Aiming at the shortcomings of traditional controllers in response speed, accuracy and robustness of complex control systems, an automatic generation control system control method based on transfer function control theory is proposed by optimizing the gain of classical controllers. Firstly, dynamic teaching and learning techniques are used to optimize the parameters of controllers. The frequency deviation and the net exchange power increment of the inter-area tie-line are controlled, and the adjustment time is shortened. Then the two-area, three-area and large disturbance systems are simulated and compared with the performance of several classical controllers. The control effect shows that the transfer controller has good sensitivity and robustness in the automatic generation control system. It is very useful for the research of automatic generation control.With the deepening of power system reform, several independent power entities have been challenged in the fierce market competition.Automatic generation control (AGC), an important auxiliary service, plays an important role in preventing grid collapse accidents, maintaining system stability and improving power quality.The main objective of AGC is to ensure that the frequency of power system is within the allowable deviation range, to ensure the balance of output and load of network generators, and to control the power exchange value of inter-regional tie-line [2].At present, scholars at home and abroad have made fruitful achievements in the research of AGC.Literature [3] The bacterial foraging optimization algorithm is used to solve the conventional AGC problem, and the effectiveness of the algorithm is proved. However, the method in this paper still needs further research to deal with the multi-region AGC system.Document [4] Using genetic algorithm fuzzy controller provides a solution for the research of multi-area AGC system. Although genetic algorithm is more effective than traditional methods, with the deepening of research, it is found that genetic algorithm still has the shortcomings of not solving the problem of large-scale computation and easy to fall into “premature”.Reference [5] Considering the bilateral contract effect of dynamics, a fuzzy logic algorithm is used to optimize the gain parameters of the controller. However, when checking the rule base of the fuzzy logic controller, this method requires a lot of computation time.Reference [6] proposes an optimal output feedback control method, which uses reduced-order observer to solve the load frequency control problem in power market environment. However, this method can not guarantee the optimal dynamic response of the controller under constraints.Nowadays, classical controllers such as integral controller (I) [7], proportional-integral controller (PI) [8] and proportional-integral-differential controller (PID) [9] are becoming more and more mature and widely used.On the basis of comparing the performance of several classical controllers, the automatic regulation of system voltage and frequency is studied, and the uncertainty of classical controllers for improving system robustness is analyzed.Up to now, there is no literature to analyze the perturbation problem of high-order systems.On the basis of optimizing the gain of classical controllers, a mathematical model of transfer controllers is proposed in this paper. Dynamic teaching and learning techniques are used to optimize the parameters of the controllers. Compared with several classical controllers in reference [11], the controller can better control the frequency deviation and the net exchange power increment of inter-regional tie lines. The adjustment time is shortened.Finally, the optimal parameters and robustness of the transfer controller are studied by simulating two-zone [12], three-zone [13] and large disturbance [14] systems.
The basic principle of the transfer controller proposed in this paper is to use the transfer function to control the frequency deviation and the net exchange power increment of the inter-regional tie lines, so as to shorten the adjustment time to the greatest extent.The area control error (ACE) of the controllers in their respective regions can be calculated according to formula (1). Power plants compete with each other in AGC system, and power supply companies choose power plants freely. This leads to a variety of power distribution modes between power plants and power supply companies. Therefore, the definition of DPM is introduced.
Each element in the matrix reflects the participation of the power supply company. Its numerical value is the percentage of the total load power purchased by the power supply company from the power supply company. The number of rows in the matrix is the number of power plants, and the number of columns in the matrix is the number of power supply companies. It can be seen that the sum of elements in the matrix is equal to 1, and the sum of each column is equal to 1.The transfer controller can not only control the frequency deviation and voltage phase angle_E, but also control the net exchange power increment, the adjustment time and the load node power of the inter-area tie-line.In addition, the inherent characteristics of the transfer function determine that the transfer controller has good sensitivity, and the controller is insensitive to the parameters of the controlled system.The control strategy of the transfer controller is to calculate the control deviation ACE of the area according to the real-time collected values of the controlled parameters_, f, and Ptie of the transfer controllers in different areas of the power grid.The total power of AGC generating units is obtained by optimizing the transfer controller.
