成都理工大学学报（自然科学版）2025，Vol.52Issue(5)：900-913,14.DOI:10.12474/cdlgzrkx.2025071801
基于机器学习与遗传算法的地热储层渗透率智能反演研究
Research on intelligent inversion of geothermal reservoir permeability based on machine learning and genetic algorithm
摘要
Abstract
Fitting the production dynamics of geothermal reservoirs is a crucial technique for inverting the reservoir permeability distribution.Traditional inversion methods require repeated numerical model simulations for trial-and-error searches,leading to high computational costs and long processing times.To address this,we propose an intelligent inversion approach that integrates machine learning surrogate models with a genetic algorithm to characterize geothermal reservoirs efficiently.Specifically,we developed intelligent prediction models using the random forest algorithm for homogeneous,anisotropic,and heterogeneous permeability reservoirs,which serve as surrogates for numerical simulations at various production stages.The surrogate models significantly improve computational efficiency while maintaining high accuracy,with prediction errors below 5%.The genetic algorithm estimates the distributions of permeability parameters based on three production indicators:production temperature,output thermal power,and accumulative extracted thermal energy.An uncertainty analysis confirmed the feasibility of the inversion model across all three reservoir types.The posterior model has reduced the dispersion of production data by more than 76%compared to the prior model,demonstrating good history-matching results.Additionally,we compared and analyzed the inversion accuracy of different permeability distribution parameters.This study contributes to the advancement of geothermal reservoir inversion theory.关键词
地热能/非均质渗透率/智能代理模型/智能反演方法/不确定性Key words
geothermal energy/heterogeneous permeability/intelligent surrogate model/intelligent inversion method/uncertainty analysis分类
能源科技引用本文复制引用
李爽,宋先知,崔启亮,王高升,杨睿月,于超..基于机器学习与遗传算法的地热储层渗透率智能反演研究[J].成都理工大学学报（自然科学版）,2025,52(5):900-913,14.基金项目
国家自然科学基金(52125401,52304057). （52125401,52304057）
