Deep convolutional generative adversarial network accelerated optimization algorithm for parameter optimization of permanent magnet synchronous generator controllers

Abstract

In permanent magnet synchronous generators (PMSG), optimized rotor-side controller (RSC) parameters improve the power coefficient. Aiming at the traditional intelligent optimization algorithms since the long optimization time and insufficient global search capability, this work proposes adaptive differential evolution variants with linear population size reduction (L-SHADE) for constrained optimization with Levy flights (COLSHADE) accelerated by using deep convolutional generative adversarial network (DCGAN). The DCGAN-COLSHADE converts the parameters of the PMSG controllers into pictures and utilizes the DCGAN alternative algorithmic iterative process to speed up the COLSHADE iterative process and accomplish a broader and deeper global optimization problem. The PMSG simulation results utilizing the maximum power point tracking strategy verify the DCGAN-COLSHADE can obtain globally optimal solutions and higher system stability. The fitness function value of DCGAN-COLSHADE is 3.96 % smaller than the comparison algorithm; the average computation time is 79.28 % less than the particle swarm optimization (PSO), 80.35 % less than the moth flame optimization (MFO), 80.75 % less than the whale optimization algorithm (WOA), 80.52 % less than gray wolf optimization (GWO) and 77.96 % less than COLSHADE. In addition, the results of rapid control prototype (RCP) hardware experiments validate the feasibility and effectiveness of the algorithm.