Closed-loop control method and system of servo driving system

Abstract

The invention relates to the technical field of servo driving, and discloses a closed-loop control method and system for a servo driving system, and the method comprises the steps: carrying out the mathematical modeling of the servo driving system, and building an initialization system model containing a motor equation and a load kinetic equation; collecting a current signal and a position signal of the initialization system model to obtain system operation characteristic data; performing feature extraction on the system operation feature data to obtain system modeling features; constructing a three-ring nested controller according to the system modeling features, and calculating a first system control parameter; performing operation test and time-frequency analysis on the servo driving system after the first system control parameter modulation to obtain system operation state data; and performing particle swarm optimization calculation on the first system control parameter based on the system operation state data to obtain a second system control parameter. According to the method, self-adaptive adjustment of the servo driving system is achieved, and the adaptive capacity and the control precision of the servo driving system under different working conditions are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of servo drive, and in particular to a closed-loop control method and system for a servo drive system. Background Art

[0002] As the core component in the field of industrial automation, the control performance of the servo drive system directly affects the overall performance of the automation equipment. The traditional servo drive system control method mainly relies on the classic PID control theory. Although it has a simple structure and is easy to implement, it is difficult to meet the dynamic performance requirements of the system under complex working conditions and high-precision control requirements.

[0003] In recent years, with the continuous improvement of industrial production requirements for servo drive system performance, the system faces more and more interference and uncertainty factors during operation, such as load disturbance, parameter perturbation and external disturbance, which seriously affect the control accuracy and dynamic response performance of the system. Traditional fixed parameter control methods are difficult to cope with these complex working conditions, and the robustness and adaptability of the system need to be improved. Summary of the invention

[0004] The present invention provides a closed-loop control method and system for a servo drive system, which realizes adaptive regulation of the servo drive system and improves the adaptability and control accuracy of the servo drive system under different working conditions.

[0005] In a first aspect, the present invention provides a closed-loop control method for a servo drive system, the closed-loop control method for a servo drive system comprising: Mathematically model the servo drive system and establish an initial system model including motor equations and load dynamics equations; Collecting the current signal and position signal of the initialized system model, and performing Clarke transformation and Park transformation processing to obtain system operation characteristic data; Extracting features from the system operation feature data to obtain system modeling features; A three-loop nested controller is constructed according to the system modeling characteristics, and a first system control parameter is calculated by Lyapunov stability theory; Performing operation test and time-frequency analysis on the servo drive system after the control parameters of the first system are modulated to obtain system operation status data; Based on the system operation status data, the first system control parameters are subjected to particle swarm optimization calculation to obtain second system control parameters. The second system control parameters are applied to a three-loop nested controller to achieve high-precision closed-loop control of the servo drive system by coordinating and controlling the three links of current, speed and position.

[0006] In a second aspect, the present invention provides a closed-loop control system of a servo drive system, the closed-loop control system of the servo drive system comprising: Mathematical modeling module, used to perform mathematical modeling on the servo drive system and establish an initialization system model including motor equations and load dynamics equations; An acquisition module, used for acquiring the current signal and the position signal of the initialization system model, and performing Clarke transformation and Park transformation processing to obtain system operation characteristic data; A feature extraction module is used to extract features from the system operation feature data to obtain system modeling features; A calculation module, used for constructing a three-loop nested controller according to the system modeling characteristics, and calculating the first system control parameter by Lyapunov stability theory; An operation test module is used to perform operation test and time-frequency analysis on the servo drive system after the control parameters of the first system are modulated to obtain system operation status data; The optimization module is used to perform particle swarm optimization calculation on the first system control parameters based on the system operation status data to obtain second system control parameters. The second system control parameters are applied to the three-loop nested controller to achieve high-precision closed-loop control of the servo drive system by coordinating the control of the three links of current, speed and position.

[0007] In the technical solution provided by the present invention, by establishing a complete mathematical model of permanent magnet synchronous motor and introducing the fourth-order Runge-Kutta method for discretization of state equations, an accurate description of the dynamic characteristics of the system is provided, a three-loop nested control structure is adopted, and the controller gain matrix is ​​designed in combination with Lyapunov stability theory to ensure the stability of the system, and the anti-interference ability of the system is enhanced through feedforward decoupling control and sliding mode observer compensation. A double-layer variable structure particle swarm optimization algorithm is proposed, which avoids the optimization process from falling into the local optimum through the competition-cooperation mechanism and adaptive mutation strategy, and improves the global convergence of parameter optimization. A performance evaluation system based on time-frequency dual-domain analysis is designed, and the system features are extracted by wavelet packet decomposition and variational mode decomposition, and the adaptive objective function of dynamic weights is combined to achieve accurate optimization of control parameters. A five-layer feedforward neural network is used for feature extraction, and residual connection and attention mechanism are introduced to improve the extraction ability of system modeling features. Through real-time analysis of system operation status data and online optimization of control parameters, adaptive adjustment of the system is achieved, and the adaptability and control accuracy of the system under different working conditions are improved.

[0008] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0009] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A schematic diagram of an embodiment of a closed-loop control method for a servo drive system in an embodiment of the present invention; Figure 2 Schematic diagram of a closed-loop control system of a servo drive system in an embodiment of the present invention. DETAILED DESCRIPTION

[0011] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0012] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device end including a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or device ends.

[0013] To facilitate understanding of this embodiment, a closed-loop control method for a servo drive system disclosed in an embodiment of the present invention is first described in detail. Figure 1 As shown, the method comprises the following steps: 101. Perform mathematical modeling on the servo drive system and establish an initialization system model including motor equations and load dynamics equations; It is understandable that the execution subject of the present invention may be a closed-loop control system of a servo drive system, or a terminal or a server, which is not specifically limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0014] Specifically, mathematical modeling is performed on the servo drive system to establish an initialization system model including motor equations and load dynamic equations. In this process, mathematical modeling is performed on the stator voltage of the permanent magnet synchronous motor in the servo drive system. The stator voltage equation of the permanent magnet synchronous motor is described by the dq axis model of the motor. In the synchronous rotating coordinate system, the stator voltages of the d-axis and q-axis are respectively expressed as: ; ; in, represents the time variable, is the stator resistance, and are the inductances on the d-axis and q-axis respectively, and are the currents of the d-axis and q-axis respectively, represents the rotor electrical angular velocity, and Represents the permanent magnet flux. Through the above mathematical modeling, the mathematical model of the motor stator voltage is obtained. This model is used to describe the electrical characteristics of the motor, reflect the relationship between the stator current, voltage and inductance, and provide a basis for subsequent servo control. On this basis, the electromagnetic torque of the motor in the servo drive system is mathematically modeled. According to the electromagnetic characteristics of the permanent magnet synchronous motor, the calculation formula of the electromagnetic torque is expressed as .in, is the electromagnetic torque, is the number of pole pairs of the motor, reflects the asymmetric characteristics of the inductor, and The term represents the torque component generated by the coupling of the d-axis and q-axis currents. Represents the direct torque component generated by the permanent magnet flux. The establishment of this equation can accurately describe the formation mechanism of electromagnetic torque and provide the necessary mathematical basis for subsequent speed control and position control. In order to describe the dynamic characteristics of the motor-driven load, a load dynamics mathematical model of the servo drive system is established. According to Newton's second law and the dynamic equation of the rotating system, the mechanical motion equation of the motor is expressed as: ; in, represents the moment of inertia of the system, is the mechanical angular velocity, is the viscous friction coefficient, Represents the external load torque. The equation shows that the angular acceleration of the motor is determined by the electromagnetic torque, load torque and viscous damping of the system. Therefore, by analyzing the dynamic model, a reasonable control strategy is designed to ensure the stable operation of the motor system. After completing the above mathematical modeling, the state space transformation of the motor stator voltage mathematical model, electromagnetic torque mathematical model and load dynamics mathematical model is performed to obtain the state equation of the system. Assume that the state vector is , the input vector is , the system state equation is expressed as , where A and B are the system matrix and input matrix respectively. The continuous-time state equation is discretized so that the control parameter calculation and algorithm implementation can be performed in the discrete time domain. In order to improve the accuracy of discretization, the fourth-order Runge-Kutta method is used to discretize the system, which effectively reduces the discretization error and improves the calculation accuracy of the system. Substitute the discrete state equation into the control period parameter to obtain the initialized system model.

