Interpretable Real-Time Prediction of Drilling Parameters Based on Improved Sequential Network
Real-time accurate prediction on variation trend of drilling parameters has important reference value for field drilling operations.In order to solve the limitations of drilling parameter availability faced by intelligent model in field operation,a drilling parameter prediction method based on Attention-Temporal Convolutional Net-work(AT-TCN)was proposed.This method not only takes into account the variation trend of mud logging curve with depth and its autocorrelation,but also embeds a highly expansible attention mechanism module,allowing the model to better capture the dynamic change of drilling parameters.Then,the field drilling data set was used to test and evaluate the effectiveness and accuracy of the model in predicting four key drilling parameters such as torque,standpipe pressure,equivalent density of drilling fluid and ROP.The research results show that AT-TCN can pre-dict equivalent density with an accuracy up to 99%.It is superior to the other four deep learning models in terms of model accuracy and computational efficiency,and it can effectively capture the variation trend of drilling parame-ters.AT-TCN also provides dual interpretability of the model,and reflects the influence of input sequence on pre-diction results from both sequential and characteristic dimensions.The research results are expected to make impor-tant contributions to the safety and efficiency of drilling operations,and have strong practical application value.
万方数据
维普
