Two-Stage Stochastic Model for Sensor Location Problem in a Large-Scale Network

Uncertainty is one of the major factors that transportation system analysts and planners have to deal with in making transportation decision. Finding a set of optimal sensor locations under uncertainty is a network design problem. This paper presents a bi-level framework to solve this problem. In the upper level, the traffic planner makes decisions on the sensor deployment in the network, to maximize the OD flow coverage and minimize the expected uncertainty of the estimated OD demand subject to the budget limitation. In the lower level, the network users make choices on the time dependent user equilibrium (DUE) routes given the sensor locations determined from the upper level and are subject to the random occurred incidents. A two-stage stochastic bi-objective model is formulated which considering link information gains (weights of each link brought to correct a-prior origin-destination flows) and flow coverage as two objectives to locate passive point sensors in a sensor network subject to budget constraint. In first stage, a set of sensor locations is identified under the normal traffic condition subject to the budget limitation, a recourse decision in the second stage is then made to correct the locations based on the locations, severity and durations of incidents occurred in the network. Recognized the location problem as a NP-hard problem, an iterative heuristic procedure, hybrid Greedy Randomized Adaptive Search Procedure (GRASP) is employed to find the near-optimal solution for this problem. A traffic simulator, DYNASMART-P, is used to propagate the vehicles along their paths in the network. The proposed approach is tested on the CHART (Washington DC- Baltimore, Maryland corridor) network and the results are analyzed.