[0015] 102. Collect the current signal and position signal of the initialization system model, and perform Clarke transformation and Park transformation processing to obtain system operation characteristic data; Specifically, the three-phase current of the motor stator voltage mathematical model in the initialization system model is sampled, and multiple sampling points are collected in each current loop control cycle according to the time sequence number of the discrete state equation. In order to ensure the high precision of the collected data, 16-bit analog-to-digital conversion is performed on each sampling point, and the analog signal is converted into a digital signal to obtain high-precision three-phase current sampling data. These data can more accurately reflect the changes in the current of each phase during the operation of the motor. At the same time, in order to accurately obtain the mechanical angular velocity in the load dynamics mathematical model, a 23-bit high-precision encoder is used for data acquisition. In each position loop control cycle, the pulse signal of the encoder is used to calculate the rotor position increment, and the sampling data of the rotor position is obtained by differential operation. After the sampling is completed, the collected three-phase current data is processed by Butterworth low-pass filtering. The characteristic of Butterworth low-pass filtering is to keep the amplitude characteristics of the signal as flat as possible within the passband range, and at the same time have good attenuation characteristics within the stopband range, so it can effectively remove high-frequency noise and prevent the signal from being disturbed by unnecessary interference in the subsequent calculation process. The three-phase current data after filtering is subjected to Clarke transformation, and the current data in the three-phase stationary coordinate system is converted into the current data in the two-phase stationary coordinate system. The transformation process can effectively reduce the complexity of calculation, so that the current control of the motor is converted from a three-phase system to a two-phase system, simplify the calculation process of the control algorithm, and improve the response speed of real-time control. After completing the Clarke transformation, in order to improve the effectiveness of control, the Park transformation is combined with the rotor position data. The role of the Park transformation is to convert the current data in the two-phase stationary coordinate system into the current data in the synchronous rotating coordinate system, so that the current control can directly act on the flux and torque of the motor, thereby optimizing the dynamic response characteristics of the motor. Through this transformation, the servo drive system controls the electromagnetic characteristics of the motor more intuitively, so that the components of the current can be independently controlled on the direct axis and the quadrature axis, respectively, to achieve more efficient torque regulation and flux regulation, and improve the operation stability and control accuracy of the motor. After completing the Park transformation, the d-axis current and q-axis current data finally obtained are combined with the previously collected rotor position sampling data to form complete system operation characteristic data. These characteristic data include key physical quantities such as current, position, and speed, which characterize the state changes of the motor during operation.

[0016] 103. Extract features from system operation feature data to obtain system modeling features; Specifically, the system operation feature data is input into the input layer of the five-layer feedforward neural network, where the input layer contains 8 neurons, which correspond to the basic features and differential features of the d-axis current, q-axis current and rotor position, respectively, so as to capture the static information and dynamic change characteristics of the system operation state. Through this process, the input layer feature vector is constructed. The input layer feature vector is processed by the first hidden layer. The first hidden layer contains 128 neurons, and the ReLU activation function is used to perform nonlinear transformation on the input of each neuron to ensure that the network can learn complex nonlinear relationships. At the same time, in order to prevent the distribution drift of data during network propagation, the first hidden layer introduces a batch normalization layer to standardize the data, make the data distribution more stable and accelerate the convergence of the network. After this layer of processing, the first layer feature map is obtained. The first layer feature map is input into the second hidden layer, which contains 256 neurons, increasing the capacity of the network and enabling it to learn more complex feature relationships. Similarly, this layer uses the ReLU activation function and batch normalization layer to maintain data stability and training efficiency. In order to enhance the information transmission effect, the second hidden layer introduces a residual connection structure, and the first layer feature map is directly superimposed on the output of this layer to form a cross-layer feature fusion. The residual structure can effectively alleviate the gradient vanishing problem that occurs during deep neural network training, and improve the accuracy of feature extraction, so that the system's operating data can be more efficiently mapped to the high-dimensional feature space. The second layer feature map is input into the third hidden layer for processing. The third hidden layer contains 256 neurons, and more representative system modeling features are extracted through deeper calculations. Since the main goal of this layer is to deeply combine the previous features, there is no need to introduce an additional residual structure, but to allow the network to make full use of a larger number of neurons and more complex connection methods to further optimize the feature representation. After calculations in this layer, the third layer feature map is obtained. The third layer feature map is processed by the fourth hidden layer, which contains 128 neurons and uses the ReLU activation function and batch normalization layer to ensure the nonlinear expression ability of the data and the stability of training. In order to improve the effectiveness of feature extraction, this layer introduces an attention mechanism to weight different important features in the feature space. Through the attention mechanism, the network automatically focuses on the most critical system parameters, such as assigning higher weights to certain features under different working conditions, making the final extracted features more targeted and discriminative, thereby improving the accuracy and generalization of modeling. After deep feature extraction in four hidden layers, the fourth layer feature map is input to the output layer for processing. The output layer contains 32 neurons, and a linear activation function is used to reduce the dimension of the high-dimensional features extracted in the previous order to ensure that the final output system modeling features can retain the most important information while reducing the dimension. Through the dimensionality reduction process, the computational complexity is effectively reduced while maintaining an accurate description of the core features of the servo drive system.The obtained system modeling characteristics can accurately reflect the operating status of the servo drive system.

[0017] 104. Construct a three-loop nested controller based on the system modeling characteristics, and calculate the control parameters of the first system through Lyapunov stability theory; Specifically, an inner loop current controller is constructed according to the system modeling characteristics to ensure fast dynamic response and stable control of the current. In order to compensate for the coupling effect in the mathematical model of the motor stator voltage, a feedforward decoupling control method is adopted to make the current of the d-axis and q-axis independently adjusted to reduce the impact of motor parameter changes on system performance. At the same time, feedback compensation control is introduced to enhance the anti-interference ability of the system, and by adjusting the current loop controller parameters, the current can quickly follow the expected value and maintain stable operation, thereby improving the overall response performance of the servo drive system. Stability analysis is performed using the Lyapunov stability theory. A Lyapunov function is constructed, and a quadratic function is selected as a candidate Lyapunov function so that the stability of the system can be analyzed using its mathematical characteristics. By calculating the time derivative of the Lyapunov function and solving the inequality of the derivative, the stability parameters of the current loop are determined to ensure the robustness and stability of the system under different working conditions. A middle loop speed controller is constructed based on the current loop stability parameters to compensate for system disturbances in real time and optimize the speed tracking performance. The speed loop controller adjusts the motor speed so that the system can quickly respond to external load changes while maintaining a stable operating state. In this process, the design of the speed controller needs to comprehensively consider the dynamic characteristics of the current loop to ensure the coordinated operation of the entire control system. Therefore, when designing the speed loop parameters, a composite Lyapunov function is constructed based on the Lyapunov theory, the Lyapunov functions of the speed loop and the current loop are combined, and the stability parameters of the speed loop are determined by solving the time derivative conditions of the composite function. The outer loop position controller is constructed based on the speed loop stability parameters to ensure that the motor can accurately track the position command. The position controller compensates for the position error and optimizes the control strategy for different operating conditions. When designing the position controller, the position command is preprocessed to smooth the target trajectory and reduce the impact of the control system. At the same time, a sliding mode observer is introduced to compensate for the position tracking error online, so that the control system can adaptively adjust the control strategy and improve the response accuracy of the servo drive system under different load conditions. After determining the parameters of the position controller, the nested Lyapunov function is used for stability analysis. The Lyapunov functions of the inner loop current controller, the middle loop speed controller and the outer loop position controller are cascaded and combined. The stability parameters of the position loop are determined by solving the stability conditions of the cascade combination function to ensure the robustness of the entire servo drive system. Based on the stability parameters of the current loop, the speed loop and the position loop, the inner loop current controller, the middle loop speed controller and the outer loop position controller are cascaded. By adjusting the gain parameters and dynamic response characteristics of the controller, the various loops can work closely together to improve the stability and dynamic response performance of the overall control system. During the tuning process, it is ensured that the time domain characteristics of each loop meet the accuracy requirements of the servo drive system, and at the same time, the system can remain stable under different loads and working conditions.After completing the cascade tuning, the obtained current loop controller parameters, speed loop controller parameters, position loop controller parameters and controller tuning parameters are combined to form the first system control parameters. The system control parameters cover the complete control strategy from current to speed to position, enabling the servo drive system to maintain high-precision and high-dynamic performance operating characteristics in complex environments, and provide a stable and reliable closed-loop control solution.

[0018] 105. Performing operation test and time-frequency analysis on the servo drive system after the control parameter of the first system is modulated to obtain system operation status data; Specifically, the control parameters of the first system are modulated to make them suitable for different operating conditions and ensure the stability and dynamic response characteristics of the system under various load conditions. The control bandwidth of the first system control parameters is modulated, and by adjusting the bandwidth of the control system, the servo drive system can maintain good control effects in different frequency ranges. The modulation of the control bandwidth mainly involves gain adjustment and filtering optimization to ensure that the system does not introduce high-frequency noise or oscillation while responding to high-speed dynamics. While modulating the bandwidth, the damping characteristics of the first system control parameters are modulated to optimize the transient response of the system and reduce steady-state errors. The core of damping characteristic modulation is to adjust the control parameters so that the system achieves a balance between fast response and stability, ensuring that the motor can maintain stable operation under different load conditions. The servo drive system is tested under load conditions based on the modulated parameters to verify the effect of the control strategy in actual operation. During the test, position commands under different load conditions are input into the system in turn, including no-load conditions, rated load conditions, and twice rated load conditions. In order to comprehensively evaluate the dynamic response performance of the system, step position commands and sinusoidal position commands are used in the test process. The step command is used to test the transient response characteristics of the system and evaluate the overshoot, rise time and steady-state error of the control system, while the sinusoidal position command is used to analyze the tracking ability of the system under periodic input and evaluate the steady-state tracking error and phase delay of the system. During the entire test process, key variables such as d-axis current, q-axis current, rotor angular velocity and rotor position are recorded in real time to obtain the actual operation data of the servo drive system and form a working condition test data set. Time-frequency analysis is performed on the working condition test data to extract the key features of the system. The collected working condition test data is decomposed by wavelet packet decomposition, and the signal is decomposed into different frequency bands using the multi-scale characteristics of wavelet packet transform to obtain the spectral characteristics at different time scales. Wavelet packet decomposition can effectively capture the dynamic characteristics of the servo drive system in different frequency ranges and identify the vibration mode and noise characteristics of the system. At the same time, in order to obtain the inherent dynamic characteristics of the system, the working condition test data is subjected to variational modal decomposition, and the signal is decomposed into multiple inherent modal components. Each modal component corresponds to a different physical phenomenon, so as to more accurately extract the operating state characteristics of the system and distinguish the normal operating mode from the abnormal mode of the system. Based on multi-scale frequency band characteristics and inherent modal characteristics, dual-domain analysis is performed, that is, in-depth mining of data in both time domain and frequency domain is carried out simultaneously to fully reveal the dynamic characteristics of the system. A feature matrix is ​​constructed to fuse the signal characteristics at different time scales and the characteristics of different modal components to form a unified mathematical description. On this basis, the main eigenvector is extracted using the singular value decomposition method, and the data is effectively reduced in dimension through singular value decomposition to extract the most representative system operation status information, avoid the interference of redundant data, and improve the recognition ability of system status characteristics.The system operation status data is obtained, which contains the core characteristics of the servo drive system under different load conditions.

