A Multilayer Perception Trained Method in Speed Control of a Linear Switched Reluctance Motor

Abstract

Switched reluctance motors are nonlinear systems with some uncertainties and unmodeled dynamics. Propulsion force and speed in these motors have inherently high fluctuations complicating their applications. The conventional controllers could not offer a precise performance for nonlinear systems because they require analytical calculations of the partial derivatives. Accordingly, in this article, a multilayer perception is presented to overcome this problem and control a linear motor. Training algorithms require a complete dataset of the system output, which complicates their implementation. To solve this problem, a Kalman filter is used to estimate uncertain parameters. Thus, the proposed control system does not require a complete dataset of the system. It can process data and predict the next values in a short time without complete observations. The proposed control strategy is implemented to a linear switched reluctance motor and the results are compared with two other conventional methods via simulation and experimental tests. The results confirm the ability and accuracy of the proposed method.