Permanent magnet synchronous motor servo control method based on nonlinear control strategy

Through nonlinear control strategy and reinforcement learning algorithm, self-optimizing load torque estimation and feedback controller bandwidth factor, the failure of the servo control system of permanent magnet synchronous motor under multiple types and multiple intensity load torques is solved, and the robustness is improved and immunity, achieving flexible dynamic response and steady-state accuracy.

CN118944506APending Publication Date: 2024-11-12HAINAN UNIV
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202410995959.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-24
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The existing permanent magnet synchronous motor servo control system is difficult to ensure control performance when facing multiple types and multiple strength load torques, and its robustness and immunity are insufficient, especially in large ranges of load torques or external interference. In the case of signal type switching, the estimated loop performance is poor.

Method used

Using a nonlinear control strategy method, reinforcement learning algorithm is used to self-optimize parameters, combined with homogeneous high gain estimator and deep reinforcement learning parameter setter, a nonlinear non-recursive controller is designed to independently schedule load torque estimation loop and feedback The bandwidth factor of the controller achieves improved system adaptability and robustness.

Benefits of technology

Under the conditions of multiple types and multiple strength load torques, the control performance and robustness of the servo system of permanent magnet synchronous motor is improved, the immunity is enhanced, the robustness is avoided, and the flexible dynamic mechanism and steady-state accuracy are achieved. .

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118944506A_ABST
    Figure CN118944506A_ABST
Patent Text Reader

Abstract

The invention relates to a permanent magnet synchronous motor servo control method based on a nonlinear control strategy. The method comprises the following steps: S1, constructing a mathematical model of a permanent magnet synchronous motor servo system; s2, constructing an estimation error dynamic system of the observer; s3, acquiring an expected mathematical output value; s4, acquiring a system after coordinate transformation; s5, obtaining a reference nonlinear non-recursive control law according to the coordinate transformation and the expected mathematical output value; s6, obtaining an output execution law based on the reference nonlinear non-recursive control law; and S7, outputting six paths of PWM signals to control a converter corresponding to the motor based on the setting. Compared with the prior art, the method has the advantages.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of permanent magnet synchronous motor servo control, and in particular to a permanent magnet synchronous motor servo control method based on a nonlinear control strategy. Background Art

[0002] Permanent Magnet Synchronous Motor (PMSM) is well-known for its high efficiency. Compared with traditional induction motors, PMSM has less loss when converting electrical energy into mechanical energy, so it can reduce energy consumption and improve the energy utilization of the system. In today's pursuit of energy conservation and environmental friendliness, the application of PMSM has received more and more attention. The servo control system enables the system to respond to external commands quickly and accurately by monitoring and adjusting the position, speed, acceleration and other parameters of the motor in real time. Permanent magnet synchronous motor servo control technology can achieve higher precision position control and faster dynamic response, so it is widely used in applications that require high-performance motion control, such as robots, CNC machine tools, printing equipment, etc. Considering the important position of PMSM in industrial production, the optimization of its servo control has also become the focus of academic attention.

[0003] The existing solutions have the following shortcomings:

[0004] 1. From the perspective of the more traditional dual closed-loop PI control strategy, its loop design is simple, the design parameter selection mechanism is concise and reliable, and it has strong applicability. However, its design parameters kp and ki are considered to be local optimal values. When the load torque changes of the permanent magnet synchronous motor servo control system are complex, it is difficult to ensure that the control performance meets the expected settings.

[0005] 2. Some existing advanced controller designs require the controlled PMSM to have high modeling accuracy, and the robustness to parameter perturbation is poor. In industrial applications, the quality of sensors will inevitably cause different degrees of measurement noise. Temperature and humidity in the operating scenario may cause controller failure. The age of use will also cause the nameplate parameters of the PMSM to shift, which will cause a decrease in control accuracy.

[0006] 3. The existing strategies based on real-time estimation and feedback of load torque guarantee the robust performance of permanent magnet synchronous motor servo control systems to a certain extent, but most of them only focus on the design of optimized control strategies for closed-loop systems after obtaining feedback information, while ignoring the design optimization of the load torque estimation loop. The quality of the estimated information also directly affects the performance of the closed-loop system.

