Dynamic Multi-Objective Evolutionary Algorithm Based On Prediction Strategy

In industrial applications and scientific research,such as job shop scheduling,combinatorial optimization,power dispatching,and investment management,decision makers often encounter dynamic multi-objective optiization problems(DMOPs)with multiple objectives that vary over time.To address these problems,dynamic multi-objective evolutionary algorithms(DMOEAs)have been widely applied.However,existing prediction-based DMOEAs have limitations in balancing convergence and diversity,and prediction methods in existing dynamic optimization techniques are greatly affected by the accuracy of the prediction model.To achieve a balance between the convergence and diversity of the algorithm and to improve prediction accuracy,this paper studies and improves the prediction-based dynamic multi-objective evolutionary algorithm.The main contributions are as follows:(1)To address the limitations of existing prediction-based DMOEAs in balancing convergence and diversity,this paper proposes a DMOEA based on particle swarm prediction strategy and prediction adjustment strategy.The particle swarm prediction strategy initializes the particle swarm position using historical information and iterates multiple times in the new environment to form a predicted population for fast convergence.In addition,this paper further designs an individual supplementary mechanism to increase diversity.The prediction adjustment strategy selects individuals with lower dominance levels and more evenly distributed in the population before the environmental changes to form the initial population in the new environment.Experimental results show that the proposed algorithm performs better than existing algorithms on most test functions and achieves a balance between convergence and diversity,especially in testing functions with position-dependent variable switching features,but its performance needs to be improved in complex three-objective problems.(2)In order to address the issue of insufficient prediction accuracy of existing dynamic multi-objective evolutionary algorithms based on prediction,this paper proposes DMOEA based on fine prediction strategy and nondominated solution-guided evolution.After the environmental changes,the fine prediction strategy implements different prediction strategies for different individuals based on the dominance relationship.For dominated individuals in the population,the opposite-based learning prediction is used,while for nondominated individuals,the differential prediction is used.In addition,this paper designs a static strategy for nondominated solution-guided evolution,which can be applied to various MOEA frameworks to improve static optimization performance.Experimental results show that the proposed algorithm performs better than the compared algorithms on most test problems and can solve complex test problems such as three-objective problems well.In summary,the two prediction-based dynamic multi-objective evolutionary algorithms proposed in this paper can efficiently solve dynamic multi-objective optimization problems by balancing convergence,diversity,and improving prediction accuracy.