Wireless Federated Learning Based Building Temperature Estimation With Latency Constraint

The paper proposes a novel approach for temperature estimation in buildings using wireless federated learning (FL) while considering latency constraints. The proposed model utilizes a hierarchical federated learning architecture within a wireless network, incorporating base stations (BS), access points (APs), and user equipment (UEs). Each UE performs local learning and shares model updates with APs, which aggregate them and forward them to the BS for final aggregation. The system aims to minimize both latency and energy consumption while ensuring accurate temperature predictions. Simulation results show the effectiveness of the proposed scheme in comparison to deep reinforcement learning (DRL) and genetic algorithm (GA) approaches. Specifically, at a latency threshold of 10 seconds, the proposed scheme achieves a prediction accuracy of approximately 0.60, while DRL reaches 0.50 and GA stays around 0.48. These results highlight the superior performance of the proposed federated learning-based method, especially in high-latency scenarios, and demonstrate its potential for real-time applications in smart building environments under wireless communication constraints.