Investigating the artificial intelligence methods for determining performance of the NZVI permeable reactive barriers

In situ groundwater remediation, which contains multiple contaminants simultaneously, is a fundamental challenge globally. One of the suitable technologies for the removal of several contaminants from groundwater is a permeable reactive barrier containing iron nanoparticles (NZVI-PRB). A laboratory-scale experimental setup is used as NZVI-PRB with a funnel- and- gate barrier. NZVI (average particle diameter (APS) = 35-70 nm) particles are used in PRB as a reactive media. NO3-N, as a dominant element of required contact time, is considered for the design of the optimum width of PRB. Then, three contaminants of nitrate (NO3), cadmium, and chromium (VI) were removed simultaneously from a shallow aquifer. With the development of Artificial Neural Network (ANN) in the context of Deep Neural Network (DNN) models and machine learning methods such as Support Vector Machine (SVM) and Linear Regression (LR), considerable progress has been made in various fields. In the final step, we assessed the performance of DNN, SVM, and LR models to estimate the residual contaminants based on preliminary experimental data. The average elimination of contaminants was stabilized at about 45% of the initial value. Our DNN model could estimate nitrate, cadmium, and chromium with a mean absolute percentage error (MAPE) of 7.05, 7.32, and 7.84, respectively. Our results showed that despite using a small dataset, utilizing a deep, fully connected network resulted in remarkably higher accuracy than the other methods. Notably, among the three contaminants, Nitrate estimation is performed more accurately and recommended for future large-scale modeling.

Automated summary

The study successfully employed a permeable reactive barrier containing iron nanoparticles to simultaneously remove nitrate, cadmium, and chromium from groundwater. By utilizing machine learning methods such as deep neural networks, a more accurate estimation of residual contaminants was achieved, resulting in higher efficiency in contaminant removal.