基于机器学习的地热采灌方案优化方法  

Optimizing development parameters of geothermal energy using machine learning technique

作　　者：王佳铖 陈进帆 赵志宏[1] 谭现锋 WANG Jiacheng;CHEN Jinfan;ZHAO Zhihong;TAN Xianfeng(Department of Civil Engineering,Tsinghua University,Beijing 100084,China;Shandong Lunan Geological Engineering Investigation Institute,Jining272100,China)

机构地区：[1]清华大学土木工程系,北京100084 [2]山东省鲁南地质工程勘察院,济宁272100

出　　处：《科技导报》2022年第20期93-100,共8页Science & Technology Review

基　　金：国家重点研发计划项目(2019YFB1504103)。

摘　　要：为了解决基于数值模拟的优化方法,通常需要大量模拟计算的问题,在比较不同机器学习方法的预测性能后,建立了基于多层感知机的代理模型,以降低计算成本,然后将其与遗传算法相结合,提出了非均质地热储层中地热对井系统采灌方案优化方法。通过地热田对井系统的案例研究,证明了所开发的采灌方案优化方法的合理性和有效性。结果表明,基于代理模型的采灌方案优化方法能以更低的计算成本,准确地找到给定开采井位置时最优的回灌井位置、采灌量和尾水温度。In order to solve the problem that the simulation-based optimization method usually requires a large number of simulations,in this paper,after comparing the prediction performance of different machine learning methods,a surrogate model based on MLP was developed to reduce the computational cost,which was then combined with genetic algorithm to develop an optimization method of development parameters for geothermal doublets in heterogeneous geothermal reservoirs.Through the case study of a doublet system,the reasonability and efficiency of the developed optimization method of development parameters were demonstrated.The results show that when given a certain position of production well,the surrogate model-based optimization method of development parameters can accurately find the optimal placement of injection well,the rate of production and injection,and the temperature of recharge water with lower computational cost.

关 键 词：地热对井系统:机器学习 采灌方案优化:代理模型 遗传算法 

分 类 号：P314[天文地球—固体地球物理学]