Proceedings of the 2nd International Conference on Computing Innovation and Applied Physics (CONF-CIAP 2023)

DOI: 10.54254/2753-8818/5/20230545

A review of reactive power optimization algorithm in power

system

Yijun Ma
School of intelligent manufacturing, Nanyang Institute Of Technology, 80 Changjiang
Road, Wancheng District, Nanyang City, Henan Province, China

2500531775@qq.com

Abstract. The paper analyses the significance of reactive power optimization, generalizes the
current situation of power system development. Various optimization algorithms were
introduced in this paper such as traditional optimization algorithm, intelligence optimization
algorithm, including the methods of linear programming, Newton’s method, heuristic
optimization algorithm, etc. This research analyzes the advantages and disadvantages of each
algorithm and its application direction by comparing their outstanding performance in solving
discrete variables and continuous variables. The purpose of the research is to find the optimal
solution of reactive power optimization algorithm, minimize the transport network loss of
power system, and improve the quality of users.

Keywords: power system, reactive power optimization, traditional algorithms, intelligence

optimization algorithm

1. Introduction
As the economy grows, so does the need for electricity, prompting the government to focus more on
improving the country's electrical infrastructure. If reactive power is allocated and dispatched
efficiently, not only can the voltage level be maintained and the stability of power system operation be
enhanced, but active power loss and reactive power loss may also be minimized. It is also a crucial
resource for helping dispatchers set up the appropriate operating mode and prepare for optimal
reactive power usage. Power savings, higher voltage quality, and more stable power grid operation are
just a few of the many tangible benefits of reactive power management. As far as the current situation
is concerned, there are generally insufficient reactive power or unreasonable reactive power
distribution in the power distribution system. What we need to do is to find the best algorithm for
rational distribution of reactive power. The research about reactive power optimization could decrease
the incidence of system breakdown and other accidents, proving that power system could operate
more efficiently and economically [1]. The research of the algorithm is to better solve the discrete
variables and continuous variables. The paper will analyze a variety of algorithms to solve the
problem of reactive power optimization. Different algorithms can be selected to solve different
problems.

2. The definition and significance of reactive power optimization

© 2023 The Authors. This is an open access article distributed under the terms of the Creative Commons Attribution License 4.0
(https://creativecommons.org/licenses/by/4.0/).

888
 Proceedings of the 2nd International Conference on Computing Innovation and Applied Physics (CONF-CIAP 2023)
DOI: 10.54254/2753-8818/5/20230545

When the system's structural parameters and load conditions are known, reactive power optimization
is the process of adjusting the system's reactive power by optimizing a subset of its control variables,
with the goal of achieving the system's optimal performance on one or more performance indicators.
Whether or not the reactive power distribution of the system reasonably and directly affects the safety
and stability of the power system, reactive power optimization is of major significance. Analyzing
from the user's perspective, the most basic purpose of power system is to provide electricity to the
user, the transmission error will have a great impact on the people's lives if it happens, it is necessary
to ensure the normal operation of the power grid. Reactive power optimization can keep the voltage
stable through the reactive power flow distribution. From the perspective of power transmission
companies, reactive power optimization can effectively reduce network loss, improve transmission
efficiency, reduce transportation costs, and help power transmission companies obtain more benefits.
Therefore, the research on reactive power optimization has the great practical significance.

3. Introduction to reactive power optimization

The reactive power optimization of power system is a complex mixed nonlinear programming
problem mathematically, which contains multiple variables and constraints, and the control variables
include two kinds, namely, discrete variables and continuous variables, and it is also a multi-peak
function models. In reactive power optimization, the model used will be different depending on the
solution target. The article will introduce several typical power system reactive power optimization
models.

