Quantifying Aquifer Contamination Risk from Casing Rupture using Backpropagation Neural Network: A Comprehensive Assessment

Abstract

Abstract Casing failure in hydraulic fracturing can lead to the leakage of fracturing fluid into the aquifer, resulting in groundwater contamination. To prevent such contamination, a universal assessment model is necessary to understand fracturing fluid transport in the subsurface. However, the complexity of the formation, parameter uncertainty, and computational challenges pose obstacles to risk assessment. In this study, we address these challenges by considering the uncertainty of hydrological parameters and the heterogeneity of the formation in the development of a new conceptual model. Numerical simulations and sensitivity analysis were performed to investigate the key factors influencing groundwater contamination. Additionally, a backpropagation neural network (BPNN) was developed as an alternative approach. Monte Carlo simulations using Latin Hypercube sampling were conducted to obtain probability distributions. Our results demonstrated strong correlations in the BPNN model, with correlation coefficients (R2) of 0.9973 and 0.9617, and low Root Mean Square Errors (RMSE) of 5.45×10− 2 mg and 3.607 days. In a ten-year risk assessment, the probability of contaminant flux in the aquifer being less than 0.651 mg was 100%, indicating a low risk, and the average time for fracturing fluid to reach the aquifer was 1,500 days. These findings provide valuable insights into the potential environmental impact of fracturing fluid contaminants and can inform the development of regulations and best practices for fracturing operations.