Observability of Origin-Destination matrices for Dynamic Traffic Assignment

Abstract

The estimation of dynamic Origin-Destination (O-D) matrices from aggregated sensor counts is one of the most important and well-researched problems in Dynamic Traffic Assignment (DTA) systems. In practice, more often than not, number of sensors are far less than the number of potential O-D pairs, and hence this problem is modeled in an optimization framework as function of historical estimates of O-D flows. However, in the absence of reliable historical O-D flows, it is critical that O-D estimation module is observable. Observability is defined as a property of the system by which it is possible to uniquely determine the (initial) state (O-D flows) of the system eventually by making regular indirect measurements of the state. In DTA systems, observability implies that given enough sensor data, it is possible to uniquely determine O-D flows without any prior information about them. This thesis develops a methodology to verify the observability property of the O-D estimation model given limited sensor coverage on the network. A case study involving a large-scale network from Los Angeles, California is used to demonstrate the feasibility of the proposed approach.

(cont.) A comprehensive off-line calibration exercise for the same network is then used to verify the validity of the conclusion.