A multi-strategy random weighted gray wolf optimizer-based multi-layer perceptron model for short-term wind speed forecasting

Abstract

Gray wolf optimizer (GWO) that is one of the meta-heuristic optimization algorithms is principally based on the hunting method and social hierarchy of the gray wolves in the nature. This paper presents the Multi-strategy Random weighted Gray Wolf Optimizer (MsRwGWO) including some effective and novel mechanisms added to the original GWO algorithm to improve the search performance. These are a transition mechanism for updating the parameter (a) over right arrow, a weighted updating mechanism, a mutation operator, a boundary checking mechanism, a greedy selection mechanism, and an updating mechanism of leader three wolves (alpha, beta, and delta wolves). We utilized some benchmark functions known as CEC 2014 test suite to evaluate the performance of MsRwGWO algorithm in this study. Firstly, during the solution of optimization problems, the MsRwGWO algorithm's behaviors such as convergence, search history, trajectory, and average distance were analyzed. Secondly, the comparison statistical results of MsRwGWO and GWO algorithms were presented for CEC 2014 benchmarks with 10, 30, and 50 dimensions. In addition, some of the popular meta-heuristic algorithms taken from the literature were compared with the proposed MsRwGWO algorithm for 30D CEC 2014 test problems. Finally, MsRwGWO algorithm was adapted to the training process of a Multi-Layer Perceptron (MLP) used in wind speed estimation and comparative results with GWO-based MLP were obtained. The statistical results of the benchmark problems and training performance of MLP model for short-term wind speed forecasting show that the proposed MsRwGWO algorithm has better performance than GWO algorithm. Source code of MsRwGWO is publicly available at https://github.com/uguryuzgec/MsRwGWO.