Hybrid excitation synchronous motor control strategy optimization method and device

Abstract

The present invention relates to the field of motor control technology, and discloses a hybrid excitation synchronous motor control strategy optimization method and device. The method includes: mathematically modeling the permanent magnet flux, excitation current and rotor parameters of the hybrid excitation synchronous motor to obtain a state observer equation; collecting current parameters and position parameters, and filtering them to obtain motor operation state parameters; inputting a five-layer deep neural network to predict current and speed parameters to obtain optimal current parameters and speed prediction parameters; constructing a multi-objective function to solve and obtain a current reference value; inputting the current reference value into a cascade PI controller, and adjusting the voltage through feedforward decoupling control and space vector pulse width modulation technology to obtain a control voltage signal; optimizing the control parameters of the control voltage signal by performance calculation to obtain target control parameters. The present invention improves the control accuracy of the hybrid excitation synchronous motor.

Description

Technical Field

[0001] The present invention relates to the technical field of motor control, and in particular to a method and device for optimizing a control strategy of a hybrid excitation synchronous motor. Background Art

[0002] The hybrid excitation synchronous motor combines the advantages of permanent magnet synchronous motor and electrically excited synchronous motor, and has the characteristics of high efficiency, high power density and adjustable magnetic field, and has broad application prospects in electric vehicles, wind power generation and other fields. However, since the hybrid excitation synchronous motor has two excitation sources, permanent magnet and excitation winding, its mathematical model is more complex, and the optimization of the control strategy faces greater challenges.

[0003] Traditional hybrid excitation synchronous motor control methods mainly use single PI control or vector control, which is difficult to take into account multiple control objectives such as copper loss minimization, torque fluctuation suppression and dynamic response under complex working conditions. In addition, existing control strategies often rely on accurate motor parameters, but in actual operation, motor parameters will change with factors such as temperature and saturation, resulting in a decrease in control performance. Summary of the invention

[0004] The present invention provides a method and a device for optimizing a control strategy of a hybrid excitation synchronous motor, and the present invention improves the control accuracy of the hybrid excitation synchronous motor.

[0005] In a first aspect, the present invention provides a hybrid excitation synchronous motor control strategy optimization method, the hybrid excitation synchronous motor control strategy optimization method comprising:

[0006] The permanent magnet flux, excitation current and rotor parameters of the hybrid excitation synchronous motor are mathematically modeled to obtain the state observer equation;

[0007] Based on the state observer equation, current parameters and position parameters are collected and filtered to obtain motor operation state parameters;

[0008] Input the motor running state parameters into a five-layer deep neural network to predict current and speed parameters, and obtain optimal current parameters and speed prediction parameters;

[0009] A multi-objective function is constructed for the optimal current parameter and the speed prediction parameter to obtain a current reference value;

[0010] The current reference value is input into a cascade PI controller, and voltage regulation is performed through feedforward decoupling control and space vector pulse width modulation technology to obtain a control voltage signal;

[0011] The control voltage signal is subjected to performance calculation and control parameter optimization to obtain target control parameters.

[0012] In a second aspect, the present invention provides a hybrid excitation synchronous motor control strategy optimization device, the hybrid excitation synchronous motor control strategy optimization device comprising:

[0013] A modeling module is used to mathematically model the permanent magnet flux, excitation current and rotor parameters of the hybrid excitation synchronous motor to obtain the state observer equation;

[0014] A processing module, used for collecting current parameters and position parameters based on the state observer equation, and performing filtering processing to obtain motor operation state parameters;

[0015] A prediction module, used for inputting the motor running state parameters into a five-layer deep neural network to predict current and speed parameters, and obtain optimal current parameters and speed prediction parameters;

[0016] A construction module, used for constructing a multi-objective function for the optimal current parameter and the speed prediction parameter, and solving to obtain a current reference value;

[0017] A regulating module, used for inputting the current reference value into a cascade PI controller, performing voltage regulation through feedforward decoupling control and space vector pulse width modulation technology, and obtaining a control voltage signal;

[0018] The optimization module is used to perform performance calculation and control parameter optimization on the control voltage signal to obtain target control parameters.

[0019] The third aspect of the present invention provides a hybrid excitation synchronous motor control strategy optimization device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the hybrid excitation synchronous motor control strategy optimization device executes the above-mentioned hybrid excitation synchronous motor control strategy optimization method.

[0020] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned hybrid excitation synchronous motor control strategy optimization method.

[0021] In the technical solution provided by the present invention, by establishing a complete mathematical model and state observer equation, combined with deep neural network prediction, the control system has self-learning ability, which effectively overcomes the influence of parameter changes; a multi-objective optimization method is adopted, and the weight coefficient is determined by a two-stage weighted least squares method, so as to achieve coordinated optimization of multiple control objectives such as copper loss minimization and torque fluctuation suppression; the Fisher information matrix and Cramér-Rao lower bound theory are introduced for performance evaluation, which provides a theoretical basis for control parameter optimization and gets rid of the limitations of traditional empirical setting; a voltage regulation scheme based on feedforward decoupling control and space vector PWM is designed, which reduces the coupling degree of the control system and improves the dynamic response performance; through the combination of deep neural network and traditional control method, an adaptive intelligent control system is constructed, which enhances the robustness and generalization ability of the control system; a hierarchical and progressive control strategy optimization method is adopted, the logical relationship between each control link is clear, and the control accuracy of the hybrid excitation synchronous motor is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0023] Figure 1 A flow chart of a hybrid excitation synchronous motor control strategy optimization method provided in an embodiment of the present application;

[0024] Figure 2 A schematic block diagram of the structure of a hybrid excitation synchronous motor control strategy optimization device provided in an embodiment of the present application;

[0025] Figure 3 A schematic block diagram of the structure of a hybrid excitation synchronous motor control strategy optimization device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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.

[0027] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change based on actual conditions.

[0028] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0029] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0030] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0031] See also Figure 1 , Figure 1 A flow chart of a hybrid excitation synchronous motor control strategy optimization method provided in an embodiment of the present application, such as Figure 1 As shown, the hybrid excitation synchronous motor control strategy optimization method provided in the embodiment of the present application includes steps S100 to S600.

[0032] Step S100, mathematically modeling the permanent magnet flux, excitation current and rotor parameters of the hybrid excitation synchronous motor to obtain a state observer equation;

[0033] It is understandable that the execution subject of the present invention may be a hybrid excitation synchronous motor control strategy optimization device, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0034] Specifically, the basic electromagnetic behavior of the motor is mathematically modeled to enable accurate state estimation and dynamic optimization in the subsequent control process. The stator voltage and flux are mathematically modeled, and the stator voltage is modeled using dq axis decomposition to obtain the voltage equations of the d-axis and q-axis. The d-axis voltage equation is ,in is the d-axis stator voltage, is the d-axis stator current, is the d-axis magnetic flux, is the electrical angular velocity, is the stator winding resistance, and is the q-axis magnetic flux, is the electrical angular velocity. The q-axis voltage equation is ,in is the q-axis stator voltage, is the q-axis stator current, is the q-axis flux. The stator flux is calculated and the d-axis and q-axis flux equations are obtained. The d-axis flux equation is ,in is the d-axis stator self-inductance, is the mutual inductance between the stator and the field winding, is the field winding current is the basic flux generated by the permanent magnet; and the q-axis flux equation is ,in is the q-axis stator self-inductance, is the q-axis stator current. Through these two sets of flux equations, the calculation formula based on the electromagnetic torque is Get the electromagnetic torque of the motor, where Te is the electromagnetic torque, is the number of motor pole pairs, and are the currents of the d-axis and q-axis respectively, is the mutual inductance between the stator and the field winding, is the basic flux generated by the permanent magnet. In order to describe the mechanical motion of the motor, the mechanical motion equation is established ,in is the rotor moment of inertia, is the mechanical angular velocity, is the external load torque, is the viscous friction coefficient of the rotor bearing, which in turn links the dynamic behavior of the motor with the electromagnetic torque and mechanical characteristics. In the current dynamic response analysis, based on the voltage equation obtained previously, the differential equations of the d-axis and q-axis currents are obtained: and These current differential equations reveal the relationship between the dynamic changes of the stator current and the voltage input. Combining the current differential equations with the mechanical motion equations, we get the state space equations ,in is the state vector, representing the current and mechanical angular velocity, is the control input vector, and are the 3×3 order state matrix and input matrix respectively. These state space equations provide the basis for describing the motor dynamics for the control system, which is used for subsequent system analysis and control design. In order to perform numerical solutions in the discrete time domain, the state space equations are discretized and numerically integrated using the fourth-order Runge-Kutta method to obtain the discrete state equations ,in is the discrete time point number, is the sampling time interval, is the nonlinear function on the right side of the state equation. Based on the discrete state equation, the state observer equation is established ,in is the state observation vector, representing the estimated values ​​of current and mechanical angular velocity, is the actual measured output vector, is the observer output, is the observer gain matrix. Through this state observer, the state of the motor is estimated without directly measuring the rotor position, improving the robustness and control accuracy of the system.

