A Data-Driven Automatic Design Method for Electric Machines Based on Reinforcement Learning and Evolutionary Optimization

The design problems of electric machines are actually treated as a kind mixed-integer problem, because the machine shapes defined by some integer variables, such number slots, and other tooth width, which here called fundamental shape respectively. To automatically solve these problems, this article presents an automatic method combining reinforcement learning evolutionary optimization. In proposed method, decision process is modeled tree structure to seek for determined result search depending on value functions nodes. Then, variables estimated from function variables. These constructed based data, generate optimization employed. As result, can be adapted unexperienced through data generation learning. applied problem linear induction motor. It shown that designs with prescribed performance given specifications obtained. Moreover, it also acceptable candidate immediately generated when specification similar previously-solved utilizing past explorations.