Research On Torque Estimation Generated By Induction Motor Based On Neural Network

Vector control is one of the most widely used induction motor control technologies,which greatly improves the accuracy and response speed of torque control.Traditional vector control is strongly dependent on motor parameters,so it often needs additional parameter estimation.In recent years,the intelligent algorithm represented by neural network has been widely used in PID control,parameter estimation,torque estimation and so on.Because neural network has strong expression ability and good robustness.In this paper,a torque estimation method is proposed by combining the neural network with the flux observation method,and its effectiveness is verified by experiment and simulation.1.A iron loss model with high precision is introduced to improve the accuracy of the algorithm to estimate torque;the influence of field orientation error on torque estimation is studied,and the given and response sequence of the system is introduced as the input of neural network to improve the dynamic accuracy and robustness of torque estimation algorithm.2.A torque estimation algorithm without flux linkage observation based on endto-end mapping and recurrent neural network(RNN)structure is proposed.The problem of difficult to convergence of long chain RNN is solved by using sliding window structure and gated cyclic unit(GRU).3.A training process of combining pre-training simulation data and fine-tuning experimental data is proposed to improve the quality of data set and efficiency of training;a step-by-step algorithm to calculate feedforward neural network based on flux observation is designed to reduce the calculation time in the interruption period.4.The simulation and experimental verification platform based on TI kit(TMS320F28335),computer(tensorflow1.9)and induction motor is built.The algorithm is verified by simulation and experiment.