3.1. Reactive power optimization model considering network loss

Objective function：
𝑛𝑙

𝑚𝑖𝑛 𝐹 = 𝑚𝑖𝑛 ∑ 𝐺𝑗𝑦 [𝑢𝑖2 + 𝑢𝑗2 − 2𝑢𝑖 𝑢𝑗 𝑐𝑜𝑠(𝛿𝑗 − 𝛿𝑖 )]#(1)

𝑘=1

Subject to:
𝑃𝑖 − 𝑈𝑖 ∑𝑛𝑗=1 𝑈𝑗 (𝐺𝑖𝑗 cos 𝛿𝑖𝑗 + 𝐵𝑖𝑗 sin 𝛿𝑖𝑗 ) = 0
𝑄𝑖 − 𝑈𝑖 ∑𝑛𝑗=1 𝑈𝑗 (𝐺𝑖𝑗 sin 𝛿𝑖𝑗 + 𝐵𝑖𝑗 cos 𝛿𝑖𝑗 ) = 0
𝑢𝑖 𝑚𝑖𝑛 ≤ 𝑢𝑖 ≤ 𝑢𝑖 𝑚𝑎𝑥 (2)
𝑄𝐺𝑖 𝑚𝑖𝑛 ≤ 𝑄𝐺𝑖 ≤ 𝑄𝐺𝑖 𝑚𝑎𝑥
𝑄𝐶𝑖 𝑚𝑖𝑛 ≤ 𝑄𝐶𝑖 ≤ 𝑄𝐶𝑖 𝑚𝑎𝑥
{ 𝑇𝑖 𝑚𝑖𝑛 ≤ 𝑇𝑖 ≤ 𝑇𝑖 𝑚𝑎𝑥
In (2): nl means total number of legs, u is voltage, Gjy is conductance, Q Gi is reactive power output
of generator, Q Ci is reactive power compensation capacity of reactive power compensation device,
Ti is the i-th on-load voltage regulating transformer ratio.

3.2. Reactive power optimization model with minimum difference between node voltage and specified
value
Objective function:
𝑛
|𝑈𝑗 − 𝑈𝑗𝑠𝑝𝑒𝑐 |
min 𝐹 = min ∑ (3)
∆𝑈𝑗𝑠𝑝𝑒𝑐
𝑘=1

In (3): N is the total number of nodes(except for balance nodes), Ujspec and ∆Ujspec represent the
given value of node voltage and its maximum offset.

3.3. Summary of reactive power optimization

In the literature [2],we know that the reactive power optimization of power system is a complex
mixed nonlinear programming problem mathematically, which contains multiple variables and

889
 Proceedings of the 2nd International Conference on Computing Innovation and Applied Physics (CONF-CIAP 2023)
DOI: 10.54254/2753-8818/5/20230545

constraints, and the control variables include two kinds, namely discrete variables and continuous
variables, and it is also a multi-peak function. There are different methods for solving reactive power
optimization problems, and the scope of application of each method is different, so it is necessary to
use appropriate methods to get the optimal solution. In general, there are two optimization algorithms,
which are traditional algorithm, and intelligent optimization algorithm.

4. Traditional algorithm
Typically, the focus of traditional optimization techniques is on structured problems with well-defined
issue and condition descriptions, as well as restrictions.