[0035] Step S200, collecting current parameters and position parameters based on the state observer equation, and performing filtering processing to obtain motor operation state parameters;

[0036] Specifically, the three-phase stator current is collected by the Hall current sensor to obtain the original three-phase current data. At the same time, the current of the excitation winding is collected by the Hall current sensor to obtain the original excitation current data. The original three-phase current data and excitation current data are processed by a fourth-order Butterworth low-pass filter. The Butterworth filter is an electronic filter used for signal filtering. Because it exhibits a flat frequency response within the bandpass frequency range, it can effectively remove high-frequency noise while retaining the low-frequency components of the signal, and is suitable for current signal processing. After filtering, smoother and more accurate current data is obtained. At the same time, a magnetic encoder is used for rotor position acquisition. The magnetic encoder can accurately measure the angle information of the rotor and obtain the original rotor position data. By combining the position information with the time difference, the M / T method (position-time method) is used to calculate the mechanical angular velocity of the motor. The M / T method calculates the mechanical angular velocity based on the relationship between the rotor position change and the sampling time. The filtered three-phase current data is transformed into the dq axis coordinate system by Clark-Park coordinate transformation. Clark-Park transformation is a motor control algorithm that converts three-phase AC current signals into DC current signals, making the control of the motor simpler and more stable. Through this coordinate transformation, the current components on the d-axis and q-axis are obtained, reflecting the state information of the motor. The M / T method speed calculation is performed based on the original position data. The rotor position data is converted into the motor speed through the M / T method to obtain the mechanical angular velocity. The dq-axis current components, excitation current and mechanical angular velocity are substituted into the previously established state observer equation. The state observer equation can obtain the state observation vector of the motor by processing these input data. The vector contains the current, speed and other key state parameters of the motor. The observer gain matrix is ​​optimized. By calculating the observation error, that is, the difference between the observed value and the actual measured value, the observer gain is adjusted according to the error feedback to improve the accuracy of the state estimation. The observer gain matrix is ​​updated by the recursive least squares method. The recursive least squares method is an online optimization algorithm that recursively updates the observation error and continuously adjusts the gain matrix, so that the system can continuously perform self-optimization in a dynamic environment and improve the observation accuracy of the motor operation state.

[0037] Step S300, inputting the motor running state parameters into a five-layer deep neural network to predict the current and speed parameters, and obtaining the optimal current parameters and speed prediction parameters;

[0038] Specifically, the motor running state parameters are standardized to obtain a standardized input vector to avoid the adverse effects of parameters of different dimensions on model training. The standardized input vector is input into the first hidden layer of the five-layer deep neural network. In order to improve the expression ability and computational efficiency of the neural network, 16 neurons are set in the first hidden layer, and the feature vector of the first layer is calculated through the first weight matrix and the first bias vector. Each neuron nonlinearly maps the input data through the activation function to generate a meaningful feature vector as the input of the next layer of the network. The feature vector of the first layer is input into the second hidden layer. In the second hidden layer, 32 neurons are set, and the function of this layer is to extract higher-order feature information. The feature vector of the second layer is calculated through the second weight matrix and the second bias vector, and carries more high-dimensional features. The feature vector of the second layer is input into the third hidden layer. The third hidden layer contains 64 neurons, and its function is deep learning and feature extraction. The feature vector of the third layer is calculated through the third weight matrix and the third bias vector. The third hidden layer has a strong learning ability and can capture the deep-level pattern of the motor running state in a large amount of training data, thereby improving the expression ability and prediction accuracy of the model. The third layer feature vector is input into the fourth hidden layer. The fourth layer is provided with 32 neurons, which refine and extract feature information and calculate the fourth layer feature vector. Through the fourth weight matrix and the fourth bias vector, the network can perform fine mapping of the input features. The feature vector of the fourth layer will be input into the fifth layer of the five-layer deep neural network for the final prediction of current and speed. The fifth layer is divided into a current prediction unit and a speed prediction unit, which are used to predict the current and speed of the motor respectively. The current prediction unit contains 3 neurons, through which the neural network can output a standardized current prediction value. In order to match the output result with the current value in the actual application, the standardized current prediction value is de-standardized to obtain the optimal current parameter. At the same time, the speed prediction unit contains a neuron, which uses a linear activation function for output mapping. Since the speed prediction value is a continuous scalar quantity, the use of a linear activation function can ensure the linear relationship of the output value. Similar to the current prediction, the speed prediction value also needs to be de-standardized to finally obtain the optimal speed prediction parameter.

[0039] Step S400, constructing a multi-objective function for the optimal current parameter and the speed prediction parameter, and solving to obtain a current reference value;

[0040] Specifically, the objective function is decomposed based on the optimal current parameter and the speed prediction parameter to obtain the copper loss function, the torque-current ratio function and the torque fluctuation function. The copper loss function is used to describe the loss generated when the current passes through the stator winding of the motor, the torque-current ratio function reflects the relationship between the current and the torque, and the torque fluctuation function describes the fluctuation of the motor torque output. Each objective function has different physical meanings, and is optimized for different aspects such as loss, efficiency, and stability in motor operation. Based on the copper loss function, the torque-current ratio function and the torque fluctuation function, a multi-objective function is constructed to obtain a composite objective function, and multiple optimization requirements in motor operation are considered at the same time, so as to find a balance point in the control and achieve the optimal control effect. Different objective functions are integrated through weighted coefficients in a weighted manner to ensure that each objective receives appropriate attention in the overall optimization. The sensitivity of the motor's operating history data sequence is calculated to obtain the sensitivity matrix of each objective function relative to the control target. The sensitivity matrix reveals the relationship between different objective functions and the motor control target, and reflects the sensitivity of each control target to the changes in different objective functions. Based on the sensitivity matrix, the first-stage weight equation is solved using the least squares method to obtain the initial weight coefficient. The least squares method is an optimization method that solves the problem by minimizing the sum of squared errors and determines the initial weight coefficients of each objective function. The initial weight coefficients reflect the importance of different objective functions in the overall optimization. The initial weight coefficients are normalized to ensure that the weights of each objective function are within a unified standard range, and the weight coefficients of the first stage are obtained. The normalized weight coefficients are substituted into the composite objective function to construct the weighted least squares equation of the second stage. By solving the weighted least squares equation, the final weight coefficients are obtained, reflecting the final importance of each objective function in the optimization process. The final weight coefficients are substituted into the composite objective function and iterative optimization calculations are performed using the gradient descent algorithm. The gradient descent algorithm is an optimization method that gradually reduces the value of the objective function by continuously adjusting the control variables until the optimal solution is reached. In this process, the optimized current parameters are adjusted as control variables so that the final current value can obtain the optimal control effect while considering multiple objectives such as copper loss, torque-to-current ratio, and torque fluctuation. After the optimization calculation is completed, the optimized current parameters are subjected to operation constraint processing to ensure that the current reference value meets the physical constraints in actual operation, such as the limitation of the current amplitude, system protection requirements, etc., and the current reference value is finally obtained.

[0041] Step S500, inputting the current reference value into the cascade PI controller, performing voltage regulation through feedforward decoupling control and space vector pulse width modulation technology, and obtaining a control voltage signal;

[0042] Specifically, the current reference value is input into the outer loop of the cascade PI controller. The outer loop PI controller calculates the voltage instructions of the d-axis and q-axis according to the input current reference value through the proportional coefficient and integral coefficient of the outer loop. The outer loop proportional control part adjusts the amplitude of the d-axis and q-axis voltage instructions according to the current error, while the integral control part is used to eliminate the long-term steady-state error of the system and provide accurate voltage instructions. The d-axis and q-axis voltage instructions are input into the inner loop of the cascade PI controller. The inner loop PI controller further adjusts these instructions to ensure that the voltage instructions of the d-axis and q-axis more accurately meet the requirements of the motor. The inner loop control is adjusted through the proportional coefficient and the integral coefficient. The role of the inner loop is to refine the output of the voltage instruction to ensure that the voltage can effectively track the expected value and provide a more stable motor operation state. Through the adjustment of the inner loop, the basic voltage instruction is obtained. The basic voltage instruction is fed forward and decoupled to eliminate the coupling effect between the d-axis and the q-axis in the system, so that the voltage control of the two axes can be independent of each other and adjusted separately under different working conditions. Through the decoupling calculation, the decoupled voltage component is obtained to eliminate the negative impact between each other and improve the control accuracy. The basic voltage command and the decoupled voltage component are superimposed by voltage synthesis to obtain the dq axis voltage command vector. The final dq axis voltage command vector is transformed by Clarke transformation from the dq coordinate system to the α-β coordinate system. Clarke transformation transforms the voltage vector from the coordinate system of the direct axis and the quadrature axis to the plane coordinate system through a transformation matrix, making the subsequent voltage control more intuitive. The voltage vector in the α-β coordinate system is judged by sector. According to the position of the voltage vector in the coordinate system, the specific sector where the voltage vector is located is determined by amplitude calculation and angle calculation. Based on the sector number and voltage vector information, the action time of adjacent basic vectors is calculated to determine the action time ratio of different voltage vectors on the motor during SVPWM modulation, so as to achieve more accurate voltage control. At the same time, the action time of the zero vector is calculated, and the action time of the zero vector will be allocated according to the control needs to ensure the smoothness of the voltage output. Based on the basic vector action time and the zero vector action time, the seven-segment SVPWM modulation method is used for timing allocation. The seven-segment SVPWM modulation generates the three-phase six-channel PWM switching signal required by the motor by reasonably allocating the action time of different vectors. Through the output of these PWM signals, the voltage signal is precisely controlled, thereby adjusting the operating state of the motor to ensure that the motor can work stably according to the set operating conditions.

