Reactive Power and Voltage Control in a Power Grid:

A Game-Theoretic Approach
Ikponmwosa Idehen Shiny Abraham Gregory, V. Murphy
Electrical and Computer Engineering Electrical and Computer Engineering Electrical Engineering
Texas A & M University Seattle University Tuskegee University
College Station, TX, USA Seattle, WA, USA Tuskegee, AL, USA
[email protected] [email protected] [email protected]

Abstract— This paper presents an approach of applying game- minimization of losses between interconnecting micro-grids in a
theoretic concepts to the systematic control of reactive power and smart grid [3] – [6]. This paper presents an overview of the
voltage in a power grid. A non-cooperative, extensive form game concept of game theory, and explores its direct application in
is used to model the complex interactions that exist among voltage- simplifying the complex interactions that exist among different
control components of a power system. An IEEE 6-bus test system
voltage-control components of a power system. The approach
has been used as a case study to study the systematic interactions
among system components. At equilibrium, component settings utilizes a systematic method of controlling reactive power flow,
minimize the reactive power injections, while satisfying load- bus while ensuring an acceptable voltage profile at all buses. The
voltage constraints. determination of optimal parameter settings for the different
voltage-control devices is an important result of this process.
Index Terms— power systems, game theory, nash equilibrium, This paper has been organized as follows: Section II presents
reactive power, voltage control. an overview of game theory and presents key terminologies used
in this work. Problem formulation and the game-theoretic model
I. INTRODUCTION are discussed in Section III. Preliminary results and conclusions
Maintaining an acceptable grid-voltage profile is an are presented in Sections IV and V, respectively.
important task for power utilities. The constantly changing state
of the grid requires that reactive voltage-control devices be
II. GAME THEORY OVERVIEW
periodically re-dispatched to ensure that voltage limits are not
violated. These devices include reactive power compensators, According to Myerson[7], Game Theory (GT) is the study
generators, and tap-positions of on-load transformer changers of mathematical models of conflict and cooperation between
(OLTCs). However, left uncontrolled, reactive power injections intelligent and rational decision makers. A game comprises of
could lead to increased line loadings and losses, resulting in agents (or players) having a set of strategies which correspond
increased stress on the system. to individual payoffs/costs or rewards/utilities. The goal is to
Optimal reactive power procurement and voltage control achieve a ‘Nash Equilibrium’, such that there are no existing
methods are important issues that have been considered in incentives for players to change their strategies. Game theory is
restructured power systems [1]. Several techniques have been broadly categorized into non-cooperative and co-operative
studied for solving reactive power dispatch problems that arise games.
largely due to complexities associated with non-linearity of the A. Non-Cooperative Games
optimization problem. These include classical methods of
This model specifies the set of actions available to each
optimization, which include nonlinear programming (NLP),
agent [8]. The basic modeling unit is the agent, such that it is
successive linear programming, Newton and quadratic
only interested in choosing strategies that will maximize its own
techniques, and evolutionary algorithms such as artificial neural
payoff (or minimize its cost). Non-cooperative games can be
networks and particle swarm optimization [2]. The
further categorized into normal- and extensive-form games:
incorporation of different system operating constraints and
1) Normal or Strategic Form: In this type of model, the
control variables (discrete and continuous) in the problem
actions (or strategies) of agents are carried out simultaneously,
formulation is often a challenge that impacts both computation
without prior knowledge of the actions carried out by other
and complexity.
agents. This is represented as a payoff matrix with payoffs for
The method of game theory introduces a new concept of
all agents, for all possible combinations of their individual
players aiming to maximize their individual goals in a multi-
strategies.
player environment. This technique has found applications in
2) Extensive or Sequential Form: Agents carry out their
competitive pricing in electricity markets, formation of
actions in a timely, sequential manner, with perfect or imperfect
coalitions in micro-grid distribution networks and the
information of the actions that have been carried out by other

