You are on page 1of 9

46 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 5, NO.

1, JANUARY 2014

Improving Sustainability of Hybrid Energy Systems


Part II: Managing Multiple Objectives With a
Multiagent System
Christopher M. Colson, Member, IEEE, M. Hashem Nehrir, Life Fellow, IEEE, Ratnesh K. Sharma, Member, IEEE,
and Babak Asghari, Member, IEEE

Abstract—Hybrid power systems and microgrids may employ a proving performance for such systems is often subjective and
mixture of dispatchable (conventional) and nondispatchable (re- can be hard to fully define, as will be discussed below.
newable) generators alongside storage. Whether in grid-connected Microgrids continue to be of interest as infrastructure compo-
or grid-isolated (islanded) modes of operation, these systems
may face multiple competing objectives when managing diverse nents that can readily connect with or separate from the utility
installed assets. Power management of hybrid energy systems, grid, sustaining continuous operations while offering broader
therefore, involves operational tradeoffs amongst Pareto-optimal opportunities to integrate locally assigned objectives for power
solutions. These attributes, including the ready implementation management. Additionally, it is expected that a prime means
of distributed renewable generation and the incorporation of for deploying distributed energy resources (DERs), especially
methods to locally manage power-networked assets, make them a
unique area of study for pursing better sustainable performance. renewables, will be in hybrid or microgrid systems. The supply
In part I of this paper, storage system round-trip efficiency and available from renewable DERs may vary greatly within a very
operational cost concepts were formulated for use in real-time short period of time, and with many such assets coupled on a
dispatch decisions towards yielding improved performance of relatively small power network, asset dispatch and control de-
overall system objectives. In this paper (part II), the concepts of cisions must be made at a more rapid pace than is common for
part I are implemented with a decentralized multiagent system
(MAS). This MAS is employed for power management of a hybrid conventional power systems. The combination of these factors,
(diesel-storage battery) microgrid in grid-connected and islanded along with a desire to incorporate local power management ob-
modes. This paper highlights the development and implementation jectives, represents a shift in the operational approach towards
of an MAS suitable for hybrid and microgrid system applications, installed assets, including the employment of storage systems.
as well as presenting an important discussion about the tradeoffs In this paper, the authors contribute an approach to address
associated with multiobjective design for power management.
The simulation results presented demonstrate improvement in the multiobjective decision-making for hybrid or microgrid sys-
sustainable performance of the hybrid system. tems that is different than conventional techniques for power
management. Here, the authors expand on the MAS from their
Index Terms—Distributed generation, microgrid, multiagent.
previous work, [1], and utilize the formulations in part I of this
paper [2], within the decision-making MAS framework, to seek
I. INTRODUCTION power management solutions for a diesel-battery storage hybrid
stationary power system. The authors intend to introduce the
complexity seeking optimal power management solutions for
M ICROGRIDS are still rare as their development is at an
early stage, but conceptually, microgrids are widely un-
derstood to be self-contained power systems of a few MW or
multiple objectives and diverse assets incorporated into a micro-
grid. MAS methods are recent and their application to the micro-
grid multiobjective problem is new. Therefore, the authors seek
less of generation, controllable and noncontrollable load cen-
to present the MAS approach used as a foundation for future
ters, and local energy storage. More common globally are hy-
works that will compare MAS performance to other techniques,
brid systems that couple two or more generation and/or storage
including centralized decision-making methods; what method-
assets together to enhance system operation. Clearly, a primary
ology or technique is best for addressing microgrid management
purpose of implementing multiple assets in a hybrid or micro-
holistically remains to be shown. The diesel-battery hybrid was
grid configuration is to improve performance of the overall en-
chosen as an example for two reasons. First, for hybrid systems
ergy system. It must be noted, however, that notions of im-
employed worldwide, diesel-battery systems are some of the
most commonly seen, especially in islanded applications. This
Manuscript received October 14, 2012; revised February 27, 2013 and May
allows ready comparison of MAS power management to con-
24, 2013; accepted June 04, 2013. Date of publication August 08, 2013; date of
current version December 12, 2013. This work was supported in part by NEC ventional control methods. Examples range from on-grid unin-
Laboratories-America, Inc., and in part by DOE Award DE-FG02-11ER46817. terruptible power supplies (UPS) [3], to off-grid rural commu-
C. M. Colson and M. H. Nehrir are with the Electrical and Computer En-
nity power [4], [5], to remote base transceiver stations (BTS)
gineering Department, Montana State University, Bozeman, MT 59717 USA
(e-mail: christopher.colson@msu.montana.edu; hnehrir@ece.montana.edu). for critical telecommunications [6]–[8]. Second, the diesel-bat-
R. K. Sharma and B. Asghari are with NEC Laboratories America, Cupertino, tery hybrid is a suitable example of a system that combines a
CA 95014 USA (e-mail: ratnesh@sv.nec-labs.com; babak@sv.nec-labs.com).
small number of controllable assets, each with sufficient com-
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. plexity of optimizable objectives, to be illustrative of MAS de-
Digital Object Identifier 10.1109/TSTE.2013.2269319 cision-making.

