A spatiotemporal surrogate model and multi-objective optimization for multi-well coalbed methane production

Authors

• QIN Yuancheng, Gas Field Company, Shaanxi Yanchang Petroleum (Group) Co., Ltd., Yan’an 716000, ChinaFollow
• WANG Mengyi, Gas Field Company, Shaanxi Yanchang Petroleum (Group) Co., Ltd., Yan’an 716000, China
• WANG Yanfeng, Gas Field Company, Shaanxi Yanchang Petroleum (Group) Co., Ltd., Yan’an 716000, China
• LI Yuan, Gas Field Company, Shaanxi Yanchang Petroleum (Group) Co., Ltd., Yan’an 716000, China
• HE Bo, Gas Field Company, Shaanxi Yanchang Petroleum (Group) Co., Ltd., Yan’an 716000, China
• FANG Junling, Tracy Energy Technologies Co., Ltd., Hangzhou 310016, China
• DUAN Qingshan, School of Earth Resources, China University of Geosciences (Wuhan), Wuhan 430074, ChinaFollow
• LIU Chen, School of Earth Resources, China University of Geosciences (Wuhan), Wuhan 430074, China

Abstract

Objective and Methods In the full-lifecycle optimization of multi-well and long-term coalbed methane (CBM) production, conventional numerical simulation methods face challenges including extremely high computational costs and difficult multi-objective trade-offs. To address these issues, this study proposed a spatiotemporal surrogate model for multi-objective optimization of CBM production. First, a heterogeneous reservoir model involving 17 production wells was constructed based on high-fidelity numerical simulation, and a sample dataset was created by applying random perturbations to the baseline production scheme. Second, a hybrid surrogate model integrating a convolutional neural network (CNN) and a Transformer decoder was established. The CNN was utilized for the global encoding of spatial reservoir heterogeneity, such as porosity and permeability. The encoding, combined with the cross-attention mechanism and causal masking of the Transformer decoder, enabled non-autoregressive parallel prediction from full-lifecycle well-controlled sequences to gas production responses. Finally, the surrogate model was embedded into the non-dominated sorting genetic algorithm II (NSGA-Ⅱ) with an elite strategy to optimize the production system across the study area within a 323-dimensional decision space. Results The constructed surrogate model achieved a coefficient of determination (R²) of up to 0.997 on the testing set. This model exhibited an average one-shot inference time of approximately 7.17 ms, representing an acceleration of about 2.5×10⁴ times compared to the time required to conduct a single numerical simulation (approximately 180 s). In the batch inference mode of NSGA-Ⅱ population evaluation, the average one-shot inference time of the model was reduced to approximately 0.344 ms. The Pareto front derived from the multi-objective optimization showed a uniform distribution, clearly reflecting the nonlinear trade-off relationship between total water drainage and total gas production. Furthermore, the representative knee points effectively avoided ineffective water production while ensuring high gas production benefits. Conclusions The entire architecture overcomes the limitations of the conventional empirical gas production control based on a single threshold while also demonstrating that a data-driven spatiotemporal surrogate model can effectively identify the nonlinear characteristics of multi-well collaborative production. The proposed architecture can be widely applied to the intelligent management and control platforms for CBM fields, providing efficient support for real-time dynamic production decision-making. Future research will focus on the characterization of geological parameter uncertainties and the development of closed-loop production control systems.

Keywords

coalbed methane (CBM), production optimization, surrogate model, Transformer, convolutional neural network (CNN), multi-objective optimization, non-dominated sorting genetic algorithm II (NSGA-Ⅱ)

DOI

10.12363/issn.1001-1986.26.04.0267

Recommended Citation

QIN Yuancheng, WANG Mengyi, WANG Yanfeng, et al. (2026) "A spatiotemporal surrogate model and multi-objective optimization for multi-well coalbed methane production," Coal Geology & Exploration: Vol. 54: Iss. 9, Article 22.
DOI: 10.12363/issn.1001-1986.26.04.0267
Available at: https://cge.researchcommons.org/journal/vol54/iss9/22

Reference

[1] 陈春琳,叶贵均. 试论煤储层气–水两相流系统[J]. 煤田地质与勘探,2003,31(5):23−26. CHEN Chunlin,YE Guijun. Double–phase liquid system of gas–water in coal reservoir[J]. Coal Geology & Exploration,2003,31(5):23−26.

[2] 秦勇,申建. 论深部煤层气基本地质问题[J]. 石油学报,2016,37(1):125−136. QIN Yong,SHEN Jian. On the fundamental issues of deep coalbed methane geology[J]. Acta Petrolei Sinica,2016,37(1):125−136.