Authorized licensed use limited to: Dalian University of Technology. Downloaded on June 24,2020 at 12:46:25 UTC from IEEE Xplore. Restrictions apply.
agents. This model is analogous to a decision tree, having its root 5) Subgame Perfect Equilibrium (SPGE): The subgame-
at the top and the leaf nodes at the extreme bottom. These leaf perfect equilibrium (SPGE) of a game G are all strategy profiles,
nodes are associated with the payoff for a specific combination S, such that for any subgame G' in G, the restriction of S to G' is
of a strategic path. a Nash equilibrium of G' [9]. This implies that at every time in
history, a player’s preferred selected strategy is always optimal,
B. Co-operative or Coalitional Games
resulting in a Nash equilibrium.
A co-operative game begins from the set of payoffs each
group of individuals can jointly achieve, hence it is coalition-
based. The basic modeling unit is a coalition or group of agents, III. PROBLEM FORMULATION AND MODEL
which is formed only if agents, at least, derive better distributed The paper assumes that reactive power control within the
payoffs than when they act alone. The emphasis is on system is determined by the amount of reactive power injected
achievements of coalitions rather than the individual agents. into the system. Reactive power injection is then controlled by
the actions of generators and shunt-connected compensators. In
Mathematically, non-cooperative games can be modeled as addition, settings of tap-changing transformers control the flow
[9]: of reactive power flowing along their connected transmission
1) Normal form game: A finite, n-person game is a tuple lines.
defined as:
‫ ܩ‬ൌ ‫ܰۃ‬ǡ ‫ܣ‬ǡ ܷ‫ۄ‬ (1) A. Objectives
1) Reactive power control: Minimize the amount of
ܰ = Set of n number of players, indexed by i. reactive power injection during voltage control activities.
‫ ܣ‬ൌ ‫ܣ‬ଵ ൈ ǥ ‫ܣ‬௡ ; ‫ܣ‬௜ set of actions available to player i.
2) Voltage security: Ensure that load bus voltages lie
ܷ ൌ ሺܷଵ ǡ ǥ ǡ ܷ௡ ሻ; ܷ௜ ǣ ‫ ܣ‬՜  is a real-valued quantity (or within specified limits thus providing good power quality.
payoff function) for player i.
2) Extensive form game: A perfect-information game is B. Players/Agents (Control Elements)
defined as:
‫ ܩ‬ൌ ‫ܰۃ‬ǡ ‫ܣ‬ǡ ‫ܪ‬ǡ ܼǡ ߯ǡ ߩǡ ߪǡ ܷ‫ۄ‬ (2) These are the reactive power control elements of the power
system. They include generators, on-load tap changing
‫ = ܪ‬Set of non-terminal choice nodes where a player makes a transformers (OLTCs), and reactive power shunt compensating
strategic decision based on available options. devices [10], [11].
ܼ = Set of terminal nodes, disjoint from H. C. Actions
߯ǣ ‫ ܪ‬՜ ʹ஺ , action function assigning a set of possible actions to These are the different, possible settings available to
a node. players during reactive power control.
ߩǣ ‫ ܪ‬՜ ܰ, player function assigning a non-terminal node to an a. Automatic voltage regulator (AVR) settings of
agent i which then chooses an action at that node. generators (continuous variable)
ߪǣ ‫ ܪ‬ൈ ‫ ܣ‬՜ ‫ܼܷܪ‬,successor function mapping a choice node b. Tap settings of the OLTCs (discrete variable).
and an action to a new choice or terminal node. c. Amount of reactive power supplied by the
C. Terminologies compensators (discrete variable).
1) Solution Concept (SC): It represents a model’s D. Assumptions
prediction of optimality. It is the manner in which the game is As a result of the infinite number of possible control
played such that an optimal result is determined from a number settings that lie within the operating range of the generator
of different possible outcomes. SCs include Nash equilibrium, AVR, discrete-step settings have been assumed. Also, unlike
sub-game perfect equilibrium, backward induction, perfect the generator or compensator which injects/absorbs reactive
Bayesian equilibrium, rationalizability and forward induction. power, the average amount of reactive power flowing across an
2) Nash equilibrium: It specifies a player i’s strategy, OLTC branch is considered.
which is a best response to the strategy of every other player, (n-
i). There is no incentive for any player to deviate from their E. Payoffs
current strategy. It is the most commonly used solution concept. These are the reactive power injections by the generator
3) Subgame: Given a perfect-information, extensive-form and compensator (Mvar), and the average reactive power flow
game G, the subgame of G at a node h is the restriction of G to across the line along which the OLTC is connected.
the descendants of h [9]. In an extensive form game (or decision
tree), the subgame is part of the overall game. F. Payoff Vector
4) Backward Induction: It is an analysis technique used in It comprises of the reactive power contributions or costs (in
solving extensive form games. If a subgame equilibrium is Mvar) for all the power system components, given any
played at any stage of the decision tree, one can iteratively combination of players’ actions.
progress up the inverted tree until the root node and final
solution is reached.

