Probabilistic Forecasting and Anomaly Detection in Sewer Systems Using Gaussian Processes

This study investigates the capability of Gaussian process regression (GPR) models in probabilistic forecasting of water flow and depth in a combined sewer system. Traditionally, deterministic methods have been implemented in sewer flow forecasting and anomaly detection which are two crucial techniques for a good wastewater network and treatment plant management. However, with the uncertain nature of the factors impacting on sewer flow and depth, a probabilistic approach which takes uncertainties into account is preferred. Using hydraulic simulation data generated in this study, a composite kernel is designed to take the flow and depth patterns based on the flow characteristics, and the hyperparameters are optimised through maximisation of the log-likelihood function. The GPR model is a multi-input, single-output model taking time and precipitation as inputs. Prediction results show reliable model outputs and are evaluated by three metrics: root mean square error, coverage and differential entropy. The model effectively predicts treatment plant inflow and manhole water levels with different training periods. Finally, the model is used for anomaly detection by identifying deviations from expected ranges, enabling the estimation of surcharge and overflow probabilities under various conditions.