油气藏评价与开发 ›› 2025, Vol. 15 ›› Issue (5): 796-806.doi: 10.13809/j.cnki.cn32-1825/te.2025.05.009

• 油气勘探 • 上一篇    下一篇

基于集成算法的玛湖凹陷油气储层价值预测模型

袁婧1(), 贾鹿1(), 许国剑2, 艾民1, 李嗣旭1   

  1. 1.中国石油新疆油田公司数智技术公司,新疆维吾尔自治区 克拉玛依 834000
    2.中国石油新疆油田公司玛湖勘探开发项目部,新疆维吾尔自治区 克拉玛依 834000
  • 收稿日期:2024-07-08 发布日期:2025-09-19 出版日期:2025-10-26
  • 通讯作者: 贾鹿(1979—),男,博士,高级工程师,现从事油田信息规划及大数据分析研究工作。地址:新疆维吾尔自治区克拉玛依市世纪大道7号,邮政编码:834000。E-mail:jialu666@petrochina.com.cn
  • 作者简介:袁婧(1998—),女,硕士,助理工程师,现从事油田大数据分析与人工智能研究工作。地址:新疆维吾尔自治区克拉玛依市世纪大道7号,邮政编码:834000。E-mail:sjyuanjing@petrochina.com.cn
  • 基金资助:
    新疆维吾尔自治区科学技术厅“天山英才”培养计划项目“油气生产现场智能应用”(2022TSYCJC0032)

Ensemble learning-based prediction model for oil and gas reservoir value in Mahu Sag

YUAN Jing1(), JIA Lu1(), XU Guojian2, AI Min1, LI Sixu1   

  1. 1. Digital Technology Company, PetroChina Xinjiang Oilfield, Karamay, Xinjiang 834000, China
    2. Mahu Exploration and Development Project Department, PetroChina Xinjiang Oilfield Company, Karamay, Xinjiang 834000, China
  • Received:2024-07-08 Online:2025-09-19 Published:2025-10-26

摘要:

位于新疆准噶尔盆地西北部的玛湖油田,是全球最大的砾岩油田之一,其储量规模已达10亿吨级。然而,油田储层物性较差、非均质性强,给油气资源的高效开发带来了巨大挑战。高效开发油气资源的关键,在于精确识别有工业生产价值的储层,即那些油气产能较高且开发成本较低的区域。针对准噶尔盆地玛湖凹陷油气储层评价的复杂性,研究提出了一种基于集成算法的油气储层价值预测模型——OGRV(Oil and Gas Reservoir Value)。研究首先深入分析了玛湖凹陷的地质特征与油气勘探现状,随后构建了一个融合随机森林(RF)、长短期记忆网络(LSTM)和卷积神经网络(CNN)的集成算法,以此提升储层评价的准确性与泛化能力。在具体实施阶段,通过系统开展预处理与特征工程,提取了关键特征参数,并结合领域专家知识,构建了增维特征,例如烃湿度比、烃平衡比和烃特征比。此外,引入滑动窗口技术追踪特征随深度的变化趋势,利用相似井的类别信息作为先验知识来增强模型的预测能力。最终,通过集成不同模型的优势,构建了一个精确且鲁棒的储层评价算法,该算法能有效识别玛湖凹陷区域中具有工业生产价值的储层,在测试集上的F1分数(F1 Score)、准确率(Accuracy)和曲线下面积(AUC)值分别达到0.847 0、0.772 5和0.781 0。研究还深入探讨了模型的可解释性,旨在为地质学家阐明模型的决策机制,助力其在油气勘探和开发领域中做出更明智的决策。

关键词: 储层预测, 玛湖凹陷, 滑动窗口, 集成模型, 可解释性

Abstract:

The Mahu oilfield, located in the northwestern part of the Junggar Basin in Xinjiang, is one of the largest conglomerate oilfields in the world, with reserves exceeding 1 billion tons. However, poor reservoir properties and strong heterogeneity present significant challenges to the efficient development of oil and gas resources. The key to efficient oil and gas development lies in accurately identifying reservoirs with industrial production value, those with higher productivity and relatively lower development costs. To address the complexity of oil and gas reservoir evaluation in the Mahu Sag of the Junggar Basin, this study proposed an oil and gas reservoir value (OGRV) prediction model based on ensemble learning. The study began with an in-depth analysis of the geological characteristics and exploration status of the Mahu Sag. Then, an ensemble model integrating random forest (RF), long short-term memory (LSTM), and convolutional neural network (CNN) was constructed to improve the accuracy and generalization ability of reservoir evaluation. During implementation, key feature parameters were extracted through systematic preprocessing and feature engineering. With expert knowledge, additional augmented features such as hydrocarbon humidity ratio, hydrocarbon balance ratio, and hydrocarbon characteristic ratio were incorporated. In addition, the sliding window technique was introduced to track the trend of features with depth variations, and the category information of similar wells was used as prior knowledge to enhance the model’s prediction performance. By leveraging the strengths of different models, a precise and robust reservoir evaluation algorithm was developed. It effectively identified reservoirs with industrial value in the Mahu Sag. The model yielded an F1-score of 0.847 0, accuracy of 0.772 5, and area under the receiver operating characteristic (ROC) curve (AUC) of 0.781 0. The study also investigated model interpretability in depth to help geoscientists better understand the model’s decision-making mechanisms and support more informed decision-making in oil and gas exploration and development.

Key words: reservoir prediction, Mahu Sag, sliding window, ensemble model, interpretability

中图分类号: 

  • TE122