基于数据驱动和机理模型的机械钻速预测  

Data Driven and Mechanistic Model Based Prediction of Rate of Penetration

作　　者：郑双进[1] 江厚顺[1] 熊梦园 孟胡 詹炜[2] 程荣升 王立辉[3] ZHENG Shuangjin;JIANG Houshun;XIONG Mengyuan;MENG Hu;ZHAN Wei;CHENG Rongsheng;WANG Lihui(School of Petroleum Engineering,Yangtze University,Wuhan,Hubei 430199,China;School of Computer Science,Yangtze University,Jingzhou,Hubei 434023,China;Dagang Oilfield Branch,China National Petroleum Corporation,Tianjin 300280,China)

机构地区：[1]长江大学石油工程学院 [2]长江大学计算机科学学院 [3]中国石油天然气股份有限公司大港油田分公司

出　　处：《钻采工艺》2025年第1期78-87,共10页Drilling & Production Technology

基　　金：国家自然科学基金青年项目“穿越井筒天然断裂及人工裂缝滑移与套管变形机理研究”(编号:52404002);油气钻采工程湖北省重点实验室开放基金项目“二氧化碳埋存井井筒屏障密封失效机理研究”(编号:YQZC202411);长江大学非常规油气省部协同创新中心开放基金项目“采用机器学习技术优化页岩气井钻井参数模型研究”(编号:UOG2022-06)。

摘　　要：为准确预测复杂工况下的机械钻速,提出了一种基于数据驱动和机理模型的机械钻速预测方法。首先对收集的8000余条钻井数据进行斯皮尔曼和曼特尔特性分析,筛选出有效施工参数,采用变分模态分解算法(VMD)进行数据降噪,然后构建时序卷积网络结合长短期记忆网络(TCN-LSTM)作为数据驱动模型,并融合多元钻速预测机理模型,通过物理约束增强数据驱动模型的准确性与可解释性,实验表明融合模型比单一数据驱动模型或机理模型预测精度更高。随后,为进一步提升模型性能,采用了改进的蜣螂优化算法(IDBO)对TCN-LSTM模型进行优化,通过改进种群初始化和更新策略,实现了参数的高效搜索。消融实验及现场应用结果表明,对比BP、RF、LSTM、TCN模型,TCN-LSTM-IDBO模型可以实现机械钻速的精确预测,并且具有较好的泛化能力,可为钻井施工人员提供有力参考。To accurately predict the rate of penetration(ROP)under complex conditions,a prediction method based on a combination of data-driven and mechanistic models was proposed.Firstly,the characteristics of Spearman and Mantel were analyzed on the collected more than 8,000 drilling data,the effective operation parameters were screened out,the variational mode decomposition(VMD)algorithm was used for data noise reduction,and then the temporal convolutional network combined with the long short-term memory network(TCN-LSTM)was constructed as the data-driven model,and the multivariate ROP prediction mechanism model was fused to enhance the accuracy and interpretability of the data-driven model through physical constraints.Experiments show that the fusion model has higher prediction accuracy than a single data-driven model or mechanistic model.Subsequently,to further improve model performance,an Improved Dung Beetle Optimization(IDBO)algorithm was applied to optimize the TCN-LSTM model.By improving population initialization and update strategies,efficient parameter search was achieved.Ablation experiments and field application results show that,compared to BP,RF,LSTM,and TCN models,the TCN-LSTM-IDBO model achieved accurate ROP prediction and exhibited strong generalization capabilities,providing valuable references for drilling engineers.

关 键 词：机械钻速预测 时序卷积网络 长短期记忆网络 变分模态分解 蜣螂优化算法 数据分析 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]