01153544

Component

This paper presents a Fuzzy Neural Network (FNN) approach to predict short-term intersection flow. The flow arriving to an intersection approach is predicted using a FNN based on flows at the upstream intersection approaches that feed into the subject intersection approach in the previous time interval. The FNN combines the strength of neural networks and fuzzy logic through classifying the input data into a number of clusters using the fuzzy system, and specifying the input-output relationship by adaptively calibrating the parameters of the fuzzy system with a neural network. The prediction performance of the FNN model is investigated for three different prediction intervals. The FNN is also compared with a widely-used back propagation neural network (BPNN) and a Time Series Model (ARIMA). The experiment conducted in the paper shows that the FNN approach produced good prediction accuracy and outperformed the BPNN and ARIMA models. Furthermore, the FNN prediction based on upstream flows is shown to produce better results than the prediction based on the flow at the current intersection location. This demonstrates an important potential of the FNN approach for adaptive signal control applications.

01147878

10-2780

English

19p

2010

Transportation Research Board 89th Annual Meeting

DVD

Figures (7) ; References (30)

Fuzzy algorithms; Neural networks; Signalized intersections; Traffic flow rate; Traffic forecasting; Traffic models; Traffic oversaturation

Data and Information Technology; Highways

Transportation Research Board Annual Meeting 2010 Paper #10-2780

BTRIS, TRIS, TRB

Jan 25 2010 11:21AM