[0043] Step S600: Calculate the performance of the control voltage signal and optimize the control parameters to obtain target control parameters.

[0044] Specifically, the performance index of the control voltage signal is calculated to obtain relevant performance indexes, including speed dynamic response time, speed regulation accuracy, torque fluctuation rate and system efficiency. These indexes reflect the dynamic characteristics and stability of the motor in operation. For example, the speed dynamic response time measures the time required for the motor to reach a stable operating state from the initial state, the speed regulation accuracy measures the accuracy of the motor in steady-state operation, and the torque fluctuation rate indicates the fluctuation of the motor output torque. The system efficiency reflects the utilization efficiency of the energy of the entire system. Based on the above performance indicators, the probability density function of the parameter system is constructed, and the log-likelihood function is established using these data. The log-likelihood function is used to describe the performance of the current system parameters under the performance indicators. By solving the log-likelihood function, the influence of the system parameters on the overall performance can be understood. The log-likelihood function can quantify the relationship between the performance indicators and the parameters. By performing the second-order partial derivative operation on it, the Fisher information matrix is ​​obtained. The Fisher information matrix is ​​a statistical tool used to measure the uncertainty of parameter estimation in the system. It is essentially a measure of the sensitivity of parameter estimation and can reveal the interaction between parameters and performance. The Fisher information matrix is ​​inverted to obtain the covariance matrix of parameter estimation. The covariance matrix is ​​used to describe the confidence interval of parameter estimation and its correlation. By solving this matrix, we can understand the estimation accuracy of each parameter and the interdependence between different parameters. Perform eigenvalue decomposition on the covariance matrix to obtain the eigenvector of parameter sensitivity. The eigenvector represents the sensitivity of different control parameters to system performance, that is, the degree of influence of the change of a certain parameter on the performance of the entire system. According to the eigenvector, the control parameters are sorted by importance to generate a sequential sequence for parameter optimization. By sorting these parameters, it is ensured that during the optimization process, the parameters with greater influence on system performance are given priority, the efficiency and accuracy of optimization are improved, and computing resources are avoided from being wasted on unimportant parameters. According to the sequential sequence of parameter optimization, the control parameters are adjusted and optimized using the gradient optimization method. The gradient optimization method calculates the gradient of the objective function with respect to each parameter and continuously adjusts the parameter value along the direction of the gradient until a solution that makes the objective function optimal is found. Through the optimization process, the optimized control parameters are obtained. The stability analysis of the optimized control parameters is performed to ensure that the optimized control parameters can maintain the stability of the system under various operating conditions and will not introduce additional oscillations or poor dynamic characteristics. Through stability analysis, the factors that cause system instability are identified, so as to correct the optimization results and ensure that the final control parameters have good stability and reliability in practical applications. Finally, the target control parameters are obtained.

[0045] In the embodiment of the present invention, by establishing a complete mathematical model and state observer equation, combined with deep neural network prediction, the control system has self-learning ability, which effectively overcomes the influence of parameter changes; a multi-objective optimization method is adopted, and the weight coefficient is determined by the two-stage weighted least squares method, so as to achieve coordinated optimization of multiple control objectives such as copper loss minimization and torque fluctuation suppression; the Fisher information matrix and Cramér-Rao lower bound theory are introduced for performance evaluation, which provides a theoretical basis for control parameter optimization and gets rid of the limitations of traditional empirical setting; a voltage regulation scheme based on feedforward decoupling control and space vector PWM is designed, which reduces the coupling degree of the control system and improves the dynamic response performance; through the combination of deep neural network and traditional control method, an adaptive intelligent control system is constructed, which enhances the robustness and generalization ability of the control system; a hierarchical and progressive control strategy optimization method is adopted, the logical relationship between each control link is clear, and the control accuracy of the hybrid excitation synchronous motor is improved.

[0046] In a specific embodiment, the process of executing step S100 may specifically include the following steps:

[0047] The stator voltage of the hybrid excitation synchronous motor is decomposed into d-q axes to obtain the d-axis voltage equation u d =R s i d +dψ d / dt-ωψ q and the q-axis voltage equation u q =R s i q +dψ q / dt+ωψ d , where u d is the d-axis stator voltage, u q is the q-axis stator voltage, R s is the stator winding resistance, i d is the d-axis stator current, i q is the q-axis stator current, ψ d is the d-axis magnetic flux, ψ q is the q-axis magnetic flux, ω is the electrical angular velocity, and t is the time variable;

[0048] The stator flux is calculated and the d-axis flux equation ψ is obtained. d =L d i d +M f i f +ψ f and the q-axis flux equation ψ q =L q i q , where L d is the d-axis stator self-inductance, L qis the q-axis stator self-inductance, M f is the mutual inductance between the stator and the field winding, i f is the field winding current, ψ f is the basic flux linkage generated by the permanent magnet;

[0049] The electromagnetic torque is calculated based on the d-axis flux equation and the q-axis flux equation, and the torque equation T is obtained. e =1.5p[(L d -L q )i d i q +(M f i f +ψ f )i q ], where T e is the electromagnetic torque, p is the number of motor pole pairs;

[0050] The mechanical motion of the rotor is described mathematically, and the mechanical motion equation J(dω m / dt)=T e -T L -Bω m , where J is the rotor moment of inertia, ω m is the mechanical angular velocity and ω m =ω / p,T L is the external load torque, B is the viscous friction coefficient of the rotor bearing;

[0051] According to the d-axis voltage equation and the q-axis voltage equation, the current dynamic response analysis is carried out and the current differential equation di is obtained. d / dt=(u d -R s i d +ωL q i q ) / Ld and di q / dt=(u q -R s i q -ω(L d i d +M f i f +ψ f )) / L q ;

[0052] Combining the current differential equation and the mechanical motion equation, we get the state space equation dx / dt=Ax+Bu, where x=[i d ,i q ,ω m ]T is the state vector, u=[u d ,u q ,i f]T is the control input vector, A is the 3×3 state matrix, and B is the 3×3 input matrix;

[0053] The state space equation is discretized and numerically solved using the fourth-order Runge-Kutta method to obtain the discrete state equation x(k+1)=x(k)+Δt·f(x(k),u(k)), where k is the discrete time point number, k+1 is the next discrete time point number, Δt is the sampling time interval, and f(·) is the nonlinear function on the right side of the state equation;

[0054] Based on the discrete state equation, the state observer equation x̂(k+1)=x̂(k)+Δt·f(x̂(k),u(k))+L(y(k)-ŷ(k)) is established, where x̂ is the state observation vector [î d ,î q ,ω̂ m ]T,î d is the observed value of d-axis current, q is the q-axis current observation value, ω̂ m is the mechanical angular velocity observation value, y(k) is the actual measured output vector [i d ,i q ,ω m ]T, ŷ(k) is the observed output vector [ d ,î q ,ω̂ m ]T, L is the 3×3 order observer gain matrix.

