Method for optimizing PI control parameters of three-phase permanent magnet synchronous motor based on EMMOA

Abstract

The invention discloses a three-phase permanent magnet synchronous motor PI control parameter optimization method based on EMMOA. The method comprises the steps of population initialization, real evaluation of a simulation model, training of an agent model by using all real evaluation solutions, population evolution, real evaluation of the simulation model and updating of the agent model. In the population evolution process, utilizing local search to optimize a local optimal solution, utilizing global search to search a potential multi-modal solution, utilizing a multi-modal detection strategy to judge whether the global search generates a new multi-modal solution or not during the period, and finally utilizing archives to screen the potential multi-modal solution; and after each evolution is finished, the potential multi-modal solution is truly evaluated and archived, and the agent model is updated. According to the method, the PI control parameters of the three-phase permanent magnet synchronous motor are optimized by combining the particle swarm evolutionary algorithm with the proxy model, the convergence and diversity of the population are enhanced by using a local and global evolutionary mode, the time cost can be effectively saved, and a satisfactory solution can be found.

Description

Technical Field

[0001] The present invention relates to the field of optimization algorithms, and in particular to an EMMOA-based three-phase permanent magnet synchronous motor PI control parameter optimization method. Background Art

[0002] Three-phase permanent magnet synchronous motor (PMSM) has been widely used in industrial automation, electric vehicles, aerospace, wind power generation and intelligent manufacturing due to its high power density, high efficiency and superior dynamic performance. Compared with other types of motors, PMSM does not require external excitation and uses permanent magnets to generate magnetic fields, which reduces the loss and structural complexity of the motor, so it has occupied a dominant position in modern high-performance motion control systems.

[0003] Although PMSM has many advantages in performance, its efficient operation depends on accurate control strategies. In the control system of PMSM, closed-loop control of speed and current is the key to achieve its efficient operation. PI (proportional-integral) controller has become the mainstream strategy for PMSM control because of its simplicity, effectiveness and easy implementation. The main task of PI controller is to adjust the error of current and speed and minimize the error as much as possible during dynamic response. However, the performance of PI controller is highly dependent on the selection of parameters. Reasonable PI parameters can significantly improve the response speed of the motor, reduce steady-state error, and improve the dynamic and static performance of the system.

[0004] Although the structure of the PI controller is simple, the adjustment of its parameters (such as proportional gain VKP, VKI and integral gain IKP, IKI) is crucial in PMSM control. In practical applications, how to obtain the optimal PI parameters suitable for different working conditions and operating conditions has become a technical difficulty. This is because the working environment and load conditions of the PMSM are complex and changeable. The parameters of the PI controller that perform well under one working condition may fail under another working condition, resulting in system instability, excessive overshoot or slow response speed.

[0005] Traditional PI parameter tuning methods, such as the empirical trial and error method, Ziegler-Nichols tuning method, or frequency domain analysis method, can provide good initial parameters in some cases, but usually do not have global optimization capabilities and are difficult to adapt to the efficient control requirements of motors under various complex working conditions. In addition, manual adjustment of PI parameters is time-consuming and labor-intensive, and relies on the operator's experience and lacks versatility. Especially in nonlinear, time-varying permanent magnet synchronous motor control systems, the selection of PI parameters has a crucial impact on the performance of the motor.

[0006] In order to better solve the problem of PI parameter adjustment, optimization methods based on intelligent algorithms have gradually attracted attention in recent years. These methods use algorithms to automatically search for global optimal or suboptimal solutions, avoiding human intervention and improving the accuracy and efficiency of parameter adjustment. Common intelligent optimization algorithms include genetic algorithms, particle swarm algorithms, simulated annealing, etc. These algorithms optimize control parameters in complex systems by simulating evolution, group behavior or physical processes in nature.

[0007] However, these optimization algorithms often require a lot of computing resources and function evaluation when dealing with motor control problems. Especially when the motor model is complex or the operating conditions vary greatly, traditional optimization algorithms are prone to fall into local optimal solutions or converge slowly.

[0008] Multimodal optimization algorithms are a type of optimization algorithm designed to solve multimodal problems. Compared with traditional single-peak optimization algorithms, multimodal algorithms can find multiple global or local optimal solutions, providing more abundant parameter choices for motor control under different working conditions. These algorithms can not only improve global search capabilities, but also provide a better balance between the diversity and complexity of solutions.

