Developing Predictive Border Crossing Delay Models Lei Lin, Ph.D. Qian Wang, Ph.D. Adel W. Sadek, Ph.D. First Annual Transportation Informatics Symposium University at Buffalo – SUNY Buffalo, NY August 14, 2015
Transportation Systems Engineering University at Buffalo The State University of New York Transportation Systems Engineering Motivation & Background Niagara Frontier International border crossing: one of North America’s busiest travel portals
Motivation & Background Economic vitality of the Golden Horseshoe Continued increase in travel demand + tighter inspection Current or instantaneous vs. Predicted or experienced
Methodology Two Steps: Border Crossing Traffic Volume Prediction Multi-server Queueing Model
Border Crossing Traffic Volume Prediction Data Processing & Analysis Individual Model Development SARIMA Model SVR Multi-model Combined Forecasting Fixed weight method Fuzzy Adaptive Variable Weight Method based on Fresh Degree Function The Spinning Network Method
Transportation Systems Engineering University at Buffalo The State University of New York Transportation Systems Engineering Border Crossing Traffic Volume Prediction Data Processing
Border Crossing Traffic Volume Prediction SARIMA Model Prediction Accuracy
Border Crossing Traffic Volume Prediction SVR Prediction Accuracy
Border Crossing Traffic Volume Prediction Multi-model Combined Forecasting SARIMA: Good for weekdays (9.84% vs % for SVR) SVR: Good for game days (9.42% vs % for SARIMA) Combining the forecasts from the two models: Fixed weight method Identifies the model that works best for a given hour Fuzzy Adaptive Variable Weight (Fresh Degree Function) weights assigned to each model based on how well each model performed on recent forecasts
Border Crossing Traffic Volume Prediction Models’ Performance Comparison
Spinning Network Forecasting Method A forecasting algorithm proposed by Huang & Sadek (2009). The method attempts to mimic some aspects of human memory and has the advantages of: Computational Efficiency Robustness
Spinning network (SPN)
Dynamic Time Warping (DTW)
Performance Evaluation
Method MAPE (%) Test Dataset (1,905 hours) MAPE (%) Hours showing Abrupt Change (36 hours) DTW-SPN Euclidean-SPN SARIMA SVR
Methodology Two Steps: Border Crossing Traffic Volume Prediction Multi-server Queueing Model
Queueing Model Development Inter-arrival Time Distribution Service Time Distribution
Queueing Models
Queueing Model---VISSIM model
Comparison between Analytical Approach and Simulation Approach Traffic Volume (vph) No. of Service Stations Number of Vehicles in the Queue (Vehicles) Simulation in VISSIM Mean Standard Variance Mean Standard Variance Mean Standard Variance
A Smartphone app, the Toronto Buffalo Border Waiting (TBBW) app, designed to collect, share and predict waiting time at the three Niagara Frontier border crossings Putting it all together – TBBW App
Future Research Compare to blue-tooth delay measurement data Modeling station opening mechanism or rules Mechanism for on-line delay prediction adjustment A comprehensive border management framework for intelligent routing and traffic load balancing
THANK YOU ! QUESTIONS ! Adel W. Sadek, Ph.D. Professor University at Buffalo – SUNY Buffalo, NY Phone: (716) FAX: (716)