[0019] 106. Based on the system operation status data, the particle swarm optimization calculation is performed on the control parameters of the first system to obtain the control parameters of the second system. The control parameters of the second system are applied to the three-loop nested controller to achieve high-precision closed-loop control of the servo drive system by coordinating the control of the three links of current, speed and position.

[0020] Specifically, a dynamic time-frequency evaluation index is constructed for the system operation status data. By mapping the energy distribution ratio of multi-scale frequency band characteristics, a frequency domain evaluation sub-function is formed, and the instantaneous amplitude variance of the inherent modal characteristics is mapped to a time domain evaluation sub-function. Since the dynamic characteristics of the system under different working conditions are time-varying, the dynamic weight coefficients of the time domain and frequency domain sub-functions are determined based on the fuzzy hierarchical analysis method, and an objective function that can be adaptively adjusted is constructed, so that the control effect under different working conditions can be taken into account during the optimization process, and the servo drive system can obtain stable and efficient control performance under various operating conditions. A double-layer variable structure particle swarm is constructed based on the adaptive objective function. The optimization algorithm adopts a double-layer particle swarm architecture, in which the outer particle swarm is mainly responsible for optimizing the bandwidth modulation parameters, while the inner particle swarm is used to optimize the damping modulation parameters. In order to avoid the outer and inner particle swarms being independent of each other during the optimization process and resulting in insufficient information utilization, the two-layer particle swarms interact with each other through a co-evolution mechanism, so that the bandwidth modulation and damping modulation can affect each other and be dynamically adjusted during the optimization process, ensuring that the optimization effect of the control parameters is more accurate. Through this hierarchical particle swarm optimization framework, the characteristics of different optimization dimensions are fully utilized, the convergence speed of the overall algorithm can be improved, and the flexibility of parameter adjustment can be enhanced, so that the servo drive system can obtain the best control performance in a complex environment. In order to improve the global search ability of the particle swarm and avoid falling into the local optimum, the hierarchical particle swarm is subjected to adaptive mutation. In this process, the mutation probability is dynamically adjusted according to the convergence degree of the particles, so that the mutation probability is higher in the early iteration to enhance the exploratory nature of the search, and the mutation probability is gradually reduced in the later convergence to improve the convergence accuracy. At the same time, in order to prevent some particles from falling into the local optimum and causing the overall search ability to decrease, Brownian mutation is applied to these particles, that is, random perturbations are used to enable particles to jump out of the local optimal area, thereby increasing the diversity of the search. Gaussian mutation is performed on the global optimal particles to perform fine-grained search near the optimal solution, thereby improving the local search ability of the algorithm and accelerating the optimization process. Through the adaptive mutation strategy, the search ability of the particle swarm is effectively improved, ensuring that different parameter spaces can be fully explored during the optimization process, and avoiding suboptimal solutions caused by early convergence. After the mutation operation is completed, a competition-cooperation mechanism is established based on the optimized particle swarm to improve the optimization effect and enhance the convergence ability. The particle swarm is divided into a competitive subgroup and a cooperative subgroup. The competitive subgroup adopts the survival of the fittest strategy to update the position, that is, in each generation of iteration, only the particles with the best performance are retained, and individuals with lower fitness are eliminated, thereby accelerating the optimization process and improving the global search ability. At the same time, the cooperative subgroup adopts a knowledge sharing strategy to adjust the speed, that is, by sharing the experience information of particles in the search process, different individuals can learn from each other and learn from the optimal search direction, thereby improving the optimization efficiency.Through the mechanism of combining competition and cooperation, the convergence speed can be improved while ensuring the global search capability, making the optimization of control parameters more stable and enhancing the adaptability to complex environments. Based on the optimized particle swarm, adaptive convergence control is performed to ensure that the optimization process stops iterating at the appropriate time and obtains the optimal solution. The convergence trend of the particle swarm is monitored in real time using the convergence judgment criterion. When the global optimal solution of the particle swarm changes less than the set threshold in multiple consecutive iterations, it is judged that the optimization process is close to convergence and the search process is terminated to prevent the increase in computational overhead and parameter overfitting caused by excessive iterations. The optimal solution mapping is performed on the converged particle swarm, and the control parameters of the optimal particles are mapped to the control model of the servo drive system to obtain the control parameters of the second system.

[0021] In the embodiment of the present invention, by establishing a complete mathematical model of permanent magnet synchronous motor and introducing the fourth-order Runge-Kutta method to discretize the state equation, an accurate description of the system dynamic characteristics is provided, a three-loop nested control structure is adopted, and the controller gain matrix is ​​designed in combination with Lyapunov stability theory to ensure the stability of the system, and the anti-interference ability of the system is enhanced through feedforward decoupling control and sliding mode observer compensation. A double-layer variable structure particle swarm optimization algorithm is proposed, which avoids the optimization process from falling into the local optimum through the competition-cooperation mechanism and adaptive mutation strategy, and improves the global convergence of parameter optimization. A performance evaluation system based on time-frequency dual-domain analysis is designed, and the system features are extracted by wavelet packet decomposition and variational mode decomposition, and the adaptive objective function of dynamic weights is combined to achieve accurate optimization of control parameters. A five-layer feedforward neural network is used for feature extraction, and residual connection and attention mechanism are introduced to improve the extraction ability of system modeling features. Through real-time analysis of system operation status data and online optimization of control parameters, adaptive adjustment of the system is achieved, and the adaptability and control accuracy of the system under different working conditions are improved.

[0022] In a specific embodiment, the process of executing step 101 may specifically include the following steps: The stator voltage of the permanent magnet synchronous motor in the servo drive system is mathematically modeled and the stator voltage equation u is established. d =R s i d +L d (di d / dt)-ω r L q i q and u q =R s i q +L q (di q / dt)+ω r L d id +ω r Ψ f , where t represents the time variable, R s Indicates stator resistance, L d and L q Represents the d-axis and q-axis inductances, i d and i q Represent the d-axis and q-axis currents, ω r represents the rotor electrical angular velocity, Ψ f Represents the permanent magnet flux, and obtains the mathematical model of the motor stator voltage; Mathematically model the electromagnetic torque of the motor in the servo drive system and establish the electromagnetic torque equation T e =1.5n p [(L d -L q )i d i q +i q Ψ f ], where T e Represents electromagnetic torque, n p Represents the number of pole pairs, and obtains the mathematical model of motor torque; Mathematical modeling of the load dynamics system in the servo drive system is carried out to establish the mechanical equation J (dω m / dt)=T e -Bω m -T L , where J represents the moment of inertia, ω m represents the mechanical angular velocity, B represents the viscous friction coefficient, T L Express the load torque and obtain the mathematical model of load dynamics; Perform state space transformation on the motor stator voltage mathematical model, motor torque mathematical model and load dynamics mathematical model to obtain the system state equation dx / dt=Ax+Bu, where x is the state vector, u is the input vector, A and B are the system matrix and input matrix respectively; The system state equation is discretized and the fourth-order Runge-Kutta method is used to convert the continuous time equation into a discrete time equation x(k+1)=x(k)+(k1+2k2+2k3+k4) / 6, where k represents the discrete time sequence number, k1, k2, k3, k4 are the Runge-Kutta coefficients, and the discrete state equation is obtained; Substitute the discrete state equation into the control period parameters to obtain the initialized system model.

