Optimization of Water Quality Sensor Placement in Sewer Networks

Wastewater surveillance has received significant attention in recent years due to the valuable environmental and epidemiological information it reveals. Yet, limited attempts have been made in previous literature to develop frameworks for optimizing the placement of sensors and/or the locations for sample collection. In this study, a simulation optimization framework is developed to optimize the placement of water quality sensors in sewer networks. The objective of the optimization process is to maximize the ability of the sensor network to detect the species of interest (e.g., a contaminant or a biomarker) throughout the sewershed (i.e., observability), and to identify the characteristics of the source junction(s) through which the species enters the system (i.e., source identifiability), including source locations and concentrations. The developed python-based framework couples a Multi-layer Perceptron Neural Network (MLP-NN) model with a Genetic Algorithm (GA) to identify the optimal design of the sensor network. The MLP-NN model is trained to forecast the concentrations of the species of interest throughout the sewershed, which are then propagated into the GA to find the sensor placement design that maximizes both the observability and source identifiability under a wide range of scenarios. The developed framework was demonstrated on a case study featuring a real-life, mid-sized sewer network. Preliminary results revealed that a clear tradeoff exists between observability and identifiability, where placing sensors in the downstream end of the network generally resulted in higher detection probability, while better identification of source characteristics was achieved by sensors placed in the upstream sections of the network.