首页|基于混合优化算法和深度神经网络模型结合的致密砂岩气藏裂缝参数优化

基于混合优化算法和深度神经网络模型结合的致密砂岩气藏裂缝参数优化

扫码查看
原文链接
  • 万方数据
  • 维普
水平井分段压裂是致密砂岩气藏的主要开发方式,其中水力压裂裂缝参数的合理设计对于气藏的经济效益开发至关重要.基于群智能优化算法和机器学习代理模型的自动优化方法存在所需数值模拟次数多、收敛速度慢和代理模型更新复杂等问题,且依靠现场工程师经验和正交实验等传统方法难以获得最佳的裂缝参数设计.为此,建立了一种新的基于混合优化算法和自适应深度神经网络(DNN)结合的致密气藏裂缝参数优化方法.首先,混合优化算法采用遗传算法(GA)和贝叶斯自适应直接搜索(BADS)之间循环迭代的混合策略.在自适应学习过程中,提出了以"最大平均距离点"作为最不确定解,同时辅以最有希望解和少量拉丁超立方采样解共同更新优化过程中的DNN代理模型.随后,将建立的优化方法用于非均质致密砂岩气藏裂缝参数优化.研究结果表明:①在标准测试函数和低维裂缝参数优化问题上,GA+BADS混合优化算法表现出了显著优于GA的寻优速度;②针对高维裂缝参数优化问题,GA+BADS混合优化算法在约1/2的GA总数值模拟次数下提高了 131万元的经济净现值(NPV),收敛速度和寻优精度都明显增加;③相比于GA+BADS混合优化算法,在获得相同NPV时,自适应DNN代理加速优化可再减少24.54％的数值模拟运算次数.结论认为,该优化方法显著提升了优化效率,为解决非常规油气藏中水力压裂裂缝参数设计问题提供了一套可行且高效的智能优化方法,将有力促进非常规油气的规模效益开发.
Fracture parameter optimization of tight sandstone gas reservoirs based on the hybrid optimization algorithm and deep neural network model
Horizontal well staged fracturing is a main development mode used for tight sandstone gas reservoirs,and the reasonable design of hydraulic fracture parameters is crucial for the cost effective development of such reservoirs.The automatic optimization methods based on swarm intelligence optimization algorithm and machine learning surrogate model often face problems such as numerous numerical simulations,slow convergence,and complex update of surrogate model.In addition,it is difficult to obtain the best fracture parameter design by traditional methods such as on-site engineer experience and orthogonal experiment.In order to solve these problems,this paper establishes a new method for optimizing fracture parameters of tight sandstone gas reservoirs based on hybrid optimization algorithm and adaptive deep neural network(DNN).First,a mixed strategy of cyclic iteration between genetic algorithm(GA)and Bayesian adaptive direct search(BADS)is adopted in the hybrid optimization algorithm.In the process of adaptive learning,the"maximum average distance point"is proposed to be the most uncertain solution,which,together with the most promising solution and a small number of Latin hypercube sampling solutions is used to update the DNN surrogate model being optimized.Subsequently,the established optimization method is applied to optimize the fracture parameters of heterogeneous tight gas reservoirs.And the following research results are obtained.First,in standard test function and low-dimensional fracture parameter optimization,GA+BADS hybrid optimization algorithm exhibits much higher optimization speed than GA.Second,in high-dimensional fracture parameter optimization,GA+BADS hybrid optimization algorithm improves the economic net present value(NPV)by CNY1.31 million in about half of the total GA numerical simulation times,and achieves a significant increase in convergence speed and optimization accuracy.Third,compared with GA+BADS hybrid optimization algorithm,adaptive DNN surrogate accelerated optimization can further reduce the number of operations in the numerical simulation by 24.54％when obtaining the same NPV.In conclusion,the proposed optimization method improves the optimization efficiency significantly,provides a feasible and efficient intelligent optimization process in solving the problems in the parameter design of hydraulic fractures in unconventional oil and gas reservoirs,and will powerfully promote the large-scale benefit development of unconventional oil and gas.

Tight gasShaximiao FmFracture parameter optimizationHybrid optimization algorithmDeep neural networkAdaptive learningSurrogate model

罗山贵、赵玉龙、肖红林、陈伟华、贺戈、张烈辉、杜诚

展开 >

西南石油大学理学院

西南石油大学人工智能研究院

油气藏地质及开发工程全国重点实验室·西南石油大学

中国石油西南油气田公司致密油气勘探开发项目部

中国石油西南油气田公司工程技术研究院

中国石油西南油气田公司川西北气矿

展开 >

致密气 沙溪庙组 裂缝参数优化 混合优化算法 深度神经网络 自适应学习 代理模型

国家自然科学基金优秀青年科学基金项目四川省杰出青年科技人才项目国家自然科学基金重点项目

522224022022JDJQ000952234003

2024

10.3787/j.issn.1000-0976.2024.09.013

天然气工业
四川石油管理局 中国石油西南油气田公司 中国石油川庆钻探工程公司

天然气工业

CSTPCD北大核心EI
影响因子：2.298
ISSN：1000-0976
年,卷(期)：2024.44(9)
  • 11