Freeway traffic data prediction using neural networks

Abstract

A multi-layer perceptron type of artificial neural network predicts congested freeway data while demonstrating robustness to faulty loop detector data. Test results on historical data from the I-5 freeway in Seattle, Washington demonstrate that a neural network can successfully predict volume and occupancy one minute in advance, as well as fill in the gaps for missing data with an appropriate prediction. The volume and occupancy predictions are used as inputs to a fuzzy logic ramp metering algorithm currently under testing.