You are on page 1of 6

Proceedings of the 36th Chinese Control Conference

July 26-28, 2017, Dalian, China

Hybrid AC/DC Microgrid Energy Management Based on Renewable


Energy Sources Forecasting
Baolei Yuan, Alian Chen, Chunshui Du, Chenghui Zhang
School of Control Science and Engineering, Shandong University, Jinan 250061
E-mail:chenalian@sdu.edu.cn

Abstract: Energy management is the core of economical and efficient operation for hybrid AC/DC microgrid. In this paper,
renewable energy sources are forecast by using BP neural network based on modified particle swarm optimization (MPSO-BP).
Compared with traditional algorithm, the modified algorithm performs better in both search ability and speed. After forecasting,
with the objectives of minimum total cost and minimum power loss, the optimal model is built. The output is optimized through
coordinating all kinds of DGs. The simulation results verified the effectiveness of the proposed model in multi-objective
optimization of hybrid microgrid. Through the energy management dispatch, the hybrid microgrid could realize the optimized
operation level.
Key Words: Hybrid AC/DC microgrid, Renewable energy sources forecasting, Energy management

In recent years, the research on microgrid energy


1 Introduction management is only limited on AC microgrid, only few
Microgrid has popularized due to its abilities of studies can be found on hybrid AC/DC microgrid. In [5] an
integrating various distributed energy sources and improving optimal power flow structure for hybrid AC/DC microgrid is
the grid power quality and reliability. Now it has already introduced, and the DC side and AC side power flow
become the future power grid development trend. AC separately in both grid-connected mode and islanding mode
microgrid is still the mainstream structure currently, however, are analyzed. In [6] an optimal operation is proposed for
with the increasing of DC loads and DC distributed energy hybrid microgrid under spot price mechanism, compared
resources (DERs), the hybrid AC/DC microgrid widely with time of use price, it could increase revenue and reduce
spread, and become more desirable for researchers to focus operating cost. In [7] an optimal day ahead scheduling is
on[1-2]. The hybrid AC/DC microgrid combines the proposed with the aim of minimizing the total cost sustained
advantages of both AC and DC microgrids, in addition, due during the day.
to the connection way that the AC loads and DERs connect In this paper, a complete hybrid AC/DC microgrid energy
to the AC bus to grid and the DC loads and DERs connect to management is presented. First, a modified PSO-BP
the DC bus, the number of required converters and the power algorithm is proposed to forecast the renewable energy
loss would be minimized so that the volume and the cost of sources output power which is verified to be more accurate
microgrid would have a significant decreased[3-4]. than the result forecast by BP or PSO-BP algorithm. Based
For operating more efficiency and economic, energy on the forecasting result, a model for the hybrid microgrid
management of hybrid AC/DC microgrid is quite necessary. optimal dispatch is built so as to minimize the total cost
The main goal of an energy management system is keeping which includes environment cost, maintaining cost,
the balance between the power output and demand and investment cost and operation cost. What’s more, although
maintaining the system security at the minimum cost. The the hybrid AC/DC microgrid has already decreased the
whole part of energy management for a hybrid AC/DC number of converters compared to AC microgrid and DC
microgrid should include forecasting and optimal dispatch, microgrid, the converter losses still exist and adversely affect
as shown in Fig.1. the efficient operation of the system to a great extent.
Therefore, the innovation is that the converter losses include
Renewable energy
sources forecasting
both DC side and AC side and interlinking converter (IC)
Interaction with
grid which is the particular unit in hybrid microgrid structure are
Demand database taken into consideration, based on which a multi-objective
Optimization module Output of DGs optimal function is implement.
Electric price
database

ESS status
2 Renewable Energy Sources Forecasting
Other database:
Energy price
Emission As shown in Fig.1, the renewable energy sources
parameters
Ă forecasting is absolutely necessary in a hybrid microgrid
Information input layer Dispatch output layer energy management system. For the uncertainty and
Fig. 1: Microgrid energy management structure randomness of renewable energy sources like solar energy
and wind energy, how to dispatch the various micro-sources
*
This work is supported by the National Natural Science Foundation of to achieve optimal state becomes an impossible task.
China under Grant 51377101 and Major International (Regional) Joint Therefore, forecasting the renewable energy sources exactly
Research Project of the National Natural Science Foundation of China is very important[8].
(NSFC) under Grant 61320106011.

