Fracture pressure prediction method of horizontal well based on neural network model

Fracture pressure is the fundamental basis for the design of the well structure and the basic parameter for the selection of the hydraulic fracturing equipment and design scheme. Logging interpretation is usually used to obtain the fracture pressure profile, but it has the problems of being difficult to obtain the parameters accurately, tedious calculation process, poor applicability and low calculation accuracy. Machine learning offers a new way to solve these problems. Therefore, four different neural network models were used to establish the nonlinear relationship between horizontal well log data and fracture pressure using log data as input parameters, the best neural network model was preferred through the comparative analysis of test set prediction results, and the model network structure and hyperparameters were optimized to realize the direct prediction of horizontal well fracture pressure. The results show that: 1) Fracture pressure shows a very strong correlation with well inclination angle and interval transit time of P- and S-waves, a strong correlation with well depth, lithology density and compensation neutron, and a weak correlation with well diameter and natural gamma. 2) Different combinations of logging parameters have significant effects on the model prediction results, and the optimal input parameters are well inclination angle, interval transit time of P- and S-waves, well depth, lithology density, and compensation neutron. 3) By comparing the multilayer perceptron, deep neural network, recurrent neural network, and long- and short-term memory(LSTM) neural network model, it is found that the LSTM model has the best prediction effect. 4) The network structure and hyperparameters of the LSTM model are optimized, and the mean absolute percentage error of the fracture pressure prediction after optimization is 0.106% and its coefficient of determination is 0.996. The LSTM model can effectively construct a nonlinear relationship between horizontal well logging parameters and fracture pressure, and can achieve accurate prediction of horizontal well fracture pressure, which is important for accurately predicting fracture pressure, simplifying the fracture pressure calculation process, and promoting the application of machine learning in the field of petroleum engineering. © 2024 Central South University of Technology. All rights reserved.