Research Of Adaptive Mutation Of Particle Swarm Optimization

With the rapid development of society, people’s needs for more advanced and efficient technology are also increasing. Therefore, how to do optimization has became a attractive topic. Accordingly, many different methods have been developed to solve the constantly emerging optimization problems, ranging from traditional optimization method based on functional properties to mimic nature’s heuristic optimization method. However, these methods always have its drawbacks. Particle Swarm is a classic efficient algorithm, which simulates the foraging behavior of birds. It is a intelligent algorithm with simple operation and high search efficiency. However, there also exists some shortcomings like pre prematurity, late slow convergence, low convergence accuracy, and easy local convergence. Our main work includes the following aspects:(1) Introduction of mutation mechanism proposed new variant conditions:the search area is divided into a number of sub-areas of the same size, if the provisions of the particles is about to enter the sub-region already has some particle, the particle choose variation. This variation makes the original prematurity disadvantage into advantage of the breadth of high search.(2) A new variant rule. According to the unknown equivalence principle, the statistical history visits of each sub-region, it can determine the next particle mutation probability in the region, so the increase in mutation particle prior knowledge make more rational and effective.(3) The proposed regional dichotomy, which extended the dichotomy from a one-dimensional to higher dimensions, inherit their accuracy controllable advantages. Regional dichotomy makes neighborhood search capability stronger, so each particle of particle swarm enhanced to improve search accuracy.(4) A single particle algorithm. On the basis of a good mutation mechanism and a strong neighborhood search mechanism, simplified search mechanism PSO algorithm, it makes the multi-particle collaboration into a single particle iteration cycle variation which greatly reduces the complexity of the algorithm improves the search breadth and efficiency.(5) Information on the estimated regional distribution of information, namely, "terrain" improves the PSO algorithm adaptive for different issues based on historical search. After searching for some time, through the "terrain" analysis to calculate the area of fitness, the selection of regions with the highest fitness as a priority search area, ranging from the search for the sub-region as a whole is reduced to thereby improve search accuracy. While retaining particles global variation prevent local optimization.(6) The new particle swarm algorithm applied to K-means algorithm, improve its shortcomings like initial center sensitive and easy to fall into local optimum.Finally, we make a summary of our main contributions in present paper and present a prospect for our future research directions.