A Real-time Recognition of Working Patterns to Fault Diagnosis Based on BP Neural Network

In the opening up oilfields, it´s an important task in petroleum industry to predict and diagnose faults under the oilfields. In this paper, an algorithm of recognizing working patterns of oil-well based on BP (back propagation) neural network was put forward, and it was applied to economical automatic monitor and control system for an oilfield. It was more accurate and reliable than direct comparison of the corresponding point in working-pump´s graph. To overcome the disfigurement and limitation of the fault recognition method at present, the Fourier descriptor of input samples was adopted to solve some problems, such as reducing the dimension of samples, increasing train speed, improving recognition rate and real time recognition. The result showed that this automatic monitor and control system was effective and satisfied with the requirement of the real-time fault diagnosis for oil pump