基于迭代PCA的油田传感器故障检测与隔离

Fault detection and isolation of sensor in oilfield based on iterative PCA

针对油田采油现场采集的传感器数据本身存在不规律动态特性,使得传统的主元分析(PCA)故障检测方法在实际应用中准确度较低、容易出现误报的问题,采用一种迭代PCA模型方法,即累积数据达到一定长度之后对PCA模型进行迭代更新,可以有效地减小误报的发生.检测出故障后,利用故障数据和残差向量的映射向量定义一个传感器故障指数,可以实现故障隔离.仿真实验表明,与传统的PCA方法相比,本文所采用的更新PCA模型的迭代方法能更好地适用于数据具有动态特性的油田传感器故障检测;通过对传感器故障指数的计算可以准确地实现故障隔离.实验表明,本文用的传感器故障检测与隔离方法可以很好地应用在实际系统中.

To solve the problem that the sensor data collected from the oil production site in oilfield has the irregular dynamic characteristics, and thus the conventional principal component analysis （PCA） method used for detecting the faults exhibits lower accuracy and easier false alarm in the practical application, an iteration PCA model method was proposed. Namely, the iterative update was conducted for the PCA model when the accumulated data reached a certain length, which could effectively reduce the occurrence of false alarm. After the faults were detected, a sensor fault index was defined with the image vector of false data and residual vector, and the fault isolation could be realized. The results of simulation experiments indicate that compared with the conventional PCA method, the iteration method with the updated PCA model can be more suitable for detecting the sensor faults in the oilfield with the dynamic characteristics data. Through calculating the sensor fault index, the faults can be accurately isolated. The experimental results show that the proposed method for both detection and isolation of sensor faults can be well applied in the actual system.