1949-3029 © 2013 IEEE

Authorized licensed use limited to: Universidad de Antioquia. Downloaded on November 14,2020 at 23:01:02 UTC from IEEE Xplore. Restrictions apply.
COLSON et al.: IMPROVING SUSTAINABILITY OF HYBRID ENERGY SYSTEMS PART II 47

The predominance of MAS literature is reviewed in [1] and


is not repeated here. For the MAS system based on [1] and
developed in this paper, the multiobjective problem is com-
plicated by knowledge that, in many instances, an improve-
ment in one objective may lead to a worsening in another; a
principle that must be taken into consideration. Additionally,
it is desirable to incorporate in situ goals defined by the mi-
crogrid owner/operator (hereafter referred to as the user) into
MAS decision-making. Therefore, Section II highlights aspects
of MAS decision-making that account for these considerations
and occurs without the benefit of a central supervisor. Further,
Section III details the application of the developed MAS to the
power management problem in an environment where tradeoffs
between objectives are key.
In [1], incremental simulations were shown to illustrate MAS
behavior when confronting a singular disturbance to a micro-
grid. There, the storage agents used cost and performance func-
tions based on a straightforward conventional charge-discharge
routine for a generic storage system. To advance the MAS given
Fig. 1. Agent decision cycle.
in [1], in Section IV, singular disturbance simulations for a grid-
connected diesel-storage battery microgrid are conducted using
an improved lead-acid battery model and better cost and per- ditions, etc. Second, the time to complete a cycle varies based
formance functions developed in part I of this paper [2]. In on many factors including computation time, intensity of nego-
Section V, the MAS is simulated for an islanded diesel-storage tiations, latency, and execution time. Ultimately, the decision
battery hybrid system under continuous disturbances as would cycle allows the agents to manage their assigned asset inde-
be expected under real-world conditions and is compared to a pendently and respond to events as they occur, without relying
conventional control technique for such systems. Sections IV upon a centralized supervisor. Collectively, agents operating au-
and V offered opportunities to present the performance of MAS tonomously can self-organize and collaborate through negotia-
power management over conventional methods for common on- tion during their decision cycles. This concept allows the MAS
and off-grid diesel-battery hybrids currently used around the to be flexible and responsive while seeking optimal solutions to
world. With these demonstrations, the authors seek to add to assigned goals.
recent research employing MAS architectures towards power An agent’s decision cycle (all references shown in Fig. 1)
systems applications, including microgrids [9]–[13]. Conclu- initiates asynchronously when an initiating event is perceived
sions are given in Section VI highlighting how agents’ improved by the agent (at ). The agent senses local conditions and con-
ability to make better asset decisions may lead to better perfor- ducts internal computation to determine a solution to its present
mance for hybrid and microgrid systems, including those that conditions during the computational period (between and ).
incorporate renewable DERs. The awareness of local conditions and the quality of solution
may lead the agent to seek cooperation with other agents, if dic-
II. MAS DECISION-MAKING tated by the agent’s algorithm; otherwise, the agent moves di-
The primary purpose of the hybrid or microgrid system power rectly to execute its decision. If cooperation is desired, the agent
management architecture is to ensure stable delivery of elec- enters a period of negotiation with one or more MAS agents
trical power to local loads, while optimizing energy production (between and ). This period may be very short due to a
towards an assigned objective(s). Here, building on [1], aspects lack of response from potential collaborators, or involve many
of the decentralized agents’ decision-making process in a mul- messages passed between agents as they settle upon a mutually
tiobjective environment are discussed. agreeable course of action. Once negotiation is complete, the
agent carries out its decision during a period of action execu-
A. Decision Cycle in a Decentralized MAS tion (between and ) that continues until a new initiating
The decision cycle is an importance concept for MAS op- event starts the decision cycle to begin again. If no negotiation
eration, introduced in [1], shown in Fig. 1. For the agents that was conducted, the period of execution follows the computation
make up the decentralized MAS, a decision cycle is defined as period directly (between and ). In this work, time between
the amount of time used by each agent to formulate, negotiate, initiating events was significantly longer (1 s or greater) than
and execute a solution for the operation of its local asset. This the total time to complete a decision cycle ( ms).
is similar in concept to trading cycles used by market agents en- In future work, initiating events that occur prior to full cycle
gaged in economic auctions, such as in [13], with two key differ- completion and, therefore, interrupt decision cycle flow will be
ences. First, as opposed to market-based auctions that occur at addressed.
regularly synchronized intervals, the decision cycle is initiated The software developed to operate the multiagent system
asynchronously by unpredictable events such as spot market was developed by the authors and utilized the Java Agent DE-
price changes, resource availability changes, emergency con- velopment (JADE) communication libraries for the Foundation

