Individual-based self-learning prediction method for dynamic multi-objective optimization

This paper proposes an individual-based self-learning prediction method for dynamic multi-objective optimization problems, called ISPM, to effectively track the time-varying Pareto-optimal set (POS) in a dynamic environment. The ISPM adjusts the reference points by individual-based self-learning, which differs from existing approaches based on fixed reference points. The self-learning reference points are given according to the information from the previous population to divide the Pareto-optimal front (POF) into the objective space as uniformly as possible. One of the ISPM’s advantages is it can improve the influence of the corresponding non-uniform POF for the population’s prediction. Furthermore, it is known that each reference point and the original point can form a vector in the objective space. Each vector can present a subregion in the objective space. The moving direction of each subregion in the last two environments is the reference when improving the population in the new environment to realize local search. Meanwhile, we roughly calculate the change degree of the environment using the self-learning reference point sets at the last two environments to realize the global search. The comprehensive experimental results show that the proposed algorithm can effectively balance convergence and diversity compared with other state-of-the-art methods.