed distributed control are the same as with the use of centralized control. We think that the
Generally, results achieved with the use of proposed distributed control are the same as with the use of centralized control. We think that the
approach based on distributed control is promising for practical applications, because of its advantages.
approach based on distributed control is promising for practical applications, because of its advantages.
Key words: microgrids, optimal operation, distributed control, rural areas.
Key words: microgrids, optimal operation, distributed control, rural areas.

1. Introduction Essential components of low voltage MG are: microsources

The topic of microgrids (MG), including low voltage (LV) ones,
is an issue, which has intensively been studied for more than a
dozen years. Microgrids are strictly connected to the develop-
ment of distributed generation (DG), and are a component of
the Smart Grid (SG) concept.
The topic of DG has already been widely described in the
literature, similarly as the topic of SG. For example, paper [1]
presents various technologies of DG for residential houses and
for municipalities, as well as “energy plus” technologies for
other buildings. In turn, paper [2] describes the issue of inte-
gration of distributed energy sources (DES) with the electric
power grid via power electronic converters (PEC). Publication
[3] is devoted to the detailed overview of the most important
issues concerning the use of PEC in the context of Smart Grids.
The issue of MG has been also discussed thoroughly in the
literature. In technical brochure [4], as well as in [5, 6] we can
find a formal definition of microgrids. In these sources detailed
characteristics of different kinds of MG have been presented.
The concept of MG has been also presented in many other pub-
lications, for example in [7, 8].
∗ e-mail:
miroslaw.parol@ien.pw.edu.pl
*e-mail: miroslaw.parol@ien.pw.edu.pl
Manuscript submitted
Manuscript submitted2019-11-30, revisedinitially
20XX-XX-XX, 2020-02-04, initiallyfor
accepted
for publication 2020-03-01,
20XX-XX-XX, published published in August
in ZZZZZZZZ 2020
2020.
(MS), electricity storage units (ES), electricity loads (control-
lable loads – CL and non-controllable loads – NCL), and LV
network infrastructure.
Microgrids can operate in grid-connected (synchronous)
mode with networks of distribution system operators (DSO), as
well as in island mode. One of the most important challenges is
proper setting of the operating points of MSs, ESs and CLs lo-
cated within a microgrid. Different aspects of the issue of MG
operation control has been already widely described in liter-
ature sources, e.g. in [7–11]. In turn, paper [12] presents an
overview of the strategies of control, grid integration and en-
ergy management in MG. In referenced publications, different
approaches for operation control have been presented. We can
distinguish two control strategies: one approach based on cen-
tralized control logic and another on decentralized (distributed)
control logic.
Usually, centralized control logic is used for operation con-
trol in microgrids. Such approach was presented for example in
[13, 14]. In turn, publications [15–17] present distributed con-
trol logic for operation control in MG. It is worth noting, that
paper [18] describes the issue of implementation of optimal op-
eration control algorithms in low voltage microgrids.
Considering the topic of this paper, we will devote the further
part of the literature review to distribution control logic. We
accepted
publication can distinguish two distributed control strategies: competitive
control and cooperative control.

Bull. Pol. Ac.:

Bull. Pol. Ac.:Tech.
Tech.68(4)
68(4)2020
2020 661
1
 M.
M.Parol,
Parol, P.
P. Kapler,
Kapler, J.
J. Marzecki,
Marzecki, R.
R. Parol, M. Połecki,
Cooperative control [19–21] concerns systems performing
engineering tasks, which can be described with the use of the
set of elements making their own decisions, having some lim-
ited possibilities with regard to data processing and making use
of local information that is available. Together with communi-
cation connections between these elements, it allows for coop-
eration and making some efforts to reach common (collective)
goals. Microgrids can also be treated as members of this class
of systems.
Distribution of information imposes the application of distri-
bution control and the performing of some necessary calcula-
tions by cooperating components. The technique that is worth
to be mentioned in this particular case is the “consensus al-
gorithm” formulated by Tsitsiklis [21]. Distributed control and
necessity of performing calculations can be perceived as an ob-
stacle when someone tries to implement predictive control in
multi-agent system [21]. Multi-agent systems (MAS) [22–24]
are composed of many interacting computer programs, called
agents, whose aim it is to achieve some individual or collective
goals. Agent is a computer application working autonomously
within its environment.
Examples of concrete solutions in the range of distributed co-
operative control in electric power grids have been presented,
among others, in [25–30]. Some of these solutions concern dis-
tributed cooperative control with the use of multi-agent sys-
tems.
One of the essential issues concerning the microgrid is its
autonomous operation. In this mode, active power balance be-
tween generated power and received power must be ensured.
This problem was a subject of article [25]. This publication
proposes the distributed control of microsources based on sub-
gradient.
Algorithm of distributed control, whose goal is the regula-
tion of output power level of big number photovoltaic systems
(generators) located in distribution grid, has been described in
work [26]. For that purpose, cooperative control method is used.
In the proposed approach of distributed cooperative control,
agents representing PV sources constitute groups, with commu-
nication within these groups. Also the communication between
the groups of agents and higher level controller takes place.
In the publication [27] the issue of the secondary control in
microgrid is considered. To implement the reliable secondary
control for the microgrid distributed cooperative control, ap-
proach with the use of multi-agent systems can be applied. The
secondary control encompassed both voltage and frequency
regulation in the microgrid. Similar concept has been presented
in [28], in which the method of the distributed cooperative con-
trol with the use of multi-agent systems was used for the sec-
ondary voltage control in microgrid.
Article [29] presents the method of decentralized control for
distributed generation sources connected to the electric power
network by the means of power electronic converters, control-
ling the level and frequency of output voltages and the level of
generated power in synchronous mode of operation, in islanded
mode as well as during synchronization.
The strategy of cooperative control of microsources and en-
ergy storage during the islanded mode of the microgrid opera-
662
2 Bull. Pol.
Parol, M. Połecki, and
and Ł.
Ł. Rokicki
Rokicki
tion has been presented, in turn, in [30]. The evaluation of the
strategy has been made there by performing appropriate exper-
imental research studies.
Some of the algorithms presented in mentioned publications
are being formulated only theoretically and evaluated in a sim-
plified manner. It seems there is still a need to study and eval-
uate control methods for MG operation, especially the ones
based on the distribution control logic.
Publication [31] provides such methods and their evaluations
for sample rural microgrids, whose topologies and assump-
tions of operation truly reflect the conditions which can be seen
in practice. This paper was an extension of the previous con-
cept research, carried out in the RIGRID (rural intelligent grid)
project, which was described, among others, in [14, 18]. In arti-
cle [31] the issue of the centralized as well as distributed control
logic of MG operation has been considered. But the main atten-
tion has been given to the method based on the centralized MG
operation control. Presented results of optimization calculations
based on decentralized control logic should be treated as initial
and approximate.
This paper is an extension of paper [31]. In this paper we de-
cided to present the approach based on the distributed control
logic for sample rural microgrid, utilizing the distributed coop-
erative control and modified Monte Carlo optimization method.
We think that this approach is promising for practical applica-
tions, because of its advantages.
The optimization problem has been formulated in the paper
with the assumption in mind, that all generating units together
with the energy storage unit belong directly to (are under full
control of) a single microgrid’s operator and realize its aims. In
this way it is possible to formulate the considered optimization
problems as ones falling into the single-criterion optimization
category. If it was assumed, that particular generating units and
controllable loads have to realize aims of their owners (pro-
sumers), then the considered optimization problem related to
the maximization of profits should be formulated as a multi-
criteria optimization one. A distributed control paradigm can
be well-fitted for this aim as well, but in the current version of
the optimization program our goal was different. The motiva-
tion of applying a distributed approach was rather to look for a
way of reducing the complexity of the problem by introducing
an algorithm, which is simple, easy to understand and imple-
ment and efficient at the same time, not to consider the problem
of many separate entities with different goals.
We are able to apply our algorithm in practice, which is par-
tially the result of our participation in one of the European re-
search project dedicated to rural electric power grids. We de-
cided to utilize a topology of a sample low voltage rural mi-
crogrid, which is typically seen in countryside conditions. For
this sample microgrid, we intensively tested our approach. This
paper presents selected observed results of the tests. The data
returned by the proposed algorithm for different microgrid op-
eration conditions was analyzed.
In this paper special attention has been paid to the impact
of the order of processing particular devices’ groups on results
of optimization calculations. Moreover, the influence of the is-
sue of continuity of communication between particular devices’
Bull. Pol.Ac.:
Ac.:Tech.
Tech.68(4)
68(4) 2020
2020
 Effective
Effective approach
approach to to distributedoptimal
distributed optimal operation
operation control
controlininrural lowlow
rural voltage microgrids
voltage microgrids

groups on the possibility of carrying out the optimization pro-
cess has been investigated. Additionally, the sensitivity of the
optimization algorithm on the changes of demand power val-
ues and changes of values of power generated by renewable en-
ergy sources in particular nodes of MG has been studied. These
changes can be caused by e.g. inaccuracy of ultra-short-term
forecasting process of the mentioned magnitudes.
Similarly as in [31], we analyzed different objective func-
tions which can be used as an optimization goal both in syn-
chronous and island operation mode of MG.
The structure of the paper is presented below.
First, introduction to the analyzed problem has been pre-
sented and review of the selected publications concerning the
subject of the paper has been done. Then, characteristics of
power loads and generation of electricity from small renew-
able energy sources appearing in rural areas have been pre-
sented. Next, the description of the proposed microgrid con-
trol algorithm, including: problem formulation (objective func-
tions, constraints) and algorithm of distributed control logic has
been done. In the further part of the paper different scenarios of
behavior of the microgrid control system with respect to the
communication loss have been discussed. Then, case study in-
cluding: description of the test microgrid, results of optimiza-
tion calculations obtained with the use of algorithm of the dis-
tributed control logic, as well as the most important observa-
tions after having carried out experiments, has been presented.
At the end of the paper summary and final conclusions have
been made available.
only 15% of this energy consumption concerns agricultural pro-
duction. Generally, the most electricity is consumed by farms
which cool or dry harvested crops. The electricity consumption
also depends on the weather and the type of fruit grown though.
The consumption of electricity for living purposes is not sub-
ject to such significant changes. Electrical devices for agricul-
tural purposes may work as single ones or in groups. Examples
of single devices are a feed mixer or a grinding mill. Group de-
vices are located in special buildings, such as a piggery, barn or
cowshed. It is estimated that about 60% of electricity is used to
drive machines for agricultural work [33].
Farm load profiles change to the rhythm of the activities per-
formed. Some loads may occur at certain times, while others
only when needed. Moreover, some devices like refrigerators
or freezers are on continuously. Sample rural households load
profiles are presented in Fig. 1. In presented profiles some char-
acteristic features can be observed. The biggest demand for
electricity occurs during the winter, while the lowest during the
summer. Also the afternoon-evening peak in the summer usu-
ally occurs later than in other seasons.

2. Characteristics of power loads and

generation of electricity from small
renewable energy sources in rural areas

2.1. Introduction. The microgrids operated in rural areas are

able to satisfy locally the energy demands of the consumers. In Fig. 1. Load profiles of rural household; elaborated on the basis of [34]
intended islanded operation the microgrids also allow for im-
provement of supply reliability of electricity for consumers who Agricultural room load profiles change to the rhythm of rou-
are connected to them [14]. tine operations. Fig. 2 presents cowshed load profile. The in-
Small wind turbine-generator sets, photovoltaic installations, crease in demand for power in the morning and afternoon is
power plants based on biomass and biogas, small hydro power associated with the process of milking cattle.
plants, as well as small reciprocating engines with internal com-
bustion (engine-generator sets) are usually used in microgrids
in rural areas [1, 6, 8, 14, 18]. Amongst electrical energy stor-
age units, battery storages are the most popular units in rural
areas.

