设备状态趋势的SVM预示技术研究  

Support Vector Machine Condition Trend Prediction Technology

作　　者：王红军[1] 徐小力[2] 张建民[1] 

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

出　　处：《机械科学与技术》2006年第4期379-381,共3页Mechanical Science and Technology for Aerospace Engineering

基　　金：国家自然科学基金项目(50375017);北京市自然基金项目(3042006);北京市重点实验室开放项目(030314);高等学校博士学科点专项科研基金项目(2005000306)资助

摘　　要：SVM采用结构风险最优化准则,预示推广能力强并有很好的鲁棒性。分析了SVM技术的理论,推导了SVM进行机电设备趋势预测的理论算法,给出了进行预测的步骤,建立了SVM用于故障趋势预示的模型。将该模型用于某机组振动烈度的预示,进行了不同核函数和不同C值和ε值的比较,证明采用径向基函数和适当的损失函数,取得了较好的预测效果。并将SVM与AR模型的提前20步预测结果进行了比较。结果证明该算法对设备状态的趋势具有较好的预示能力。With structural risk minimization principles, support vector machine（SVM） has excellent generalization ability, good forecasting accuracy and robustness. After the SVM theory is analyzed, a novel SVM trend prediction method is presented, and the SVM prediction model and procedures are put forward. The SVM model has been applied to the fault trend prediction of the vibration from a machine set. Through comparing different kernel functions and parameters C and 6, good prediction results have been obtained by using radial base functions （RBF） and proper loss functions. Furthermore, the advanced-20- step prediction result of the SVM model is compared with that of auto regression （AR） for the machine set, which shows that the SVM has excellent machinery condition trend prediction ability.

关 键 词：设备状态 趋势预示 SVM 预测模型 

分 类 号：TH17[机械工程—机械制造及自动化]