A permanent magnet synchronous motor speed regulation control method based on an improved cerebellar model neural network controller, a storage medium and equipment

Abstract

The application belongs to the technical field of motor speed regulation control, and designs a permanent magnet synchronous motor speed regulation control method based on an improved cerebellar model neural network controller, a storage medium and equipment, which comprises the following steps: firstly, a mathematical model of a permanent magnet synchronous motor is established, and then an improved cerebellar model neural network controller is designed and applied to a speed loop in a vector control structure of the permanent magnet synchronous motor; the controller improves the weight adjustment mode and learning rate of the original network, adopts a gradient descent method combined with an inverse relationship of learning times to update the weight of the cerebellar model neural network, replaces a calculation mode of division by a network generalization parameter C for average distribution, and improves the learning efficiency of the network; meanwhile, the value of the network learning rate is dynamically adjusted according to the motor operating state, and the rapidity and stability of the system are improved. Results show that the improved cerebellar model neural network controller effectively improves the response speed of the system and greatly improves the dynamic stability of the permanent magnet synchronous motor.

Description

Technical Field

[0001] This invention relates to the field of permanent magnet synchronous motor control technology, specifically to a speed control method, storage medium, and device for a permanent magnet synchronous motor based on an improved cerebellum model neural network controller. Background Technology

[0002] Permanent magnet synchronous motors (PMSMs) possess numerous advantages such as high efficiency, small size, and light weight, leading to their widespread application in electric vehicles, home appliances, and medical equipment. Currently, traditional PID control remains the most widely used method for speed control systems of PMSMs. However, due to the inherent time-varying, nonlinear, and strongly coupled characteristics of PMSMs, and the difficulty in automatically adjusting parameters during PID control, PID control cannot achieve ideal control results. This has brought intelligent control methods, such as neural network control, into focus for PMSM applications.

[0003] The Cerebellar Model Articulation Controller (CMAC) is a neural network based on local approximation. It possesses strong generalization and learning capabilities, and its learning speed is faster than other neural networks, making it highly suitable for real-time control. CMACs are often used in parallel with PID controllers in permanent magnet synchronous motor control systems to mitigate the effects of nonlinearity and parameter uncertainty on the control system, thereby improving its stability. However, in conventional CMACs, the error correction value is evenly distributed among all activated address units, which can lead to overlearning and slow down the network convergence. Furthermore, setting the learning rate too high or too low will affect the controller's learning speed and the motor's operational stability. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention provides an improved cerebellar model neural network controller for permanent magnet synchronous motor speed control, which effectively improves the dynamic performance of permanent magnet synchronous motor control and the learning speed of the network used.

[0005] The technical solution adopted by this invention to solve the technical problem is as follows:

[0006] A speed control method for a permanent magnet synchronous motor based on an improved cerebellum model neural network controller includes the following steps:

[0007] S1: Establish the state equation of the permanent magnet synchronous motor in the two-phase rotating coordinate system dq axis; assuming the direct axis current is set to 0, i.e., i d *The vector control strategy with 0 = 0 results in the dq-axis current being processed by two PI controllers in the current loop to obtain the dq-axis voltage components in the two-phase rotating coordinate system. The three-phase current i is obtained after inverse Park transform, vector pulse width modulation, and inverter. a i b i c Then, the actual output dq-axis current i of the motor is obtained through Clark transformation and Park transformation. d i q Based on the actual motor speed n rpm and a given rotational speed n ref The difference is defined as the speed tracking error of a permanent magnet synchronous motor, e(k) = n ref -n rpm The improved cerebellar model neural network controller of the velocity loop is fed back to obtain the control output u(k) of the velocity loop, and i q * =u(k), realizing dual closed-loop control of permanent magnet synchronous motor; the improved cerebellum model neural network controller includes two parts: a cerebellum model neural network controller and a PD controller; in the cerebellum model neural network controller part, the given speed n is... ref actual motor speed n rpm As the input to the cerebellar model neural network controller, the output u of the cerebellar model neural network controller is obtained sequentially through input linear quantization, concept mapping algorithm, cerebellar model neural network, and output calculation. nn (k) In the PD controller section, a positional PD is used to obtain the PD control output u. PD (k), u nn (k) and u PD The sum of the two outputs (k) yields the total output u(k) of the improved cerebellar model neural network controller;

[0008] S2: Update the weights ω of the cerebellar model neural network. i and number of learning times f i The gradient descent method combined with the inverse relationship of the number of learning iterations is used to replace the calculation method of dividing by the average distribution of the network generalization parameter C; at the same time, the value of the network learning rate is dynamically adjusted according to the motor operating status to improve the speed and stability of the system.

