深度强化学习驱动的压裂参数多目标优化方法
Multi-objective optimization method for fracturing parameters driven by deep reinforcement learning
浏览(5723) 下载(23)
- 引用格式:
-
张学敏,魏莉,杨丽娜.深度强化学习驱动的压裂参数多目标优化方法[J].天然气与石油,2025,43(6):145-151.doi:10.3969/j.issn.1006-5539.2025.06.018
ZHANG Xuemin, WEI Li, YANG Lina.Multi-objective optimization method for fracturing parameters driven by deep reinforcement learning[J].Natural Gas and Oil,2025,43(6):145-151.doi:10.3969/j.issn.1006-5539.2025.06.018
- DOI:
- 10.3969/j.issn.1006-5539.2025.06.018
- 作者:
- 张学敏1,2 魏莉1,2 杨丽娜1,2
ZHANG Xuemin1,2, WEI Li1,2, YANG Lina1,2
- 作者单位:
- 1. 海洋油气勘探国家工程研究中心, 北京 100028; 2. 中海油能源发展股份有限公司工程技术分公司, 天津 300452
1. National Engineering Reasearch Center of Offshore and Gas Exploration, Beijing, 100028, China; 2. CNOOC EnerTech-Drilling & Production Co., Tianjin, 300452, China
- 关键词:
- 深度强化学习;压裂参数优化;多目标决策;地质可行性指数;页岩气开发
Deep Reinforcement Learning; Fracturing parameter optimization; Multi-objective decision-making; Geological feasibility index; Shale gas development
- 摘要:
针对页岩气水平井压裂设计中地质非均质性显著、工程约束复杂、经济目标冲突的难题,提出一种基于深度强化学习驱动的压裂参数多目标优化方法。通过构建融合地质—工程—经济三域特征参数状态空间,设计融合NSGA-Ⅲ算法改进的奖励函数机制,并结合首创的地质可行性指数量化模型,实现压裂参数的智能优化。基于A区块18口水平井压裂数据的实验表明,相比传统粒子群优化算法,基于深度强化学习驱动的压裂参数多目标优化方法使估算最终采收率提升12.7%,单井成本降低9.3%,裂缝复杂度指数提高41%,净现值标准差减少27%。研究成果可为智能压裂决策提供理论支撑,推动地质工程一体化技术发展。
To address the challenges of significant geological heterogeneity, complex engineering constraints, and conflicting economic objectives in shale gas horizontal well fracturing design, this paper proposes a multi-objective optimization method for fracturing parameters driven by deep reinforcement learning. By constructing a state space integrating geological, engineering, and economic domain features, designing a reward function mechanism enhanced by the NSGA-Ⅲ algorithm, and incorporating an innovative “geological feasibility index” quantification model, intelligent optimization of fracturing parameters is achieved. Experiments using fracturing data from 18 horizontal wells in Block A demonstrate that, compared to the traditional particle swarm optimization algorithm, the proposed method increases estimated ultimate recovery by 12.7%, reduces single-well costs by 9.3%, improves the fracture complexity index by 41%, and decreases the standard deviation of the net present value by 27%. This research provides theoretical support for intelligent fracturing decision-making and promotes the development of integrated geological and engineering technologies.
