Study On Adaptive Inversion Method Of Azimuthal Electromagnetic Logging-While-Drilling

The azimuthal electromagnetic(AEM)logging-while-drilling(LWD)tool is able to detect the formation boundary within a few meters around the borehole due to its azimuth sensitivity,which makes it become a significant tool for the exploration and development of offshore oil and unconventional complex reservoirs.And the fast and accurate inversion of AEM LWD measurements is the key for real-time geosteering.However,the following problems still exist in data processing of AEM LWD:(1)Accuracy of gradient inversion algorithms heavily depend on initial value,which make they tend to fall into local minimum with the increase of inversion parameters.(2)The stochastic optimization algorithms need a large number of samplings to ensure the global convergence,so it is still necessary to further optimize and speed up to meet the needs of real-time data processing.(3)The formation model used in current inversion methods is fixed,which make it difficult to balance speed and accuracy.In this paper,we carry out researches on data processing methods for AEM LWD measurements.Firstly,we propose several inversion acceleration strategies applicable to different layered models and study its applicability.Then a novel inversion method that can adaptively select the number of model layers according to the drilling geological structure is developed.Thus the speed and accuracy of AEM LWD measurements processing have improved a lot,which will technically provide basis for real-time geosteering.In the second chapter,the detection performance,response characteristics and sensitivity of formation parameters in layered formation model based on AEM LWD tool are systematically studied.The results show that the charts of detection capability can be used to optimize the initial model.The azimuthal geosignals of the AEM LWD is very sensitive to the formation interface.As for the formation resistivity,anisotropy and relative dip,the response sensitivity gradually weakens,which provides a theoretical basis for the establishment of the inversion model and objective function.In the third chapter,we propose an optimized random initial model selection strategy for gradient algorithms,which make it capable of global optimization.Then it is combined with dynamic constraint of tool detection performance to avoid invalid initial value.And a dualinterface fast inversion method for AEM LWD is developed based on Levenberg-Marquardt algorithm,and it is proved two times faster and more accurate than traditional methods.In the fourth chapter,an improved PSO algorithm based on a parallel intervention mechanism is developed for the fine inversion of multi-layer models,which improves the algorithm’s global search capability,reduces the initial particle swarm size,and increase the convergence speed of the algorithm.Finally,an adaptive trans-layers optimization method is established after a comprehensive analysis of the applicability of the dual-interface real-time algorithm and the multi-interface fine algorithm.The numerical example results show that the adaptive trans-layers optimization method has good applicability and accuracy in anisotropic Oklahoma model with different thicknesses.