An entropy maximising model for estimating trip matrices from traffic counts
The main objective of this research is to develop and test a technique for estimating trip matrices from traffic counts. After discussing conventional methods for obtaining trip matrices an analysis is made of the problem of estimating them from traffic counts: it is found that in general the problem is underspecified in the sense that there will be more than one trip matrix which, when loaded onto a network, may reproduce a set of observed counts. A review is made of some models put forward to estimate a trip table from volume counts, the majority of them based on a travel demand model. A new model is then developed by the author within an entropy maximising formalism. The model may be interpreted as producing the most likely trip matrix consistent with the information contained in the counts and a prior trip matrix if available. This model does not require counts on all links in the network, can make efficient use of outdated trip matrices and other information, and is fairly modest in computer requirements. The model is then tested against real data collected by the Transport and Road Research Laboratory in the central area of Reading. Considerable temporal variability was found in the sampled trip matrices. The matrices estimated by the model are not very close to the observed ones but their errors are in general within the daily variations of the sampled matrices. A number of tests on the sensitivity of the model to errors and availability of traffic counts and route choice models used are also reported. A technique has been developed to rank links according to their potential contribution to the improvement of an estimated trip matrix. This scheme may be used to select new counting sites. The availability of a reasonable prior estimate of the trip matrix considerably improves the accuracy of the origin-destination matrix generated by the model. Some suggestions for extensions and further research are presented towards the end of this work.