Risk prediction of new energy vehicle based on dynamic-static feature fusion

To support the goals of low-carbon and sustainable development, new energy vehicles (NEVs) are being increasingly adopted. However, the frequency of traffic accidents involving NEVs also shows a rising trend. To address this challenge, this paper proposes an accident risk prediction method for new energy vehicles based on dynamic-static feature fusion. First, direct and indirect data strongly related to accident risk are extracted from the full-year accident data of a province in 2021, including environmental factors (weather and road type), dynamic operating data (speed), vehicle alarm status, and historical accident characteristics. Then, to quantify and capture the potential risk characteristics of the vehicle, LSTM layers are used to construct dynamic and static feature vectors representing vehicle accident risk. Moreover, the accident risk probability is calculated based on fully connected layers and the sigmoid activation function. Finally, the proposed accident risk prediction model is tested and validated with real accident data. The results show that the model achieves a prediction accuracy of 85% for new energy vehicle accidents, which is a 24% improvement over traditional models based on weather and road types. The model can timely warn drivers before accidents occur, helping them take necessary safety measures to reduce accident probability.