Comparison of Four Modeling Techniques for Short-Term AADT Forecasting in Hong Kong

Abstract

In Hong Kong, the annual traffic census report is published in the middle of the year and used to present the results of traffic volume recorded at the automatic traffic counter stations. The type of traffic volume data being widely used is the annual average daily traffic (AADT), which is estimated on the basis of the daily flows by 12 months in the whole surveyed year. In this paper, time series, neural network, nonparametric regression, and Gaussian maximum likelihood (GML) methods were adapted to develop four models for short-term prediction of the daily traffic flows by day of week and by month, as well as the AADT for the whole current year. The historical data (1994–1998) and available current-year data for 1999 partial daily flows are the input data used for model development. The results of the four models were compared with the real data for validation. The daily flows estimated by the four models were used to calculate the AADT for the current year of 1999. Based on the comparison results, the GML model appears to be the most promising and robust of these four models for extensive applications to provide the short-term traffic forecasting database for the whole territory of Hong Kong.