[0009] Expensive Multimodal Optimization Algorithms (EMMOA) are a class of algorithms that are particularly suitable for dealing with high computational cost problems. This type of algorithm significantly reduces the computational burden by reducing the number of function evaluations, and is particularly suitable for application scenarios with limited computing resources and the need for precise parameter adjustment, such as PI parameter optimization in PMSM control systems. In motor control, EMMOA can effectively cope with the nonlinear and strongly coupled characteristics of the motor system, while providing the best PI parameter combination for different operating conditions.

[0010] In the prior art, the PI parameter optimization method for PMSM has been widely studied, and some intelligent optimization algorithms have been applied in this field. However, the prior art still has the following shortcomings:

[0011] Limitations of parameter adjustment: Traditional PI parameter adjustment methods are mostly based on experience or single-objective optimization, lack globality, are difficult to adapt to complex changes in operating conditions, and easily lead to unstable system performance.

[0012] Computing resource consumption: Existing intelligent algorithms, such as genetic algorithms and particle swarm algorithms, have strong global search capabilities, but when dealing with complex problems, they often require a large amount of function evaluation and computing resources, making it difficult to implement them in real time in actual engineering applications.

[0013] Unsatisfactory optimization results: Existing optimization methods based on intelligent algorithms have limited optimization results when faced with large changes in working conditions or scarce data, and fail to fully utilize the potential of multimodal optimization algorithms. Summary of the invention

[0014] In order to solve the above problems, the present invention provides a new EMMOA-based three-phase permanent magnet synchronous motor PI control parameter optimization method, which can provide a more accurate PI parameter optimization scheme for the three-phase permanent magnet synchronous motor under different working conditions by combining the global search capability of the multi-modal optimization algorithm with the efficient utilization of computing resources of the expensive optimization algorithm. The present invention can not only achieve a better balance between dynamic response and steady-state error, but also reduce the computational complexity of the motor control system and improve the real-time performance and stability of the system.

[0015] The technical solution adopted by the present invention is:

[0016] The PI control parameter optimization method of a three-phase permanent magnet synchronous motor based on EMMOA includes the following steps:

[0017] S1, population initialization; using the speed loop PI and current loop PI of the three-phase permanent magnet synchronous simulation model as optimization variables, Latin hypercube is used to generate the initial population, and the initial population size is set;

[0018] S2, conduct a true evaluation of the simulation model; the simulation model first obtains the initial variables generated in S1, and assigns the variables to the simulation model through the assignin function; the simulation model is run to obtain the absolute time error IATE of the model, and the IATE is obtained through the sim function as the true evaluation of the algorithm;

[0019] S3, train the proxy model using the solutions of all true evaluations;

[0020] S4, population evolution;

[0021] S5, the simulation model truly evaluates and updates the proxy model; by assigning the multimodal solution variables in S4 to the simulation model, IATE is obtained as the true evaluation. In addition, all individuals with true evaluations will be archived in the database DB, and the proxy model will be updated in time for subsequent prediction processes.

[0022] Furthermore, the specific steps of S4 include:

[0023] S41, based on the obtained optimal solution, use the built-in function of Matlab to determine and eliminate similar optimal solutions;

[0024] S42, optimizing the local optimal solution using a local search strategy;

[0025] S43, judging the number of the current optimal solutions, if it is less than the set value 5, directly performing a global search, otherwise performing a global search every two generations;

[0026] S44, further screening potential optimal solutions using the optimal solutions saved in the archive and the solutions of the screened areas;

[0027] S45, storing the searched potential optimal solution in an archive;

[0028] S46, using an improved modal detection strategy to determine whether the searched optimal solution is actually found;

[0029] S47, determine whether the maximum number of iterations of the population has been reached; if so, stop the evolution; otherwise, go to S41 to continue the iterative evolution process.

