An adaptive prediction method based on VMD-GRU for future driving condition of vehicle

Driving conditions prediction plays an important role in energy-saving control strategy for electric vehicle. However, the complexity of the changes in road conditions poses a great challenge to the accurate prediction of driving condition. To address this problem, this paper proposes an adaptive Sliding Window (SW) and Gated Recurrent Unit (GRU) algorithm to predict the driving conditions in a short period, and the algorithm enables to adjust the size of the SW promptly when the driving conditions change frequently. A smaller window is adopted in the case of drastically changing speeds, and a larger window is adopted in the case of smooth speeds. Firstly, Principal Component Analysis (PCA) and k-means clustering are used to construct sample driving conditions with the same characteristics. Then the instantaneous frequency of the sample driving conditions is calculated by Hilbert transform and Variational Mode Decomposition (VMD), and the optimal window size applicable to the conditions with different frequencies is quantitatively calculated. The model provides precise predictions with root mean square error (RMSE), mean absolute error (MAE) and mean absolute percentage error (MAPE) of 0.8799, 0.5443 and 0.8362%, respective. The ablation experiments show that the improved SW enables the GRU prediction model to capture the trends of different driving conditions more accurately, and improves the accuracy and robustness of the prediction model.