Evolutionary Many-objective Optimization Algorithm Based On Population Decomposition And Its Application

When the number of the objectives that the optimization problem contains is two or three, the classical optimization algorithm will have a good effect. But in dealing with many-objective optimization problems that contain five or more than five objectives, the effect of these algorithms is not ideal. Main reasons are:1) The growth of the objective’s dimension will lead to the exponential growth of the non-dominated solutions in the population, which makes the algorithm cannot choose good solutions from the population, thus affecting the convergence performance of the algorithm.2) Taking the distribution into consideration, most of the distribution maintaining strategies usually prefer some extreme individuals. This characteristic will influence searching ability of the algorithm in the many-objective optimization problem, which affects the distribution performance of the algorithm.In order to make up for the convergence performance and distribution performance of the traditional multi-objective optimization algorithm when dealing with the many-objective optimization problem, this paper puts forward a evolutionary many-objective optimization algorithm(KD-MOEA) based on the improved K-dominated sorting and population decomposition. The algorithm combined with the knowledge of the population decomposition, decomposes the whole area, thus decomposes the population, which is conducive to improve the distribution of the algorithm on the whole, at the same time reduces the amount of calculation to a large extent; We improved K-dominance relationship and sorting method to avoid the phenomenon of circular dominance. In contrast with the Pareto dominance the improved K-dominance greatly increased the selection pressure of the algorithm; Using the new density estimation methods, improves the accuracy of local density estimation. The contrast experiments of proposed algorithm and NSGA-II are conducted by optimizing the DTLZ series test functions. The simulation results show that the proposed algorithm has obvious advantages, which can not only improves the population convergence, but also ensure a good distribution along the Pareto-optimal Front.After verifying the validity of the proposed algorithm, we will apply the new algorithm to the practical problem-car side impact optimization problem. When using the new algorithm, considering the particularity of optimization problem, We will add constraint handling strategy-penalty function, so that it is better to solve this optimization problem. From the simulation results, we can see that the solution obtained by the new algorithm have the smaller weight and the higher score of safety, and the safety parameter values of these solutions have larger space from the upper bound constraints, namely the solution has good applicability. The solutions with above properties can provide a good reference for actual production design, so that the new algorithm can effectively solve the optimization problem of vehicle side impact.