Particle Swarm Optimization And Its Research On Multi-Learning Strategies

With the development of information technology, a large number of high-dimensional, strong coupling parameters problems are appeared.These problems often can not be solved by the traditional optimization methods. Therefore, people's demand for efficient optimization control technology and information technology is more and more urgent. Swarm intelligence optimization algorithm with adaptive capacity have been proposed, where particle swarm optimization (PSO) is one of the most famous and popular of swarm intelligence optimization algorithm.Particle swarm optimization has been widely used in many fields, such as pattern recognition, operations research, computer information networks, and so on. Particle swarm optimization (PSO) is an effective tool to solve the global optimization problems. However, the standard particle swarm optimization (PSO) has the disadvantage of poor convergence precision and easy to fall into the local optimal solution. To enhance the convergence precision of PSO and avoid premature convergence phenomenon are very important.In this paper, the research background of particle swarm optimization (PSO) is introduced. The basic principles of particle swarm optimization (PSO), bare-bones particle swarm optimization (BPSO)and particle swarm optimization variants are introduced in detail in the second chapter. Based on the standard particle swarm optimization(PSO),bare-bones particle swarm optimization(BPSO) and Comprehensive Learning Particle Swarm Optimization(CLPSO), two different learning strategies of particle swarm optimization algorithms are proposed, which are simple comprehensive learning particle swarm optimization(also called Improved Modified Particle Swarm Optimization, IMPSO) and Gaussian elite perturbation particle swarm optimization(also called New Modified Particle Swarm Optimization, NMPSO).Firstly, a modified particle swarm optimization (MPSO) is proposed in the third chapter. On this basis, a simple comprehensive learning strategy is applied and the parameters are discussed. The same particles in different dimensions random select other particles' history optimal solution as own learning models, rather than simply learning their own history optimal solution. This learning strategy can increase the diversity of the swarm and can improve the probability of jumping out of the local trap. To verify the effects and the benefits of the proposed algorithms and strategies, a set of well known benchmark functions are employed and compared with some competitive PSO variants. Experimental results indicate that MPSO and IMPSO perform better than the traditional PSO, BPSO, CLPSO and MBPSO algorithm.Then the Gaussian elite perturbation particle swarm optimization algorithm is proposed in the fourth chapter, which makes Gaussian perturbation to the historical optimal solution and the global optimal solution. The strategy makes the population always in a non-equilibrium state, so that the particle can continue to explore the high quality solution. Thus this strategy improves the searching ability of the algorithm. Experiment results indicate that NMPSO performs better than the traditional PSO, BPSO and MBPSO algorithm.