Speed control method for alternating current servo system of permanent magnet synchronous motor

Abstract

The invention discloses a PMSM (permanent magnet synchronous motor) AC servo system speed control method based on improved particle swarm optimization optimization RBF (radial basis function) fuzzy neural network PID control, the system is composed of a speed outer ring and a current inner ring, and the current ring adopts a PI regulator. A speed loop is designed to be an RBF fuzzy neural network PID speed controller, an RBF fuzzy neural network is optimized by improving a particle swarm optimization algorithm, and optimal parameters kp, ki and kd suitable for the PID controller are output. Simulation analysis and experiment results show that compared with a common PID control strategy, a BP algorithm optimization fuzzy neural network PID control strategy and a particle swarm algorithm optimization RBF fuzzy neural network PID control strategy, the particle swarm algorithm optimization RBF fuzzy neural network PID control strategy based on the improved particle swarm algorithm has optimal control performance; the method has the advantages of being high in dynamic response speed, good in control stability, resistant to load disturbance and high in robustness, and the excellent speed control effect of the PMSM alternating current servo system can be achieved.

Description

Technical Field

[0001] The invention relates to the field of control of an AC servo system of a permanent magnet synchronous motor, and in particular to a speed control method of an AC servo system of a permanent magnet synchronous motor based on an improved particle swarm algorithm to optimize an RBF fuzzy neural network PID. Background Art

[0002] Permanent magnet synchronous motor (PMSM) has the characteristics of simple structure, low energy consumption, good speed regulation performance, high operation reliability, and has been widely used in AC servo systems such as industrial robots and CNC machine tools in the field of industrial control. AC servo systems are generally systems that are controlled based on errors. Since the control structure and algorithm principle of ordinary PID regulators are relatively simple and easy to implement, and have certain control accuracy, they have been widely used in process control in the industrial field for a long time. However, ordinary PID control is suitable for clear model control objects or linear control systems, while permanent magnet synchronous motors are multivariable, time-varying parameters, and strongly coupled serious nonlinear systems. It is difficult to achieve the ideal control requirements and control effects of PMSM AC servo systems by adopting ordinary PID control. Fuzzy neural networks combine the logical reasoning function of fuzzy control and the self-learning ability of neural networks. They have powerful structural knowledge expression and the ability to adjust and optimize their own parameters. Many scholars have conducted research on this and achieved certain research results and effects. Based on the analysis of the mathematical model of permanent magnet synchronous motor, Zhu Qingchuan proposed a method to control the permanent magnet synchronous motor servo system using BP neural network; Wei Jiejie adopted the permanent magnet synchronous motor weak magnetic control strategy of fuzzy PI to overcome the problems of large motor speed overshoot and poor anti-interference performance; Wang Peiyu et al. designed a fuzzy radial basis function neural network PID controller to replace the speed loop PID control in the permanent magnet synchronous motor servo control system, and used the gradient descent method to adjust the speed loop PID parameters. Wang Bing et al. proposed a permanent magnet synchronous motor fuzzy PI control method using genetic algorithm to optimize fuzzy rules to improve the dynamic response performance of the motor servo system. Qiao Weide proposed to apply the recursive fuzzy neural controller optimized by immune genetic algorithm to the speed controller in the permanent magnet synchronous motor AC servo system, and achieved a certain control effect. Li Ruiqi et al. used radial basis function (RBF) neural network to establish a reference model for online identification of permanent magnet synchronous motor and adjustment of PID control parameters to enhance the adaptive ability of permanent magnet synchronous motor control system. Wu Feifei took the permanent magnet synchronous motor vector control system as the research object, combined fuzzy control and radial basis function neural network to form a fuzzy RBF neural network, and optimized the performance indicators of the permanent magnet synchronous motor control system by controlling the PID controller parameters. Liu Hongshan et al. proposed an adaptive control strategy based on radial basis function neural network for the speed control problem of permanent magnet synchronous motor. Wang Fujie conducted a comparative study on the two control schemes of permanent magnet synchronous motor fuzzy PID and BP neural network PID. The simulation analysis showed that the anti-disturbance and real-time performance of neural network PID control were better than fuzzy PID control. Hu Wentao et al. used BP neural network and particle swarm algorithm to optimize the driving performance of permanent magnet synchronous motor, significantly reducing the motor torque pulsation and vibration displacement amplitude.