[0030] Further, the specific steps of S42 are: determine the number of local search areas according to the number of current optimal solutions; after determining the number, select 10 particles close to each optimal solution in the database DB to form a local population, record the local optimal Lbest and the individual optimal Pbest, and update the particles according to the following particle swarm update method:

[0031]

[0032]

[0033] Where i represents the sequence of the local population, d represents the dimension, n represents the particle in the local population, rand represents the random number generated in [0,1]; t represents the number of generations, v represents the speed, w, c1, c2 are the parameters of the particle swarm algorithm;

[0034] After the position of the particle is updated, the proxy model is used to predict the fitness value of the particle, update the local optimal Lbest and the individual optimal Pbest, meet the local cycle end condition, and terminate the local search strategy; finally, the potential local optimal solution is truly evaluated through the simulation model and stored in the database DB; when the searched local optimal solution is better than the actual local optimal solution, the searched local optimal solution replaces the actual local optimal solution.

[0035] Further, the specific steps of S43 are:

[0036] When executing the global search process, the number of optimal solutions is first pruned. When it is greater than the set value of 5, the best 5 individuals are selected as the optimal solution. Similar to the local search, 10 particles close to each optimal solution are selected from the database DB to form a local population, and then potential multimodal solutions are explored in the evolutionary way of the ring topology. The specific particle update method is as follows:

[0037]

[0038] Among them, seed represents the optimal solution, localpop represents randomly selecting a particle from the next local population; when n=η, n+1 is set to 0, so that the entire evolutionary form presents a ring topology structure; at this time, the size of the offspring population may be less than 100, so offspring individuals are randomly generated in space, and finally a offspring population of size 100 is formed; finally, AP clustering is used to screen out potential multimodal solutions.

[0039] Further, the specific steps of S44 are:

[0040] The archive selection strategy will bind the potential multimodal solutions to the particles in the archive respectively, calculate the Euclidean distance value with the closest distance, and then sort the target value from large to small to select the potential multimodal solution with the largest distance; the potential global optimal solution with the largest distance will be truly evaluated through the simulation model and stored in the database DB.

[0041] Further, in S46, the improved modal detection strategy includes calculating the distance between the potential optimal solution selected in S44 and the actual optimal solution, sorting them from large to small, and selecting the actual multimodal solution with the smallest distance as the basis for modal detection; calculating the center point coordinates of the potential multimodal solution and the actual multimodal solution in space, and selecting particles in the database DB within the radius r of the center point to determine whether a new multimodal solution is generated; when there are particles within the radius r, the particles closest to the center point are selected as the basis for judgment; when there are no particles within the radius r, the center point is directly used as the basis for judgment; the center point is truly evaluated through a simulation model and stored in the database DB.

[0042] Furthermore, a specific method for determining whether a new multimodal solution is generated is: when the fitness value of the center point is less than the fitness values ​​of the potential multimodal solution and the actual multimodal solution at the same time, the potential multimodal solution becomes the actual multimodal solution; otherwise, it is deemed that no new multimodal solution is generated.

[0043] The beneficial effects produced by the present invention are:

[0044] 1. Building a global search and local search framework can effectively improve the diversity and convergence of solutions;

[0045] 2. The predictive characteristics of the proxy model can effectively save running time costs;

[0046] 3. Save and select multimodal solutions with greater potential to avoid wasting computing resources;

[0047] 4. Improved multimodal detection strategy to timely screen the true multimodal solutions;

[0048] 5. It has good interpretability and is easy to understand the population evolution process. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 is a flow chart of the method of the present invention;

[0050] Figure 2 Convergence curve comparison of EMMOA in the present invention;

[0051] Figure 3 It is a schematic diagram of the interaction between the algorithm of the present invention and the simulation model data. Specific implementation plan

[0052] The present invention will be further described below in conjunction with the accompanying drawings.

[0053] like Figure 1 As shown, the present invention is a three-phase permanent magnet synchronous motor PI control parameter optimization method based on EMMOA, comprising the following steps:

[0054] S1, population initialization; the speed loop PI and current loop PI of the three-phase permanent magnet synchronous simulation model are used as optimization variables, and the Latin hypercube is used to generate the initial population, and the initial population size is set to 20.

[0055] The speed loop PI is recorded as variables VKP and VKI, and the current loop PI is recorded as variables IKP and IKI.

[0056] S2, conducts a true evaluation of the simulation model; the simulation model first obtains the initial variables generated in S1, and assigns the variables to the simulation model through the assignin function; the simulation model is run to obtain the absolute time error IATE of the model, and the IATE is obtained through the sim function as the true evaluation of the algorithm.

