A new hybrid prediction model with entropy-like kernel function for dynamic multi-objective optimization

Dynamic multi-objective problems (DMOPs) permeate all aspects of daily life and practical applications. As the variables of the search space or target space alter in pace with time, savants are also deepening the research on DMOPs, among which methods based on prediction mechanisms have been extensively developed. The historical optimal solutions can effectively predict the trend and location of the optimal solutions in the future. In this paper, a new hybrid prediction model (HPM) integrating the fuzzy linear prediction model with entropy-like kernel function and the one-step prediction model is developed to sort out DMOPs. In the method, the predicted center by the HPM prediction model is combined with the approximate manifold of PS to generate a trail population, and the linear one-step prediction model is utilized to generate another trail population. When the environment changes, the initial PS at the next moment is obtained by randomly crossing these two trail populations. To assess the proposed HPM model, it is compared with the reinitialization strategy, feedforward prediction strategy, population prediction strategy, T-S nonlinear regression strategy with multistep prediction and individual-based transfer learning under different MOEA optimizers for 22 benchmark problems. The results indicate that HPM has great advantages in solving these dynamic optimization problems.

Automated summary

Dynamic multi-objective problems are prevalent in daily life and practical applications. This paper proposes a new hybrid prediction model (HPM) to solve these problems, and the results show that HPM outperforms other strategies in dynamic optimization.