Function Optimization And Application Of Improved Genetic Algorithm

Many engineering problems’mathematical model is a multi-modal function, such as optimal parameters’design of fuzzy systems, control systems, radar target identification, and so on. Genetic algorithm is an effective method for solving multi-modal function problems, but basic genetic algorithm needs to supplement by optimization strategy to designed questions for multimodal function, otherwise basic genetic algorithm often fall short of the requirements of multi-modal functions’ optimization. Support vector machine (SVM) is a rising learning machines that has good prospects in recent years. Selection of parameters of SVM in the international community has not recognized a mature, perfect theory to guide, which brought great inconvenience to the wide application of SVM. SVM parameters selection in fact is a multi-modal function optimization problem, and by modern intelligent algorithms to dynamically obtain the corresponding optimum parameters of SVM problem is the current trend. For the features for multi-modal function optimization problem, the paper is to improve the basic genetic algorithm that supplemented by some optimization strategy and aim at solving multi-modal function optimization problems. Improvements in this paper and main works are the following:(1) Redesign of fitness of functions, so that every extreme point has the same fitness, but also can distinguish different levels that close to the extreme points of points. Using crossover strategy based on fitness, which can adjust dynamically global search intensity of algorithm based on individual fitness condition. Designing adaptive mutation operator, so that different individuals can adjust their local search accuracy based on fitness to accelerate the convergence. In order to verify the validity and performance of algorithm, we take some tests for 2 single peak functions,7 unconstrained multi-modal functions and 1 constrained multi-modal function.(2) For the particularity of SVM parameters’selecting problem, this paper modifies this improved genetic algorithm for multi-modal function optimization problems and assists with extra evolution strategy to get the best feasible solutions with generalization capacity of supreme classification right rate. For the problem of the penalty within a certain range corresponding separating hyper plane has the same classification accuracy in the same kernel parameter value case, we can take a phased evolution strategy. In order to verify the feasibility of this improved genetic algorithm for SVM parameters’selection problem in this paper and the correctness of the judgments to the rules, this paper takes a test for 13 data sets from UCI library, and compares with some improved algorithms of other references, experiments show that improved genetic algorithm of this paper is good at solving SVM parameters selection.