基于长短时记忆网络的腐蚀工况下抽油杆剩余使用寿命预测

Remaining Useful Life Prediction of Sucker Rod under Corrosion Condition Based on Long Short-term Memory Network

抽油杆工作环境复杂,失效形式多样,其工作寿命受诸多因素影响,如何提高抽油杆在腐蚀工况下的剩余使用寿命(remaining useful life,RUL)显得尤为关键。使用深度学习方法中的长短时记忆网络(long short-term memory,LSTM),根据已有的油田生产数据,选取了15项与抽油杆腐蚀密切相关的变量,通过参数优化、网络训练,构建了基于LSTM的抽油杆剩余寿命预测模型,对20口生产井数据的测试显示,模型预测结果的平均误差为36%,同时与双向LSTM和深度LSTM模型进行对比表明,LSTM预测模型具有更好的预测能力,结果表明:LSTM模型在抽油杆寿命预测领域具有良好的应用前景。

The working environment of sucker rod is complex and the failure forms are various.Its working life is affected by many factors.How to improve the remaining useful life of sucker rod under corrosion condition is particularly critical.Long short-term memory(LSTM)was used in deep learning method.Fifteen variables which were closely related to sucker rod corrosion were selected according to the existing oilfield production data.Through parameter optimization and network training,the remaining useful life prediction model of sucker rod based on LSTM was constructed.The training test of 20 production wells shows that the average error of LSTM model prediction results is 36%.At the same time,the comparison with BILSTM and DEEPLSTM shows that LSTM prediction model has better prediction ability.The research results show that LSTM prediction model has a good application prospect in the field of sucker rod life prediction.