基于支持向量回归的设备故障趋势预测

Fault Trend Prediction of Device Based on Support Vector Regression

介绍了故障趋势预测的研究现状及支持向量回归的基本原理,将支持向量回归用于滚动轴承故障趋势的预测,并与其他方法(BP神经网络、灰色模型及灰色-AR模型)进行比较。结果表明,BP神经网络预测结果不稳定,易出现过学习和局部极小问题;支持向量回归预测结果稳定,在预测精度上优于BP神经网络、灰色模型、灰色-AR模型,是故障趋势预测的一种有效方法。

The research condition of fault trend prediction and the basic theory of support vector regression（SVR）were introduced.SVR was applied to the fault trend prediction of roller bearing,and compared with other methods（BP neural network,gray model, and gray-AR model）.The results show that BP neural network tends to overlearn and gets into local minimum so that the predictive result is unstable.It also shows that the predictive result of SVR is stabilization,and SVR is superior to BP neural network, gray model and gray-AR model in predictive precision.SVR is a kind of effective method of fault trend prediction.