Role-Dependent Mission-Energy Reserve Control for Federated Learning in Solar-Powered Maritime Edge Networks

Abstract

Federated learning (FL) enables collaborative maritime perception without transferring raw observations, but solar-powered clients must preserve energy for platform-specific duties. Existing energy-aware FL generally applies a common battery constraint and cannot jointly protect persistent buoy service and the safe return of mobile nodes. This paper presents a mission-safe FL framework using risk-aware, role-dependent mission-energy reserves. It constructs a service-continuity reserve for every client, adds a safe-return reserve for mobile clients, and exposes only the remaining energy to joint client–workload control and safety-gated transmission. Execution-time checks prevent FL actions from drawing protected mission energy. A trace-driven evaluation combines maritime mobility and link traces generated with the Mission-Oriented Operating Suite–Interval Programming (MOOS-IvP), National Solar Radiation Database (NSRDB) irradiance records, and Jetson Orin Nano workload profiles. The proposed scheme achieved a final mean intersection over union (mIoU) of 87.11%, delivered 94.4% of scheduled updates, and attained a learning-energy efficiency of 0.061 percentage points per Wh. It reached 95% of the independently trained centralized reference performance in 9.32 days. Stationary service shortfall, mobile service shortfall, and safe-return violation were 0.56%, 0.44%, and 0.77%, respectively. Ablation and sensitivity results confirm that role differentiation, uncertainty-aware reserves, and client-side safety checks jointly sustain learning while protecting heterogeneous maritime missions. Operationally, the framework gives mission managers a tunable means of admitting FL only after continuity and recovery energy have been protected, while empirical coverage monitoring indicates when the reserve model requires recalibration.