[0055] Specifically, the stator voltage of the motor is decomposed into dq axes. Dq axis decomposition is a method for synchronous motor control. By converting the three-phase stator voltage of the motor into the direct axis (d axis) and the quadrature axis (q axis), the motor dynamic equation is simplified and the motor control is made more intuitive. After the dq axis decomposition, two voltage equations are obtained, namely the d-axis voltage equation and the q-axis voltage equation:

[0056] ;

[0057] ;

[0058] in, and are the stator voltages of the d-axis and q-axis respectively, is the resistance of the stator winding, and are the stator currents of the d-axis and q-axis respectively, and are the magnetic flux of d-axis and q-axis respectively, is the electrical angular velocity, is a time variable. The stator flux is calculated by components to obtain the expressions of d-axis and q-axis flux. The calculation of stator flux involves the contribution of stator self-inductance, mutual inductance and permanent magnet. The d-axis flux equation is expressed as:

[0059] ;

[0060] in, is the d-axis stator self-inductance, is the mutual inductance between the stator and the field winding, is the field winding current, is the basic flux generated by the permanent magnet. Similarly, the q-axis flux equation is:

[0061] ;

[0062] in, is the q-axis stator self-inductance. Through these equations, the components of the stator flux on the d-axis and q-axis are obtained. Based on the flux equations of the d-axis and q-axis, the electromagnetic torque is calculated. The electromagnetic torque is a key physical quantity in the operation of the motor, and its calculation formula is:

[0063] ;

[0064] in, is the electromagnetic torque, is the number of pole pairs of the motor. This formula describes the source of the motor torque, including the torque contribution from the stator self-inductance, as well as the torque contribution from the excitation current and the permanent magnet flux. The mechanical motion of the rotor is mathematically described. The mechanical motion of the rotor is determined by its moment of inertia, torque, and external load torque. The equation of the rotor mechanical motion is expressed as:

[0065] ;

[0066] in, is the rotor moment of inertia, is the mechanical angular velocity (and ,in is the electrical angular velocity, is the number of motor pole pairs), is the external load torque, is the viscous friction coefficient of the rotor bearing. This equation describes the relationship between the change in rotor angular velocity and the electromagnetic torque, load torque and friction. In order to further analyze the dynamic characteristics of the motor, the current dynamic response analysis is performed based on the d-axis and q-axis voltage equations. By solving the differential equations of the d-axis and q-axis currents, the time response of the current is obtained. The d-axis current differential equation is expressed as:

[0067] ;

[0068] The q-axis current differential equation is:

[0069] ;

[0070] These equations describe the variation of the stator current on the d-axis and q-axis, and take into account the influence of factors such as voltage, stator resistance, flux linkage, mutual inductance, etc. Combining the current differential equation with the rotor mechanical motion equation, a complete state space equation is obtained. The state space equation is a tool for describing dynamic systems. It combines the input, output and state variables of the system to facilitate the solution of the dynamic response of the system. The state space equation is expressed as:

[0071] ;

[0072] Among them, the state vector Contains d-axis current, q-axis current and mechanical angular velocity, control input vector Contains d-axis voltage, q-axis voltage and excitation current, is the state matrix, is the input matrix. In order to solve numerically, the state space equation is discretized. The fourth-order Runge-Kutta method is used for numerical solution, which can efficiently solve nonlinear differential equations and ensure high accuracy. The discretized state equation is expressed as:

[0073] ;

[0074] in, is the sequence number of the discrete time point, is the sampling time interval, is the nonlinear function on the right side of the state equation. The state observer equation is established based on the discretized state equation. The state observer is used to estimate the internal state of the system through the output information of the system. The state observer equation is expressed as:

[0075] ;

[0076] in, is the state observation vector, is the actual measured output vector, is the observed output vector, is the observer gain matrix. The state observer can effectively estimate the system states such as current and speed, and provide support for subsequent control strategies. Through the above steps, the dynamic behavior of the hybrid excitation synchronous motor is accurately modeled, and analyzed and controlled by combining the state space method and numerical solution method.

[0077] In a specific embodiment, the process of executing step S200 may specifically include the following steps:

[0078] The three-phase stator current is collected by a Hall current sensor to obtain the original three-phase current data, and the excitation winding current is collected by a Hall current sensor to obtain the original excitation current data;

[0079] The rotor position is collected by a magnetic encoder to obtain the original position data, and the original three-phase current data and the original excitation current data are subjected to fourth-order Butterworth low-pass filtering to obtain filtered current data;

[0080] The Clark-Park coordinate transformation is performed on the filtered three-phase current data to obtain the dq axis current components, and the M / T method speed calculation is performed based on the original position data to obtain the mechanical angular velocity;

[0081] Substitute the dq axis current components, the excitation current and the mechanical angular velocity into the state observer equation to obtain the state observation vector;

[0082] The observation error is calculated according to the state observation vector, and the observer gain matrix is ​​updated by the recursive least square method to obtain the motor operation state parameters.

[0083] Specifically, the three-phase stator current and the excitation winding current are collected. The three-phase stator current is measured by a Hall current sensor, which measures the current intensity in real time and obtains the original three-phase current data by outputting a voltage signal proportional to the current, which is recorded as , representing the current of stator winding phase A, B and C respectively. Similarly, the current of the excitation winding is collected by the Hall current sensor to obtain the original excitation current data , which controls the excitation system of the motor. In order to reduce the influence of noise and smooth the data, the original three-phase current data and excitation current data Filtering is performed. A fourth-order Butterworth low-pass filter is used. Butterworth filter is a filter with a flat frequency response. Its characteristic is that it effectively transmits low-frequency signals and attenuates high-frequency noise signals. The fourth-order Butterworth filter can effectively reduce high-frequency noise in the signal by designing a transfer function with four poles. Set the cutoff frequency to , then the filtered current data is expressed as:

[0084] ;

[0085] These filtered signals are used as input data for subsequent processing. The filtered three-phase current data is converted to the dq axis coordinate system through Clark-Park coordinate transformation. Clark-Park transformation is a coordinate transformation method that converts the current data in the three-phase stationary coordinate system to the rotating coordinate system (dq axis), thereby simplifying the mathematical modeling and control design of the motor. Clark transformation converts the three-phase current Converted into the current components in the two-phase stationary coordinate system , the formula is:

[0086] ;

[0087] ;

[0088] The Park transformation Converted to the current component in the rotating coordinate system (dq axis) The formula for Park transformation is:

[0089] ;

[0090] ;

[0091] in, The rotor position is obtained by a rotor position sensor (such as a magnetic encoder). The magnetic encoder accurately measures the rotor position and converts it into a digital signal to obtain the rotor position data. . Using this data, calculate the angular velocity of the rotor (unit: rad / s), i.e., the speed, is calculated using the M / T method (i.e., the mechanical speed method):

[0092] ;

[0093] At this time, the d-axis current is obtained , q-axis current and mechanical angular velocity The measured current component and mechanical angular velocity are substituted into the state observer equation to obtain the state observation vector of the motor. The purpose of the state observer is to estimate the internal state of the motor, especially when it cannot be measured directly, and relies on the signal obtained by the sensor for inference. The state observer model used is as follows:

[0094] ;

[0095] in, is the state observation vector, containing the observed value of the d-axis current , q-axis current observation value and the observed value of the mechanical angular velocity , is the actual measured output vector, is the observed output vector. Gain matrix is an important parameter that determines the convergence speed and accuracy of the observer. By introducing the gain matrix , the state observer can update its estimated value at each time step and gradually approach the actual motor state. Based on the observation vector, the observation error is calculated, and the observer gain matrix is ​​calculated using the recursive least squares method. Update. Recursive least squares is a method for dynamically adjusting model parameters, optimizing parameter estimates by minimizing the sum of squares of observation errors. At each iteration, the recursive least squares method updates the gain matrix based on the current error and historical data, as follows:

[0096] ;

[0097] in, is the learning rate, is the error covariance matrix. The gain matrix is ​​updated by iteration , can effectively estimate the operating state of the motor and obtain accurate motor state parameters such as current, speed, etc.

[0098] In a specific embodiment, the process of executing step S300 may specifically include the following steps:

[0099] Standardize the motor operation state parameters to obtain a standardized input vector;

[0100] The standardized input vector is input into the first hidden layer of the five-layer deep neural network, the first hidden layer is set with 16 neurons, and the first layer feature vector is calculated by the first weight matrix and the first bias vector;

[0101] The first layer feature vector is input into the second hidden layer of the five-layer deep neural network, the second hidden layer is set with 32 neurons, and the second layer feature vector is calculated by the second weight matrix and the second bias vector;

[0102] The second layer feature vector is input into the third hidden layer of the five-layer deep neural network, the third hidden layer is set with 64 neurons, and the third layer feature vector is calculated by the third weight matrix and the third bias vector;

[0103] The third layer feature vector h3 is input into the fourth hidden layer of the five-layer deep neural network, the fourth hidden layer is set with 32 neurons, and the fourth layer feature vector is calculated by the fourth weight matrix and the fourth bias vector;

[0104] The fourth layer feature vector is input into the fifth layer of the five-layer deep neural network. The fifth layer is divided into a current prediction unit and a speed prediction unit. The current prediction unit contains 3 neurons. The standardized current prediction value is obtained, and the standardized current prediction value is de-standardized to obtain the optimal current parameter.

[0105] The fourth layer feature vector is input into the speed prediction unit, which contains 1 neuron. The linear activation function is used for output mapping and denormalization to obtain the speed prediction parameters.