[0023] Specifically, the stator voltage of the permanent magnet synchronous motor in the servo drive system is mathematically modeled and the stator voltage equation is established to describe the electrical characteristics of the motor in the synchronous rotating coordinate system. - The shaft model is used for analysis, so the stator voltage equation is expressed as:

[0024]

[0025] in, and Respectively Axis and Stator voltage on the shaft, represents the resistance of the motor stator winding, and They are Axis and The shaft current, and Respectively Axis and The equivalent inductance on the axis, is the electrical angular velocity of the rotor, and Represents the flux of the permanent magnet, that is, the magnetic flux generated by the permanent magnet in the rotor. This model reflects the electrical state of the motor in the rotating coordinate system. Next. The electromagnetic torque of the motor in the servo drive system is mathematically modeled to describe the electromagnetic torque characteristics of the motor. The electromagnetic torque of the permanent magnet synchronous motor is generated by the interaction between the stator current and the flux, and its mathematical model is expressed as: ; in, represents the electromagnetic torque of the motor, is the number of pole pairs of the motor, which represents half of the number of magnetic poles on the rotor. The term indicates that due to Axis and The torque component generated by the shaft current coupling is The term represents the main torque component generated by the interaction between the permanent magnet flux and the stator current. When , the first term disappears, and the motor torque is only The torque of the motor is generated In actual control, it is usually hoped to maintain , so that the torque is only The shaft current determines the motor efficiency, which can maximize the motor efficiency. The load dynamics mathematical model of the motor is established to describe the motion behavior of the motor rotor. According to Newton's second law, the angular acceleration of the rotor is determined by the electromagnetic torque, load torque and viscous friction torque, and its mathematical expression is:

[0026] in, represents the motor's moment of inertia, is the mechanical angular velocity, i.e. the rotation speed of the motor shaft, Represents the viscous friction coefficient of the system, which is used to describe the effect of bearing and air resistance on motor rotation. represents the load torque, that is, the rotational torque applied by the external load to the motor shaft. This equation can describe the mechanical motion state of the motor and provide a basis for speed loop and position loop control. After the mathematical model of the motor is established, the state space of the system is transformed so that it can be expressed in the form of a standard state equation. Assume that the state vector is the input vector for: ; ; Then convert the system equations into standard state space expression:

[0027] in, is the system matrix, describing the internal coupling relationship of the system, and is the input matrix, describing how the control input affects the system state. In this form, the mathematical model of the system is directly used in the design and optimization of the control algorithm. Since the actual servo control system adopts digital control, the state equation is discretized for implementation in the computer controller. Discretization methods include forward Euler method, backward Euler method and fourth-order Runge Kutta method. Among them, the fourth-order Runge Kutta method is a high-precision numerical integration method, and its calculation formula is as follows:

[0028] in, ;

[0029]

[0030] ; in, is the sampling period, represents the right-hand side function of the system state equation, where represents the state vector at the kth discrete time point, is a discrete time sequence number, indicating the sampling moment. , , and The four coefficients in the Runge-Kutta method are used for numerical integration calculation. The fourth-order Runge-Kutta method uses multi-step iterative calculation to improve numerical stability while ensuring calculation accuracy, thereby ensuring the reliability of the control system in practical applications. is the sampling period. Substituting the discrete state equation into the control period parameters, we get the complete initialization system model.

[0031] In a specific embodiment, the process of executing step 102 may specifically include the following steps: The three-phase current of the motor stator voltage mathematical model in the initialization system model is sampled, and multiple sampling points are collected in each current loop control cycle according to the time sequence number k of the discrete state equation, and each sampling point is converted into a 16-bit analog-to-digital conversion to obtain the three-phase current sampling data; Initialize the mechanical angular velocity ω of the load dynamics mathematical model in the system model m Perform 23-bit encoder acquisition. In each position loop control cycle, calculate the rotor position increment based on the encoder pulse signal, and obtain the rotor position sampling data through differential operation. Perform Butterworth low-pass filtering on the three-phase current sampling data to obtain filtered three-phase current data, and perform Clarke transformation on the filtered three-phase current data to convert the current of the three-phase stationary coordinate system into the α-β axis current of the two-phase stationary coordinate system to obtain the α-β axis current data; The α-β axis current data is processed by Park transformation according to the rotor position sampling data, the current of the two-phase stationary coordinate system is converted into the dq axis current of the synchronous rotating coordinate system to obtain the dq axis current data, and the dq axis current data and the rotor position sampling data are combined into the system operation characteristic data.

[0032] Specifically, the three-phase current of the motor stator voltage mathematical model in the initialization system model is sampled, and in each current loop control cycle, according to the time sequence number of the discrete state equation Collect multiple sampling points to ensure that the system can fully capture the dynamic characteristics of the motor current. During the sampling process, the three-phase current signals are , and , as the current components of the motor stator winding, they directly affect the electromagnetic torque and operating stability of the motor. Since the actual system usually uses a digital signal processing unit for control, these analog current signals are converted into digital signals so that they can be processed by the digital controller. In order to improve the sampling accuracy, a 16-bit analog-to-digital converter (ADC) is used at each sampling point to ensure that the resolution of the sampled data is high enough to provide an accurate current feedback signal. The converted data is used for subsequent filtering, coordinate transformation, and control calculations. At the same time, in order to obtain the mechanical angular velocity of the load dynamics mathematical model in the initialization system model , using a high-precision rotary encoder for data acquisition. The encoder uses a 23-bit incremental encoding mode, and a series of pulse signals are generated in each position loop control cycle. These pulse signals are used to calculate the incremental change of the rotor position. Assume that the total number of encoder pulses is , the pulse count increment in the current cycle is , then the motor rotor angle increment is expressed as:

[0033] in, represents the rotor electrical angle, is the pulse increment measured by the encoder in one control cycle, Represents the resolution of the encoder, that is, the total number of pulses per revolution. In order to obtain the rotor angular velocity , using differential operation, that is

[0034] in, Is the sampling period of the controller. The high-precision pulse signal provided by the encoder is used to calculate the rotation speed of the motor rotor in real time and provide it to the subsequent coordinate transformation and control loop for use. After completing the current signal sampling, the collected three-phase current data is processed by Butterworth low-pass filtering to remove high-frequency noise and retain useful current signal components. The design goal of the Butterworth filter is to provide a flat amplitude-frequency response curve so that the signal maintains its original characteristics within the passband range while effectively suppressing high-frequency noise. Set the cutoff frequency of the filter Slightly higher than the operating frequency of the motor to ensure that the effective signal part of the motor current is not weakened. The filtered current signal is smoother, which can effectively reduce the impact of interference on the control system and improve the stability of the servo drive system. After filtering, the three-phase current data is transformed by Clarke to convert the current in the three-phase stationary coordinate system into the current in the two-phase stationary coordinate system. -β axis current. The mathematical expression of Clarke transformation is as follows:

[0035] in, , and is the original three-phase current, and is the current component in the two-phase stationary coordinate system after transformation. The function of this transformation is to project the original three-phase current onto an equivalent two-dimensional plane, making the control calculation simpler and enabling a more intuitive analysis of the motor's current characteristics. - After the shaft current data is obtained, Park transformation is performed in combination with the rotor position sampling data to convert the current in the stationary coordinate system into the current in the synchronous rotating coordinate system. - Shaft current. The mathematical expression of Park transformation is as follows:

[0036] in, represents the angle of the motor rotor, and The function of this transformation is to project the current into a coordinate system that rotates synchronously with the motor rotor, so that The shaft current is mainly used to control the flux linkage, while The shaft current is mainly used to control the torque. Through Park transformation, the flux linkage and torque of the motor are decoupled, making the control more intuitive and effective. - The shaft current data is combined with the rotor position sampling data to form complete system operation characteristic data, including key parameters such as current, position, and speed.

[0037] In a specific embodiment, the process of executing step 103 may specifically include the following steps: The system operation characteristic data is input into the input layer of the five-layer feedforward neural network. The input layer contains 8 neurons, which correspond to the basic characteristics and differential characteristics of the d-axis current, q-axis current and rotor position, respectively, to obtain the input layer feature vector. The input layer feature vector is processed by the first hidden layer. The first hidden layer contains 128 neurons. The ReLU activation function is used to perform nonlinear transformation on the input of each neuron, and the data offset is eliminated through the batch normalization layer to obtain the first layer feature map; The first layer feature map is processed by the second hidden layer. The second hidden layer contains 256 neurons, uses the ReLU activation function and batch normalization layer, and introduces the residual connection structure to directly superimpose the first layer feature map to obtain the second layer feature map; The second layer feature map is processed by the third hidden layer, the third hidden layer contains 256 neurons, and the third layer feature map is obtained; The fourth hidden layer is processed on the third layer feature map. The fourth hidden layer contains 128 neurons, uses the ReLU activation function and batch normalization layer, and introduces the attention mechanism to weight the features to obtain the fourth layer feature map; The fourth layer feature map is processed in the output layer, which contains 32 neurons and uses a linear activation function for dimensionality reduction mapping to obtain system modeling features.

