Intelligent Spectrum Allocation Based on Deep Reinforcement Learning for Power Emergency Communications

In order to improve system performance of power emergency communication systems, this paper studies an intelligent spectrum allocation scheme based on multi-agent reinforcement learning (MARL) to allocate limited spectrum resources to different users according to their spectrum requirements. When the users access communication channels, whether the communication is successful is judged according to the channel feedback information, which provides rewards for learning training process. A spectrum allocation scheme based on MARL is proposed to intelligently share the limited spectrum resources among different users. Simulation results show that the proposed MARL scheme can achieve better system performance compared to traditional reinforcement schemes such as Deep Q Network (DQN). The proposed scheme provides an efficient spectrum usage paradigm for power emergency communications.