Ahead prediction of kinematics of vehicles under various collision circumstances by application of ARMAX autoregressive model

摘要

In this paper we present the application of regressive models to simulation of a full-scale vehicle-to-pole impact as well as virtual vehicle-to-barrier collision. The capability of an ARMAX model to reproduce vehicle kinematics was examined. Regressive model parameters were established by minimizing a weighted sum of squares of prediction errors. The prediction horizon was assigned to evaluate model's robustness and verify its time series data forecasting performance. It was found that the ARMAX model does not only reproduce the signal which was used for its establishment (i.e. real vehicle's acceleration) but it predicts another signal as well (i.e. virtual vehicle's acceleration). Moreover, such estimation technique preserves all characteristic information relevant for a given collision, since integration of the estimated acceleration pulse yields plots of velocity and displacement which closely follow the reference ones.