1. Introduction
In recent years, the continuous operation of large generator sets and the large-scale use of high-gain and fast excitation systems have improved the transient level of the system, but also caused the problem of low frequency oscillation of the power system due to insufficient damping, which has increasingly become an important factor restricting the overall interconnection of the power grid [
1,
2,
3]. A power system stabilizer (PSS) can not only offset the negative damping torque generated by the regulator, but also provide additional positive damping for the system, which is the most efficient and economical measure for suppressing the low-frequency oscillation of the power system [
4,
5]. Especially in the context of a large power grid, PSS has been widely used. The configuration of PSS, that is, the research theory of the installation site, has been perfected, and another key issue is the parameter coordination, which is also one of the hot topics discussed by scholars. It has been shown that the damping of local and interval oscillation modes can be optimized to a certain extent if the PSS can obtain appropriate parameters [
6].
In control theory, the phase compensation approach can be used to design a power system stabilizer, but for multiple power system stabilizers in the system, the traditional phase compensation method usually carries out parameter setting sequentially, which will inevitably cause the interaction effect between each stabilizer [
7]. To overcome this phenomenon, some studies have pointed out that the parameter tuning of PSS is equivalent to an optimal problem based on the eigenvalues of electromechanical modes, that is, to move the eigenvalues to the left half plane of the complex plane as much as possible. In recent years, using the global optimization ability of artificial intelligence algorithms to solve this phenomenon has provided a new idea for the coordination design of a controller and has become a hot research topic. Particle swarm optimization (PSO), a basic and powerful approach that many researchers have studied and applied to power systems, is a commonly used optimization algorithm [
8]. In Ref. [
9], PSS parameters were optimized based on a PSO algorithm. Although the optimization principle is simple, the optimized particles in the optimization iteration process are prone to precocity and easily fall into local optimum, which is also the common fault of most algorithms. For this reason, Ref. [
10] proposed an improved PSO algorithm based on particle swarm theory. Although the performance of the algorithm has been improved to a certain extent, the particles do not completely tend to the global optimal value, but tend to a relatively good value. Ref. [
11] proposed the gray wolf optimization algorithm to optimize PSS parameters, which has fewer algorithm parameters and faster convergence speed. However, in the later optimization period, with the decrease of population number, the optimization rate decreases, and at the same time, it is possible to fall into local optimal. Therefore, Ref. [
12] proposed an improved gray wolf optimization algorithm, which overcomes the problem that it is easy to fall into local optimum and strengthens the robustness of the algorithm. However, the optimization algorithm search process tends to be complicated, so the reliability of the optimization result needs to be further improved.
In this paper, a method combining the quasi-affine transformation evolutionary algorithm [
13] and the simulated annealing algorithm is proposed to coordinate the optimization design of stabilizer parameters of multi-machine power system. The main contribution of the paper lies in the following:
The proposed SA-QUATRE algorithm not only overcomes the defect of slow convergence of particle cluster optimization algorithm, it also has stronger cooperation and reduces time complexity, and accepts inferior solutions with a certain probability to avoid falling into local minimum in the process of searching.
To overcome the interaction between power system stabilizers, in this paper, the parameter coordination tuning problem of the power system stabilizer is transformed into the optimal problem based on the characteristic value of oscillation mode and damping coefficient, and the dynamic stability of the power system is improved.
The validity and feasibility of the proposed method are verified by simulation with a test function and the IEEE 4-machine 11-node system. It can be found that, compared with the existing multi-machine power system with PSS parameters, the multi-machine and multi-node power systems with the SA-QUATRE algorithm designed and optimized under stabilizer suppression has better system dynamic performance.
The rest of the paper is organized as follows:
Section 2 introduces PSS structure and each link, as well as its transfer function, objective function and constraints of the parameter optimization of PSS for the multi-machine system.
Section 3 introduces the proposed SA-QUATRE algorithm in detail and PSS parameter optimization based on it.
Section 4 discusses the simulation example and the experiment results.
Section 5 presents the final conclusion.
