考虑全过程优化的支持向量机预测方法

发布时间：2018-11-07 08:27

【摘要】：针对支持向量机(support vector machine,SVM)预测过程中影响因素选择、输入特征集优化、核函数选择及参数优化方面存在的问题,提出了一种全过程优化方法。首先使用频繁模式增长关联规则分析和模糊贝叶斯网络组合模型来解决影响因素选择中存在的主观性问题,然后使用在异常值处理和类内距离与类间距离方面进行改进的模糊C均值聚类算法优化输入特征集,减小支持向量机预测模型冗余度及训练样本集过修正度,通过比较各核函数的特点选择径向基核函数作为SVC的核函数,改进了粒子群优化算法中微粒速度和位置函数及惯性权重值算法,使用该方法优化SVM参数并建立预测模型。最后,通过案例运算和分析,证明该文方法具有更高的预测精度。
[Abstract]:Aiming at the problems of factors selection, input feature set optimization, kernel function selection and parameter optimization in the prediction process of support vector machine (support vector machine,SVM), a full process optimization method is proposed. Firstly, we use frequent pattern growth association rule analysis and fuzzy Bayesian network combination model to solve the subjective problem in the choice of influencing factors. Then the input feature set is optimized by using the improved fuzzy C-means clustering algorithm in the aspects of outlier processing and the distance between classes and intra-class distance. The redundancy of SVM prediction model and the over-correction of training sample set are reduced. By comparing the characteristics of each kernel function, the radial basis function is selected as the kernel function of SVC. The particle velocity and position function and inertial weight value algorithm in particle swarm optimization algorithm are improved. The SVM parameters are optimized and the prediction model is established by using this method. Finally, it is proved that the proposed method has higher prediction accuracy by case analysis.
【作者单位】：装甲兵工程学院技术保障工程系;中国人民解放军68207部队;中国国防科技信息中心;
【基金】：武器装备预先研究基金军队技术基础项目(A157167) 军队维修改革科研项目(2012SC49,2014BZ54)资助课题
【分类号】：TP18

【相似文献】