Iterative learning of stochastic disturbance profiles using Bayesian networks

In this paper we present an iterative method for learning data of stochastically occurring disturbances using Bayesian networks. Our methodology can be used for learning the complete disturbance profile of a given road segment by processing information gathered from multiple passages of road vehicles over the given segment. After the learning process the data can be used to predict disturbances during a new passage using inference in Bayesian networks. By means of this information the driving performance is to be improved. We test this new method on an X-by-wire test vehicle called “Chameleon”. The iterative learning method is applied to a quarter-vehicle model of this innovative vehicle, which is sufficient for the purpose of evaluation. We have also used an observer to estimate system states that cannot be measured directly. The results achieved with our learning method show, that the occurrence or non-occurrence of disturbances can be predicted correctly in 90% of the analyzed cases.