A two-step method for the calibration of semicompensatory models is presented. To demonstrate the use of the method, it is applied to a model that represents the process of choosing modes for work trips. The calibration of semicompensatory models, such as the one presented here, is not a trivial process because it involves finding the best set of parameters for two functions while satisfying a series of inequalities. In the example shown here, the inequalities are used to determine whether the modal choice predicted by the model corresponds to the user's choice. The best set of parameters is that corresponding to the fewest differences between the observed and predicted choices. The first stage in the proposed calibration process is a preliminary fitting, which attempts to find the maximum of a deterministic function using a process that resembles the maximum likelihood calibration method. The second stage uses the first parameters determined in the first stage as an initial solution and then tries to find the best fit through an exhaustive search around the initial guess. The justification of this two-step procedure is that the efficiency of the calibration process will be increased, since the technique used in the first stage is faster than that used in the second stage. The proposed procedure ensures that an accurate answer is obtained in a reasonable time while allowing the user to determine the sensivity of each calibration parameter. The calibrated model was able to correctly predict more than 85 percent of the modal choices observed.
Abstract