基于电子鼻的SVM模式识别算法的研究与实现(2)
This article mainly expounds the data pre processing of machine smell，we applied min-max linear transformations for data normalization,and used PCA to reduce dimension.Referring to various domestic and overseas studies and combining the theories and experimental schemes of various scholars, we propose three parameters optimization schemes:genetic algorithm, Particle Swarm Optimization (PSO) algorithm,
grid search.Besides，the study adopts the method of cross validation experiments which can be set in proportion to the training and testing to verify the accuracy of models.
Finally,this article compared the results of the three parameter optimization methods in 6 kinds gas data.Firstly the study compared on whether do dimension reduction data pre processing or not,and analyzed the necessity of dimension, because of the high dimension data sets (128 d).The study found that after dimension reduction,the parameter optimization and training efficiency increased over 50% of the time and the accuracy increased by 1% ~ 2%,which means it’s correct and very necessary to do dimension reduction on high dimension data.Secondly,the study compared the time efficiency and accuracy,and analyzed the advantages and disadvantages of three methods of parameter optimization. When the data set contains small number of samples, the grid search algorithm is ideal to be used to optimize the parameters of support vector machine(SVM).when the data set contains a larger number of samples,it’s better to choose genetic algorithm and particle swarm algorithm parameters of support vector machine(SVM). Genetic algorithm (GA) optimization of support vector machine parameters is not as good as Particle swarm optimization (pso) algorithm on running time, but is superior in the classification accuracy.
Keywords: electronic nose; support vector machine (SVM); feature extraction; pattern recognition
目录
摘要 I
Abstract II
第一章 绪论 1
1.1研究背景 1
1.2研究意义 2
1.3国内外研究现状 2
1.4 论文结构 2
第二章 数据预处理 3
2.1 归一化 3
2.2 主成分分析（PCA） 3
2.2.1.主成分分析的基本思想 4
2.2.2 主成分分析的步骤 5
第三章 支持向量机............. 6
3.1最优超平面理论 6
3.2线性支持向量机 8
3.3非线性支持向量机 10
3.4遗传算法 12
3.5粒子群算法 15
3.6网格搜索法 16
第四章 算法的优化与验证 17
4.1 SVM参数优化需要优化的参数 17
4.1.1惩罚因子C 17
4.1.2核函数参数gamma 18
4.1.3交叉验证（CV）参数K值 19
4.2实验环境................. 19
4.2.1 libsvm工具箱简介 19
4.2.2气体数据集简介 20
4.3 数据预处理 22
4.4基于遗传算法的参数寻优 25
4.5基于粒子群算法的参数寻优 29
4.6基于网格算法的参数寻优 33
4.7 数据预处理和交叉验证参数的实验对比 36
第五章 结论 38
致谢： 39
参考文献： 40
第一章 绪论
1.1研究背景
机器嗅觉是一门结合了计算机科学、仿生学以及应用数学的交叉学科，近年来发展迅猛，在人们的日常生活中发挥着越来越重要的作用。机器嗅觉系统俗称电子鼻，是一种模仿生物嗅觉系统的智能装置，它由具备部分专一性的化学传感器构成的阵列和适当的模式识别系统组成[3]，可以通过对未知气体的检测、分析、识别和评判来测定简单或复杂气体的信息。