Feature Information Prediction Algorithm for Dynamic Multi-objective Optimization Problems

Abstract Dynamic multi-objective optimization problems (DMOPs) contain multiple conflicting goals while tracking the changing Pareto-optimal front (PF) or Pareto-optimal set (PS). Most algorithms treat the solutions of DMOPs as if they were dealing with static multi-objective optimization problems. However, solutions under different environments may obey different distributions. To solve some of the existing limitations of currently available methods, a dynamic multi-objective optimization algorithm based on feature information prediction (FIP) is proposed. To identify the distribution of solutions after an environmental change, Joint Distribution Adaptation (JDA) is used to construct a mapping function. The feature information, which is extracted from the objective space at the current time step, is mapped to a higher dimensional space. Then the feature information of decision space at the next time step is obtained using the interior point method. Based on this information, the initial population at the next time step is generated when a change is detected. The performance of FIP is validated by comparing it with respect to four state-of-the-art evolutionary algorithms on eight benchmark functions. Experimental results demonstrate that FIP can quickly cover the front with rapidly changing environments.