Research On Quantile-Guided Hybrid Of Strategies For Dynamic Multi-Objective Optimization

In the real world,common dynamic multi-objective optimization problems such as portfolio,production scheduling and dynamic path planning need to find the best solution under the uncertain or dynamic factors.To better solve these problems,dynamic multi-objective optimization algorithms are proposed by researchers.Using historical information to find the position of Pareto-optimal front in the new environment is an effective way to solve these problems.Through the quantile,historical information is fully employed to search the Pareto-optimal front when environmental change happens.The main contents of research are as follows:(1)Quantile-guided dual prediction strategy: Most dynamic multi-objective evolutionary algorithms solve the DMOPs solely from the perspective of decision space or objective space,which will also lead to underutilization of population information.As a result,the speed of the convergence of population will decrease.Therefore,a novel quantile-guided dual prediction strategy for dynamic multi-objective optimization(NQDPEA)is proposed in this paper.In NQDPEA,evolution of population is guided by quantiles,the location of quantiles in a new environment is predicted by historical quantile information.Then,the new solution set is expanded according to the position of the new quantile.Its prediction strategy not only predicts the Pareto-optimal set by quantile in the decision space,but also predicts the Pareto-optimal front by historical quantile in the objective space,and then mapping back to the decision space.A large number of experimental results on different types of test functions show that NQDPEA has better performance than contrast algorithms and makes full use of the population information in these two spaces.(2)Quantile-guided multi-strategy algorithm: Most dynamic multi-objective optimization problems have certain patterns of historical environmental changes,thus it is necessary to find these patterns to accelerate the population convergence.Therefore,this paper proposes a quantile-guided multi-strategy algorithm(QMA)for dynamic multi-objective optimization.Due to the robustness of quantiles to outliers,quantiles are used to express the characteristics of data.In QMA,the historical quantile information of the decision space is used to find the new position of the quantile.Then expand the solution set based on the new quantile.In addition,by selecting the mapping matrix as retrieval judgment,the historical search information is precisely used to assist population evolution.Quantile-guided intensity of environmental change detector is employed to determine the number of retained individuals and the number of randomly generated individuals when there is no sufficient information to retrieve.Experimental results on various dynamic multi-objective optimization problems show that,compared with the five state-of-the-art comparison algorithms,QMA has excellent performance,which demonstrates that it can effectively utilize the patterns of historical environment change to accelerate the convergence of population.