作者：李华洋1,2,3, 曹志鹏1,2, 吴小龙3,4, 朱施杰3,4, 邓金根1,2, 张水良5
作者单位：1. 中国石油大学（北京）石油工程学院，北京 102249;
2. 中国石油大学（北京） 油气资源与探测国家重点实验室，北京 102249;
3. 中国科学院武汉岩土力学研究所 岩土力学与工程国家重点实验室，湖北 武汉 430071;
4. 重庆大学 煤矿灾害动力学与控制国家重点实验室，重庆 400044;
5. 中海油天津分公司，天津 300459
关键词：破裂压力;机器学习;LightGBM算法;压力预测
摘要：
针对传统的地层破裂压力预测方法预测精度较低、普适性不高等问题，提出基于LightGBM机器学习算法构建破裂压力智能预测模型。以井深、地层密度和孔隙压力当量密度作为模型的输入层数据，以S区块中相邻的3口直井为例验证模型的预测效果，并将LightGBM模型与常用的声波测井资料法进行预测结果的对比分析，最后进行模型的参数敏感性分析。研究结果表明，LightGBM模型的预测精度和稳定性均很好，模型的泛化能力强，5项评价指标均表现得十分优越。LightGBM模型的预测相对误差不超过2%，小于声波测井资料法。所有输入层数据中地层密度对于破裂压力的预测最为敏感。利用LightGBM机器学习算法所建立的破裂压力预测模型不受地质环境的影响，其预测精度也大于声波测井资料法。
Prediction method of formation fracture pressure based on LightGBM algorithm and its application
LI Huayang1,2,3, CAO Zhipeng1,2, WU Xiaolong3,4, ZHU Shijie3,4, DENG Jin’gen1,2, ZHANG Shuiliang5
1. Petroleum Engineering College, China University of Petroleum ( Beijing), Beijing 102249, China;
2. State Key Laboratory of Petroleum Resource & Prospecting, China University of Petroleum (Beijing), Beijing 102249, China;
3. State Key Laboratory of Geomechanics and Geotechnical Engineering, Institute of Rock and Soil Mechanics, Chinese Academy of Science, Wuhan 430071, China;
4. State Key Laboratory of Coal Mine Disaster Dynamics and Control, Chongqing University, Chongqing 400044, China;
5. CNOOC Tianjin Branch, Tianjin 300459, China
Abstract: Aiming at the problems of low prediction accuracy and low universality of traditional formation fracture pressure prediction methods, an intelligent prediction model of fracture pressure based on LightGBM machine learning algorithm is proposed. Taking well depth, formation density and pore pressure equivalent density as the input layer data of the model, the prediction effect of the model is verified by taking three adjacent vertical wells in S block as an example, and the LightGBM model is compared with the commonly used acoustic logging data method. Finally, the parameter sensitivity analysis of the model is carried out. The results show that the prediction accuracy and stability of LightGBM model are very good, the generalization ability of the model is strong, and the five evaluation indexes are very superior. The prediction relative error of LightGBM model is less than 2 %, which is less than that of acoustic logging data method. Among all the input layer data, the formation density is the most sensitive to the prediction of fracture pressure. The fracture pressure prediction model established by LightGBM machine learning algorithm is not affected by geological environment, and its prediction accuracy is also greater than that of acoustic logging data method.
Keywords: fracture pressure;machine learning;LightGBM algorithm;pressure prediction
2024, 50(4):134-143 收稿日期: 2022-12-04;收到修改稿日期: 2023-02-18
基金项目: 国家自然科学基金项目(52174040)
作者简介: 李华洋(1999-),男,辽宁葫芦岛市人,硕士研究生,专业方向为地质力学、机器学习等。
参考文献
[1] 张东东, 刘文汇, 王晓锋, 等. 深层油气藏成因类型及其特征[J]. 石油与天然气地质, 2021, 42(5): 1169-1180.
ZHANG D D, LIU W H, WANG X F, et al. Genetic types and characteristics of deep oil and gas plays[J]. Oil & Gas Geology, 2021, 42(5): 1169-1180.
[2] HUBBERT M K, WILLIS D G. Mechaics of hydraulic fracturing[J]. Petroleum Transactions, 1957, 210: 153-168.
[3] EATON B A. Fracture gradient prediction and its application in oilfield operations[J]. Journal of Petroleum Technology, 1969, 21(10): 1353-1360.
[4] STEPHEN R D. Prediction of fracture pressures for wildcat wells[J]. Journal of Petroleum Technology, 1982, 34(4): 863-872.
