Position Sensorless Control For Switched Reluctance Motor Based On Support Vector Machine

Switched reluctance motor (SRM) has the advantages of simple construction, high reliability, high efficiency and low cost. Switched reluctance motor drive system (SRD) which is a new sort of AC drive systems possesses excellent speed regulation performance and has aroused a widespread attention in the world. SRD is a position closed loop system. The rotor position information is needed by the controller to ensure the corresponding winding is switched on or off at appropriate time. Conventional way of obtaining rotor position signal is to detect directly by position sensor. The existence of position sensor not only brings on structure complication, but also weakens the reliability of the system at high speed. Then, searching for new methods without position sensors to detect rotor position has become the focus of SRD research all over the world.Based on the analysis of nonlinear model of SRM, a dynamic simulation model of the SRD is established in MATLAB/Simulink environment. This paper presents a novel approach of rotor position estimation for switched reluctance motor based on support vector machine (SVM). For the nonlinear property of SRM, this approach takes advantage of SVM with better solution for small-sample learning problem and well generalization property. Through the off-line learning, a better support vector machine structure is formed to realize an efficient nonlinear mapping among phase current, phase flux linkage and the corresponding rotor position, and then it facilitates the rotor position estimation. The simulation results show that correct rotor position estimation can be achieved with this method.On the base of work mentioned above, a method of position sensorless control for switched reluctance motor based on support vector machine associated with RBF neural network is proposed. After the support vector machine structure is obtained and according to the equivalence of support vector machine and neural networks, the configuration and the parameter of RBF neural network can be obtained effectively by conversion. This approach combines the advantages of SVM with better solution for small-sample learning problem and RBF neural network that can modify the parameter on-line. Through the off-line learning associated with on-line learning, the learned network is purposed to achieve the sensorless control of SRM. The simulation results show that this method can obtain the correct commutation signal, and thus the sensorless control of SRM is realized.Finally, the hardware of SRM experiment system based on DSP TMS320F2812 as the core of controller is introduced, and it facilitates further research.