Vehicle Road Grade Prediction Based on CNN-LSTM

Abstract

Road grade is an important parameter for application of modern vehicle intelligent control technology. At present, the estimation methods of road grade are mainly based on the vehicle longitudinal dynamics or sensor information. However, for practical applications, relying only on the recognition of road grade, without map information, will be a post estimation, which is not conducive to the implementation of intelligent control strategies. To meet the needs of predictive intelligent control algorithms for vehicle, it is necessary to predict future road grade changes. This study considers the road grade variation experienced by the vehicle as a time series and proposes a prediction method that combines a convolutional neural network (CNN) and a long short-term memory neural network (LSTM). The road grade information collected by the vehicle driving sensors is used to train the designed network to predict the future road grade and the prediction model is compared with LSTM prediction model. The results of experimental data analysis show that the method designed in this study is superior to LSTM prediction model, and the MAPE (Mean Absolute Percentage Error) of the road grade prediction after 5 s is 11.6 %, which is 15.1 % lower than the MAPE of the LSTM prediction result. The generalization ability and computational time of the model are also compared, and the CNN-LSTM demonstrated better performance.