How can we trust in the o-d matrix correction procedure using traffic counts?

Abstract

Planning and design of transport systems are based on the application of systems of transport simulation models, whose forecast reliability and goodness of fit strongly influence the results and the quality of the planned/designed interventions. The implementation of a reliable and good system of models should be based on the disaggregate estimation of each one of the model components (supply, demand, assignment). The resulting estimated system of models should be then validated as a whole by comparing its output with corresponding observed measures (normally link flows). This validation generally fails for a number of reasons. Therefore, the observed measures are generally used to correct and to improve the model (usually the o-d matrix), normally through a GLS estimator. This procedure is so widely applied and trusted in practice that researchers and practitioners often adopt sub-models already estimated somewhere else, therefore leading to a further approximation. As a consequence, the model reliability is in practice almost entirely handled by means of the model correction through trafficcounts. In spite of that, few researchers have focused their attention ona systematic analysis of to what extent this procedure is able to correctthe whole system of models and consistently guarantee its forecast reliability.ents wherein a demand matrix, a supply and an assignment model as well as the corresponding link flows (i.e. resulting from demand assignment to the network) are all assumed to be true. This allows carrying out a series of experiments wherein both the true o-d matrix and the whole set of(observed) unbiased link flows are available for the o-d matrix correction. For instance, the following issues can be deepened: (a) given a random perturbation of the true o-d demand, check the capability of different subsets of link flows to reproduce the starting true demand through the GLS estimator; (b) introducing a random perturbation of link flows in the previous experiment so to mimic assignment/sampling errors. The paper proposes a number of different experiments of this kind in order to explore the capability of the correction procedure to correct the o-d matrix in presence of different possible biases involving each demand choice dimension (generation, distribution, mode choice), and with reference to different levels of precision in the input data (i.e. perturbations in link flows for mimicking possible supply/assignment errors and approximations). Very interestingly, some preliminary results on a real network show the o-d matrix correction procedure to provide for a reliable demand correction only in very few specific cases. For the covering abstract see ITRD E135582.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 3 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.