Short-term vehicle speed prediction based on BiLSTM-GRU model considering driver heterogeneity

Short-term vehicle speed prediction is an essential part of Intelligent Transportation Systems (ITS), which influences the critical parameter for high-level energy management of electric vehicles. Accurate predictions of vehicle speed contribute to take timely countermeasures and enhance energy application efficiency. Deep learning is a hot research method in current prediction, which can already accurately predict vehicle speed. However, the prediction accuracy of the fixed algorithm is difficult to further improve after reaching a certain accuracy, and overfitting may occur in the process of improving the prediction accuracy. At the same time, driving behavior of drivers will affect the prediction effect to varying degrees. In order to verify the difference of speed prediction under different driving characteristics, a hybrid prediction model K-BiLSTM-GRU is proposed, which is combined the adaptive ability of K-means to reasonably classify samples and the advantage of bidirectional long short-term memory network (BiLSTM) and gated recurrent unit (GRU) to solve long-range dependencies and reduce overfitting. Firstly, a two-step method is used to denoise the NGSIM dataset, and K-means clustering method is used to cluster the data related to the car-following (CF) teams in the selected lane. After obtaining three types of drivers, the driving characteristics of the different types of drivers are analyzed. Secondly, the construction, training and prediction of the neural network is completed in the deep learning framework Keras. Finally, the model performance of verified by vehicle speed prediction through the actual speed dataset. The proposed hybrid model is compared with lots of current mainstream deep learning algorithms, the effectiveness of the K-BiLSTM-GRU method is validated. Meanwhile, the prediction performance of timid drivers is better than that of aggressive and neutral types. The results may provide some potential insights for vehicle speed prediction and electric vehicle energy consumption about different driving characteristics.