The Fault Prediction Based On Principal Component Analysis And Support Vector Machine For Regression

Fault diagnosis is an important aspect of the industrial processes, but the more complexity of the modern industrial system put forward new requirements for the fault diagnosis. While the traditional fault diagnosis which based on mathematical model is no longer practical, on the contrast, the fault diagnosis which based on data-driven attract scholars more and more attention. This paper proposed a PCA-SVR fault prediction model base on the data-driven fault diagnosis.First, for the shortage of the traditional PCA multivariate monitoring method, this paper use the combine monitoring of the traditional Hotelling’s statistic and Q statistic for fault prediction, which not only avoids the simple use of one of the fault monitoring that may cause the failure to the fault diagnosis, but also simplifying the failure prediction model.Second, this paper used a moving time window support vector machines for fault prediction, the support vector machine has advantage in solving the small sample and non-linear problems, while in the actual industrial the fault data for fault diagnosis is limited, the monitoring indicators is also non-linear, which makes the support vector machine becoming the preferred tool for fault prediction. Moving time window method for support vector machine not only can effectively updated the training data information in real time, but also can reduce the size of the training data, thereby reducing support vector machine model training time and the amount of computation.Finally, using the PCA-SVR mode on the Tennessee Eastman simulation process and the actual batch reactor, the results of the TE process simulation and the batch reactor show that the PCA-SVR fault prediction model can track the TE process and the batch reactor fault development trend well.