Lane-based short-term urban traffic forecasting with GA designed ANN and LWR models

摘要

Short-term traffic forecasting has received special attention in the past decade due primarily to their vital role in supporting route choice decisions, traffic management and control. In this study, genetic algorithms (GAs) are used to design artificial neural network (ANN) models and locally weighted regression (LWR) models. The modeling approach proposed relies on a combination of Genetic algorithm, neural network and locally weighted regressions to achieve optimal prediction performance under various input and traffic condition settings. The GA designed ANN (GA-ANN) and GA designed LWR (GA-LWR) aggregate and disaggregate models were used to predict short-term traffic (5-minute) for four lanes of an urban road in Beijing, China. The GA-ANN models developed in this study show most of the average errors, are less than 5-6% and 95th percentile errors are mostly less than 15% for all lanes. Whereas overall the GA-LWR models developed in this study show a better performance with their average errors mostly less than 5% and 95th percentile errors lower than 10%. Study results show that such accurate predictions would be useful for highway authorities to put through their statewide ATIS.