This research project develops a fuzzy logic ramp metering algorithm utilizing artificial neural network (ANN) traffic data predictors. Considering the highly beneficial effects of ramp metering, such as reduced travel times and lower accident rates, optimizing metering rates is of great importance. The research objective is to overcome limitations of the current Seattle ramp metering algorithm, which reacts to existing bottlenecks rather than preventing them. An algorithm with predictive capabilities can help prevent or delay bottleneck formation. Hence, an accurate 1-minute ANN prediction provides a powerful asset to the ramp metering algorithm. The research project divides into two stages: the ANN traffic data predictor and the fuzzy logic ramp metering algorithm. This research focuses primarily on the ANN traffic data predictors, but also lays the groundwork for the fuzzy logic ramp metering concepts and algorithm.
The ANN predicts 1 minute in advance significantly better than previous techniques in the Seattle area, as well as demonstrates robustness to faulty loop detector data. A multi-layer perceptron (MLP) type of ANN predicts congested mainline volume and occupancy for a station when given past values of volume and occupancy for that particular station and the adjacent upstream station. This data prediction provides an input to the fuzzy logic ramp metering algorithm. The ramp metering rate is then based on both current and predicted traffic flow. By considering the freeway as a control system instead of one section at a time, the new algorithm should avoid an oscillatory ramp metering rate, and achieve equilibrium more quickly and smoothly.
Washington State Transportation Center (TRAC)
Algorithms, Bottlenecks, Fuzzy logic, Neural networks, Prevention, Ramp metering, Traffic data, Traffic forecasting.