Font Size: a A A

Research On Dynamic Multi-objective Evolutionary Algorithm Based On Predictive Strategy

Posted on:2022-06-26Degree:MasterType:Thesis
Country:ChinaCandidate:X X LiFull Text:PDF
GTID:2518306536490754Subject:Control Science and Engineering
Abstract/Summary:PDF Full Text Request
In real life,a large number of multi-objective optimization problems,such as logistics distribution,vehicle path planning,portfolio optimization and so on,need to be solved under the constantly changing environment.These optimization problems are called dynamic multi-objective optimization problems.Due to the existence of dynamic or uncertain factors,traditional multi-objective optimization algorithms are hard to optimize these problems.The dynamic multi-objective optimization algorithm based on prediction,which can capture the law of dynamic change according to the historical environmental information,are effective algorithms to respond to environmental change.Therefore,on the basic of the deep study on prediction strategy,two algorithms based on prediction are developed.The main contents of this paper being outlined as follows:(1)Aiming at the problems of long training period and low accuracy of prediction model,a dual prediction strategy with inverse model is proposed.Considering that the dimension of the objective space is relatively low,the algorithm uses the information of population center from the previous two moments to directly predict the individuals in the objective space,which can not only improve the running speed of the algorithm,but also make the generated population evenly distributed along the real Pareto front.The inverse model is established to map the population from the objective space back to the decision space,which guides the search toward desired regions.In order to minimize the error of the population mapping from the target space to the decision space and improve the prediction accuracy of the algorithm,the inverse model is adjusted adaptively with the help of the historical mapping law.Based on the simulation of FDA,dMOP and Fun test problems,the proposed algorithm has good convergence and distribution on different dynamic multi-objective optimization problems.(2)Focusing on the weakness of slow convergence speed of the algorithms in solving complex nonlinear dynamic optimization problems,a hybrid prediction strategy based on decision variable classification is developed.Depending on the impact of the environmental change on each dimension of the decision space,the decision variables are divided into three types,that is,similar,micro and macro variables.The maintenance strategy is used to maintain the optimal value of the similar variables to improve the algorithm's convergence speed.Different prediction methods are used to estimate the new locations of micro and macro variables,thereby increasing the diversity of the population.To enable the population adapt to the changing environment quickly,the adjustment strategy that makes full use of new environment information to guide population evolution.To verify the effectiveness of the algorithm,F and DF benchmark instances are selected for simulation research.The experimental results show that the proposed algorithm achieves good exploration and tracking ability,and shows better adaptability in complex or dramatically changing dynamic environment.
Keywords/Search Tags:Dynamic multiobjective optimization, Prediction, Inverse model, Decision variable classification, Adjust strategy
PDF Full Text Request
Related items