Authorized licensed use limited to: Universidad de Antioquia. Downloaded on November 14,2020 at 23:01:02 UTC from IEEE Xplore. Restrictions apply.
48 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 5, NO. 1, JANUARY 2014

variables. This means that the Pareto set of solutions in one time
increment may be dramatically different from the Pareto set in
the next increment. This phenomenon, as well as the microgrid
MAS ability to find strong solutions along the Pareto front, will
be shown in Section IV.

III. MAS POWER MANAGEMENT PRINCIPLES OF OPERATION


As the MAS operates, cost and performance functions are cal-
culated by the agents for their respective assets at each computa-
Fig. 2. Representation of a multiobjective, multiconstraint optimization tional increment. As discussed in [1] and part I of this paper [2],
problem and resulting Pareto frontier.
a cost measure that incorporates fixed and variable cost com-
ponents is, alone, inadequate to fully define the optimal oper-
for Intelligent Physical Agents (FIPA)-compliant protocols—a ation of hybrid or microgrid system assets. It is noted for all
public domain software. simulations, cost functions for microgrid assets were limited to
energy costs and fixed overhead; maintenance and other vari-
B. Pareto Decision-Making able costs were omitted, but may be readily included in more
Often, decision-making requires analysis of numerous com- detailed cost functions. Subsequently, each agent determines a
peting goals simultaneously. Multiobjective optimization seeks performance measure that reflects how well its asset is achieving
an optimal solution amidst a combination of objective functions, optimal operation based on its given objectives. The charac-
described by maximizing or minimizing as follows: teristics of the cost and performance functions vary dynami-
cally depending on the asset and conditions including loading,
state-of-charge (SOC), discharge history, etc. Cost and perfor-
(1) mance information may be shared between agents according to
the decision cycle, described in Section II, to assist in cooper-
subject to ation. In order to solve for the best operating condition for an
asset, the agent assigned to it first uses local information to de-
(2) velop its cost and performance functions, and , re-
spectively. These functions are unique for every asset and de-
where is the vector of objective functions containing objec- pend on different decision variables . To simplify the problem
tive functions ( to ), is the -length vector of variables presentation, the set of decision variables used for the simula-
to be optimized known as the decision variables, and rep- tions in Sections IV and V consisted of the individual microgrid
resents the -length vector of inequality constraints. Equality asset’s dispatch power. However, the set of decision variables
constraints were not applicable to the power management con- may be broader including voltage, frequency, and other factors
sidered in this paper. However, in most cases, a global best of of operation. For this paper, each agent’s overall objective func-
any particular individual objective function may not be a sat- tion was
isfactory solution for the remaining objectives [14]. Because
of this, we need to alter our concept of optimality for prob- (3)
lems such as for hybrid or microgrid power management. Sim-
ilar to economic systems, Pareto optimal solutions each max- where is the combinational objective function for agent
imize or minimize the objective function , but given it containing cost objective function and performance ob-
is composed of multiple objectives, each Pareto optimal so- jective function , weighted according to the user-defined
lution implies that any one objective’s satisfaction cannot be and . The details of the functions and variables used are
raised any further without lowering another objective’s satis- shown in [1] for the diesel genset and in [2] for the lead-acid
faction, or vice versa. In other words, many “optimal” solu- battery. This linear sum of assigned objectives defines the deci-
tions exist where the value of cannot be improved any fur- sion-making search space which the agent must solve.
ther. To illustrate this, consider two decision variables , Within the MAS framework, two agents are assigned to a
three constraints (dashed lines), and two objective functions storage asset; one producer and one consumer. Each decision
shown in Fig. 2. A particular solution in the re- cycle, these agents independently seek solutions to the dispatch
gion of feasible solutions, shown on the left side of Fig. 2, maps of their asset, guided by goal information provided by the user.
to the objective functions graph, projection shown by the dotted User goals directly influenced the weights assigned to
line to the right side of Fig. 2, where the particular solution’s rel- the components of performance utilized by individual agents,
ative achievement of each objective is shown. The Pareto front described further in Section IV, scenario 2 below. However, be-
represents the combinational minimization of the two objective cause two agents are responsible for the same asset, negotiation
functions, shown by the dark line in Fig. 2. Along this front, is forced to arise each decision cycle. This ensures that merits
is at a minimum. It is noted that no direct method to de- of charging or discharging in each computational increment are
rive the Pareto set of solutions is known. properly evaluated. The result of storage agents’ negotiations
Unfortunately, for the hybrid or microgrid system, conditions leads to an operating decision based upon a mutually agreed
can change very rapidly changing the constraints and decision upon solution.