[0038] Specifically, the system operation feature data is input into the input layer of the five-layer feedforward neural network, and through multi-layer nonlinear transformation and feature mapping, the feature vector used for system modeling is obtained. When constructing the neural network, the input layer contains 8 neurons, each of which corresponds to Shaft current , Shaft current and rotor position And their differential characteristics, namely , and These characteristic quantities describe the electromagnetic state of the servo motor and the rotor motion information, where Used to describe the flux control of the motor, Directly affects the electromagnetic torque, Reflects the position information of the rotor at different time steps, while the differential feature , and It can capture the dynamic change characteristics of the system, thereby enhancing the neural network's modeling ability for the motor's operating state. Represents the current discrete time point / current sampling moment, Indicates the previous discrete time point / previous sampling moment. The specific meaning is: represents the d-axis current value at the current time k, represents the d-axis current value at the previous moment (k-1), and Δid represents the change in the d-axis current between two adjacent sampling moments. Similarly: and Represent the q-axis current value at the current moment and the previous moment respectively, represents the change in q-axis current, and Respectively represent the rotor position angle value at the current moment and the previous moment, Represents the change in the rotor position angle. These six features and two zero vector elements constitute the feature vector of the input layer, denoted as: ; The data in the input layer is normalized to ensure that the scales of all eigenvalues ​​are similar, so as to avoid the phenomenon that some features dominate the optimization process during network training, resulting in uneven gradient updates. After the input layer feature vector is formed, the data is passed to the first hidden layer for processing. The hidden layer contains 128 neurons and uses ReLU (rectified linear unit) as the activation function. The expression of the ReLU function is: ; It can effectively enhance the nonlinear expression ability of neural networks while avoiding the gradient vanishing problem. The calculation process of the first hidden layer is expressed as: ; in, is the weight matrix from the input layer to the first hidden layer, is the bias vector, and is the result of the first layer feature mapping. In order to improve training stability and reduce the problem of internal covariate shift, batch normalization is introduced in this layer. The processing method is as follows:

[0039] in, and The normalization operation can speed up training and improve the generalization ability of the network. The first layer feature map results are input into the second hidden layer, which contains 256 neurons and also uses the ReLU activation function and batch normalization layer to maintain the stability of calculation. This layer introduces residual connection, that is, the first layer feature map is Directly superimposed on the output of the second layer, the feature calculation formula of the second layer becomes: ; in, is the weight matrix of the second hidden layer, Represents the bias vector of the second hidden layer. The introduction of the residual structure can effectively alleviate the gradient vanishing problem of the deep network, making information efficiently transmitted between different layers while retaining shallow features, thereby improving the expression ability of the entire network. The second layer feature mapping Entering the third hidden layer, this layer contains 256 neurons and uses a standard fully connected calculation method. The calculation formula is as follows: ; in, represents the weight matrix of the third hidden layer, Represents the bias vector of the third hidden layer. After the data is deeply transformed, the representation ability of the feature vector is further enhanced, and it can more comprehensively capture the operating status information of the servo drive system. Passed to the fourth hidden layer, which contains 128 neurons and uses ReLU activation function and batch normalization layer to ensure the stability of calculation. This layer introduces attention mechanism to adaptively weight features. In the attention mechanism, the attention weight of the feature vector is calculated :

[0040] in, It is a trainable attention weight matrix. This mechanism can automatically learn the importance of each feature, thereby enhancing the influence of key features in the feature mapping process, reducing the weight of redundant features, and finally obtaining an attention-weighted feature map. :

[0041] This operation can improve the sensitivity of the neural network to the key dynamic characteristics of the servo drive system and improve the overall prediction performance of the model. Input to the output layer, which contains 32 neurons and uses a linear activation function for dimensionality reduction mapping. The expression of the linear activation function is: ; in, is the output layer weight matrix, is the bias vector of the output layer. Since the output layer adopts linear mapping, it ensures that the final system modeling characteristics will not be affected by the nonlinear activation function, so that good computational stability can be maintained in the continuous numerical domain.

[0042] Among them, before obtaining the system modeling features to construct the three-loop nested controller, it also includes: performing timing feature analysis on the system modeling features, extracting multiple continuous sampling points of dq axis current, rotor angular velocity and rotor position based on the sliding time window, constructing a dynamic timing feature matrix, and obtaining the system timing features; constructing a Bayesian prior distribution for the system timing features, using Gaussian process to perform probability modeling on the system parameters, establishing a priori probability density function, and obtaining a Bayesian prior model; inputting the Bayesian prior model into a double variation deep kernel network, using a deep kernel function to construct a feature mapping layer, and updating the model parameters through variational inference to obtain a fault prediction model; constructing a stability discriminator based on the fault prediction model, and projecting the system operation trajectory onto the Lyapunov The stability domain is obtained, the stability margin index is calculated, and the system stability assessment result is obtained; the system stability assessment result is classified, and a dual-path prediction mechanism is constructed. Prediction branches are established for stable working conditions and potential fault conditions respectively to obtain the classified prediction result; the trajectory of the system state is predicted based on the classified prediction result, and the kernel regression algorithm is used to predict the system response of multiple time steps in the future to obtain the system prediction trajectory; the system prediction trajectory is corrected and compensated, the prediction weight is dynamically adjusted according to the stability assessment result, and the prediction result under the potential fault condition is corrected online to obtain the corrected system trajectory; the corrected system trajectory is integrated with the system modeling features, and the two types of feature information are combined using an adaptive weighting method to obtain enhanced system modeling features.

[0043] In a specific embodiment, the process of executing step 104 may specifically include the following steps: An inner loop current controller is constructed according to the system modeling characteristics, and feedforward decoupling control and feedback compensation control are performed on the mathematical model of the motor stator voltage to obtain the current loop controller parameters; The Lyapunov function is constructed for the current loop controller parameters, a quadratic function is selected as a candidate Lyapunov function, and the derivative inequality of the candidate Lyapunov function is solved to determine the current loop stability parameters; Based on the current loop stability parameters, the mid-loop speed controller is constructed to compensate for system disturbances and obtain the speed loop controller parameters; The speed loop controller parameters are constructed using a composite Lyapunov function. The Lyapunov functions of the speed loop and the current loop are combined to obtain a composite function. The derivative condition of the composite function is solved to determine the speed loop stability parameters. An outer loop position controller is constructed based on the speed loop stability parameters, the position command is preprocessed, and a sliding mode observer is introduced to compensate the position tracking error online to obtain the position loop controller parameters. The position loop controller parameters are constructed by nesting Lyapunov functions, and the Lyapunov functions of the inner loop current controller, the middle loop speed controller and the outer loop position controller are cascaded to obtain a cascade combination function. The position loop stability parameters are determined by solving the conditions of the cascade combination function. Based on the current loop stability parameter, speed loop stability parameter and position loop stability parameter, the inner loop current controller, the middle loop speed controller and the outer loop position controller are cascade tuned to obtain the controller tuning parameters; The current loop controller parameters, the speed loop controller parameters, the position loop controller parameters and the controller tuning parameters are combined to form the first system control parameters.

[0044] Specifically, an inner loop current controller is constructed according to the system modeling characteristics so that it can accurately adjust the current dynamic response of the motor. In the mathematical model of the permanent magnet synchronous motor, the stator voltage equation of the motor is:

[0045]

[0046] in, and Respectively Axis and Stator voltage on the shaft, is the stator resistance, and They are Axis and The shaft current, and for Axis and The inductance on the axis, is the electrical angular velocity of the rotor, represents the flux of the permanent magnet. On this basis, in order to eliminate the coupling effect between currents and improve the control accuracy of the current loop, feedforward decoupling control is adopted to make Axis and The current regulation of the axis can be performed independently. The feedforward decoupling control strategy is designed so that the compensation control voltage is as follows:

[0047]

[0048] in, is the reference value of the d-axis compensation control voltage, is the reference value of the q-axis compensation control voltage. Through this decoupling control strategy, - The cross-coupling effect of the d-axis and q-axis makes the current regulation more accurate. The compensation control voltage in the feedforward decoupling control strategy is mainly used to eliminate the coupling effect between the d-axis and q-axis of the permanent magnet synchronous motor. In the control of the permanent magnet synchronous motor, there is cross-coupling between the d-axis and q-axis voltages / currents, which will affect the control performance. By adding the compensation voltage and This coupling effect can be offset so that the d-axis and q-axis can be controlled independently. After obtaining the current loop controller parameters, the stability of the system is verified using the Lyapunov stability theory. Select the quadratic Lyapunov candidate function:

[0049] This function represents the energy function of the motor stator current, and it is always positive. When the system is stable, the derivative of the Lyapunov function needs to satisfy:

[0050] By solving this inequality, the stability parameters of the current loop are determined, and then the gain of the current loop controller is optimized. After determining the stability parameters of the current loop, the mid-loop speed controller is constructed to compensate for the disturbance in the system and optimize the speed tracking performance of the motor. The torque equation of the motor is:

[0051] in, is the electromagnetic torque, is the pole pair number, Reflects Axis and The difference in shaft inductance, is a cross term, and is the main torque generating term. Using the torque balance equation, the mechanical motion equation is obtained:

[0052] in, is the moment of inertia, is the mechanical angular velocity, is the viscous friction coefficient, is the load torque. In order to improve the anti-disturbance ability of the system, a PI controller is used to adjust the speed, obtain the speed loop control parameters, and construct a composite Lyapunov function to combine the speed loop with the stability function of the current loop:

[0053] And solve its derivative:

[0054] Thus, the stability parameters of the speed loop are determined. After obtaining the stability parameters of the speed loop, an outer loop position controller is constructed to ensure that the motor can accurately track the position command. The position command is preprocessed to make its trajectory smooth to avoid sudden acceleration commands that affect the dynamic response of the system. At the same time, in order to compensate for the position tracking error caused by uncertain factors, a sliding mode observer is used to estimate and adjust the position error in real time, making the position loop control more accurate. The control equation of the position loop is:

[0055] in, is the position of the motor rotor. In order to minimize the position tracking error, the position error is defined as:

[0056] in, The reference value of the position command is represented, and a sliding mode observer is designed to compensate the error online. After obtaining the position loop controller parameters, a nested Lyapunov function is constructed to cascade the stability functions of the current loop, speed loop, and position loop to form an overall stability judgment function:

[0057] in, Indicates the proportional gain coefficient of the position loop controller. In the position loop PID control, k p Used to adjust the response speed of the position error and solve its derivative inequality:

[0058] To ensure the global stability of the entire control system. Based on the obtained stability parameters of the current loop, speed loop and position loop, the inner loop current controller, the middle loop speed controller and the outer loop position controller are cascaded and the controller gain is adjusted to make the system have better dynamic response and steady-state accuracy. The optimized current loop controller parameters, speed loop controller parameters, position loop controller parameters and tuning parameters are combined to form the first system control parameters, thereby completing the high-precision closed-loop control of the servo drive system.

