基于驾驶风格的前撞预警系统报警策略

With consideration of driving style, an optimized prediction model for longitudinal relative distance is designed, based on which an alarm strategy for the frontal-crash warning system is improved. The combination of quantile method and information entropy method is adopted for driving style classification to extract features in different ways, and k-means method is used to cluster sample data. Based on long short-term memory model, the encoder-decoder model is designed for prediction. All the data of above classifications are used to train the sharing parameters of model for improving its generalization ability, while the personalized parameters are trained with a higher learning rate by the corresponding data set of three driving styles. Utilizing the above-mentioned prediction models, the warning strategy for frontal crash based on European NCAP-AEB test protocol is improved, and as a result, the number of false alarms reduces from 123 to 50.