Research On A Three-dimensional Wellbore Trajectory Intelligent Steering Method Based On Deep Learning

Geosteering is an interactive drilling technology consisting of high-precision drill bits,steering decision software,measurement while drilling and logging while drilling technology,and power steering tools,which is in widespread application for exploration of petroleum,natural gas and other liquid and gaseous minerals.The growing depth of resource burial causes the growth of stratigraphic complexity and drilling difficulty,which leads to the decrease of drilling efficiency.With the advancement of drilling technology,well-drilling technology is in the direction of automation,digitalization and intelligence.Intelligent drilling is a drilling technology that uses intelligent decision models to guide and thus achieve autonomous closed-loop control,and drilling visualization technology can intuitively display the formation structure and borehole trajectory.In this context,this thesis proposes a 3D wellbore trajectory intelligent steering method based on deep learning for geosteering drilling decision and 3D visualization.The main contributions of this thesis are as follows:Firstly,due to the existence of a large number of redundant features in the original logging data,this thesis combines random forest and recursive feature elimination algorithm to perform feature selection on stratigraphic data.In addition,for the drilling decision-making problem in 3D scenarios,this thesis designs a steerable decision rule for labeling training data,and proposes a 3D wide-angle detection mechanism to expand the spatial perception range of data while drilling.Secondly,aiming at the real-time steering decision-making of 3D borehole trajectory in complex and special geological scenarios,this thesis proposes an intelligent steering decision-making algorithm based on logging while drilling data.The asymmetric depthwise separable convolution model and peephole mechanism are introduced in the LSTM unit for extracting spatio-temporal features of formation data,and the selfattention mechanism is incorporated for key feature extraction,which provides an efficient and accurate steering decision-making method for drilling process.Finally,for the 3D visualization of formation and borehole trajectory,this thesis designs and implements an interactive intelligent drilling simulation system for simulating drilling process and interface interaction.In addition,in order to solve the problem of insufficient formation data samples,this thesis proposes a time-series formation data generation model by combining GAN and LSTM models to construct a simulated drilling environment.Ablation experiments and comparison experiments were performed with other models on the own data set and public data set.The experimental results show the drilling accuracy rate of the model in this thesis is increased by about from 2% to 4%,and the drilling encounter rate is increased by about from 3% to 10%,which verifies the advancedness of the method in this thesis.Finally,the drilling simulation visualization process is demonstrated to further demonstrate the effectiveness of intelligent steering decision-making algorithm from visualization.