Planning mobile sensing cycles for urban data collection

This study proposes an optimization model for mobile sensing in cities, where data is collected by moving sensors rather than stationary ones. Examples include collecting on-street parking occupancy with onboard vehicle sensors and gathering pavement crack data using maintenance vehicles along assigned inspection routes. The reliability of the collected data depends on the frequency of successive visits to each location, determined by the deployment of sensors, hereafter referred to as agents. The model determines the sequence of locations to visit and the assignment of agents to cycles to achieve reliable predictive accuracy from the collected data. Three solution methods are proposed: an exact mixed-integer programming (MIP) model, a logic-based Benders decomposition (LBBD) method, and a heuristic algorithm. Numerical results obtained from the Solomon dataset show that the LBBD serves as a computational benchmark, providing faster performance than the exact MIP while delivering solutions of comparable quality. Results also indicate that increasing the sensing headway reduces the required number of agents and their total travel time. Increasing the number of cycles can enhance spatial coverage and reduce travel time per agent; however, excessive subdivision of the service area into cycles yields only marginal efficiency gains and diminishing returns in overall system performance in some scenarios. A case study in Toronto, based on spatial data on permit parking zones, demonstrates the practical applicability of the proposed framework for monitoring on-street parking occupancy.