[0007] In response to the first deficiency, the patent (CN113556068B, a permanent magnet synchronous motor servo control method and servo system, 2021.7.12) introduced a third-order extended state observer to identify and compensate for the system disturbance and current load conditions, and use expert experience to determine the fuzzy reasoning rules and establish a fuzzy rule setting table for the parameters. Subsequently, the disturbance identification information is used to adjust the permanent magnet synchronous motor PID control parameters, thereby improving the performance of the servo control system. In addition, the patent (CN111934585A, a permanent magnet synchronous motor servo control system based on fuzzy PI control, 2020.6.23) proposes a direct fuzzy PID control strategy that integrates PI control and fuzzy control, taking the best of both, improving the system's anti-interference ability, and effectively improving the steady-state and dynamic performance of the permanent magnet synchronous motor servo control system.

[0008] Regarding the second point, the patent (CN107171609A, a method and system for identifying the moment of inertia of a permanent magnet synchronous motor, 2017.2.15) proposes a method for identifying the moment of inertia. By constructing a rotor speed observer and inputting a model reference adaptive identification algorithm, the identified moment of inertia is obtained, which improves the stability of identification and user experience. The patent (CN114744946B, a method and system for controlling a permanent magnet synchronous motor based on a width learning disturbance observer, 2022.5.18) constructs a width learning disturbance observer by acquiring the input and output data of the disturbance observer under different control states of the permanent magnet synchronous motor over a certain period of time, and performs online compensation for the disturbance of the permanent magnet synchronous motor control, thereby achieving high-performance operation of the motor under multiple working conditions and reducing parameter setting adjustments.

[0009] Regarding the third point, the literature (Zhang Xiaozhen, et al. Permanent magnet fault-tolerant vernier rim propulsion motor position sensorless control system based on adaptive sliding mode observer) constructed an adaptive sliding mode observer to estimate the rotor position to improve the control accuracy, and the system has good dynamic and static performance.

[0010] However, the above strategy does not take into account the situation of large-scale load torque or external interference, and switching of multiple signal types, so the estimated loop may have poor performance, which will bring a performance control burden to the feedback loop. Summary of the invention

[0011] The purpose of the present invention is to realize autonomous scheduling of parameters to meet the requirements of testing permanent magnet synchronous motors under multiple types and multiple intensities of load torque to ensure better performance, and to provide a permanent magnet synchronous motor servo control method based on a nonlinear control strategy. By utilizing the parameter self-optimization capability granted by the reinforcement learning algorithm, the bandwidth factor of the external load torque estimation loop and the feedback controller design loop can be jointly and autonomously scheduled to the most appropriate value.

[0012] The purpose of the present invention can be achieved by the following technical solutions:

[0013] A permanent magnet synchronous motor servo control method based on a nonlinear control strategy, the method comprising the following steps:

[0014] S1. Construct a mathematical model of the permanent magnet synchronous motor servo system;

[0015] S2. Based on the mathematical model, the system disturbance is reconstructed using a nonlinear high-bandwidth gain load torque and disturbance estimator to construct an estimation error dynamic system of the observer;

[0016] S3, based on the estimated error dynamic system of the observer, a mathematical steady-state model of the system is established to obtain the expected mathematical output value;

[0017] S4, introducing coordinate transformation of the mathematical model through the mathematical steady-state model to obtain a system after coordinate transformation;

[0018] S5. Obtaining a benchmark nonlinear non-recursive control law according to the coordinate transformation and the expected mathematical output value;

[0019] S6. Obtaining an output execution law based on a benchmark nonlinear non-recursive control law;

[0020] S7. A homogeneous high-gain estimator, a reinforcement learning parameter setter and a nonlinear non-recursive controller corresponding to the estimated error dynamic system are set in the permanent magnet synchronous motor system. When the permanent magnet synchronous motor is working, the actual feedback angle position of the motor is obtained and converted into an actual angular velocity. At the same time, the actual d-axis current is obtained. The reinforcement learning parameter setter outputs the set parameters l and τ to the homogeneous high-gain estimator according to the current state, and outputs the set parameters τ and ι to the nonlinear non-recursive controller. The differences between the reference angle position, the reference angular velocity and the reference d-axis current and the actual feedback angle position, the actual angular velocity and the actual d-axis current are calculated, respectively. The calculated differences are input into the nonlinear non-recursive controller. The nonlinear non-recursive controller outputs an executable q-axis voltage. The executable q-axis voltage is input into the homogeneous high-gain estimator. The homogeneous high-gain estimator feeds back the estimated parameters to the nonlinear non-recursive controller. The executable q-axis voltage and the d-axis voltage are input into the transformation module and the SVPWM module. The output PWM signal controls the three-phase inverter circuit connected to the motor.

