基于神经网络的莺歌海盆地DF区块黄流组储层压力预测与成因分析

Pressure prediction and genesis analysis of Huangliu Formation reservoir in DF block of Yinggehai Basin based on neural networks

宁卫科 鞠玮 相如

NING Weike JU Wei XIANG Ru

1. 煤层气资源与成藏过程教育部重点实验室, 江苏 徐州 221008; 2. 中国矿业大学 资源与地球科学学院, 江苏 徐州 221116; 3. 中国科学院 地理科学与资源研究所, 北京 100101

1. Key Laboratory of Coalbed Methane Resources and Reservoir Formation Process, Ministry of Education, Xuzhou, Jiangsu 221008, China; 2. School of Resources and Geosciences, China University of Mining and Technology, Xuzhou, Jiangsu 221116, China; 3. Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China

在油气勘探开发及生产过程中，储层压力对油气聚集、分布及运移的过程起着重要作用，异常高压储层甚至会造成井壁崩落、井涌、井喷等钻井事故。传统的储层压力测井预测主要采用经验公式法、有效应力法等，存在参数确定难、主观性强等问题。为此，以莺歌海盆地DF区块为例，在实测数据基础上，构建基于BP神经网络和卷积神经网络的储层压力预测模型，建立测井曲线与实测储层压力之间的隐式直接关系，对储层压力进行了预测并分析了其超压成因。研究结果表明：(1)构建的卷积神经网络模型预测储层压力精度高，最优模型的均方根误差为0.27 MPa；(2)预测莺歌海盆地DF区块黄流组储层压力为53.26～55.60 MPa，平均压力系数为1.66～1.95，呈现为超压；(3) DF区块黄流组超压成因机制为以流体膨胀作用为主，欠压实作用为辅。

In the process of oil and gas exploration, development and production, reservoir pressure plays a crucial role in the accumulation, distribution and migration of oil and gas. Abnormally high-pressure reservoirs can lead to drilling accidents such as wellbore collapse, kicks and blowouts. Traditional methods for predicting reservoir pressure, mainly based on well logging calculations using empirical formula and effective stress methods, suffer from drawbacks including complex parameter identification and significant subjectivity. Consequently, the paper uses the DF block in the Yinggehai Basin as a case study, building a reservoir pressure prediction model based on real-time pressure data using both the BP neural network and convolutional neural network. This process established an implicit direct relationship between logging curves and real-time reservoir pressure, allowing for the prediction of reservoir pressure and an analysis of the causes of overpressure. The results of the study indicate that: (1) The established convolutional neural network model demonstrates high accuracy in predicting reservoir pressure, with a root mean square error of 0.27 MPa for the optimal model. (2) The predicted reservoir pressure range for the Huangliu Formation in the DF block of the Yinggehai Basin is 53.26-55.60 MPa, with an average pressure coefficient of 1.66-1.95, consistent with overpressure. (3) The mechanism behind the overpressure in the Huangliu Formation, DF block, is mainly due to fluid expansion, supplemented by undercompaction.

国家自然科学基金项目(42372185，41971335)资助。

https://doi.org/10.11781/sysydz2024051088