CN-122001239-A - Control method of multilevel inverter, motor driving system, and readable storage medium

CN122001239ACN 122001239 ACN122001239 ACN 122001239ACN-122001239-A

Abstract

The application discloses a control method of a multi-level inverter, a motor driving system and a readable storage medium, wherein the method obtains a system state of the multi-level inverter at k time, approximates nonlinear dynamics of the inverter system through a neural network according to the system state, predicts the system state according to a nonlinear item prediction value by combining a prediction model of the inverter system, obtains a current prediction value of k+1 corresponding to different inverter switching states, obtains a current prediction value of k+2 corresponding to the current prediction value of k+1 through a system control law according to the current prediction value of k+1, evaluates all possible inverter switching states by using a cost function in combination with current reference values at k+1 time, and selects an optimal switching state with the minimum cost function to be applied to the multi-level inverter. The method realizes high-precision control of the multi-level inverter without depending on an accurate model.

Inventors

MA JIEN
GUO ZEPU
FANG YOUTONG
LIU XING
ZHANG ZEYU
QIU LIN
ZHAO PENGBO

Assignees

浙江大学

Dates

Publication Date: 20260508
Application Date: 20260408

Claims (10)

1. A control method of a multilevel inverter, the control method comprising: acquiring a system state of an inverter system at k time of the multi-level inverter, wherein the system state comprises a load current at k time of the inverter system; According to the system state, approaching the nonlinear dynamics of the inverter system through a neural network, and obtaining a nonlinear item predicted value; According to the nonlinear item predicted value, carrying out system state prediction by combining a predicted model of the inverter system to obtain current predicted values at the moment k+1 corresponding to different inverter switch states; according to the current predicted value at the corresponding k+1 moment under the different inverter switching states, obtaining the current predicted value at the corresponding k+2 moment through a system control law; Based on the current predicted values of the corresponding k+1 moment and the corresponding k+2 moment in the different inverter switching states, in combination with the current reference value of the k+1 moment, evaluating all possible inverter switching states by using a cost function, wherein the cost function is constructed based on a scalar function, and the scalar function is derived based on a stability theory and reflects the current error convergence trend of an inverter system; an optimal switching state that minimizes the cost function is selected and applied to the multilevel inverter.
2. The method for controlling a multilevel inverter according to claim 1, The neural network comprises an input layer, a hidden layer, an output layer and a context layer, The input layer is used for receiving the system state, and outputting the nonlinear item predicted value at the output layer through the transformation of the hidden layer to the system state; the context layer is used for recording the activation state of the hidden layer of the previous time step, and updating the current activation state of the hidden layer according to the recorded activation state of the hidden layer.
3. The control method of a multilevel inverter according to claim 2, wherein the neural network approximates the nonlinear dynamics of the inverter system by the following formula: ; in the formula, Representing the predicted value of the non-linear term, 、、 The weight is represented by a weight that, 、 The offset is indicated as being a function of the offset, Indicating the activation function of the hidden layer, Representing an input vector consisting of the current system state and a reference state of the inverter system, Indicating the hidden layer activation status of the context layer record.
4. A control method of a multilevel inverter according to claim 3, characterized in that the predictive model of the inverter system adopts a super local predictive model in the form of an integrator.
5. The control method of a multilevel inverter according to claim 4, wherein the current prediction value at the corresponding k+1 time under different inverter switching states is obtained by the following formula; ; in the formula, The predicted value of the current at time k is indicated, A nonlinear term prediction value representing the k moment; To control the input gain; Is a positive constant; representing a current estimation error; Representing voltage vectors corresponding to different switch states.
6. The control method of a multilevel inverter according to claim 1, characterized in that the control method further comprises constructing the cost function by: Acquiring an error state reflecting a deviation between the predicted current and the reference current; Performing time differentiation on the error state to obtain an error change rate; constructing a lyapunov function for evaluating the stability of the inverter system based on the sum of squares of the error states; and differentiating the Lyapunov function to derive the change rate of the Lyapunov function, wherein the change rate of the Lyapunov function directly correlates the product relationship between the error state and the error change rate, and converts the product relationship into the cost function.
7. The method for controlling a multilevel inverter according to claim 6, The cost function satisfies the sum of the products of the current error and the error change rate at the time k+1.
8. A multi-level inverter employing the control method of a multi-level inverter according to any one of claims 1 to 7, the multi-level inverter being any one of the following types of inverters: An active neutral-point clamped inverter, a T-type inverter, a cascaded H-bridge inverter, and a modular multilevel converter.
9. A motor drive system, characterized in that the motor drive system includes a motor and a multi-level inverter that drives the motor, the motor drive system further comprising a controller that performs the control method of the multi-level inverter according to any one of claims 1 to 7 to control the multi-level inverter.
10. A readable storage medium, characterized in that it stores a computer program executable by an electronic device, which when run on the electronic device, causes the electronic device to perform the control method of the multilevel inverter according to any one of claims 1 to 7.