Authorized licensed use limited to: Universidad de Antioquia. Downloaded on November 14,2020 at 23:01:02 UTC from IEEE Xplore. Restrictions apply.
COLSON et al.: IMPROVING SUSTAINABILITY OF HYBRID ENERGY SYSTEMS PART II 49

TABLE I TABLE II
MICROGRID PARAMETERS SCENARIO PARAMETERS

that was only collected for testing and was not available to the
agents. While asset dispatch decisions were made exclusively
by the agents, the user could impose their will upon the storage
agents by specifying a relative importance between
revenue and battery health. The agents made dispatch decisions
independently, but communicated as appropriate.
In scenarios 1 and 2 below, simulation results are presented in
Fig. 3. Concept of operations for simulated two-asset microgrid. three increments: initial conditions, disturbance, and final con-
ditions. The results are presented this way for the purpose of
clarity, illustrating the MAS decision-making outcomes and al-
IV. SIMULATION AND RESULTS—GRID CONNECTED CASE lowing easier comparison of system conditions upon a system
This section builds on the analytic work developed in [1] and disturbance. In Section V, the results for a similar system (is-
[2]. Here, a simulated grid-connected microgrid includes a hy- landed case) are presented and aggregated outcomes are com-
brid system composed of a diesel genset and a lead-acid storage pared with the outcomes from a conventional case.
battery bank serving a fixed load center with noncontrollable
loads (parameters shown in Table I). The load was fixed in these A. Scenario 1: Spot Market Price Change
scenarios to simplify the comparative results which would be The purpose of this scenario was to demonstrate how agents
affected by changing demand. The microgrid is connected to sense and respond to a system disturbance, using local cost
the utility through a point of common coupling (PCC), either and performance information to guide their decision-making.
supplying or consuming energy across the interconnection. The Here, the microgrid was simulated at an initially stable con-
MAS used for this study employed a producer agent assigned dition, given in Table II, interconnected with the utility and
to the diesel genset, both a producer agent and consumer agent all objectives weighted equally . Initially given the
assigned to the storage asset, and an observer agent assigned to high cost of utility electricity supply (Fig. 3), the microgrid
pass utility spot price information to the MAS collective, shown was both operating its diesel and discharging the battery to
in Fig. 3. offset purchases from the utility required to power the local
In the decentralized decision-making process, two objective load. Then, system conditions were disturbed by a significant
functions were imposed upon each agent within the system: increase in the market price of electricity supplied by the
minimize operational cost and maximize system perfor- utility, from $0.31 to $0.38 per kWh, yielding a total cost of
mance . Operational cost was defined by fuel consump- load operation increasing from $31/h to $38/h. Given a new
tion and the cost of battery charge/discharge, respectively. Per- price environment, the diesel producer agent manipulated its
formance was defined by operational efficiency and environ- asset to operate more profitably and with better performance
mental emissions, namely total hydrocarbons (THC) and carbon within constraints. During the short time period of grid price
dioxide (CO ). The solutions decided by the agents translate change, SOC of the battery did not change appreciably. After
into actual operation of their assets. For the purpose of evalu- the battery producer and consumer agents communicated, their
ating the solution quality of the whole microgrid, the individual decision was to continue discharging at predisturbance levels.
cost and performance results of each agent’s solution was aggre- Ultimately, as grid price rose in this scenario, new operating
gated. This information is highlighted by shading in Tables II parameters determined by the agents lead to better overall mi-
and III. It is important to note, however, that this overall mea- crogrid performance and more operational savings than prior to
sure of microgrid performance is a form of global knowledge the disturbance. As described earlier in this section, the overall

Authorized licensed use limited to: Universidad de Antioquia. Downloaded on November 14,2020 at 23:01:02 UTC from IEEE Xplore. Restrictions apply.
50 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 5, NO. 1, JANUARY 2014

TABLE III
SCENARIO PARAMETERS

Fig. 5. Screenshot showing the discrete search space for postdisturbance con-
ditions of scenario 1. The objective functions, to minimize cost and maximize
performance, form the plot axes. Each circle represents a solution.

