Dynamic spectrum sharing based on federated learning in smart grids and power communication networks

Abstract

As the guaranteed basis for providing communication services, the power communication network plays a vital role in the smart grid. However, during natural disasters, wired communication networks have inherent limitations and come with substantial construction and maintenance costs, which makes it difficult to function effectively. Therefore, it is imperative to apply wireless communication to smart grids and power communication networks in emergency scenarios. To solve the problems of spectrum resource scarcity and insufficient spectrum utilization in wireless communication, the integration of cognitive radio networks (CRNs) into smart grids and power communication networks is considered, which can effectively solve the problems and promote their development. Based on the deep reinforcement learning (DRL) and federated learning (FL) algorithms, this paper proposes a novel dynamic spectrum sharing framework which is applied to smart grids and power communication networks in emergency scenarios. In the proposed framework, the maximum entropy based multi-agent actor-critic (ME-MAAC) algorithm is used as the local learning model, which can not only improve system performance but also help power users to choose an optimum dynamic spectrum sharing strategy. It can be seen from the simulation results that the proposed scheme has better performance in reward value, access rate, and convergence speed.