APPLICATION OF EXPLAINABLE ARTIFICIAL INTELLIGENCE IN WASTEWATER TREATMENT PLANTS

Abstract

Wastewater treatment plants (WWTPs) play a crucial role in protecting public health and the environment by removing contaminants before releasing them back into the environment. However, efficient management of these plants can be challenging because of the complex nature of wastewater treatment processes. Machine learning (ML), a subfield of artificial intelligence (AI), can predict WWTP variables by extracting correlations between variables from historical data. Accurate effluent variable prediction through ML can facilitate efficient adjustment of operational variables, thus minimizing operation cost while effectively meeting effluent quality standards. However, relying solely on ML models without understanding the contexts of the predictions is not ideal, especially in WWTPs where operators need to understand the reasons behind model predictions to increase their confidence in real-world applications. While recent studies have focused on predicting variables in WWTP using ML, research on implementing explainable artificial intelligence (XAI) is still developing. XAI is a promising approach for enhancing the transparency and interpretability of ML models in complex environmental systems, such as WWTPs. This study investigated the performance of widely used ML algorithms in predicting the key effluent quality variables across four WWTPs in Wisconsin, USA. Data was collected from both small- and large-scale WWTPs to evaluate the scalability and generalizability of the models. Feature selection (FS) techniques were used to identify the most relevant features for each target variable in the dataset. XAI tools, including SHapley Additive Explanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME), were used to interpret the influence of input variables on the model outputs. This study highlights how ML can accurately predict water quality outcomes, thereby allowing plant operators to better manage their facilities. The study also demonstrates how XAI can clarify the reason behind specific ML models' predictions, enhancing operator confidence in decision-making. These XAI methods identify which factors, i.e., nutrient levels, flow rates, and organic material loads, most significantly impact effluent quality. Using these transparent predictive tools, wastewater facilities can optimize operations, meet regulatory standards, reduce operating costs, and ultimately support healthier communities and ecosystems.