An Improved Multi-objective Genetic Algorithm Based on Agent

An improved multi-objective Genetic Algorithm based on agent is offered. In the improved algorithm, agents were co-evolution with different control parameters to increase the diversity of candidate solutions. Two kinds of crossover strategies of Arithmetic and Simulated binary (SBX) were introduced in order to complete the competition behavior of the agent, these strategies increased the choice range of the agent, and improved the search performance. To construct non-dominated set, Arena Principle (AP) was used in the process of self-learning behavior, and the clustering method was used to narrow the non-dominated set, so as to obtain the set of Pareto optimal front. The idea of the elitism retaining was used to quicken the convergence rate, and then formed the elitism of individuals, this performance was similarly to the local climbing for self-learning operation. Finally, we saved these individuals to the elitism population, several standard test functions are used to verify this improved algorithm. The results indicated that the improved Genetic Algorithm (GA) obtained good performance.