2. Parameter Optimization of PSS for Multi-Machine System
Voltage stabilizers suppress oscillations by adding a signal to the excitation system that produces a positive damping torque to offset the negative damping torque produced by the Voltage regulator [
14]. The principle and excitation systems are shown in
Figure 1, where, taking the rotor angle deviation
of the generator as the input signal, the PSS and excitation system transfer function can be expressed as [
15]:
where the subscript
i represents the generator
i;
is the amplification gain,
is the high-pass filter time constant, usually set as 5 s or 10 s; in this paper,
. Two lead-lag compensators are used to eliminate the delay between excitation and electromagnetic torque. In practical applications, the two lead-lag compensators can compensate the low frequency and high frequency phases respectively. After each input signal passes through PSS, its output signal can provide the corresponding reference voltage for the excitation system, and its reference voltage serves as the reference modulation signal provided to the excitation system, so that the negative damping or weak damping caused by the fast excitation system can be compensated accordingly [
16,
17].
PSS plays an effective role mainly by adjusting and the inertia time constant of the lead-lag compensators, and the other parameters for a generator are usually unchanged. Therefore, it is necessary to tune parameters reasonably, otherwise the effect of suppressing low frequency oscillation will be counterproductive.
In general, a linear differential equation can be used to describe the dynamic behavior of a power system:
where the column vector
is the state vector,
is the external input vector, and the derivative of the state variable to time is represented by
. When the variables of the system remain constant, the system is in equilibrium. The nonlinear system equation can then be linearized at the equilibrium point [
18], where the equation is as follows:
where
is the state vector,
is the output vector,
is the input vector,
A is the state matrix,
B is the control matrix,
C is the output matrix,
D is the feedforward matrix. For the power system,
is not a direct function of
, that is
. The stability of the system can be determined by judging the position of eigenvalues of the state matrix
A on the complex plane [
19].
Many studies show that the generator controller is effective in changing the real part of the oscillation mode; in the meantime, it has little influence on the imaginary part. As shown in
Figure 2, the dotted line is equal to the damping ratio line. After a lot of engineering practice, the Ontario Electric Power Bureau of Canada proposed that
is the critical state in the normal operation of the power system. Thus,
is set as 0.03 in this paper [
20]. In fact, the optimization of PSS is to move the eigenvalues of the state matrix of the system to the left part of the complex plane through continuous optimization, as far away from the virtual axis as possible. According to this rule, we can define the optimization objective function of the stabilizer and measure the dynamic characteristics of the system by the damping ratio of the system.
In the complex plane, each such conjugate complex root
has the following relationship with the system damping ratio
:
Define the objective function
f:
where
n represents the number of operating modes,
k is the number of oscillation modes,
is the preset minimum damping coefficient,
represents the
ith electromechanical oscillation mode damping coefficient in the
jth operating mode. Combined with other constraints, the parameter coordination of PSS can be expressed as the following eigenvalue optimization form:
where
,
,
=
= 1,
,
and
are three variables to be optimized and the remaining parameters are given in advance.
3. PSS Parameter Optimization Based on SA-QUATRE Algorithm
The evolution formula of the quasi-affine transformation is analogous to the affine transformation in geometry. The affine transformation function
in geometry is as follows:
The evolutionary structure of the QUATRE algorithm is to use
. Supposing a population of N particles is searching in a D-dimensional space, and
X in Formula (8) is used to represent the particle’s position. If the position of particle
i is
, and the population size is ps, the population position can be expressed as
. Then the particle position update formula is as follows:
where
B represents the evolutionary guidance matrix,
represents the bitwise multiplication of matrix elements.
and
are generated by randomly arranging the row vectors of the matrix
X. Their difference is the difference matrix, which is used to represent the particle search radius. This search method helps to adapt to different search dimensions,
c is the coefficient factor or step size of the differential matrix. If the
ith particle obtains the best fitness value, it is recorded as
. Then the global best coordinate matrix for each particle is shown below, and its size is
ps * 1.
is the cooperative search matrix. represents the incidence matrix of which is the core of the algorithm. This algorithm transforms all the ordinary individuals of the population into the optimal individuals and extends the global optimal individual search method to the whole population. Although this operation achieves individual equivalence, the complexity of the algorithm will become n times. Therefore, the collaborative architecture is introduced to reduce the search complexity of the algorithm, which is implemented by the M matrix.