Description

Control method of multilevel inverter, motor driving system, and readable storage medium Technical Field The present application relates to the field of power electronic control, and in particular, to a control method of a multilevel inverter, a motor driving system, and a readable storage medium. Background Along with the continuous improvement of the requirements of industrial application on the electric energy quality and the driving performance, the three-level and multi-level inverter is widely applied to the fields of medium-high voltage and high power occasions such as motor driving, new energy power generation grid connection and the like by virtue of the advantages of low output voltage harmonic wave, small device voltage stress and the like. However, the control performance of the method faces a core challenge that the control performance of the traditional finite control set model predictive control (FCS-MPC) method based on an accurate mathematical model is seriously dependent on the accuracy of system parameters, and in actual operation, uncertain factors such as perturbation of inverter parameters, abrupt load change and the like can lead to model mismatch, so that steady-state error increase, dynamic response deterioration and even system instability are caused. In order to reduce the model dependency, the prior art proposes methods such as model independent predictive control based on an Extended State Observer (ESO), but the approximation capability of complex nonlinear dynamics is limited and lacks theoretical guarantee of stability, or a simple neural network is adopted for compensation, but the dynamic learning capability and the closed-loop stability are still insufficient. Disclosure of Invention The embodiment of the application provides a control method, a motor driving system and a readable storage medium of a multi-level inverter, which can realize high-precision, fast response and robust current tracking control of the multi-level inverter under the condition of not depending on an accurate model. In a first aspect, the present application provides a control method of a multilevel inverter, the control method comprising: acquiring a system state of an inverter system at the k moment of the multi-level inverter, wherein the system state comprises a load current at the k moment of the inverter system; according to the system state, approaching the nonlinear dynamics of the inverter system through a neural network, and obtaining a nonlinear item predicted value; according to the nonlinear item predicted value, carrying out system state prediction by combining with a predicted model of an inverter system to obtain current predicted values at the time k+1 corresponding to different inverter switching states; According to the current predicted value at the corresponding k+1 moment under different inverter switching states, obtaining the current predicted value at the corresponding k+2 moment through a system control law; Based on the current predicted values of the corresponding k+1 moment and the corresponding k+2 moment in different inverter switch states, and combining the current reference value of the k+1 moment, evaluating all possible inverter switch states by using a cost function, wherein the cost function is constructed based on a scalar function, and the scalar function is derived based on a stability theory and reflects the current error convergence trend of an inverter system; An optimal switching state that minimizes the cost function is selected and applied to the multilevel inverter. In one embodiment, the neural network includes an input layer, a hidden layer, an output layer and a context layer, The input layer is used for receiving the system state, and outputting a nonlinear item predicted value at the output layer through the transformation of the hidden layer to the system state; the context layer is used for recording the activation state of the hidden layer in the last time step, and updating the current activation state of the hidden layer according to the recorded activation state of the hidden layer. In one embodiment, the neural network approximates the nonlinear dynamics of the inverter system by the following formula: ; in the formula, Representing the predicted value of the non-linear term,、、The weight is represented by a weight that,、The offset is indicated as being a function of the offset,Indicating the activation function of the hidden layer,Representing an input vector consisting of the current system state and a reference state of the inverter system,Indicating the hidden layer activation status of the context layer record. In one embodiment, the predictive model of the inverter system employs a super local predictive model in the form of an integrator. In one embodiment, the current prediction value at the time k+1 corresponding to the different inverter switching states is obtained by the following formula; ; in the formula, The predicted value of