Authorized licensed use limited to: Dalian University of Technology. Downloaded on June 24,2020 at 12:46:25 UTC from IEEE Xplore. Restrictions apply.
G. Constraints ܳ௚௞ ൌ െ σே
௡ୀଵหܸ௚௞ ܸ௡ ܻ௚௞௡ ห•‹ሺߠ௚௞௡ ൅ ߜ௡ െ ߜ௚௞ ሻ (4)
The state (bus voltages and power flows) of the system is
determined by solving the power flow equations at each of the ܸ௡ ‫ߜס‬௡ and ܻ௚௞௡ ‫ߠס‬௚௞௡ are the phasor representations of the
buses. However, the limits for device operation and allowable voltage at the ݊௧௛ bus, and the admittance matrix entry of the
bus voltages are defined as follows [2]: line connecting buses ݃݇ and ݊ respectively.

Generator reactive power: ܳ௚௠௜௡ ൑ ܳ௚ ൑ ܳ௚௠௔௫

Generator AVR settings: ܸ௚௠௜௡ ൑ ܸ௚ ൑ ܸ௚௠௔௫ 2) Tap-changing transformer: The complex power
OLTC settings: ‫ݐ‬௞௡௠௜௡ ൑ ‫ݐ‬௞௡ ൑ ‫ݐ‬௞௡௠௔௫ (3) flowing between two buses ݊ and ݇ in the directions ݊݇ is given
Var Compensators: ܳ௦௛௠௜௡ ൑ ܳ௦௛ ൑ ܳ௦௛௠௔௫ as ܵ௡௞ ൌ ܸ௡ ݅௡௞ ǡsuch that
Load bus voltage: ܸ௟௠௜௡ ൑ ܸ௟ ൑ ܸ௟௠௔௫
௒೙ೖ ௒೙ೖ
݅௡௞ ൌ ሺܸ௡ െ ܸ௞ ሻ ൅ ܸ௡ ሺ‫ ݐ‬െ ͳሻ (5)
௧ ௧
ܳ௚ is the generator reactive power input; ܸ௚ is the AVR setting
of the generator; ‫ݐ‬௞௡ is the transformer tap setting; ܳ௦௛ is the ‫ = ݐ‬tap ratio.
value of the shunt reactive power compensator; ܸ௟ is the load bus
voltage. The average amount of reactive power flowing across the line is
H. Power System Model then computed as:
ொ ାொ
The IEEE 6-bus test system model has been used for this ܳ௡௞ǡ௔௩௚ ൌ ೙ೖ ೖ೙ Ǣܳ௡௞ ൌ ܵ௡௞ ‫ߜ݊݅ݏ‬௡௞ (6)
ଶ
work, and is illustrated in Figure 1. The system line and bus data
are presented in [1], and the operating limits of the control 3) Shunt reactive compensator: The amount of reactive
elements are given in Table I. power ሺܳ௦௛ ሻ injections is in discrete steps based on a given step
size. Within a range of control ሺܳ௦௛௠௜௡ ǡ ܳ௦௛௠௔௫ ሻ, the
compensator is able to support the system.
J. Payoffs
In this paper, the interaction among the power system
control elements is modeled using a perfect information, non-
cooperative, extensive form game model as shown in Figure 2.

