基于ConvLSTM神经网络的有杆抽油系统故障诊断研究

Fault diagnosis of rod pumping system based on ConvLSTM neural network

针对复杂抽油环境中的有杆抽油系统(rod pumping system,RPS)故障问题,提出一种基于深度学习算法的故障诊断方法。通过边缘计算设备,采集中国西北部油田12种油井工况的11961个抽油杆悬点载荷数据,构建油井使用年限范围内的常见工况数据集。利用卷积长短期记忆神经网络(convolutional long short-term memory neural networks,ConvLSTM)模型,建立抽油杆载荷数据与工作状态之间的特征关联,将算法模型部署在边缘计算装置中,从而实现RPS故障的边缘诊断。为了验证该方法的有效性,采集了5126例生产数据进行测试。结果表明,该故障诊断方法具有更高的准确率及更好的泛化能力,可对RPS进行接近实时的故障诊断。

A rod pumping system(RPS)fault diagnosis method based on deep learning algorithm was proposed for the problem of RPS fault in complex pumping environment.The data of 11961 sucker rod suspension loads for 12 oil wells in oilfields in Northwest China were collected by edge calculating device,and the common working condition data set within the service life of the oil well was constructed.The characteristic relationship between the sucker rod load data and the working condition was established by using convolutional long short-term memory neural networks(ConvLSTM)model,and the algorithm model was deployed in the edge calculating device to realize the edge diagnosis of RPS fault.The method was validated experimentally using 5126 working condition data queues collected from the oil field.The results show that the proposed fault diagnosis method has higher accuracy and better generalization ability,and can perform near-real time fault diagosis of RPS.