The output power of each AGC generating unit in the region is allocated according to the ACE participation factor APF of the generating units. The dynamic coordinated control of the AGC generating units in the power grid is realized. Finally, the purpose of frequency regulation of the system and net exchange power control of the inter-regional tie lines is achieved.The system control structure of the transfer controller is shown in Figure 1.In this paper, dynamic teaching and learning optimization technology is used to search all spatial parameters.This method is a heuristic swarm intelligence algorithm to simulate the teaching effect of teachers on class students. In the optimization process, a class is a population, and the total number of teachers and students in the class is the number of individuals contained in the population. All teachers and students are divided into group D to study their respective subjects, and group D is the dimension of the population, and corresponds to group d, respectively. Optimize parameter variables.Students are regarded as a group of design variables to optimize parameters, and each teacher is the best solution to optimize parameters variables (in fact, the level of each individual in the population can not reach the level of teachers). Teachers improve the average performance of students by sharing knowledge with students.The fitness value is calculated by the grade of the class, and the best solution is the optimal value of the objective function.The simulation optimization algorithm consists of “teaching stage” and “learning stage”.At this stage, teacher Xt tries to improve the average score of Xs of all the students in the class.The students who satisfy formula (17) have better learning performance, and the algorithm will continue to iterate until the optimal value is reached; the students who satisfy formula (18) have poor learning ability, and they can continue to improve their learning ability through formula (15) of repetition teaching stage, thereby improving their learning performance.The flow chart of the algorithm is shown in Figure 2.Using MATLAB/SIMULINK to realize the dynamic optimization algorithm of teaching and learning and test it. The test methods are as follows: the total number of teachers and students in the class is Np=212 and the grouping number d=2. The first group has one teacher and 10 students, the second group has one teacher and 200 students. The same optimization parameters are divided into two groups to optimize teaching and learning.The test results show that there is little difference between the two groups in the final convergence value of learning optimization; the first group of teachers and students learn 30 times iteratively, and the convergence result reaches the optimal value, while the second group of teachers and students have to learn 50 times iteratively before reaching the same result.Therefore, in the following case analysis, each group of teachers and students take one, 10.When the difference between the optimal solution of the optimized parameter and the standard optimal solution is less than the preset value, it is considered that the optimization is successful and the cycle output optimal value is terminated. Otherwise, return to the “teaching” stage and “learning” stage, repeat the iterative process, and increase the number of iterations by 1.Compared with traditional optimization technology, dynamic teaching and learning method can achieve optimal learning effect by teachers leading students to study in groups and reasonably setting learning times. It can significantly improve the ability of searching the optimal value of control parameters in all space, avoid abnormal convergence caused by numerical over-iteration and local optimization, and improve the convergence speed. The calculation time is shortened and the method is simple and feasible.To verify the effectiveness of the proposed method, the IEEE39 node system in Figure 3 is simulated and compared with the PID and PI controllers.Among them, the generator at node 30-35 is AGC unit.Based on the number of partitions specified artificially, this paper uses the multi-objective quantitative evaluation algorithm of document [15] to partition the IEEE39-node system internally.The principle of this method is: based on the results of the electrical distance matrix between the nodes of each level system, the initial partition of grid generators is carried out by using K-means according to the specified number of partitions, and then the objective solution is obtained by constructing the fitness function suitable for different operation modes.In the two-zone system, the AGC units numbered 30, 31 and 32 nodes belong to Area 1, while the AGC units of the other nodes belong to Area 2, and the lines between nodes 14 and 15 are the inter-regional contact lines.Area 1:2 000 MW, area 2:2 000 MW, and in each control area, there are two power supply companies and two power plants.In the three-area system, the AGC units numbered 30 and 31 nodes belong to area 1, 32 and 34 nodes belong to area 2, the AGC units numbered 33 and 35 nodes belong to area 3, the lines between node 4 and 14 belong to area 1 and area 2, and the lines between node 15 and 16 belong to area 2 and area 3.Area 1:2 000 MW, area 2:2 000 MW, area 3:2 000 MW, and there are two power supply companies and two power plants in Region 1, and only one power supply company and one power plant in Region 2 and Region 3 respectively.The system is simulated by using MATLAB/SIMULINK. Considering the randomness of the optimization algorithm, a large number of iteration learning times can be set to 1000 times at the first run of the program, and the test results can be recorded.After the results are obtained, the number of iterations is reduced to the number of iterations close to the convergent stable value.
