Energy consumption prediction strategy for electric vehicle based on LSTM-transformer framework

As the market share of electric vehicles (EVs) continues to grow, driving range capability has emerged as a primary concern for drivers, car manufacturers, and policymakers. Accurate real-time energy consumption prediction is essential in mitigating range anxiety and fostering the adoption of EVs. Consequently, an EV energy consumption prediction framework that comprehensively considers vehicle and environmental factors, with special attention to individual driving styles and driving conditions, constructed based on long short-term memory (LSTM) and Transformer models is proposed. This framework addresses the issue of long-term dependencies and captures characteristics of time-series data with increased efficiency. Upon rigorous evaluation, the model achieved the mean absolute percentage error (MAPE) of 4.63 % for state of charge (SOC) of EV's battery in experimental dataset, outperforming LSTM and multivariate regression (MLR). The ablation experiment shows that the MAPE of the model is reduced by 18.47 % and 15.27 % respectively after considering the individual driving style of drivers and the driving conditions of vehicles for both two types of vehicles. Based on this framework, a long-distance EV energy consumption prediction strategy based on short-distance is proposed, with a MAPE of 6.7 % when SOC value is reduced from 90 to 40 in the selected dataset.