Research On Dynamic Multi-objective Optimization Technology Based On Pivot Individual Prediction

Evolutionary Algorithms(EA)are a class of heuristic random search algorithms based on Darwinian evolutionary theory.The evolutionary algorithm(EA)regards the objective as the individual,simulates the evolution of species through genetic operators such as selection and mutation,and iteratively searches the feasible region to obtain a set of approximate optimal solutions.In real life,we may encounter such a type of problem,which not only has many conflicting objective functions to be optimized,but also these functions change over time.Objective Optimization Problems(DMOPs).The various conditions of dynamic multi-objective optimization problems(DMOPs)may change over time,so solving such problems is very challenging.In order to solve such problems,researchers have designed some strategies or mechanisms of dynamic multi-objective optimization algorithms(DMOEAs)according to the characteristics of such problems.Among them,memory strategy and multi-swarm strategy are common methods used to solve dynamic multi-objective optimization problems(DMOPs)and have achieved certain results.However,in a dynamic environment,as algorithm optimization proceeds,the population may face abrupt changes in Pareto front(PF)/Pareto solution set(PS)or aperiodic Pareto front(PF)/Pareto solution set(PS)At this time,the memory strategy will reduce its performance due to the insufficient utilization of its stored historical effective information,and thus the effective guidance of the population will be reduced.In addition,it may also face a dynamic environment in which the Pareto front(PF)/Pareto solution set(PS)changes drastically.At this time,multiple population strategies may not have enough time to search the feasible region,which will lead to a decrease in population convergence.In addition,there are some existing diversity-based strategies,which pay too much attention to the diversity of the population.When the population faces a relatively complex dynamic environment,it may need a precise evolutionary direction to guide the evolution of the population.The evolutionary direction may lead to stagnation of population optimization.Based on the above considerations,this paper believes that introducing some representative individuals as guide individuals,and using the guide individuals to give other individuals in the population a more accurate evolutionary direction can effectively improve the above problems.Therefore,this paper proposes a Pivot Individual Prediction-based Response Strategy(HPPDS)and its improvement strategy(PGPS).The strategy proposed in this paper is characterized by looking for some representative individuals as guiding individuals,hoping to guide other individuals in the population to respond to the dynamic environment through the traction of guiding individuals,so as to achieve the purpose of solving dynamic multi-objective optimization problems.In this paper,prediction strategies include prediction of pivot individuals(representative individuals selected through a dynamic selection mechanism)and prediction of non-dominant individuals.Then,a diversity preservation strategy is used to generate random search individuals in the decision space to ensure the diversity of the population.To verify that the strategy proposed in this paper is competitive,the proposed algorithm and several other state-of-the-art algorithms are experimentally compared on 12 different types of test problems.Through the analysis of experimental results,it can be known that HPPDS and PGPS have good prospects in solving dynamic multi-objective optimization problems.