2.2. Characteristics of loads in rural areas. Farms are es-

sential entities in rural areas. A farm is an arrangement of land
and buildings with devices, whose task is to produce agricul-
tural products. The demand for electricity is strongly depen-
dent on the size of the farm and the type of agricultural produc-
tion. Consumed electricity is used both for domestic purposes
(like lighting or heating) and for conducting agricultural activ-
ities (like breeding or fruit-growing). According to the Polish
Central Statistical Office data [32], rural households in Poland
consumed 12101 GWh of electricity in 2017. Approximately Fig. 2. Cowshed load profile; elaborated on the basis of [35]

Bull. Pol. Ac.:

Bull. Pol. Ac.:Tech.
Tech.68(4)
68(4)2020
2020 6633
 M.
M.Parol,
Parol, P.
P. Kapler,
Kapler, J.
J. Marzecki,
Marzecki, R.
R. Parol, M. Połecki,
Parol, M. Połecki, and
and Ł.
Ł. Rokicki
Rokicki

There may also be small shops or service points in the coun- demand at that time. The presented generation profile is typical
tryside. Their load profiles are usually similar to rural house- for Polish conditions.
holds profiles as shown in Fig. 3. The farm load profiles with a PV installation are expected to
flatten during the hours of the largest generation. If some part
of the energy available during peak hours is not used, it can
.
be transferred to the distribution network. Photovoltaic installa-
tion can usually power processes which do not require large
amounts of electricity, such as irrigation, poultry lighting or
supplying small appliances for commercial services.
. Unlike a photovoltaic installation, a small wind turbine-
generator set can generate electricity continuously. In agricul-
ture, small wind turbine-generator sets with rated power from
5 kW to 20 kW are usually used. They can be connected to the
.
distribution network or work in a separate network as well.
In contrast to photovoltaic installation, small wind turbine-
generator sets have lower investment outlays. Also it will not
occupy space for planting plants or grazing animals.
Small wind turbine generation profiles in June and January
Fig. 3. Load profile of small service point in rural area; elaborated on are presented in Fig. 5. The presented generation profile is typ-
the basis of [36] ical for Polish conditions.

Knowledge of farm load profiles can be helpful in changing .

the seller of electricity or in the selection of renewable energy

.
sources. These types of energy sources in a farm can be used to
cover the energy needs of living in households, as well as for
the needs of agricultural production. Photovoltaic (in short –
.
PV) installations and small wind turbines are the most popular
renewable energy sources. .

.
2.3. Characteristics of photovoltaic installations and small
wind turbine-generator sets in rural areas. Each photo- .

voltaic installation consists of a PV module, an energy receiver

and auxiliary devices. An example place for PV installation can
be the roof surface of the piggery or barn. As a result, arable
land is not used. Solar installation generation profiles in June Fig. 5. Generation profiles of 5 kW small wind installation during June
and January are presented in Fig. 4. Generation peaks can be and January; elaborated on the basis of [37]
observed at 12:00–13:00 in winter and 11:00–14:00 in summer.
Electricity is generated mainly during the day, which predis- Analysis of the profile presented in Fig. 5 shows that in win-
poses the photovoltaic installation to cover part of the farm’s ter the turbine-generator set is able to generate noticeably more
electricity than in summer. An important aspect is the place of
turbine installation. The proximity of buildings or forests may
. cause slowdown in wind speed, whereas the occurrence of hills
– acceleration. Disadvantages of using small wind turbines in
the countryside are significant fluctuations in generated power
. and small wind resources at low altitudes.
Analysing the presented load profiles, it can be concluded
that any type of building in a rural area can successfully use
. renewable energy sources to cover part of its demand. One of
the key requirements is the knowledge of a typical load profile.
This knowledge is also necessary for the proper selection of an
. energy storage device due to the lack of continuity of generation
by PV installation. Farm load profiles should be adjusted so that
as many agricultural processes as possible are carried out dur-
ing the hours of the highest generation from renewable sources.
Fig. 4. Generation profiles of 5 kW PV installation during June and The use of PV installation can significantly affect the load pro-
January; elaborated on the basis of [37] file of a given building. The sudden decline in the PV generation

664
4 Bull. Pol.
Bull. Pol.Ac.:
Ac.:Tech.
Tech.68(4)
68(4) 2020
2020
 Effective
Effective approach
approach to to distributedoptimal
distributed optimal operation
operation control
(Fig. 4) coincides with the occurrence of the afternoon-evening
peak (Fig. 1, 2, and 3). As a result, there is a possibility of
changing the shape of that peak to a larger and steeper one than
when there is not a PV installation. Partial coverage of demand
by renewable energy sources can reduce the cost of purchasing
electricity and ensures supply during rural distribution network
failures.
It is worth considering the use of hybrid systems to simul-
taneously obtain electricity from solar radiation and wind en-
ergy. This will be especially important for farms with increased
electricity consumption in the morning and afternoon when the
generation from photovoltaic system in not large.
3. Optimization problem formulation
The distributed algorithm for determination of the set of
(sub)optimal operating points for all the devices composing a
microgrid, which has been presented in the further part of the
paper, is based on several assumptions. In the current section
we are going to present all the most important ones.
Firstly, we assume that the algorithm takes advantage of the
cooperative control approach [29]. In this hierarchical scheme
of both organization and coordination of distributed entities,
different units (in our case devices composing a microgrid)
form groups managed by some group controllers, which are
responsible for controlling the behavior of the groups. In a
case of our microgrid we will have 4 different groups of simi-
lar devices: controllable loads (CL), controllable microsources
(CM), non-controllable microsources (NCM) and energy stor-
ages (ES). Each group controller is responsible for exchanging
information with other group controllers, making appropriate
calculations and on their basis, for making decisions on the
proper choice of the operating points for all the devices be-
longing to the group it is in charge of. In our version of the
cooperative control approach each device also has its own local
controller responsible for communication with the group con-
troller, listening to its orders and applying the demanded operat-
ing point on the device. Apart from group controllers and local
controllers there exists also a single master controller (MC) re-
sponsible for monitoring the state of different network compo-
nents (lines, transformers, switchgears and so on) and inform-
ing group controllers about all failures and incidents within the
network, as well as about the mode of the microgrid operation.
Secondly, for each device an operating point can be set. We
understand the operating point as a pair of active power P and
reactive power Q values which can be applied on the device.
In a case of different types of devices we have different lev-
els of flexibility when choosing the operating points. In a case
of all energy storages, controllable loads and controllable mi-
crosources we can choose any operating point we want from
the available – individual for each i-th device – range of reg-
ulation possibilities spanning from Pmin, i to Pmax, i and from
Qmin, i to Qmax, i . For all non-controllable microsources (renew-
able energy sources) the choice is discrete. We can turn the
i-th device on and then it will work with the operating point
(Pcurrent, i ; Qcurrent, i ) determined mostly by the current weather
Bull. Pol. Ac.:
Bull. Pol. Ac.:Tech.
Tech.68(4)
68(4)2020
2020
controlininrural lowlow
rural voltage microgrids
voltage microgrids
conditions and the individual technical characteristics of the de-
vice. We can also turn it off and apply on the device the oper-
ating point (0; 0). We have also decided that all devices should
work with some individually chosen, fixed for them, constant
values of power factors (tan(ϕ)), which will not be changed
during the whole optimization process. This way we only need
to choose the level of generated or consumed active power for
each device – the level of reactive power will be automatically
determined. Such an assumption does not seem to be a big lim-
itation – it is a common practice to keep the power factor of the
device unchanged for a given short time interval. However, it
will simplify our optimization process significantly, by turning
our optimization problem into one-dimensional one (only the
levels of active powers need to be chosen).
Thirdly, we take into account that the microgrid should be
able to work in two different operation modes – synchronous
one and island one. We are going to consider 7 different ob-
jective (criterion) functions (CF) [14, 18, 31]: (1)/(2) min-
imization/maximization of the amount of energy imported
from/exported to the network of the distribution system op-
erator, (3) minimization of active power losses, (4)/(5) maxi-
mization/minimization of the amount of energy generated from
renewable/non-renewable energy sources, (6) minimization of
costs related to the operation of the microgrid, (7) maximiza-
tion of profits resulting from the operation of the microgrid.
Two of them: (3) and (7) will be analyzed in much more detail
in the paper. The mathematical formulation of them is presented
below [14, 31].
1. Minimization of active power losses in the microgrid (CF3).
The objective function has the following form:
Fobj = min(∆P), (1)
where Fobj is the objective function and ∆P is the sum of all
active power losses in the microgrid.
2. Maximization of all profits related to the operation of the
microgrid (CF7). The objective function has the following
form:
Fobj = max(PRopMG ), (2)
PRopMG = Atot · ps −CopMG , (3)
where: PRopMG are the profits related to the operation of the mi-
crogrid, ps is a price, per unit, of the energy sold to the clients,
including also the DSO power grid, Atot is the sum of energy
sold, CopMG are the costs related to the operation of the micro-
grid within the optimization period.
Last but not least, we assume that the final solution being
the result of our optimization process needs to be a feasible
one – it cannot violate any of the optimization constraints. The
feasibility of solutions is checked with the use of Newton-
Raphson method of calculating power flows. The set of the
following optimization constraints should be met: (a) the long-
term current-carrying capacities of all lines cannot be exceeded,
(b) the rated powers of all transformers should not be exceeded,
(c) nodal voltages for all nodes should be within permissible
limits, (d) levels of energy stored in energy storages should be
6655
 M.
M.Parol,
Parol, P.
P. Kapler,
Kapler, J.
J. Marzecki,
Marzecki, R.
R. Parol, M. Połecki,
within the permissible ranges of values. In the publications [8,
31] detailed forms of the constraints are given. If the candidate
solution does not violate any of these constraints it is treated
as a feasible one and can compete with other ones to become a
(sub)optimal solution returned by the decentralized algorithm.
4. Description of proposed microgrid distributed
control logic algorithm
The general idea of the proposed approach of decentralized
determination of operating points in a microgrid is based on
a quite simple heuristic algorithm belonging to the family of
Monte Carlo methods. At the same time we also want to
take advantage of the cooperative control approach, in case of
which each group of devices (in our case: energy storages, non-
controllable microsources, controllable microsources and con-
trollable loads) will be managed by a group controller, which
is responsible for calculating and setting the operating points
of all devices belonging to the group. In our approach we want
each group controller to sum all power generation or power con-
sumption capabilities of the devices which it is in charge of and
to generate a set of candidate solutions which cover uniformly
the range of power generation or power consumption capabil-
ities of the group treated as a whole. Within this process only
active powers need to be taken into account – because the val-
ues of power factors are fixed for all devices, reactive powers
will always be determined automatically. Each group controller
cares only about the devices composing its own group – it treats
devices belonging to other groups as the ones, for which the
operating points have been already optimized and set by other
group controllers. All group controllers perform such compu-
tations one-by-one, in a synchronized or a non-synchronized
manner. Synchronization is not necessary. However, taking into
account that the whole approach is an iterative one – many invo-
cations of the algorithm on all group controllers are needed to
obtain the (sub)optimal solution, it can speed up the whole op-
timization process. After some number of algorithm iterations
on all group controllers a balance point should be achieved and
the (sub)optimal solution should be found.
The details on how a single group controller creates its candi-
date solutions set is illustrated in Fig. 6. Let us assume the group
of devices consists of N devices. First it sorts them according to
their power generation/power consumption capabilities. Then it
divides the whole group into K separate subgroups. After that,
(K × L) + 2 candidate solutions are generated, L for each sub-
group. The numbers K and L should be chosen arbitrarily, in-
dependently of the value of the parameter N. Subgroups are
processed in order, one-by-one, beginning from the subgroups
containing the devices with the biggest capabilities and ending
with the subgroups containing the “smallest” devices. When
generating L candidate solutions for the subgroup with index
k, all devices belonging to the subgroups with indices lower
than k must work with their full power generation/power con-
sumption capabilities, while all devices belonging to the sub-
groups with indices greater than k must be turned off. Devices
belonging to the subgroup k work with a fraction of their full
666
6 Bull. Pol.
Parol, M. Połecki, and
and Ł.
Ł. Rokicki
Rokicki
Fig. 6. An illustration of the generation of candidate solutions set in
case of a single group controller
potential. In case of each of L candidate solutions we gener-
ate operating points for these devices in a totally random way.
At the end of the candidate solutions generation procedure we
also add two special solutions to the whole candidate set – the
one where all devices belonging to the group are turned on and
work with their full potential and the one where all devices are
turned off. After the generation part, all (K × L) + 2 solutions
need to be evaluated. Different objective functions are possible.
We propose 7 different functions. The feasibility of candidate
solutions is checked by power flow calculations procedure per-
formed with the use of Newton-Raphson method. The best can-
didate solution is chosen as the (sub)optimal one for the group.
The operating points are then applied on the physical devices
composing the group.
The last question to consider is how often to invoke the op-
timization algorithm on a single group controller. We have al-
ready said that the same algorithm needs to be executed on each
of the group controllers. The invocations on different group
controllers do not need to be synchronized (of course they can
be, but it is not required).
Many iterations consisting of single algorithm invocations
for all group controllers (one invocation per group controller)
are needed to achieve the balance point, when the final solu-
tion proposed by all group controllers will be one and the same
(group controllers will somehow agree on the final solution).
How many iterations will be needed to achieve this point and
find the (sub)optimal solution? How often to invoke the algo-
rithm on a single group controller? In the Fig. 7 three possi-
ble approaches are presented. Let’s assume that the length of
the optimization period is fixed and set to M minutes. We call
the approach presented in the upper part of Fig. 7 the “initial-
phase” approach. In case of this approach, at the beginning of
each optimization period a fixed number of iterations is sched-
uled. We seek for the best possible solution in terms of its
quality (objective function value) in each iteration. In case of
two other approaches the optimization process is being con-
ducted during the whole optimization period. We invoke the
Bull. Pol.Ac.:
Ac.:Tech.
Tech.68(4)
68(4) 2020
2020
 Effective
Effective approach
approach to to distributedoptimal
distributed optimal operation
operation control
Fig. 7. An illustration of how often the optimization algorithm can be
invoked on a single group controller. Top: initial-phase approach, mid-
dle: continuous-dense approach, bottom: continuous-sparse approach
optimization algorithm continuously at regular intervals. How-
ever, changing operating points all the time, turning the devices
on and then after a while turning them off and then turning
them on once again is not what we look for. We would rather
like to calculate some set of (sub)optimal operating points at
the beginning of the optimization period and then listen for
some changes in the microgrid operational conditions (chang-
ing load demands, changing solar and wind generation capa-
bilities) and be able to react to them. That is why instead of
only looking at the value of the objective function, we can cal-
culate some combined score which takes into account both the
objective function value and the distance between the currently
applied solution and the newly proposed ones. At the beginning
of the optimization period the objective function value compo-
nent is of a greater importance, while at the end the distance
component should be much more significant. This pattern re-
sembles the behavior known from the simulated annealing al-
gorithm, with the relative importance of components playing
the role of the temperature. The “continuous-dense” approach
(presented in the middle part of Fig. 7) is the exact realiza-
tion of the scheme described above. However, this approach
can quickly become a “computation-heavy” one. An acceptable
compromise can be a “continuous-sparse” approach (presented
in the lower part of Fig. 7) – computations are performed dur-
ing the whole optimization period, but the intervals between
them are much longer. At the same time, we still calculate the
score based on two separate components. In this approach, the
temperature is stable and has the same value during the whole
optimization period.
5. Scenarios of behavior of the microgrid control
system in case of the communication loss
This section presents different scenarios of behavior of the mi-
crogrid control system (MCS) with respect to the communica-
Bull. Pol. Ac.:
Bull. Pol. Ac.:Tech.
Tech.68(4)
68(4)2020
2020
controlininrural lowlow
rural voltage microgrids
voltage microgrids
tion failure. Details are presented in Fig. 8. It shows features of
MC, MGCs, LCs and interrelations between them.
The proposed algorithm is based on two independent com-
munication channels. The primary channel is used to control
and gather information from LCs. The secondary channel is
used to transmit data measured by independent grid analyzers.
This approach ensures MCS may operate with full observability
in case of the communication fault of a channel.
We may distinguish the following classes of the main group
controllers:
• controllable generation units (CGU),
• controllable loads (CL),
• energy storage units (ESU),
• uncontrollable generation units (uCGU),
• uncontrollable loads (uCL),
• circuit breakers (CB).
Circuit breakers can be opened or closed remotely. Uncon-
trollable loads and generation units can be switched on or off,
but generally under normal conditions, they follow their de-
mand or generation profiles. Typical uncontrollable generation
units are photovoltaic panels and wind turbine-generator sets.
In the event of losing the continuity of communication be-
tween particular groups of devices and between individual de-
vices within the group we can distinguish several cases:
• uncontrollable LCs (uLCs),
• controllable LCs,
• uncontrollable MGCs (uMGCs),
• controllable MGCs,
• master controller – MC.
The loss of communication with uLCs and uMGCs changes
the functioning of optimal control algorithm marginally. The
only malfunction which happens during the communication
failure is the disability of uLCs to be switched on/off remotely
in the unintended island operation. Due to the fact that protec-
tion systems requirements (selectivity, sensitivity and timeous)
are based on local parameters, the discussed case should not
cause any problems during disturbances.
More complicated issues concerning communication losses
occur when controllable LCs or MGCs are affected. The gen-
eral rule of repair algorithms is a transfer of devices with broken
communication to a different group of control. Details are pre-
sented in Table 1 and Fig. 8. When the controllable MGCs loses
its communication with MC, repair algorithm is split into two
different modes of operation. The first mode is activated when
the duration of communication loss is shorter then optimiza-
tion period. In this mode, MGC tries to optimize each LC in its
group based on the last data received from MC. Otherwise, if
the duration of communication loss is greater than optimization
period a second mode of operation is activated – each LC in
a group with failed MGC follows the transfer pattern given in
Table 1 and Fig. 8.
Despite the lack of communication with selected objects, the
optimization process can proceed. However, the objective func-
tion value may not be as accurate as in the case when the com-
munication is fully operative.
Another failure, which may happen, is the loss of connec-
tion between all MGCs and MC. In such a case optimization
6677
 M.
M.Parol,
Parol, P.
P. Kapler,
Kapler, J.
J. Marzecki,
Marzecki, R.
R. Parol, M. Połecki,
Parol, M. Połecki, and
and Ł.
Ł. Rokicki
Rokicki

