An Improved Grey Prediction Evolution Algorithm for Real-world Engineering Optimization Problems

Abstract

Grey predictive evolutionary algorithm (GPE), which is developed with the inspiration of the grey prediction theory, is a promising candidate for solving complex optimization problems. Due to the complexity and challenge of constraint search space of real-world engineering problems, an improved GPE algorithm (called GPEde) incorporated the constraint handling methods of an ε-constrained level and a gradient-based repair is proposed to effectively handle real-world engineering constrained problems in this paper. The core idea of GPEde is to guide inferior individuals through predicted information produced by GPE. The proposed algorithm is developed by the following two processes. Firstly, half of offspring individuals carrying the predicted information are generated by GPE using three consecutive generation subpopulation, where the subpopulation consists of the top 50% excellent individuals in each generation. Secondly, another half of offspring individuals are produced by adopting a prediction-based learning strategy that is constructed by using the new 50% of offspring individuals to guide inferior individuals of the last 50% of the current population. The effectiveness and superiority of GPEde are validated on 19 mechanical engineering problems extracted from the CEC 2020 real-world single-objective constrained optimization problems. The experimental results demonstrate that GPEde is a competitive optimizer for solving constrained optimization problems compared with differential evolution algorithm, GPE, three variants of GPE and four state-of-the-art algorithms. If this paper is accepted, MATLAB codes associated with this paper will be uploaded to https://github.com/Zhongbo-Hu/Prediction-Evolutionary-Algorithm-HOMEPAGE.