A hybrid neural network integrating attention mechanism for time series and non-time series multi-factor electric vehicle energy consumption prediction

Introduction In recent years, electric vehicles (EVs) have garnered increasing consumer favor due to their low energy consumption and mechanical simplicity; however, the persistent limitation of short driving range has not been fundamentally resolved and continues to fuel drivers’ range anxiety. To enhance the accuracy of EV energy-consumption prediction, this paper categorizes influencing factors from multiple perspectives and proposes a hybrid neural-network prediction model that integrates temporal features and an attention mechanism. Methods The model first partitions the dataset into time-series and non-time-series subsets based on temporal correlation. A convolutional neural network (CNN) is then employed to extract and reconstruct features from the time-series data to reduce computational complexity, after which an attention-enhanced bidirectional gated recurrent unit (AtBiGRU) further captures sequential dependencies. The resulting fitted representations, together with the non-time-series variables, are fed into a deep neural network (DNN) for ensemble learning, yielding precise energy-consumption predictions. By processing sequential and non-sequential data separately, the method effectively improves computational efficiency and model expressiveness. Results Experimental results demonstrate that the proposed CNN–AtBiGRU–DNN hybrid model achieves higher prediction accuracy and faster convergence than baseline algorithms. Conclusion The proposed model validates its effectiveness and advancement in EV energy-consumption prediction.