Utilizing Intelligent O&M to Harness Machine Learning Across the Water and Wastewater Industry

BACKGROUND AND LEARNING OBJECTIVE This presentation will describe a successful approach taken for developing and deploying wide-scale machine learning algorithms in the wastewater industry. Attendees will learn the level of effort and investment required to truly leverage the power of Artificial Intelligence (AI) at scale and all the accompanying challenges, benefits, and considerations. This session focuses on the utilization of Foundry (a data delivery software developed by Palantir) to facilitate the development, deployment, and monitoring of machine learning at two wastewater treatment facilities (WWTF) — Agua Nueva WWTF as well as Wilmington WWTP. These implementations have been active for over a year and are utilizing the models to support day to day operations consistently. With the case study at both Agua Nueva and Wilmington WWTF, disinfection plays a large role in the successful treatment of wastewater and has significant cost implications dependent on operations. Using Foundry, Jacobs unlocked the power of streaming sensor data, integrated with Jacobs Smart Algorithms and subject matter expertise, models can be custom tailored and translated to simple human readable recommended actions. Real-world feedback, including operator messages model error, and sensor drift, are captured in platform to continually improve performance over time. PROJECT OBJECTIVES The objectives were to: 1.Increase operator engagement with tools developed by data scientists and process engineers to model the complexities of treatment facilities and translate those models into actionable workflows that operators can use to make their jobs easier while improving plant efficiency with chemical usage. Discussion will present how lead Jacobs operators at two different wastewater facilities leverages these tools developed by data scientist and process engineers to facilitate chemical savings. 2.Build technical infrastructure that involves heavy data engineering to configuring data connectors and pipelines to ingest SCADA historian data, maintenance data, HACH WIMs data, and financial data all into a single unified platform and using data pipelines and cloud technology. 3.Develop and deploy machine learning that ingest all this information and provide actionable insights and recommendations to operators to fully leverage big data, cloud technology, data science, and wastewater subject matter expertise. Figure 1 shows a mapping of all the processes used in the foundry toolkit to facilitate the objectives from start to finish. METHODOLOGY, RESULTS, AND CONCLUSION The modeling for the case studies at Agua Nueva and Wilmington follow a 3 part modeling format: -multivariate bayesian inferencing to statistically quantify risk profiles for disinfection and bacterial exceedances (Figure 2). This allowed for designing a dosing strategy for the facility that would result in as close to 0% probability of exceedance as statistically definable based on the available data streams and past performance of the facility considering things like CT, contact time, dose, wind speeds, pH, temperature, and other conditions at the facility. -machine learning modeling to predict plant flows and other future conditions to estimate (Figure 3). Using weather forecasts, temporal data, and plant data allowed for successful prediction of flow concentrations with R2 values > 0.9. -estimating required dosages based on aforementioned steps and using machine learning to estimate downstream implications such as residual concentrations. (Figure 4) There are many components to how the models are built and how model metrics are tracked. The main metrics tracked in foundry are root mean squared errors (RMSE), mean absolute percentage errors (MAPE), and coefficients of determinations (r2). These are tracked in real time in Foundry and can alert to model health and notify Data Scientists for any sensor drift or model retraining. Additionally, Figure 6 shows several randomized back tests of the response variable (total residual chlorine concentrations) that depict how well models remain relevant to optimizing for compliance and savings. As of current, both sites are on track to having 10-30% energy Savings at Agua Nueva with close to $100,000 in annual energy costs for Agua Nueva and around $250,000 in savings for Wilmington. For Agua Nueva and Wilmington respectively, Figure 5 and Figure 6 shows the results of this on a gal/MGD (gallons of hypo used per million gallons of water treated) for basis where the red line shows the historical average usage and the green line shows the chemical usages after implementation of the tool. Savings are the area between the two lines and shaded in blue and the trends are continuing to be positive. In addition to this, implementation of the technology is also tracked over time and both site exhibit model utilization rates of over 70% shown in Figure 7 and Figure 8. This case-study demonstrates significant progress in the successful implementation of AI at a treatment facility. The strong engagement with operators has advanced their skillset such that they are submitting abstracts to their local operator's conferences at their own volition.