Authors: Jin Xing*, Newcastle University, Philip James, Newcastle University , Stuart Barr, Newcastle University
Topics: Geographic Information Science and Systems, Spatial Analysis & Modeling, Urban Geography
Keywords: Sewage Prediction; Recurrent Neural Network; Real-Time Computing; Smart City
Session Type: Paper
Start / End Time: 3:05 PM / 4:45 PM
Room: Roosevelt 3, Marriott, Exhibition Level
Presentation File: No File Uploaded
Understanding in real-time the operational capability of sewage networks is critical in order to assess risk to transport disruption and public health. However, relatively few sewage networks are instrumented. In such cases, a network of rainfall gauges may be used to record instantaneous rainfall and precipitation accumulation which in turn could potentially be used to infer sewage discharge spatially across the network. However, to date the ability to predict in real-time sewage discharge has been limited because traditional time series analysis is unable to dynamically respond to abrupt high magnitude changes that one might experience during extreme rainfall events. In this work we demonstrate how a new real-time sewage discharge prediction framework based on the combination of recurrent neural network and real-time data streaming rainfall gauges. Our work utilises a Long-Short Term Memory (LSTM) machine learning approach that dynamically adjusts weights among previous measurements and their current values, to account abrupt changes. The framework has been implemented using the real-time data collection service provided by the UK Newcastle University Urban Observatory and validated with sewage discharge measurement from Northumbrian Water Group. The high accuracy of the prediction result has indicated the promising future of integrating deep learning with real-time computing for Smart City research.