期刊文献+

融合物理约束的压裂水平井产能智能预测框架构建与应用

Establishment and application of an intelligent productivity prediction framework for fractured horizontal wells with physical constraints
下载PDF
导出
摘要 水平井分段压裂是实现非常规油气藏有效开发的关键技术,准确预测压裂水平井产能对于井位优选和压裂参数优化至关重要。随着历史开发数据的不断积累和人工智能技术的迅速发展,数据驱动的人工智能方法为压裂水平井产能预测提供了新的渠道。为此,从压裂水平井生产物理过程出发,分析了产能智能预测的物理约束,提出了与物理过程相匹配的产能智能预测框架,并结合四川盆地南部(以下简称川南)地区页岩气开发生产数据开展了实例验证。研究结果表明:(1)压裂水平井产能智能预测需要以压裂段为单位进行特征融合处理,单井产能是由初始压裂段到末尾压裂段依次累计作用的结果,各压裂段之间存在顺序关系,各单井的因素输入维度存在差异;(2)采用循环神经网络能够完全匹配压裂段之间的顺序关系和汇聚作用,而Mask屏蔽机制则能够解决各单井压裂段数量不统一的矛盾。结论认为:(1)该智能预测模型能够学习各单井输入序列与产能之间的复杂映射关系,训练集相对误差为0.098、测试集相对误差为0.117,较循环神经网络(RNN)模型误差的下降幅度分别为37.6%和37.0%,较多层感知机(MLP)模型误差的下降幅度分别为77.3%和77.4%,展现出优异的预测性能;(2)该研究成果能够为非常规油气藏压裂水平井产能预测的技术进步和快速发展提供新的思路与借鉴。 Staged fracturing in horizontal wells is a key technology for the efficient development of unconventional oil and gas reservoirs.Accurate productivity prediction of fractured horizontal wells is crucial for the selection of well locations and the optimization of fracturing parameters.With the continuous accumulation of historical exploitation data and the rapid development of artificial intelligence technologies,data-driven artificial intelligence methods have provided new channels for the productivity prediction of fractured horizontal wells.Based on the physical production process of fractured horizontal wells,this paper analyzes the physical constraints of intelligent productivity prediction and proposes an intelligent productivity prediction framework matching the physical process.In addition,case verification is carried out based on the shale gas exploitation data in the southern Sichuan Basin.The following research results are obtained.First,intelligent productivity prediction of fractured horizontal wells requires feature fusion with fracturing stage as a unit.Single well productivity is the cumulative result from the initial stage to the final stage,and there is a sequential relationship between different stages.There are differences in the factor input dimensions of different wells.Second,the recurrent neural network(RNN)can fully match the sequential relationship and aggregation effect among fracturing stages,and the Mask mechanism can solve the contradiction of inconsistent stage number in different wells.In conclusion,this intelligent prediction model can learn the complex mapping relationship between input sequence and productivity of each well,demonstrating an excellent predictive performance.Its relative errors of training set and testing set are 0.098 and 0.117,respectively,which are 37.6%and 37.0%lower than those of RNN model and 77.3%and 77.4%lower than those of multi-layer perceptron(MLP)model.What's more,the research results can provide new ideas and references for the technological progress and rapid development of the productivity prediction of fractured horizontal wells in unconventional oil and gas reservoirs.
作者 卢聪 罗扬 郭建春 曾凡辉 LU Cong;LUO Yang;GUO Jianchun;ZENG Fanhui(State Key Laboratory of Oil&Gas Reservoir Geology and Exploitation//Southwest Petroleum University,Chengdu,Sichuan 610500,China;Zhenhua Oil Co.,Ltd.,Beijing 100031,China)
出处 《天然气工业》 EI CAS CSCD 北大核心 2024年第9期99-107,共9页 Natural Gas Industry
基金 国家自然科学基金面上项目“深层页岩压裂人工干预转向多裂缝动态扩展机制”(编号:52374044)。
关键词 四川盆地南部 页岩气 水平井 分段压裂 特征融合 产能预测 人工智能 循环神经网络 物理约束 Southern Sichuan Basin Shale gas Horizontal well Staged fracturing Feature fusion Productivity prediction Artificial intelligence Recurrent neural network Physical constraint
  • 相关文献

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部