Research on ε-dominance Multi-Objective Evolutionary Algorithms in the Application of Optimization Problems
|School||Wuhan University of Technology|
|Course||Applied Computer Technology|
|Keywords||Multi-objective optimization Pareto optimal solution Evolutionary Algorithms ε dominance Super grid|
In real life, most people encounter optimization problem is multi-objective optimization problem, but most of these goals are conflicting. Obtained during optimization solution set is called Pareto optimal solution set. Most of the multi-objective optimization algorithm is used to find approximate Pareto optimal solution set. Algorithm to obtain the solution not only to distribution, but also to try to close Pareto optimal front. However, most of the existing optimization algorithm in this regard, but there are some flaws. For example, some algorithms may spend a lot of time to obtain a uniform distribution of the optimal solution; Some algorithms although the time is shorter, but the optimal solution obtained is not very evenly distributed. In this paper, we propose a fast and stable multi-objective evolutionary algorithm (Ⅰ-MOEA), while improving the current A better algorithm-NSGA-Ⅱ algorithm. Ⅰ-MOEA algorithm is based on Deb. Introduced in 2002 the concept of ε-dominance, the algorithm also applied elitist strategy, archiving update, ultra mesh segmentation method and the G vector control strategies. Elitist strategy is now recognized as an effective strategy for maintaining the diversity of individuals. Archive update strategy into online and offline archiving archiving algorithm applied online archive. This paper proposed super mesh segmentation method and the G vector control strategy is used to solve the problem of diversity of individuals. Mesh segmentation method is divided into a plurality of target space over the grid, while ensuring that each of the super-grid dominated by only one individual non-occupied, this is intended to facilitate ε dominance relationship between the individual comparison may to maintain the diversity of solutions to solve the problems. G vector control strategy is used to select non-dominated elite individuals can ensure the most representative elected elite, which solves the problem of individual diversity. Pareto dominance and ε dominance with selected non-dominated individuals selected representative elite individuals, so that you can reduce the Pareto optimization of regional candidate solutions. Size of the archive does not pre-set value, but according to the tuning parameters λ and ε depending on the product, this can shorten the update time of the archive, and can obtain the desired Pareto optimal solution, thereby reducing the whole algorithm computation time. From the results of simulation experiments, the algorithm has good versatility and efficiency of the algorithm is also high. In the improved algorithm NSGA-Ⅱ (Ⅰ-NASG-Ⅱ algorithm), first NSGA-Ⅱ algorithm analyzes the defects, while maintaining the diversity of solutions that there is a defect. So with SPEA2 in Clustering method instead NSGA-Ⅱ algorithm Crowding approach. From the test results, Ⅰ-NSGA-Ⅱ algorithm to obtain the solution of distribution than NSGA-Ⅱ algorithm to obtain more homogeneous solution; while Ⅰ-MOEA algorithm both in unconstrained two goals, two goals of constraint or three tests on the target to get the optimal solution are ideal. From the simulation results point of view, these two algorithms have achieved good results.