Figure 1. IEEE 6-bus system

TABLE I. OPERATING LIMITS OF REACTIVE POWER COMPONENTS

Control Lower Upper Initial

Component
Variable Limit Limit Value
Generator Voltage ܸ௚ଵ 1.05 1.10 1.050
(pu) ܸ௚ଶ 1.10 1.15 1.100
Compensator VAR ܳସ 0.00 5.00 0.000
Injection (pu) ܳ଺ 0.00 5.50 0.000
‫ݐ‬ସଷ 0.90 1.10 1.100
Tap Changer
‫଺ݐ‬ହ 0.90 1.10 1.025

I. Reactive power contributions and non-cooperative

extensive form game model
The reactive power contributions of each component player
in the system are as follows:
1) Generator: At a generator bus ݃݇, the reactive power
Figure 2. Non-cooperative extensive form game model
injection is given as:

Authorized licensed use limited to: Dalian University of Technology. Downloaded on June 24,2020 at 12:46:25 UTC from IEEE Xplore. Restrictions apply.
 Two each of the generators, tap-changing transformers and IV. RESULTS/DISCUSSIONS
reactive power compensators constitute the six players involved MATLAB was used to carry out the power flow
in game. The mathematical representation is given in equation computation and the backward induction process of
(2). This model ensures that every combination of players’ Algorithm1. The preliminary results obtained are presented and
actions is considered in a separate power flow computation, and discussed in this section.
indicated as a strategy profile or path. Game theory game 1 (GT1) models the extensive-form
ܵ௡ҧ is the payoff vector of the nth possible strategy path, and game interactions among players using a top-down hierarchical
located at the leaf node. It comprises of the reactive power order of generator-transformer-compensator actions as
contributions of all six players if that nth path was chosen. illustrated in Figure 2. In setting up the game, the acceptable
load voltage is set to be within 0.98 -1.02 p.u.
ܵ௡ҧ ൌ ሺܳ௚ଵ௡ ǡ ܳ௚ଶ௡ ǡ ܳ௧ସଷ௡ ǡ ܳ௧଺ହ௡ ǡ ܳ௖ସ௡ ǡ ܳ௖଺௡ ሻ (12) A total of 41,616 control combinations and possible
solutions are obtained for the different settings of the voltage-
ܳ௚ଵ௡ ǡ ܳ௚ଶ௡ : nth possible reactive power injection payoffs by control devices, after which the backward induction algorithm
generators 1 and 2. is applied to only those combinations whose load voltages
ܳ௧ସଷ௡ ǡ ܳ௧଺ହ௡ : nth possible reactive power flowing along lines satisfy the required limits. These are the feasible solutions.
4-3 and 6-5. GT1: Beginning with the first player in the backward
ܳ௖ସ௡ ǡ ܳ௖଺௡ : nth possible reactive power injections by induction algorithm, compensator 6 (player 6) attempts to
compensators at buses 4 and 6. maximize its payoff at all the different subgames. This is
achieved by choosing leaf nodes whose settings minimize its
Algorithm 1 describes the steps used to select the feasible reactive power contribution to the system. The other leaf nodes
combination settings which satisfy the voltage requirements, with higher reactive power requirements are removed from the
