In order to effectively predict the fuel consumption of vehicles, improve fuel economy and promote energy saving and emission reduction, a Hyperband-CNN-BiLSTM-based motor vehicle fuel consumption prediction method was proposed. Firstly, based on the vehicle operating status data and fuel consumption data collected from the actual road test, the salient factors affecting the fuel consumption of vehicles were analyzed. Secondly, combining the powerful feature extraction capability of convolutional neural network(CNN) and the advantages of bidirectional long and short-term memory network (BiLSTM) in dealing with the time-series data, a combined model of vehicle fuel consumption prediction based on CNN-BiLSTM was constructed. Then, in order to improve the model prediction accuracy, the combined model was optimized by Hyperband optimization algorithm, and the vehicle fuel consumption influencing factors were taken as the model input features to train the model to realize the modeling and prediction of vehicle fuel consumption. Finally, CNN, LSTM, BiLSTM, CNN-LSTM and CNN-BILSTM were selected as comparison models to evaluate the effect of Hyperband-CNN-BiLSTM prediction model. The results show that compared with other models, the Hyperband-CNN-BiLSTM model has the smallest mean absolute error (MAE) and root mean squared error (RMSE). They are 0.057 69 and 0.119 25, respectively. R² is the largest (0.991 76), and the model has the best prediction effect.