On the use of longitudinal data techniques for modeling the behavior of a complex system

Abstract

This work presents the use of longitudinal data analysis techniques to fit the accelerations of a real car in terms of some previous throttle pedal measurements and of the current time. Different repetitions of the same driving maneuvers have been observed in a real car, which constitute the data used to learn the model. The natural statistical framework to analyze these data is to consider it as a particular case of longitudinal data. Different fits are given and tested as a first step in order to explain the relationship between variables describing the control of the car by the driver and the final variables describing the movement of the vehicle. Results show that the approach can be valid in those cases in which a temporal implicit dependency can be assumed and in which several realizations of the experiment in similar conditions are available; in such cases an analytical model of the system can be obtained which has the ability to generalize, i.e. to show a robust behavior when faced to input data not used in the model construction phase. © 2002 Springer-Verlag Berlin Heidelberg.