Intelligent Prediction Method Of Overflow Risk Based On Drill-Logging Data

Deep and ultra-deep oil and gas resources are important areas for increasing oil and gas reserves and production,which account for 34% of total oil and gas resources in China,and they are also the important basis to ensuring national energy security.However,they pose greater challenges to oil and gas exploration and development technologies with complex geological structure and formation pressure,narrow safe density window,and are easy to occur risks such as overflow during the drilling process.For the problem such as the low accuracy of traditional methods to calculate formation pore pressure and the poor timeliness of overflow risk prediction,this thesis carried out research on intelligent prediction of formation pore pressure and overflow risk based on machine learning and deep learning.First,use big data methods to clean and process data such as drilling and logging to establish high-quality data sets.Considering the sequence of formation deposition and drilling process,a series model of Long-Short Time Memory(LSTM)and Back Propagation Neuron Network(BP)was constructed.The models were trained and tested by drilling-logging data,and a real-time calculation model of formation pore pressure was established.Compared with the LSTM and BP models,the results show that LSTMBP performs best,with the mean relative error of 2.76% for the single well model and2.35% for the offset well model.Secondly,a real-time advance prediction model of formation pore pressure was developed based on logging while drilling data and dual-stage attention-based recurrent neural network(DA-RNN),and compared with the LSTM model.The accuracy and computational efficiency of the LSTM model are slightly higher than that of the DA-RNN model,but the DA-RNN model is better than the LSTM model in terms of model stability and effectiveness.Finally,combining with the random forest and XGBoost algorithm,the overflow risk diagnosis model is established based on the drilling data and formation pore pressure data.The results show that the XGBoost model performs better,with an accuracy rate of97.13%.And the overflow risk can be predicted in advance by entering the current depth drilling data and the formation pore pressure data of the next 1m into the diagnostic model.Among them,the XGBoost model has the best prediction effect,with an accuracy rate of90.45%,missed alarm rate of about 12.5%,and false alarm rate of 9.14%.In addition,relative importance analysis shows that formation pore pressure and total hydrocarbons have the greatest impact on model performance.The research results can accurately predict the formation pore pressure and overflow risk,and provide a reference for safe and efficient drilling and the adjustment of drilling plans.