AI/ML tools for early decision making in water system operations: managing non-stationarity water quality

Abstract

The assumption that natural systems oscillate within a stationary range of variability has traditionally guided water system management and allowed water utilities to experience steady state operations. These steady state operations are viewed as the 'normal state' of the water system. However, non-stationarity events such as wildfires, droughts, and floods shift water systems to new 'states' that negatively impact water quality and complicate water treatment decision-making and performance. Many water system managers do not account for these variations systemically and instead respond reactively as watersheds shift from perceived normal states. To enhance their operational resilience and develop adaptive and robust methodologies, water utilities must gain knowledge of non-stationarity states. Artificial Intelligence (AI) and its subset, Machine Learning (ML), are emerging as key tools for addressing the impacts of non-stationarity events on water system operations. This thesis responds to this gap by investigating how AI and its subset, ML, are emerging as key tools for addressing the impacts of non-stationarity events on water system operations. It provides a review of how AI supports decision-making in drinking water treatment systems when non-stationary water quality states occur due to perturbations. This study provides a summary and observations on: (1) Understanding the boundary influences on water quality due to non-stationarity events and their implications for drinking water treatment processes. (2) Exploring how AI/ML methods inform stationarity and non-stationarity system state patterns. (3) Applying AI/ML models to develop a TOC predictive tool and assess their potential use to address non-stationarity water quality states.