Deep learning-accelerated optimization algorithm for controller parameters optimization of doubly-fed induction generators

Abstract

In this work, a cooperative Gray wolf Optimizer with adaptive differential Evolution (GOE) is proposed for the multimodal controller parameters optimization of doubly-fed induction generators (DFIGs) based on maximum power point tracking (MPPT) strategies. Moreover, the optimization process of the GOE is accelerated by a deep fully connected model (DFCM). The GOE contains a cooperative gray wolf optimizer (GWO) and adaptive differential evolution (ADE). The cooperative GWO contains alpha, beta, delta, and omega wolves to explore and exploit optimization problems and achieves optimization tasks wider and deeper than GWO. The ADE cooperates with the cooperative GWO to solve global optimization over continuous spaces. The simulation results on seven uni-model benchmark functions show that the GOE accelerated by DFCM obtains acceptable fitness values with 39.99% lesser computation time than the symmetry adapted stochastic search (SASS) algorithm and 80.72% lesser computation time than the Lévy flights-success-history based adaptive differential evolution with constraint handling technique (COLSHADE) algorithm, which are the winners of the CEC2020 Competition on Real-World Single Objective Constrained Optimization. Furthermore, the simulation results on DFIG with MPPT strategies in three real-world cases verify that the GOE accelerated by DFCM can effectively obtain global optimization solutions for non-smooth problems with 99.51% lesser average computation time than the SASS algorithm, 99.63% less than the COLSHADE algorithm, and 89.52% less than other methods. In addition, the accelerated GOE algorithm by DFCM has the feature of faster convergence.