Fig. 8. Structure and functionalities of MCS

process could be less accurate. An unintentional island mode
is impossible to last and an intentional island mode as well as
a synchronous mode of operation may last. The accuracy of
the optimization process (and possible balance) depends on the
ramps demand and generation by uCGU.
To sum up, the cooperative control is more robust than the
centralized one in case of communication faults. The pro-
posed repair algorithms allow for the continuation of optimiza-
tion process while different layers of control structure are af-
fected.

668
8 Bull. Pol.
Bull. Pol.Ac.:
Ac.:Tech.
Tech.68(4)
68(4) 2020
2020
 Effective approach
Effective to distributedoptimal
optimal operation control in rural low voltage microgrids
Effective approach
approach to
to distributed
distributed optimal operation
operation control
control in
in rural
rural low
low voltage
voltage microgrids
microgrids

Table
Table 11 mization
mization calculations
calculations were
were done
done for
for summer
summer period
period (Wednes-
(Wednes-
Transfer
Transfer to different group of control
to different group of control after
after communication
communication loss day,
loss day, 12:00). We consider both the synchronous and
12:00). We consider both the synchronous and island
island op-
op-
eration
eration modes,
modes, asas well
well as
as 77 different
different objective
objective functions.
functions. The
The
Before the
the loss
loss After the
the loss
loss details
of
Before
communication of
After
communication details on
on the
the specification
specification of
of different
different objective
objective functions
functions and
and
of communication of communication constraints have been already given in the paper [31].
constraints have been already given in the paper [31].
controllable generation units
controllable generation units uncontrollable generation units
uncontrollable generation units
controllable loads
loads uncontrollable loads
loads 6.2.
6.2. Results
Results of of optimization
optimization calculations
calculations with with the the use
use of of al-
al-
controllable uncontrollable
gorithm of distributed control logic.
gorithm of distributed control logic. In the current discus- In the current discus-
energy storage
energy storage systems
systems
uncontrollable loads
loads sion,
(charging)
(charging)
uncontrollable sion, wewe willwill focus
focus onon the
the minimization
minimization of of active
active power
power losseslosses
and
and the maximization of profits (the difference between the
the maximization of profits (the difference between the sum
sum
energy storage
energy storage systems
systems
(discharging)
uncontrollable generation
uncontrollable generation units
units of
of the
the revenues
revenues from from selling
selling thethe energy
energy to to customers
customers and and to to the
the
(discharging) DSO
DSO network
network and and the
the sum
sum of of costs
costs related
related to to generating
generating the the
energy
energy in in microsources
microsources or or purchasing
purchasing the the energy
energy from
from the the DSO
DSO
6.
6. Case
Case study network
study network (also (also thethe distribution
distribution costs
costs are
are included)).
included)). The The feasibil-
feasibil-
ity
ity of all the (sub)optimal solutions proposed by our distributed
of all the (sub)optimal solutions proposed by our distributed
6.1.
6.1. Description
Description of of a
a test
test microgrid.
microgrid. To To evaluate
evaluate the
the correct-
correct- approach
approach has has been
been checked
checked withwith thethe Newton-Raphson
Newton-Raphson method method
ness and efficiency of the proposed approach
ness and efficiency of the proposed approach some calculations some calculations for calculating power
for calculating power flows. flows.
have
have been
been made
made withwith the
the use
use ofof aa sample
sample test
test microgrid
microgrid located
located In
In Table
Table 22 we we can
can seesee the
the comparison
comparison of of the
the objective
objective
in
in a typical countryside (see Fig. 9). The test microgrid
a typical countryside (see Fig. 9). The test microgrid has has al-
al- function
function values obtained from the calculations based on
values obtained from the calculations based on both:
both:
ready
ready been described in detail in the paper [31], where the main
been described in detail in the paper [31], where the main the method based on the centralized control
the method based on the centralized control logic described logic described
attention
attention was was given
given toto methods
methods based based on on thethe centralized
centralized con-con- in
in detail
detail in in [31]
[31] and
and thethe currently
currently discussed
discussed decentralized
decentralized
trol
trol paradigm, and the current approach based on
paradigm, and the current approach based on the
the decen-
decen- algorithm.
algorithm. In case of the distributed algorithm
In case of the distributed algorithm the the compu-
compu-
tralized
tralized control
control logic
logic was
was discussed
discussed only only very
very briefly.
briefly. The
The test
test tations
tations consisted of 10 iterations. In each iteration 4
consisted of 10 iterations. In each iteration 4 group
group
network
network consists of 22 non-controllable loads (farms with the
consists of 22 non-controllable loads (farms with the controllers
controllers have have been
been invoked,
invoked, always
always in in the
the following
following order: order:
total
total amount
amount of of demanded
demanded active active power
power equal
equal to to 50.451
50.451 kW),kW), controllable
controllable loads, loads, controllable
controllable microsources,
microsources, non-controllable
non-controllable
22 controllable
controllable loads
loads (public
(public utility
utility buildings
buildings withwith the
the range
range of of microsources
microsources and energy storages. We
and energy storages. We assume,
assume, that that the the
total
total demanded active power between 6.4 kW and 9.6 kW), one
demanded active power between 6.4 kW and 9.6 kW), one distributed
distributed algorithm works in its “initial-phase” variant,
algorithm works in its “initial-phase” variant,
controllable
controllable microsource
microsource (with (with thethe active
active power
power generation
generation ca- ca- which
which is is the
the simplest
simplest one,one, always
always seeking
seeking for for the
the best
best quality
quality
pacity
pacity up to 49.0 kW), 7 non-controllable microsources making
up to 49.0 kW), 7 non-controllable microsources making solution.
solution. The starting point for the whole optimization process
The starting point for the whole optimization process
use
use ofof solar
solar and
and wind
wind energy
energy (with
(with thethe sum
sum of of the
the installed
installed ac-ac- was
was the the test test microgrid
microgrid with with the the controllable
controllable microsource
microsource
tive
tive power equal to 40.0 kW and the sum of active power gener-
power equal to 40.0 kW and the sum of active power gener- working
working with the full power generation capacity,
with the full power generation capacity, the the energy
energy
ation
ation capacities
capacities predicted
predicted for for the
the time
time interval
interval of of making
making calcu-
calcu- storage
storage working with the full power charging capacity, all
working with the full power charging capacity, all
lations
lations equal to 17.413 kW) and one energy storage (with
equal to 17.413 kW) and one energy storage (with pos-
pos- controllable
controllable loads loads with
with their
their operating
operating points
points in in the
the middle
middle
sible
sible operating
operating points
points between
between charging
charging 20 20 kW kW ofof active
active power
power of
of their
their possible
possible power
power consumption
consumption ranges, ranges, half half of of thethe
and
and discharging
discharging 20 20 kW
kW of of active
active power).
power). All All devices
devices work
work with
with non-controllable
non-controllable microsources working with their full power
microsources working with their full power
constant
constant tan(ϕ) factors – in case of all non-controllable mi-
tan(ϕ) factors – in case of all non-controllable mi- generation
generation potential
potential and
and thethe other
other half
half of
of microsources
microsources turned turned
crosources
crosources and and the
the energy
energy storage
storage the the factor
factor is is equal
equal to to 0.0,
0.0, off.
off. We can see that for all the considered objective
We can see that for all the considered objective functions
functions
while
while forfor other
other devices
devices it it takes
takes some some positive
positive values
values (power
(power and
and operation
operation modes modes thethe results
results areare always
always almost
almost exactly
exactly the the
factor
factor tan(ϕ) equal to 0.395 for each load and power
tan(ϕ) equal to 0.395 for each load and power factor
factor same
same for both approaches being compared: the centralized al-
for both approaches being compared: the centralized al-
tan(ϕ)
tan(ϕ) equal
equal to to 0.75
0.75 forfor controllable
controllable microsource).
microsource). The The opti-
opti- gorithm
gorithm and and thethe distributed
distributed algorithm.
algorithm. Because
Because the the centralized
centralized

