Multi-objective Evolutionary Algorithm For Complex Pareto Front Problem And Its Application

With the development of science,technology and social progress,the real world in a variety of systems increasingly complex and diverse,the traditional mathematical method can not fulfil the requirements of multi-objective optimization problem in our real-world complex system.Evolutionary algorithm is a population evolution using global search method which is based on group intelligence.Evolutionary algorithm is widely used for the advantages of all approximation solutions that can be solved by one run.Recently,MOEA/D,a multi-objective evolutionary algorithm based on decomposition,is superior to many other algorithms in performance.In recent years,the decomposition-based multi-objective evolutionary algorithm(MOEA / D)has been used in many multi-objective evolutionary algorithms.In MOEA/D,a multi-objective optimization problem is decomposed into a series of single-objective optimization problems.However,MOEA / D algorithm in the complex problems still exist on many shortcomings,such as the loss of diversity,slow convergence and so on.The study of complex MOPs is still a difficult problem of multi-objective evolutionary algorithm.Complicated MOPs often includes not only the complexity of the objective space,but also the complexity of decision space.Taking the two kind of complexity aspects into consideration,and based on the in-depth study of MOEA / D algorithm.The main work is as follow:1.Aiming at the improvement of MOEA / D algorithm,an evolutionary multiuser algorithm for complex Pareto front problem is proposed.Firstly,this thesis studies the nonlinear relationship between the weight vector and the demapping vector and the method of the reference point are the reasons why the MOEA / D algorithm solves the complex Pareto Front problem is affected by the PF shape.On the above analysis,it is proposed to use the reciprocal Chebyshev decomposition method to solve the nonlinear relationship between the weight vector and the demapping vector firstly.And what's more,we propose a method of using the utopian point instead of the reference point to weaken the influence of the PF shape on the distribution of the solution and give the calculation method of the utopian point.Finally,the corresponding weight vector mapping strategy and the objective function normalization method are given for the value method of utopian point.In this chapter,the validity of the improved algorithm is verified by a large number of experiments and the recommended range of the utopian coefficients is given by experiments.2.In this part,a multi-objective algorithm for reservoir flood control based on utopia algorithm is proposed for Reservoir flood control operation(RFCO)problems.we first studies Reservoir flood control operation(RFCO)problems and theoretically prove the complexity of the multiple conflicting objectives and the interdependent chain correlation of decision variables.By using the evolutionary multi-objective algorithm for complex Pareto Front problem for solving the RFCO problems,the comparison experiment is used to prove the good diversity and convergence of the algorithm in solving complex problems.3.In the last part,aiming at the complex problem of reservoir scheduling model,a large-scale multi-objective reservoir flood control algorithm is proposed.First,the complexity of the reservoir problem and the schemes to the existing large-scale problem are analyzed.Based on the interdependent correlation between the decision variables in RFCO problems of second part,a large-scale reservoir flood control algorithm based on scheduling cycle transition is proposed.In this algorithm,the variables of the variables between the parent and child populations are transformed with each other,so as to realize the co-evolution between multiple parent and child populations.In this part,we propose a transformation method of "nearest distance interpolation method" to map the decision variables between father and son populations,and realize the co-evolution between father and son populations.Finally,through the co-evolution of multiple parent-child populations,the algorithm obtains a good non-dominated solution and good convergence speed when solving the complex problem with high dimension.