4.1. Linear programming method

The reactive power optimization problem can be linearly approximated by applying the principle of
linear programming, which entails using the Taylor formula to expand the objective function and
constraint conditions, respectively, while ignoring the higher order term and keeping only the constant
term and the first order term. The issue of reactive power optimization has been addressed with this
strategy in the past [3]. The simplified gradient approach is advantageous for solving the large-scale
optimal power flow problem because the algorithm is straightforward, the beginning value is
unnecessary, it is straightforward to implement, and it is convenient to solve. The disadvantage of this
method is that it is easy to appear sawtooth phenomenon, the convergence performance is poor, and
the closer to the optimal point, the slower the convergence, and the huge amount of calculation. In
reference [4], an algorithm is used to modify the step size, and the convergence characteristics are
checked every time the optimization is carried out. Another method is called interior point
method.The idea behind this technique is to begin in the feasible zone and advance in the direction
that will lead to the fastest possible decrease in the objective function, all the while searching for the
best possible answer. The optimization process of this algorithm is always carried out in the feasible
region. Affine scaling method is a kind of interior point method that is widely used and developed.
The original affine scaling method and the dual affine scaling method are the most frequent, and the
original dual affine scaling interior point method (also known as the path following method) is the
most promising. Good robustness and convergence are hallmarks of the path following approach, and
the calculation time is relatively insensitive to changes in system size[5]. The path following method
was successful in solving the document [6]. Optimization of reactive power as a linear programming
problem. The experimental results show that the algorithm has polynomial time complexity, stable
iteration convergence times, and is not significantly affected by the system size. Therefore, when the
system size is relatively large, the algorithm has advantages over the exponential time complexity
simplex method. The optimization path, however, must adhere to the primal dual path, and the starting
iteration point must be within the feasible region. The optimization procedure will also be influenced
by the step size of the control variable. By enhancing the algorithm in document [6], the initial point is
no longer constrained to the feasible region, and the optimization process is not restricted to the
primal dual path; yet, the algorithm is still able to converge to the optimal solution, and its
convergence performance is stable. Since linear programming has been developed to a pretty high
standard, its physical notion is well understood, its mathematical model is straightforward, its
calculation time is short, and its calculation scale is effectively infinite. But the true mixed non-
linearity is approximatively linearized by this procedure [7].

4.2. Newton’s method

Convergence is faster and the search direction is more natural in the Newton method because it uses
the Jacobian matrix generated by the first derivative of the system power flow equation and the
Hessian matrix formed by the second derivative of the objective function. The disadvantage of
Newton method is that if the deviation of initial value selection is too large, the iterative direction will
be wrong and the optimal solution cannot be obtained. Newton's method does not guarantee that the
function value will decline at each iteration, nor does it guarantee certain convergence. Therefore,

890
 Proceedings of the 2nd International Conference on Computing Innovation and Applied Physics (CONF-CIAP 2023)
DOI: 10.54254/2753-8818/5/20230545

some remedial measures are proposed, one of which is the line search technology, that is, searching
the optimal step size. The specific method is to select some typical discrete values, such as 0.0001,
0.001, 0.01, etc., and compare which value will decrease the function value the fastest as the optimal
step size. Compared with gradient descent method, Newton method has faster convergence speed, but
the cost of each iteration is also higher. In each iteration, in addition to the gradient vector, the
Hessian matrix is also calculated, and the inverse matrix of the Hessian matrix is solved. At the same
time, it takes a lot of time to solve the Hessian matrix, and the conjugate gradient method can solve
this problem, but it only shows convergence in the region where the quadratic nature of the objective
function is obvious, so the use of this method has limitations [8].

4.3. Dynamic programming

The principle of the dynamic programming method is to divide the problem into stages where Hong
Kong people are connected according to the time or space order. First, analyze each stage separately,
and then obtain the overall optimal solution through balance and coordination. Compared with linear
programming and nonlinear programming, the objective function and constraints of dynamic
programming can often obtain the global optimal solution with good convergence even if they do not
strictly observe the linearity and convexity. The idea of this method is clear, but if there are too many
variables, the modeling will become complicated, the calculation speed will be slow, and there will be
"dimension disaster", which restricts its application in practical projects. In addition, as long as the
"time period" factor is artificially introduced, the dynamic programming method can also deal with
some static optimization problems that are not related to time [9].
From the analysis of the above algorithms, we can see that each traditional optimization algorithm has
the same defect in varying degrees: it requires high accuracy of the mathematical model, but the more
accurate the model is, the more complex the corresponding is, which is not conducive to practical
application, cannot solve the discrete variable problem well, and requires high initial values, "curse of
dimensionality". To solve these problems, various intelligent algorithms have emerged as the times
require.

