Methodology for efficient real time OD demand estimation on large scale networks
Tamara Djukic, Hans van Lint, Serge Paul Hoogendoorn · Data Archiving and Networked Services (DANS) · 2014
In previous work, we have explored the idea of dimensionality reduction and approximation of OD demand based on principal component analysis (PCA). In particular, we have shown how we can apply PCA to linearly transform the high dimensional OD matrices into the lower dimensional space without significant loss of accuracy. Next, we have defined a new transformed set of variables (demand principal components) that is used to represent the OD demand in lower dimensional space. These new variables are defined as state variable in a novel reduced state space model for real time estimation of OD demand. In this paper, we review previous work and continue this line of research. Based on the previous results, we demonstrate the quality improvement of OD estimates using this new formulation and a so-called, ’colored’ Kalman filter approach for OD estimation, in which correlated observation noise is accounted. Moreover, we provide a thorough analysis of the model performance and computational efficiency using real data from a large network, and method for obtaining a reduced set of state variables.