Qurashi, M., Lu, Q., Cantelmo, G., & Antoniou, C. (2020) PC-SPSA: Exploration and assessment of different historical data–set generation methods for enhanced DTA model calibration. In 9th Symposium of the European Association for Research in Transportation (hEART2020).
Date:
This paper explores multiple historical data-set estimation methods which are crucial for the calibration performance for principal component analysis (PCA) based algorithms. We first propose multiple sets of historical data-set generation methods with probable calibration scenarios (which replicate more realistic changes within the structure of the demand) and later explore the performance of all the proposed historical data-sets with PC-SPSA to understand the importance of different historical data-set generation parameters. As per the current results, more correlatedly generated historical estimates (i.e. method 3 and 6) outperform other simplified techniques but it will be further interesting to explore and analyze their performance calibrating other different sets of scenarios. Next steps, to be shown in hEART2020 conference, will include, the exploration of all the proposed methods on the possible demand scenarios to identify the best most generically wellperforming data-set generation technique, and later validating that technique on a larger network of Munich city (with a network of 8689 links, 706 detector location and demand of OD matrix [73 × 73] or 5329 OD pairs) with different demand scenarios and also other network information e.g. travel times etc. Finally, results proposed in this study are still based on synthetic experiments. This is a limitation, as we aim to test PCA based algorithms when historical data sets are not available (or information is not reliable). To do so, we will use real traffic data from Munich to generate a benchmark scenario that is assumption free - e.g. the “true” network state is derived from real data and not from syntetic functions. This will allow us to validate our probability function against real data in an assumption free scenario.