On-line calibration of behavior parameters for behavior-consistent route guidance

Abstract

This paper calibrates on-line the parameters of a controller-estimated driver behavior model used in a deployable behavior-consistent approach for real-time route guidance by checking the consistency between the time-dependent actual and estimated system states. The behavior model has a fuzzy multinomial logit structure where the systematic utility component is obtained using aggregate behavioral if–then rules. The weights of these rules are calibrated through a fuzzy on-line calibration model using the unfolding traffic volume measurements. The on-line calibration is done within the deployment framework of the behavior-consistent approach where the drivers’ likely response is factored in determining the route guidance strategies. The generalized structure of the calibration component enables it to simultaneously incorporate other sources of state inconsistency such as traffic flow model parameters. The results indicate that the calibration model can enhance the accuracy of system state estimation, leading to the increased effectiveness of the behavior-consistent route guidance. It provides the ability to more accurately predict drivers’ likely route choices by using aggregate if–then rules, and consequently, aggregate level data. This is attractive in a deployment context as it implies reduced data needs at a disaggregate level, a difficult proposition in the real world.