Short Duration Traffic Flow Prediction of an Urban Road Segment

Abstract

The study involved estimating short-duration traffic flow count using a mathematical filtering technique named Kalman filtering. The area of study was taken Mirpur road, Dhaka city near Sobhanbagh mosque. There is the traffic of heterogeneous mix in the traffic stream. Though the efficiency of the Kalman filtering technique (KFT) is tested for homogeneous traffic already, however efficacy of KFT under heterogeneous traffic yet to be explored. A short-duration traffic estimation is a useful tool for traffic operation and transportation system management. Route guidance and advanced traveler information system can use the outcome of short duration traffic count for travel time estimation. The proposed technique is implemented using a python library developed by Kalman.py library API. The library is widely implemented in the advanced modeling of databases in the KFT framework. The data were obtained from 1-hour traffic count of the vehicle. The heterogeneous traffic count was converted into equivalent passenger car unit (PCU) as per the Indian road congress manual. The PCU obtained over every 5 minutes aggregation then used as the dataset for the KFT model. The proposed model has a mean absolute error (MAE) of 15.39%, which represents that the KFT model has reasonably good forecasting capacity. The root mean square error (RMSE) shows 19.36% accuracy. The developed model has an R2 value of 0.543 i.e. the model can explain 54.3% variability of the dataset. The proposed estimation technique can be implemented in the application tool developed for travel time prediction and traffic flow count estimation dynamically. The application of KFT is tested for both motorized and non-motorized vehicles. The study can be extended to other geographic locations. Also, traffic under the various levels of service can be studied for a wide range of validation of the study.