Rethinking the differential evolution algorithm

Abstract

Selection operation plays a significant role in differential evolution algorithm. A new differential evolution algorithm based on an improved selection process is presented in this work. It was studied that there was neither a practical method to maintain the distribution of population nor a correction to the variables out of bounds in mutation process in a standard differential evolution algorithm. The fast non-dominated sorting approach and the spatial distance algorithm which were applied to the beginning of the selection process, as well as a method to fix the transboundary variables in the mutation process, were adopted to optimize the differential evolution algorithm. The reformative algorithm could obtain a uniformly distributed and effective Pareto-optimal sets when applied to the classical multi-objective test functions; it performed prominently in the experiment of optimizing the quality, the cost and the time in a construction project compared with the previous work.