Research on Multi-Mode Power Management Technology

To address the diversified power consumption demands arising from differences in devices and tasks under low-power, multi-mode operating conditions of flight control (FC) systems, this study proposes a historical data-driven work state prediction method. Considering the inconsistent data characteristics across various flight modes, the method employs Long Short-Term Memory (LSTM) networks to learn from partial power consumption and environmental data. To further enhance prediction accuracy and generalization, adaptive hyperparameter tuning techniques are introduced to dynamically adjust the network structure and training parameters. A flight control system power simulation platform, integrated with external sensors such as an anemometer, hygrometer, and barometer, is used to construct a multi-mode dataset. Experimental evaluations are conducted on this dataset, with performance measured using metrics such as Root Mean Square Error (RMSE) on the validation set. The results demonstrate the robustness, accuracy, and feasibility of the proposed method for predicting operating states and optimizing power consumption in flight control systems.