Artificial neural networks have been considered as an alternative to existing techniques across a broad range of disciplines including transport. Neural computing has advanced considerably in recent years. Viable solutions to forecasting and pattern recognition tasks are now emerging in increasing numbers. This paper reports on research commissioned by the Department of Transport to investigate the potential use of neural networks to predict (the onset of) congestion in an urban context. Key issues are identified as being neural network techniques and their existing applications in transportation and traffic control, sources of data available to a forecasting model and congestion definition and management. congestion definition and management. To establish whether a neural network forecasting model can be successfully developed and integrated into an urban traffic management system, all these issues must be considered in a coordinated approach. A broad scope of potential applications exists within the remit of forecasting (the onset of) urban congestion. The paper attempts to provide a generic commentary on the options and opportunities available. A focus is achieved through development of a demonstrator application which forecasts the Congestion Index (journey time/cruise time) on a given link for a future time horizon. State-of-the-art neural network software is used to produce models which are calibrated and validated using data collected from SCOOT networks in Southampton with consideration also given to SCOOT data from London.
Abstract