[0057] S3, trains the proxy model using the solutions of all true evaluations.

[0058] S4, population evolution; the specific steps include:

[0059] S41, based on the obtained optimal solution, use the built-in function of Matlab to determine and eliminate similar optimal solutions;

[0060] S42, optimize the local optimal solution using the local search strategy; the specific steps are: determine the number of local search areas according to the number of current optimal solutions. After the number is determined, select 10 particles that are close to each optimal solution in the database DB to form a local population, record the local optimal Lbest and the individual optimal Pbest, and update the particles according to the following particle swarm update method:

[0061]

[0062] Among them, i represents the sequence of the local population, d represents the dimension, n represents the particle within the local population, rand represents the generation of random numbers in [0,1]; t represents the number of generations, v represents the speed, and w, c1, and c2 are the parameters of the particle swarm algorithm.

[0063] After the particle position is updated, the proxy model is used to predict the particle fitness value, update the local optimal Lbest and the individual optimal Pbest, reach the local cycle end condition, and terminate the local search strategy. Finally, the potential local optimal solution is truly evaluated through the simulation model and stored in the database DB. When the searched local optimal solution is better than the actual local optimal solution, the searched local optimal solution replaces the actual local optimal solution.

[0064] S43, judging the number of the current optimal solutions, if it is less than the set value 5, directly perform global search, otherwise perform global search every two generations; the specific steps are:

[0065] When executing the global search process, the number of optimal solutions is first pruned. When it is greater than the set value of 5, the best 5 individuals are selected as the optimal solution. Similar to the local search, 10 particles close to each optimal solution are selected from the database DB to form a local population, and then potential multimodal solutions are explored in the evolutionary way of the ring topology. The specific particle update method is as follows:

[0066]

[0067] Among them, seed represents the optimal solution, localpop represents randomly selecting a particle from the next local population; when n = η (η is the number of true multimodal solutions), n + 1 is set to 0, so that the entire evolutionary form presents a ring topology structure. At this time, the size of the offspring population may be less than 100, so offspring individuals are randomly generated in space, and finally a offspring population of 100 is formed. Finally, AP clustering is used to screen out potential multimodal solutions.

[0068] S44, using the optimal solutions saved in the archive and the solutions of the screened areas, further screen potential optimal solutions; the specific steps are:

[0069] The archive selection strategy will bind the potential multimodal solutions to the particles in the archive respectively, calculate the Euclidean distance value with the closest distance, and then sort the target values ​​from large to small to select the potential multimodal solution with the largest distance. The potential global optimal solution with the largest distance is truly evaluated through the simulation model and stored in the database DB.

[0070] S45, storing the searched potential optimal solution in an archive;

[0071] S46, using the improved modal detection strategy to determine whether the real optimal solution is really searched; the specific steps are: the improved modal detection strategy includes calculating the distance between the potential optimal solution selected in S44 and the actual optimal solution, sorting from large to small, and selecting the actual multimodal solution with the smallest distance as the basis for modal detection. Calculate the center point coordinates of the potential multimodal solution and the actual multimodal solution in space, and select particles in the database DB within the radius r of the center point to determine whether a new multimodal solution is generated. There are two cases, one is that there are particles within the radius r, and the particle closest to the center point is selected as the basis for judgment; the other is that there are no particles within the radius r, and the center point is directly used as the basis for judgment. The center point is truly evaluated through the simulation model and stored in the database DB. The specific method for determining whether a new multimodal solution is generated is: when the fitness value of the center point is less than the fitness value of the potential multimodal solution and the actual multimodal solution at the same time, the potential multimodal solution becomes the actual multimodal solution; otherwise, it is deemed that no new multimodal solution is generated.

[0072] S47, determine whether the maximum number of iterations of the population has been reached; if so, stop the evolution; otherwise, go to S41 to continue the iterative evolution process.

[0073] S5, the simulation model truly evaluates and updates the proxy model; by assigning the multimodal solution variables in S4 to the simulation model, IATE is obtained as the true evaluation. In addition, all individuals with true evaluations will be archived in the database DB, and the proxy model will be updated in time for subsequent prediction processes.

