Evaluation of Speed-Based Travel Time Estimation Models

Abstract

Travel time estimation models, which rely on speed data provided from point detectors (usually inductive loops), have application in travel time prediction and network performance monitoring. Unfortunately there are limited, and at times counterintuitive, results in the literature about their performance. This paper focuses on the field evaluation of four speed-based travel time estimation models, namely, the instantaneous model, the time slice model, the dynamic time slice model, and the linear model. Those models are evaluated using data from two operational motorways in Melbourne, Australia. Travel time estimation errors are quantified against actual travel times measured using a timed number plate survey and time-stamped toll tag data. There was little difference in the travel time estimation error across the models and they were all found to underestimate actual travel times. Errors ranged from about 7% in the off peak up to 15% in the peak. Marginal improvements in model performance were achieved through careful selection of which detectors provide input for each section (upstream, downstream, or the average of those values) and by conversion of the inputs from time mean speed to an estimate of space mean speed.