Improved Discrete Particle Swarm Optimization And Application In Attribute Reduction

The particle swarm optimization (PSO) is a swarm intelligence optimal algorithm, proposed by Ebethart and Kennedy in 1995. It was quickly recognized by the academic community because the advantages of simple concept, convenient operation, less parameters, and strong global convergence. Currently, it has been widely applied in many fields, such as function optimization, fuzzy system control and industrial production.With the information age coming, the amount of data stored in the database increases quickly, a large amount of data exists in our life and production fields, people need for an efficient data processing method because the traditional statistical techniques and data management tools are no longer suitable for analyzing the massive amount of data set. Attribute reduction theory of rough set is an important method of data preprocessing, it can not only reduce the data size, save storage space and processing time, but also preserve the useful information of the original data, so that the results are more credible and more accuracy.This thesis has maily done the two aspects:(1) To deal with the problem of discrete particle swarm optimization (DPSO) that the particles search blindly and can not carry out local search deeply around the current optimical solution, the traditional DPSO algorithm is improved. In order to escape the local optima, mutation operator with multi-scale possibilities is applied, The mutation operator with large-scale possibilities can be utilized to quickly localize the global optimized space at the early evolution, the novel scale-changing strategy produces a smaller multi-scale mutation operators according to the variation of the fitness value and make mutation operators with smaller-scale possibilities implement local accurate minima solution search at the late evolution. The algorithm can escape local optimum area because it can not only guarantee the local mining ability but also improve the exploration ability. The experiment studies on several standard benchmark functions compare with other DPSO algorithm to prove the superiority of the algorithm.(2) Recently, PSO has also been widely used in minimum attribute reduction. A novel minimum rough-set attribute reduction algorithm based on virus-coordinative discrete particle swarm optimization inspired by virus evolutionary theory in nature is presented. In this algorithm, evolutions of the virus swarm are performed in coordination with the particle swarm, and virus swarm keeps coordinative relations with the particle swarm by virus infection operations and best virus seed extraction operation in order to improve the ability of local search of DPSO. To raise the probability with which the algorithm finds the minimum attribute reduction optimum or a satisfied solution, the cut operator is introduced in the virus swarm's self-renewal process. A proper fitness function is defined and theoretical analysis and the experimental results on UCI dataset attribute reduction show the proposed method appears better than other evolution attribute reduction algorithms, and the searching efficiency and convergence rate for the globally optimum are greatly improved as well.