Drillability Prediction Method for Deep Undrilled Formation Based on SAE and LSTM Neural Network
It is very necessary to predict the drillability of the formation before drilling when formulating the scheme of drilling speed increase in deep formation.The existing rock drillability prediction models have low accuracy and are difficult to meet the requirements of drilling technology.Therefore,a combined model based on SAE and LSTMneural networks is proposed to predict the drillability of deep undrilled formations.The training time and prediction results of SAE-LSTMcombined model are compared with those of BP neural network,support vector machine,random forest and single LSTMmodels.The results show that the SAE-LSTMcombined model has the shortest training time and the smallest error between the predicted value and the actual measured value.The root mean square error RMSE of fitting result is only 0.081,the average absolute percentage error MAPE is 1.189,and the determination coefficient R2is 0.966,with the smallest RMSE and MAPE and the largest R2.The SAE-LSTMcombined model has higher prediction accuracy than oth-er models.This method brings a new way to the prediction of formation parameters,and can improve the problems of low prediction effi-ciency and low prediction accuracy of the previous prediction methods in dealing with complex formations.
deep formation drillingrock drillabilityprediction modelstacked autoencoderLSTM neural networkdeep learning
万方数据
维普
