Prediction of Fracture Morphology Parameters in Unconventional Reservoir Fracturing: Innovative Insights Driven by Deep Learning

ABSTRACT: Accurately and efficiently predicting the fracture propagation patterns in hydraulic fracturing is crucial for optimizing the development of unconventional oil and gas resources. However, traditional models primarily rely on empirical formulas and simplified assumptions, leading to high computational costs and limiting a deeper understanding of complex fracture network behaviors. This study presents a deep learning surrogate model framework based on the multi-head self-attention mechanism of Transformer for predicting the total fracture volume, a key parameter in fracture propagation. Fracture propagation simulation data is generated using the unconventional fracture model, which is then used to train and optimize the surrogate model. Experimental results show that the surrogate model achieves a high prediction accuracy on the test set, with a coefficient of determination of 0.994, root mean square error of 10.82, and mean absolute percentage error of 1.26%. Moreover, the model's computational efficiency is significantly improved, reducing the prediction time by 100,000 to 500,000 times compared to traditional numerical simulation methods. This approach not only enhances prediction accuracy but also provides a powerful tool for efficient prediction of hydraulic fracturing parameters, offering new theoretical support and technical tools for the development of unconventional resources.