[0009] S3: Set an acceptable error range Δ, when ||u(k)-u nn When (k)||≥Δ, continuously update the weights until the set precision value is reached;

[0010] Furthermore, in step S1:

[0011] Ignoring the effects of rotor magnetic circuit saturation and core losses in the motor, the state equations of the permanent magnet synchronous motor in the two-phase rotating coordinate system dq-axis are established as follows:

[0012]

[0013] Among them, u d u q These are the dq-axis components of the stator voltage, i d i q These are the dq-axis components of the stator current, where R is the stator resistance and ω is the dq-axis component. e L is the electric angular velocity. d L q These are the dq-axis components of the stator inductance, ψ f It is a permanent magnet flux chain. For the differentiation operation, T e For electromagnetic torque, P n Where J is the number of pole pairs of the permanent magnet synchronous motor, ω is the moment of inertia, and J is the number of pole pairs of the motor. m T is the rotor's mechanical angular velocity. L Where B is the load torque and B is the damping coefficient;

[0014] Furthermore, the improved cerebellar model neural network controller is specifically as follows:

[0015] Input linear quantization is:

[0016]

[0017] Among them, X i S i Let Q be the i-th dimension input, and let Q be the input quantization value. i Let X be the quantization level of the input. imax X imin These are the maximum and minimum values ​​of the i-th dimension input, respectively, and floor() is the floor function.

[0018] The concept mapping algorithm is represented as:

[0019]

[0020] Where r is the address of the input vector in space M, C is the network generalization parameter, ceil(·) is the floor function; k and l represent the mapping dimension; Q l S represents the quantization level of the l-th dimension of the input; i represents the input dimension; S k S represents the quantized value of the k-th dimension input; n This represents the quantized value of the nth dimension input;

[0021] The output u controlled by the cerebellar model neural network nn (k) is:

[0022]

[0023] Where, α i ω is the binary selection vector, C is the network generalization parameter, i.e., the number of activated address units in a cerebellar model neural network during one learning cycle. i (k) represents the weights of the cerebellar model neural network at time k, where the i-th address unit is located.

[0024] Output u of position-type PD control PD (k) is:

[0025] u PD (k)=K p e(k)+K d [e(k)-e(k-1)](7)

[0026] Among them, K p K d These are the proportional coefficient and differential coefficient of a positional PD, respectively;

[0027] The controller is designed using the PD algorithm.

[0028] Adding equations (4) and (5), we obtain the total output u(k) of the cerebellar model neural network controller as follows:

[0029] u(k)=u nn (k)+u PD (k)(8)

[0030] Take i q * =u(k);

[0031] The tuning metrics for the cerebellar model neural network are:

[0032]

[0033] Where E(k) represents the network error at time k;

[0034] Furthermore, step S2 is as follows:

[0035] The weights ω of the cerebellar model neural network are updated using gradient descent combined with the inverse relationship of the number of learning iterations. i (k), the specific method is as follows:

[0036]

[0037] ω i (k)=ω i (k-1)+Δω i (k) (11)

[0038] Where, Δω i β(k) is the weight correction amount of the cerebellar model neural network of the i-th address unit at time k; β(k) is the network learning rate at time k, and the learning rate at time zero is β(0), β(k)∈(0,1); σ is the balance learning coefficient, which is used to improve the learning efficiency of the network.

[0039] The network learning rate β(k) is dynamically adjusted according to the motor's operating status. The specific implementation method is as follows:

[0040]

[0041] Where ρ is the adaptive gain adjustment coefficient, ρ>0, ω e (k) represents the electric angular velocity of the motor rotor at time k, T L (k), T L (k-1) represent the load torque of the motor at time k and time k-1, respectively;

[0042] In the cerebellar model neural network, the weights ω i (k) and number of learning times f i The initial values ​​are all set to 0; the maximum number of learning iterations, f, is set. max When f is satisfied i <f max At that time, f i Add 1;

[0043] Furthermore, step S3 is as follows:

[0044] Set an acceptable error range Δ, specifically:

[0045]

[0046] Among them, i qmax The maximum desired current along the q-axis;

[0047] A computer-readable storage medium storing at least one instruction, at least one program, a code set, or an instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the above-described speed control method for a permanent magnet synchronous motor based on an improved cerebellum model neural network controller.

[0048] A server includes a processor and a memory, the memory storing at least one instruction, which is loaded and executed by the processor to implement the above-described speed control method for a permanent magnet synchronous motor based on an improved cerebellum model neural network controller.