Fig.
Fig. 9.
9. Key
Key diagram
diagram of
of the
the LV
LV microgrid
microgrid on
on aa typical
typical countryside
countryside (MV_L
(MV_L – – MV
MV overhead
overhead line,
line, RE
RE (SG)
(SG) –– reciprocating
reciprocating engine
engine with
with internal
internal
combustion
combustion (engine-generator set), BES – battery energy storage, PV – photovoltaic panels, WT – wind turbine-generator set); elaborated on
(engine-generator set), BES – battery energy storage, PV – photovoltaic panels, WT – wind turbine-generator set); elaborated on the
the
basis of [31]
basis of [31]

Bull. Pol. Ac.:

Bull. Pol. Ac.:Tech.
Tech.68(4)
68(4)2020
2020 6699
Bull. Pol. Ac.: Tech. 68(4) 2020 9
 M.
M. Parol, P. Kapler, J. Marzecki, R. Parol, M.
M. Połecki, and
and Ł. Rokicki
Rokicki
M. Parol,
Parol, P.
P. Kapler,
Kapler, J.
J. Marzecki,
Marzecki, R.
R. Parol,
Parol, M. Połecki,
Połecki, and Ł.
Ł. Rokicki

algorithm makes use of more sophisticated mechanisms, it is a microsources and the energy storage work with their full active
more time-taking approach. It checks for a bigger number of power generation capabilities. In case of the synchronous mode
possible solutions. We can then assume that the centralized al- of operation, the energy purchased from the DSO is cheaper
gorithm is accurate. And since the results are the same, so is the than the one generated by the reciprocating engine (controllable
distributed one. This way we can prove our distributed approach microsource), so the import of the energy from the DSO net-
behaves in a proper way and yields correct results. work takes place. In case of the island operation mode we do
not have this option. Also, in case of synchronous work with
Table
Table 22 the DSO network, increasing the amount of energy purchased
The
The comparison
comparison ofof the
the results
results of
of optimization
optimization calculations
calculations for
for test
test mi-
mi- from the DSO to deliver this energy to the controllable loads
crogrid
crogrid in summer obtained with the use of centralized control logic
in summer obtained with the use of centralized control logic is often profitable, but only to a point. After that point the ad-
and
and distributed
distributed control
control logic;
logic; based
based partially
partially on
on [31]
[31] ditional profits will be lower than the costs resulting from the
bigger level of active power losses.
Centralized
Centralized control
control logic
logic Distributed
Distributed control
control logic
logic In Fig. 10, 11 we can see how the objective function values
Criterion
Criterion
function Summer,
Summer, Wednesday,
Wednesday, 12:00
12:00 Summer,
Summer, Wednesday,
Wednesday, 12:00
12:00 changed over all optimization process iterations, when the cho-
function
no.
no. Synchronous
Synchronous Island
Island Synchronous
Synchronous Island
Island sen objective function was the minimization of active power
mode
mode mode
mode mode
mode mode
mode losses and the test microgrid worked synchronously with the
11 0.000 –– 0.000 –– DSO network. In Fig. 10 all group controllers have been pre-
0.000 0.000
sented separately. In Fig. 11 the whole optimization process
22 7.087
7.087 –– 7.087
7.087 –– is presented as a single curve. However, also the second curve
33 0.764
0.764 0.776
0.776 0.764
0.764 0.776
0.776 for the island mode of operation has been presented. The simi-
44 4.353 4.353 4.353 4.353 lar curves for the maximization of profits have been presented
4.353 4.353 4.353 4.353
in Fig. 12 and 13. We can see that the minimization of active
55 0.000
0.000 5.054
5.054 0.000
0.000 5.054
5.054 power losses is a little harder objective function to optimize.
66 4.613
4.613 5.450
5.450 4.613
4.613 5.451
5.451 Also, in case of both objective functions 3 or 4 iterations (12–
77 2.787 1.940 2.787 1.940 16 of distributed algorithm invocations) is usually enough for
2.787 1.940 2.787 1.940
group controllers to reach and agree on the (sub) optimal solu-
tion.
In Table 3 we can see which operating points have been cho- We can see in Fig. 14 to 17 and 18 to 21 how the specific
sen by different group controllers for different objective func- group controllers worked over time – which operating points
tions and different operation modes. In a case of active power they chose – for both objective functions and the synchronous
losses, to minimize them, the main motivation is to reduce mode of operation. For minimization of active power losses all
the amount of energy transmitted over the different network non-controllable sources were turned on in the first iteration.
branches and as a result to reduce the amount of energy con- Similarly, the amount of energy delivered to the controllable
sumed by all loads – that is why all controllable loads work loads was reduced to the absolute minimum at the start. Over
always with their minimum operating points. Also, local en- time we also observed the decreasing usage of the controllable
ergy sources are promoted over the DSO network – that is why microsource and the gradual switch of the operation mode of
all non-controllable microsources are turned on. In case of the the energy storage – from charging to the discharging. In the
island mode of operation we cannot make use of energy deliv- case of the maximization of profits, the group controllers de-
ered by the DSO network anymore, so the usage of the control- cided to increase the usage of cheaper energy sources (non-
lable microsource and the energy storage is more prominent. In controllable microsources and the energy storage) to the maxi-
case of the maximization of profits the main motivation is to use mum at the start. However, setting the right operating points for
cheaper sources of energy first – that is why all non-controllable other groups of devices was a little harder.

Table
Table 33
Operating
Operating points
points chosen
chosen for
for groups
groups by
by respective
respective group
group controllers
controllers –– the
the comparison
comparison for
for different
different objective
objective functions
functions and
and operation
operation modes
modes
Controllable
Controllable Controllable
Controllable Non-controllable
Non-controllable Energy storages Active
Active power
power
Criterial
Criterial function,
function, Energy storages imported from Total
Total active
active Objective
Objective
loads
loads active
active microsources
microsources microsources
microsources imported from power losses function value
operation
operation mode
mode active power
active power power losses function value
power
power active power
active power active power
active power the DSO network
the DSO network
Active
Active power
power losses,
losses, 6.400000
6.400000 25.719077
25.719077 17.413000
17.413000 −10.866746
−10.866746 3.616462
3.616462 0.764
0.764 0.764399
0.764399
synchronous
synchronous
Active
Active power
power losses,
losses, 6.400000
6.400000 26.076424
26.076424 17.413000
17.413000 −14.133197
−14.133197 0.000000
0.000000 0.776
0.776 0.775676
0.775676
island
island
Profits,
Profits, synchronous
synchronous 7.517190
7.517190 0.000000
0.000000 17.413000
17.413000 −20.000000
−20.000000 21.639229
21.639229 1.084
1.084 2.787106
2.787106
Profits,
Profits, island
island 6.400000
6.400000 20.215482
20.215482 17.413000
17.413000 −20.000000
−20.000000 0.000000
0.000000 0.782
0.782 1.939689
1.939689
Notes:
Notes: Negative
Negative sign
sign (minus)
(minus) in
in case
case of
of energy
energy storage
storage active
active power
power means
means that
that this
this unit
unit is
is in
in discharging
discharging mode;
mode; active
active power
power values
values as
as well
well as
as
active power losses values are given in [kW].
active power losses values are given in [kW].

670
10 Bull. Pol. Ac.:
Bull.
Bull. Pol.
Tech. 68(4)
Pol. Ac.:
Ac.: Tech.
Tech. 68(4)
2020
68(4) 2020
2020
 Effective
Effective approach
approach to to distributedoptimal
distributed optimal operation
operation control
controlininrural lowlow
rural voltage microgrids
voltage microgrids

Minimization of active power losses - consecutive iterations Maximization of profits - synchronous mode vs island mode

1.6 4

1 .5 3
Controllable loads group controller
1 .4 2
Controllable microsources group controller
1 .3 1
Objective function value

Non-controllable microsources group controller

Objective function value

1 .2 0
1 5 9 13 17 21 25 29 33 37
Energy storages group controller
-1
1 .1
Synchronous
-2
1
Island
-3
0.9

-4
0.8
-5
0.7
-6
0.6
1 2 3 4 5 6 7 8 9 10 -7
Iterations Executions of algorithm

Fig. 10. The illustration of the process of optimizing the objective func- Fig. 13. The illustration of the process of optimizing the objective func-
tion value by different group controllers. All group controllers shown tion value. The efforts of all group controllers shown together with a
separately. Minimization of active power losses chosen as an objective single curve. Maximization of profits chosen as an objective function.
function Two different operation modes considered: synchronous one and is-
land one
Minimization of active power losses - synchronous mode vs island mode

1.6
Minimization of active power losses - controllable loads

1.4 12
Synchronous

Island
1.2
10
Objective function value

8
0.8
Active power [kW]

0.6 6

0.4 Controllable loads group controller

4
Min
0.2
Max
2
0
1 5 9 13 17 21 25 29 33 37
Executions of algorithm
0
1 2 3 4 5 6 7 8 9 10
Fig. 11. The illustration of the process of optimizing the objective func- Iterations
tion value. The efforts of all group controllers shown together with a
single curve. Minimization of active power losses chosen as an objec- Fig. 14. The operating point chosen for the group of controllable loads
tive function. Two different operation modes considered: synchronous by respective group controller. Minimization of active power losses
one and island one chosen as an objective function

Maximization of profits - consecutive iterations

Minimization of active power losses - controllable microsources
4
60
3

50
2
Controllable microsources group controller
1 40
Min
Objective function value

