Inverse Model based Prediction for Evolutionary Dynamic Multiobjective Optimization

Dynamic multiobjective optimization (DMO) requires that the optimization algorithm to be able to keep track of the Pareto optimal front (PF) over time and find a series of Pareto optimal set (PS) at different times, then can respond effectively and timely when environmental changes are detected. In this paper, a prediction strategy based on inverse model (IMP) is developed to solve dynamic multiobjective optimization problems (DMOPs). Specifically, the inverse model closely links the decision space and the objective space, which can guide the search for promising decision areas. When a change occurs, the IMP first predicts individuals in the objective space, so that the predicted initial population will be close to the new PF. Secondly, the inverse model is established to map the population from the objective space back to the decision space, resulting in the population close enough to PS. To exam the performance of the proposed IMP, eleven benchmark test problems with different types of difficulties are simulated and evaluated. The statistical results indicate that IMP is promising for addressing complex DMOPs.