and the backward induction used to identify all component game before control is passed to compensator 4 (player 5).
player optimal settings (ࡿࢋ࢚࢕࢖࢚) at the Nash equilibrium point. Similarly, player 5 maximizes its payoff at each of the higher-
level subgame. Iteratively, the payoffs of other system players
are maximized at successive, higher-level subgames until the
Algorithm1: Reactive Power Game Control
final decision node is reached. Table II shows the final 2
strategy profiles at the decision node level for player 1
Input: N-players (ܲଵ ǡ ǥ ǡ ܲே ሻ with ሺܽǡ ܾǡ ܿ ǥ ሻ number of actions
(Generator 1).
respectively;
Define constraints (3);
Both voltage settings of Generator 1, 1.05 or 1.1 p.u.,
Define ‫ ݓ‬load bus voltages i.e. ࢂࡸǡࡹࡻࡺ ൌ ሼܸ௟ଵ ǡ ǥ ǡ ܸ௟௪ ሽ , from either of the two strategy profiles 34596 and 45504 satisfy
Define component settings, ࡿ࢚࢚ࢍ࢙ ൌ ሼܸ௚ଵ ǡ ܸ௚ଶ ǡ ‫ݐ‬ସଷ ǡ ‫଺ݐ‬ହ ǡ ܳ௦ସ ǡ ܳ௦଺ ሽ the voltage requirements at all load buses. However, the option
Output: Optimal settings, ࡿࢋ࢚࢕࢖࢚ to select the voltage set point of 1.05 p.u (rather than 1.1 p.u) is
contingent on the smaller amount of reactive power injection
Invalidate combination settings with unacceptable voltages by generator 1 into the system (20.93 Mvars). Player 1 utilizes
1. For ݅ ൌ ͳǣ ‫( ܭ‬Total combinations - ሺܽ ൈ ܾ ൈ ܿ ǥ ሻ) a minimal amount of reactive power to provide satisfactory
2. Run power flow program; voltage levels at all buses, while still ensuring a higher reactive
3. Prune: If all ‫ ݓ‬load bus voltages satisfy voltage limits in (3) power in its reserve. Thus, strategy profile 34596 constitutes the
4. Save: i , ࢂࡸǡࡹࡻࡺ , ܵ௡ҧ , ࡿ࢚࢚ࢍ࢙ Nash equilibrium strategy for the game.
5. Else set:ܵ௜ҧ ൌ ͻͻͻͻ ‫ݏ݁݊݋ כ‬ሺͳǡ ܰሻ Arbitrary large Q’s
6. Proceed to i +1 TABLE II. STRATEGY PROFILES FOR PLAYER 1 (GENERATOR 1)
7. End
Execute hierarchical GT Strategy profiles 34596 45504
8. For ݅ ൌ ܰǣ െͳǣ ͳ (Reverse order) Voltage setting gen 2 (p.u.) 1.05 1.10
9. ̷‫ܪ‬ሺܲ௜ ሻ (Non-terminal choice node)
Tap setting of OLTC 4-3 0.9625 0.975
10. Total sub-trees at player i, ݅݀‫ ݔ‬ൌ ̴݅‫݈ܽݐ݋ݐ‬ൗܴሺ݅ሻ
Tap setting of OLTC 6-5 1.0000 0.9625
11. For ݆ ൌ ͳ ‫( ݔ݀݅ ׷‬index of sub-tree)
12. ܳ௜௠௜௡ ൌ ‹ሺܵ௝ҧ ሼܲ݅ሽ) Vars from compensator 4 (p.u.) 5.0 5.0
13. If ܳ௜ ൑ ݉݅݊݅ then Vars from compensator 6 (p.u.) 5.5 5.5
14. Update: ݉݅݊݅ ՚ ܳ௜௠௜௡
15. Subtree assignment: 1. Update: ܵ௝ҧ ሼܲ௜ ሽ ՚ ݉݅݊݅ Voltage setting gen 1 (p.u.) 1.05 1.10
16. 2. Save: i , ࢂࡸǡࡹࡻࡺ , ࡿ࢚࢚ࢍ࢙ Vars from gen 1 (Mvar) 20.93 41.91
17. Update: subtree as new terminal node, ܵ௝ҧ
18. End
GT2: In addition, another game was considered, however
19. End
20. End with a reversed hierarchy of control i.e. compensator-
21. ࡿࢋ࢚࢕࢖࢚= ࡿ࢚࢚ࢍ࢙ሼܵଵҧ ሽܿ‫ܲ݋ݐ ݃݊݅݀݊݋݌ݏ݁ݎݎ݋‬ଵ  transformer-generator. Table III shows comparison of all bus
voltages obtained for the base case, and equilibrium settings for