[0049] The beneficial effects of this invention are:

[0050] This invention rationally allocates weights to correct errors based on the different learning counts (i.e., confidence levels) of each activated address unit, avoiding the average distribution of weight corrections and improving the network's learning efficiency. Simultaneously, a balanced learning coefficient is introduced to further optimize the network structure. Finally, the network learning rate is adaptively processed, enabling the motor to maintain good dynamic performance under various complex operating conditions. This method combines a classic PID controller with a cerebellum model neural network controller. The cerebellum model neural network control implements feedforward control, realizing the inverse dynamic model of the controlled object, while the PID control implements feedback control, ensuring system stability while suppressing disturbances. The design process is simple and efficient. Attached Figure Description

[0051] Figure 1 It is a speed-current dual closed-loop vector control framework for permanent magnet synchronous motors, where u α u β For the voltage components along the α-β axes in the stationary coordinate system, i α i β For the α-β axis current components in the stationary coordinate system, θ e 1 / S is the electric angle of the motor rotor.

[0052] Figure 2 This is a schematic diagram of the improved cerebellar model neural network controller structure of the present invention.

[0053] Figure 3 The speed loops show the motor speeds under sudden load increases and decreases when using PI control, conventional CMAC control, model-free sliding mode control (MFSMC), and improved cerebellar model neural network control (MC-CMAC).

[0054] Figure 4 These are the motor torque response curves for sudden load increases and decreases under different speed loop control methods: PI control, conventional CMAC control, model-free sliding mode control (MFSMC), and improved cerebellar model neural network (MC-CMAC). Detailed Implementation

[0055] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0056] like Figure 1 As shown, the permanent magnet synchronous motor control achieves high-performance control by improving the speed loop controller.

[0057] Figure 2 This is a schematic diagram of the improved cerebellar model neural network controller (MC-CMAC) of the present invention, which includes two parts: cerebellar model neural network control and PD control.

[0058] In this embodiment, a MATLAB simulation model is used to simulate the speed control method of permanent magnet synchronous motor based on an improved cerebellum model neural network according to the present invention.

[0059] The parameters of the MATLAB simulation model are shown in Table 1 below:

[0060] Table 1 Simulation Model Parameters

[0061]

[0062] The speed loop comparison curves for sudden load increases and decreases of the motor using PI control, conventional CMAC control, model-free sliding mode control (MFSMC), and the improved cerebellar model neural network control (MC-CMAC) of this invention are shown below. Figure 3 As shown, the torque response curve is as follows: Figure 4 As shown.

[0063] from Figure 3 As can be seen, the PI control motor exhibits a speed overshoot of 141 r / min upon startup, which drops to 148 r / min after a sudden load increase, with a speed recovery time of 0.22 s. After a sudden load decrease, the speed changes by 73 r / min, with a speed recovery time of 0.18 s. The MFSMC control motor shows no speed overshoot upon startup, with speed changes of 63 r / min and 20 r / min after sudden load increases and decreases, respectively, with speed recovery times of 0.06 s and 0.05 s. The CMAC control motor shows no speed overshoot upon startup, with speed changes of 58 r / min and 30 r / min after sudden load increases and decreases, respectively, with speed recovery times of 0.06 s and 0.05 s. Under the MC-CMAC method, the system speed shows no speed overshoot after no-load startup and reaches the given speed faster than CMAC and MFSMC. After sudden load increases and decreases, the speed changes by 38 r / min and 20 r / min, respectively, with speed recovery times of 0.03 s and 0.03 s, respectively, representing the shortest speed recovery time.

[0064] from Figure 4 As can be seen, when the load is suddenly increased or decreased, the torque recovery time of the motor controlled by the PI method is 0.2s and 0.16s, respectively; the torque recovery time of the MFSMC control method is 0.06s and 0.04s; the torque recovery time of the CMAC control method is 0.07s and 0.06s; and the torque recovery time of the MC-CMAC control method is 0.03s and 0.028s. The MC-CMAC method recovers to the pre-disturbance state faster and shows better resistance to load disturbances.

