The Research Of Multi-objective Particle Swarm Optimization Based On Fast-sorting And Its Application

Multi-objective optimization problem(MOP), in which there are multiple conflicting objectives, exists in most area of engineering and science. Different from single optimization problem, the optimal solution in multi-objective optimization problem is not a single one, the optimal of MOP consists of a set of solutions which are not non-dominated by each other. It is not easy for traditional mathematical methods to solve MOP. With the develop of computational intelligence, more and more evolutionary algorithms and swarm intelligence algorithms are implemented in MOP which have shown the advantage in solving MOP.The contents in this thesis include:First of all, the research background and significance of MOP is introduced. This thesis also introduces evolutionary algorithms in the research status in the field of MOP, development course, and hot research direction in the future. Some basic concepts of particle swarm optimization(PSO) are introduced as well as the basic idea and characteristics of PSO, then is the development and research direction of PSO. After that, multi-objective particle swarm optimization(MOPSO) is introduced. A classic algorithm in multi-objective optimization, NSGA-II, is also briefly introduced in this thesis.Secondly, a variant of PSO which is called comprehensive learning particle swarm optimization(CLPSO) is introduced, and the advantage of CLPSO, compared with the other PSOs, is displayed. Study shows that the key of multi-objective particle swarm optimization(MOPSO) is how to fix the external archive and how to choose the global optimal in external archive. In this thesis, an new strategy called Fast-sorting is applied in multi-objective comprehensive learning particle optimization(MOCLPSO) to fix the external archive. This method aims to improve the convergence and diversity of the algorithm, what’s more, it can boost the speed of algorithm. How parameters influence the algorithm is also analyzed. The improved algorithm and another two classic algorithms are tested on benchmark functions, and the result shows the improved algorithm performs better than the other two ones both in convergence and diversity. The speed of improved algorithm is also faster than the other ones.At last, portfolio optimization problem is used to check whether the improved algorithm is able to solve real problem. Markowitz mean-variance model is chosen as the problem to be optimized. Multi-objective comprehensive learning particle swarm optimization based on fast sorting is used to optimize the model. The experiment results show than non-dominated solutions obtained by the improved algorithm is better than the other classic algorithms both in convergence and diversity, more importantly, the simulation time is decreased.