深度学习神经网络在电潜泵井口排量预测与工况诊断中的应用

Application of Deep Learning Neural Network in Wellhead Discharge Prediction and Working Condition Diagnosis of ESP

针对潜油电泵井运行状态的多变量、非线性、强耦合的复杂时序系统,传统的电流卡片、参数阈值方法难以建立精确的工况诊断与产量预警模型,提出一种基于深度学习循环神经网络的长短时记忆算法对电泵井运行状态建模。利用海上油田电泵井大量实时数据,基于Tensorflow框架搭建神经网络模型,实现电泵井井口产量预测,并实现多种异常工况的诊断。

For multivariable, non-linear and strongly coupled complex time series system of ESP wells, traditional current card and parameter threshold methods can not establish accurate condition diagnosis and production forewarning model. The paper proposes long and short term memory algorithm based on deep learning cycle neural network to model operation state of electric pump wells. Based on large number of real-time data of electric pump wells in offshore oilfields, it builds neural network model based on Tensorflow framework to predict wellhead production of electric pump wells and diagnose various abnormal conditions.