Research of sensor fault detection based on the residual flitered for the oilfield petroleum exploited system

Conventional principal component analysis (PCA) method is used in detecting the sensor fault without regard to the noise. The uncertain noise and abrupt faults in the process data can lead to a mass of misstatement and false alarms ultimately. This paper deals with how the rate of false alarms can be reduced by using an improved PCA sensor fault detection method with filter EWMA (exponentially weighted moving average) application to the oilfield system. The method manage the residuals by means of filtering. An EWMA filter is used to the model residuals in this paper. To improving the accuracy of the results, a new fault detection index f is proposed in the residual subspaces. The new sensor fault index(SFI) with the filtered residual vectors can reduce the possibilities of false alarms in sensor fault detection effectively. Simulation results of sensor fault detected for the petroleum exploited system are compared between conventional SPE and the new sensor fault index. Conclusions can be summarized that the latter one is more accuracy and the filtered residual vectors can effectively lower the false alarms caused by noise or abrupt faults.