CNN-LSTM Driving Style Classification Model Based On Driver Operation Time Series Data

This paper aims to establish a driven g style recognition m eth od that is highly accurate, fast and generalizable, considering the la ck o f d a ta types in driven style classification task a n d the lo w recognition accuracy of widely u sed u n supervised clustering algorithms and single convolutional neural network methods. First, we propose a method to collect the inform a t ion on drive r\\\'s operation time sequence in view of the imperfect driving data, a n d then extract the drive r\\\'s style features through convolutional n e u ra l network. Then, for the collected temporal data, the Lo n g S h ort T e rm Memory networks (L ST M) m od u le is added to encode and transform the driven features, to a chive the driven style classification. T h e results show that accuracy of driving style recognition reaches over 9 3 %, while the speed is improv ed significantly.