[0059] In this embodiment, before constructing a three-loop nested controller according to system modeling characteristics to obtain the first system control parameter and performing operation test and time-frequency analysis on the servo drive system after the first system control parameter is modulated, it also includes: decomposing the first system control parameter, mapping the bandwidth parameter and the damping parameter to the dynamic performance equivalent circuit and the steady-state performance equivalent circuit respectively, to obtain a dual equivalent circuit model; performing characteristic analysis on the dynamic performance equivalent circuit, constructing a current dynamic response function based on the mathematical model of the motor stator voltage, and introducing the current loop controller parameters to calculate the overshoot and the adjustment time to obtain the dynamic characteristic index; performing characteristic analysis on the steady-state performance equivalent circuit, constructing a position tracking error function based on the load dynamics mathematical model, and introducing the position loop controller parameters to calculate the steady-state error and the disturbance suppression capability to obtain the steady-state characteristic index; establishing a dynamic characteristic index based on the dynamic characteristic index and the steady-state characteristic index Coupled transfer function, using Laplace transform to convert the time domain response characteristics to s domain, and construct the system quality factor to obtain the performance evaluation function; according to the performance evaluation function, the sensitivity of the control parameters of the first system is calculated, and the parameter-performance mapping matrix is ​​established, and the main influencing factors are extracted through singular value decomposition to obtain the parameter sensitivity matrix; the parameter sensitivity matrix is ​​subjected to eigenvalue decomposition, the dominant eigenvector is extracted, and the weight coefficient of each parameter is determined based on the energy distribution to obtain the parameter influence index; according to the parameter influence index, the parameter adjustment criterion is constructed, the balance factor of dynamic performance and steady-state performance is set, and the parameter adjustment interval is established based on the multi-objective optimization theory to obtain the parameter optimization constraint condition; the parameter optimization constraint condition and the performance evaluation function are combined into a parameter optimization theoretical model to guide the subsequent servo drive system parameter optimization process and obtain the parameter optimization guidance plan.

[0060] In a specific embodiment, the process of executing step 105 may specifically include the following steps: Performing control bandwidth modulation on the first system control parameter to obtain a bandwidth modulation parameter, and performing damping characteristic modulation on the first system control parameter to obtain a damping modulation parameter; Based on the bandwidth modulation parameters and damping modulation parameters, the servo drive system is tested under load conditions. Position step commands and sinusoidal position commands under no-load, rated load and 2 times rated load conditions are input in sequence, and the real-time responses of dq axis current, rotor angular velocity and rotor position are recorded to obtain the working condition test data. Perform wavelet packet decomposition on the working condition test data to obtain multi-scale frequency band characteristics, and perform variational mode decomposition on the working condition test data to obtain inherent mode characteristics; Based on the multi-scale frequency band characteristics and the inherent modal characteristics, a dual-domain analysis is performed, the characteristic matrix is ​​constructed, and the main eigenvectors are extracted through singular value decomposition to obtain the system operation status data.

[0061] Specifically, the control bandwidth of the first system control parameter is modulated to determine the appropriate bandwidth parameter so that the system can meet the dynamic response requirements while avoiding the interference of high-frequency noise. In the servo drive system, the control bandwidth is determined by the closed-loop gain of the current loop, speed loop and position loop. Therefore, the gain of each loop is adjusted to make the frequency response of the system meet the design requirements. For the current loop, its control bandwidth Set to the motor electrical time constant About 10 times the reciprocal of , that is:

[0062] in, is the motor stator resistance, is the inductance. For the speed loop, the control bandwidth It needs to meet the requirement of being at least one order of magnitude lower than the current loop and be selected as 1 / 10 of the current loop bandwidth, that is:

[0063] For the position loop, the bandwidth It needs to be one order of magnitude lower than the speed loop to ensure the stability of the system, so it is set as follows:

[0064] Through these formulas, reasonable bandwidth parameters are calculated and modulated in the control system to make the system have faster dynamic response capabilities while avoiding system oscillation and noise sensitivity problems caused by excessive bandwidth. The damping characteristics of the first system control parameters are modulated to optimize the transient response characteristics of the system. Damping ratio It is an important parameter to describe the characteristics of the system transition process. Its value determines the overshoot and response speed of the system. For a standard second-order system, its characteristic equation is:

[0065] in, is the natural frequency of the system, and represents the damping ratio, represents the complex frequency variable in the Laplace transform. In order to optimize the transient response of the servo drive system, it is usually desirable to select Between 0.6 and 0.8 to ensure faster response speed and lower overshoot. Therefore, the gain parameter of the controller is adjusted so that the closed-loop poles of the system (the closed-loop poles are the roots of the above characteristic equation) satisfy the following relationship: ; in, It represents the closed-loop poles of the system, which is the characteristic equation The root of the controller is obtained by adjusting the damping characteristics of the controller in this way, so that the system can maintain better dynamic characteristics under different load conditions. After completing the bandwidth modulation and damping characteristic modulation, a load condition test is performed to verify the optimization effect of the controller. During the test, position step commands and sinusoidal position commands are input under no-load, rated load, and twice rated load conditions to evaluate the transient and steady-state response characteristics of the system. The mathematical expression of the step position command is:

[0066] in, is the step position setting value, is the step amplitude, is a unit step function. The sinusoidal position command is expressed as: ; in, is the amplitude of the sinusoidal command, is the angular frequency of the command signal, Set the value for the sine position. During the test, record - Shaft current and , rotor angular velocity and the rotor position The real-time response is obtained and these data are used as working condition test data. Wavelet packet decomposition is performed on the working condition test data to extract multi-scale frequency band features. Wavelet packet decomposition is a time-frequency analysis method that decomposes the signal into different frequency band levels to extract feature information in different frequency ranges. For a given signal , wavelet packet decomposition is expressed as:

[0067] in, is the wavelet packet coefficient, is the wavelet basis function, represents the scale index of the decomposition, Represents the translation parameter at this scale, which is used to locate the time position. By calculating the energy distribution of different frequency bands, the multi-scale frequency band characteristics are obtained to analyze the dynamic characteristics of the system. The variational modal decomposition is performed on the working condition test data to decompose the signal into multiple intrinsic modal components (IMFs), and its mathematical model is expressed as:

[0068] in, Representative The intrinsic mode components, represents its center frequency, represents the gth intrinsic mode function IMF, and Relationship: yes The time domain expression of the g-th IMF represents the change of the g-th IMF over time. VMD decomposes the signal by iterative optimization so that each modal component has the minimum bandwidth in a specific frequency band, and obtains the inherent modal characteristics of the system. Dual-domain analysis is performed based on multi-frequency band characteristics and inherent modal characteristics to integrate information in the time domain and frequency domain. A feature matrix is ​​constructed to combine signal characteristics of different time scales and characteristics of different modal components to form a unified mathematical description. Assumptions is the time domain feature matrix, is the frequency domain feature matrix, then the feature matrix is ​​expressed as:

[0069] Singular value decomposition is used to reduce the dimension of the feature matrix to extract the most representative main eigenvector. The calculation formula of singular value decomposition is as follows:

[0070] in, and is an orthogonal matrix, is a diagonal matrix whose diagonal elements are singular values. By selecting the eigenvector corresponding to the largest singular value, the main features of the system operation status are extracted, and finally the system operation status data is obtained.

[0071] In a specific embodiment, the process of executing step 106 may specifically include the following steps: The dynamic time-frequency evaluation index is constructed for the system operation status data, the energy distribution ratio of the multi-scale frequency band characteristics is mapped into the frequency domain evaluation sub-function, the instantaneous amplitude variance of the inherent modal characteristics is mapped into the time domain evaluation sub-function, and the dynamic weight coefficients of the time domain and frequency domain sub-functions are determined based on the fuzzy hierarchical analysis to obtain the adaptive objective function; Based on the adaptive objective function, a double-layer variable structure particle swarm is constructed. The outer particle swarm is responsible for optimizing the bandwidth modulation parameters, and the inner particle swarm is responsible for optimizing the damping modulation parameters. The two-layer particle swarms interact with each other through the co-evolution mechanism to obtain a hierarchical particle swarm. Adaptive mutation operation is performed on the hierarchical particle swarm, the mutation probability is dynamically adjusted according to the convergence degree of the particles, Brownian mutation is applied to the particles trapped in the local optimum, and Gaussian mutation is performed on the global optimal particles to obtain the mutated particle swarm; Based on the mutated particle swarm, a competition-cooperation mechanism is established to divide the particles into a competition sub-group and a cooperation sub-group. The competition sub-group adopts the survival of the fittest strategy to update the position, and the cooperation sub-group adjusts the speed through the knowledge sharing strategy to obtain the optimized particle swarm. The optimized particle swarm is adaptively converged to obtain a converged particle swarm, and the converged particle swarm is optimally mapped to obtain the control parameters of the second system.