[0106] Specifically, the motor operating state parameters are standardized to eliminate the dimensional differences between different parameters, so that the neural network can converge faster and improve accuracy during training and prediction. The standardization process uses the mean standardization or minimum-maximum normalization method. Assuming that the motor operating state parameters are , its standardized form is expressed as:

[0107] ;

[0108] in, is the standardized input data, is the mean of the original data, is the standard deviation of the original data. After standardization, the standardized input vector As the input of the deep neural network. The normalized input vector Input to the first hidden layer of the five-layer deep neural network. The first hidden layer contains 16 neurons, which means that this layer has 16 feature maps. The output feature vector of this layer is determined by the weight matrix and the bias vector. The specific calculation formula is:

[0109] ;

[0110] in, is the first layer feature vector, is the weight matrix of the first hidden layer, and its dimension is ( is the dimension of the input vector), is the bias vector of the first hidden layer, is an activation function (such as the ReLU function), which introduces nonlinearity so that the network can learn complex features. The input vector is mapped to a 16-dimensional feature space, and the bias vector is used to adjust the mapping result so that the model has better generalization ability. Input to the second hidden layer of the five-layer deep neural network. The second hidden layer contains 32 neurons, and the calculation formula of its output feature vector is:

[0111] ;

[0112] in, is the second layer feature vector, is the weight matrix of the second hidden layer, and its dimension is is the bias vector of the second hidden layer, The second hidden layer allows the model to further extract complex features from the input data and enhance the understanding of nonlinear relationships in the data. It is input to the third hidden layer, which has 64 neurons and is the largest layer in the entire network. Its function is to enrich the model's ability to express features. The calculation formula for its output feature vector is:

[0113] ;

[0114] in, is the third-layer feature vector, is the weight matrix of the third hidden layer, with dimension is the bias vector of the third hidden layer. By increasing the number of neurons, the third hidden layer better captures high-order features and complex patterns in the data. The third layer feature vector is input to the fourth hidden layer. The fourth hidden layer contains 32 neurons, which is used to refine the extracted high-order features. The calculation formula of its output feature vector is:

[0115] ;

[0116] in, is the fourth-layer feature vector, is the weight matrix of the fourth hidden layer, with dimension is the bias vector of the fourth hidden layer. Through this layer, the model extracts and compresses the feature vector. Input to the fifth layer. The fifth layer contains two different prediction units: current prediction unit and speed prediction unit. The current prediction unit contains 3 neurons, and its output is the standardized current prediction value. The calculation formula is:

[0117] ;

[0118] in, is the normalized current prediction value, which contains three components: is the weight matrix of the current prediction unit, with dimension is the bias vector of the current prediction unit, is the activation function. In order to obtain the final current prediction value, the standardized current is de-standardized:

[0119] ;

[0120] in, For the The final predicted value of the current component, and are the standard deviation and mean used in standardization. The de-standardization process can restore the output of the model to its original dimension, so that the predicted value has actual physical meaning. At the same time, the fourth layer feature vector Input to the speed prediction unit. The speed prediction unit contains 1 neuron, which is used to predict the mechanical angular velocity. The calculation formula is:

[0121] ;

[0122] in, is the normalized speed prediction value, is the weight matrix of the speed prediction unit, with the dimension , is the bias scalar of the speed prediction unit. Since the speed is a continuous quantity, a linear activation function is used for output mapping, that is, no nonlinear transformation is performed on the output. The normalized speed is denormalized to obtain the final speed prediction parameter:

[0123] ;

[0124] in, is the final predicted value of mechanical angular velocity, and are the standard deviation and mean of the rotation speed during normalization, respectively.

[0125] In a specific embodiment, the process of executing step S400 may specifically include the following steps:

[0126] The objective function is decomposed based on the optimal current parameter and the speed prediction parameter to obtain the copper loss function, the torque-current ratio function and the torque fluctuation function;

[0127] Based on the copper loss function, torque-current ratio function and torque fluctuation function, a multi-objective function is constructed to obtain a composite objective function;

[0128] The sensitivity of the operation history data sequence of the hybrid excitation synchronous motor is calculated to obtain the sensitivity matrix of each objective function relative to the control target;

[0129] Based on the sensitivity matrix, the first-stage weight equation is solved by the least square method to obtain the initial weight coefficient, and the initial weight coefficient is normalized to obtain the first-stage weight coefficient;

[0130] Substitute the weight coefficient of the first stage into the composite objective function, construct the weighted least squares equation of the second stage, and obtain the final weight coefficient by solving it;

[0131] The final weight coefficient is substituted into the composite objective function, and the optimization calculation is performed through gradient descent iteration to obtain the optimized current parameters, and the optimized current parameters are subjected to operation constraint processing to obtain the current reference value.

[0132] Specifically, the objective function is decomposed based on the optimal current parameter and the speed prediction parameter. The energy consumption and performance indicators of the motor are quantified and decomposed into three parts: copper loss function, torque-current ratio function, and torque fluctuation function. The copper loss function is used to measure the electric energy loss of the motor, which is mainly determined by the power loss generated by the current in the stator winding. The mathematical expression of the copper loss function is:

[0133] ;

[0134] in, represents copper loss, is the stator winding resistance, and are the currents of the d-axis and q-axis respectively. The purpose of the copper loss function is to minimize the power loss to improve the overall efficiency of the system. Next is the torque-current ratio function, which reflects the relationship between the motor output torque and the input current, with the purpose of improving the torque generation efficiency. The torque-current ratio function is defined as:

[0135] ;

[0136] in, is the torque-to-current ratio function, is the electromagnetic torque, and Still the d-axis and q-axis currents. In order to obtain the largest possible torque output at the same current, the ratio is maximized during the optimization process. By adding a negative sign to the formula, it can adapt to the criteria of the minimization problem. The third objective function is the torque fluctuation function, which is used to measure the fluctuation of the output torque of the motor during operation. Stable torque is essential for the smooth operation of the motor, so the torque fluctuation is minimized. The torque fluctuation function is expressed as:

[0137] ;

[0138] in, represents the torque fluctuation function, is the electromagnetic torque at a certain moment, is the average electromagnetic torque, is a cycle time. This function quantifies the degree of torque fluctuation by calculating the mean square error of the torque deviation from the mean value, and the goal is to minimize it. A composite objective function is constructed based on the copper loss function, the torque-current ratio function, and the torque fluctuation function , which is used to balance multiple objectives. The composite objective function is constructed as follows:

[0139] ;

[0140] in, are the weight coefficients of the copper loss function, torque-current ratio function, and torque fluctuation function, respectively. These weight coefficients are used to adjust the relative importance of different objectives in the optimization process. In order to optimize the weight coefficients, a sensitivity analysis is performed on the operating history data of the hybrid excitation synchronous motor to obtain the sensitivity matrix of each objective function relative to the control target. The sensitivity matrix describes the response of each objective function to changes in the control parameters, and its elements are calculated by numerical differentiation, for example:

[0141] ;

[0142] in, Indicates The objective function is The sensitivity of the control parameters, is the objective function, is the control parameter. After obtaining the sensitivity matrix The first stage weight equation is solved using the least squares method to determine the initial weight coefficients The goal of the least squares method is to make the composite objective function have the smallest overall error under different control parameters, which is achieved by solving the following equation:

[0143] ;

[0144] in, represents the initial weight vector, is the target response vector. The initial weight coefficients are normalized to ensure that the sum of all weights is 1, and the normalized first-stage weight coefficients are obtained:

[0145] ;

[0146] Substitute the weight coefficients of the first stage into the composite objective function to construct the weighted least squares equation of the second stage. By solving the weighted least squares equation, the weight coefficients are optimized, making the composite objective function more stable under different operating conditions. The weight solution formula for the second stage is as follows:

[0147] ;

[0148] in, is a regularization parameter used to balance between keeping the weight unchanged in the first stage and minimizing the error. The final weight coefficient is substituted into the composite objective function and optimized by the gradient descent algorithm to obtain the optimal current parameter. The gradient descent algorithm continuously calculates the gradient of the composite objective function with respect to the control parameter and adjusts the control parameter value along the direction of the gradient descent. Its iterative formula is:

[0149] ;

[0150] in, For the The control parameters after iterations are is the learning rate, which determines the step size of each iteration. Through continuous iteration, the gradient descent algorithm eventually finds the control parameters that minimize the composite objective function. The optimized current parameters are processed by operation constraints to ensure that they meet the physical constraints and operation safety requirements of the motor. The current amplitude is limited to the range allowed by the motor design to prevent overheating or damage to the motor due to excessive current. Constraint processing is achieved through projection algorithms, for example:

[0151] ;

[0152] in, is the d-axis current reference value after constraint processing, To optimize the obtained d-axis current, and are the maximum and minimum allowed values ​​of the d-axis current, respectively. Similar constraints apply to the q-axis current .

