The Research On Multi-objective Optimization Based On Swarm Intelligence Algorithm

This thesis proposes a multi-objective particle swarm optimization algorithm based on game strategies for continuous multi-objective optimization problem, and a co-evolutionary particle swarm optimization algorithm for Multi-Objective Traveling Salesman Problem (MOTSP) with discrete variables. Focusing on the swarm intelligent algorithms for solving multi-objective optimization problems, the main achievements of this thesis can be summarized as follows:(1)A novel multi-objective particle swarm optimization algorithm inspired from Game Strategies (GMOPSO) has been proposed, where those optimized objectives are looked as some independent agents which tend to optimize own objective function. Therefore, a multi- player game model is adopted into the multi-objective particle swarm algorithm, where appropriate game strategies could bring better multi-objective optimization performance. In the algorithm, novel bargain strategy among multiple agents and nondominated solutions archive method are designed for improving optimization performance. Moreover, the algorithm is validated by several simulation experiments and its performance is tested by different benchmark functions.(2) Inspired from the co-evolutionary, a new hybrid particle swarm evolutionary algorithm for solving TSP is proposed. The algorithm adopts an effective code schema and defines a new addition operation of the particle's position in order to exchange information among the particles. Here a mutation operator is designed to keep the population's diversity.(3) In order to keep a balance between enhancing convergence speed and improving species diversity, a new mutation operator is designed, which consist of local pheromone updating theory in ant colony algorithm and disturbance operator in simulated annealing algorithm. New algorithm has a better performance than the above algorithm.(4) A new algorithm is designed for solving multi-objective TSP that is prevalent existed in real life, which adopts new archiving mechanism by improving the above algorithm. The performance of algorithm is validated through a series of simulation experiments. PSO is a new evolutionary optimization method, which could reduce computational burden of large-scale multi-objective optimization problem. The achievements of this thesis have important theoretic and realistic significance in researching multi objective particle swarm optimization, especially in realistic engineering optimization problem with discrete variables.