Qurashi, M., Lu, Q., Cantelmo, G., & Antoniou, C. (2020). PC-SPSA: Implementation assessment and exploration of different historical data-set generation methods for enhanced DTA model calibration. In 3rd Symposium on Management of Future Motorway and Urban Traffic Systems (MFTS).
Date:
Calibrating DTA models is complex due to the involved undeterminedness, non-linearity, and dimensionality, restricting calibration approaches especially when calibrating larger networks. Simultaneous perturbation stochastic approximation (SPSA) has been proposed for the DTA model calibration, with encouraging results, for more than a decade with multiple variants trying to improve its application scalability on larger networks. Recently, PC-SPSA has been proposed, combining Principal Component Analysis (PCA) with SPSA to reduce the problem dimensions and non-linearity by limiting the search space in lower dimension space based on orthogonal Principal Components evaluated upon a set of historical estimates. In this paper, we further explore PC-SPSA implementation by assessing its sensitivity towards SPSA parameters definition, its performance in calibrating synthetic problems of different dimensions and non-linearity, and formulating multiple OD historical data–set generation methods for improved calibration (in case of non-existent or irrelevant historical estimates). The performance of each method is compared calibrating an urban network of Munich with similar PC-SPSA settings, depicting more correlated generation techniques perform better consistently than simplified ones.