[0072] Specifically, a dynamic time-frequency evaluation index is constructed for the system operation status data. Since the operation status of the servo drive system not only involves the instantaneous change characteristics in the time domain, but is also affected by different frequency components, a comprehensive evaluation index is established by combining the information in the time domain and the frequency domain. The system operation status data is decomposed by wavelet packets to extract multi-scale frequency band features and calculate the energy distribution ratio of each frequency band. The frequency domain evaluation subfunction is defined as:

[0073] in, is the number of frequency bands of wavelet packet decomposition, Indicates The energy of the frequency band, is the total energy in all frequency bands, is the corresponding weight coefficient. The frequency domain evaluation subfunction reflects the energy distribution characteristics of the system in different frequency ranges, thereby identifying the resonant frequency and harmonic characteristics of the system. At the same time, in order to characterize the dynamic characteristics of the system in the time domain, the inherent modal characteristics are analyzed and the variance of the instantaneous amplitude is calculated to characterize the volatility of the system operation state. The time domain evaluation subfunction is defined as:

[0074] in, is the number of modal components obtained by variational mode decomposition, Representative The instantaneous amplitude variance of the modal components is, is the corresponding weight coefficient. The time domain evaluation sub-function reflects the instantaneous fluctuation degree of the system, thereby identifying the stability characteristics of the system. Since the weights of the time domain and frequency domain evaluation sub-functions are dynamically adjusted according to the specific working conditions, the fuzzy hierarchical analysis method is used to determine the dynamic weight coefficient. Assume and are the weights of the time domain and frequency domain sub-functions respectively, then the adaptive objective function is expressed as:

[0075] Among them, the weight coefficient satisfies , and adjust according to the dynamic characteristics of different working conditions, so that the optimization process can more accurately capture the key characteristics of the system state. A double-layer variable structure particle swarm is constructed based on the adaptive objective function to optimize the bandwidth modulation parameters and damping modulation parameters of the control system. The outer particle swarm is used to optimize the bandwidth modulation parameters , while the inner particle swarm is used to optimize the damping modulation parameters ,in, represents the current loop, Indicates the speed loop, Represents a position ring. Defines individual particles The status is:

[0076] And set the fitness function as:

[0077] The iterative update rules of the particle swarm are as follows:

[0078]

[0079] in, represents the fitness value of the i-th particle, represents the position of the ith particle at time t, It is The speed of a particle, Represents a discrete time point in the particle swarm algorithm, that is, the time point when the algorithm executes to the tth iteration. At each time point t, the algorithm updates the speed and position of all particles, and then enters the next time point t+1 to continue iterative calculation until the convergence condition is met or the maximum number of iterations is reached. is the historical optimal position of the particle, is the global optimal position, and is the learning factor, and is a random number, is the inertia weight. In order to enhance the global search capability of the particle swarm, an adaptive mutation operation is performed. For particles that converge too quickly, Brownian mutation is used, that is:

[0080] in, It is a variable asynchronous length. is the standard normal distribution noise. For the global optimal particle, Gaussian mutation is used to perform local search near the optimal solution:

[0081] in, is the mutation rate, and are the mean and variance of the Gaussian distribution, Represents a mean value of , the variance is Normal distribution. After mutation optimization, a competition-cooperation mechanism is established to improve the search efficiency of the particle swarm. The particle swarm is divided into a competitive subgroup and a cooperative subgroup. The competitive subgroup adopts the strategy of survival of the fittest, that is, it is sorted by fitness, eliminates the individuals with the lowest fitness, and replaces them with the best individuals. The cooperative subgroup adopts a knowledge sharing strategy, so that each particle adjusts its speed through information exchange to avoid falling into the local optimum. Based on the optimized particle swarm, adaptive convergence control is performed to ensure that the optimization process can converge within a reasonable time. The convergence judgment criterion is based on the rate of change of the global optimal value of the particle swarm. Setting the threshold: ; in, is the convergence threshold. When the optimization process meets the convergence condition, the converged particle swarm is optimally mapped, that is, the control parameters of the optimal particles are mapped to the servo drive system controller to obtain the final control parameters of the second system.

[0082] The closed-loop control method of the servo drive system in the embodiment of the present invention is described above. The closed-loop control system of the servo drive system in the embodiment of the present invention is described below. Figure 2 , an embodiment of a closed-loop control system of a servo drive system in an embodiment of the present invention includes: Mathematical modeling module 201, used for mathematical modeling of the servo drive system, and establishing an initialization system model including motor equations and load dynamic equations; The acquisition module 202 is used to acquire the current signal and position signal of the initialization system model, and perform Clarke transformation and Park transformation processing to obtain system operation characteristic data; The feature extraction module 203 is used to extract features from system operation feature data to obtain system modeling features; A calculation module 204 is used to construct a three-loop nested controller according to the system modeling characteristics, and calculate the first system control parameter by Lyapunov stability theory; The operation test module 205 is used to perform operation test and time-frequency analysis on the servo drive system after the first system control parameter is modulated to obtain system operation status data; The optimization module 206 is used to perform particle swarm optimization calculation on the first system control parameter based on the system operation status data to obtain the second system control parameter. The second system control parameter is applied to the three-loop nested controller to achieve high-precision closed-loop control of the servo drive system by coordinating the control of the three links of current, speed and position.

[0083] Through the cooperation of the above components, a complete mathematical model of permanent magnet synchronous motor is established, and the fourth-order Runge-Kutta method is introduced to discretize the state equation, which provides an accurate description of the system dynamic characteristics. The three-loop nested control structure is adopted, and the controller gain matrix is ​​designed in combination with Lyapunov stability theory to ensure the stability of the system. The anti-interference ability of the system is enhanced through feedforward decoupling control and sliding mode observer compensation. A double-layer variable structure particle swarm optimization algorithm is proposed. Through the competition-cooperation mechanism and adaptive mutation strategy, the optimization process is prevented from falling into the local optimum and the global convergence of parameter optimization is improved. A performance evaluation system based on time-frequency dual-domain analysis is designed. The system features are extracted by wavelet packet decomposition and variational mode decomposition. Combined with the adaptive objective function of dynamic weights, the precise optimization of control parameters is achieved. A five-layer feedforward neural network is used for feature extraction, and the residual connection and attention mechanism are introduced to improve the extraction ability of system modeling features. Through real-time analysis of system operation status data and online optimization of control parameters, the adaptive adjustment of the system is realized, and the adaptability and control accuracy of the system under different working conditions are improved.

[0084] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0085] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0086] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A closed-loop control method for a servo drive system, characterized in that: The method comprises: Mathematically model the servo drive system and establish an initial system model including motor equations and load dynamics equations; Collecting the current signal and position signal of the initialized system model, and performing Clarke transformation and Park transformation processing to obtain system operation characteristic data; Extracting features from the system operation feature data to obtain system modeling features; A three-loop nested controller is constructed according to the system modeling characteristics, and a first system control parameter is calculated by Lyapunov stability theory; Performing operation test and time-frequency analysis on the servo drive system after the control parameters of the first system are modulated to obtain system operation status data; Based on the system operation status data, the first system control parameters are subjected to particle swarm optimization calculation to obtain second system control parameters. The second system control parameters are applied to a three-loop nested controller to achieve high-precision closed-loop control of the servo drive system by coordinating and controlling the three links of current, speed and position.

2. The closed-loop control method of the servo drive system according to claim 1, characterized in that: The servo drive system is mathematically modeled to establish an initialization system model including motor equations and load dynamic equations, including: The stator voltage of the permanent magnet synchronous motor in the servo drive system is mathematically modeled and the stator voltage equation u is established. d =R s i d +L d (di d / dt)-ω r L q i q and u q =R s i q +L q (di q / dt)+ω r L d i d +ω r Ψ f , where t represents the time variable, R s Indicates stator resistance, L d and L q Represents the d-axis and q-axis inductances, i d and i q Represent the d-axis and q-axis currents, ω r represents the rotor electrical angular velocity, Ψ f Represents the permanent magnet flux, and obtains the mathematical model of the motor stator voltage; Mathematically model the electromagnetic torque of the motor in the servo drive system and establish the electromagnetic torque equation T e =1.5n p [(L d -L q )i d i q +i q Ψ f ], where T e Represents electromagnetic torque, n p Represents the number of pole pairs, and obtains the mathematical model of motor torque; Mathematical modeling of the load dynamics system in the servo drive system is carried out to establish the mechanical equation J (dω m / dt)=T e -Bω m -T L , where J represents the moment of inertia, ω m represents the mechanical angular velocity, B represents the viscous friction coefficient, T L Express the load torque and obtain the mathematical model of load dynamics; Performing state space transformation on the motor stator voltage mathematical model, the motor torque mathematical model and the load dynamics mathematical model to obtain a system state equation dx / dt=Ax+Bu, where x is a state vector, u is an input vector, and A and B are a system matrix and an input matrix, respectively; The system state equation is discretized, and the continuous time equation is converted into a discrete time equation x(k+1)=x(k)+(k1+2k2+2k3+k4) / 6 by using the fourth-order Runge-Kutta method, where k represents the discrete time sequence number, k1, k2, k3, k4 are Runge-Kutta coefficients, and a discrete state equation is obtained; Substituting the discrete state equation into the control period parameter, an initialization system model is obtained.