5. Intelligence optimization algorithm

A heuristic algorithm is a type of intelligent optimization method. To identify the optimal solution to
an optimization problem, this is a crucial piece of knowledge to have as soon as possible [10]. This
method is a strategy that ignores some limitations to achieve the goal, and makes a fast response with
the highest precision. Compared with the complex method, it can save the most time [11]. Heuristic
algorithm is an algorithm developed by people inspired by nature.

5.1. Genetic algorithm

Genetic algorithm is an intelligent algorithm that simulates the evolution and mutation of organisms.
The specific method is as follows: first, a series of initial solutions are randomly generated, then each
group is decomposed into a "chromosome", and several chromosomes form a group, and after
comparing with the preset target value, the maximum value is selected to replicate, and the next
generation of chromosomes is generated by cross-mutation. The next generation of chromosomes will
inherit the excellent genes of the previous generation and continue to develop in the optimal direction.
The advantage of this method is that it is robust, it can find the global optimal solution with the
maximum probability, and there is no problem of dimension disaster. Its shortcomings are also
obvious. Too many iterations will lead to slow optimization, which makes it difficult to optimize
large-scale power systems. In literature[12], an improved immune genetic algorithm is proposed,
which can enhance the local optimization ability and improve the convergence speed and accuracy of
the algorithm.

891
 Proceedings of the 2nd International Conference on Computing Innovation and Applied Physics (CONF-CIAP 2023)
DOI: 10.54254/2753-8818/5/20230545

5.2. Annealing algorithm

Using the same concept as metal annealing, the annealing algorithm is a heuristic algorithm that may
be used to quickly and efficiently identify the global optimal solution. The algorithm's guiding idea is,
first, precise temperature regulation of the object, second, a rough search of the range, and third, a
precise search of the immediate vicinity. The theory behind this approach is straightforward, which
makes it easy to put into practice. However, because it has no memory function, there will be local
repeated search, and the larger the scale of the system, the more complex the algorithm, which will
lead to slower computing speed[13]. Combined with the actual operation characteristics of the power
system and the properties of the simulated annealing algorithm, the method in Ref. [14] improves the
original simulated annealing algorithm and improves the search efficiency and optimization speed. A
method is introduced in Ref.[15], which combines the memory-guided search method with the pattern
method for local optimization. The pattern method first optimizes each link, and finally selects the
optimal value of all the search results as the output result. This algorithm not only avoids repeated
optimization to some extent, but also makes up for the weakness of random search method with poor
local optimization ability.

5.3. Particle swarm optimization algorithm

The optimization technique known as particle swarms was first conceived as a way to simulate the
way birds hunt for food. By modeling the social system and creating a particle swarm in a high-
dimensional space, we can ensure that each individual particle is always moving in the right direction
at the right speed and following the best course of action for the swarm as a whole[16]. The basic
iterative equation is [17]:
Vi (k + 1) = ωVi (k) + c1 rand( )(X pid − X id ) + c2 rand( )(X gd − X id ) (4)
X i (k + 1) = X i (k)+Vi (k + 1) (5)
In(4): Vi =(vi1 + vi2 , … , viN ) is the random velocity of particles, X p = (xp1 + xp2 + xp3 , … , xpN ) is
the optimal position of a particle, X g = (xg1 + xg2 + xg3 , … , xgN ) is the optimal position of particle
swarm, c1 and c2 are acceleration factors, ω is the inertia factor, rand() represents a random number
between 0 and 1, where the speed of the example is less than the latter equal to the maximum speed of
the particle.
This algorithm lacks mutation and crossover, yet each particle follows the best particle to seek.
This approach has good convergence, rapid computation speed, and no dimension disaster, thus it can
find the global optimal solution. This algorithm easily finds the local optimal solution. The particle
swarm optimization approach enhances calculation accuracy, convergence stability, and calculation
time [18].