2870
2.1 Modified PSO-BP Algorithm
Start
The BP algorithm is widely used in the field of renewable
Define the BP
energy sources forecasting because of its good capability of weight as
nonlinear mapping and self-learning. Nevertheless, the particles
Initialize the
particles
gradient descent method is used in traditional BP algorithm
leading to slow convergence[9-10]. Using PSO to optimize BP Take the error as
fitness function,
could accelerate convergence, but it is prone to be in local Calculate the fitness
value Obtain the best
minimum easily. To solve the foregoing problem, a BP particle

algorithm which is optimized by modified particle swarm is


Calculate the
proposed. individual best
Calculate the
The inertial weight ɷ, which has an important influence on and global best
error
the searching strategy, is a dependent parameter in PSO. In
Update the velocity
traditional PSO algorithm, it is defined as follows: and position of No
(ω -ω ) u t each particle Error<Expected
ω ωmax - max min (1)
T Yes

where ɷmax, ɷmin is the maximal value and minimum value No Yes
Satisfy the condition Output
forecasting
respectively, t is the current iteration time, and T is the total or not?
result
iteration time.
As formula (1) shown, particle search step size decreases PSO part BP part

gradually with the increase of the time of iteration, which can


Fig. 2: MPSO-BP algorithm flow chart
make the particles avoid falling into local optimum in a
certain extent, but once the particles fall into local optimum, 2.2 Renewable Energy Sources Forecasting Curve
it is very difficult to jump out. Furthermore, the change of ɷ
Taking wind power forecasting as an example, the sample
is only related to the iteration time linearly so that it cannot
is chosen every fifteen minutes and the figure only shows the
adapt with complex nonlinear optimization problem well. To
result every hour. The result of using traditional BP, PSO-BP,
this end, a modified PSO is present as formula (2) which
MPSO-BP method respectively are shown as Fig.3, Fig.4
adjusts ɷ based on fitness value variation.
and Fig.5.
1
Zt D  e- Kt / Kt1 (2) Expected
2 Forecasted
16
1 n
Kt
ni1
¦ ( ft ( X i )  ft ( X best ))2 (3) 14

where α is a random number from 0 to 1, n is the total number


Wind Power/kW

12
of particles, f t ( X i ) is the fitness value of the ith particle in tth
10
iteration, f t ( X best ) is the best fitness value in tth iteration.
In formula (3), Kt is the variance indicates the dispersion 8

degree of fitness value of the ith particles in tth iteration. The 6


bigger Kt is, the more dispersed the particles are. When Kt >
Kt-1 , namely the tth iteration is dispersed compared with the 4
0 5 10 15 20
t-1th iteration, it will be seen from formula (2) that ɷt will be Time/h
smaller, and the search step will be smaller, too. Conversely,
the ɷt will be bigger, and the search step will be bigger. Fig. 3: Wind power forecasting result by using BP
Using e as the base can limit the ɷ from 0 to 1, and for more MAPE=0.076619
efficient search, 1/2α is added to widen the range from 0 to
Expected
1.5. Forecasted
The MPSO-BP algorithm flow chart is shown, as in Fig. 2. 16

14

12
Wind Power/kW

10

0 5 10 15 20
Time/h

Fig. 4: Wind power forecasting result by using PSO-BP


MAPE=0.054708

2871
Expected 3 Hybrid AC/DC Microgrid Energy Management
Forecasted Mathematic Formulation
16

The structure of hybrid AC/DC microgrid is shown in


14
Fig.8. As various micro-sources and kinds of optimal
objectives in microgrid, the operation optimization is a
Wind Power/kW

12
typical non-linear programming problem with complex
10
constraint conditions. Therefore, the optimization model can
8
be expressed as
min[ f1 ( x), f 2 ( x), , f n ( x)] (4)
6
s.t. ­h(x) 0
4 ® (5)
0 5 10
Time/h
15 20 25
¯gmin d g(x) d gmax
where f1 ( x), f 2 ( x), , f n ( x) are the objective functions,
Fig. 5: Wind power forecasting result by using MPSO-BP
>h1 (x), h2 (x), , hm (x)@ is the equation constraint
7
MAPE=0.04132 h(x)
function, g(x) > g1 (x), g2 (x), , gm (x)@7 is the inequation
The mean absolute percent error (MAPE) of the
forecasting model is 7.7%, 5.5%, and 4.1%, respectively. constraint function. Based on the structure, the specific
Compared with BP algorithm and PSO-BP algorithm, model is established.
MPSO-BP’s MAPE is smaller, the forecasting result is more
AC-bus DC-bus
exact. Meanwhile, the convergence of PSO and MPSO are Center controller

compared in Fig.6. The search ability of modified algorithm


PCC
is enhanced, and its convergence speed is faster. DC/DC

MPSO Photovoltaic(PV)
PSO AC/DC/AC
350 DC/DC
Wind-turbine(WT)
Battery
300
Interlinking
AC/DC/AC converter
250
Micro-turbine(MT)
Fitness Value