[5] 黄荣樽. 地层破裂压力预测模式的探讨[J]. 华东石油学院学报, 1984, 8(4): 335-347.
HUANG R Z. A model for predicting formationfracture pressure[J]. Journal of China University of Petroleum（Edition of Natural Science）, 1984, 8(4): 335-347.
[6] 何伊丽, 贺华, 侯彬彬, 等. 延长气田地层破裂压力模型研究[J]. 内蒙古石油化工, 2021, 47(8): 98-100.
HE Y L, HE H, HOU B B, et al. Study on the formation fracture pressure model in Yanchang gas field[J]. Inner Mongolia Petrochemical Industry, 2021, 47(8): 98-100.
[7] 马先林, 周德胜, 蔡文斌, 等. 基于可解释机器学习的水平井产能预测方法[J]. 西南石油大学学报(自然科学版), 2022, 44(4): 81-90.
MA X L, ZHOU D S, CAI W B, et al. An interpretable machine learning approach to prediction horizontal well productivity[J]. Journal of Southwest Petroleum University(Science & Technology Edition), 2022, 44(4): 81-90.
[8] 孙岿. 基于改进KNN算法的潜山复杂岩性测井识别方法[J]. 特种油气藏, 2022, 29(3): 18-27.
SUN K. Logging identification method of complex lithology in buried hill based on the improved KNN algorithm[J]. Special Oil & Gas Reservoirs, 2022, 29(3): 18-27.
[9] 宋先知, 姚学喆, 李根生, 等. 基于LSTM-BP神经网络的地层孔隙压力计算方法[J]. 石油科学通报, 2022, 7(1): 12-23.
SONG X Z, YAO X Z, LI S G, et al. A novel method to calculate formation pressure based on the LSTM-BP neural network[J]. Petroleum Science Bulletin, 2022, 7(1): 12-23.
[10] 李华洋, 吴惠梅, 南冲, 等. 南海深水L-1井区地层压力预测技术研究及应用[J]. 石油地质与工程, 2020, 34(3): 113-117.
LI H Y, WU H M, NAN C, et al. Study and application of formation pressure prediction technology in deep water L-1 well area of south China sea[J]. Petroleum Geology and Engineering, 2020, 34(3): 113-117.
[11] 林涛, 严寒, 赵丹阳. 基于IWOA-LightGBM的主轴轴承故障预警研究[J]. 中国测试, 2023, 49(5): 82-88.
LIN T, YAN H, ZHAO D Y. Research on fault early warning of spindle bearing based on IWOA-LightGBM[J]. China Measurement & Test, 2023, 49(5): 82-88.
[12] 张航, 史兆培, 束垠, 等. 基于OOB-BO-LightGBM的风电机组故障诊断方法[J/OL]. 中国测试: 1-7[2022-12-04]. http://kns.cnki.net/kcms/detail/51.1714.TB.20220623.1608.040.html.
ZHANG H, SHI Z P, SHU Y, et al. Fault diagnosis method of wind turbine based on OOB-BO-LightGBM[J/OL]. China Measurement & Test: 1-7[2022-12-04]. https://kns.cnki.net/kcms/detail/51.1714.TB.20220623.1608.040.html.
[13] 陈维刚, 张会林. 基于RF-LightGBM算法在风机叶片开裂故障预测中的应用[J]. 电子测量技术, 2020, 43(1): 162-168.
CHEN W G, ZHANG H L. Application of RF-LightGBM algorithm in early warning of fan blade cracking[J]. Electronic Measurement Technology, 2020, 43(1): 162-168.
[14] 夏宏泉, 张元泽, 陈平, 等. 碳酸盐岩地层破裂压力的测井预测研究[J]. 天然气工业, 2004(8): 32-35.
XIA H Q, ZHANG Y Z, CHEN P, et al. Log prediction of carbonate formation breakdown pressure[J]. Natural Gas Industry, 2004(8): 32-35.
[15] 李昌盛, 宋海, 肖莉, 等. 基于遗传算法优化BP神经网络的地层破裂压力预测方法[J]. 西安石油大学学报(自然科学版), 2015, 30(5): 75-79,10.
LI C S, SONG H, XIAO L, et al. Prediction method of formation fracture pressure based on BP neural network optimized by genetic algorithm( GA)[J]. Journal of Xi’an Shiyou University（Natural Science Edition）, 2015, 30(5): 75-79,10.