[3] 马东民. 煤层气吸附解吸机理研究[D]. 西安: 西安科技大学, 2008. MA Dongmin. Research on the adsorption and desorption mechanism of coalbed methane[D]. Xi'an: Xi'an University of Science and Technology, 2008.

[4] 夏阳,朱敏,刘玉石,等. 基于人工智能的煤层气排采制度优化方法[J]. 中国煤层气,2025,22(3):37−39. XIA Yang,ZHU Min,LIU Yushi,et al. Artificial intelligence–based optimization method for coalbed methane drainage systems[J]. China Coalbed Methane,2025,22(3):37−39.

[5] 彭守建,许江,陶云奇,等. 煤样渗透率对有效应力敏感性实验分析[J]. 重庆大学学报,2009,32(3):303−307. PENG Shoujian,XU Jiang,TAO Yunqi,et al. Coefficient of sensitiveness between permeability and effective pressure of coal samples[J]. Journal of Chongqing University,2009,32(3):303−307.

[6] SHI J Q,DURUCAN S. Drawdown induced changes in permeability of coalbeds: A new interpretation of the reservoir response to primary recovery[J]. Transport in Porous Media,2004,56(1):1−16.

[7] 李金海,苏现波,林晓英,等. 煤层气井排采速率与产能的关系[J]. 煤炭学报,2009,34(3):376−380. LI Jinhai,SU Xianbo,LIN Xiaoying,et al. Relationship between discharge rate and productivity of coalbed methane wells[J]. Journal of China Coal Society,2009,34(3):376−380.

[8] 侯世辉. 煤层气井排采过程中不同煤体结构储层渗透率动态变化特征研究[D]. 武汉: 中国地质大学, 2018. HOU Shihui. Dynamic variation characterization of permeability of different coal structure reservoir during depletion[D]. Wuhan: China University of Geosciences, 2018.

[9] PALMER I,MANSOORI J. How permeability depends on stress and pore pressure in coalbeds: A new model[J]. SPE Reservoir Evaluation & Engineering,1998,1(6):539−544.

[10] 王彦迪. 基于改进遗传算法的煤层气井网优化研究[D]. 徐州: 中国矿业大学, 2020. WANG Yandi. Coalbed methane well pattern optimization based on improved genetic algorithm[D]. Xuzhou: China University of Mining and Technology, 2020.

[11] GUYAGULER B,HORNE R N,ROGERS L,et al. Optimization of well placement in a gulf of Mexico waterflooding project[J]. SPE Reservoir Evaluation & Engineering,2002,5(3):229−236.

[12] ZUBAREV D I. Pros and cons of applying proxy–models as a substitute for full reservoir simulations[C]//SPE Annual Technical Conference and Exhibition. New Orleans: Society of Petroleum Engineers,2009: SPE–124815–MS.

[13] 徐慧. 粒子群优化算法改进及其在煤层气产能预测中的应用研究[D]. 徐州: 中国矿业大学, 2013. XU Hui. Research on particle swarm optimization algorithm improvement and its application in the CBM production forecast[D]. Xuzhou: China University of Mining and Technology, 2013.

[14] 谢坤,田轩硕,刘长龙,等. 机器学习方法预测油气产量技术发展现状及前景展望[J]. 特种油气藏,2025,32(4):14−24. XIE Kun,TIAN Xuanshuo,LIU Changlong,et al. Status and prospects of machine learning methods for predicting hydrocarbon production[J]. Special Oil & Gas Reservoirs,2025,32(4):14−24.

[15] 吴淑红,毕剑飞,范天一,等. 油藏数值模拟智能化技术思考与展望[J]. 石油科技论坛,2024,43(6):96−106. WU Shuhong,BI Jianfei,FAN Tianyi,et al. Thoughts and prospects on the intelligent transformation of reservoir simulation[J]. Petroleum Science and Technology Forum,2024,43(6):96−106.

[16] ALIREZA S,SHAHAB M. 智能代理在油藏建模中的应用[J]. 石油勘探与开发,2020,47(2):372−382. ALIREZA S,SHAHAB M. Applications of smart proxies for subsurface modeling[J]. Petroleum Exploration and Development,2020,47(2):372−382.

[17] 李卓,刘斌,刘铁男. 支持向量机在油田产量预测中的应用[J]. 大庆石油学院学报,2005,29(5):96−97. LI Zhuo,LIU Bin,LIU Tienan. Application of supportive vector machine to the prediction of production in oil fields[J]. Journal of Daqing Petroleum Institute,2005,29(5):96−97.

