Research On Dynamic Robust Evolutionary Optimization Method

Many real-world optimization problems are often influenced by dynamic factors such as production conditions and operating environment,so as to form dynamic optimization problems.The common method solving this kind of problem is to trace the moving optima.When it detects the environment change,the optimization process is re-triggered to find the optimal solution adapted to the new optimization model quickly and accurately.Tracking optimal solution method can effectively solve the dynamic optimization problem,however,when the dynamic optimization problems have complex objective function or large search space,the evolutionary solution process is often difficult to obtain optimal solutions in finite time because of time consuming.In addition,in some practical dynamic optimization problems,when the dynamic factors change,implementing the new optimal solution need to adjust a large number of related personnel or resources,and result in a larger switching costs.Based on this,this dissertation presents a dynamic robust evolutionary optimization method for solving dynamic optimization problems.The core idea is to solve the dynamic optimization problems in the continuous change environment,and find a set of robust optimal solution sequences over time that satisfy the user's requirements.When the environment changes,according to the user's acceptability,the robust solutions in previous adjacent environment are directly used as the optimal solution in the current environment,rather than re-finding the optimal solution in the new environment.This method can effectively reduce the optimization cost in the new environment,and meet the needs of limited resources deployment in actual production.For dynamic single-objective optimization problems,there are two kinds of robustness indexes in the dynamic robust optimization method,such as the survival time and the average fitness.On this basis,a two-stage multi-objective evolutionary optimization model with two kinds of robust performance indexes is constructed.The robust optimal solution sequence is obtained by using non-dominated sorting genetic algorithm.In the first stage,multi-objective evolutionary optimization method is used to obtain the Pareto solutions in each dynamic environment with the above two aspects of performance;In the second stage,the evolutionary individuals are combined by the Pareto solution obtained from the first stage according to the environmental changes.Considering the average survival time and average fitness of the solution sequence,a multi-objective evolutionary optimization method is used to obtain the practical dynamic optimal solution set,and apply it to solve the improved moving peak problem.For dynamic multi-objective optimization problems,the concept of multiobjective robustness on time scale is given,the robust Pareto optimal solution set over time is defined,and two performance measures: time robustness and performance robustness are given.Thus building a dynamic robust multi-objective optimization model,then the multi-objective evolutionary algorithm based on decomposition is used to solve its robust Pareto optimal solution set.Furthermore,taking into account both the performance robustness and the time robustness of the individual,the dynamic robust multi-objective constrained optimization model is constructed for the dynamic robust multi-objective optimization problem.In the robustness evaluation of the individual,it is necessary to consider fitness of the Pareto solution in the current and future adjacent dynamic environment.In order to effectively estimate the fitness value of an individual in the future environment,the time series based on the historical information are established.Using the moving average forecasting,autoregressive prediction and nearest neighbor prediction,the ensemble prediction model is constructed.The ultimate goal of multi-objective optimization is to find the solution meeting the decision makers' demands from the Pareto solution set.In order to improve the evolutionary efficiency,it is not necessary to obtain all the Pareto optimal solutions in each dynamic environment,and only the optimization process needs to be concentrated in the region where the decision makers are interested.Thus,the decision maker's preference information is incorporated into the search process to guide the population toward the preference region.In addition,in the robust performance evaluation,the decision maker's preference information is transformed into the target stability thresholds,which are used to guide robust individual selection.Using the above preference information,a robust multi-objective evolutionary optimization method based on decomposition is used to obtain a robust optimal solution set that satisfies the decision maker's preference.By using the traditional tracking optimal solution method or dynamic robust optimization method,there are some limitations in solving the dynamic multi-objective optimization problem with complex environmental changes.Therefore,according to the dynamic environment history information,the environmental change sequences will be constructed to predict the future environmental changes.Furthermore,according to the environmental change information,the decision criterion is established to determine what kind of dynamic evolution optimization method is used in the optimization process.Finally,a hybrid dynamic multi-objective optimization framework incorporating two dynamic evolutionary optimization methods is established.In view of the dynamic optimization problems,this dissertation proposes the concept of robust Pareto optimal solution set over time,and constructs the dynamic multiobjective robust evolution optimization framework,and gives the corresponding evolutionary optimization methods.The research results not only enrich the dynamic evolutionary optimization theory,but also provide a new idea and more effective mehtods for the actual dynamic optimization problem,therefore,have important theoretical and practical values.