Dynamic O-D travel time estimation using an artificial neural network

Although the minimum Q-D travel time path in a dynamic traffic network can be calculated using standard minimum path algorithms there are a number of transportation applications which require a quick estimate of this information. These applications include such areas as real time vehicle dispatching systems where potential routes between a large number of origins and destinations have to be continually updated for a variety of vehicles throughout the day. The objective of this paper is to demonstrate the feasibility of using an artificial neural network (ANN) to estimate the O-D travel time in a dynamic traffic network. Three feed forward neural networks were developed to model the travel time behavior during different time periods of the day: AM peak, PM peak and off peak. These ANN models were subsequently trained and tested using a network from the City of Edmonton, Alberta. A comparison of the ANN model with a traditional statistical model is then presented. Lastly, the computational efficiency of the proposed ANN model compared to two shortest path algorithms is demonstrated. The statistical results show that the ANN models are appropriate for estimating dynamic O-D travel times and are significantly faster than the exact minimum path algorithms.