Addressing Parameter Variation Of PMSM Drive With Multi Network Policy Based Control For Electric Vehicle Application

Abstract

Since the advancement of DRL algorithms in the last decade, there is an increased focus on their application in several different fields like autonomous driving, health data monitoring, NLP, image processing, etc. Likewise a subset of these algorithms have found application in replacing the traditional control techniques, with improved performance and efficiency. PMSM drives utilized for traction application in modern day electric vehicles require a highly accurate control scheme. The performance of the controller directly contributes to the performance and efficiency of the entire system. The drawback of PMSM drives is the nonlinearity and that its parameters are dynamic in nature. The controller used in traditional control techniques (PI, PID controller) have good control action for system which are linear and time invariant. As mentioned the PMSM does have a parameter variation issue which results due to the change in ambient condition like temperature, humidity, pressure, etc. The control efficiency of traditional controller diminishes with change in parameters of the system. Hence to address these issues we adopt the FOC technique with an actor-critic agent based controller. The actor critic agent is trained with multi network policy based control algorithm. In this paper we compare and discuss the control action of a traditional control vs RL based control.