200

150 DC load
AC load

100 Fig. 8: Structure of hybrid AC/DC microgrid


50
3.1 Objective Function
0
0 20 40 60 80 100 120 140 The first objective, related to the minimization of the total
Evolutional Generation
cost expressed as follows:
Fig. 6: Convergence of PSO and MPSO fobj ,1 CF  Ce  CID  Cm (6)
The same method is applied on the solar power forecasting where CF is operation cost, Ce is environmental cost, CID
as in Fig.7.
is investment depreciation cost, Cm is maintaining cost.
Expected

18
Forecasted CF , Ce , CID and Cm is expressed as follows respectively:
16 1. Operation cost CF :
CFMT ( PMT (t ))  CFGRID ( PGRID (t ))
14
CF (7)
Solar Power/kW

12
Cnl P (t ) u 't
10
CFMT ( PMT (t )) u ¦ MT (8)
8 L KMT
PGRID (t ) u C price
6
CFGRID ( PGRID (t )) (9)
4

2 where ηMT is the efficiency of micro-turbine (MT), PMT is the


0
net output power, Cnl is the price of natural gas, L is
0 5 10 15 20 coefficient, take the value of 9.7kWh/m3, Cprice is the price of
Time/h real time electricity price.
Fig. 7: Solar power forecasting result by using MPSO-BP
MAPE=0.069867 2. Environmental cost Ce :
For the structure of this paper, the pollution emission
mainly comes from MT which can be calculated as follows:
m
Ce PMT (t ) u ¦ E pj u D j (10)
j 1

2872
where E pj represents the emission price for the jth emission max
where f obj max
,1 and f obj ,2 is the value of function which is not
types of MT, α is the emission coefficient. The data is shown optimized. α and β is the weight coefficients of each
in Table 1. objective function respectively, and meet the condition of
Table 1: Emission factors of pollution α+β=1. The weighting factor represents the degree of
preference for each objective function affecting the
Emission types Emission coefficient Emission price optimization results.
for MT kg/MWh yuan/kg
NOx 0.20 8.00 3.2 Constraint
SO2 0.0036 6.00 The hybrid microgrid operation constraints are classified
CO2 724 0.023
as DC side and AC side.
DC constraints:
The power balance at the DC side of the hybrid microgrid
3. Investment depreciation cost CID : is:
24
CINIi u CRF (l , r ) Pconv (t )  PPV (t )  PBA (t ) Pload _ dc (t ) (20)
CID ¦
t 1 8760 u Ci
u Pi (t ) (11)
where Pconv(t) is the power injected from AC side to DC side
through the interlinking converter.
r (l  r )l The operation of battery should satisfy the following
CRF (l , r ) (12)
(l  r )l  l constraints:
PYi 0 d PBA (t ) d Pmax
disch
(t ) (21)
Ci u PDi (13)
8760 ch
Pmax (t ) d PBA (t ) d 0 (22)
where CINIi is the ith micro-source initial investment cost of
SOCmin (t ) d SOC(t ) d SOCmax (t ) (23)
unit capacity, Ci is the coefficient of capacity, PYi is the
ch
annual output power, and PDi is the rated capacity. For where P (t ) is the maximum value of battery when it is
max

calculation of capital recovery factor CRF(l,r), r represents disch


charging, and Pmax (t ) is the maximum value when it is
rate, and l represents the service life of power supply. discharging. SOC represents the state of charge.
4. Maintaining cost Cm : For the other DGs of DC side, the constraints are as
Cm Kmi u Pi (t ) (14) followings:
DC
Pi min (t ) d P DC (t ) d Pi max
DC
(t ) (24)
where K mi is the maintaining coefficient for ith i

