Research on Fault Early Warning and User Behavior of Electric Vehicles Based on Time Series and Machine Learning

With the global energy transition, new energy electric vehicles have become a key direction in transportation. Their complex electronic control systems and high-voltage battery components pose new safety risks. Using real driving data of a new energy vehicle from January to September, this study built a multidimensional dataset that covered the vehicle operating status, motor performance metrics, and battery parameters. It analyzed the coupling relationships between dynamic parameters, developed fault early warning models, and revealed patterns in user charging and driving behaviors. The decision tree and random forest models identified the key factors affecting fault alarms, with cumulative mileage, temperature, motor controller input voltage, and insulation resistance as the core variables. A comparative framework for decision trees, random forests, MLPs, and CNNs was established. The results showed that MLP and CNN performed excellently in the majority class (class $\mathbf{0}, \mathbf{F} 1=\mathbf{0. 9 9}$). Based on the above, user behavior suggestions were proposed: avoid charging during high-temperature periods, keep the battery SOC between $20 \%$ and $80 \%$, adjust driving behaviors, and implement preventive maintenance strategies.