Neighbour corridors travel time estimation: Concept and a case study

Abstract

An approach is presented to estimate corridor travel time in urban areas using available data of a nearby corridor. The purpose is to improve the efficiency of real-time traveller information systems when the number of data collection sensors is limited. Field travel time data were collected for two corridors in downtown Vancouver, British Columbia. The association between the travel times of the two corridors was found significant. Models were then developed to estimate travel times on one corridor using data of the other corridor. The developed models included regression, Artificial Neural Network (ANN), Neuro-fuzzy, and K-Nearest Neighbours (KNN). The estimation accuracy was considered satisfactory as the Mean Absolute Percentage Error (MAPE) of all models ranged between 13.7% and 17.6%. It was concluded that the concept of estimating travel time from nearby corridors is promising. The type of modelling technique had a little impact on the results with the KNN method producing slightly better results.