[0021] Furthermore, the expected mathematical output value is:

[0022]

[0023] Among them, θ ref represents the reference angle position, B represents the viscous friction coefficient, J represents the moment of inertia, z 1,3 、z 1,2and z 2,2 They represent the estimated value of the first-order derivative of the load torque, the estimated value of the load torque and the estimated value of the lumped disturbance w2, respectively. p represents the pole pair number, ψ f represents the rotor flux, L represents the stator inductance, R s Indicates the stator resistance.

[0024] Furthermore, the reference nonlinear non-recursive control law is:

[0025]

[0026] Among them, ι and τ are the parameter values ​​to be designed output by the reinforcement learning parameter setter, k1, k2, k3 are the design gain values,

[0027]

[0028] Where ω represents the actual angular velocity, i q represents the q-axis current.

[0029] Furthermore, the estimation error dynamic system of the observer is:

[0030]

[0031] in,

[0032]

[0033] Symbol z 1,1 , z 1,2 , z 1,3 , z 1,4 are the estimated values ​​of speed ω, load torque and their first and second order derivatives respectively; z 2,1 , z 2,2 , z 2,3 They are x3, the estimated values ​​of the lumped disturbance w2 and its first-order derivative, and w1 is the load disturbance of the permanent magnet synchronous motor system.

[0034] Furthermore, the estimated parameters fed back by the homogeneous high gain estimator to the nonlinear non-recursive controller are:

[0035] Furthermore, the reinforcement learning parameter setter includes an agent, and the agent is trained using the DDPG algorithm. The training process of the agent is:

[0036] The current state information for training is obtained. The action network of the agent outputs the current action based on the current state information for training. The reward function is output based on the current action. The action network is trained based on the reward function. The above steps are repeated until a trained reinforcement learning parameter setter is obtained.

[0037] Furthermore, the current status information is:

[0038] S t ={θ,θ-θ ref ,w1-z 1,2 ,w2-w 2,2}.

[0039] Furthermore, the current action is: a t ∈{a min ;c;a max}, where a min , a max are the lower and upper bounds of the parameters l, ι and τ respectively, and c is the minimum incremental step size of the action set.

[0040] Furthermore, the reward function is:

[0041]

[0042] Among them, R is the reward function.

[0043] Furthermore, the number of hidden layers of the action network is set to two, the number of nodes in the first hidden layer is set to 64, and the number of nodes in the second hidden layer is also 64, and the activation functions of the hidden layer and the output layer are two ReLU functions and one linear function.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] The present invention utilizes the parameter self-optimization capability given by the reinforcement learning algorithm in the traditional nonlinear non-recursive controller, and the bandwidth factor of the external load torque estimation loop (including the homogeneous high-gain estimator) and the feedback controller design loop can be jointly and autonomously scheduled to the most appropriate value to meet the test of the permanent magnet synchronous motor under multiple types and multiple strengths of load torque to ensure better performance requirements. At the same time, the homogeneous high-gain estimator of the present invention is also obtained through reinforcement learning, which improves the adaptability of the permanent magnet synchronous motor servo system under different load torque conditions, and further enhances the anti-interference ability through the parameter self-configuration mechanism, ensuring a certain robustness of the robust adaptive enhanced permanent magnet synchronous motor servo control system, and the flexible dynamic mechanism avoids robustness redundancy as much as possible. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a flowchart of the present invention;

[0047] Figure 2 The load torque is cos(t)+sin(t); 0≤t≤1, and the expected position is θ ref = Actual position response curve when cos(t)+sin(t);

[0048] Figure 3 The load torque of the present invention is cos(t)+sin(t); 1≤t≤2, and the expected position is θ ref = Actual position response curve when 0;

[0049] Figure 4 The load torque of the present invention is cos(t)+sin(t); 2≤t≤3, and the expected position is θ ref =2e -t The actual position response curve when

[0050] Figure 5 The load torque of the present invention is cos(t)+sin(t); 3≤t≤4, and the expected position is θ ref The actual position response curve when =2. DETAILED DESCRIPTION

[0051] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0052] The present invention proposes a permanent magnet synchronous motor servo control method based on a nonlinear control strategy. The flow chart of the method is as follows: Figure 1 As shown, the method comprises the following steps:

[0053] S1. Construct a mathematical model of the permanent magnet synchronous motor servo system;

[0054] S2. Based on the mathematical model, the system disturbance is reconstructed using a nonlinear high-bandwidth gain load torque and disturbance estimator to construct an estimation error dynamic system of the observer;

[0055] S3, based on the estimated error dynamic system of the observer, a mathematical steady-state model of the system is established to obtain the expected mathematical output value;

