Assignment Matrix Free Algorithms for On-line Estimation of Dynamic Origin-Destination Matrices

Dynamic Traffic Assignment (DTA) models represent fundamental tools to forecast traffic flows on road networks, assessing the effects of management and transport policies. As biased lead incorrect predictions, which can cause inaccurate evaluations huge social costs, calibration DTA is an established active research field. When it comes estimating Origin-Destination (OD) demand flows, perhaps most important input for models, one algorithm suggested outperform all others real-time applications: Kalman Filter (KF). This paper introduces a non-linear framework online dynamic OD estimation that reduces number variables easily incorporate heterogeneous data sources better explain relationship between time-dependent OD-flows. Specifically, we propose model takes advantage Principal Component Analysis (PCA) capture spatial correlations exploit local nature specific KF recently proposed in literature, Local Ensemble Transformed filter (LETKF). The main LETKF gain not explicitly formulated means that, differently from other approaches there no need compute assignment matrix or its approximation. shows different sources, such as counts link speeds. Additionally, thanks PCA, identify patterns within correlation data. effectiveness methodology demonstrated first through synthetic experiments where functions are used benchmark conditions then real-world network Vitoria, Spain (2,884 nodes, 5,799 links) using mesoscopic simulator Aimsun. Results show method leads state performances with respect Ensemble-based filters, providing improvements high 64% terms reproduction 17-fold problem dimensionality reduction.