0
Active power [kW]

Max
1 2 3 4 5 6 7 8 9 10 30
-1
Controllable loads group controller
20
-2
Controllable microsources group controller
-3 10
Non-controllable microsources group controller

-4
Energy storages group controller
0
-5 1 2 3 4 5 6 7 8 9 10

-6 -10
Iterations Iterations

Fig. 12. The illustration of the process of optimizing the objective func- Fig. 15. The operating point chosen for the group of controllable
tion value by different group controllers. All group controllers shown microsources by respective group controller. Minimization of active
separately. Maximization of profits chosen as an objective function power losses chosen as an objective function

Bull. Pol. Ac.:

Bull. Pol. Ac.:Tech.
Tech.68(4)
68(4)2020
2020 671
11
 M.
M.Parol,
Parol, P.
P. Kapler,
Kapler, J.
J. Marzecki,
Marzecki, R.
R. Parol, M. Połecki,
Parol, M. Połecki, and
and Ł.
Ł. Rokicki
Rokicki

Minimization of active power losses - non-controllable microsources Maximization of profits - controllable microsources

20 60

50

15

Non-controllable microsources group controller Controllable microsources group controller

40

Min Min
Active power [kW]

Active power [kW]

10
30

Max Max

20
5

10

0
1 2 3 4 5 6 7 8 9 10
0
1 2 3 4 5 6 7 8 9 10

-5
Iterations -10
Iterations

Fig. 16. The operating point chosen for the group of non-controllable Fig. 19. The operating point chosen for the group of controllable
microsources by respective group controller. Minimization of active microsources by respective group controller. Maximization of profits
power losses chosen as an objective function chosen as an objective function

Minimization of active power losses - energy storages Maximization of profits - non-controllable microsources

25
20

20

15
15
Energy storages group controller
Non-controllable microsources group
10 Min controller

Max Min
Active power [kW]

5
Active power [kW]

10

0 Max
1 2 3 4 5 6 7 8 9 10
-5
5

-10

-15
0
1 2 3 4 5 6 7 8 9 10
-20

-25
Iterations -5
Iterations

Fig. 17. The operating point chosen for the group of energy storages Fig. 20. The operating point chosen for the group of non-controllable
by respective group controller. Minimization of active power losses microsources by respective group controller. Maximization of profits
chosen as an objective function chosen as an objective function

Maximization of profits - controllable loads Maximization of profits - energy storages

12
25

20
10
15
Energy storages group controller

8 10 Min
Active power [kW]

Max
Active power [kW]

5
6
0
Controllable loads group controller
1 2 3 4 5 6 7 8 9 10
Min -5
4
Max
-10

2
-15

-20
0
1 2 3 4 5 6 7 8 9 10
Iterations -25
Iterations

Fig. 18. The operating point chosen for the group of controllable loads Fig. 21. The operating point chosen for the group of energy storages
by respective group controller. Maximization of profits chosen as an by respective group controller. Maximization of profits chosen as an
objective function objective function

672
12 Bull. Pol.
Bull. Pol.Ac.:
Ac.:Tech.
Tech.68(4)
68(4) 2020
2020
 Effective
Effective approach
approach to to distributedoptimal
distributed optimal operation
operation control
controlininrural lowlow
rural voltage microgrids
voltage microgrids

We have also decided to study more advanced aspects of
our decentralized approach. First, the impact of different group
devices orderings (in which order the groups calculate their
(sub)optimal operating points) on the speed of the optimization
process has been analyzed. The results have been illustrated in
Fig. 22 and 23. The most important finding is that no matter
what the ordering is, the time of 3 or 4 iterations (12–16 of dis-
tributed algorithm invocations) is still enough to find the even-
tual solution. We can also see that the ordering in which the
Energy Storages group comes first seems to be faster than the
others, but we have too little evidence to treat is as a definite
rule – some further research is necessary.
Minimization of active power losses - different groups orderings
information on the currently applied operating points. The re-
sults are displayed in Fig. 24 and 25. We have a common curve
for all group controllers (40 consecutive algorithm invocations).
The assumption is that the communication is lost during the it-
erations from 2 to 5. In the iteration no. 6 the communication
is back and all group controllers can talk to each other without
any problems. For both the minimization of active power losses
and the maximization of profits we can observe some charac-
teristic cycles for iterations from 2 to 5 (algorithm invocations
from 5 to 20). Within such cycles all properly communicating
with one another groups have their own single operating point
(and the corresponding objective function value). The operat-
ing point (and the corresponding objective function value) of

1.6 Minimization of active power losses - lost communication

1.4 1.6

Standard Standard
1.2 1.4
Alternative 1 Lost CL
Objective function value

1 Alternative 2 1.2 Lost CM

Alternative 3 Lost NCM

Objective function value

0.8 1
Lost ES

0.6 0.8

0.4 0.6

0.2 0.4

0 0.2
1 5 9 13 17 21 25 29 33 37
Executions of algorithm
0
1 5 9 13 17 21 25 29 33 37
Fig. 22. The illustration of the process of optimizing the objective func- Executions of algorithm

tion value. The efforts of all group controllers shown together with a
Fig. 24. The illustration of the process of optimizing the objective func-
single curve. Minimization of active power losses chosen as an objec-
tion value. The efforts of all group controllers shown together with a
tive function. The comparison for different group orderings
single curve. Minimization of active power losses chosen as an ob-
jective function. The comparison for different experiments, when dif-
ferent group controllers lose the communication with all other groups
Maximization of profits - different groups orderings during the iterations from 2 to 5

3
Maximization of profits - lost communication
2
4

1
3
Objective function value

0
1 5 9 13 17 21 25 29 33 37 2

-1
1
Objective function value

-2
Standard 0
1 5 9 13 17 21 25 29 33 37
-3 Alternative 1
-1
Standard
-4 Alternative 2
-2 Lost CL
Alternative 3
-5 Lost CM
-3
Lost NCM
-6
Executions of algorithm -4 Lost ES

Fig. 23. The illustration of the process of optimizing the objective func- -5

tion value. The efforts of all group controllers shown together with a
single curve. Maximization of profits chosen as an objective function.
The comparison for different group orderings
Moreover, we also have investigated the scenario when one
of the groups was losing the communication with all other
groups and as a result was neither able to send nor to receive the
-6
Executions of algorithm
Fig. 25. The illustration of the process of optimizing the objective func-
tion value. The efforts of all group controllers shown together with a
single curve. Maximization of profits chosen as an objective function.
The comparison for different experiments, when different group con-
trollers lose the communication with all other groups during the itera-
tions from 2 to 5

Bull. Pol. Ac.:

Bull. Pol. Ac.:Tech.
Tech.68(4)
68(4)2020
2020 673
13
 M.
M.Parol,
Parol, P.
P. Kapler,
Kapler, J.
J. Marzecki,
Marzecki, R.
R. Parol, M. Połecki,
Parol, M. Połecki, and
and Ł.
Ł. Rokicki
Rokicki

the isolated group is different. What is interesting, none of these
operating points is the one that is actually applied to the devices
at that moment – no group has a full and true overview of what
is happening in all devices within the network. What is more,
we can see the optimization potential is locked during those cy-
cles. When the communication is back this potential becomes
unlocked and further improving of the objective function values
is possible.
Last but not least, we have also decided to analyze the impact
of the uncertainty of predictions of demanded active power lev-
els of controllable and non-controllable loads and active power
levels generated by non-controllable microsources. It is worth
recalling that we have made an assumption that our test mi-
crogrid makes use of the “initial-phase” variant of the decen-
tralized algorithm. It means all calculations are made only at
the beginning of every 15-minute-long optimization period (this
time interval is arbitrary and may be shorter if needed) and once
applied they will not change during that period. It means when
making the computations we do not know the true values of de-
manded active power levels and active power levels generated
by photovoltaic power plants and wind turbine-generator sets.
We need to rely on predictions and these, sometimes, are not
very precise. That is why we have decided to check what will
happen when we change all predicted values (for 24 loads and
7 non-controllable microsources) by at most 20% in a random
way. We can observe the results in Tables 4 and 5. We can see
that even if the predictions are a little bit different for single de-
vices, the operating points chosen by group controllers remain
more or less the same.
6.3. Results of optimization calculations proving the sta-
bility of the algorithm. In order to prove the reliability and
stability of our algorithm we decided to make two additional
experiments. Our intention here was to analyze the impact of
some unexpected and accidental changes of the microgrid’s
work conditions on the functioning of our algorithm.
In the first experiment we analyze a situation when a short-
circuit takes place in one of the lines of our sample microgrid
(somewhere in the upper half – upper feeder of the microgrid).
We assume here that during the experiment the microgrid
operates in the island mode. As a result, the protection systems
of the microgrid start to work and the whole upper half of the

Table 4
Operating points chosen for groups by respective group controllers for different sets of predictions of demanded active power for loads and
weather conditions for non-controllable microsources; the minimization of active power losses chosen as an objective function
Controllable loads

Controllable loads

Controllable loads
max active power
Non-controllable

Non-controllable

Non-controllable

Energy storages
loads requested

max requested

function value
min requested

microsources

microsources

microsources
active power

active power

active power

active power

active power

active power

active power
Controllable

Active power

Objective
losses

Standard 50.450933 6.400000 9.600000 17.413000 6.400000 25.719077 17.413000 −10.866746 0.764399
Alternative 1 52.513632 6.157218 9.354789 18.884147 6.157218 26.386413 18.884147 −10.706908 0.805831
Alternative 2 52.203232 7.305418 9.759093 16.197076 7.305417 27.643244 16.197076 −12.426787 0.881627
Alternative 3 52.803455 6.691778 11.214232 17.048570 6.691778 27.376252 17.048570 −11.960431 0.842023

Table 5
Operating points chosen for groups by respective group controllers for different sets of predictions of demanded active power for loads and
weather conditions for non-controllable microsources; the maximization of profts chosen as an objective function
Controllable loads

Controllable loads

Controllable loads
max active power
Non-controllable

Non-controllable

Non-controllable

Energy storages
loads requested

max requested

function value
min requested

microsources

microsources

microsources
active power

active power

active power

active power

active power

active power

active power
Controllable

Objective

Profits

Standard 50.450933 6.400000 9.600000 17.413000 7.517190 0.000000 17.413000 −20.000000 2.787106
Alternative 1 52.513632 6.157218 9.354789 18.884147 8.694338 0.000000 18.884147 −20.000000 2.983249
Alternative 2 52.203232 7.305418 9.759093 16.197076 8.882640 0.000000 16.197076 −20.000000 2.648194
Alternative 3 52.803455 6.691778 11.214232 17.048570 8.483295 0.000000 17.048570 −20.000000 2.757713

