A First-Order Difference Model-Based Evolutionary Dynamic Multiobjective Optimization

This paper presents a novel algorithm to solve dynamic multiobjective optimization problems. In dynamic multiobjective optimization problems, multiple objective functions and/or constraints may change over time, which requires a multiobjective optimization algorithm to track the moving Pareto-optimal solutions and/or Pareto-optimal front. A first-order difference model is designed to predict the new locations of a certain number of Pareto-optimal solutions based on the previous locations when an environmental change is detected. In addition, a part of old Pareto-optimal solutions are retained to the new population. The prediction model is incorporated into a multiobjective evolutionary algorithm based on decomposition to solve the dynamic multiobjective optimization problems. In such a way, the changed POS or POF can be tracked more quickly. The proposed algorithm is tested on a number of typical benchmark problems with different dynamic characteristics and difficulties. Experimental results show that the proposed algorithm performs competitively when addressing dynamic multiobjective optimization problems in comparisons with the other state-of-the-art algorithms.