On the algorithms of adaptive neural network-based speed control of switched reluctance machines

A switched reluctance machine (SRM) drive is a time-varying, strongly nonlinear system. High performance control can no longer be achieved by using linear techniques. This paper describes the back-propagation (BP) neural network-based proportional-integral-derivative (PID) speed control of the SRM. It's the interest of this paper to explore the utilization of the prior empirical knowledge as guidance in the initializing and training of the neural networks. The purpose is to make the networks less sensitive on the initial weights. Two modified algorithms are presented and simulation experiments show some interesting findings about their control effects and their corresponding sensitivity on the initial weights of the networks.