Identification of network sensor locations for estimation of traffic flow

This paper addresses the network sensor location prohlem (NSLP) for identifying the set of sensor locations that minimizes the variability in estimation of traffic flow given budget constraints. The trace of the covariance matrix is adopted as a measure of variability in traffic flow. On the basis of the trace of the covariance matrix in the posterior estimation of traffic flow conditional on a given set of sensor locations, the general form of the NSLP is derived. As an illustration, the multivariate normal distribution for the prior estimation of traffic flow is assumed. In this case, the actual value of the counted flows is not required. Furthermore, an incremental method that can avoid matrix inversion and give priorities of the identified sensor locations is presented to solve the NSLP. Finally, a numerical example based on the Nguyen-Dupuis network illustrates the NSLP approach and clarifies some of its implementation details.