After testing, the maximum number of iteration learning times evaluated by the objective function is 30. The optimal value obtained in operation is chosen as the optimal solution of the controller parameters and the convergence stability value of the control variables.The traditional two-area system analysis is to simulate simple AGC research between two regions (for example, AGC control between two adjacent power supply enterprises).This paper studies the system in which load changes only occur in area 1, that is, only power supply company 1 and power supply company 2 participate in the transaction, thermostatic element and assumes that power supply company 3 and power supply company 4 in area 2 do not have any power load in other power plants.The dispatching center calculates the ACE value according to the real-time information of the tracked system_, f, _and_Ptie. Through the transfer of the controller, a total power instruction is obtained. According to the APF and DPM data of AGC units, the power output of each unit is allocated. The four control parameters of ACE are optimized by using the dynamic teaching and learning algorithm in this paper, and the ACE is optimized. Output, to achieve economic and reasonable control of AGC units.The class population parameters Np=44 and d=4 were set.Among them, the number of teachers in the class is 4, and the number of students is 40.All the trainees were divided into four groups, 10 in each group. One teacher was followed to learn four subjects of AGC control parameters_, Delta f, Delta and Delta Ptie, i.e. four groups of teachers and students learned to optimize four parameters.Enter the stage of “teaching”.Students enter the “learning” stage.Students in each subject use different learning strategies to learn.If the optimal solution is less than 0.005, the optimal termination will be optimized and the optimal value will be output. Otherwise, step 3 will be repeated and the number of iterations will be increased by 1.Finally, the aim of controlling AGC is realized by using the optimal ACE value of the output.The optimization algorithm is compared with the traditional particle swarm optimization algorithm and artificial neural network algorithm, and the comparison results are shown in Table 1.By analyzing the optimization results of the three algorithms in the table, it can be clearly concluded that the dynamic teaching and learning algorithm has better optimal value than the other two algorithms in the optimization process.In order to study the dynamic response characteristics of the transfer controller, a 10% step load change in region 1 is set at t = 0 s.The dynamic response results are tested and compared with those of PID and PI controllers in the same environment, as shown in Fig. 4, Fig. 5 and Fig. 6.The transfer controller has better control effect and higher sensitivity, and the dynamic response performance of the controller has been significantly improved.The convergence speed of the transfer controller has obvious advantages: 30% higher than that of the PID controller and 35% higher than that of the PI controller.Fig. 5 shows that when 10% step load changes in zone 1, 10% step frequency offset dynamic synchronous response is generated in zone 2.It is clear from the figure that when the position of step load changes, the transfer controller has better dynamic response characteristics than the PID and PI controllers.Fig. 6 shows that the peak overshoot of the transfer controller, the net exchange power of the interconnection lines and the deviation of the transfer controller are significantly improved due to the inherent characteristics of the transfer function when a 10% step load change is applied in region 1.The controller parameters are given in Table 2.Compared with ITSE = 0.573 6 of PID controller and ITSE = 0.783 7 of PI controller, the transfer controller has a minimum ITSE value of 0.384 9 and absolute advantage in regulating time.With the expansion of smart grid scale, the research of three-area and multi-area systems is to simulate the AGC problem of large-scale complex networks (e.g. between provinces).In order to simplify the calculation, this paper still uses IEEE39 node system to simulate and analyze.In order to verify the cooperative ability of transfer controllers and other types of controllers to control multi-source, multi-area interconnected power systems, simulation is extended to three-area systems, and each area includes generator units with different types of controllers, and has a high voltage direct current (HVDC) transmission system.At t = 0 s, there are 10% step load changes in the three regions at the same time.The optimal gain of the transfer controller is obtained by dynamic teaching and learning optimization technology.The simulation results are compared with the dynamic response data of the classical controller. The results are shown in figs. 7, 8, 9, 10 and 11.From the data in the figure, it can be seen that the transfer controller has good compatibility with other types of controllers.Figures 9 and 10 show that the net exchange power increment curve of interconnected lines is parallel in HVDC systems.Compared with Fig. 5, it can be seen that if there are no HVDC transmission lines and equipment in the system, the net exchange power increment curve of the inter-area tie lines also proves that the ITSE objective function is used to improve the sensitivity of the controller.Compared with PID and PI controllers, in the same test environment, the transfer controller can obtain the minimum value of the contact line power transaction.The simulation results of the optimal value of ITSE objective function and adjustment time for different kinds of controllers are given in Table 3.It can be seen that in the same test environment, using ITSE objective function to obtain the optimal parameters of the transfer controller has excellent performance: adjusting time and ITSE optimal values change within a reasonable range, and close to the corresponding values obtained by nominal system parameters.Robustness refers to the characteristic that the control system maintains some other performances under certain parameter perturbations (structure, size).If the transfer controller is a robust controller, there is no need to readjust the parameters of the controller itself when the load condition of the system changes or the parameters of the system change.In order to verify the good robustness of the transfer controller in case of large disturbance, the parameters and load conditions of the two-area system are simulated. The parameters range from -150% to 150% of the nominal value of the system, and 35% of the step load varies in the region 1.
The optimal gain of the transfer controller is not changed, and the region 1 is assumed.
Ten percent of contracts are in default.In order to show the superior robustness of the transfer controller to large disturbances, the results are compared with those of the PID and PI controllers. The dynamic responses of the frequency deviation of the controller and the net exchange power deviation of the inter-area tie lines are compared as shown in figs. 12 and 13.It can be seen from the graph that when the system has large disturbance, the transfer controller has better stability than the PID and PI controllers, and the influence of the large fluctuation of load conditions on the system response can be neglected, and the influence of the large change of system parameters on the performance index of the controller can also be neglected.
Under nominal load, the optimal value of controller parameters obtained by nominal parameters need not be reset because of the great change of system load or system parameters, nor need to reset system load and system parameters, nor need to change the location and scope of default.Table 4 gives the comparative data of controller performance when large disturbances occur.Under the same large disturbance, the transfer controller has better robustness than the PID and PI controllers.In this paper, the transfer controller is applied to the multi-area AGC system in the power market environment, and the dynamic teaching and learning optimization technology is used to optimize the parameters of the controller at the same time. Finally, the IEEE39 node system is simulated.Compared with the performance of classical controllers, the controller has certain advantages in reducing frequency deviation, net exchange power increment of interconnection lines, shortening adjustment time and obtaining the optimal value of objective function, and has good stability and robustness in both two-zone or three-zone systems and large disturbance systems. 。It has a good use value for the study of AGC system control.