Probability-based Region Prediction For Dynamic Multi-objective Optimization Algorithm

Many complex dynamic multi-objective optimization problems occur in many situations in the real-world,such as in production scheduling,combinatorial optimization,engineering design,large-scale data processing and many other areas.The objective functions of these problems are not only related to the decision variables,but also affected by environmental change,so it is very difficult to design a common method for all problems.In recent years,various methods h ave been proposed to address these problems and one of the most promising methods is the prediction based method,especially,the point-to-point prediction method.However,this method neglects the properties of dynamic multi-objective optimization problems,such as the relative structure information between Pareto optimal solutions and the probability distribution information of these solutions.The current Pareto optimal solutions become increasingly scattered when compared to the history of Pareto optimal solutions,this method is not able to achieve the expected prediction effect.Currently,dynamic multi-objective optimization algorithm includes three parts: environmental detection,multi-objective optimization,and change reaction.When the environment is detected changes,according to the characteristics of dynamic multi-objective optimization problems,this paper proposes a probability-based region prediction strategy for dynamic multi-objective optimization problem.Considering the relative structure information between Pareto optimal solutions,utilize the historical Pareto optimal solutions for K-mediods clustering to generate several subsets.Considering the probability distribution information of Pareto optimal solutions,then employ the quantum probability distribution model to generate "probability-based regions".Inspired b y quantum evolutionary algorithm,this paper puts forward a "probability-based region" prediction strategy which is able to predict "probability-based regions" of next time quickly and efficiently.At the same time,this paper proposes a competition mechanism,by which the new population is able to close to the true optimal solutions of new environment.Finally,some contrast experiments are conducted,the statistic results manifest that the proposed algorithm is capable of tracking the Pareto front over di fferent environment efficiently and effectively.Finally,it carries on the full text summary,explains the deficiencies of this paper,and looks forward to the future research work of the development.