Authorized licensed use limited to: Dalian University of Technology. Downloaded on June 24,2020 at 12:46:25 UTC from IEEE Xplore. Restrictions apply.
GT1 and GT2. Table IV is the reactive power injections from in future contingencies. Also, by prioritizing the sources of
the generators and compensators. reactive power, we can allocate fast-acting devices to areas in
need of emergency control, while still maintaining acceptable
TABLE III. VOLTAGE PROFILES (P.U.) voltage profiles.
Buses 1 2 3 4 5 6 The optimal device settings obtained from this game is
Base 1.05 1.10 0.871 0.966 0.941 0.957 largely dependent on the game objective. In future work, the
GT1 1.05 1.15 1.006 0.980 0.989 0.982 algorithm computation time is improved by utilizing only
GT2 1.1 1.15 1.020 0.996 1.018 0.983 critical players in the game and employing the use of
approximate power flows. Also, the implementation of the
TABLE IV. REACTIVE POWER INJECTIONS (MVARS) game-theoretic method in a distributed framework is currently
being tested on an IEEE 39-bus and other larger systems.
Case Gen 1 Gen 2 Q4 Q6 Total
Base 29.617 38.233 0.0 0.0 67.85 ACKNOWLEDGMENT
GT1 20.92 32.48 5.0 5.5 63.90 This research was supported by the Center for Ultra-Wide-
GT2 41.68 21.21 0.0 0.0 62.89 Area Resilient Electric Energy Transmission Networks.
(CURENT).
As observed, the injection of arbitrary large amounts of
reactive power into the system in the base case (67.85 Mvars) REFERENCES
does not guarantee improved voltage profiles, thus invalidating [1] N. K. Sharma, D. Suresh Babu, and S. Choube, “Application of particle
the base case component settings. However, controlled reactive swarm optimization technique for reactive power optimization,” in
power injections, using the game theoretical approach, ensures Advances in Engineering, Science and Management (ICAESM), 2012
International Conference on. IEEE, 2012, pp. 88–93.
that only component settings which result in acceptable voltage [2] J. G. Vlachogiannis and K. Y. Lee, “A comparative study on particle
profiles are considered. swarm optimization for optimal steady-state performance of power
Reactive power control is also achieved as implementing systems,” Power Systems, IEEE Transactions on, vol. 21, no. 4, pp.
GT1 ensures reactive support is shared by the compensators, 1718–1728, 2006.
[3] B. F. Hobbs, C. B. Metzler, and J.-S. Pang, “Strategic gaming analysis
thus reducing the demand from both generators. In the case of for electric power systems: An mpec approach,” Power Systems, IEEE
GT2, reactive support is re-distributed between both generators. Transactions on, vol. 15, no. 2, pp. 638–645, 2000.
Component settings from both GT1 and GT2 result in the [4] M. Saguan, N. Keseric, P. Dessante, and J.-M. Glachant, “Market power
injection of lower amounts of reactive power into the system in power markets: Game theory vs. agent-based approach,” in
Transmission & Distribution Conference and Exposition
than the base case. In addition, these settings are able to control [5] T. Krause, E. V. Beck, R. Cherkaoui, A. Germond, G. Andersson, and
load bus voltages within the required limits. D. Ernst, “A comparison of nash equilibria analysis and agent-based
The implementation of the above game-theoretic methods modelling for power markets,” International Journal of Electrical
results in a systematic reduction and control of the reactive Power & Energy Systems, vol. 28, no. 9, pp. 599–607, 2006.
[6] M. Saguan, N. Keseric, P. Dessante, and J.-M. Glachant, “Market power
injections into the system. Transmission lines are not unduly in power markets: Game theory vs. agent-based approach,” in
overloaded, and the system losses are reduced. Simultaneously, Transmission & Distribution Conference and Exposition: Latin
increase in the reactive power reserves ensures that in the event America, 2006. TDC’06. IEEE/PES. IEEE, 2006, pp. 1–6.
of future contingencies, voltage-control components can [7] Myerson, R., 1997. Game Theory: Analysis of Conflict, Cambridge:
Harvard University Press.
provide support to the system state. [8] Osborne, M.J. & Rubinstein, A., 2011. A Course in Game Theory,
Cambridge: MIT Press
V. CONCLUSION [9] Leyton-Brown, K. & Shoham, Y., 2008. Essentials of Game Theory: A
In this work, a non-cooperative, extensive form game was Concise Multidisciplinary Introduction, California: Morgan and
Claypool.
applied to model a simple and systematic interaction among [10] G. Bakare, G. Venayagamoorthy, and U. Aliyu, “Reactive power and
power system components, with an objective of controlling voltage control of the nigerian grid system using micro-genetic
reactive power injections and voltage. The method was algorithm,” in Power Engineering Society General Meeting, 2005.
implemented on an IEEE 6-bus system, with reactive power IEEE. IEEE, 2005, pp. 1916–1922.
[11] Kundur, P., 1995. Power Systems Stability and Control Edited., New
control elements chosen as players of the game. Preliminary York: Mc-Graw Hill.
results support the goal of reactive power conservation for use

Authorized licensed use limited to: Dalian University of Technology. Downloaded on June 24,2020 at 12:46:25 UTC from IEEE Xplore. Restrictions apply.