A Research On Multi-objective Optimization Based On Evolutionary Mechanism

Evolutionary algorithm based group heuristic search is a kind of random search method. It simulates the natural selection and natural evolution in biological process of the evolution. Its versatility makes it suitable for complex nonlinear and low dimensional problems. However, with the requirement of solutions' performance increasing and the complexity of problems enhancing, the solutions obtained are required to close to the real problem solution.Based on the above background, this paper proposes an improved differential evolution-based hybrid evolutionary algorithm for solving multi-objective optimization problem of low dimension. In addition, based on the study of high-dimensional multi-objective optimization problem, we propose a correlative selection mechanism and differential evolution-based evolutionary algorithm for many-objective optimization. The specific work mainly is as follows:1. An improved hybrid evolutionary multi-objective optimization algorithm is proposed. Firstly, the typical representative evolutionary algorithm for solving multi-objective optimization problem--NSGA-? algorithm is given. Then, its disadvantage in not well search accuracy and blind target space is in-depth studied. The proposed algorithm adopts the technology of Latin Hypercube Sampling to ensure that the distribution of initial population can be uniform. And it uses differential evolution operator to replace crossover operator in NSGA-? to enhance the local search ability and search accuracy, while retaining the mutation operator of NSGA-? to improve diversity of population. The test results of ZDT1, ZDT2, ZDT3 and ZDT4 multi-objective optimization functions show that the proposed algorithm has great advantage on convergence and it also performs better than the compared algorithms of NSDE, NSGA-? and MODE on diversity.2. A correlative selection mechanism and differential evolution-based evolutionary algorithm for many-objective optimization is proposed. Firstly, the relative concepts abount reference point and individual in correlative selection mechanism are given. Then, the correlation-based differential evolution and polynomial mutation selection mechanism is proposed, and it maintains the diversity of the population by the number of reference point's relative individuals. Finally, the correlative selection of population update mechanism is proposed, and it uses the penalty function of individual to ensure the convergence of population. Through the IGD indicator test results of DTLZ1, DTLZ2, DTLZ3 and DTLZ4 having 3,8 and 15-dimensional objectives, the overall performance of the proposed algorithm is better than NSGA-? and MOEA/D algorithm.