A generation-based optimal restart strategy for surrogate-assisted social learning particle swarm optimization

Abstract Evolutionary algorithm provides a powerful tool to the solution of modern complex engineering optimization problems. In general, a great deal of evaluation effort often requires to be made in evolutionary optimization to locate a reasonable optimum. This poses a serious challenge to extend its application to computationally expensive problems. To alleviate this difficulty, surrogate-assisted evolutionary algorithms (SAEAs) have drawn great attention over the past decades. However, in order to ensure the performance of SAEAs, the use of appropriate model management is indispensable. This paper proposes a generation-based optimal restart strategy for a surrogate-assisted social learning particle swarm optimization (SL-PSO). In the proposed method, the SL-PSO restarts every few generations in the global radial-basis-function model landscape, and the best sample points archived in the database are employed to reinitialize the swarm at each restart. Promising individual with the best estimated fitness value is chosen for exact evaluation before each restart of the SL-PSO. The proposed method skillfully integrates the restart strategy, generation-based and individual-based model managements into a whole, whilst those three ingredients coordinate with each other, thus offering a powerful optimizer for the computationally expensive problems. To assess the performance of the proposed method, comprehensive experiments are conducted on a benchmark test suit of dimensions ranging from 10 to 100. Experimental results demonstrate that the proposed method shows superior performance in comparison with four state-of-the-art algorithms in a majority of benchmarks when only a limited computational budget is available.