基于PSO聚类和特征贡献度的油液监测信息特征选择方法

发布时间：2018-01-08 23:01

  本文关键词：基于PSO聚类和特征贡献度的油液监测信息特征选择方法　出处：《润滑与密封》2016年01期 　论文类型：期刊论文

  更多相关文章： 机械装备 油液监测 特征选择 粒子群聚类 特征贡献度

【摘要】：特征选择是实现油液监测多技术手段综合应用的关键问题之一。针对油液监测信息特点,提出一种油液监测信息特征选择方法。该方法首先采用K均值PSO聚类算法对样本实施无监督聚类,实现样本的预先分类;然后采用定义的特征贡献度,计算各特征对聚类结果的贡献度,并以此作为特征选择的依据,实现无监督的过滤式特征选择。通过在某型柴油机润滑油原子发射光谱和红外光谱信息中的应用表明,该算法能够很好的实现油液监测信息的特征选择,减少特征指标数量,而且能够避免由于油液监测信息依存度和相关度高的特点而造成特征选择时可能会将重要信息删除的问题。
[Abstract]:Feature selection is one of the key problems in realization of oil monitoring and comprehensive application of multiple technologies. According to the characteristics of oil monitoring information, proposes a feature selection method of oil liquid monitoring information. The method first uses K PSO means clustering algorithm to sample the implementation of unsupervised clustering, sample pre classification; then the definition of feature contribution degree the contribution of each feature, the calculation of the clustering results, and as a feature selection based on selection of filter characteristics. Through unsupervised in a certain type of diesel engine lubricating oil by atomic emission spectroscopy and infrared spectral information shows that this algorithm can achieve very good characteristics of oil monitoring information, reduce the number of features, and can avoid the oil monitoring information dependency degree and high correlation characteristics caused by feature selection may issue important information will be deleted.

【作者单位】： 海军工程大学青岛油液检测分析中心;武汉理工大学能源与动力工程学院;
【基金】：国家自然科学基金项目(51309185)
【分类号】：TK421.9
【正文快照】： j縥縥縥縥縥縥，

本文编号：1398987