A neuro-fuzzy system approach for forecasting short-term freeway traffic flows

Abstract

Because the neuro-fuzzy system (NFS) combines the learning capability of neural networks and the decision structure of fuzzy inference systems, it is very useful in the modeling, control, and forecasting of complex systems such as traffic systems. This paper proposes a form of neuro-fuzzy systems (NFS) and applies it to forecast short-term traffic flows. Different learning algorithms for the NFS have been tested and evaluated using actual traffic data collected from the Loop 3 Freeway in Beijing, China. These test results indicate that the NFS based approach is an effective method for short-tern traffic flow forecasting. To demonstrate the advantage of the proposed approach, a comparison with a typical neural network based approach has been made.