Fig. 6. Screenshots of the user input interface for the relative importance of
revenue versus battery health/longevity. Balanced initial conditions of scenario
2 are left; disturbance conditions (favoring battery health/longevity) are shown
on the right.

Fig. 6. The visual depiction of a weighing scale offers the user


an intuitive sense of relative importance, choosing which goal
to more heavily emphasize, thereby adjusting and . The
Fig. 4. Screenshot showing the discrete search space for predisturbance condi-
tions of scenario 1. The objective functions (minimize cost and maximize per- dollar sign (the left of the scale in Fig. 6) represents revenue;
formance) form the axes. Each circle represents a solution. the caduceus (the right of the scale) represents battery health
and longevity.
Initially, the microgrid conditions are identical to the final
cost and performance of the microgrid operating solution are conditions from scenario 1, shown in Table III. The predis-
shown in the shaded rows of Tables II and III. turbance cost and performance functions for the storage pro-
The aggregated solution can be difficult to see, however, out- ducer and consumer agents are shown in Fig. 7. The distur-
side the context of the search space that the agents have avail- bance occurs when the user manipulates their interface (initially
able to them. Subsequently, in Fig. 4, the initial search space is ), making battery health more important than cost/
displayed, showing all possible discrete solutions for dispatch profit ( ). Immediately, sensing a shift in objectives, the
on the microgrid; an arrow points to the operating solution col- storage agents commence negotiations on how to best achieve
lectively decided upon by microgrid agents. In Fig. 5, the post- them. Meanwhile, the grid price has not changed and the diesel
disturbance search space is shown, reflecting the change in oper- genset, already operating at its optimal state given the condi-
ational conditions that deepens the set of all possible solutions. tions, does not modify its operation. Likewise, the grid cost
Again, the operating solution determined by the MAS is anno- ($0.38/kWh) remains below the charging cost threshold ($0.45/
tated by the black arrow. In both cases, before and after the dis- kWh). However, when battery agent negotiation is completed,
turbance, the agents have settled upon operating parameters for the initial 5-kW battery discharge rate changes to charging at
their individual assets that yield a total microgrid solution that 25 kW. In this scenario, as user goals changed to favor battery
is along (or very proximate) to the Pareto set of solutions. The health, the overall system performance improved from 80% to
progress of events for scenario 1 is summarized in Table II. 99%, while operational savings changes from $6/h to -$3/h, in-
dicating a shift from savings to expense.
B. Scenario 2: Change of User-Assigned Battery Goals At the outset, the system was at an equilibrium state where
The purpose of this scenario was to show how agents can each agent has completed a decision cycle and is awaiting an
not only incorporate changing operational conditions and con- initiating event. When the disturbance occurs altering the ob-
straints, but goals that adjust unpredictably, as well. Here, the jectives for the storage agents, the performance functions deter-
user manipulates the relative importance of microgrid revenue mined independently by each storage agent reveal a scenario
to the health and longevity of the lead-acid battery. In other where no point of discharge would yield better performance
words, the user may influence MAS decision-making by con- than any charging point. This is illustrated in Fig. 8 where the
veying goals dynamically. The user accomplishes this by inter- producer agent’s discharging performance curve is less
acting with microgrid agents through a user interface, shown in than the consumer agent’s charging performance curve

Authorized licensed use limited to: Universidad de Antioquia. Downloaded on November 14,2020 at 23:01:02 UTC from IEEE Xplore. Restrictions apply.
COLSON et al.: IMPROVING SUSTAINABILITY OF HYBRID ENERGY SYSTEMS PART II 51

Fig. 7. Screenshot showing the predisturbance cost and performance characteristics for the storage producer and consumer agents in scenario 2. The solid vertical
line represents actual battery dispatch; the dotted vertical line represents the dispatch limit.