[0153] In a specific embodiment, the process of executing step S500 may specifically include the following steps:

[0154] The current reference value is input into the outer loop of the cascade PI controller, and the d-axis voltage command and the q-axis voltage command are obtained by calculating the outer loop proportional coefficient and the outer loop integral coefficient;

[0155] The d-axis voltage command and the q-axis voltage command are subjected to inner-loop PI control, and the inner-loop proportional coefficient and the inner-loop integral coefficient are used to adjust the control to obtain the basic voltage command;

[0156] Perform feedforward decoupling calculation on the basic voltage command to obtain the decoupled voltage component, perform voltage synthesis on the basic voltage command and the decoupled voltage component, and obtain the final voltage command vector of the dq axis through superposition operation;

[0157] The final voltage command vector of the dq axis is subjected to Clarke transformation, and the voltage vector in the α-β coordinate system is obtained through the transformation matrix. The voltage vector in the α-β coordinate system is subjected to sector judgment, and the sector number of the space voltage vector is obtained through amplitude calculation and angle calculation.

[0158] Calculate the adjacent basic vector action time based on the sector number and voltage vector information to obtain the basic vector action time and the zero vector action time;

[0159] The basic vector action time and the zero vector action time are modulated by seven-segment SVPWM, and three-phase six-channel PWM switching signals are generated through timing distribution to obtain control voltage signals.

[0160] Specifically, the current reference value is input into the outer loop of the cascade PI controller, and the outer loop PI controller generates a corresponding voltage command according to the current reference value. and the integral coefficient , calculate the error signal and get the voltage command of d-axis and q-axis. The error signal is defined as the difference between the actual current and the reference current. For example, for the d-axis, the current error is expressed as:

[0161] ;

[0162] After obtaining the error, the outer loop PI controller calculates the d-axis voltage command according to the following formula: :

[0163] ;

[0164] Similarly, the q-axis voltage command The corresponding q-axis current error These voltage instructions As the input of the next inner loop PI control, the voltage is further refined and optimized to obtain a more accurate basic voltage command. For the inner loop PI control, the voltage commands of the d-axis and q-axis are input. and The inner loop PI controller is similar to the outer loop, and also adjusts the input voltage command through proportional and integral operations to improve the response speed and steady-state accuracy of the system. The proportional coefficient and integral coefficient of the inner loop are and The output of the inner loop PI controller is the basic voltage command of the d-axis and q-axis and , the formula is as follows:

[0165]

[0166] in, is the inner loop voltage error, is the current actual voltage value. Through the adjustment of the inner loop PI control, more accurate basic voltage instructions for the d-axis and q-axis are obtained. and The basic voltage command is fed forward and decoupled to eliminate the coupling between the d-axis and the q-axis, so that the current control of the d-axis and the q-axis can be performed independently. In the synchronous rotating coordinate system, there are cross-coupling terms between the d-axis and the q-axis, which are mainly caused by the rotor angular velocity and the mutual inductance. The feedforward decoupling voltage is expressed as:

[0167] ;

[0168] ;

[0169] in, and are the decoupled voltage components of the d-axis and q-axis respectively, is the electrical angular velocity, and are the inductances of the d-axis and q-axis respectively, is the magnetic flux generated by the permanent magnet. The basic voltage command and the decoupled voltage component are superimposed to obtain the final voltage command vector on the dq axis:

[0170] ;

[0171] ;

[0172] At this time, the final voltage command vectors of the d-axis and q-axis are obtained , and then transform it to the α-β coordinate system for use in space vector pulse width modulation (SVPWM) control. Perform Clarke transformation on the dq axis voltage command vector and transform it to the α-β coordinate system. The formula for Clarke transformation is:

[0173] ;

[0174] in, is the Clarke transformation matrix, usually in the form of:

[0175] ;

[0176] What you get and It is the voltage component in the α-β coordinate system. In order to perform SVPWM control, the voltage vector in the α-β coordinate system is judged by sector, and the sector number where the voltage vector is located is determined by amplitude calculation and angle calculation. The SVPWM algorithm divides the entire α-β plane into six sectors, each sector corresponds to three adjacent voltage vectors, and the amplitude of the voltage vector is calculated. and angle , determine which sector it is located in. According to the sector number where the voltage vector is located and the amplitude and angle of the voltage vector, calculate the action time of the adjacent basic vectors. The calculation formula for the basic vector action time is as follows:

[0177] ;

[0178] ;

[0179] in, and are the action time of two adjacent basic vectors, is the PWM cycle time, is the DC bus voltage, is the angle between the voltage vector and the reference axis. The action time of the zero vector Then:

[0180] ;

[0181] Based on basic vector action time and zero vector action time , seven-segment SVPWM modulation is performed on the motor. The seven-segment SVPWM modulation method generates a synthetic voltage vector by controlling the switching signals of the three-phase bridge arm, so that the output voltage is as close to the required reference voltage as possible. The seven-segment modulation method includes a combination of two basic vectors and a zero vector. These vectors are allocated in a specific order within a PWM cycle to achieve the best voltage synthesis effect. The generated three-phase six-way PWM switching signal is expressed as:

[0182] ;

[0183] in, They are the switching signals of the three-phase bridge arms, which control the switching state of the inverter and finally obtain the required control voltage signal. The function of the PWM signal is to control the conduction time of each phase so that the average output voltage is equal to the required voltage command, thereby achieving precise control of the motor.

[0184] In a specific embodiment, the process of executing step S600 may specifically include the following steps:

[0185] Calculate the performance index of the control voltage signal to obtain the performance index, which includes: speed dynamic response time, speed regulation accuracy, torque fluctuation rate and system efficiency;

[0186] Based on the performance index, the probability density function of the parameter system is constructed to obtain the log-likelihood function, and the second-order partial derivative of the log-likelihood function is calculated to obtain the Fisher information matrix;

[0187] Based on the Fisher information matrix, matrix inversion operation is performed to obtain the covariance matrix of parameter estimation, and the covariance matrix is ​​decomposed by eigenvalue to obtain the parameter sensitivity eigenvector;

[0188] The importance of parameters is sorted according to the characteristic vector to obtain a parameter optimization sequence, and the control parameters are gradient optimized based on the parameter optimization sequence to obtain optimized control parameters;

[0189] Perform stability analysis on the optimized control parameters to obtain the target control parameters.

[0190] Specifically, the performance indicators of the control voltage signal are calculated to comprehensively evaluate the dynamic and static performance of the motor under different operating conditions to ensure that the optimized control parameters can effectively improve the overall performance of the motor. Performance indicators include speed dynamic response time, speed regulation accuracy, torque fluctuation rate and system efficiency. The speed dynamic response time measures the time required for the motor to reach the target speed from a stationary or certain speed state, reflecting the reaction speed of the system; the speed regulation accuracy measures the error between the actual speed and the target speed of the motor during stable operation, reflecting the steady-state accuracy of the system; the torque fluctuation rate describes the fluctuation of the output torque of the motor during load changes, and a smaller fluctuation rate means smoother operation; and the system efficiency is the ratio between input power and output power, reflecting the energy utilization of the system. Construct the probability density function of the parameter system based on the performance indicators. Statistical modeling is performed on the performance indicators to describe the probability distribution characteristics under different conditions. Assuming that these performance indicators are random variables, the corresponding probability density functions are obtained by fitting the historical operating data of these indicators, which are recorded as ,in is a set of control parameters, is the observed performance index data. The log-likelihood function is constructed based on the probability density function to analyze the parameter characteristics of the system. The log-likelihood function is obtained by taking the logarithm of the probability density function, and its form is as follows:

[0191] ;

[0192] in, represents the log-likelihood function, It is Observation data, is the total number of observations, and the log-likelihood function is used to measure the control parameter The degree of fit between the observed performance indicator data. In order to analyze the sensitivity of the parameters, the second-order partial derivative of the log-likelihood function is solved to obtain the Fisher information matrix. The Fisher information matrix is ​​defined as the expectation of the second-order partial derivative of the log-likelihood function, and its formula is as follows:

[0193] ;

[0194] in, is the Fisher information matrix, Represents mathematical expectation operation. Fisher information matrix is ​​used to measure the degree of uncertainty of parameters in the model. The elements in the information matrix represent the degree of influence of different parameters on the system performance indicators. The larger the value, the greater the influence of the corresponding parameter on the system performance. The Fisher information matrix is ​​inverted to obtain the covariance matrix of parameter estimation. The covariance matrix describes the correlation between different parameters and the size of the estimation error. Its formula is:

[0195] ;

