Research On Preference-inspired Coevolutionaryalgorithms For Many-objective Optimization

Multi-objective optimization problem has always been an important research subject in fields of science and engineering because of its strong ability in global search.Recently,many-objective optimization algorithm has developed rapidly with the complicated actual application problem.However,many-objective optimization algorithm has a general conflict between fast convergence and population diversity.Firstly,with the number of objectives increased,the rate of non-dominated populations increases rapidly,which leads to the ability of searching population weaken and couldn't convergence to Pareto front.Secondly,with the number of objectives increased,the number of solutions required increasing exponentially for approximating the entire Pareto front.Lack of the enough number of solutions may block the algorithm approximate to Pareto front.Thirdly,High computational complexity makes the solutions distributed non-uniformly.The co-evolutionary mechanism emphasizes the interaction between populations and environment,and it involves the role of competition and cooperation between evolutionary populations,so it could not only improve population diversity but also speed up convergence rate.To solve these problems,this paper proposed that the knee point guided population coevolution toward Pareto front as a preference for decision maker based on coevolution mechanism.At the same time,this paper proposed a many-objective optimization via cooperative coevolution with adaptive preferences to decrease the computation complexity.1.This paper proposed a knee point guided many-objective optimization via coevolution with preferences.Firstly the randomly generated preferences were introduced to guide the population and population selects preferences usingcooperative mechanism to realize coevolution toward Pareto front.Further,to improve the decision maker's ability of choosing the region of interest without knowing the preference information,this paper proposes a knee point guided many-objective optimization via coevolution with preferences,which using knee point guided the population to approximate the entire Pareto front.In the algorithm performance evaluation experiments,the proposed algorithm is been used to solve ZDT and DTLZ testing functions,the experimental results verify it's effectiveness and convergence.2.This paper proposed a many-objective optimization via cooperative coevolution with adaptive preferences.Firstly,focusing on how to use multi-preference to guide the solutions,a collaborative evolutionary strategy is built to implement the co-evolutionary of the preferences and the solutions.Secondly,the statistic analysis of the frequency of multi-preference is analysed to acquire the optimal synergy model between multi-preference and solutions.In the algorithm performance evaluation experiments,the proposed algorithm is been used to solve2-10 objectives testing functions,the experimental results verify it's effectiveness and superiority.