[0056] S4, introducing coordinate transformation of the mathematical model through the mathematical steady-state model to obtain a system after coordinate transformation;

[0057] S5. Obtaining a benchmark nonlinear non-recursive control law according to the coordinate transformation and the expected mathematical output value;

[0058] S6. Obtaining an output execution law based on a benchmark nonlinear non-recursive control law;

[0059] S7. A homogeneous high-gain estimator, a reinforcement learning parameter setter and a nonlinear non-recursive controller corresponding to the estimated error dynamic system are set in the permanent magnet synchronous motor system. When the permanent magnet synchronous motor is working, the actual feedback angle position of the motor is obtained and converted into an actual angular velocity. At the same time, the actual d-axis current is obtained. The reinforcement learning parameter setter outputs the set parameters l and τ to the homogeneous high-gain estimator according to the current state, and outputs the set parameters τ and ι to the nonlinear non-recursive controller. The differences between the reference angle position, the reference angular velocity and the reference d-axis current and the actual feedback angle position, the actual angular velocity and the actual d-axis current are calculated, respectively. The calculated differences are input into the nonlinear non-recursive controller. The nonlinear non-recursive controller outputs an executable q-axis voltage. The executable q-axis voltage is input into the homogeneous high-gain estimator. The homogeneous high-gain estimator feeds back the estimated parameters to the nonlinear non-recursive controller. The executable q-axis voltage and the d-axis voltage are input into the transformation module and the SVPWM module. The output PWM signal controls the three-phase inverter circuit connected to the motor.

[0060] In the above steps, step 1 is to construct a mathematical model of the permanent magnet synchronous motor servo system.

[0061] First, according to the physical model of the permanent magnet synchronous motor servo system, the motor's feedback angle position θ and the motor's feedback real-time angular velocity ω are obtained through the angular position sensor on the PMSM rotor, and real-time current feedback is performed through the hardware analog-to-digital converter (ADC). Then, the Clark / Park transformation is used to obtain the q-axis and d-axis current i q ,i d At this point, the mathematical model M1 of PMSM in the dq coordinate system can be constructed, which is specifically in the following form:

[0062]

[0063] Where J = 7.066 × 10 -6 kg·m 2 is the moment of inertia; n p =4 is the number of pole pairs; ψ f =0.0064Wb is the rotor flux; B = 2.637×10 -6 N·m·s / rad is the viscous friction coefficient; T L is the load torque; L s =2×10 -4 mH and R s =0.36Ω are stator inductance and stator resistance respectively; u d ,u q are the d-axis and q-axis voltages in the dq coordinate system.

[0064] In order to obtain the maximum torque-to-current ratio, i d =id * = 0 is usually regarded as the most ideal situation, and the system M2 is:

[0065]

[0066] Therefore, according to system M2, using the feedback angle position θ, the feedback real-time speed ω of the motor and the instantaneous q-axis current i q The information of the mathematical system is constructed as follows:

[0067]

[0068] After derivation, the mathematical model M3 of the permanent magnet synchronous motor servo system can be constructed as:

[0069]

[0070] The system output is expressed as:

[0071]

[0072] w1=-T L / J

[0073] is the load disturbance of the permanent magnet synchronous motor system, w2 is the model parameter perturbation considered in system modeling, and the high-order unmodeled dynamics are regarded as lumped disturbances. Considering that the control target is to adjust the position of the permanent magnet synchronous motor, the output of the system mathematical model is x1.

[0074] In step S2, the lumped interference term of the mathematical model is estimated.

[0075] Based on the mathematical permanent magnet synchronous motor servo control system constructed in step 1, the system disturbance is reconstructed using a nonlinear high-bandwidth gain load torque and disturbance estimator, which can be expressed in the following specific form:

[0076]

[0077] where α 1,1 , α 1,2 , α 1,3 , α 1,4 ; α 2,1 , α 2,2 , α 2,3 ; l; τ are the parameters of the designed nonlinear high-bandwidth gain interference estimator, l is a sufficiently large positive value; τ is a negative parameter, ranging from an open interval of -1 / 3 to 0, and the specific values ​​of these two parameters are obtained by the reinforcement learning algorithm in the following steps. sgn is a sign function, and sgn(0) is defined as 0; symbol z 1,1 , z 1,2 , z 1,3 , z1,4 are the estimated values ​​of speed ω, load torque and their first and second order derivatives respectively; z 2,1 , z 2,2 , z 2,3 They are x3, the estimated value of the aggregate interference w2 and its first-order derivative. Definition

[0078] By defining:

[0079]

[0080] Construct the observer’s estimation error dynamics system:

[0081]

[0082] In step S3, a desired position model for permanent magnet synchronous motor servo control is constructed.