[18] KLEIJNEN J P C. Kriging metamodeling in simulation: A review[J]. European Journal of Operational Research,2009,192(3):707−716.

[19] 付天予,胡慢谷,张骁俊,等. 基于TLE–UNet的高温水冷岩石裂纹分割与特征提取方法[J]. 煤田地质与勘探,2026,54(2):172−182. FU Tianyu,HU Mangu,ZHANG Xiaojun,et al. TLE–UNet–based image segmentation and feature extraction of water–cooling–induced complex cracks in high–temperature rock masses[J]. Coal Geology & Exploration,2026,54(2):172−182.

[20] 梁潇,喻高明,辛显康,等. 基于GRU循环神经网络的稠油油藏产量预测新方法[J]. 西安石油大学学报(自然科学版),2020,35(3):54−59. LIANG Xiao,YU Gaoming,XIN Xiankang,et al. A new method for production prediction of heavy oil reservoirs based on GRU circulation neural network[J]. Journal of Xi'an Shiyou University (Natural Science Edition),2020,35(3):54−59.

[21] 高亚军,唐力辉,王振鹏,等. 基于循环神经网络和数据差分处理的油田产量预测方法[J]. 中国海上油气,2023,35(3):126−136. GAO Yajun,TANG Lihui,WANG Zhenpeng,et al. A method for oilfield production prediction based on recurrent neural network and data differential processing[J]. China Offshore Oil and Gas,2023,35(3):126−136.

[22] 马承杰. 基于循环神经网络的油藏产量预测方法[J]. 内蒙古石油化工,2021,47(5):108−112. MA Chengjie. The prediction method of reservoir production based on recurrent neural network method[J]. Inner Mongolia Petrochemical Industry,2021,47(5):108−112.

[23] 赵誉兴,李向文. Bi–RNN与GMS耦合的金属露天矿井涌水量预测[J]. 煤田地质与勘探,2024,52(12):155−169. ZHAO Yuxing,LI Xiangwen. Predicted water yield of open–pit metal mines based on a Bi–RNN and GMS coupling model[J]. Coal Geology & Exploration,2024,52(12):155−169.

[24] 韩江峡,薛亮,位云生,等. 基于深度自回归神经网络的多井产量概率预测[J]. 石油科学通报,2024,9(4):679−689. HAN Jiangxia,XUE Liang,WEI Yunsheng,et al. Multiple well production rate probabilistic forecasting using deep autoregressive recurrent networks[J]. Petroleum Science Bulletin,2024,9(4):679−689.

[25] 贾靖,樊庆虎,王李昌,等. 低效井重复压裂产量深度时间序列预测方法综述[J]. 绿色矿山,2025,3(1):14−36. JIA Jing,FAN Qinghu,WANG Lichang,et al. Prediction of post–refracture production of low–productivity wells using deep time series models: A critical review[J]. Journal of Green Mine,2025,3(1):14−36.

[26] LI Xianxian,LI Xijian,XIE Honggao,et al. Enhanced coalbed methane well production prediction framework utilizing the CNN–BL–MHA approach[J]. Scientific Reports,2024,14:14689.

[27] HAN Jiangxia,XUE Liang,WEI Yunsheng,et al. Physics–informed neural network–based petroleum reservoir simulation with sparse data using domain decomposition[J]. Petroleum Science,2023,20(6):3450−3460.

[28] WANG Lian,YAO Yuedong,WANG Kongjie,et al. A novel surrogate–assisted multi–objective optimization method for well control parameters based on tri–training[J]. Natural Resources Research,2021,30(6):4825−4841.

[29] 王欣,吴晓茹,李杰,等. 基于Transformer和产量递减模型的井口压力智能预测方法[J]. 中国海上油气,2025,37(6):80−89. WANG Xin,WU Xiaoru,LI Jie,et al. Intelligent prediction method of wellhead pressure based on Transformer and production decline model[J]. China Offshore Oil and Gas,2025,37(6):80−89.

[30] ZHANG Jinding,KANG Jinzheng,ZHANG Kai,et al. An efficient transformer–based surrogate model with end–to–end training strategies for automatic history matching[J]. Geoenergy Science and Engineering,2024,240:212994.

[31] 刘军,任静思,胡南,等. 基于Transformer模型的页岩气井积液预测方法[J]. 天然气勘探与开发,2025,48(6):104−117. LIU Jun,REN Jingsi,HU Nan,et al. A Transformer–model–based method for predicting liquid loading in shale gas wells[J]. Natural Gas Exploration and Development,2025,48(6):104−117.