Neural Kalman filter for estimating dynamic velocity and headway distance in vehicle platoon system

A dynamic feedback system is developed for estimating the velocity and headway distance in a longitudinal three-vehicle platoon. The estimation system is modeled using an extended Kalman filter (EKF) and a Neural Kalman filter (NKF) that estimate the velocity and headway distance by measuring the acceleration rate of some selected vehicles in the platoon. State equations of the EKF are analytically defined by a discrete conservation equation of vehicle speed and headway distance, whereas the measurement equation is based on a conventional car-following model. The NKF, however, defines both equations using artificial neural network models (ANNs) that enable both equations to be defined without using any analytical equations. Numerical analysis showed that the NKF reduces the estimation errors in most cases even under unexpected car-following situations as the ANNs have the capability of describing nonlinear car-following phenomena. However, some difficulties still remain unsolved in optimizing NKF parameters. It was found that alternate approaches may be required to yield more accurate estimates instead of using NKF.