Fig. 8. Screenshot showing the postdisturbance cost and performance characteristics for the storage producer and consumer agents in scenario 2. The solid vertical
line represents the actual battery dispatch.

for all possible dispatch powers. This presents conditions where


objectives of minimizing cost and maximizing performance are
nearly directly opposed. In other words, because the user has
emphasized the goal of battery health and there is no feasible
solution that achieves a better objective function value, ,
for discharging than for charging, the battery must charge. Using
the performance functions and through negotiation, the storage
agents arrive at this conclusion collaboratively. Despite being
costly to charge, the storage agents’ solution best achieves the Fig. 9. Screenshot showing the search space for postdisturbance conditions of
scenario 2. The search space is greatly condensed given operating conditions.
user goals given the operating conditions. This decision is re-
flective of the user’s intentions to maintain the battery health
and longevity over cost implications, dictated by the user inter- performance measure, partially a function of SOC, contains op-
face shown by Fig. 6. In Fig. 9, the postdisturbance search space erating points within its feasible region that represent perfor-
has tightened compared to the predisturbance search space for mance values better than for charging. This indicates an oppor-
scenario 2, shown in Fig. 5. tunity to improve battery operating performance; the agents rec-
In this scenario, the focus was on an isolated disturbance ognize this and conclude to resume discharging. In this way,
within a short period of time. Of course, as time moves ahead, using only local information, the agents autonomously follow
the battery continues to charge until SOC improves to about the goals set forth by the user and dispatch the battery to best
75% in this case, shown in Fig. 10. At this point, the discharge achieve them.

Authorized licensed use limited to: Universidad de Antioquia. Downloaded on November 14,2020 at 23:01:02 UTC from IEEE Xplore. Restrictions apply.
52 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 5, NO. 1, JANUARY 2014

Fig. 10. Screenshot showing the cost and performance characteristics for the storage producer and consumer agents after the battery charging has recovered
SOC to 75% from the postdisturbance conditions of scenario 2. The solid vertical line represents the actual battery dispatch; the dotted vertical line represents the
dispatch limit.

Fig. 12. Normalized load demand profiles for 50-home village [15].

Fig. 11. Concept of operations for simulated two-asset islanded microgrid.


TABLE IV
HYBRID SYSTEM ASSET CAPACITY OPTIONS

V. SIMULATION AND RESULTS—ISLANDED CASE


The computation culminating in agent action during a single
decision cycle, described in Section II, is difficult to illus-
trate. To that end, however, two scenarios were presented in
Section IV for the purpose of showing how the MAS may
respond to a disturbance. In this section, rather than focusing
on an isolated disturbance, the MAS was simulated under
constantly changing conditions, as would be expected in a increment. Only results from the Summer Weekend profile are
real-world situation, continuing the analytic work developed in given in Table V, but overall simulation outcomes were similar
[1] and [2]. In this case, a grid-isolated (islanded) hybrid system for each daily profile.
composed of a diesel genset and a lead-acid storage battery The MAS used for this study was nearly identical to that used
bank, serving a noncontrollable load center, shown in Fig. 11, in Section IV, with a major exception. In the grid-connected
was simulated over a full day. The load center was modeled as case, excess power produced locally by assets on the micro-
a residential neighborhood or village of 50 all-electric homes, grid could be absorbed by the utility. Conversely, inadequate
daily load profiles of which are shown in Fig. 12 [15]. The use local supply could be supplemented with utility supply. In the
of a realistic load profile was intended to simulate a demand islanded case, however, this is not possible and the local assets
that varied in real-time; the MAS had no foreknowledge of load are constrained to meet demand at all times. The consequence
changes and reacted to demand information from increment to of this constraint emphasizes the operational tradeoffs between

Authorized licensed use limited to: Universidad de Antioquia. Downloaded on November 14,2020 at 23:01:02 UTC from IEEE Xplore. Restrictions apply.
COLSON et al.: IMPROVING SUSTAINABILITY OF HYBRID ENERGY SYSTEMS PART II 53

TABLE V
FOUR SIMULATION SCENARIOS FOR 24-HOUR SUMMER WEEKEND DAY (BEST METRICS IN SCENARIO SHADED IN DARK GREY)