[0003] The methods listed in the above research literature have achieved certain control effects in the permanent magnet synchronous motor servo system, but there are still defects and problems. For example, the BP algorithm and genetic algorithm have slow convergence speed, large amount of calculation, low search efficiency, and the particle swarm algorithm is prone to "premature maturity" and fall into local extreme values, and cannot achieve global optimization, thus affecting the speed control performance of the permanent magnet synchronous motor AC servo system. Summary of the invention

[0004] In order to overcome the above problems and shortcomings, and further improve the dynamic performance of the permanent magnet synchronous motor AC servo system, in view of the fact that the RBF neural network has stronger nonlinear mapping ability and global approximation performance than the BP neural network, based on a comprehensive analysis of the existing research literature results, the present invention proposes a PID controller based on a radial basis function (RBF) fuzzy neural network to replace the speed regulator in the permanent magnet synchronous motor AC servo system, and adopts an improved particle swarm algorithm (IPSO) to optimize the training of the fuzzy neural network, which can achieve more accurate permanent magnet synchronous motor speed control and meet the performance requirements of the permanent magnet synchronous motor AC servo control system. The technical solution of the present invention is:

[0005] A speed control method for a permanent magnet synchronous motor AC servo system. The system structure consists of a speed outer loop and a current inner loop. The speed outer loop is a speed controller, and the current inner loop is a current regulator. The current regulator adopts a general PI regulator. The input of the current loop depends on the control output of the speed loop. The speed loop is designed as a radial basis fuzzy neural network PID regulator. The control performance of the radial basis fuzzy neural network PID speed controller directly affects the overall control performance of the current loop and the permanent magnet synchronous motor AC servo system. The online learning of the improved particle swarm algorithm is used to continuously adjust the connection weights and other parameters of the RBF fuzzy neural network according to the system error objective function, and the optimal control parameter k of the PID controller is output. p , k i , k d , enhance the self-learning function of the system, ensure the real-time and stability of the neural network online training process, and thus meet the continuity and stability requirements of the system speed control.

[0006] Furthermore, the design of the RBF fuzzy neural network PID speed controller is divided into three parts: PID controller, RBF fuzzy neural network, and improved particle swarm algorithm optimization training. The parameters that need to be learned in the RBF fuzzy neural network structure include network connection weights ω jk ,ω ko And the central value m of the Gaussian membership function ij and width value δ ij , the improved particle swarm algorithm is used to adjust the RBF fuzzy neural network parameters online until the specified speed target error requirements are achieved.

[0007] Furthermore, the improved particle swarm algorithm is based on the basic particle swarm algorithm and optimizes the learning factor r 1 、r 2 And the particle's motion inertia weight ω, the specific improvement method is as follows:

[0008] The formula for iterating particle speed and position in the basic particle swarm algorithm is as follows:

[0009] V ij (t+1)=ωV ij (t)+r 1 *rand 1 *(P j (t)-S ij (t)) +r 2 *rand 2 *(G j (t)-S ij (t)) (5)

[0010] S ij (t+1)=S ij (t)+V ij (t+1) (6)

[0011] In formula (1) and (2), t is the number of optimization iterations; S ij (t) is the current position of the particle at generation t; V ij (t) is the velocity of the particle at generation t; ω is the inertia weight of the particle motion; r 1 is the individual learning factor, r 2 is the social learning factor; rand 1 、rand 2 It is any value between 0 and 1.

[0012] The particle speed and position depend on ω and r 1 、r 2 The present invention improves the basic particle swarm algorithm by making necessary adjustments to the three control parameters, ω adopts a new nonlinear decreasing adjustment strategy, r 1 、r 2 A nonlinear exponential adaptive dynamic adjustment method is adopted to better enhance and balance the global and local search capabilities of the particle swarm algorithm, accelerate the particle convergence speed, and overcome the "premature" phenomenon that is prone to occur in the basic particle swarm algorithm.