[0196] in, is the covariance matrix. By inverting the Fisher information matrix, we can get the confidence interval of each parameter and the covariance relationship between different parameters. The diagonal elements of the covariance matrix represent the variance of each parameter, while the non-diagonal elements represent the covariance between different parameters. Perform eigenvalue decomposition on the covariance matrix to obtain the sensitivity eigenvector of the parameter. The formula for eigenvalue decomposition is as follows:

[0197] ;

[0198] in, is the eigenvector matrix, It is an eigenvalue diagonal matrix. Each eigenvector in the eigenvector matrix represents the sensitivity in a specific direction, and the eigenvalue represents the variance size in that direction. Through eigenvalue decomposition, we can identify which parameters have the greatest impact on the system and which parameters are less sensitive to the performance indicators. Based on the obtained eigenvectors, the importance of the parameters is sorted to determine the priority of parameter optimization. The importance of the parameters is sorted according to the size of the eigenvalue. The larger the eigenvalue, the higher the sensitivity of the parameter to the system and the higher the optimization priority. The parameter optimization sequence is determined by arranging the eigenvalues ​​from large to small. Based on the parameter optimization sequence, the control parameters are adjusted using the gradient optimization method to find the optimal control settings. The process of gradient optimization is to calculate the gradient of the composite objective function with respect to each parameter and adjust the parameter value along the direction of the gradient. The iterative formula of gradient optimization is as follows:

[0199] ;

[0200] in, Indicates The parameter value after iterations, is the learning rate, which controls the step size of parameter updates, is the gradient of the composite objective function with respect to the parameters. Through multiple iterations, the gradient optimization finally finds the optimal control parameters that minimize the composite objective function. Stability analysis is performed on the optimized control parameters to ensure that they can maintain the stability of the system under different operating conditions. Through eigenvalue analysis or simulation testing, it is verified whether the optimized parameters will cause unstable behavior under dynamic loads and changing conditions. In order to perform stability analysis, a state space model is constructed and the Jacobian matrix of the system is calculated , whose eigenvalues ​​are used to determine the stability of the system. If the real parts of all eigenvalues ​​are negative, the system is asymptotically stable, otherwise the parameters need to be further adjusted.

[0201] See also Figure 2 , Figure 2 A schematic block diagram of the structure of a hybrid excitation synchronous motor control strategy optimization device 200 provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, the hybrid excitation synchronous motor control strategy optimization device 200 includes:

[0202] A modeling module 210 is used to perform mathematical modeling on the permanent magnet flux, excitation current and rotor parameters of the hybrid excitation synchronous motor to obtain a state observer equation;

[0203] The processing module 220 is used to collect current parameters and position parameters based on the state observer equation, and perform filtering processing to obtain motor operation state parameters;

[0204] Prediction module 230, used for inputting motor operation state parameters into a five-layer deep neural network to predict current and speed parameters, and obtaining optimal current parameters and speed prediction parameters;

[0205] A construction module 240 is used to construct a multi-objective function for the optimal current parameter and the speed prediction parameter to obtain a current reference value;

[0206] The regulating module 250 is used to input the current reference value into the cascade PI controller, perform voltage regulation through feedforward decoupling control and space vector pulse width modulation technology, and obtain a control voltage signal;

[0207] The optimization module 260 is used to perform performance calculation and control parameter optimization on the control voltage signal to obtain target control parameters.

[0208] Through the collaborative cooperation of the above-mentioned components, by establishing a complete mathematical model and state observer equation, combined with deep neural network prediction, the control system has self-learning ability, effectively overcoming the impact of parameter changes; a multi-objective optimization method is adopted, and the weight coefficient is determined by the two-stage weighted least squares method, so as to achieve coordinated optimization of multiple control objectives such as copper loss minimization and torque fluctuation suppression; the Fisher information matrix and Cramér-Rao lower bound theory are introduced for performance evaluation, which provides a theoretical basis for control parameter optimization and gets rid of the limitations of traditional empirical tuning; a voltage regulation scheme based on feedforward decoupling control and space vector PWM is designed to reduce the coupling degree of the control system and improve the dynamic response performance; through the combination of deep neural networks and traditional control methods, an adaptive intelligent control system is constructed, which enhances the robustness and generalization ability of the control system; a hierarchical and progressive control strategy optimization method is adopted, and the logical relationship between each control link is clear, which improves the control accuracy of the hybrid excitation synchronous motor.

[0209] See also Figure 3 , Figure 3 A schematic block diagram of the structure of a hybrid excitation synchronous motor control strategy optimization device 300 provided in an embodiment of the present application, wherein the hybrid excitation synchronous motor control strategy optimization device 300 includes a processor 301 and a memory 302, wherein the processor 301 and the memory 302 are connected via a device bus 303, wherein the memory 302 may include a non-volatile storage medium and an internal memory.

[0210] The non-volatile storage medium can store a computer program. The computer program includes program instructions, and when the program instructions are executed by the processor 301, the processor 301 can execute any of the above hybrid excitation synchronous motor control strategy optimization methods.

[0211] The processor 301 is used to provide computing and control capabilities to support the operation of the entire hybrid excitation synchronous motor control strategy optimization device 300 .

[0212] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor 301, the processor 301 can execute any of the above-mentioned hybrid excitation synchronous motor control strategy optimization methods.

[0213] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the scheme of the present application, and does not constitute a limitation on the hybrid excitation synchronous motor control strategy optimization device 300 involved in the scheme of the present application. The specific hybrid excitation synchronous motor control strategy optimization device 300 may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0214] It should be understood that the processor 301 may be a central processing unit (CPU), and the processor 301 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0215] It should be noted that, technical personnel in the relevant field can clearly understand that, for the convenience and simplicity of description, the specific working process of the hybrid excitation synchronous motor control strategy optimization device 300 described above can refer to the corresponding process of the aforementioned hybrid excitation synchronous motor control strategy optimization method, and will not be repeated here.

[0216] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by one or more processors, the one or more processors implement the hybrid excitation synchronous motor control strategy optimization method provided in the embodiment of the present application.

[0217] The computer-readable storage medium may be an internal storage unit of the hybrid excitation synchronous motor control strategy optimization device 300 of the aforementioned embodiment, such as a hard disk or memory of the hybrid excitation synchronous motor control strategy optimization device 300. The computer-readable storage medium may also be an external storage device of the hybrid excitation synchronous motor control strategy optimization device 300, such as a plug-in hard disk, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped with the hybrid excitation synchronous motor control strategy optimization device 300.

[0218] 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.

[0219] 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 application is essentially or the part that contributes to the prior art or all 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 execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0220] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. 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 application.

Claims

1. A hybrid excitation synchronous motor control strategy optimization method, characterized in that: include: The permanent magnet flux, excitation current and rotor parameters of the hybrid excitation synchronous motor are mathematically modeled to obtain the state observer equation; Based on the state observer equation, current parameters and position parameters are collected, and filtering is performed to obtain motor operation state parameters; specifically, the following steps are performed: the three-phase stator current is collected by a Hall current sensor to obtain original three-phase current data, and the excitation winding current is collected by a Hall current sensor to obtain original excitation current data; the rotor position is collected by a magnetic encoder to obtain original position data, and the original three-phase current data and the original excitation current data are subjected to fourth-order Butterworth low-pass filtering to obtain filtered current data; the filtered three-phase current data is subjected to Clark-Park coordinate transformation to obtain dq-axis current components, and the M / T method speed calculation is performed based on the original position data to obtain mechanical angular velocity; the dq-axis current components, the excitation current and the mechanical angular velocity are substituted into the state observer equation to obtain a state observation vector; the observation error is calculated according to the state observation vector, and the observer gain matrix is ​​updated by the recursive least squares method to obtain the motor operation state parameters; Input the motor running state parameters into a five-layer deep neural network to predict current and speed parameters, and obtain optimal current parameters and speed prediction parameters; A multi-objective function is constructed for the optimal current parameter and the speed prediction parameter to obtain a current reference value; The current reference value is input into a cascade PI controller, and voltage regulation is performed through feedforward decoupling control and space vector pulse width modulation technology to obtain a control voltage signal; The control voltage signal is subjected to performance calculation and control parameter optimization to obtain target control parameters.

