Font Size: a A A

Research On Dynamic Multi-objective Evolutionary Algorithms Based On Prediction And Diversity Maintenance Strategy

Posted on:2019-09-07Degree:MasterType:Thesis
Country:ChinaCandidate:G RuanFull Text:PDF
GTID:2428330548481924Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
Evolutionary Algorithm(EA)is a heuristic global optimization Algorithm based on the theory of Darwinian evolution.It takes the population as the information carrier,and simulates the evolution of biology in nature through natural selection and heredity,so as to search the whole solution space in an iterative way.In practical engineering applications,there is a kind of problem whose objective functions conflict with each other and objective function or its parameters may change over time,this kind of problem is called dynamic multi-objective optimization problems,DMOPs.To resolve DMOPs,scholars in this field have proposed a series of related techniques and theoretical methods of dynamic multi-objective evolutionary algorithm(DMOEAs).Existing DMOEAs mainly include random initialization of population,diversity maintaining mechanism and variation method as well as multiple population strategy,prediction and memory mechanism.However,these algorithms also have corresponding disadvantages,which are mainly reflected in the following aspects.First of all,in order to increase the diversity of population,the introduction of the super mutation and random initialization and dynamic migration in enhancing diversity random blindness,do not bring in the course of the evolution of species proper guidance.Then,the prediction methods which use some forecasting model of population information to study and forecast the whole population of next environmental change,although can achieve very good effect,the accuracy of prediction is an important difficulties to achieve this method,which needs to design a proper prediction model for the existing problems.Additionally,the existing prediction models always have high time complexity,so improving the efficiency of prediction model of prediction is top priority.In this paper,a hybrid diversity retention strategy based on simple prediction model is proposed to solve the dynamic multi-objective optimization problem.This method consists of three steps:the first step is to use the prediction strategy based on the direction of the central point to redistribute some individuals to the new Pareto surface after the next environmental change.On the basis of the self-defined lowest and highest points of the POS in the article,the second step applies gradual search strategy to produce some good individuals in the decision space,so as to improve the accuracy of the prediction strategy in the first step.In the third step,some well-diversified individuals are randomly generated in the next POS area to enhance the diversity of the population.In the end,individuals generated by the three steps of are held together and non-dominated sort will be conducted to select some good individuals as the next initialization population of the optimization.Thus,the prediction strategies become more accurate because the strategy chooses some individuals with good convergence and diversity.In experimental comparison,the algorithm put forward in this article are compared with three other state-of-the-art dynamic multi-objective evolutionary algorithm in a series of dynamic multi-objective benchmark problems.The experimental results prove that this algorithm show significant competitiveness in convergence and diversity as well as the speed of responding to environmental changes.
Keywords/Search Tags:evolutionary multi-objective optimization, evolutionary dynamic multi-objective optimization, response mechanism, prediction strategy, gradual search strategy, diversity maintenance mechanism
PDF Full Text Request
Related items