Nonlinear autoregressive neural networks to predict hydraulic fracturing fluid leakage into shallow groundwater

Abstract

Hydraulic fracturing of horizontal wells is an essential technology for the exploitation of unconventional resources, but led to environmental concerns. Fracturing fluid upward migration from deep gas reservoirs along abandoned wells may pose contamination threats to shallow groundwater. This study describes the novel application of a nonlinear autoregressive (NAR) neural network to estimate fracturing fluid flow rate to shallow aquifers in the presence of an abandoned well. The NAR network is trained using the Levenberg-Marquardt (LM) and Bayesian Regularization (BR) algorithms and the results were compared to identify the optimal network architecture. For NAR-LM model, the coefficient of determination (R2) between measured and predicted values is 0.923 and the mean squared error (MSE) is 4.2 × 10-4, and the values of R2 = 0.944 and MSE = 2.4 × 10-4 were obtained for the NAR-BR model. The results indicate the robustness and compatibility of NAR-LM and NAR-BR models in predicting fracturing fluid flow rate to shallow aquifers. This study shows that NAR neural networks can be useful and hold considerable potential for assessing the groundwater impacts of unconventional gas development.