In the very near future, intelligent cruise control systems will be introduced to the automotive market. These systems will include sensors which can measure the distance to the vehicle in front. Using the distance measurement, an appropriate setpoint for the vehicle speed can be autonomously selected. The end goal is to automatically maintain a safe inter-vehicle spacing. This functionality has to work over the full range of vehicle speed. A number of empirically derived parametric models of longitudinal vehicle dynamics are compared. The vehicle concerned is a fully instrumented Mercedes-Benz commercial lorry. The main focus is on a class of models known as local model networks. In this structure, a number of simple local linear models are combined (interpolated) by scheduling on a number of physical variables which are known to, or can be found to, capture the system nonlinearities. Following extensive experimentation involving the use of constructive search for operating regime decomposition, it is concluded that scheduling on gear and throttle angle leads to the best overall model. Local first-order linear ARX models are found to be sufficient. For comparison, a nonlinear model of Hammerstein type and a multi-layer perceptron neural-network are also identified.
Constructive empirical modelling of longitudinal vehicle dynamics using local model networks
Empirische Modellbildung des Fahrverhaltens eines Kraftfahrzeuges mit in der kleinen Arbeitspunktumgebung gültigem neuronalem Netz
Control Engineering Practice ; 4 , 2 ; 167-178
1996
12 Seiten, 11 Bilder, 11 Tabellen, 11 Quellen
Article (Journal)
English
Modelling longitudinal vehicle dynamics with neural networks
Taylor & Francis Verlag | 2020
|Towards Constructive Modelling and Control with Neural Networks
British Library Conference Proceedings | 1995
|Wiley | 2012
|Springer Verlag | 2011
|Springer Verlag | 2006
|