Variation Based Online Travel Time Prediction Using Clustered Neural Networks

Abstract

This paper proposes a variation-based online travel time prediction approach using clustered Neural Networks with traffic vectors extracted from raw detector data as the input variables. Different from previous studies, the proposed approach decomposes the corridor travel time into two parts: 1) the base term, which is predicted by a fuzzy membership-value-weighted average of the clustered historical data to reflect the primary traffic pattern in the corridor; and 2) the variation term, which is predicted through the calibrated cluster-based artificial neural network model to capture the actual traffic fluctuation. To evaluate the effectiveness of the proposed approach, this paper has conducted intensive numerical experiments with simulated data from the microscopic simulator CORSIM. Experimental results under various traffic volume levels have revealed the potentials for the proposed method to be applied in online corridor travel time prediction.