One of the most important fundamental challenges in islanded microgrids (IMs) is frequency stability.
The reduction in inertia and damping in this system is due to the increasing penetration of renewable energy sources.
Virtual inertia control (VIC) is a solution to improve frequency stability in IM.
Control methods that have been used for VIC in IM have problems with parameter uncertainty and disturbances and are not adaptable.
The results indicate that the proposed method has achieved an average improvement of 42% in the microgrid frequency deviation and 52% in the settling time of the frequency deviations.
One of the most important fundamental challenges in islanded microgrids (IMs) is frequency stability. The reduction in inertia and damping in this system is due to the increasing penetration of renewable energy sources. Virtual inertia control (VIC) is a solution to improve frequency stability in IM. Control methods that have been used for VIC in IM have problems with parameter uncertainty and disturbances and are not adaptable. In this paper, a new method based on improved deep reinforcement learning (IDRL) has been used in the VIC structure, which is obtained by combining three components: (1) The TD3 algorithm has been used to train frequency stability in IM, (2) The hierarchical attention mechanism has been used to focus on the microgrid frequency deviation and reduce the impact of disturbances on it, (3) Meta-learning has been used to quickly adapt to changes in system parameters and disturbances. The proposed method (IDRL) has been compared in several scenarios with the ANN-PI controller and BEL controller in terms of performance against disturbances and parameter uncertainty. The results indicate that the proposed method has achieved an average improvement of 42% in the microgrid frequency deviation and 52% in the settling time of the frequency deviations.