674
14 Bull. Pol.
Bull. Pol.Ac.:
Ac.:Tech.
Tech.68(4)
68(4) 2020
2020
 Effective approach to distributed optimal operation control in rural low voltage microgrids
Effective approach to distributed optimal operation control in rural low voltage microgrids
Effective approach to distributed optimal operation control in rural low voltage microgrids
sample microgrid becomes Effective
Effective approach
disconnected.
approach to to distributed
We optimal operation
are particularly
distributed optimal operation control
controlininrural lowlowvoltage
Maximization
rural of profits microgrids
- short circuit, group controllers
sample microgrid becomes disconnected. We are particularly Maximization ofvoltage
profits - short microgrids
circuit, group controllers
interested here in the situation when such a short-circuit event 3
interested
happensmicrogrid here in the
between the consecutive
situation when such aof
iterations short-circuit event
sample becomes disconnected. We theareoptimization
particularly 3 Maximization of profits - short circuit, group controllers
happens
sample
process making between
microgrid the
usethe consecutive
becomes
ofsituation
our distributed iterations
disconnected. algorithm. of
We the
are optimization
particularly
If such a situa-
2
Maximization of
of profits
profits -- short
short circuit,
circuit, group
group controllers
controllers
interested here in when such a short-circuit event 2 Maximization
process
interested making
tion takesbetween here use
in
place beforetheof our distributed
situation when
or after the iterations algorithm.
such a If such
short-circuit
start of theofalgorithm, a situa-
event
there is
3
1
happens the consecutive the optimization 3
1
tion
happens
nothing takes place
between
new before
in suchtheof or after the
a consecutive
scenario. start ofequivalent
It isiterations
fully theofalgorithm, there is
theIf optimization
to perform-
3
2
0
process making use our distributed algorithm. such a situa- 1 2 3 4 5 6 7 8 9 10

value
nothing
process new
making
ing optimization in such
use a
of scenario.
our
calculations It
distributedis fully equivalent
algorithm. If to
such perform-
a situa-
2
0
2
1
tion takes place before or after for the another
start of the microgrid
algorithm, of slightly
there is 1 2 3 4 5 6 7 8 9 10

value
-1

function
ing
tion optimization
takes
differentnew place
topology calculations
before or after for
the another
start of microgrid
the algorithm, of slightly
there is
1
-1
1
nothing in suchand differentIt choice
a scenario. is fully of devices.toIt perform-
equivalent is much 0

valuefunction
-2
1 2 3 4 5 6 7 8 9 10

value
different
nothing
moreoptimizationtopology
new in such
interesting to and
see different
a scenario.
how the choice
fully of
Italgorithm
is devices.to
equivalent
behaves It perform-
when is much
such
0
-2 Controllable loads group controller

Objective
ing calculations for another microgrid of slightly 0
-1
-3 1 2 3 4 5 6 Controllable
7 loads group
8 controller 9 10

function
1 2 3 4 5 6 7 8 9 10

value
more interesting
change to see how the algorithm inbehaves when such

Objective
ing
a serious
differentoptimization
topology calculations
ofandtopology
different for
takesanother
choiceplaceofmicrogridthe middle
devices. of isslightly
It of our
much -1
-3
-1
-2 Non-controllable microsources group controller

function
function
-4
acalculations
serious topology
different change of scenario
topology the takes placeof indevices.
theafter
middle of our Non-controllable
loadsmicrosources group controller
toandsee different choice Ittheis 5.
much Controllable group controller
more interesting (in our change
how the algorithm starts
behaves when and
such

Objective
-2
-4
-2
-3 Energy storages group controller
calculations
more 6.(in
interesting our
to scenario the algorithm
changeand starts after till
thethe 5.such
and -5 Controllable loads
loads group
group controller
controller
of see ofhowour the inbehaves when

Objective
Controllable
abefore
seriousthechangeiteration topology algorithm
takes place lasts
the upmiddle of end
our Energy storages group controller

Objective
-3
-5 Non-controllable microsources group controller
-3

abefore the 6. iteration of our algorithm and lasts up till the end -4
-6
serious change
of the calculations).
calculations of topology takes place in
(in our scenario the change starts after the 5. and the middle of our -4
-6
Non-controllable microsources
Non-controllable microsources group
Energy storages group controller
group controller
controller

of the calculations).
calculations (in our scenario the change starts after the 5. and
-4
-5
-7
before the26,
In Fig. 6. 27,
iteration
28, and of 29ourwe algorithm
can see how and lasts up till
the value of the
the end
ob- -5
-7
Iterations Energy storages group controller
Energy storages group controller

of In
theFig.
before
jective the26,
6. 27, 28, and
iteration
calculations).
function changes 29
ourwe
of both can the
algorithm
from see perspective
how the value
and lasts upoftill of the
the the ob-
end
whole
-5
-6 Iterations

jective
of function changes both from the perspective of the whole Fig.
theFig.
calculations).
-6 28. The illustration of the process of optimizing the objective func-
-6
-7
In 26, 27, 28, and 29 we can see how the value of the ob- Fig. tion
-7
28. The
value byillustration of the controllers.
different group process of optimizing
Iterations
All group the objectiveshown
controllers func-
In Fig. 26,
jective function 27, 28, ofand
changes
Minimization active29
bothwe
power can
from
losses see
the
- short how the
perspective
circuit, group valueof
controllers of
the the ob-
whole tion
-7
value by
separately. different group
Maximization of controllers.
profits
Iterations
chosen
Iterations Allasgroup an controllers
objective shown
function.
Minimization of active power losses - short circuit, group controllers Fig. 28. The illustration of the process of optimizing the objective func-
jective function changes both from the perspective of the whole separately. Fig.
tion 28. The
Maximization
The comparison
illustration of
offorprofits chosen in
short-circuit asisland
an objective
mode function.
1.6
Fig. value
28. Theby different
illustration of the
groupthe process
controllers.
process of
of optimizing
All group
optimizing the objective
controllers
the objectiveshown func-
func-
The comparison for short-circuit in island mode
1.6 Minimization of active power losses - short circuit, group controllers
tion
tion value
value by
separately. different
different group
Maximization
by controllers.
of profits
group All
Allasgroup
chosen
controllers. controllers
an objective
group shown
function.
controllers shown
1.4 Minimization of
of active
active power
power losses
losses -- short
short circuit,
circuit, group
group controllers
controllers
1.4
1.6
Minimization
Controllable loads group controller
separately. The
separately. Maximization offorprofits
comparisonof
Maximization profits chosen in
short-circuit
chosen asisland
as an objective
an objective function.
mode function.
1.2
1.6
1.6
Controllable loads group controller The comparison
The comparison for
for short-circuit
short-circuit in
in island
island mode
mode
1.2
1.4 Non-controllable microsources group Maximization of profits - short circuit, consecutive iterations
value

controller
Non-controllable Maximization of profits - short circuit, consecutive iterations
1
1.4 Controllable loadsmicrosources group
group controller
value

1.4 controller
Energy storages group controller 3
function

1
1.2 Controllable
Controllable loads
loads group
group controller
controller
Energy storages
Non-controllablegroup controller
microsources group 3 Maximization of profits - short circuit, consecutive iterations
valuefunction

0.8
1.2
1.2
value

controller 2
0.8
1 Non-controllable microsources
microsources group
group Maximization of
Maximization of profits
profits -- short
short circuit,
circuit, consecutive
consecutive iterations
iterations
Objective

Non-controllable 2
Energy storages group controller
controller 3
function
value

controller
Objective

0.6
1 1
1
0.6
0.8 Energy storages
Energy storages group controller
group controller 1
3
function

3
function

2
0
Objective

0.4
0.8
value

0.8 0
2 1 5 9 13 17 21 25 29 33 37
0.4
0.6 2
1
value
Objective

-1 1 5 9 13 17 21 25 29 33 37
Objective

function

0.2
0.6 -1
1
0.6 1
0
valuefunction

0.2
0.4 -2
value

1 5 9 13 17 21 25 29 33 37
-2
0
Objective

0
0.4 0
-1
0.4 -3 1 5 9 13 17 21 25 29 33 37
function
value

1 5 9 13 17 21 25 29 33 37
Objective

0
0.2 1 2 3 4 5 6 7 8 9 10
-3
-1
1 2 3 4 5 Iterations 6 7 8 9 10 -1
-2
function

-4
function

0.2
0.2 Iterations
Objective

0 -4
-2
-2
Fig.0 26.1The illustration
2 3 of the
4 process
5 of 6optimizing
7 the
8 objective
9 func-
10
-3
-5
Objective

Fig.
tion 26. The
byillustration of the process of 66optimizing the objective func-
Objective

0 Iterations -5
value
1
1 different
2
2 3
3 group4
4 controllers.
5
5 All group
7
7 controllers
8
8 9
9 shown
10
10
-3
-3
-4
-6
tion value by
separately. different group
Minimization of controllers.
active
Iterations All group controllers shown
power
Iterations
losses chosen as an objective -6
Fig. 26. The illustration of the process of optimizing the objective func- -4
-4
-5
separately. Minimization
function. of active
The comparison forpower losses chosen
short-circuit as an
in island modeobjective -7
Executions of algorithm
Fig.
Fig. 26.
tion The
value
26. byillustration
The different
illustration of
of the
group process
the controllers.
process of
of optimizing
All group
optimizing the objective
controllers
the objective func-
shown
mode func-
-7
-5
-5
function. The comparison for short-circuit in island -6 Executions of algorithm
tion value
tion value by
separately. by different group
Minimization
different group controllers.
of active power losses
controllers. All group
All group controllers
chosen shown
as an objective
controllers shown Fig.
-6
-6 29. The illustration of the process of optimizing the objective
-7
separately. Minimization
function.
separately. of active
The comparison
Minimization of active power
forpower losses chosen
short-circuit
losses chosen as an
in island
as an
modeobjective
objective Fig. 29. value
function The illustration
by differentofgroup
the process
Executions of optimizing
of algorithm
controllers. The effortsthe
of objective
all group
-7
function.
function. The
The comparison
comparison for for short-circuit
short-circuit in in island
island mode
mode function
-7
controllersvalue by
shown different group controllers.
Executions
togetherofwith
of algorithm
algorithm The efforts of allofgroup
the aprocess
single curve. Maximization prof-
Executions of
Fig. 29. The illustration of optimizing the objective
Minimization of active power losses - short circuit, consecutive iterations controllers
its chosen shown
as an togetherfunction.
with a single curve. Maximization inofisland
prof-
Minimization of active power losses - short circuit, consecutive iterations Fig. 29. value
function
Fig. 29. The
The byobjective
illustration ofgroup
differentof
illustration the The case
the process
process
controllers. of of short-circuit
of optimizing
optimizing
The the
effortsthe objective
of objective
all group
its chosen as an objective function. mode The case of short-circuit in island
1.6 function value
controllers
function value by different
shown
by different groupa single
together with
group controllers.
controllers. The
curve.The efforts of
of all
Maximization
efforts allofgroup
group
prof-
1.6 Minimization of active power losses - short circuit, consecutive iterations mode
controllers
its chosen as
controllers shown together
an objective
shown with
with aa single
togetherfunction. The case
single curve.
curve.ofMaximization
short-circuit inof
Maximization prof-
ofisland
prof-
1.4 Minimization of
Minimization of active
active power
power losses
losses -- short
short circuit,
circuit, consecutive
consecutive iterations
iterations
its
its chosen
chosen as
as an
an objective
objective function.
mode
function. The
The case
case of
of short-circuit
short-circuit in
in island
island
1.4
1.6
microgrid and particular group controllers. We conducted sep-
1.2
1.6 microgrid and particular group mode
mode controllers. We conducted sep-
1.6
1.2
1.4 microgrid and particular
arate experiments for both group controllers.
objective We conducted
functions: the minimiza- sepa-
arate ofexperiments forforboth
both objective functions: the
of minimiza-
value

1
1.4 rate
tion experiments
microgrid active
andpower
particularlosses objective
and
group the functions:
maximization
controllers. Wetheconducted
minimization
profits. In
sep-
value

1.4
tion of active power losses theand thepower
maximization of profits. In
function

1
1.2
microgrid
of active
case
arate of the and
power particular
losses
minimization
experiments forand
both group
of active controllers.
maximization
objective Weprofits.
of
losses
functions: conducted
wethecan Insee
case
minimiza-sep-
of
that
valuefunction

0.8
1.2
case of the minimization of active power losses we can see that
value

1.2
0.8
1 arate
the
tion experiments
afterminimization
ofthe short-circuit
active power offor both
active
event
losses objective
power
thethe
and losses
algorithmfunctions:
we can
maximization seethe
converges that
of minimiza-
after
to some
profits. the
In
Objective

after the short-circuit event the algorithm converges to some

function
value

tion of active power losses and the maximization of profits. In

Objective

0.6
1
1
0.6
short-circuit
(sub)optimal
case event
solution
of the minimization the algorithm
which is converges
of active better
power than tothe
lossessome one
we (sub)optimal
before
can the
see that
0.8
(sub)optimal solution which is better than the one before the
function
function

case of
incident.
solution
after the
thewhich minimization
There is better
is
short-circuit nothing of
than
event active
surprising
the one power
here
before
algorithm losses
– after we can
half oftothe
the incident.
converges see that
mi-
There
some
Objective

