A Modified Differential Evolution Algorithm Based on Improving A New Mutation Strategy and Self-Adaptation Crossover

The differential evolution algorithm is one of the promising natural inspired population-based metaheuristic algorithms that attracted the attention of researchers in the recent years. This pa-per presents a new mutation strategy called DE/current-to-best/2 that presents a new mutated vector based on utilizing the distance between the best vector and the current vector along with another random vector. In addition, the crossover procedure is self-adapted to cover low locality and high locality based on the iteration number. To obtain the best results of the proposed mod-ified differential evolution algorithm, design of experiments is done to optimize its parameters. The comparative results are done using 11 optimization problems to compare the classical ver-sion of differential evolution algorithm with the new modified version and the results show high efficiency of the proposed DE algorithm in terms of CPU time, evaluation, and accuracy The outline of the work done in this paper can be shown as follows:& BULL; The paper produces a new modification of one of the most promising metaheuristics algo-rithms, the differential evolution algorithm. & BULL; The mutation strategy of the algorithm is modified to work with the current solution, the global best solution, and a random solution. The resulted mutated vector from this procedure is used to produce a new modified crossover solution. & BULL; The crossover procedure is self-adapted to cover low locality and high locality based on the iteration number, where in case of the odd iterations, the high locality is applied to obtain more diversity, and in case of the even iterations the low locality is applied to obtain local neighbor solutions. The comparison is done with the classical version of the algorithm, and the results show efficiency in terms of CPU time, evaluation, and accuracy.