Dynamic Multi-objective Optimization Algorithm Based on RNN Information Accumulation
Dynamic multi-objective optimization problems exist widely in real life.After the environment changes,it is necessary for the evolutionary algorithm to have the abilities of fast convergence,fast tracking Pareto optimal frontier and maintaining di-versity.For severe and frequent environmental changes,the traditional forecasting method can not effectively obtain Pareto opti-mal frontier solution.For this problem,a dynamic multi-objective optimization algorithm based on recurrent neural networks in-formation accumulation(IA-RNN)is proposed.Firstly,a nonlinear prediction method based on RNN information accumulation is proposed,which uses RNN recursion for information accumulation,improves the utilization rate of historical information and en-hances the ability of prediction.Secondly,a linear prediction method based on individual is designed,which uses parameter matrix to predict the linear changes of individual.Linear prediction and RNN nonlinear prediction co-evolve,which can quickly track the Pareto optimal frontier.Finally,a parameter correction strategy based on the least square method is designed to guide the parame-ter correction by the approximate Pareto optimal frontier solution in the current environment,which reduces the influence of error accumulation.IA-RNN is compared with five representative dynamic multi-objective optimization algorithms on 14 DF benchmark problems.Experiments show that the IA-RNN algorithm has better convergence and diversity.
万方数据
维普
