Multi-objective integrated optimization of electrical machines using Monte Carlo tree search

Abstract

This paper introduces a novel optimization method for the integrated design of electrical machines. The proposed method can simultaneously optimize the discrete variables that define the machine configurations and the continuous variables describing the detailed shape of parts. The discrete variables are represented by a tree structure and the continuous variables are determined at the leaf node using Monte Carlo Tree Search (MCTS). Multi-objective optimization is performed in conjunction with MCTS to obtain Pareto solutions composed of different machine structures and shapes. The obtained Pareto solutions are stored in the tree and evaluated to update the search strategy to effectively select the combinations of the discrete variables. This method is applied to the optimization of permanent magnet (PM) motors. It is shown that this method is effective for the multi-objective integrated optimization of the PM motors.