0.4
0.8
0.8
0.4 incident.
after
crogrid There
the(upper is nothing
short-circuit
feeder) event surprising
becomes the here
algorithm
disconnected of –theafter half oftothe
converges
also mi-
some
is nothing surprising here – after half onea half
microgrid of
(upper
0.6
(sub)optimal solution which is better than the before the
Objective
Objective

0.2
0.6 crogrid (upper
(sub)optimal
loads cannot feeder)
solution
be becomes
which disconnected
isalso
better than also
the loads
one a half
before of the
the
feeder) becomes issupplied
disconnected with electricity
a half –anymore
of the half(further
cannot we
be
0.6
0.2
0.4 incident. There nothing surprising here after of the mi-
loads
incident.
care onlycannot
There be supplied
isthenothing
aboutelectricityhalfbecomeswith
ofanymore electricity
surprising
the grid –here
lower anymore
– we
after
feederhalf (further
of the we
mi-
0
0.4
0.4 supplied
crogrid with
(upper feeder) (further
disconnected care
also ainonly
which
half about
of no
the
0
0.2 1 5 9 13 17 21 25 29 33 37
care only
crogrid
the half ofabout
(upper
short-circuits thebe the half
feeder)
happened),
grid ofsothe
becomes
– lower thegrid
feeder –inlower
disconnected
total amount
which feeder
also ain which
half
of short-circuits
no energy of no
to the
be
1 5 9 13 Executions
17 of21algorithm 25 29 33 37 loads cannot supplied with electricity anymore (further we
0.2
0.2 Executions of algorithm short-circuits
loads
deliveredcannot
happened), is happened),
be
lower
so thethesupplied
and
total the
amount so
withthe
total sumtotal
electricity
of amountanymore
active of
powerenergy
(further
losses to be
we
be-
Fig.0 127. The illustration 13 of the
0
17 process of 25optimizing the objective care only about half of the of energy
grid – lower to befeeder
delivered is lower
in which no
Fig.
5 9
0 27. The illustration of the process
21 29
of 25optimizing
33 37
delivered
care only is lower
about the and
of half the
In of total
thethe sum
ofgrid of active
–maximization
lower power
feeder losses
in profits
which be-
no
function
1 value
5 by 99different
13
Executions
group
17
of algorithm
controllers.
21 effortsthe
25 The 29
29
objective
33of all 37
group
37 comes
and the lower
total
short-circuits as
sum well. active
happened), casepower
so the
losses
total becomes
amount of
of lower
energy as to all
well.
be
1
function
5
value by together
13
different with
17
group
21
controllers. The
33
efforts of all group comes
the
In grouplower
short-circuits as well.
happened),
controllers
case of isthelowermaximizationIn
are case
so
also of
theablethe
total
of profits tomaximization
amount
agree on of
some of
energyprofits
new to
com-all
be
controllers
Fig. shown
27. The illustration
Executions
of theaprocess
single
Executions of algorithm
of algorithm
curve. Minimization
of optimizing of active
the objective delivered and the total sum ofall the group
active powercontrollers
losses be-
controllers shown together with a single curve. Minimization of active the group
delivered
mon
are controllers
is lower
(sub)optimal
alsolower
able to and are
the
solution also
total able
after sum to
the ofagree
activeon
short-circuit some
power new
losses
takes com-
be-
place.
power27.
Fig.
Fig. losses
27.
function The chosen
value
The as an of
illustration
by different
illustration objective
of the process
group
the function.
process
controllers. The
Thecase
of optimizing
of optimizingofthe
efforts short-circuit
objective
of objective
all group comes asagree
well. on someofnew
In case thecommon
maximization (sub)optimal
of profits solu-
all
power losses chosen as an objective function. The case ofthe
short-circuit mon
comes
Here (sub)optimal
thelower
new as well.
solutionsolution
Inis case after
worse, of the
butthemaximization
short-circuit oftakes
profitsplace.
all
function value
controllers
function value by together
shown
by in
different with
different island
group
group mode
controllers.
a single
controllers.curve.TheThe efforts of
Minimization
efforts of all
all group
of active
group tion after
the group the short-circuit
controllers are alsotakesable toit agree
place. could
Hereon bethe
better
somenew as
new well
solution
com- –
in island mode Here
the
is the
group
worse, new solution
controllers
but it could is
are worse,
also
be better ablebut
as well toit could
agree onbe better
some
– everything takes as
new
depends well
com-on–
controllers
power losses
controllers shown
chosen
shown together with
with aa single
as an objective
together function.
single curve.
curve.The Minimization of
of active
case of short-circuit
Minimization active mon (sub)optimal solution after the short-circuit place.
power
power losses
losses chosen
chosen as as an
an objective
in islandfunction.
objective mode The
function. The case
case of
of short-circuit
short-circuit mon
the (sub)optimal
Hereactual
the new topology
solutionsolution
and theafter
is worse, proportion
buttheit short-circuit
of the
could losttakes
be better place.
generation
as well –
Bull. Pol. Ac.: Tech. 68(4) 2020
in island
in island mode mode Here the new solution is worse, but it
capabilities from the renewable energy sources in comparison could be better as well 15–
Bull. Pol. Ac.: Tech. 68(4) 2020 15
Bull. Pol. Ac.: Tech. 68(4) 2020 15
Bull. Pol. Ac.:
Pol.
Bull. Pol. Ac.:Tech.
Ac.: Tech.68(4)
Tech. 68(4)2020
68(4) 2020
2020 675
15
 M.
M.Parol,
Parol, P.
P. Kapler,
Kapler, J.
J. Marzecki,
Marzecki, R.
R. Parol, M. Połecki,
Parol, M. Połecki, and
and Ł.
Ł. Rokicki
Rokicki

to the total amount

everything dependsofonenergy the actualthat should
topology be and
received by all the
the proportion
disconnected
of the lost generation loads, butcapabilities
after the incident
from the does not needenergy
renewable to be
delivered. We assume to
sources in comparison here,
the that
total such
amount a random
of energy unpredictable
that should
event does not
be received bylead to any
all the obligation of
disconnected paying
loads, butanyafterpenalties
the inci- to
the
dentenddoes consumers.
not need When we take a look
to be delivered. We assumeat the operating
here, thatpoints
such
chosen
a random by unpredictable
particular group controllers
event does not(figures lead tonot anypresent here
obligation
because
of payingofany the penalties
limited space) to the we endcan see that in
consumers. case we
When of both
take
objective
a look at functions
the operating rightpoints
from the start up
chosen by till the end group
particular all the con-
trollable loads work
trollers (figures not at their minimum
present here because consumption
of the limitedcapabilities,
space)
while
we canallsee thethatcurrently
in case available non-controllable
of both objective functionsenergy sources
right from the
are
startturned
up tillon. theInend case
allof thethe minimization
controllable loads of active
work at power
theirlosses
mini-
it
mumresults from the fact
consumption that we need
capabilities, whiletoall minimize
the currentlythe amount
available of
energy being conveyed
non-controllable energyand to promote
sources are turned the local
on. In generation
case of the of
energy.
minimizationIn caseofofactive profits it results
power losses from the fact
it results thatthe
from thefact
energy
that
generated
we need tobyminimize the reciprocating
the amount engine
of energyis the most
beingexpensive
conveyedone, and
so we need to Fig. 31. The illustration of the process of optimizing the objective
to promote thedecrease the total consumption
local generation of energy. Inofcase energy as much
of profits it
function value by different group controllers. The efforts of all group
as possible
results fromand alsothat
the fact try the
to make
energyuse of cheaper
generated by theenergy (from
reciprocat-
controllers shown together with a single curve. Minimization of active
renewable
ing engine energy is the most sources) first. one,
expensive The missing
so we need parttoofdecrease
the energy the power losses chosen as an objective function. The case of lost PVs in
will
total be always provided
consumption of energy by both
as much the as reciprocating
possible andengine also tryand to island mode
the
make energy
use of storage.
cheaper In energy
case of (from
the minimization
renewable energy of active power
sources)
losses
first. Theit does
missing not matter
part of which
the energy of the will devices
be alwayswill provided
produce the by
missing
both the energy (only the
reciprocating distance
engine and between
the energy thestorage.
generation and
In case
the
of theconsumption
minimization matters),
of active while powerin case losses of itthedoes
maximization
not matter
of profits
which of the thedevices
energy will storage is always
produce prioritized
the missing energy(works
(only at the
its
maximum
distance between both before and after the
the generation and incident) as the cheaper
the consumption one.
matters),
Theinsecond
while case of experiment considersofa situation
the maximization profits thewhen energybecause
storageof
some
is alwaysrandom unpredictable
prioritized (workssituation
at its maximum we lose some both big fraction
before and
of devices
after of the same
the incident) as thetype (loads
cheaper one. or microsources). Here we
analyze a scenario
The second when during
experiment considersthe execution
a situationof the distributed
when because of
algorithm
some random (more interestingsituation
unpredictable dynamicwe scenario)
lose some clouds appear
big fraction
on the sky of
of devices andthe allsame
the photovoltaic
type (loads panels stop to produce
or microsources). Here anywe
energy
analyze(between
a scenariothe wheniterations
duringno. the 4execution
and no. 7ofinclusively).
the distributed In
Figs. 30, 31,(more
algorithm 32, and 33 we candynamic
interesting see how the value ofclouds
scenario) the objective
appear
function
on the sky behaves.
and all For the both objectivepanels
photovoltaic functionsstop first we always
to produce any
find
energy some (sub)optimal
(between solution,
the iterations no.then4 and during
no. 7 the incident we
inclusively). In Fig. 32. The illustration of the process of optimizing the objective func-
switch
Fig. 30,to 31,some32, and other,
33 we always
can see worse
how one and then
the value of thewe switch
objective tion value by different group controllers. All group controllers shown
back
functionto the original
behaves. For solution. The solution
both objective functions duringfirstthe
weincident
always separately. Maximization of profits chosen as an objective function.
The comparison for lost PVs in island mode
Minimization of active power losses - lost PVs, group controllers

1.6

1.4

1.2
Objective function value

0.8

0.6
Controllable loads group controller
0.4
Non-controllable microsources group controller

0.2
Energy storages group controller

0
1 2 3 4 5 6 7 8 9 10
Iterations

Fig. 30. The illustration of the process of optimizing the objective func- Fig. 33. The illustration of the process of optimizing the objective func-
tion value by different group controllers. All group controllers shown tion value by different group controllers. The efforts of all group con-
separately. Minimization of active power losses chosen as an objective trollers shown together with a single curve. Maximization of profits
function. The comparison for lost PVs in island mode chosen as an objective function. The case of lost PVs in island mode