The
M matrix is transformed from its initialization matrix
.
is a D-dimensional Boolean matrix whose elements of the lower triangular matrix are all 1 and the rest elements are 0, as shown in Equation (11). Its stacking method is determined according to the population size
ps, and the dimension
D. There are three stacking methods: if
ps =
D, if
ps = n * D,
is n times the previous case of vertical stack, if
ps = n * D+ k and
ps%
D = k (% means remainder), the
n * ps row of
is consistent with the second method, and the last
k rows are the first
k rows of the Boolean matrix in the first method. After the stacking is completed, the conversion is achieved through two consecutive operations. The first step is to randomly arrange the row elements of the matrix
and perform independent operations on each row element, The second step is to randomly arrange the row vectors of the matrix without changing the row elements [
21]. So,
M is compared with
, the only thing that changes is the position of 0 and 1 in each row and the position of each row vector.
This cooperative structure can effectively solve the defects that individuals in the population cannot achieve information sharing and high coordination in the evolution process due to the existence of two states of global search and local search.
Although QUATRE can gain performance advantages and reduce time complexity by multiplying the population, it will also fall into a local optimum in the later stage of the algorithm due to the reduction of population diversity. The simulated annealing algorithm is a global search algorithm, which accepts the difference with a certain probability during the search process and jumps out of the local optimum. Based on selecting the appropriate temperature parameter T, this paper uses a simulated annealing algorithm to improve QUATRE, which can improve the deficiencies of QUATRE’s later local optimization, speed up the algorithm’s process, and better design the parameters of the power system stabilizer.
Figure 3 shows the flow of the improved quasi-affine transformation evolution algorithm. The steps are summarized as follows:
Step1: Initialize the population and randomly initialize the particle position , limit the search range, set the difference matrix coefficient factor c, and set the initial position as the individual historical optimal solution of each particle as and the optimal global solution ;
Step2: Calculate the objective function of each particle;
Step3: Find the optimal value in the value of the initial particle objective function, and calculate the initial temperature T of the simulated annealing, ;
Step4: Calculate the sudden change probability at each temperature. The method is first to calculate the sudden change probability of each population corresponding to the best individual and then divide by the sum of the jump probability of all populations;
Step5: Randomly set and generate a probability. If the jump probability of a specific population is greater than or equal to this random value, update the ;
Step6: Evolve the parent population according to formula (10), update the position and generate the offspring population;
Step7: Calculate the optimal fitness value after each update, update the temperature of the simulated annealing algorithm. Output the optimal particles after meeting the conditions of the number of iterations or convergence accuracy.
This article applies the SA-QUATRE algorithm introduced above to the coordinated design of power system stabilizers. For a multi-machine system, all the PSS parameter tuning and optimization processes are carried out simultaneously. The parameters to be tuned for each PSS are
and
.
Figure 4 shows the flow chart of optimizing PSS parameters by SA-QUATRE algorithm. Under the objective function and constraint conditions of Equation (6), the optimization steps are summarized as follows:
- (1)
First, import the basic power flow and dynamic data of the grid and configure the number and position of power system stabilizers according to the electromechanical oscillation characteristics of the power system to be analyzed. In the power system, installing power system stabilizers for each generator is unrealistic. So, the residue method is used to choose the installation location of PSS in this article [
22];
- (2)
Set the parameters and operating mode of the power system stabilizer, linearize the system model and calculate the initial damping ratio in this mode through eigenvalues. If all are greater than ξmin, then end, otherwise continue;
- (3)
Initialize the parameters of the SA-QUATRE algorithm and simulated annealing, and at the same time, generate a set of initial solutions ;
- (4)
Assign each solution Xi to the variable in the PSS and calculate the damping coefficient of each electromechanical oscillation mode under this operating state;
- (5)
Evaluate the group according to the objective function based on characteristic value;
- (6)
Use the improved quasi-affine transformation evolutionary algorithm to continuously update and search, generate offspring populations, and update next-generation candidate solutions;
- (7)
The operation ends when reaching the maximum number of iterations, otherwise, it returns to the (4) step and enters the next cycle;
- (8)
Finally, obtain the global optimal parameter combination of PSS and the fitness value of the objective function. The specific process is as follows: