Study Of Permanent Magnet Synchronous Motor Control Strategy Based On Retina Neural Network

Permanent Magnet Synchronous Motor (PMSM) is widely applied in the industrial control field, which is due to its unique advantages such as small volume, light weight, high efficiency, small torque ripple, big power density, strong overload capacity and fast dynamic response. However, the control system of PMSM is restricted by factors like strong nonlinearity, time-varying and multivariable. Therefore, the control method and strategy for PMSM should be further studied to achieve high accuracy speed control.Artificial Neural Network (ANN) has good ability of self-learning and adapting, it is very suitable to solve the control problem of complex nonlinear systems. For BP network, less parameter require to be determined offline, and its self-learning ability is perfect. Extensive research has been done on the control strategies of PMSM based on BP neural network, and lots of achievements have been made. But the BP neural network has shortcomings such as complex structure, low speed of learning, local minimum problem; this could not meet the real-time control request of system.In view of these problems in motor control field, a type of completely new neural network-Retina Neural Network (RNN) is introduced in the paper. The RNN adjusts its network parameters in the way of non-supervised learning; its structure is simple, and it needn't studying repeatedly, so the convergence rate for learning is fast. In order to explore the real-time performance of RNN, the paper attempts to make some research for the control strategy of PMSM based on retina neural network.First, the control strategy of PMSM based on BP neural network is proposed, after the BP neural network finishes training, the method which combines BP network and PI control is adopted in the speed control system of PMSM, and the simulation model of control system is established. The simulation results are obtained, which provides theoretical basis for the comparative study afterwards.Then, the model and learning rules of RNN is given, and the network is trained through different data samples to prove its fast convergence. Based on this, the RNN controller is designed, and the simulation model of speed control system with RNN for PMSM is established. Meanwhile, the simulation results are obtained.Finally, the simulation results under two control methods are compared; it is not difficult to discover that when using BP-PI method to control the speed of PMSM, the dynamic response is slow, ripple is large, and the error is big; whereas, when using RNN-PI method to achieve the online real-time control of PMSM speed, the speed error is smaller, accuracy is higher, dynamic response is faster, and most importantly the real-time performance is better. Thus, the effectiveness and rationality of the method proposed in this thesis are well conformed. Therefore, for the filed of high control accuracy and real-time performance, this method is greatly valuable for further academic research and has a broad application prospect.