Short-term traffic speed prediction plays a key importance in urban traffic management and operation. Literature review indicates that lane-based short-term urban traffic forecasting is still rare. In this study, genetic algorithms (GAs) are used to optimize the input data sets for artificial neural network (ANN) models and locally weighted regression (LWR) models to achieve optimal prediction under various input and traffic settings. The GA designed ANN (GA-ANN) and GA designed LWR (GA-LWR) models are used to predict 5 minute short-term traffic speed prediction for four lanes of an urban road. For the peak and off-peak time traffic speed prediction, both GA-ANN and GA-LWR aggregate and disaggregate models are developed and tested, with the former forecasting traffic speed of every 5 min with a 30 min period (e.g., 7:25–7:55 am morning peak) and the latter for forecasting traffic speed of a particular 5-min of each weekday (e.g., 7:25–7:30 am of Monday to Friday). In addition, for peak and off-peak traffic speed prediction, the GA-ANN disaggregate model results in most cases average errors of 3-4% and the 95th percentile errors of lower than 8% for each of the four lanes. Meanwhile, for the peak and off-peak time traffic speed prediction, the GA-LWR disaggregate model results in most cases average errors of 1-2% and the 95th percentile errors of lower than 4% for each of the four lanes. When compared to previous models that usually offer average errors more than 6-15%, such empirical findings is promising and instrumental for transportation authorities to put through their city- or state-wide ATIS.
An Optimized Hybrid Lane-Based Short-Term Urban Traffic Forecasting Using Artificial Neural Network and Locally Weighted Regression Models
17th COTA International Conference of Transportation Professionals ; 2017 ; Shanghai, China
CICTP 2017 ; 3734-3748
2018-01-18
Conference paper
Electronic Resource
English
Taylor & Francis Verlag | 2018
|Short Term Traffic Flow Forecasting Based on Artificial Neural Network Combined Predictor
British Library Conference Proceedings | 2007
|