Abstract:
This paper proposes a new logistic regression model to predict a vehicle's queue/platoon state based on data that can be collected from a single loop detector positioned at the stop line of signalised intersections. The study focuses on estimating the queue length at the end of each cycle by predicting whether a vehicle was queued or platooned prior to passing over the detector. Different model forms were explored using data from an enhanced NGSIM dataset. These data were filtered to mimic data from a stop line detector loop. The best four models from 20 resulted in an accuracy ranging from 83% to 95% of correctly predicting a discharging vehicle's queue/platoon state, during the preceding red phase. When combined with a logical filter to group sequential vehicles, it will enable a traffic controller to estimate the most likely queue length. The queue predictor model will form part of a new offset algorithm under development.