Research On Prediction Of Drilling Point Geological Information Based On Drilling And Logging While Drilling Data

Non-renewable resources such as oil and natural gas are of great significance to the survival and development of mankind.Drilling is a vital part of the exploration and production of oil and natural gas resources.How to improve drilling efficiency and reduce drilling cost has become a hot issue in today’s oil field.If the geological information of the formation can be detected in real-time during the drilling process,various strategic adjustments can be made based on the geological characteristics of the formation,thereby achieving geological guided drilling,improving drilling efficiency and reducing drilling costs.However,conventional logging while drilling tools do not detect the position of the drill bit,but rather at a distance of several meters from the drill bit.This results in a lack of real-time data collection and a delay in time,which can affect the entire drilling decision and the entire drilling process of drilling guidance.In order to solve this problem,this paper proposes a prediction method of drilling point geological information based on drilling and logging while drilling data on the basis of real-time engineering data at the bit,and uses deep learning method to predict and correct the lagging geological information measured by logging while drilling instruments.By predicting the geological information from the detection point to the drilling point,the lag problem of inconsistent geological information was solved,and the method was tested and evaluated with actual data.The main research contents of this paper are as follows:(1)In this paper,based on the timing characteristics of drilling engineering parameters and logging while drilling data,the deep learning sequence model is introduced and the RNN and LSTM models are constructed.On top of this,in order to improve the robustness of the model to noise and enhance the generalization ability of the model,this paper introduces the selfattention mechanism,adjusts part of the structure of the Transformer model,and applies it to this time sequence task.In addition,through comparison and analysis of actual data,this paper proposes a model structure combining MLP and Transformer based on the characteristics of drilling engineering parameters and logging while drilling data,which can achieve higher prediction accuracy than conventional methods.(2)The time step analysis experiment was designed,and a reasonable sliding window was set for the input data of the four models,so as to ensure the rationality of the comparison experiment between the subsequent models.In the actual data test,in order to meet the actual location differences of different logging instruments while drilling,the experiment adopted the multi-output strategy,the output sample step length is 10 steps,the total depth span is 5 meters.The experimental results show that the input data time step values of RNN,LSTM,Transformer and MLP-Transformer models are reasonable when they are 30,60,40 and 60 respectively.(3)Based on the results of the time step analysis experiment,the comparison experiment between the models was designed.Among them,the prediction accuracy of MLP-Transformer model is the best.MAPE of the model in 10 prediction steps is all less than 0.18,indicating that the model can accurately predict geological information 5 meters down.In addition,the comprehensive RMSE,MAE and MAPE of MLP-Transformer model in 10 prediction steps are20.7458,14.5525 and 0.1453,respectively.Compared with RNN,LSTM and Transformer model,the prediction accuracy is higher and the prediction performance is better.It can solve the lag problem of physical equipment detection to a certain extent,and provide a more scientific basis for the follow-up research of drilling point geological information prediction.