基于增量学习的储气库井底压力快速计算方法  

Rapid calculation method of bottom hole pressure for gas storage based on incremental learning

作　　者：郭海伟 GUO Haiwei(Informatization Management Center,Zhongyuan Oilfield Company,SINOPEC,Puyang 457001,China)

机构地区：[1]中国石化中原油田分公司信息化管理中心,河南濮阳457001

出　　处：《断块油气田》2025年第2期292-299,共8页Fault-Block Oil & Gas Field

基　　金：中石化中原油田分公司科技项目“基于数据驱动的储气库注采能力评价”(2024BY05)。

摘　　要：针对储气库井底压力预测的效率和精度问题,文中创新地提出了一种基于增量学习的井底压力快速计算方法。该方法首先构建垂直管流方程,深入分析注气、采气和关井3种不同工况下决定井底压力计算准确性的关键因素。随后,运用极致梯度提升(XGBoost)模型融合了这3种状态下的传统理论计算方法,形成了一个综合性的机器学习模型。通过增量学习策略,并结合实际测量的井底压力数据对模型进行了优化。该方法在计算速度上远超传统技术,能够实时处理井口压力和流量数据,模拟了一个虚拟的井下永久压力计,大幅减少了测压成本。在中原油田卫11储气库的注采井中进行了验证。该方法成功构建了一个融合数据和传统理论的井底压力综合计算代理模型,显著提升了预测的效率和精确度。Aiming at the efficiency and accuracy of bottom hole pressure prediction in gas storage,a rapid calculation method of bottom hole pressure based on incremental learning is proposed.The method firstly constructs vertical pipe flow equations and deeply analyzes the key factors determining the accuracy of bottom hole pressure calculation under three different working conditions,namely,gas injection,gas production and well shut-in.Following this,XGBoost algorithm is used to integrate the traditional theoretical calculation methods in these three states to form a comprehensive machine learning model.The model is optimized by incremental learning strategy and combining with actual measured bottom hole pressure data.This method is much faster than the traditional technology,and can process the wellhead pressure and flow data in real time,simulating a virtual downhole permanent pressure gauge,and greatly reduce the cost of pressure measurement.It is verified in the injection and production well of Wei 11 gas storage in Zhongyuan Oilfield that the method successfully constructs a comprehensive bottom hole pressure calculation proxy model combining data and traditional theory,which significantly improves the prediction efficiency and accuracy.

关 键 词：增量学习 垂直管流 井底压力 储气库 极致梯度提升 

分 类 号：TE19[石油与天然气工程—油气勘探]