[0013]

[0014] In formulas (3), (4), and (5), t, t max are the current iteration number and the maximum iteration number of the particle, ω min ,ωmax are the minimum and maximum inertia weights, r 1max 、r 2max are the maximum values ​​of individual learning factor and social learning factor, r 1min 、r 2min are the minimum values ​​of individual learning factor and social learning factor, respectively.

[0015] The beneficial effects of the present invention are summarized as follows:

[0016] (1) Aiming at the nonlinearity and uncertainty in the permanent magnet synchronous motor AC servo system and the problem that the motor is easily affected by load disturbance during operation, the present invention proposes a speed controller based on RBF fuzzy neural network PID, and uses an improved particle swarm algorithm to online optimize the center value m of the Gaussian membership function of the RBF fuzzy neural network. ij and width value δ ij , the connection weight between the membership function layer and the rule layer ω jk , the connection weight between the rule layer and the output layer ω ko The improved particle swarm algorithm overcomes the shortcomings of the basic particle swarm algorithm and exhibits stronger global optimization ability and faster convergence speed.

[0017] (2) Compared with the common PID, BP algorithm optimized fuzzy neural network PID, and particle swarm algorithm optimized RBF fuzzy neural network PID control strategy, the improved particle swarm algorithm optimized RBF fuzzy neural network PID control strategy has the best control performance and can output the optimal parameter k of the PID controller. p , k i , k d It has the advantages of fastest dynamic response speed, best control stability, strongest resistance to load disturbance and robustness, and can largely meet the control requirements of permanent magnet synchronous motor AC servo system. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a schematic diagram of the principle of permanent magnet synchronous motor AC servo system.

[0019] Figure 2 This is a schematic diagram of the RBF fuzzy neural network structure.

[0020] Figure 3 The speed response of the system no-load permanent magnet synchronous motor with different control strategies.

[0021] Figure 4 Speed ​​response of permanent magnet synchronous motor for system load with different control strategies

[0022] Figure 5 The speed response of the variable speed permanent magnet synchronous motor for the system with different control strategies.

[0023] Figure 6 This is the experimental structure diagram of the permanent magnet synchronous motor AC servo system.

[0024] Figure 7 This is the experimental speed response waveform of the permanent magnet synchronous motor AC servo system. DETAILED DESCRIPTION

[0025] The present invention is further explained in detail below with reference to the accompanying drawings and embodiments.

[0026] The embodiment of the present invention describes a permanent magnet synchronous motor AC servo system based on an improved particle swarm algorithm to optimize the RBF fuzzy neural network PID. The basic structure of the system is as follows: Figure 1 As shown in the figure, the system structure consists of a speed outer loop and a current inner loop. The speed outer loop is a speed controller, which is designed as a radial basis fuzzy neural network PID regulator; the current inner loop is a current regulator, which adopts a general PI regulator. r is the given speed of the system, n is the actual speed of the PMSM, and e is the given speed n r The deviation from the actual speed n, ec is the rate of change of the deviation; x 1 and x 2 is the input of the RBF fuzzy neural network, and its domain interval value is [0,1]; the task of X mapping is mainly to map the actual e and ec into x with domain [0,1]. 1 and x 2 ; The function of F mapping is to transform the output k of the RBF fuzzy neural network with the domain [0,1] p , k i , k d Mapped to the proportional integral derivative parameter K required for the PID regulator to operate P , K I , K D The deviation between the desired motor current and the actual motor current generated by the PID regulator is regulated and controlled by the current regulator. Figure 1 J in it represents the system error objective function. By improving the online learning of the particle swarm algorithm, the connection weights and other parameters of the RBF fuzzy neural network are continuously adjusted according to the system error objective function to enhance the self-learning function of the system and ensure the real-time and stability of the neural network online training process, thereby meeting the continuity and stability requirements of the speed control of the permanent magnet synchronous motor AC servo system.

