The Simultaneous Perturbation Stochastic Approximation (SPSA) algorithm has been used for solving the off-line dynamic origin-destination (OD) estimation problem. While the algorithm can be used with very general formulations of the problem, it can also be unstable. The paper proposes methods and evaluates their effectiveness in improving the SPSA performance at two levels: a) scaling the step size and using a hybrid gradient estimation; and b) proposing alternative clustering strategies to be used with the c-SPSA version of the algorithm, where OD flows are estimated in clusters. The proposed enhancements are evaluated through simulation experiments on a real network.
Tympakianaki, A., Koutsopoulos, H. N., & Jenelius, E. (2018). Robust SPSA algorithms for dynamic OD matrix estimation. In Procedia Computer Science (Vol. 130, pp. 57–64). Elsevier B.V. https://doi.org/10.1016/j.procs.2018.04.012