[0074] In order to further illustrate the characteristics of the present invention in processing the PI parameter optimization of the three-phase permanent magnet synchronous motor Figure 2 The comparison between the SAPSO of the present invention and two expensive multimodal algorithms MaMPSO (Multisurrogate-assisted multitaskingparticle swarm optimization for expensive multimodal problems) and DSCPSO-MMO (Dual-surrogate-assisted cooperativeparticle swarm optimization for expensive multimodal problems) is given, wherein the comparison index selected is the time absolute error integral IATE, which is used to measure the performance index of the overall system.

[0075] To further illustrate the data interaction between EMMOA, Simulink simulation model and Ansys three-phase permanent magnet synchronous motor, Figure 3 The details are shown. The Simulink simulation model is a PMSM vector control model based on PI regulator. EMMOA controls the speed loop PI and the current loop PI as optimization variables, and the absolute time error generated by the speed loop PI regulator and the current loop PI regulator is fed back to EMMOA as the optimization target. The PMSM is built by Ansys.

[0076] In this embodiment, in the PMSM vector control problem based on PI regulator, Latin hypercube is used to generate random initial population, each algorithm runs independently 10 times, and finally takes the average value. The initial size of the population is set to 20, and the final real evaluation number RFEN is set to 200. Table 1 contains the four parameters and target values ​​that need to be optimized this time. The parameters VKP and VKI represent the proportion and integral of the speed loop, respectively, and the parameters IKP and IKI represent the proportion and integral of the current loop, respectively. It can be seen from Table 1 that SAPSO provides us with 3 solutions, which can be selected as needed. At the same time, Figure 1 In the convergence curve of , SAPSO is obviously better than MaMPSO and DSCPSO-MMO. These results show the superiority of the proposed method in the PI parameter optimization of three-phase permanent magnet synchronous motor.

[0077] Table 1 Multimodal solutions provided by SAPSO

[0078] VKP VKI IKP IKI IATE Solution 1 14.523 3.966 28.147 5.200 0.04358 Solution 2 9.376 0.215 87.931 1.954 0.04374 Solution 3 6.304 5.589 87.805 5.863 0.04374

[0079] The real evaluation of practical problems is often costly, so it is difficult for current environmental assessments that require a large number of evaluations to effectively deal with such problems. Therefore, lightweight computing methods that use agents instead of actual evaluations have been applied to evolutionary computing. This method is called the surrogate model-assisted evolutionary algorithm (SAEA). SAEA has become a powerful tool for solving expensive optimization problems, in which objective values ​​are estimated by model prediction rather than actual evaluation. Surrogate models assist evolutionary algorithms in solving expensive multimodal problems, which can effectively save running time costs, find multimodal solutions faster, and effectively search for global optimal or near-optimal solutions in complex spaces. Surrogate models are now widely used to solve various practical problems due to their high efficiency and strong interpretability. The global search and local search strategies are combined with each other, and the diversity and convergence of solutions are guaranteed. After that, potential solutions are further screened through archives, and then the improved modal detection strategy is used to determine whether a true multimodal solution is generated. The effective combination of multiple strategies with the surrogate model gives the evolutionary algorithm a powerful search performance.

[0080] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A three-phase permanent magnet synchronous motor PI control parameter optimization method based on EMMOA, characterized in that: The following steps are involved: S1, population initialization; using the speed loop PI and current loop PI of the three-phase permanent magnet synchronous simulation model as optimization variables, Latin hypercube is used to generate the initial population, and the initial population size is set; S2, conduct a realistic evaluation of the simulation model; The simulation model first obtains the initial variables generated in S1, and assigns the variables to the simulation model through the assignin function; runs the simulation model to obtain the absolute time error IATE of the model, and obtains IATE as the true evaluation of the algorithm through the sim function; S3, train the proxy model using the solutions of all true evaluations; S4, population evolution; S5, the simulation model truly evaluates and updates the proxy model; by assigning the multimodal solution variables in S4 to the simulation model, IATE is obtained as the true evaluation. In addition, all individuals with true evaluations will be archived in the database DB, and the proxy model will be updated in time for subsequent prediction processes.