[0027] The present invention is introduced in depth from four aspects: RBF fuzzy neural network structure, improved particle swarm learning algorithm and optimized training process of RBF fuzzy neural network, AC servo system simulation experiment, and AC servo system experimental verification.

[0028] (1) RBF fuzzy neural network structure

[0029] The topological structure of the radial basis fuzzy neural network is Figure 2 As shown, there are 4 layers in total, namely input layer (i.e., layer i), fuzzification layer (i.e., layer j), fuzzy reasoning layer (i.e., layer k), and output layer (i.e., layer o).

[0030] ① Input layer. The input layer has two neuron nodes, x 1 and x 2 , they are directly passed to the next layer without processing. The input and output formulas of the input layer are as follows, where The input variable x 1 、x 2 , are e and ec respectively.

[0031]

[0032] ② The fuzzification layer, also known as the membership function layer, is responsible for fuzzifying the input variables on the nodes using linguistic variables. The input variables of this layer are all represented by fuzzy subsets. The fuzzy subsets have 7 fuzzy linguistic variables, namely {negative large, negative medium, negative small, zero, positive small, positive medium, positive large}, recorded as {ND, NM, NX, ZO, PX, PM, PD}. The output of this layer is the membership function of the input variables of each node of the corresponding fuzzy subset. The membership function selects Gaussian type [15-16] The membership function layer is designed to have 14 nodes. The operation rules of node output and input are shown in the following formula, where m ij is the central value of the Gaussian membership function, δ ij is the width of the Gaussian membership function.

[0033]

[0034] ③ Fuzzy reasoning layer, also known as fuzzy rule layer. Each node of the fuzzy rule layer corresponds to a fuzzy operation rule. This layer performs fuzzy logic reasoning operations between nodes through logical connection with the fuzzification layer. By analyzing the characteristics of the PMSM AC servo system and the actual needs of PMSM speed control, the PID regulator parameters need to follow the following logical conditions when adjusting: If the motor speed deviation e is large, in order to speed up the system response, first select a larger proportional coefficient k p1 and the smaller differential coefficient k d1 At the same time, in order to avoid large overshoot, the integral effect should be limited, and the integral coefficient k i Can be 0. If e is not too large, enter a smaller k first p , speed up the system response and limit the system overshoot. If e is relatively small, select a larger k p , k i ; When the deviation change rate ec is small, select a larger k d, and if ec is larger, choose a smaller k d , to improve system stability, where k p , k i , k d The fuzzy control rules of e and ec are shown in Tables 1 to 3 respectively.

[0035] Table 1k p Fuzzy rules with e and ec

[0036]

[0037] Table 2k i Fuzzy rules with e and ec

[0038]

[0039] Table 3k d1 Fuzzy rules with e and ec

[0040]

[0041] Through the analysis of the above fuzzy rule table, k p , k i , k d There are 49 fuzzy logic rules for each of them, so the three tuning parameters of the PID regulator generate a total of 147 fuzzy rules, that is, there are 147 node neurons corresponding to the fuzzy rule layer. The output formula of the fuzzy rule layer is shown in formula (3), where is the jth input of this layer, ω jk is the node connection weight between the fuzzy rule layer and the previous layer.

[0042]

[0043] ④ Output layer. The output layer has 3 output nodes. The output of each node is the sum of all outputs of the fuzzy reasoning layer after mathematical processing. The processing rules are shown in formula (4). is the kth input of the output layer, ω ko is the connection weight between the fuzzy inference layer and the output layer node, the output layer Output Corresponding to the PID regulator setting parameter k p , k i , k d .

