Adaptive Online Traffic Flow Prediction Using Aggregated Neuro Fuzzy Approach

Abstract

Short term prediction of traffic flow is one of the most essential elements of all proactive traffic control systems. Although various methodologies have been applied to forecast traffic parameters, several researchers have showed that compared with the individual methods, hybrid methods provide more accurate results . These results made the hybrid tools and approaches a more common method for prediction of various traffic parameters. In this paper, an aggregated approach is proposed for traffic flow prediction. The approach is based on the adaptive neuro-fuzzy inference system (ANFIS) and the macroscopic traffic flow model (METANET). Macroscopic modeling tool, METANET, is used to simulate the Hemmat highway/Tehran. After simulation, validation is done using real measurements to show the reliability of the simulation results. In order to calibrate the model, genetic algorithm was followed. The outcome suggests that the proposed approach as a hybrid method obtains a more accurate forecast than the neuro-fuzzy model alone.