Claims

1. A speed control method for a permanent magnet synchronous motor based on an improved cerebellum model neural network controller, characterized in that, Includes the following steps: S1: Establish the permanent magnet synchronous motor in a two-phase rotating coordinate system The state equation under the shaft; using a setting of the direct-axis current as 0, i.e. Vector control strategy, The shaft current passes through two PI controllers in the current loop to obtain a two-phase rotating coordinate system. Axis voltage components , After the reverse The three-phase current is obtained through transformation, vector pulse width modulation, and inverter. Then through Transformation and Transformation to obtain the actual output of the motor shaft current Based on the actual speed of the motor and given speed The difference is defined as the speed tracking error of a permanent magnet synchronous motor. The improved cerebellar model neural network controller of the velocity loop receives the control output of the velocity loop. ,make This achieves dual closed-loop control of the permanent magnet synchronous motor; the improved cerebellum model neural network controller comprises two parts: a cerebellum model neural network controller and a PD controller; in the cerebellum model neural network controller part, the given speed is... actual motor speed As the input to the cerebellar model neural network controller, the output of the cerebellar model neural network control is obtained sequentially through input linear quantization, concept mapping algorithm, cerebellar model neural network, and output calculation. In the PD controller section, a position-type PD is used to obtain the PD control output. , and The sum of the two outputs yields the total output of the improved cerebellar model neural network controller. ; S2: Update the weights of the cerebellar model neural network and number of times of learning The gradient descent method combined with the inverse relationship of the number of learning iterations is used to replace the calculation method of dividing by the average distribution of the network generalization parameter C; at the same time, the value of the network learning rate is dynamically adjusted according to the motor operating status to improve the speed and stability of the system. S3: Set an acceptable error range ,when The weights are continuously updated until the set precision value is reached. The improved cerebellum model neural network controller is as follows: Input linear quantization is: (4) in, , For the first Dimensional input, input quantization value, The quantization level of the input. The first The maximum and minimum values ​​of the input are specified, and floor() is the floor function. The concept mapping algorithm is represented as: (5) in, Let M be the address of the input vector in space M, and C be the network generalization parameters. It is a rounding function; , Indicates the dimension of the mapping; Indicates the first The quantization series of the input dimension; Indicates the input dimension; Indicates the first Dimensional input quantization value; Indicates the first Dimensional input quantization value; Cerebellum model neural network control output for: (6) in, Here, C is the binary selection vector, and C is the network generalization parameter, which is the number of activated address units in a cerebellar model neural network during one learning cycle. for Time of day Weights of a cerebellar model neural network with one address unit; Output of positional PD control for: (7) in, , These are the proportional coefficient and differential coefficient of a positional PD, respectively; The controller is designed using the PD algorithm. Adding equations (4) and (5) together yields the total output of the cerebellar model neural network controller. for: (8) Pick ; The tuning metrics for the cerebellar model neural network are: (9) in, express Timing network error.

2. The speed control method for a permanent magnet synchronous motor based on an improved cerebellum model neural network controller according to claim 1, characterized in that, In step S1: Ignoring the effects of rotor magnetic circuit saturation and core losses in the motor, the state equations of the permanent magnet synchronous motor in the two-phase rotating coordinate system dq-axis are established as follows: (1) (2) (3) in, , These are the dq-axis components of the stator voltage, respectively. , These are the dq-axis components of the stator current, and R is the stator resistance. Electric angular velocity, , These are the dq-axis components of the stator inductance. It is a permanent magnet flux linkage. For the differentiation operation, For electromagnetic torque, Where J is the number of pole pairs of the permanent magnet synchronous motor, and J is the moment of inertia. The rotor's mechanical angular velocity, Where is the load torque and B is the damping coefficient.

3. The speed control method for a permanent magnet synchronous motor based on an improved cerebellum model neural network controller according to claim 1, characterized in that, The process of step S2 is as follows: The weights of the cerebellar model neural network are updated using gradient descent combined with the inverse relationship of the number of learning iterations. The specific method is as follows: (10) (11) in, for Time of the first Weight correction amount of a cerebellar model neural network per address unit; for The network learning rate at time 0, and the learning rate at time 0. , ; To balance the learning coefficients, thereby improving the learning efficiency of the network; The network learning rate is dynamically adjusted based on the motor's operating status. The specific implementation method for determining the value of is as follows: (12) in, This is the adaptive gain adjustment coefficient. , for At any given moment, the electric angular velocity of the motor rotor, These respectively indicate the motor in Time and Load torque at any given moment; Weights in a cerebellar model neural network and number of times of learning The initial values ​​are all set to 0; the maximum number of learning iterations is set. When satisfied hour, Add 1.

4. The speed control method for a permanent magnet synchronous motor based on an improved cerebellum model neural network controller according to claim 1, characterized in that, The process of step S3 is as follows: Set an acceptable error range Specifically: (13) in, This represents the maximum expected current along the q-axis.

5. A computer-readable storage medium, characterized in that, The storage medium stores at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the speed control method for a permanent magnet synchronous motor based on an improved cerebellum model neural network controller as described in any one of claims 1-4.

6. A server, characterized in that, The server includes a processor and a memory, the memory storing at least one instruction, which is loaded and executed by the processor to implement the speed control method for a permanent magnet synchronous motor based on an improved cerebellum model neural network controller as described in any one of claims 1-4.

Images