Deep Reinforcement Learning-Based Energy-Efficient Sensor Scheduling for Remote State Estimation in Islanded Microgrids

This paper investigates the issue of remote state estimation in islanded microgrids (ImGs), considering the limited transmission resources. Specifically, to effectively monitor the internal state of the ImGs, measurements of multiple sensors are transmitted to a remote workstation via wireless channels for state estimation. However, this process is constrained by two factors stemming from the limitations of transmission resources. One such factor is the limited network bandwidth, which allows only a subset of the sensors' measurements to be transmitted at each time step. Another factor is the restricted transmission energy of the sensors, which requires each sensor to select an appropriate transmission power level at each moment, with different power levels resulting in varying packet loss rates. To reduce transmission energy while achieving high-precision state estimation, the optimal transmission scheduling problem for multiple sensors must be addressed. In this paper, the problem is first modeled as a Markov decision process (MDP) with a finite action space. Subsequently, a new sensor scheduling algorithm is proposed, based on a dueling double deep Q-network, to address this MDP. Finally, numerical simulations on an ImG with two distributed generation units are conducted, demonstrating the proposed scheduling algorithm's superior effect compared to established methodologies.