Optimal Selection of Traffic Sensors: An Information-Theoretic Framework

This paper presents an information-theoretic framework for the optimal selection of sensors across a traffic network. For the selection of sensors a set covering integer programming (IP) problem is developed. A measure of correlation between random variables, reflecting a variable of interest, is introduced as a “distance” metric to provide sufficient coverage and information accuracy. The ultimate goal is to select sensors that are most informative about unsensed locations. The Kullback-Leibler divergence (relative entropy) is used to measure the dissimilarity between probability mass functions corresponding to different solutions of the IP program. Efficient model selection is a trade-off between the Kullback-Leibler divergence and the optimal cost of the IP program. The proposed framework is applied to the problem of developing sparse-measurement traffic flow models with empirical inductive loop-detector data of one week from a central business district with about sixty sensors. Results demonstrate that the obtained sparse-measurement rival models are able to preserve the shape and main features of the full-measurement traffic flow models.