2. The hybrid excitation synchronous motor control strategy optimization method according to claim 1 is characterized in that: The state observer equation is obtained by mathematically modeling the permanent magnet flux, excitation current and rotor parameters of the hybrid excitation synchronous motor, including: The stator voltage of the hybrid excitation synchronous motor is decomposed into d-q axes to obtain the d-axis voltage equation u d =R s i d +dψ d / dt-ωψ q and the q-axis voltage equation u q =R s i q +dψ q / dt+ωψ d , where u d is the d-axis stator voltage, u q is the q-axis stator voltage, R s is the stator winding resistance, i d is the d-axis stator current, i q is the q-axis stator current, ψ d is the d-axis magnetic flux, ψ q is the q-axis magnetic flux, ω is the electrical angular velocity, and t is the time variable; The stator flux is calculated and the d-axis flux equation ψ is obtained. d =L d i d +M f i f +ψ f and the q-axis flux equation ψ q =L q i q , where L d is the d-axis stator self-inductance, L q is the q-axis stator self-inductance, M f is the mutual inductance between the stator and the field winding, i f is the field winding current, ψ f is the basic flux linkage generated by the permanent magnet; Electromagnetic torque is calculated based on the d-axis flux equation and the q-axis flux equation to obtain the torque equation T e =1.5p[(L d -L q )i d i q +(M f i f +ψ f )i q ], where T e is the electromagnetic torque, p is the number of motor pole pairs; The mechanical motion of the rotor is mathematically described, and the mechanical motion equation J(dω m / dt)=T e -T L -Bω m , where J is the rotor moment of inertia, ω m is the mechanical angular velocity and ω m =ω / p,T L is the external load torque, B is the viscous friction coefficient of the rotor bearing; According to the d-axis voltage equation and the q-axis voltage equation, the current dynamic response analysis is performed to obtain the current differential equation di d / dt=(u d -R s i d +ωL q i q ) / L d and q / dt=(u q -R s i q -ω(L d i d +M f i f +ψ f )) / L q ; The current differential equation and the mechanical motion equation are combined to obtain the state space equation dx / dt=Ax+Bu, where x=[i d ,i q ,ω m ] T is the state vector, u=[u d ,u q ,i f ] T is the control input vector, A is a 3×3 state matrix, and B is a 3×3 input matrix; The state space equation is discretized and numerically solved by the fourth-order Runge-Kutta method to obtain a discrete state equation x(k+1)=x(k)+Δt·f(x(k),u(k)), where k is the discrete time point number, k+1 is the next discrete time point number, Δt is the sampling time interval, and f(·) is the nonlinear function on the right side of the state equation; The state observer equation is established based on the discrete state equation in is the state observation vector is the d-axis current observation value, is the q-axis current observation value, is the mechanical angular velocity observation value, y(k) is the actual measured output vector [i d ,i q ,ω m ] T , Output vector for observation L is a 3×3 order observer gain matrix.

3. The hybrid excitation synchronous motor control strategy optimization method according to claim 1, characterized in that: The step of inputting the motor running state parameters into a five-layer deep neural network to predict current and speed parameters to obtain optimal current parameters and speed prediction parameters includes: Standardizing the motor operating state parameters to obtain a standardized input vector; Input the standardized input vector into the first hidden layer of a five-layer deep neural network, where 16 neurons are set in the first hidden layer, and obtain the first layer feature vector by calculating the first weight matrix and the first bias vector; Inputting the first layer feature vector into the second hidden layer of the five-layer deep neural network, the second hidden layer is provided with 32 neurons, and calculating the second layer feature vector through the second weight matrix and the second bias vector; Input the second layer feature vector into the third hidden layer of the five-layer deep neural network, the third hidden layer is provided with 64 neurons, and the third layer feature vector is calculated by a third weight matrix and a third bias vector; Input the third layer feature vector h3 into the fourth hidden layer of the five-layer deep neural network, the fourth hidden layer is provided with 32 neurons, and the fourth layer feature vector is calculated by a fourth weight matrix and a fourth bias vector; Inputting the fourth layer feature vector into the fifth layer of the five-layer deep neural network, the fifth layer is divided into a current prediction unit and a speed prediction unit, the current prediction unit includes 3 neurons, obtaining a standardized current prediction value, and performing de-standardization processing on the standardized current prediction value to obtain an optimal current parameter; The fourth layer feature vector is input into the speed prediction unit, which includes 1 neuron, and a linear activation function is used to perform output mapping and denormalization to obtain a speed prediction parameter.

4. The hybrid excitation synchronous motor control strategy optimization method according to claim 3 is characterized in that: The constructing of a multi-objective function for the optimal current parameter and the speed prediction parameter to obtain a current reference value includes: Decomposing the objective function based on the optimal current parameter and the speed prediction parameter to obtain a copper loss function, a torque-current ratio function and a torque fluctuation function; A multi-objective function is constructed based on the copper loss function, the torque-current ratio function and the torque fluctuation function to obtain a composite objective function; Performing sensitivity calculation on the operation history data sequence of the hybrid excitation synchronous motor to obtain a sensitivity matrix of each objective function relative to the control target; Based on the sensitivity matrix, solving the first-stage weight equation by least square method to obtain an initial weight coefficient, and normalizing the initial weight coefficient to obtain a first-stage weight coefficient; Substituting the weight coefficient of the first stage into the composite objective function, constructing the weighted least squares equation of the second stage, and obtaining the final weight coefficient by solving it; The final weight coefficient is substituted into the composite objective function, and optimization calculation is performed through gradient descent iteration to obtain optimized current parameters, and operation constraint processing is performed on the optimized current parameters to obtain current reference values.

5. The hybrid excitation synchronous motor control strategy optimization method according to claim 4, characterized in that: The current reference value is input into the cascade PI controller, and voltage is regulated by feedforward decoupling control and space vector pulse width modulation technology to obtain a control voltage signal, including: Input the current reference value into the outer loop of the cascade PI controller, calculate through the outer loop proportional coefficient and the outer loop integral coefficient, and obtain the d-axis voltage command and the q-axis voltage command; Performing inner loop PI control on the d-axis voltage command and the q-axis voltage command, adjusting them through an inner loop proportional coefficient and an inner loop integral coefficient, to obtain a basic voltage command; Performing feedforward decoupling calculation on the basic voltage command to obtain a decoupled voltage component, performing voltage synthesis on the basic voltage command and the decoupled voltage component, and obtaining a dq axis final voltage command vector through superposition operation; The dq axis final voltage command vector is subjected to Clarke transformation, and a voltage vector in an α-β coordinate system is obtained through a transformation matrix, and a sector judgment is performed on the voltage vector in the α-β coordinate system, and a sector number of the space voltage vector is obtained through amplitude calculation and angle calculation; Calculate the adjacent basic vector action time based on the sector number and the voltage vector information to obtain the basic vector action time and the zero vector action time; The basic vector action time and the zero vector action time are modulated by seven-segment SVPWM, and three-phase six-channel PWM switch signals are generated through timing distribution to obtain a control voltage signal.

6. The hybrid excitation synchronous motor control strategy optimization method according to claim 5, characterized in that: The step of performing performance calculation and control parameter optimization on the control voltage signal to obtain target control parameters includes: Calculating the performance index of the control voltage signal to obtain the performance index, wherein the performance index includes: speed dynamic response time, speed regulation accuracy, torque fluctuation rate and system efficiency; Constructing a parameter system probability density function based on the performance index to obtain a log-likelihood function, and calculating a second-order partial derivative of the log-likelihood function to obtain a Fisher information matrix; Performing a matrix inversion operation based on the Fisher information matrix to obtain a covariance matrix for parameter estimation, and performing eigenvalue decomposition on the covariance matrix to obtain a parameter sensitivity eigenvector; Sorting the importance of parameters according to the characteristic vectors to obtain a parameter optimization sequence, and performing gradient optimization on the control parameters based on the parameter optimization sequence to obtain optimized control parameters; A stability analysis is performed on the optimized control parameters to obtain target control parameters.

7. A hybrid excitation synchronous motor control strategy optimization device, characterized in that: A method for optimizing a hybrid excitation synchronous motor control strategy according to any one of claims 1 to 6, comprising: A modeling module is used to mathematically model the permanent magnet flux, excitation current and rotor parameters of the hybrid excitation synchronous motor to obtain the state observer equation; A processing module, used for collecting current parameters and position parameters based on the state observer equation, and performing filtering processing to obtain motor operation state parameters; A prediction module, used for inputting the motor running state parameters into a five-layer deep neural network to predict current and speed parameters, and obtain optimal current parameters and speed prediction parameters; A construction module, used for constructing a multi-objective function for the optimal current parameter and the speed prediction parameter, and solving to obtain a current reference value; A regulating module, used for inputting the current reference value into a cascade PI controller, performing voltage regulation through feedforward decoupling control and space vector pulse width modulation technology, and obtaining a control voltage signal; The optimization module is used to perform performance calculation and control parameter optimization on the control voltage signal to obtain target control parameters.

8. A hybrid excitation synchronous motor control strategy optimization device, characterized in that: The hybrid excitation synchronous motor control strategy optimization device comprises: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instructions in the memory to enable the hybrid excitation synchronous motor control strategy optimization device to execute the hybrid excitation synchronous motor control strategy optimization method according to any one of claims 1 to 6.

9. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the hybrid excitation synchronous motor control strategy optimization method according to any one of claims 1 to 6 is implemented.

Images