Surrogate Model Assisted Multi-objective Differential Evolution Algorithm for Performance Optimization at Software Architecture Level

This paper proposes a surrogate model assisted differential evolutionary algorithm for performance optimization at the software architecture (SA) level, which is named SMDE4PO. In SMDE4PO, different strategies of crossover and mutation are adopted to enhance the algorithm’s search capability and speed up its convergence. Random forests are used as surrogate models to reduce the time of performance evaluation (i.e., fitness evaluation). Our comparative experiments on four different sizes of cases between SMDE4PO and NSGA-II are conducted. From the results, we can conclude that (1) SMDE4PO is significantly better than NSGA-II according to the three quality indicators of Contribution, Generation Distance and Hyper Volume; (2) By using random forests as surrogates, the run time of SMDE4PO is reduced by up to 48% in comparison with NSGA-II in our experiments.