Intelligent Prediction of Drilling Rate of Penetration Based on Method-Data Dual Validity Analysis

Abstract

Summary The rate of penetration (ROP) is a critical parameter in drilling operations, essential for optimizing the drilling process and enhancing drilling speed and efficiency. Traditional and statistical models are inadequate for predicting ROP in complex formations, as they fail to conduct a comprehensive analysis of method validity and data validity. In this study, geological conditions parameters, mechanical parameters, and drilling fluid parameters were extracted as prediction parameters, and an intelligent ROP prediction method was constructed under method-data dual validity analysis. The effectiveness of the ROP prediction method is studied by comparing five machine learning algorithms. The data validity of ROP prediction is also studied by changing the input data type, input data dimension, and input data sampling method. The results show that the effectiveness of the long short-term memory (LSTM) neural network method was found to be superior to support vector regression (SVR), backpropagation (BP) neural network, deep belief neural network (DBN), and convolutional neural network (CNN) methods. For data validity, the best input data type for ROP prediction is geological conditions parameters after principal component analysis (PCA) combined with mechanical parameters and drilling fluid parameters. The lower limit of input data dimension validity is seven input parameters, and the accuracy of prediction results increases with the increase of data dimension. The optimal data sampling method is one point per meter, and the error of the prediction result increases and then decreases with the increase of sampling points. Through step-by-step analysis of method validity, input data type, input data dimension, and input data sampling method, the range, size, and mean of error values of ROP prediction results were significantly reduced, and the mean absolute percentage error (MAPE) of the prediction results of the test set is only 18.40%, while the MAPE of the prediction results of the case study is only 11.60%. The results of this study can help to accurately predict ROP, achieve drilling speedup in complex formations, and promote the efficient development of hydrocarbons in the study area.