Large Initial Population and Neighborhood Search incorporated in LSHADE to solve CEC2020 Benchmark Problems

The single objective multimodal bound-constrained optimization problems in CEC (IEEE Congress on Evolutionary Computation) competitions pose tremendous challenges to the researchers in finding the global optimum. This paper introduces the orthogonal array-based initialization of population and neighborhood search strategy in LSHADE (linear population reduction technique in success history based adaptive differential evolution) algorithm to solve the CEC2020 competition problems. In the novel algorithm, named as O-LSHADE, a large number of population members have been initialized in the search space using the orthogonal design. Thereafter, a small Euclidean neighborhood is defined for each member in the population. The adaptive mutation and crossover are performed within the neighborhood during first phase, i.e., predominantly exploration phase of the algorithm. The process ensures that the search space is adequately explored by the algorithm as the neighborhoods are not mixed and each neighborhood performs the mutation and crossover operations independently. At a later stage of the algorithm, a step reduction in population is done to perform the operations like an adaptive differential evolution. The proposed algorithm is found efficient, effective and robust when applied to the CEC2020 benchmark problems for single objective bound constrained optimization. In most cases, the algorithm could achieve the best error value of less than 100. This implies that the method could guide the search process towards the global basin.