[0083] Based on the estimated error dynamic system of the observer in step 2, for the estimation of the load torque and other unknown lumped disturbances applied to the permanent magnet synchronous motor servo, the unknown information that may exist in the system has been fully evaluated. At this point, the mathematical steady-state model of the system can be established and expressed as follows:

[0084]

[0085] At the same time, the expected mathematical output value u* is:

[0086]

[0087] In step S4, the coordinate transformation of the mathematical model is performed.

[0088] By constructing a mathematical steady-state model and introducing the coordinate transformation of the mathematical model, the permanent magnet synchronous motor servo control problem can be converted into the output trajectory traceability problem in control theory. The specific coordinate transformation is as follows:

[0089]

[0090] Where l is a sufficiently large normal parameter to be designed.

[0091] At this point, the present invention introduces system M4, which is expressed as:

[0092]

[0093] In step S5, a reference nonlinear non-recursive control law is set.

[0094] Based on steps 1 to 4, the benchmark nonlinear non-recursive control law of the present invention can be explicitly constructed in the following form:

[0095]

[0096] Among them, ι is the parameter value to be designed; τ is a negative number with an open interval from -1 / 3 to 0; k1, k2, k3 are design gain values, which are usually determined as a set of large positive values.

[0097] In step S6, output execution law design is performed.

[0098] Combined with the system output and the q-axis voltage u q The relationship between them is:

[0099]

[0100] Then the executable q-axis voltage expression is:

[0101]

[0102] In step S7, it is necessary to design a parameter tuning method based on deep reinforcement learning.

[0103] The design process of the deep reinforcement learning controller based on the deep deterministic policy gradient (DDPG) algorithm in this invention is mainly divided into four parts.

[0104] Where S t Represents the state space, that is, the various state information that the agent can obtain from the environment. t , a t , a t+1 ) is a reward function, which is used to evaluate the training benefit at time t and give feedback to the agent to make it decide the next action. t It is a discrete action space, which is the command signal required by the present invention and is only related to the previous action.

[0105] For the permanent magnet synchronous motor servo control system, its main control target is to adjust the angle of the motor, and the angle is related to the angular velocity, d-axis current, q-axis current, d-axis voltage, q-axis voltage, load torque and lumped disturbance of the whole system. In the design process of the present invention, in order not to increase the computational complexity of the neural network and to reduce the difficulty of training as much as possible, the constructed deep reinforcement learning controller does not consider the inherent parameters of the environment. Then the state space is defined as:

[0106] S t ={θ,θ-θ ref ,w1-z 1,2 ,w2-w 2,2}

[0107] The q-axis voltage u required by the permanent magnet synchronous motor servo control system at each moment q The key design parameter in is the action space a consisting of the policy set π t The strategy provides a possible action distribution for each situation. The action range and action reference value are determined by a traditional nonlinear non-recursive controller, specifically a t ∈{a min ;c;a max}, where a min ,a max are the lower and upper bounds of the parameters l, ι and τ respectively, and c is the minimum incremental step size of the action set.

[0108] In order to make the output position of the permanent magnet synchronous motor servo control system approach the reference value as quickly as possible, the present invention sets a set of reward functions related to the position tracking of the permanent magnet synchronous motor servo control system, which can be expressed in the specific form as follows:

[0109]

[0110] The number of hidden layers of the neural network of the present invention is set to two layers, the number of nodes in the first hidden layer is set to 64, and the number of nodes in the second hidden layer is also 64. The activation functions of the hidden layer and the output layer are two ReLU functions and one linear function.

[0111] So far, the main design of the position angle adjustment control for the permanent magnet synchronous motor servo system has been completed. Next, in order to facilitate the actual application of engineers in engineering scenarios, the control parameter adjustment rules involved in the applied robust adaptive enhanced permanent magnet synchronous motor servo controller design patent and the impact of the corresponding parameter changes on the control system performance can be simply summarized as follows:

[0112] 1. Parameter tuning of nonlinear high bandwidth gain disturbance estimator

[0113] (The parameter to be adjusted is α 1,1 , α 1,2 , α 1,3 , α 1,4 ; α 2,1 , α 2,2 , α 2,3 ; l; τ)

