Multivariable grey prediction evolution algorithm: A new metaheuristic

Abstract

The theoretical foundation of the grey prediction system, proposed by Deng J. in 1982, is built on the fact that appropriate conversion can transform unordered data to series data with an approximate exponential law under certain conditions. Inspired by the grey prediction theory, this paper introduces a novel evolutionary algorithm based on the multivariable grey prediction model MGM(1,n), called MGPEA. The proposed MGPEA considers the population series of an evolutionary algorithm as a time series. It first transforms the population data to series data with an approximate exponential law and then forecasts its next population using MGM(1,n). Philosophically, MGPEA implements the optimizing process by forecasting the development trend of the genetic information chain of a population sequence. The performance of MGPEA is validated on CEC2005 benchmark functions, CEC2014 benchmark functions and a test suite composed of five engineering constrained design problems. The comparative experiments show the effectiveness and superiority of MGPEA.The proposed MGPEA could be regard as a case of constructing metaheuristics by using the grey prediction model. It is hoped that this design idea leads to more metaheuristics inspired by other prediction models.