Study Of Multi-Objective Evolutionary Optimization Algorithm Based On Double Selection Mechanism And Angle Neighbor Punishment Mechanism

Compared with the single-objective optimization problems (SOPs), there are a large number of non-linear and highly complex engineering problems in real life, these problems are multi-objective optimization problems (MOPs). The objective number of Multi-objective optimization problem is greater than or equal to 2; the objective vectors must be optimized at the same time, moreover, the objective vectors are often conflicting. The evolutionary algorithms mainly solving multi-objective optimization problems are called multi-objective evolutionary algorithms (MOEA). This paper puts forward two algorithms which based on the Pareto dominance relation and the decomposition algorithms.According to the matching selection problems of individuals and sub-problems in advanced MOEA, MOEA/D-DRA, which is based on the aggregation-based MOEA, MOEA/D. this paper analyzes and demonstrates the inherent nature of sub-problems in theory, and puts up with a method to select the best sub-problems. Then it introduces the competitive selection to promote information communication of individuals and sub-problems, and finally proposes the decomposition of multi-objective evolutionary algorithm with double selection (MOEA/D-SS), which employs SPEA2 to acquire a set of uniform weight vectors and finds the best sub-problem of each individual by the inherent information of sub-problems. From the experimental results, MOEA/D-SS has good convergence and distribution over MOEA/D-DRA, and it has been proved to have a good effect on multi-objective algorithms based on decomposition.The Pareto-based multi-objective evolutionary algorithms have disadvantages of decrease of selection pressure and early convergence, and hence we propose an angle neighbor punishment mechanism based evolutionary algorithm (ANPMEA) to deal wi-th these deficiencies. Firstly, select the elite individual which nearest to the idea point (The ideal point is composed of the smallest objective values in every objectives of the population); Secondly, Design a neighborhood shape based on the including angle between two vectors. Lastly, punish the individuals in the neighborhood of the elite, so as to reduce the opportunity of these individuals of being selected. In this paper, compare our algorithm with NSGA-II and other two MOEAs, AR+DMO, AR+CD'. The results show that the proposed algorithm is better than other three algorithms in terms of convergence and diversity, and the effect is more obvious with the increasing of the number of objective.