[0114] The bandwidth factor l in the nonlinear high-bandwidth gain disturbance estimator needs to satisfy the constraint l>0 in the rough configuration in the project, and it can be a sufficiently large positive constant. In order to ensure that the designed robust adaptive enhanced permanent magnet synchronous motor servo controller can ensure that the motor system has good position regulation performance under various working conditions, engineers usually need to use traditional nonlinear non-recursive controllers, guided by the motor position regulation performance, to test and determine the maximum value of l lmax and the minimum value l min , and use reinforcement learning to self-adjust its min ,l max ] is the optimal value within the range. Similarly, the maximum and minimum values ​​of the robust factor τ can be determined from a theoretical perspective as (-1 / 3, 0). In the permanent magnet synchronous motor servo control system, its specific parameters are also adjusted by reinforcement learning self-learning. Finally, α 1,1 , α 1,2 , α 1,3 , α 1,4 , α 2,1 , α 2,2 ,

[0115] α 2,3 , roughly speaking, it only needs to satisfy the condition of a sufficiently large positive number. It can also be set in detail using the pole configuration method, where the pole position directly affects the convergence speed of the estimator. Normally, the smaller its value, the faster the estimator converges, corresponding to the stronger robustness of the closed-loop system and the stronger ability to handle medium and high frequency time-varying disturbances. However, the side effect is that the steady-state response effect may have more obvious glitches or even oscillations, and in the early stage of the estimator convergence, it will bring about a large overshoot, so this is a set of parameters worth coordinating.

[0116] 2. Reference feedback nonlinear non-recursive controller parameter tuning: (parameters to be tuned k1, k2, k3, ι)

[0117] For the design parameters κ1, κ2, and κ3 of the reference feedback nonlinear non-recursive controller, only small positive values ​​need to be given in engineering sense. When fine adjustment is required, a set of control gains can also be set through the idea of ​​pole configuration. Generally speaking, the selection of poles is related to the robustness of the system. The poles are in the left half plane of the s plane, and the farther they are from the imaginary axis, the stronger the robustness of the controlled permanent magnet synchronous motor servo system, the less affected by the load torque fluctuation, and the faster the angle adjustment convergence of the permanent magnet synchronous motor system. However, it will bring many side effects. Being too far away from the imaginary axis will cause the system to be over-robust to a certain extent. When the permanent magnet synchronous motor servo system enters the steady state, a more obvious angle oscillation will occur. At the same time, when the load torque does not change much, it is easy to have excessive overshoot in the transient process. Therefore, this set of parameters is also a parameter worth weighing, and it can usually be flexibly changed according to the deviation between the actual response and the expected value.

[0118] 3. Deep reinforcement learning controller parameter tuning: (the parameters to be tuned are a min , a max , c):

[0119] a min , amax As the lower and upper bounds of the output space duty cycle, the present invention uses the parameter selection control effect of the nonlinear non-recursive controller to learn from a large number of simulations and experiments, and finds out the optimal duty cycle by trial and error as an empirical guide. min ,a max The selection requirements of are relatively strict, and its size directly determines the convergence speed and convergence effect of the reinforcement learning algorithm. Therefore, in engineering, the difference between the two is generally small, and the training pressure given to reinforcement learning will not be too great. However, if the difference is too small, the search space will be small and better control effects cannot be obtained. c is the minimum increment of each step of action change. In the present invention, it is set to 0.01 to make the action as smooth as possible, while ensuring a certain search efficiency.

[0120] The following is an actual experimental verification. In order to verify the effectiveness of the position adjustment control of the permanent magnet synchronous motor servo system of the present invention and the obvious position response performance improvement effect, the present invention sets three groups of operating conditions:

[0121] 1) Load torque is cos(t)+sin(t); 0≤t≤1, and the desired position is θ ref =cos(t)+sin(t); 2≤t≤3, the expected position is θ ref =2e -t ;

[0122] 2) Load torque is cos(t)+sin(t); 1≤t≤2, and the desired position is θ ref =0; 3≤t≤4, the expected position is θ ref =2;

[0123] 3) Based on working conditions (1) and (2), the control controller of this working condition adopts the traditional double closed-loop PI controller.

[0124] According to the parameter selection rules given above, combined with the angle adjustment control system of the permanent magnet synchronous motor servo system selected in this example, the simulation control parameter settings of the present invention are shown in Table 1.

[0125] Table 1 Simulation parameters

[0126]

[0127]

[0128] The simulation results of working condition (1) are as follows: Figure 2 and Figure 4As shown in the result graph, when the load torque remains unchanged and the desired position changes and is presented in a time-varying form, both the robust adaptive enhanced control scheme proposed in the present invention and the dual closed-loop PI controller can respond to the desired position, and this group of working condition results effectively illustrates the performance advantages of the method proposed in the present invention in compensating for real-time disturbances and adapting to system parameter perturbations, that is, the algorithm proposed in the present invention obviously performs better in adjustment time, and the overshoot is also greatly reduced. At the same time, the steady-state tracking effect is also better than the dual closed-loop PI, which is reflected by the improvement in robustness brought by the feedforward loop.

