石油地球物理勘探2025，Vol.60Issue(3)：630-642,13.DOI:10.13810/j.cnki.issn.1000-7210.20240291
基于自适应损失均衡梯度增强的物理信息神经网络微地震定位
Adaptive-loss-weighting gradient-enhanced physics-informed neural network for microseismic localization
摘要
Abstract
Microseismic localization is a major challenge in microseismic monitoring tasks and is helpful for analy-zing the effect of hydraulic fracturing.Physics-informed neural network(PINN)can achieve microseismic loca-lization.However,the trade-off among multiple loss terms plays a crucial role in the training stage of PINN.Thus,this paper proposes a novel microseismic localization method based on an adaptive-loss-weighting gradient-enhanced PINN.First,a combined loss function is constructed by integrating the residuals of relative arrival time and the eikonal equation.Second,an adaptive term is introduced to automatically update the loss weights,and gradient information is also incorporated to enhance the performance of the network.Finally,network training is performed to obtain the traveltime distribution across the computational domain,and the hypocenter location is predicted by identifying the minimum traveltime.Test results demonstrate that this method can enhance the trai-ning stability and prediction accuracy of the network and achieve a reliable microseismic localization effect.关键词
微地震/物理信息神经网络/相对到时/程函方程/自适应损失均衡梯度增强Key words
microseismic/physics-informed neural network/relative arrival time/eikonal equation/adaptive-loss-weighting gradient-enhanced分类
地质学引用本文复制引用
潘登,唐杰,范忠豪,产嘉怡,彭婧妍..基于自适应损失均衡梯度增强的物理信息神经网络微地震定位[J].石油地球物理勘探,2025,60(3):630-642,13.基金项目
本项研究受国家自然科学基金项目"基于石油勘探面波与P-导波的近地表纵横波速度一体化反演"(42174140)资助. （42174140）
