Multimodal Optimization And Application Of Particle Swarm Optimization

There are many problems that are multi-modal in the field of computer, electric and chemical industry, such as the design of the parameters of the fuzzy system, and the problem of this type often has many effective solutions. SVM is a relatively new type of learning machine in recent years; there are many practical applications in pattern recognition field. SVM is often used to solve classification problems in data mining field and its penalty factor and kernel function parameters have important influence on the classification performance of SVM classifier. Selection of SVM parameters is also a multi-modal problem, and there is no perfect mature theory to guide the selection of its parameters, and the use of modern intelligent algorithm is a relatively good method of dynamic. Particle swarm algorithm is simple and easy to implement, efficient operation, but if used directly for solving multi-modal problems there will be many difficulties, such as difficult to find multiple different extreme, not up to multimodal optimization requirements. In view of the characteristics of the multi-modal optimization problem, the basic particle swarm optimization algorithm is combined with some strategies in order to solve the multi modal optimization problem:(1) The different social learning strategies that change the basic particle swarm optimization algorithm only refer to the lack of a single optimal particle. It can improve the diversity of the population and to avoid the single particle swarm and search easily trapped in local minima, leading to search cannot continue to move forward.(2) A new adaptive value assessment method that the extreme position has to adapt to the same value. This assessment method can inform us the position close to the extreme point. Combining with the first point to improve content, it can improve the chances of finding all extreme points of algorithm in theory. In order to verify the effectiveness of the improved measures and the performance of the algorithm, in the second chapter,8 multi peak functions are used to test.(3) SVM is a multi-modal parameter selection problem. Besides it has some certain particularity. We modify the particle swarm algorithm that is improved for multimodal function optimization according to the selection of the SVM parameters with special additional impairment to search strategy to find the parameters with correct classification rate and the best generalization ability in second chapter. In order to verify the effectiveness of several measures and the performance of the algorithm for SVM parameter selection problem, we take a test by using 13 data from the UCI database in the third chapter and compare with other algorithms. The comparison results show that the SVM parameter selection problem and some improved measures are effective and the performance of the algorithm is also good.