The operational safety and endurance of unmanned aerial vehicles (UAVs) are strongly affected by lithium-ion battery degradation under extreme thermal environments. However, conventional physics-based models often rely on simplified assumptions, whereas purely data-driven methods usually lack physical interpretability and robust generalization. To address these limitations, this study proposes a Physics-Informed Deep Neural Network (PIDNN) for predicting UAV battery degradation under complex environmental conditions. The proposed framework integrates thermodynamic and fluid dynamic principles with deep neural networks by incorporating physical constraints derived from heat generation, heat conduction, and convective heat transfer into the loss function. This design enables the model to capture nonlinear degradation patterns while maintaining consistency with fundamental physical laws. Comprehensive simulation-based experiments were conducted under high-temperature (45 °C), low-temperature (−20 °C), and room-temperature (25 °C) conditions, together with varying discharge rates, humidity levels, wind speeds, and multi-factor coupled scenarios. The results show that the proposed PIDNN consistently outperforms conventional physics-based models and several representative data-driven methods, including SVM, LSTM, and GAN-based approaches. It achieves lower prediction errors across all evaluated conditions, as reflected by reduced mean absolute error and root mean square error. By providing physically consistent predictions of capacity fade, internal resistance growth, and remaining useful life, the proposed framework supports degradation-aware monitoring and early warning for intelligent battery management systems. These findings provide a robust methodological basis for improving the reliability, safety, and service life of UAV power systems operating in complex climatic environments.
Duan et al. (Wed,) studied this question.