Electromagnetic Design Optimization of a PMSG Using a Deep Neural Network Approach

Abstract

This article presents the electromagnetic design optimization of a permanent magnet synchronous generator (PMSG) based on machine learning (ML). First, the optimization methodology is presented; then, a correlation and a sensitivity analysis are carried out to determine the set of design variables. The optimization goal is maximizing efficiency, which is equivalent to minimizing electrical PMSG losses. It also considers the core and copper materials by minimizing their weight. A deep neural network (DNN) architecture is developed and trained using PMSG 2D-FE data. The DNN is based on the nonlinear rectified linear unit (ReLU). The resulting DNN was later used to construct the PMSG objective function, which was then solved using non-sorting genetic algorithms. Numerical results and comparisons between two genetic algorithms are given to demonstrate the validity of the proposed approach.