[0044]

[0045] (2) Learning algorithm of RBF fuzzy neural network

[0046] The parameters that the RBF fuzzy neural network needs to learn include the network connection weight ωjk ,ω ko And the central value m of the Gaussian membership function ij and width value δ ij In view of the shortcomings of the basic particle swarm (PSO) algorithm, such as being prone to falling into local optimal solutions and lacking dynamic speed adjustment, the present invention proposes an improved particle swarm (IPSO) algorithm based on the basic particle swarm algorithm, and adopts the improved particle swarm algorithm to optimize the learning of RBF neural network. The formula for iterating particle speed and position in the basic particle swarm algorithm is as follows:

[0047] V ij (t+1)=ωV ij (t)+r 1 *rand 1 *(P j (t)-S ij (t)) +r 2 *rand 2 *(G j (t)-S ij (t)) (5)

[0048] S ij (t+1)=S ij (t)+V ij (t+1) (6)

[0049] In the formula, t is the number of optimization iterations; S ij (t) is the current position of the particle at generation t; V ij (t) is the velocity of the particle at generation t; ω is the inertia weight of the particle motion; r 1 is the individual learning factor, r 2 is the social learning factor; rand 1 、rand 2 It is any value between 0 and 1.

[0050] The particle speed and position depend on ω and r 1 、r 2 The present invention improves the basic particle swarm algorithm by making necessary adjustments to the three control parameters, ω adopts a new nonlinear decreasing adjustment strategy, r 1 、r 2 A nonlinear exponential adaptive dynamic adjustment method is adopted to better enhance and balance the global and local search capabilities of the particle swarm algorithm, accelerate the particle convergence speed, and overcome the "premature" phenomenon that is prone to occur in the basic particle swarm algorithm.

[0051]

[0052] In the formula, t, t maxare the current iteration number and the maximum iteration number of the particle, ω min ,ω max are the minimum and maximum inertia weights, r 1max 、r 2max are the maximum values ​​of individual learning factor and social learning factor, r 1min 、r 2min are the minimum values ​​of individual learning factor and social learning factor, respectively.

[0053] (3) Optimization training of RBF fuzzy neural network

[0054] The process of optimizing the training of RBF fuzzy neural network using improved particle swarm algorithm is as follows:

[0055] Step 1: Set relevant parameters, namely the initial position, velocity, group size, maximum and minimum inertia weights, maximum and minimum individual and social learning factors, maximum number of iterations, training target error, etc.

[0056] Step 2: Determine the topological structure of the RBF fuzzy neural network, and encode the connection weights and structural parameters such as the center and width of the radial basis function into real number vectors to represent individual particles in the population.

[0057] Step 3: According to the improved particle swarm algorithm formulas (5) to (9), find the individual extreme value and the group extreme value, and update the speed and position of the particles.

[0058] Step 4: Take the inverse of the absolute value of the error between the network output and the expected output as the particle fitness function value, calculate the particle fitness value, and search for the optimal position of the particle.

[0059] Step 5: If the current position reaches the maximum number of iterations of the particle swarm algorithm, the global optimal particle position is output and mapped to the ω of the RBF neural network. jk ,ω ko and m ij、 δ ij Wait for the optimal initial structural parameters, otherwise go to Step 3.

[0060] Step 6: Repeatedly train the RBF fuzzy neural network, continuously adjust the connection weights of the RBF neural network, etc., until the training error meets the target accuracy requirements, and output the optimal tuning parameter k of the PID regulator p , k i , k d .

[0061] (3) PMSM AC servo system simulation experiment

[0062] The simulation model of PMSM AC servo system was built using MTLAB / Simulink. The parameters of PMSM used for simulation were: rated voltage 220V, rated speed 1000r / min; stator phase winding resistance 0.85Ω, moment of inertia 0.0683kg·m 2 . The parameters of the improved particle swarm algorithm are: the group size is 300, the maximum number of iterations is 500, the maximum and minimum inertia weights are 0.85 and 0.35 respectively, the maximum and minimum individual learning factors are 2.85 and 1.65 respectively, the maximum and minimum social learning factors are 3.50 and 1.55 respectively, the RBF neural network structure is 2-14-49-3, and the training target error is 0.0001. During the simulation experiment, the current controller still adopts the traditional control PI regulator, while the speed controller adopts regulators of four different control strategies respectively. The regulator parameters adopted by different control strategies are shown in Table 4. When the AC servo system is under no-load and certain load conditions, simulation comparison experiments are carried out on different control strategies of the AC servo system to verify the superiority of the control strategy of the present invention.

