This thesis provides an alternate online calibration approach for dynamic traffic as- signment (DTA) model. This approach has been used so far to estimate the OD demand and can be extended for other supply and demand parameters. The purpose of online calibration is to dynamically update the model parameters using the ob- served traffic flow data. The proposed approach gains the advantage against other developed approaches in terms of its computational performance and application scale. Online calibration approaches are being mostly restricted due to the prob- lems of non-linearity and dimensionality. This online calibration approach, named as PC-SPSA, combines a stochastic approximation algorithm i.e. Simultaneous Per- turbation Stochastic Approximation (SPSA) with a dimension reduction technique i.e. Principal component analysis (PCA). A set of prior estimates are used to cal- culate their variance in form of PC-directions. Then, these PC-directions are used to evaluate PC-scores of a latest previous estimate. These PC-scores are then cali- brated based on the observed traffic flows using SPSA. SPSA has been widely used as an optimization algorithm for model calibration. However, being a random search algorithm, its performance deteriorates as the problem dimension increases. The application of PCA on SPSA provides two major advantages. First, it reduces the number of variables to be estimated significantly. Secondly, it also narrows down the search area of SPSA from a higher dimensional OD flow vector to lower dimensional PC scores, improving SPSA’s performance considerably. Case studies of synthetic non-linear problems with different dimensions and a network of Vitoria, Spain are used to test the proposed PC-SPSA approach. The empirical results from these case studies show that PC-SPSA not only performs very well in reducing the error to a very low value, but it also does it rapidly with very few iterations, making it an effective online calibration approach.
«
Login
BibTeX