676
16 Bull. Pol.
Bull. Pol.Ac.:
Ac.:Tech.
Tech.68(4)
68(4) 2020
2020
 Effective approach to distributed optimal operation control in rural low voltage microgrids
is always worse because there is no possibility to make use of
some cheap energy from the local microsources situated close
to the places of the consumption. The analysis of the operating
points chosen by particular group controllers (also not shown
here for the sake of limited space) exposes generally the same
behavior that could be observed in the previous experiment.
7. Summary and conclusion
In the paper we have presented a distributed method of deter-
mination of (sub)optimal operating points for devices compos-
ing a microgrid. The algorithm itself is rather not complicated
– applied logic, similar to the ones known from Monte Carlo
methods, is much simpler than in many other more sophisticated
centralized approaches. The results obtained for our decentral-
ized algorithm have been compared to the ones produced by
a centralized method. We can see that in all cases the objective
function values are exactly the same. We have analyzed the func-
tioning of the decentralized algorithm in detail. Apart from the
standard behavior we have tested different additional scenarios:
a variant when the communication between group controllers
is lost, a variant when the predictions of both the demands for
active power and the solar and wind generation are not precise
and a variant when the order in which group controllers execute
the algorithm is chosen in a totally random way.
We have described and presented typical load and generation
profiles and their characteristics. The knowledge of detailed daily
profiles of all loads composing a microgrid together with daily
profiles of wind and solar energy which is expected to be gen-
erated by different types of renewable energy sources which are
possible to be installed can help to choose the right set of devices
and design a microgrid in such a way, that its exploitation can
be more effective and cheaper. If exact profiles are not known
in advance, some typical profiles known from the literature can
be used instead. The knowledge of load and generation profiles
also helps in the task of predicting both the demands for active
power and the generation capabilities of microsources for a spe-
cific future point in time. It is needed for many algorithms to
work properly, especially when the set of operating points is
being calculated in advance for a future optimization time period.
Having precise predictions on the total demanded active
power and the active power which can be generated from the
different renewable energy sources for sure helps in determina-
tion of the set of (sub)optimal operating points for every single
existing microgrid. However, in case of our distributed algo-
rithm we have proved that even if all predictions are not very
precise and differ from the true (accurate) values by around
20%, the set of calculated operating points remains more or
less the same. It shows that our algorithm is not very sensitive
to the precision of the predictions.
Our studies on the potential loss of communication between
the different group controllers show that establishing reliable
communication between the entities is necessary to fully opti-
mize the value of the chosen objective function. However, even
without the possibility to exchange messages between one of
the group controllers and all others, the process of optimiz-
Bull. Pol. Ac.: Tech. 68(4) 2020
ing the objective function value can still take place, at least to
some point. When the communication is back, the distributed
algorithm unlocks the optimization potential which could not
be utilized during the communication loss period. Knowing the
difference between the objective function values which can be
achieved with and without the properly working communication
network, the designer of the microgrid can choose if it is more
beneficial to invest heavily in the reliable communication infra-
structure or to accept that during some communication network
failures the algorithm will choose a little worse set of operating
points.
In our opinion, the cooperative control is more robust than the
centralized one in case of communication faults. The proposed
repair algorithms allow for the continuation of the optimization
process while different layers of control structure are affected.
We have also showed that the choice of the order in which
different group controllers make their calculations can be chosen
randomly in case of our distributed algorithm. The final value
of the objective function will still be more or less the same.
However, in case of some orders the optimization process will
be complete faster. Proper synchronization of executions of the
algorithm on different entities can speed up the optimization
process if time constraints are very strict.
Two additional, carried out experiments (the simulation of
a short-circuit in one feeder of the test microgrid and the sim-
ulation of a loss of power generated by all photovoltaic pan-
els in this microgrid) proved the reliability and stability of our
algorithm.
In the future, a further research should be mostly focused on
some further and more detailed investigation of the communica-
tion problems of different types.We have analyzed the situation
of communication loss, but for example we have not taken into
account the issue of communication delays (for example the
situation when operating points from other group controllers are
known, but always only from the previous iteration, not from
the current one). Also, our current approach relies strongly on
the assumption that the values of all power factors are constant,
so the optimization problem becomes a one-dimensional one
(only the optimization of active power is needed). The natu-
ral extension of our method will be partial or total giving up
of this assumption and finding out how to generate candidate
solutions with different and desirable values of reactive power
in a smart way.
In our current approach we did not consider the problem of
many separate entities (owners of generating units and energy
storage units) with different goals. This issue can be an inter-
esting subject of any further research related to the problem of
optimization of microgrid operation in rural areas.
Acknowledgements. The paper was partially financed from the
funds earmarked for the RIGRID (Rural Intelligent Grid) proj-
ect realization. The authors gratefully acknowledge funding of
this research from the ERA-Net Smart Grid Plus initiative, with
support from the European Union’s Horizon 2020 research and
innovation program. Printing the paper was financed from the
funds of the Warsaw University of Technology earmarked for
science and research.
677
 M. Parol, P. Kapler, J. Marzecki, R. Parol, M. Połecki, and Ł. Rokicki
R eferences
[1] J. Kiciński, “Do we have a chance for small-scale energy gener-
ation? The examples of technologies and devices for distributed
energy systems in micro & small scale in Poland”, Bull. Pol. Ac.:
Tech. 61(4), 749–756 (2013).
[2] I. Wasiak and Z. Hanzelka, “Integration of distributed energy
sources with electrical power grid”, Bull. Pol. Ac.: Tech. 57(4),
297–309 (2009).
[3] G. Benysek, M.P. Kazmierkowski, J. Popczyk, and R. Strzelecki,
“Power electronic systems as a crucial part of Smart Grid infra-
structure – a survey”, Bull. Pol. Ac.: Tech. 59(4), 455–473 (2011).
[4] CIGRÉ Working Group C6.22 Microgrids evolution roadmap,
Microgrids 1: Engineering, economics & experience, TB 635,
2015.
[5] Ch. Marnay, S. Chatzivasileiadis, Ch. Abbey, R. Iravani, G. Joos,
P. Lombardi, P. Mancarella, and J. von Appen, “Microgrid evolu-
tion roadmap. Engineering, economics, and experience”, Paper
presented at the 2015 Int. Symp. on Smart Electric Distribution
Systems and Technologies (EDST15), CIGRE SC C6 Colloquium,
Austria, 2015.
[6] https://building-microgrid.lbl.gov/about-microgrids. Accessed 05
Feb 2020.
[7] Microgrids: architectures and control, ed. N.D. Hatziargyriou,
IEEE Press, Wiley, 2014.
[8] Low voltage microgrids. Joint publication ed. M. Parol, Publish-
ing House of the Warsaw University of Technology, Warsaw,
2013, [in Polish].
[9] F. Katiraei, R. Iravani, N.D. Hatziargyriou, and A.L. Dimeas,
“Microgrids management. Control and operation aspects of
microgrids”, IEEE Power & Energy Mag. 6(3), 54–65 (2008).
[10] F. Katiraei and R. Iravani, “Power management strategies for
a microgrid with multiple distributed generation units”, IEEE
Trans. Power Syst. 21 (4), 1821–1831 (2006).
[11] A.L. Dimeas and N.D. Hatziargyriou, “Operation of a multiagent
system for microgrid control”, IEEE Trans. Power Syst. 20(3),
1447–1455 (2005).
[12] Y. Li and F. Nejabatkhah, “Overview of control, integration and
energy management of microgrid”, J. Mod. Power Syst. Clean
Energy, 2(3), 212–222, doi:10.1007/s40565-014-0063-1 (2014).
[13] D.E. Olivares, C.A. Cañizares, and M. Kazerani, “A Centralized
Energy Management System for Isolated Microgrids”, IEEE
Trans. Smart Grid 5(4), 1864–1875 (2014).
[14] M. Parol, Ł. Rokicki, and R. Parol, “Optimal operation control in
low voltage microgrids in rural areas functioning on the basis of
centralized control logic”, Przegląd Elektrotechniczny (Electrical
Rev.) 94(3), 134–138, doi:10.15199/48.2018.03.26 (2018).
[15] Y. Han, K. Zhang, H. Li, E.A. Alves Coelho, and J.M. Guerrero,
“MAS-based distributed coordinated control and optimization in
microgrid and microgrid clusters: a comprehensive overview”,
IEEE Trans. Power Electron. 33(8), 6488–6508 (2018).
[16] T. Morstyn, B. Hredzak, and V.G. Agelidis, “Control strategies
for microgrids with distributed energy storage systems: an over-
view”, IEEE Trans. Smart Grid 9(4), 3652–3666 (2018).
[17] Q. Li, C. Peng, M. Chen, F. Chen, W. Kang, J.M. Guerrero,
and D. Abbott, “Networked and distributed control method with
optimal power dispatch for islanded microgrids”, IEEE Trans.
Ind. Electron. 64(1), 493–504 (2017).
[18] R. Parol, M. Parol, and Ł. Rokicki, “Implementation issues
concerning optimal operation control algorithms in low voltage
microgrids”, in Pendrive Proc. of the 5th Int. Symp. on Electrical
and Electronics Engineering, Romania, 2017, pp. 7
678
R eferences
[19] V. Kumar, N. Leonard, N.E. Leonard, and A.S. Morse, Cooper-
ative control. Springer, Berlin/Heidelberg, 2005.
[20] Z. Qu, Cooperative control of dynamical systems. Springer, Lon-
don U. K. 2009.
[21] Cooperative Control of Distributed Multi-Agent Systems. Ed. J.S.
Shamma, Wiley, 2007.
[22] M. Wooldridge, An Introduction to Multi Agent Systems. John
Wiley& Sons, 2002.
[23] J.P. Bigus and J. Bigus, Constructing Intelligent Agents Using
Java. 2nd Edition. John Wiley & Sons, 2001.
[24] S. Russell and P. Norvig, Artificial Intelligence. A Modern
Approach. The Intelligent Agent Book. Second Edition. Prentice
Hall, 2003.
[25] Yinliang Xu, Wei Zhang, Wenxin Liu, Xin Wang, Frank Ferrese,
Chuanzhi Zang, and Haibin Yu, “Distributed Subgradient-Based
Coordination of Multiple Renewable Generators in a Microgrid”,
IEEE Trans. Power Syst. 29(1), 23–33 (2014).
[26] Huanhai Xin, Zhihua Qu, John Seuss, and Ali Maknouninejad,
“A Self-Organizing Strategy for Power Flow Control of Photovol-
taic Generators in a Distribution Network”, IEEE Trans. Power
Syst. 26(3), 1462–1473 (2011).
[27] A. Bidram, F.L. Lewis, and A. Davoudi, “Distributed control
systems for small-scale power networks using multiagent cooper-
ative control theory”, IEEE Control Systems Magazine 34, 56–77
(2014).
[28] A. Bidram, A. Davoudi, F.L. Lewis, and J.M. Guerrero, “Distrib-
uted cooperative secondary control of microgrids using feedback
linearization”, IEEE Trans. Power Syst. 28(3), 3462–3470 (2013).
[29] R. Jalizade Hamidi, H. Livani, S.H. Hesseinian, and G.B. Ghare-
hpetian, “Distributed cooperative control system for smart micro-
grids”, Electr. Pow. Syst. Res. 130, 241–250 (2016).
[30] J.-Y. Kim, J.-H. Jeon, S.-K. Kim, C. Cho, J. Ho Park, H. Kim,
and K.-Y. Nam, “Cooperative control strategy of energy storage
system and microsources for stabilizing the microgrid during
islanded operation”, IEEE Trans. Power Electron. 25(12), 3037–
3048 (2018).
[31] M. Parol, L. Rokicki, and R. Parol, “Towards optimal operation
control in rural low voltage microgrids”, Bull. Pol. Ac.: Tech.
67(4), 799–812 (2019).
[32] Statistical Yearbook of the Republic of Poland, Central Statistical
Office, (2018).
[33] J. Godlewska, Guide for good practices energy mangement on
the agricultural farm, Białystok School of Economics, (2011), [in
Polish].
[34] A. Grzybek, M. Ćwil, Z. Ginalski, D. Raczkiewicz, and J. Sta-
rościk, Effective management electricity and heat on the farm,
Foundation for the Development of Polish Agriculture, (2017),
[in Polish].
[35] The results of monitoring electricity consumption on farms par-
ticipating in the OZERISE project, https://ozerise.pl. Accessed
05 Feb 2020.
[36] P. Banasiak, A. Gorczyca-Goraj, and M. Przygrodzki, “Analysis
of load diagrams developed for selected segments of low voltage
consumers”, Energetics 70(1), 23–27 (2017), [in Polish].
[37] A. Curkowski, “Problems and challenges related to energy supply
in commercial fruit and vegetable farms, including those using
storage rooms and cold stores”, https://www.lodr-bratoszewice.
pl/sites/default/files/attachment/2.problemy_i_wyzwania_zwi-
azane_z_zaopatrzeniem_w_energie_w_towarowych_gospodarst-
wach_rolnych-ac-fin.pdf. Accessed 05 Feb 2020, [in Polish].
Bull. Pol. Ac.: Tech. 68(4) 2020