Research On Intelligent Guided Method And Interactive Drilling Simulation Method Based On Deep Learning

Due to the complexity and invisibility of the stratum,the drilling operation heavily relies on the two-way signal transmission on the well and underground,as well as the experience and decision-making of human experts.However,for complex well conditions such as deep well and ultra-deep well,high-quality real-time signal transmission is difficult to achieve,and the rock mechanics model based on theory is not enough to support real-time decision-making.Therefore,this thesis proposes an intelligent geosteering method based on deep learning,which is expected to provide a new idea for real-time decision making of geological guidance.To this end,the main works are as follows:a)Propose a method of perception enhancement based on feature selection and wide-angle eye mechanism.This method can improve the data quality and reduce the online computing cost.And the method use the wide-angle eye mechanism to broaden the drill’s field of vision to obtain the surrounding geological information and enhance the drill’s perception ability;b)Propose an Asymmetric Peephole Connection Long Short-Term Memory(APCLSTM)approach for geosteering drilling decision.APC-LSTM can capture the spatial-temporal correlation between different strata for drilling decision making.Contrasted with existing methods,proposed model achieved better performance.c)In order to verify the accuracy of the proposed model,we design a simulation drilling interaction mechanism based on generative adversarial networks.This mechanism can generate simulation logging data according to the data of bit location,and simulate the interactive drilling process of bit in actual operation.The simulation test is carried out in the simulated drilling environment to verify the actual performance of the model.The experimental result indicates that the proposed model has high prediction accuracy.Moreover,it has achieved good results in the simulation environment.