Investigating the Performance of SPSA in Simulation-Optimization Approaches to Transportation Problems


While optimization models play a key role in transportation analysis, the objective function to be optimized, however, cannot be defined analytically. It is therefore necessary to resort to non-differentiable optimization methods that usually pivot on evaluating the objective function. Special cases of particular interest are those in Dynamic OD Estimation, which cannot evaluate the objective function analytically and thus the formulation falls in the computational framework of Simulation-Optimization. SPSA is not limited to the inputs from conventional traffic counts and can be easily extended to account for the measurements of traffic variables supplied by emerging sensors exploiting Information and Communication Technologies (ICT). Numerical experiments have been conducted, and the results have been analyzed from two different perspectives: performance and solution quality. This allows understanding the behavior of the SPSA algorithm and new variants, which altogether contribute to the aim of adding ICT measures in the future. Their sensitivity to the initial values, the effect of bounding the variables and scaling techniques are analyzed. This paper will report on the results of the numerical experiments, their analyses, conclusions and further research.

Transportation Research Procedia