Development of a Data-Driven Framework for Wellhead Pressure Prediction in Hydraulic Fracture Operations

Pressure prediction is essential for optimizing hydraulic fracturing (HF) performance and mitigating operational risks. By analyzing complex patterns and processing large volumes of data, data-driven methods exhibit better accuracy and greater potential than traditional methods, thereby ensuring optimal prediction during HF operations. The present study is an attempt to provide the possibility of predicting wellhead pressure (WHP) using data-driven models (DDMs), including random forest (RF), convolutional neural networks (CNN), and support vector machine (SVM). The effectiveness of the models was evaluated based on operational data derived from the McCully gas field. The prediction results demonstrated that the RF model had the highest accuracy with an R-squared correlation (\(\:{\text{R}}^{2}\)) of 0.9517 for the experimental dataset. Also, it has mean absolute error (MAE) and root mean square error (RMSE) values of 0.37 and 0.081, respectively, indicating the minimum error of the model in WHP prediction. In addition, analyzing the injection rate (IR) and pressure drop trend using the RF model could help properly diagnose the behavioral pattern of sudden pressure changes. These results proved the reliability and effectiveness of the RF model for WHP prediction, which can contribute considerably to HF design optimization, operational risk mitigation, and reservoir performance maximization in the future.