3. The closed-loop control method of the servo drive system according to claim 2, characterized in that: The collecting of the current signal and the position signal of the initialization system model, and performing Clarke transformation and Park transformation processing to obtain system operation characteristic data includes: Sampling the three-phase current of the motor stator voltage mathematical model in the initialization system model, collecting multiple sampling points in each current loop control cycle according to the time sequence number k of the discrete state equation, and performing 16-bit analog-to-digital conversion processing on each sampling point to obtain three-phase current sampling data; The mechanical angular velocity ω of the load dynamics mathematical model in the initialization system model m Perform 23-bit encoder acquisition. In each position loop control cycle, calculate the rotor position increment based on the encoder pulse signal, and obtain the rotor position sampling data through differential operation. Performing Butterworth low-pass filtering on the three-phase current sampling data to obtain filtered three-phase current data, and performing Clarke transformation processing on the filtered three-phase current data to convert the current of the three-phase stationary coordinate system into the α-β axis current of the two-phase stationary coordinate system to obtain the α-β axis current data; The α-β axis current data is subjected to Park transformation processing according to the rotor position sampling data, the current of the two-phase stationary coordinate system is converted into the dq axis current of the synchronous rotating coordinate system to obtain the dq axis current data, and the dq axis current data and the rotor position sampling data are combined into system operation characteristic data.

4. The closed-loop control method of the servo drive system according to claim 1, characterized in that: The extracting the system operation characteristic data to obtain system modeling characteristics includes: Inputting the system operation characteristic data into the input layer of a five-layer feedforward neural network, wherein the input layer comprises 8 neurons corresponding to the basic characteristics and differential characteristics of the d-axis current, the q-axis current and the rotor position, respectively, to obtain an input layer characteristic vector; Performing first hidden layer processing on the input layer feature vector, the first hidden layer includes 128 neurons, using ReLU activation function to perform nonlinear transformation on the input of each neuron, and eliminating data offset through batch normalization layer to obtain first layer feature map; Performing second hidden layer processing on the first layer feature map, the second hidden layer includes 256 neurons, adopts ReLU activation function and batch normalization layer, and introduces residual connection structure to directly superimpose the first layer feature map to obtain the second layer feature map; Performing third hidden layer processing on the second layer feature map, wherein the third hidden layer includes 256 neurons, to obtain a third layer feature map; Performing fourth hidden layer processing on the third layer feature map, the fourth hidden layer includes 128 neurons, adopts ReLU activation function and batch normalization layer, and introduces attention mechanism to weight the features to obtain the fourth layer feature map; The fourth layer feature map is subjected to output layer processing, wherein the output layer includes 32 neurons, and a linear activation function is used to perform dimensionality reduction mapping to obtain system modeling features.

5. The closed-loop control method of the servo drive system according to claim 1, characterized in that: The three-loop nested controller is constructed according to the system modeling characteristics, and the first system control parameter is calculated by Lyapunov stability theory, including: An inner loop current controller is constructed according to the system modeling characteristics, and feedforward decoupling control and feedback compensation control are performed on the motor stator voltage mathematical model to obtain current loop controller parameters; Performing Lyapunov function construction on the current loop controller parameters, selecting a quadratic function as a candidate Lyapunov function, and solving the derivative inequality of the candidate Lyapunov function to determine the current loop stability parameter; A mid-loop speed controller is constructed based on the current loop stability parameters to compensate for system disturbances and obtain speed loop controller parameters; The speed loop controller parameters are constructed by a composite Lyapunov function, the Lyapunov functions of the speed loop and the current loop are combined to obtain a composite function, and the derivative condition of the composite function is solved to determine the speed loop stability parameter; An outer loop position controller is constructed based on the speed loop stability parameters, the position command is preprocessed, and a sliding mode observer is introduced to perform online compensation for the position tracking error to obtain the position loop controller parameters; The position loop controller parameters are constructed by nesting Lyapunov functions, the Lyapunov functions of the inner loop current controller, the middle loop speed controller and the outer loop position controller are cascaded to obtain a cascade combination function, and the position loop stability parameters are determined by solving the conditions of the cascade combination function; Based on the current loop stability parameter, the speed loop stability parameter and the position loop stability parameter, cascade tuning is performed on an inner loop current controller, a middle loop speed controller and an outer loop position controller to obtain controller tuning parameters; The current loop controller parameter, the speed loop controller parameter, the position loop controller parameter and the controller tuning parameter are combined to form a first system control parameter.

6. The closed-loop control method of the servo drive system according to claim 1, characterized in that: The servo drive system after the control parameter of the first system is modulated is subjected to operation test and time-frequency analysis to obtain system operation status data, including: Performing control bandwidth modulation on the first system control parameter to obtain a bandwidth modulation parameter, and performing damping characteristic modulation on the first system control parameter to obtain a damping modulation parameter; Based on the bandwidth modulation parameter and the damping modulation parameter, a load condition test is performed on the servo drive system, position step commands and sinusoidal position commands under no-load, rated load and 2 times rated load conditions are input in sequence, and real-time responses of dq axis current, rotor angular velocity and rotor position are recorded to obtain condition test data; Performing wavelet packet decomposition on the working condition test data to obtain multi-scale frequency band features, and performing variational modal decomposition on the working condition test data to obtain inherent modal features; Based on the multi-scale frequency band characteristics and the inherent modal characteristics, a dual-domain analysis is performed, a characteristic matrix is ​​constructed, and the main characteristic vector is extracted through singular value decomposition to obtain system operation status data.

7. The closed-loop control method of the servo drive system according to claim 1, characterized in that: The particle swarm optimization calculation is performed on the first system control parameter based on the system operation status data to obtain the second system control parameter, and the second system control parameter is applied to the three-loop nested controller to achieve high-precision closed-loop control of the servo drive system by coordinating and controlling the three links of current, speed and position, including: A dynamic time-frequency evaluation index is constructed for the system operation status data, the energy distribution ratio of the multi-scale frequency band characteristics is mapped into a frequency domain evaluation sub-function, the instantaneous amplitude variance of the inherent modal characteristics is mapped into a time domain evaluation sub-function, and the dynamic weight coefficients of the time domain and frequency domain sub-functions are determined based on fuzzy hierarchical analysis to obtain an adaptive objective function; Based on the adaptive objective function, a double-layer variable structure particle swarm is constructed, wherein the outer particle swarm is responsible for optimizing the bandwidth modulation parameters, and the inner particle swarm is responsible for optimizing the damping modulation parameters. The two-layer particle swarms exchange information through a co-evolution mechanism to obtain a hierarchical particle swarm. An adaptive mutation operation is performed on the hierarchical particle swarm, the mutation probability is dynamically adjusted according to the convergence degree of the particles, Brownian mutation is applied to the particles trapped in the local optimum, and Gaussian mutation is performed on the global optimum particles to obtain a mutated particle swarm; Based on the mutated particle swarm, a competition-cooperation mechanism is established to divide the particles into a competition subgroup and a cooperation subgroup. The competition subgroup adopts a survival of the fittest strategy to update the position, and the cooperation subgroup adjusts the speed through a knowledge sharing strategy to obtain an optimized particle swarm. The optimized particle swarm is subjected to adaptive convergence control to obtain a converged particle swarm, and the converged particle swarm is subjected to optimal solution mapping to obtain a second system control parameter.

8. A closed-loop control system for a servo drive system, characterized in that: A closed-loop control method for a servo drive system according to any one of claims 1 to 7, the system comprising: Mathematical modeling module, used to perform mathematical modeling on the servo drive system and establish an initialization system model including motor equations and load dynamics equations; An acquisition module, used for acquiring the current signal and the position signal of the initialization system model, and performing Clarke transformation and Park transformation processing to obtain system operation characteristic data; A feature extraction module is used to extract features from the system operation feature data to obtain system modeling features; A calculation module, used for constructing a three-loop nested controller according to the system modeling characteristics, and calculating the first system control parameter by Lyapunov stability theory; An operation test module is used to perform operation test and time-frequency analysis on the servo drive system after the control parameters of the first system are modulated to obtain system operation status data; The optimization module is used to perform particle swarm optimization calculation on the first system control parameters based on the system operation status data to obtain second system control parameters. The second system control parameters are applied to the three-loop nested controller to achieve high-precision closed-loop control of the servo drive system by coordinating the control of the three links of current, speed and position.

Images