micro-source. R DC
idown u 't d Pi (t )  Pi DC (t  1) d Riup
DC DC
u 't (25)
The second objective is aimed at improving the energy where R DC
and R DC
is the rate of increase or decrease in
idown iup
efficiency of the microgrid by minimizing power losses of
converters: 't .
AC constraints:
fobj ,2 dc / ac
Ploss ( Pi (t ))  Ploss
ac / dc
( Pi (t ))  Ploss
dc / ac
( Pi (t )) (15) The power balance at the AC side of the hybrid microgrid
is:
Pi (t ) PWT (t )  PMT (t )  PGRID (t ) Pload _ ac (t )  Pconv (t ) (26)
Ploss ( Pi (t )) u (1  Kconv ) (16)
Kconv For the other DGs of AC side, the constraints are as
where ηconv is the efficiency of converter. The efficiency of followings:
AC/DC, DC/AC converter and IC can be expressed in the AC
Pi min (t ) d Pi AC (t ) d Pi max
AC
(t ) (27)
same equation given by (17)[11]:
u
AC
Ridown u 't d Pi AC (t )  Pi AC (t  1) d Riup
AC
u 't (28)
K1 u100% (17)
0.004  1.002u  0.018u 2 A constraint also imposed on the maximum values of the
where u is the input power ratio. power that can be imported from/exported to the distribution
Meanwhile, The efficiency of DC/DC converter and grid as:
bi-directional DC/DC converter can be expressed in the
GRID
Pmin (t ) d PGRID (t ) d Pmax
GRID
(t ) (29)
same equation given by (18): The plus represents the power should be bought from the
u grid, and the minus represents the power can be sold to the
K2 u100% (18)
0.004  1.0017u  0.007u 2 grid.
Above mentioned, the objective function can be expressed 3.3 Algorithm
as
Based on the objective function and various constrains, a
f obj ,1 f obj ,2 genetic algorithm can be used to find an approximate
min f D( max
)  E( max
) (19)
solution. Genetic algorithm (GA)[12] is a kind of search and
f obj ,1 f obj ,2
optimization algorithm, which is based on the simulation of
biological evolution and heredity. GA simulates the
phenomena of selection, crossover and mutation in the

2873
process of natural selection and genetic process. Genetic 4.1 Numerical Simulations
operators are used to generate the best individuals from
When only optimizing the cost minimum (the first
generation to generation, and finally the best individuals are
objective), the optimal dispatch is shown as follows.
searched for. Specific steps are as follows:
120
z Initialization IC
100 MT
A group of individuals are randomly generated to form GRID
the initial population. Each individual is encoded in the 80

form of the initial solution, and each individual is valued 60


by the fitness value.

Power/kW
z Judgment 40

The result is judged to see whether it meet the 20

convergence criteria of the algorithm or not. if satisfied, 0


then output the search results, or continue to perform the
-20
following steps.
z Genetic manipulation -40

Based on the fitness value, genetic manipulation is 0 5 10 15 20


Time/h
executed.
Fig. 11: Power flow of single-objective optimization
z Iteration
Discharging
Return to the second step. Charging

20
4 Numerical Simulations and Discussions
The simulation example is based on the hybrid microgrid 10

Battery Power/kW
shown in Fig.3, including a 60 kW MT, a 20 kW battery, a 20
kW PV, a 20 kW WT, the maximum power of tie-line 0
between microgrid and external grid is 100kW, and the
maximum power of flow through the interlinking converter -10
is 40kW. The output of PV and WT is forecast by the
algorithm described as 2.1. The other data information is
-20
shown as follows:
0 5 10 15 20
4.0
Time/h
3.5 Fig. 12: Battery output power of single-objective optimization
SOC
3.0
Real-time Electric Price

2.5 100

2.0
80
1.5

1.0 60
SOC/%

0.5
40
0.0
0 5 10 15 20 25
Time/h 20

Fig. 9: Spot price curve


0
0 5 10 15 20
AC Load
Time/h
DC Load
120 Fig. 13: SOC of single-objective optimization
In the case of single-objective, the hybrid AC-DC
100
microgrid is considered as a whole: WT and PV have the
right to give priority to generation as clean renewable energy
Load Power/kW

80
which do not participate in the scheduling with other
60 distributed generations. From the Fig.12 we can see that
although the spot price has the promoting effect of battery,
40
because of the consideration of investment depreciation cost
20 and maintaining cost of battery, the battery can not produce
benefit and only charges at the valley price time and
0 discharges at the peak price time. The MT generates
5 10 15 20
Time/h
electricity in the peak price to reduce the purchase of
electricity from the grid and sell power to increase microgrid
Fig. 10: Daily load curve
revenue.

