结合相空间和LS-SVM的风机状态预测方法  

Trend prediction for condition of fans based on phase space and least squares support vector machine

作　　者：王晓景[1] 黎敏[1] 阳建宏[1] 王蓬[1] 李春杰 

机构地区：[1]北京科技大学机械工程学院,北京100083 [2]中国机械设备工程股份有限公司,北京100055

出　　处：《中国科技论文》2013年第8期743-746,共4页China Sciencepaper

基　　金：高等学校博士学科点专项科研基金资助项目(20090006120007);国家自然科学基金资助项目(51004013)

摘　　要：针对风机运行过程中存在的非线性非平稳特征,提出一种相空间与最小二乘支持向量机(LS-SVM)相结合的风机状态预测方法。首先利用相空间重构方法将一维的时间序列拓展到高维相空间中,还原出风机运行的动力学行为;然后将高维空间中的拓扑结构输入到最小二乘支持向量机中,利用其非线性拟合的优势,最终实现风机状态的趋势预测。利用该方法与BP神经网络方法分别对工业现场的风机振动信号进行对比分析,最大预测误差从7.22%下降到3.75%,说明在相同样本数的条件下,新方法能够更准确地预测风机的振动状态,可为维修决策提供更可靠的数据支持。According to the characteristics of nonlinearity and nonstationarity which occur in the working process of fans a trend prediction method is proposed based on the combination of phase space and least squares support vector machine （LS-SVM） for condition of fans. Firstly, a time series is extended into the high dimensional phase space using the phase space reconstruction method. This high dimensional phase space can restore the dynamic behavior of condition of fans; then the topology of the high dimensional phase space is input to the LS-SVM. Due to the LS-SVM＇s advantage of nonlinear fitting, the condition of fans is thus predicted. The vibration signal of fans in industrial field is analyzed. The results show that, compared with BP neural net-work method, the maximum prediction error using the proposed method drops from 7. 22% to 3.75%. As our method can predict the condition of fans more accurately under the same sample size, it can provide more reliable data support for the maintenance decision.

关 键 词：风机 故障诊断 相空间方法 最小二乘逼近 支持向量机 

分 类 号：TH43[机械工程—机械制造及自动化] TH165.3