the diesel and battery assets that are not present in the grid-con- sider diesel or battery efficiency, emissions, or costs. For the
nected case. In other words, when grid-connected, each asset is purpose of comparison, this conventional power management
fairly free to operate where the assigned agent determines the scheme is used as the base case. In Table V, the conventional
best operating point to be. However, in the islanded case, the case is compared to the operation of the MAS under three sep-
agents must coordinate their operating points to precisely meet arate power management approaches:
load demand. 1) Battery operation is favored over the diesel.
Clearly, the capacity ratings of the diesel and battery assets 2) Diesel operation is favored over the battery.
matter significantly in the islanded case. After all, the asset rat- 3) Neither asset is favored; dispatch optimization decisions
ings determine how much energy storage is available to sup- are made relative to capacity rating.
plement dispatchable power, as well as how long it takes to re- In order to appropriately compare the operational results be-
plenish stored energy. To that end, two sizes of diesel gensets tween the power management methods, metrics such as effi-
and two sizes of batteries were considered in this study. Sizes ciency, cost, and emissions make up the rows of Table V. The
were dictated by load demand, shown in Table IV. Using each darkly shaded boxes indicate the power management method
possible combination of asset capacities, four unique scenarios that achieved the best results within the scenario. Below the dark
were used for simulations: Large Diesel Small Battery; Large horizontal line, the last four rows give the combined microgrid
Diesel Large Battery; Small Diesel Small Battery; and results. For each scenario, the best result within a row is high-
Small Diesel Large Battery. lighted with dark shading. It can be noted from Table V that the
Conventionally, diesel-battery hybrid systems operate with a MAS that used a blend of both assets’ cost and performance ob-
fairly simple control strategy. In terms of power management, jectives to guide agent decisions, rather than focus on attempting
the method may be summarized by the following rules: to optimize a single asset, consistently yielded the best results
1) The battery supplies power to the load until reaching a low (achieves 10 of the 16 possible best solutions for system metrics
voltage threshold, indicating low SOC. across the four scenarios). However, some observations are im-
2) At the low voltage threshold, the diesel runs, powering the portant to note. For the multiobjective problem, there are many
load in addition to fully recharging the battery. cases where conditions arise where superior results for one goal
3) As the battery charges, typically at a set rate, voltage rises may be sacrificed for another. This finding is consistent with
and charging power decreases. the Pareto-optimal discussion in Section II, and can be seen by
4) The diesel shuts off at a set low-load threshold. some of the results. For example, in the Large Diesel Large
This conventional control method for diesel-battery hybrid Battery scenario, the conventional dispatch method achieves the
operations can be problematic [16], and clearly does not con- best system efficiency. However, the MAS using mixed objec-

Authorized licensed use limited to: Universidad de Antioquia. Downloaded on November 14,2020 at 23:01:02 UTC from IEEE Xplore. Restrictions apply.
54 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 5, NO. 1, JANUARY 2014

tives showed better costs and emissions for the same scenario. [7] M. D. Bhawan and J. L. N. Marg, Consultation Paper on Green
In other words, in this case, although the MAS using mixed ob- Telecommunications, Telecom Regulatory Authority of India, New
Delhi, India, 3/2011, Feb. 2011.
jectives had poorer efficiency, it had better results overall. This [8] J. Rijssenbeek, H. Wiegman, D. Hall, C. Chuah, G. Balasubramanian,
case highlights the importance of considering multiple objec- and C. Brady, “Sodium-metal halide batteries in diesel-battery hybrid
tives for power management and tailoring these objectives to telecom applications,” in Proc. 2011 IEEE Int. Telecommunications
Energy Conf. (INTELEC), Amsterdam, The Netherlands.
the user’s goals. Another example of this phenomenon is seen [9] A. L. Dimeas and N. D. Hatziargyriou, “Operation of a multiagent
for the Large Diesel Small Battery scenario. Although the system for microgrid control,” IEEE Trans. Power Syst., vol. 20, no.
MAS that uses mixed objectives yields the best efficiency and 3, pp. 1447–1455, Aug. 2005.
[10] M. Pipattanasomporn, H. Feroze, and S. Rahman, “Multi-agent sys-
emissions results, it does so at higher cost than other power man- tems in a distributed smart grid: Design and implementation,” in Proc.
agement methods. Again, in most cases tradeoffs exist when op- IEEE Power Systems Conf. Exposition, 2009, pp. 1–8.
timizing multiple objectives. [11] Z. Jun, L. Junfeng, W. Jie, and H. W. Ngan, “A multi-agent solution to
energy management in hybrid renewable energy generation system,”
Renew. Energy, vol. 36, pp. 1352–1363, 2011.
VI. CONCLUSION [12] K. Huang;, S. Srivastava, and D. Cartes, “Decentralized reconfigura-
Making the grid, as well as hybrid and microgrid systems, tion for power systems using multi agent system,” in Proc. 2007 IEEE
Systems Conf., Honolulu, HI.
smarter requires advances on many fronts. In this paper, we [13] B. Ramachandran, S. K. Srivastava, D. A. Cartes, and C. S. Edrington,
have explored the exploitation of an MAS for decentralized “Distributed energy resource management in a smart grid by risk based
multiobjective power management within a hybrid or microgrid auction strategy for profit maximization,” in Proc. 2010 IEEE Power
and Energy Society General Meeting, Minneapolis, MN, USA.
system. Dynamically incorporating user goals into MAS deci- [14] S. Gass, Linear Programming: Methods and Applications, 5th ed.
sion-making, as well as handling rapid changes in operational New York, NY, USA: Dover, 2003, pp. 224–250.
conditions, demonstrate the capabilities of this approach to [15] C. Colson and M. H. Nehrir, “An alternative method to load mod-
eling for obtaining end-use load profiles,” in Proc. 41st North American
achieving real-time optimal solutions. Further, the authors have Power Symp., Starksville, MS, USA, Oct. 2009.
presented the importance of using many objectives, including [16] Bagen and R. Billinton, “Evaluation of different operating strategies in
round-trip efficiency and battery system operational cost, to small stand-alone power systems,” IEEE Trans. Energy Convers., vol.
20, no. 3, pp. 654–660, Sep. 2005.
assist in making better power management decisions for hybrid
or microgrid systems that incorporate storage. The consider-
ation of how tradeoffs between objectives affect the overall Chris M. Colson (S’07–M’13) received the B.S.
degree in control systems engineering from the
power management solution is often difficult to deduce, but United States Naval Academy, in 1999, the MEM
discussion presented herein offers some insight into addressing degree from Old Dominion University, in 2006,
tradeoffs towards a more capable and responsive power system, and the Ph.D. degree in electrical engineering from
Montana State University, in 2012.
especially when augmented by hybrid or microgrid systems. He served as a Submarine Warfare Officer in the
The combination of multiple assets into microgrids and the U.S. Navy and was a National Science Foundation
use of MAS architectures, such as shown in this paper, offer (NSF) Graduate Research Fellow. His research inter-
ests include microgrids, power system integration of
opportunities to embed decentralized intelligence into emerging emerging technologies, and intelligent computational
power systems and, ultimately, can provide significant “smarts” methods.
at the lowest hierarchical level of the power system.

