Lessons Learned in 18 Months of Deploying Machine Learning for Predictive Operational Support

This manuscript summarizes the past 18 months of deploying a predictive operational dashboard leveraging machine learning (ML) to forecast inflow into Citizens Energy Group’s DigIndy Tunnel system. We will present our lessons learned in the deployment of the ML tool in the cloud, managing the ML tool, and the integration of meter data to evaluate the accuracy of the forecast. Citizens Energy Group is the combined water, wastewater, natural gas, steam, and chilled water utility in Indianapolis. Citizens is presently implementing a 20-year CSO consent decree with required completion in 2025. Consent decree compliance is primarily achieved through the DigIndy Tunnel system, a 28-mile long, 18-foot diameter tunnel network. Citizens is also implementing a pair of storage facilities in the Upper White River and Upper Pogues Run watersheds. Our ML tool consists of an artificial neural network fit to long-term datasets to predict inflows to the DigIndy Tunnel using the 72-hour NOAA radar rainfall forecast (Sutton et. al, 2022). The neural network and automated tools to obtain actual gauged rainfall and metered tunnel inflow are fully contained within Citizens’ Azure Synapse enterprise cloud environment, with data storage in Azure Data Lake. The forecasted rainfall and tunnel inflow and actual data is presented in an online Power BI dashboard hosted in Citizens’ enterprise architecture that is accessible to all engineering and operations staff.