Research On Fuzzy Support Vector Machine Optimization Based On Unacertained Information

In general, uncertainty is always relate to quantitative and qualitative factors in practicalproblems. Fuzzy support vector machine (SVM) is reduce the uncertainty information impactto the standard support vector machine (SVM) in learning precision and generalization ability.This machine learning method is used fuzzy set theory, based on statistical learningtheory.However, the existing fuzzy support vector machine method has the low accuracyanalysis and the large-scale data "bottleneck" problems when is large-scale data "bottleneck"problems. Therefore, this thesis is firstly analysis the uncertainty information characteristics,the support vector machine and fuzzy support vector machine theory and technology. Then thispaper is research on data preprocessing model, fuzzy kernel function and the dynamic parameteroptimization algorithm based on nature data and geometric structure characteristics ofuncertainty information. The work and innovation achievements of this thesis mainly includethe following five aspects:(1) Starting from nature data and geometric structure characteristics of uncertaintyinformation, the fuzzy number measurement formula is built based on the fuzzy set theory.Using this formula, the input data can be preprocessed effective. Then the unascertainedquadratic programming in the standard model of fuzzy support vector machine is improved inthe target decision function and the constraint conditions according to the improved fuzzynumber measurement formula.(2) High dimensional feature space mapping response study the similarity betweensamples. After analyzing the relationship of high dimensional feature space mapping, similaritymeasure and kernel function, the improved Gregson fuzzy similarity measurement method isconstructed to applicable to fuzzy kernel function. This methods for uncertainty study sampleshave a more accurate mapping and considering the different situation of uncertainty informationsamples. The works could provide the research foundation for fuzzy support vector machinekernel function.(3) According to the theory of kernel function, the high-dimensional space mappingcorresponding to low dimensional space vector inner product, vector inner product can beexpressed by using to satisfy the Mercer theorem of kernel function. The performance of thefuzzy support vector machine depends largely on the choice of kernel function. The fuzzysimilar kernel is proposed based on the improved Gregson fuzzy similarity measurement method. Compared with the existing fuzzy kernel function, the proposed fuzzy similar kernelcan be more accurate for the analysis and research of the uncertainty information.(4) The performance of fuzzy support vector machine rather depends on the selection ofgood model parameter. This paper is put forward an adaptive PSO-GA parameters selectionand optimization method by using Genetic Algorithm (GA) method and Particle SwarmOptimization (PSO) method. Then, the method used for the analysis of the unascertained fuzzysupport vector machine algorithm model research. This method can effectively solve theproblem of easily plunged into local optimal solution and the existing defects of slowconvergence speed in PSO algorithm and GA algorithm and introducing the adaptive factorconsidering the uncertainty information characteristic. This method improved the searchingcapability of the parameter optimization algorithm, and algorithm robustness and calculatetime-consuming has obvious advantages. Also, computing precision, computing speed and thestability of algorithm is superior to the standard PSO and GA algorithm.(5) In order to prove the advantage of the proposed and improved unascertained fuzzysupport vector machine algorithm, the Material Removal Rate (MRR) in rotary ultrasonicmachining is used to verify the optimization algorithm of data preprocessing and fuzzy similarkernel function in fuzzy support vector machine; the rolling bearing test table bearing failuredata is used to verify the adaptive PSO-GA parameters optimization algorithm. Through thesepractical problems, and real industrial data information, the methods, which are proposed andimproved in this paper, can prove improve the balance of the fuzzy support vector machinelearning precision and generalization ability in the face of uncertainty information analysis.