[0063] Table 4 Regulator parameter values ​​for different control strategies

[0064]

[0065] ①1①Simulation analysis of the system under no-load condition

[0066] The PMSM motor speed response of the AC servo system in the no-load operation state is as follows: Figure 3 As shown in the figure, curves ①②③④ are the motor response curves under the action of PID regulator, fuzzy neural network PID regulator optimized by BP algorithm (BP-FNN PID for short), RBF fuzzy neural network PID regulator optimized by particle swarm algorithm (PSO-RBF-FNN PID for short), and RBF fuzzy neural network PID regulator optimized by improved particle swarm algorithm (IPSO-RBF-FNN PID for short).

[0067] from Figure 3The simulation analysis shows that the AC servo system has a maximum overshoot of 24.43% in the ordinary PID control under no-load operation, and the system is difficult to reach a relatively stable state; in the no-load operation of BP-FNN PID control, the system basically reaches a stable state in about 14.85s, and the maximum overshoot is 10.86%; under the no-load state of PSO-RBF-FNN PID control, the system needs 5.26s to reach a stable state, and its maximum overshoot is 2.12%; while under the no-load condition of IPSO-RBF-FNN PID control, the system only needs 4.35s to reach a stable state, and its overshoot is 0, and the speed response error approaches 0 after 4.35s. The response indicators of different control strategies under the no-load condition of the system are shown in Table 5. It can be seen that the control performance of the RBF fuzzy neural network PID regulator optimized by the improved particle swarm algorithm is the best among several different control strategies, indicating that the control strategy of this paper has a fast response speed and the best control stability.

[0068] Table 5 Response indicators of different control strategies in the system no-load state

[0069]

[0070] ②2②Simulation analysis under system load state

[0071] In order to analyze the response performance of the PMSM AC servo system under certain load conditions, a load of 50N·m is suddenly added after 15s of no-load operation. The system speed response curve is as follows: Figure 4 shown.

[0072] Depend on Figure 4 Observation and analysis show that when the load is suddenly added, due to the disturbance of the external load, the speed response of the ordinary PID control produces an adjustment change fluctuation of 4.56s, with a fluctuation amplitude of 182r / min; the BP-FNN PID control has a speed change adjustment fluctuation of 4.39s, with a fluctuation amplitude of 158r / min; the speed fluctuation adjustment time of the PSO-RBF-FNN PID control is 4.12s, with a fluctuation amplitude of 60r / min; the IPSO-RBF-FNN PID control has an adjustment fluctuation of only 2.57s, with a fluctuation amplitude of 32r / min. When the control system is adjusted for 2.58s, it returns to a stable state. From the above analysis, it can be seen that the RBF fuzzy neural network PID regulator control system optimized by the improved particle swarm algorithm has the strongest load disturbance resistance ability, showing its strong robustness. The response index values ​​of different control strategies under system load conditions are shown in Table 6.

[0073] Table 6 Response indicators of different control strategies under system load conditions

[0074]

[0075] ③3③Simulation analysis of the system under variable speed state

[0076] When the system is under load condition, the motor speed is reduced from 1000r / min to 600r / min at t=22s. The speed response of the PMSM system under the above four control strategies is as follows: Figure 5 As shown. Figure 5 It is obvious that the motor variable speed tracking response curve under the action of the RBF fuzzy neural network PID regulator optimized by the improved particle swarm algorithm has the fastest dynamic response speed and the strongest robustness.

