Grid Search for Operator and Parameter Control in Differential Evolution

Evolutionary Algorithms constitute a very active research branch of Computational Intelligence. Typically, such algorithms are used for the detection of (sub-) optimal solutions in difficult optimization problems. Numerous works have provided experimental evidence of the remarkable efficiency of Evolutionary Algorithms. However, their performance has proved to be strongly connected to their proper parametrization. Various approaches have been proposed for (offline) tuning and (online) control of their parameters. Recently, a grid-based technique was proposed for parameter adaptation during the algorithm's run without user intervention, and it was validated on the Differential Evolution algorithm, which is widely known for its parameter sensitivity. Experimental results on high-dimensional test problems verified the effectiveness of the technique on controlling the scalar parameters and crossover type of the algorithm. The present work extends that study by considering another crucial component of the algorithm, namely the mutation operator type. Extensive experiments enrich and verify the previous evidence, suggesting that grid-based search can maintain competitive performance while absolving the user from the laborious parameter-tuning phase.