Accurate diagnosis of sewer inflow and infiltration (I/I) is crucial for ensuring the safe transportation of sewage and the stability of wastewater treatment processes. Identifying periods impacted by I/I is essential for I/I diagnosis, but current methods lack a standard criterion and require adaptation to specific conditions, resulting in low accuracy, complexity, and limited generalizability. This paper proposes a novel approach to distinguish I/I periods from time series of sewer measurements based on anomaly detection theory through an iterative use of a time-series reconstruction model. This method eliminates the need for external data such as rainfalls and avoids intensive manual data analysis. Operating directly on in-sewer data, it enhances accuracy compared to existing approaches and is applicable to various external factors such as rainfall, snowmelt, and seawater intrusion. The method can be applicable to a broad range of monitoring data, including flow rate, temperature, and conductivity. Validated through simulation studies and demonstrated via real-life applications, this method offers an efficient solution for I/I detection, facilitating further I/I diagnosis, including I/I quantification and location identification.
Keywords: Anomaly detection; Data; Dry/ wet weather; Inflow and infiltration; Time series.
© 2024 The Authors. Published by Elsevier Ltd.