Dynamic multi-objective evolutionary optimization algorithm based on two-stage prediction strategy

Tracking pareto-optimal set or pareto-optimal front in limited time is an important problem of dynamic multi-objective optimization evolutionary algorithms (DMOEAs). However, the current DMOEAs suffer from some deficiencies. In the early optimization process, the algorithms may suffer from random search. In the late optimization process, the knowledge which can accelerate the convergence rate is not fully utilized. To address the above issue, a DMOEA based on the two-stage prediction strategy (TSPS) is proposed. TSPS divides the optimization progress into two stages. At the first stage, multi-region knee points are selected to capture the pareto-optimal front shape, which can accelerate the convergence and maintaining good diversity at the same time. At the second stage, improved inverse modeling is applied to search the representative individuals, which can improve the diversity of the population and is beneficial to predicting the moving location of the pareto-optimal front. Experimental results on dynamic multi-objective optimization test suites show that TSPS is superior to the other six DMOEAs. In addition, the experimental results also show that the proposed method has the ability to respond quickly to environmental changes.