REFERENCES
M. Hashem Nehrir (S’68–M’71–SM’89–F’10–LF’13) is a Professor of Elec-
[1] C. M. Colson and M. H. Nehrir, “Comprehensive real-time microgrid trical and Computer Engineering at Montana State University. His research in-
power management and control with distributed agents,” IEEE Trans. terests include modeling and control of power systems, alternative energy power
Smart Grid, vol. 4, no. 1, pp. 617–627, Mar. 2013. generation systems, and application of intelligent control to power systems.
[2] C. M. Colson and M. H. Nehrir, “Improving sustainability of hybrid
energy systems—Part I: Incorporating battery round-trip efficiency and
operational cost factors,” IEEE Trans. Sustain. Energy, accepted for
publication.
[3] C. V. Nayar, M. Ashari, and W. W. L. Keerthipala, “A grid-interactive Ratnesh K. Sharma (M’11) has over 12 years experience in sustainable energy
photovoltaic uninterruptible power supply system using battery storage management for distribution systems in buildings and transportation. He is the
and a back up diesel generator,” IEEE Trans. Energy Convers., vol. 15, founding head of the Energy Management Department at NEC Laboratories
no. 3, pp. 348–353, Sep. 2000. America, Cupertino, CA, USA. He has authored more than 200 papers/technical
[4] R. W. Wies, R. A. Johnson, A. N. Agrawal, and T. J. Chubb, “Simulink reports and holds over 70 U.S. patents on energy management and related areas.
model for economic analysis and environmental impacts of a PV with
diesel-battery system for remote villages,” IEEE Trans. Power Syst.,
vol. 20, no. 2, pp. 692–700, May 2005.
[5] P. M. Crimp, S. Colt, and M. A. Foster, Renewable Power in Rural Babak Asghari (S’06–M’09) received the Ph.D. degree in electrical engi-
Alaska: Improved Opportunities for Economic Deployment, Institute neering from the University of Alberta, Canada. He is currently a researcher in
of the North—Arctic Energy Summit, Anchorage, AK, USA, 2008. the Energy Management Department at NEC Laboratories America, Cupertino,
[6] E. M. Nfah and J. M. Ngundam, “Evaluation of optimal power options CA, USA. His research interests include design of energy management systems
for base transceiver stations of mobile telephone networks cameroon,” for microgrids and real-time simulation and control of power systems and
Solar Energy, vol. 86, pp. 2935–2949, 2012. electric drives.

Authorized licensed use limited to: Universidad de Antioquia. Downloaded on November 14,2020 at 23:01:02 UTC from IEEE Xplore. Restrictions apply.

You might also like