Telemetry-driven physics-informed prediction for on-orbit thermal–electrical performance of solar panels in satellite power system

For small satellites and CubeSat missions, body-mounted solar panels serve as the primary and often sole source of onboard power. Accurate prediction of the on-orbit thermal–electrical performance of small satellite solar panels is critical for power system management and health monitoring. However, it remains challenging due to complex orbital environments, unstable attitude motion, and limited availability of high-fidelity telemetry information. This paper proposes a physics-informed machine learning (PIML) prediction model for satellite solar panels under uncertain orbital environment conditions. The proposed approach avoids reliance on detailed flight information by converting measured telemetry into physically admissible information. A telemetry-driven orbital environment inversion strategy is developed to infer the solar panel’s thermal environment from real-time telemetry data. Based on the equivalent environmental states, a thermal–electrical soft constraint model is constructed to define a limiting scope of operation. The resulting physics-informed constraints are fused with telemetry data and used as hybrid inputs to PIML prediction models, encoded by recurrent neural networks (RNN) and long short-term memory networks (LSTM). The prediction model is evaluated using real-time telemetry from the TURUS satellite. Multiple prediction scenarios are designed, and the results demonstrate that the PIML prediction model is stable, physical consistent, and generalized. This study provides a practical solution for on-orbit thermal–electrical performance prediction under telemetry-limited conditions.