Application of Artificial Neural Networks and Neighboring Sections Relationship in the Short-Term Travel Time Prediction on Urban Roadways

Abstract

In general, travel times on urban road networks are highly stochastic due to various disturbance factors from surrounding environment. However, most of the existing short-term travel time prediction techniques on these roadways are very simple methods e.g. using the travel time from historical database to expect the forthcoming travel time according to a specified date and time, or using the current travel time for representing the forecasted travel time of the study section. There was lack of consideration on the forming and dissipating trends of traffic congestion, and the effects from the neighboring road sections which commonly influences the traffic conditions and travel times of the study sections. This study aims to develop the robust travel time prediction models by using Artificial Neural Networks (ANNs) technique together with the integrating of data from neighboring sections as the candidates for model inputs. The testing results from the real field data on urban roadway networks in Bangkok CBD indicated the proposed technique was superior to the baseline models in all testing scenarios. Furthermore, in the case that data from target section was missing, this paper also suggested the model that could address the travel time prediction problem with acceptable results.