[0129] Figure 3 and Figure 5 The simulation results of working condition (2) are shown. From this set of results, the strategy proposed by the present invention is still stronger than the double closed-loop PI control strategy, which is reflected in the adjustment time and steady-state accuracy of the control. In contrast, the control effect of the double closed-loop PI controller is obviously slower than the control strategy of the present invention in terms of adjustment time. The performance of overshoot and steady-state accuracy is improved compared with the time-varying expected position. Therefore, it can be considered that the PI controller has a certain control effect when tracking the constant position.

[0130] Compared with the prior art, the beneficial effects of the technology of the present invention are as follows:

[0131] 1. The robust adaptive enhanced permanent magnet synchronous motor servo controller method proposed in the present invention utilizes the parameter self-optimization capability granted by the reinforcement learning algorithm in the traditional nonlinear non-recursive controller. The bandwidth factor of the external load torque estimation loop and the feedback controller design loop can be jointly and autonomously scheduled to the most appropriate value to meet the test of the permanent magnet synchronous motor under multiple types and multiple intensities of load torque to ensure better performance requirements.

[0132] 2. The training range of the bandwidth factor of the reinforcement learning algorithm in the present invention ensures a certain robustness of the robust adaptive enhanced permanent magnet synchronous motor servo control system, while the flexible dynamic mechanism avoids robustness redundancy as much as possible.

[0133] 3. The external load torque estimator based on reinforcement learning improves the adaptability of the permanent magnet synchronous motor servo system under different load torque conditions, and further enhances the anti-disturbance capability through the parameter self-configuration mechanism.

[0134] The physical model and mathematical model of the permanent magnet synchronous motor servo system constructed in step 1 of the present invention take into account the perturbation of the system circuit parameters, and dynamically integrate it with the irrelevant modeling inside the system as lumped interference, and do not require high accuracy in PMSM system modeling.

[0135] In step 2, a feedforward compensation loop was built to correct the unknown information contained in the system in real time, taking into account the high-order estimated information of load torque and external disturbance, and having a stronger ability to identify the two, effectively enhancing the robustness of the permanent magnet synchronous motor servo system.

[0136] The steady-state information construction and mathematical model coordinate transformation proposed in steps 3 and 4 utilize the idea of ​​nonlinear output regulation, and do not adopt the recursive idea of ​​nonlinear term feedback elimination, so the control law can be directly and clearly derived.

[0137] The nonlinear non-recursive control law proposed in steps 5 and 6 has certain robustness performance and makes the closed-loop system converge faster.

[0138] Step 7 integrates the idea and strategy of deep reinforcement learning, and has certain adaptive performance when the circuit parameters are perturbed and the load power is switched. It can realize the autonomous and flexible adjustment of the dual-loop control bandwidth factor, thereby ensuring that the control system has good dynamic response and steady-state accuracy.

[0139] In step 7, the action space design combines traditional control and permanent magnet synchronous motor system design experience to flexibly adjust the action space parameters. For the nonlinear high-bandwidth gain disturbance estimator parameters and the reference feedback nonlinear non-recursive controller parameter tuning, the pole configuration method can be used to select, which provides engineers with a better design experience.

[0140] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A permanent magnet synchronous motor servo control method based on nonlinear control strategy, characterized in that: The method comprises the following steps: S1. Construct a mathematical model of the permanent magnet synchronous motor servo system; S2. Based on the mathematical model, the system disturbance is reconstructed using a nonlinear high-bandwidth gain load torque and disturbance estimator to construct an estimation error dynamic system of the observer; S3, based on the estimated error dynamic system of the observer, a mathematical steady-state model of the system is established to obtain the expected mathematical output value; S4, introducing coordinate transformation of the mathematical model through the mathematical steady-state model to obtain a system after coordinate transformation; S5. Obtaining a benchmark nonlinear non-recursive control law according to the coordinate transformation and the expected mathematical output value; S6. Obtaining an output execution law based on a benchmark nonlinear non-recursive control law; S7. A homogeneous high-gain estimator, a reinforcement learning parameter setter and a nonlinear non-recursive controller corresponding to the estimated error dynamic system are set in the permanent magnet synchronous motor system. When the permanent magnet synchronous motor is working, the actual feedback angle position of the motor is obtained and converted into an actual angular velocity. At the same time, the actual d-axis current is obtained. The reinforcement learning parameter setter outputs the set parameters l and τ to the homogeneous high-gain estimator according to the current state, and outputs the set parameters τ and ι to the nonlinear non-recursive controller. The differences between the reference angle position, the reference angular velocity and the reference d-axis current and the actual feedback angle position, the actual angular velocity and the actual d-axis current are calculated, respectively. The calculated differences are input into the nonlinear non-recursive controller. The nonlinear non-recursive controller outputs an executable q-axis voltage. The executable q-axis voltage is input into the homogeneous high-gain estimator. The homogeneous high-gain estimator feeds back the estimated parameters to the nonlinear non-recursive controller. The executable q-axis voltage and the d-axis voltage are input into the transformation module and the SVPWM module. The output PWM signal controls the three-phase inverter circuit connected to the motor.