2874
On the basis of the above optimization goal, taking the Table 2: Result comparison
losses into consideration, the dispatch curves are as follows.
Total cost/yuan Power loss/kW
120
IC Single-objective 1014.406 38.501
100 MT
GRID Multi-objective 1044.272 25.509
80

60 5 Conclusions
Power/kW

40
Hybrid AC/DC microgrid is becoming more attractive
20 options than AC or DC microgrid. In this paper, a complete
0 energy management is introduced, followed with renewable
energy sources forecast by using the algorithm of MPSO-BP
-20
which is verified to have better performance, and
-40
establishing the mathematical model of multi-objective
0 5 10 15 20 optimization to minimize the operation cost and loss of
Time/h
converter. With the comparison of taking the losses as
Fig. 14: Power flow of multi-objective optimization
objective or not, the results show that minimizing loss of
Discharging converter may increase operational cost, which must be
Charging

20
balanced during the actual planning process. Numerical
simulations revealed the feasibility of the proposed model.
By reasonable optimization, the result can achieve an effect
10
that a small increase in cost with a big discount of the loss.
Battery Power/kW

0 References
[1] Eghtedarpour N, Farjah E. Power Control and Management
-10 in a Hybrid AC/DC Microgrid[J]. IEEE Transactions on
Smart Grid, 2014, 99(3):1494-1505.
[2] Kurohane K, Senjyu T, Yona A, et al. A Hybrid Smart
-20
AC/DC Power System[J]. IEEE Transactions on Smart Grid,
0 5 10 15 20 2010, 1(2):199-204.
Time/h [3] Bo D, Li Y, Zheng Z. Energy management of hybrid DC and
Fig. 15: Battery output power of multi-objective optimization AC bus linked microgrid[J]. Power Electronics for
Distributed Generation Systems IEEE International
SOC
Symposium O, 2010, 24(1):713 - 716.
100
[4] Lotfi H, Khodaei A. Hybrid AC/DC microgrid planning[J].
Energy, 2017, 118:37-46.
[5] Qachchachi N, Mahmoudi H, El Hasnaoui A. Optimal power
80
flow for a hybrid AC/DC microgrid[C]// Renewable and
Sustainable Energy Conference. IEEE, 2015:559 - 564.
60
[6] Li P, Hua H, Di K, et al. Optimal operation of AC / DC hybrid
SOC/%

microgrid under spot price mechanism[C]// IEEE Power and


40 Energy Society General Meeting. 2016:1-5.
[7] Teimourzadeh Baboli P, Shahparasti M, Parsa Moghaddam
20 M, et al. Energy management and operation modelling of
hybrid AC–DC microgrid[J]. Iet Generation Transmission &
0 Distribution, 2014, 8(10):1700-1711.
0 5 10 15 20 [8] Wan C, Zhao J, Song Y, et al. Photovoltaic and solar power
Time/h forecasting for smart grid energy management[J]. Power &
Fig. 16: SOC of multi-objective optimization Energy Systems Csee Journal of, 2015, 1(4):38-46.
[9] Zhang J W, Zhang Z J. Short-Term photovoltaic system
In the case of multi-objective, due to the losses on IC are power forecasting based on PSO-BP neural network[J].
taken into consideration, hybrid AC-DC microgrid is Kezaisheng Nengyuan(Renewable Energy Resources), 2012,
considered separately as AC side and DC side. The MT 30(8): 28-32.
should fully generate during the electricity price peak period [10] Rong F U, Wang W Q, Gui-Xiong H E. The forecasting of
and the rest of the time should considerate the losses, and in wind speed in wind farm based on the meteorological factors
order to reduce the power losses flowing through the IC, the with BP neural network[J]. Renewable Energy Resources,
battery also need to strike a balance between making full use 2009, 5: 029.
of electricity price and reducing losses. [11] Kakigano H, Miura Y, Ise T, et al. Basic sensitivity analysis
of conversion losses in a DC microgrid[C]//Renewable
The optimal result with the different objective is shown as
Energy Research and Applications (ICRERA), 2012
Table 2, when we take cost as the objective only, the cost is International Conference on. IEEE, 2012: 1-6.
minimum, while the power loss on converter is 38.501 kW. [12] Peng D, Qiu H, Zhang H, et al. Research of Multi-objective
When we take the power loss as objectives at the same time, optimal dispatching for microgrid based on improved Genetic
although the cost added 29.8 yuan, but the loss is reduce to Algorithm[C]// IEEE, International Conference on
13kW, dropped 33.7% percent. Networking, Sensing and Control. IEEE, 2014:69-73.

2875

You might also like