Research On Dynamic Multi-objective Evolutionary Algorithm Based On Hybrid Response Mechanism

The environment of Dynamic Multi-objective Optimization Problems(DMOPs)usually changes frequently with the evolution process,which is usually expressed as Pareto optimal Set(POS)or Pareto optimal Front(Pareto optimal Front).,POF)change over time.Such problems bring higher challenges to evolutionary algorithms,because DMOPs require evolutionary algorithms to have the ability to continuously track a moving POF at any time.When tracking a moving POF,the speed of convergence is the core performance,since the time interval between two environmental changes can be very short.Therefore,whether the population can quickly and effectively track and converge to the POF in the new environment is the criterion for judging the performance of the dynamic multi-objective evolutionary algorithm.It is well known that the prediction-based response mechanism is a common method for dealing with environmental changes,which meets the requirements of rapid response to environmental changes,but is dedicated to generating individuals within the predicted new POS after the change occurs.This type of method is only suitable for changes that follow some fixed pattern(predictable).Individuals in the population may have converged to the approximate Pareto optimal area one by one.After the environment changes,the population may not find new POS and the diversity is very likely to be insufficient.It can be seen that the imbalance of population diversity and convergence is exacerbated in the process of tracking the dynamically changing POF.For this reason,another great challenge for DMOEAs is to keep the population well diverse,i.e.widely and evenly distributed in the search space.A hybrid prediction strategy and an accuracy-controlled mutation strategy are combined as a new change response mechanism to address DMOPs.Specifically,the hybrid forecasting strategy has the ability to quickly adapt to predictable environmental changes,and coordinate the central point-based forecasting strategy and the guided individual-based forecasting strategy to improve the forecasting accuracy.This hybrid forecasting strategy,while somewhat crude,offers the benefit of more exploratory search than some of the more refined forecasting strategies that use various models to achieve.In addition,the precision-controlled mutation strategy improves the diversity exploration of the population by controlling the mutation precision of the solution to deal with unpredictable changes.In this way,the proposed change response mechanism can adapt to various environmental changes of DMOPs.The proposed new change response mechanism HPPCM is integrated into a multi-objective distribution estimation algorithm to jointly optimize DMOPs.The statistical results of comparison experiments with some excellent dynamic multi-objective evolutionary algorithms on multiple test benchmark problems verify that the proposed change response mechanism is effective and competitive.