2. A permanent magnet synchronous motor servo control method based on nonlinear control strategy according to claim 1, characterized in that: The expected mathematical output value is: Among them, θ ref represents the reference angle position, B represents the viscous friction coefficient, J represents the moment of inertia, z 1,3 、z 1,2 and z 2,2 They represent the estimated value of the first-order derivative of the load torque, the estimated value of the load torque and the estimated value of the lumped disturbance w2, respectively. p represents the pole pair number, ψ f represents the rotor flux, L represents the stator inductance, R s Indicates the stator resistance.

3. The permanent magnet synchronous motor servo control method based on nonlinear control strategy according to claim 2 is characterized in that: The reference nonlinear non-recursive control law is: Among them, ι and τ are the parameter values ​​to be designed output by the reinforcement learning parameter setter, κ1, κ2, κ3 are the design gain values, Where ω represents the actual angular velocity, i q represents the q-axis current.

4. The permanent magnet synchronous motor servo control method based on nonlinear control strategy according to claim 1 is characterized in that: The estimation error dynamic system of the observer is: in, Symbol z 1,1 , z 1,2 , z 1,3 , z 1,4 are the estimated values ​​of speed ω, load torque and their first and second order derivatives respectively; z 2,1 , z 2,2 , z 2,3 They are x3, the estimated values ​​of the lumped disturbance w2 and its first-order derivative, and w1 is the load disturbance of the permanent magnet synchronous motor system.

5. The method for controlling a permanent magnet synchronous motor servo based on a nonlinear control strategy according to claim 4, characterized in that: The estimated parameters of the homogeneous high-gain estimator feeding back to the nonlinear non-recursive controller are: and 6. The permanent magnet synchronous motor servo control method based on nonlinear control strategy according to claim 1, characterized in that: The reinforcement learning parameter setter includes an agent, which is trained using the DDPG algorithm. The training process of the agent is as follows: The current state information for training is obtained. The action network of the agent outputs the current action based on the current state information for training. The reward function is output based on the current action. The action network is trained based on the reward function. The above steps are repeated until a trained reinforcement learning parameter setter is obtained.

7. The permanent magnet synchronous motor servo control method based on nonlinear control strategy according to claim 6, characterized in that: The current status information is: S t ={θ,θ-θ ref ,w1-z 1,2 ,w2-w 2,2 }。 8. The permanent magnet synchronous motor servo control method based on nonlinear control strategy according to claim 7, characterized in that: The current action is: a t ∈{a min ;c;a max }, where a min , a max are the lower and upper bounds of the parameters l, ι and τ respectively, and c is the minimum incremental step size of the action set.

9. The method for controlling a permanent magnet synchronous motor servo based on a nonlinear control strategy according to claim 8, characterized in that: The reward function is: R=r1+r2+r3 Among them, R is the reward function.

10. The method for controlling a permanent magnet synchronous motor servo based on a nonlinear control strategy according to claim 9, characterized in that: The number of hidden layers of the action network is set to two, the number of nodes in the first hidden layer is set to 64, and the number of nodes in the second hidden layer is also 64. The activation functions of the hidden layer and the output layer are two ReLU functions and one linear function.

Citation Information

Patent Citations

  • Method and system for identifying moment of inertia of permanent magnet synchronous motor

    CN107171609A

  • Permanent magnet synchronous motor servo control system based on fuzzy PI control

    CN111934585A

  • A servo control method and servo system for a permanent magnet synchronous motor

    CN113556068B

  • A Permanent Magnet Synchronous Motor Control Method and System Based on Width-Learning Interference Observer

    CN114744946B