Research Based On Preference Information And Prediction Mechanism For Dynamic Multi-objective Evolutionary Optimization Algorithm

Evolutionary Algorithm(EAs)is a kind of self-organizing and self-adapting heuristic algorithm.It is drawing lessons from the evolutionary ideas of biological competition in natural,choosing individuals which can better fit the environment,through the improvement of each generation,the individuals which participate in evolution process will approaching the optimal solution set of the problem step by step.The two features of EA: group search strategy and information exchange between individuals make EAs has high robustness and wide applicability compared with traditional optimization algorithms,and has the ability to effectively and quickly deal with complex real-life problems which difficult to solve with traditional mathematical methods.EAs have been diffusely adapted in industrial applications and scientific research to solve practical problems such as traffic route planning,satellite system layout optimization,and workshop scheduling.Dynamic multi-objective optimization problems(DMOPs)is a branch of multi-objective optimization problems(MOPs)whose objective functions changes over time.Many optimization problems in real life have the characteristics of DMOPs.Due to the dynamic change of the Pareto-optimal set(POS)with time,to track the continuous POSs,an effective algorithm must not only have the ability to track the optimal solution,but also need to quickly detect environmental changes and make respond.Because of lacking the ability to adapt to changing environments,it is difficult for traditional multi-objective optimization algorithms to solve DMOPs well.Most existing dynamic multi-objective evolutionary algorithms(DMOEAs)anchor in three strategies: introduce of population diversity,prediction mechanism for population evolution direction,and retention of historical information.In most real-life problems,the solutions(Region Of Interest,ROI)that are of interest to decision makers(DMs)often account for only a small part of POS.Most existing DMOEAs focus on estimate the complete POS,which obviously will cause great waste of computational resources and also increases the selection pressure of DMs.Hence,in this paper we present a novel dynamic algorithm based on the decision maker's preference information,involving an improved r-dominance relation and a response strategy.It focuses on searching ROI according to the DM's preference information in dynamic multi-objective optimization.The improved r-dominance relation adapts the angle to measure the closeness between the solution and the preference information,which solves the convergence problem of the r-dominance relation since the original r-dominance relation has difficulty converging to the true Pareto-optimal front(POF)when the preference information is located in the feasible objective region.The prediction mechanism is based on the movement of the population's special points;it helps the population make adjustments in its moving direction and moving step size towards the new POF when a change is detected.From the experimental results we can find that the proposed algorithm can properly solve DMOPs,and it is competitive compared to state-of-the-art methods.