基于支持向量机的机械系统状态组合预测模型研究  

Study on combination trend prediction technology for mechaninery system based on SVM

作　　者：王红军[1] 张建民[2] 徐小力[1] 

机构地区：[1]北京机械工业学院机械工程系,北京100085 [2]北京理工大学机械与车辆工程学院,北京100081

出　　处：《振动工程学报》2006年第2期242-245,共4页Journal of Vibration Engineering

基　　金：国家自然科学基金资助(50375016);北京市自然基金资助(3042006);北京市重点实验室开放基金资助(030314);高等学校博士学科点专项科研基金资助(20040007029)

摘　　要：提出了一种新的支持向量机(Support V ectorM ach ines,SVM)机械系统状态组合预测模型。应用FPE(F ina lP rinc ip le E rror)准则优化样本的维数,采用时域内的振动烈度和频域内的特征频率分量作为预测机械系统状态的敏感因子,构建了预测模型。支持向量机采用新型的结构风险最优化准则,预测能力强、鲁棒性好。采用径向基函数和ε损失函数,将该模型应用于实验台和旋转注水机组的状态预测,取得了较好的效果。这表明利用支持向量机的组合预测模型,可以降低设备维修代价,提高设备的安全性和可靠性。A novel combination prediction method of machinery condition trend based on SVM is presented, for which the vibration velocity in time domain and the vibration amplitude of the feature frequency in frequency domain are employed as sensitive factors and the condition trend is predicted via SVM theory. And then final prediction＇error （FPE） principle is used to determine the embedding dimension. SVM adopts new type structural risk minimization principle and has good robustness and high forecasting accuracy, which is used for experimental system as well as water injection sets. The good predicting results were obtained with radial basis function （RBF） and ε loss function. The combination method based on SVM is proved to be a method with excellent performance of condition trend prediction for machinery and good for machinery maintenance.

关 键 词：故障诊断 状态监测 机械系统 组合预测模型 支持向量机 

分 类 号：TH165.3[机械工程—机械制造及自动化]