Calibration of parameters for a combined gravity and traffic assignment model

The transport planning process usually consists of four stages, trip generation, trip distribution, modal split and traffic assignment. This paper deals only with road traffic, so that there is no modal split. It describes a method for calibrating, from observed traffic counts, the parameters of a joint model, combining trip generation and distribution with traffic assignment. The objective is to minimise the total deviation of the values F estimated by the model cl(f, F), from the observed values f. A maximum likelihood model is formulated to calibrate the parameters and thus estimate the values F. The sampling distribution of traffic counts on each individual link is assumed to be Poisson. The global solution to calibrate the relationship cannot be found for this non-convex problem using conventional techniques such as hill-climbing. A genetic algorithm (GA) is introduced to reach a point close to the global optimum, and then Newton's method is used to search for an optimal or nearly optimal solution. Finally, three aggregate measures are applied to test the goodness of fit of the calibrated model. Tests for the individual parameters are performed using the Wald statistic.