Path Choice Estimation in Urban Rails : Asimulation based optimisation for frequency-based assignment model

Sammanfattning: Transit system have a large importance in modern urban cities, with urban rail often acting as the central system with it efficient travel time and great capacity. As cities grow in population, so to does the usage of urban rail resulting in increased crowding on the platform and in the trains. Since crowding level is directly correlated to the experience of travel as well as a safety issue, much research has been done to improve it. Currently its common to utilise transit assignment models (TAM) to evaluate and research transit system but for them to work optimally requires weight parameters connected to perceived time spent on the journey. To get the weight parameters for a system requires surveys to be preformed which is costly and not always possible. Therefor its attractive to find these weights through optimisation using available data. Most transit system uses automated fare collection (AFC), which can be used to create origin-destination (OD) data, and automated vehicle location (AVL) together with link-load data. This project aims to develop a simulation-based optimisation (SBO) that automatically finds the weights for a frequency-based assignment model using OD and link-load as input arguments. The SBO will evaluate five different algorithm, genetic algorithm (GA), simulated annealing (SA), Nelder-Mead method (NM), simultaneous perturbation stochastic approximation (SPSA), and Bayesian optimisation (BO), using a fitness model based on KolmogorovSmirnov test. Synthetic data was implemented to evaluate the algorithms where result needed to be within a margin of error of the set weight. No algorithm was however able to converge during the simulation, therefor not optimising the weights to within the margin of error. A longer simulation was evaluated to see if the length needed to reach convergence was to short but achieved the same results. While the cause was not found, the standard deviation of the TAM could be the problem since the deviation was larger than the change of weight parameters achieved. Even if this project could not achieve its objective of developing a SBO method, it can be used for future research and work as a guide on further development on TAM research.