Fuzzy Binary PSO for Traffic Sensor Location Problem with Error-Propagation Control for Large Scale Networks

This study addresses the Traffic Sensor Location Problem for complete link-flow observability under non-uniform sensor measurement uncertainty. The proposed framework minimizes the accumulated error propagated from observed link flows to inferred unobserved link flows while preserving the structural conditions required for complete network observability. Its methodological novelty lies in combining a structured new-link selection procedure with a fuzzy-enhanced Binary Particle Swarm Optimization (FBPSO) algorithm that adaptively balances exploration and exploitation. An ILU-preconditioned GMRES procedure is also incorporated to efficiently solve the sparse linear systems generated during the evaluation of candidate sensor configurations. The proposed framework is evaluated using the Fishbone and Sioux Falls benchmark networks and the large-scale Austin transportation network, which contains 7388 non-centroid nodes and 18,961 directed links. Its performance is compared with standard Binary Particle Swarm Optimization (BPSO) and the Binary Bat Algorithm (BBAT) under uniform and non-uniform measurement-error conditions. For the Fishbone network, all three methods reach the same minimum accumulated inference error of 89.21, indicating agreement on the best solution for this small test case. For the Sioux Falls network, FBPSO obtains an inference error of 634.46, compared with 641.78 for BPSO and 640.94 for BBAT. For the Austin network, FBPSO achieves the lowest final inference error and continues improving after the comparison methods reach prolonged plateaus. These findings demonstrate that the proposed framework provides an effective and scalable approach for uncertainty-aware traffic-sensor placement and reliable network-wide link-flow inference.