Study On Evolutionary Dynamic Multi-objective Optimization Algorithm Based On Dynamic Diversity Introduction

Optimization problems are often involved in industrial applications and scientific research.Most of them have multiple objectives need to optimized simultaneously,which are called multi-objective optimization problems.However,there are also many multi-objective optimization problems with time-varying objective functions,parameters or constraints,which are usually called dynamic multi-objective optimization problems.Due to the uncertainty of dynamic multi-objective optimization problems,there are few universal approaches,which make it more difficult to be extended to practical applications.As a kind of heuristic search algorithm,evolutionary algorithm has good universality,implicit parallelism,robustness and so on.It is very befitting for solving and has been successfully applied to multi-objective optimization.However,evolutionary algorithms still face many challenges in solving dynamic multi-objective optimization problems,and there is a big research space.Based on the review of dynamic evolution multi-objective algorithms,this paper proposes new effective algorithms for dynamic multi-objective optimization about change detection,change adaptation and benchmark functions.The main work of this thesis is as follows:1.A dynamic multi-objective optimization algorithm based on multi-objective optimization algorithm with decomposition and inverse model is proposed in the third chapter.The algorithm adopts the improved inverse model as diversity introduction method.Then new individuals generated by inverse models will be inserted in population for adapting the new environment.Too much diversity introduction will lost useful solutions,while too little introduction will lead to premature convergence.In order to determine a appropriate diversity introduction proportion,the algorithm integrates evaluations of environmental changes intensity and feedback information of performance indicator.Compared with four representative algorithms,the proposed algorithm shows better robust search performance in dynamic environment.2.A dynamic multi-objective optimization algorithm based on multi-objective algorithm with decomposition and dynamic diversity introduction strategy is proposed in the fourth chapter.This algorithm has been improved in environmental change detection and adaptation.The new environmental detection mechanism not only judges whether the environment has changed,but also the type(strong or weak change)and the intensity of changes.Faced with guidance information provided by the detection mechanism,inverse model and partial random initialization are adopted for different environmental changes subsequently.Experimental results show that the proposed algorithm has obvious advantages over other representative multi-objective optimization algorithms.3.In the current research,most dynamic multi-objective optimization problems change in a certain type in evolutionary optimization.In order to construct dynamic multi-objective optimization problems with complex environmental changes,it is proposed that make different combinations based on the given parameters change rate and change frequency.In addition,the dynamic multi-objective optimization algorithm based on dynamic diversity introduction strategy is adopted to solve these dynamic multi-objective optimization problems.On the one hand,the dynamic tracking performance of the algorithm has been verified.On the other hand,this work makes an attempt to extend the dynamic multi-objective optimization to more complicated practical problems.