A Data-Driven Automatic Design of Induction Motors Based on Tree Search and Reinforcement Learning Considering Multiple Objectives

The design problems of electric machines are mixed-integer ones, which are generally difficult to solve. To automatically obtain the whole designs of electric machines, in this article, a multi-objective automatic design method for electric machines is presented based on the tree search-based reinforcement learning and multi-fidelity (MF) technique. In this method, various candidate designs are generated based on the tree search using low-fidelity model. The value functions of the nodes are learned from the generated data. Moreover, high-fidelity evaluations are applied to a limited number of better designs, and then, the MF metamodel is constructed to correct the low-fidelity evaluation results. The present method is applied to the design problem of an induction motor. It is shown that various designs are obtained considering tradeoff relation between two objectives based on high-accuracy evaluations.