深度学习的电泵井产液量动态预测模型

电泵井产量的动态预测对认识油井供排协调、电泵设备工况、改善工作制度、提高产量和节能降耗具有重要指导意义。根据电泵井静态数据、生产动态数据、举升设备的工况数据，利用皮尔逊相关系数分析方法，分析了影响潜油电泵工作特性因素的关联性，根据主成分分析方法(PCA)进行数据降维确定主控参数，并综合考虑了电泵机组设备工况的变化趋势和前后关联性，应用长短期记忆神经网络(LSTM)建立了电泵井产量时序预测模型。利用某油田现场实际数据对产液量进行预测，并与BP神经网络的预测结果相比较。研究结果表明，基于LSTM模型的电泵井产液量预测值与现场实际值高度一致，预测模型拟合效果更好，预测精度更高，考虑因素更全面、应用更方便、结果更可靠，进而为潜油电泵生产的产液量动态预测提供了一种新的方法，为潜油电泵工作制度调整以及合理选泵设计提供依据。

Deep learning based dynamic prediction model of ESP well production

The dynamic prediction of electric submersible pump (ESP) well production is of guiding significance to recognize supply-discharge coordination of oil well and working condition of ESP equipment and improve working system, production rate and energy saving. In this paper, the relationships between the factors influencing the operating performance of ESP were analyzed by means of the Pearson correlation coefficient analysis method, based on the static and production dynamic data of ESP well and the working condition data of lifting equipment. Then, the data dimension was reduced by means of the principal component analysis method (PCA) to determine the main control parameters. In addition, the time series prediction model of ESP well production was established by using the long short-term memory (LSTM) and considering the change trend and relevance of the working conditions of ESP equipment comprehensively. Finally, the liquid production was predicted based on the on-site actual data of one certain oilfield and compared with the prediction result of BP neural network. The results show that the ESP well production predicted by LSTM model is highly accordant with the on-site actual value, indicating the prediction model has better fitting results and higher prediction accuracy. It considers factors more comprehensively, its application is more convenient and its prediction result is more reliable. It provides a new dynamic prediction method of ESP production and a basis for the adjustment of ESP’s working system and the reasonable selection and design of ESP.