2. The method for optimizing PI control parameters of a three-phase permanent magnet synchronous motor based on EMMOA according to claim 1, characterized in that: The specific steps of S4 include: S41, based on the obtained optimal solution, use the built-in function of Matlab to determine and eliminate similar optimal solutions; S42, optimizing the local optimal solution using a local search strategy; S43, judging the number of the current optimal solutions, if it is less than the set value 5, directly performing a global search, otherwise performing a global search every two generations; S44, further screening potential optimal solutions using the optimal solutions saved in the archive and the solutions of the screened areas; S45, storing the searched potential optimal solution in an archive; S46, using an improved modal detection strategy to determine whether the searched optimal solution is actually found; S47, determine whether the maximum number of iterations of the population has been reached; if so, stop the evolution; otherwise, go to S41 to continue the iterative evolution process.

3. The method for optimizing PI control parameters of a three-phase permanent magnet synchronous motor based on EMMOA according to claim 2 is characterized in that: The specific steps of S42 are: Determine the number of local search areas according to the number of current optimal solutions; after determining the number, select 10 particles close to each optimal solution in the database DB to form a local population, record the local optimal Lbest and individual optimal Pbest, and update the particles according to the following particle swarm update method: Where i represents the sequence of the local population, d represents the dimension, n represents the particle in the local population, rand represents the random number generated in [0,1]; t represents the number of generations, v represents the speed, w, c1, c2 are the parameters of the particle swarm algorithm; After the position of the particle is updated, the proxy model is used to predict the fitness value of the particle, update the local optimal Lbest and the individual optimal Pbest, meet the local cycle end condition, and terminate the local search strategy; finally, the potential local optimal solution is truly evaluated through the simulation model and stored in the database DB; when the searched local optimal solution is better than the actual local optimal solution, the searched local optimal solution replaces the actual local optimal solution.

4. The method for optimizing PI control parameters of a three-phase permanent magnet synchronous motor based on EMMOA according to claim 2 is characterized in that: The specific steps of S43 are: When executing the global search process, the number of optimal solutions is first pruned. When it is greater than the set value of 5, the best 5 individuals are selected as the optimal solution. Similar to the local search, 10 particles close to each optimal solution are selected from the database DB to form a local population, and then potential multimodal solutions are explored in the evolutionary way of the ring topology. The specific particle update method is as follows: Among them, seed represents the optimal solution, localpop represents randomly selecting a particle from the next local population; when n=η, n+1 is set to 0, so that the entire evolutionary form presents a ring topology structure; at this time, the size of the offspring population may be less than 100, so offspring individuals are randomly generated in space, and finally a offspring population of size 100 is formed; finally, AP clustering is used to screen out potential multimodal solutions.

5. The method for optimizing PI control parameters of a three-phase permanent magnet synchronous motor based on EMMOA according to claim 2 is characterized in that: The specific steps of S44 are: The archive selection strategy will bind the potential multimodal solutions to the particles in the archive respectively, calculate the Euclidean distance value with the closest distance, and then sort the target value from large to small to select the potential multimodal solution with the largest distance; the potential global optimal solution with the largest distance will be truly evaluated through the simulation model and stored in the database DB.

6. The method for optimizing PI control parameters of a three-phase permanent magnet synchronous motor based on EMMOA according to claim 2, characterized in that: In S46, the improved modal detection strategy includes calculating the distance between the potential optimal solution selected in S44 and the actual optimal solution, sorting them from large to small, and selecting the actual multimodal solution with the smallest distance as the basis for modal detection; calculating the center point coordinates of the potential multimodal solution and the actual multimodal solution in space, and selecting particles in the database DB within the radius r of the center point to determine whether a new multimodal solution is generated; when there are particles within the radius r, the particles closest to the center point are selected as the basis for judgment; when there are no particles within the radius r, the center point is directly used as the basis for judgment; the center point is truly evaluated through a simulation model and stored in the database DB.

7. The method for optimizing PI control parameters of a three-phase permanent magnet synchronous motor based on EMMOA according to claim 6, characterized in that: The specific method for judging whether a new multimodal solution is generated is: when the fitness value of the center point is less than the fitness values ​​of the potential multimodal solution and the actual multimodal solution at the same time, the potential multimodal solution becomes the actual multimodal solution; otherwise, it is deemed that no new multimodal solution is generated.

Images