Inversion for logging-while-drilling extra-deep azimuth electromagnetic measurement is an important technique to characterize formation parameter information.The inversion method for logging-while drilling electro-magnetic wave measurement based on regularization(physics driven)is widely used in field interpretation,but it needs to utilize forward modeling for many times in the iterative process,which takes a long time to calculate and fails to obtain real time inversion results.Therefore,an efficient inversion method is urgently needed for real-time inversion of logging while-drilling electromagnetic data.In recent years,a deep learning(data-driven)inversion algorithm of azimuthal logging-while drilling electromagnetic wave measurement has attracted widespread attention in the field of oil and gas exploration,but the algorithm relies too much on data,overlook-ing the Maxwell theory during the training process.Consequently,the effect of deep learning inversion is not good when the data set is not complete.In this paper,a hybrid inversion method coupling physics driven and data-driven methods is proposed for two-dimensional anisotropic formations:a network is trained using ran-domly generated datasets comprising models with and without faults,based on the data from logging while-drilling extra-deep azimuthal electromagnetic wave measurement;and model predictions are then made using the trained network.Compared with traditional deep learning methods,the prediction accuracy of the pro-posed method is significantly improved.Test results also show that,under the influence of different test noises,the physics-driven deep learning inversion method achieves favorable outcomes for resistivity models,exhibiting strong robustness and enhanced generalization ability.
关键词
物理驱动/超深探测/随钻测井/电磁场/参数反演
Key words
physics-driven/extra-deep detection/logging while drilling/electromagnetic response/parameter inversion
维普
万方数据