6. Conclusion
First of all, this paper introduces the general reactive power optimization model. Secondly, the
principles, advantages and disadvantages of various basic optimization algorithms are summarized
and analyzed, and some methods to improve the algorithm are listed. However, the convergence rate,
the accuracy of the results and the final optimization effect still can not be fully taken into account.
With the continuous exploration of the algorithm, the system model can be made more accurate, the
active power loss can be minimized, and the transmission efficiency and power quality can be
improved. In the future, reactive power optimization research, a series of considerations should be
carried out in combination with the emerging problems, so that the reactive power optimization
algorithm can keep pace with the times and more and more adapt to the development of power
system.

References
[1] Xiong Yuhui, Reactive Power Optimization of Power System [D]; Nan Chang University.2005

892
 Proceedings of the 2nd International Conference on Computing Innovation and Applied Physics (CONF-CIAP 2023)
DOI: 10.54254/2753-8818/5/20230545

[2] Chen Laiying, Lin Yu, Research on Intelligent algorithm for reactive Power Optimization of
Power system,Adhesion,2020
[3] Hoboson E,et al. Linear programming for power system real time control calculation[C],IFAS
Conference Melbourne,1977
[4] Albert M． Sasson, Combined use of the parallel and fletcher-powell, non-linear for optimal
load flows [J] IEEE Transactions on Power ,Apparatus and Systems, 1969, 10: 1530-1537
[5] Jabr R A,Coonick A H, Cory B J． A primal-dual interior Point Method for Power Flow
Dispatching [J] IEEE Transactions on Power System, 2002,17( 3) : 654-662．
[6] Liu MingBo, Chen Xuejun. Prime-dual affine scaling interior point method based reactive
power optimization in power systems[J]. Power System Technology, 1998, 22( 3): 21-28
[7] Liu MingBo, Chen Xuejun. Improved interior point method for reactive power optimization in
power systems [J] Automation of Electric Power Systems, 1998, 22( 5): 33-36．
[8] Andreas Rieder, Inexact Newton Regularization Using Conjugate Gradients as Inner Iteration,
SIAM Journal on Numerical Analysis, 2005
[9] Tang Huanwen, Qin Xuezhi. Practical Optimization Method[M]. Dalian: Dalian University of
Technology Press, 1994
[10] M. Dehghani, M. Mardaneh, O. P. Malik , Following Optimization Algorithm for Solving
Power Engineering Optimization Problems, Journal of Operation and Automation in Power
Engineering,pp 57-64,2020.
[11] A. Lazar, ”Heuristic knowledge discovery for archaeological data using genetic algorithms
and rough sets, ”Heuristics Optim. Knowl. Discovery,pp. 263-278, 2002.
[12] Guo Chuangxin, Zhu Chengzhi, Zhao Bo, et al. Power system reactive power optimization
based on an improved immune algorithms [J], Automation of Electric Power System
,2005,29(15):23-28
[13] Peter J. M. van Laarhoven , Emile H. L. Aarts ,Simulated annealing, Simulated Annealing:
Theory and Applications, pp 7–15,1987
[14] Zhou Xiaojuan, Zhou Hui, The improved SA algorithm in VAR and voltage control of power
system[I]. Relay,2004,32(1):24-27
[15] Jia Dexiang, Tang Guoqing, Han Jing, Reactive power optimization of power system based on
modified simulated annealing algorithm[J], Relay, 2004, 32(4):32-35
[16] Eberhart R ， Kennedy J.A new optimizer using particle swarm theory[C]//
MHS'95.Proceedings of the Sixth,International Symposium on Micro Machine and
Human,Science.IEEE, 2002
[17] Shi Y.A Modified Particle Swarm Optimizer[C]// Proc.of IEEE ICEC conference ，
Anchorage.1998.
[18] Liu Shukui, Reactive power optimization of power system based on adaptive focused particle
swarm optimization[D],Cheng Du: Southwest Jiaotong University.2009

893