Fracture morphology is the fundamental data to optimize hydraulic fracturing design. However, the physical modeling and simulation is time consuming involving data preparation, model parameter setting and validation, time-consuming running. Machine Learning (ML) methods have the potential to integrate multi-source data and enhance calculation efficiency, thus enable to be critically useful in fracture morphology prediction. This paper aims to create a ML-based surrogate model for quick fracture morphology prediction. The approach involves the numerical simulation of hydraulic fracturing in two blocks wells, the combination of geological and engineering parameters to be a dataset. Particularly, the pumping time-series data were integrated by encoding and the Conditional Generative Adversarial Networks were used to expand the samples in order to enhance the model's capabilities. Following this, the correlation and principal component analyses were used for the feature engineering to identify key factors for stimulation effectiveness and surrogate model inputs. The comparative analysis demonstrated the excellent predictive ability of the fused Recurrent Neural Network and Multi-Layer Perceptron model, reducing validation set errors by approximately 15%. This work demonstrates the feasibility of ML-based surrogate models for predicting fracture morphology, with the advantages of fast, efficient and integrated multi-source data.

This content is only available via PDF.
You can access this article if you purchase or spend a download.