[0077] (4) Experimental verification of PMSM AC servo system

[0078] In order to verify the control performance of the RBF fuzzy neural network PID regulator optimized by the improved particle swarm algorithm, the PMSM experimental platform structure is as follows: Figure 6 As shown. DSP TMS320F2812 is used as the control chip, which is mainly responsible for data acquisition, signal processing and generating PWM signals to control the drive circuit, so as to achieve the purpose of controlling PMSM. The peripheral circuits mainly include power drive circuit, inverter, current and speed detection circuit, host computer, oscilloscope, etc. The PMSM is given a speed of 1000r / min in the experiment. When it runs at no load for 15s, a load of 50N·m is suddenly added, and the system speed changes from 1000r / min to 600r / min at t=22s. The four control strategies used in the simulation experiment are respectively adopted for experimental verification. The experimental results are basically similar to the simulation results. The motor speed response curve under the action of the RBF fuzzy neural network PID regulator optimized by the improved particle swarm algorithm is shown as follows. Figure 7 As shown (the experimental curves of other control strategies are omitted). Figure 7 It can be concluded that the system takes a short time to reach a steady state when unloaded, which is only 4.4s, and the overshoot is almost zero. After adding the load, the PMSM speed fluctuation of the servo system is 36r / min, and it takes 3.0s of load adjustment time for the system to reach a stable state again, with good anti-disturbance. When the system speed changes, the motor has a faster speed dynamic response capability and strong robustness.

[0079] The above contents are only preferred embodiments of the present invention and do not impose any form of limitation on the present invention. Any simple modifications and equivalent changes made within the scope of the patent application of the present invention should be covered by the present invention.

Claims

1. A speed control method for a permanent magnet synchronous motor AC servo system, characterized in that: The AC servo system consists of a speed outer loop and a current inner loop. The speed outer loop is a speed controller, and the current inner loop is a current regulator. The current regulator adopts a general PI regulator, and the speed loop is designed as an RBF fuzzy neural network PID regulator. The control performance of the RBF fuzzy neural network PID speed controller directly affects the overall control performance of the current loop and the permanent magnet synchronous motor AC servo system. The RBF fuzzy neural network is optimized using an improved particle swarm algorithm to output the optimal parameter k of the PID controller. p , k i , k d .

2. The speed control method of a permanent magnet synchronous motor AC servo system according to claim 1, characterized in that: The design of the RBF fuzzy neural network PID speed controller is divided into three parts: PID controller, RBF fuzzy neural network, and improved particle swarm algorithm optimization training. The parameters that need to be learned in the RBF fuzzy neural network structure include network connection weights ω jk ,ω ko And the central value m of the Gaussian membership function ij and width value δ ij , the improved particle swarm algorithm is used to adjust the RBF fuzzy neural network parameters online until the specified speed target error requirements are achieved.

3. The speed control method of a permanent magnet synchronous motor AC servo system according to claim 2 is characterized in that: The improved particle swarm algorithm is based on the basic particle swarm algorithm, and optimizes and improves the learning factors r1, r2 and the particle's motion inertia weight ω. The specific improvement method is as follows: The formula for iterating particle speed and position in the basic particle swarm algorithm is as follows: V ij (t+1)=ωV ij (t)+r1*rand1*(P j (t)-S ij (t))+r2*rand2*(G j (t)-S ij (t))(5) S ij (t+1)=S ij (t)+V ij (t+1)(6) In formula (5) and (6), t is the number of optimization iterations; S ij (t) is the current position of the particle at generation t; V ij (t) is the speed of the particle at generation t; ω is the inertia weight of the particle motion; r1 is the individual learning factor, r2 is the social learning factor; rand1 and rand2 are arbitrary values ​​in the range of 0 to 1; The particle speed and position are dependent on the three parameters of ω, r1 and r2. The present invention improves the basic particle swarm algorithm by making necessary adjustments to the three control parameters. ω adopts a new nonlinear decreasing adjustment strategy, and r1 and r2 adopt a nonlinear exponential adaptive dynamic adjustment method, so as to better enhance and balance the global and local search capabilities of the particle swarm algorithm, accelerate the particle convergence speed, and overcome the "premature" phenomenon that is easy to occur in the basic particle swarm algorithm. In formulas (7), (8), and (9), t, t max are the current iteration number and the maximum iteration number of the particle, ω min ,ω max are the minimum and maximum inertia weights, r 1max 、r 2max are the maximum values ​​of individual learning factor and social learning factor, r 1min 、r 2min are the minimum values ​​of individual learning factor and social learning factor respectively.

Images