Hybrid AC/DC microgrid can effectively coordinate the power distribution of microgrid through mutual power support between two sub-grids, and improve the system’s ability to suppress power fluctuations. In order to realize the reasonable mutual power support between the two sub-grids, this paper proposes a mutual power support strategy for hybrid AC/DC microgrid used in island operation mode. Firstly, in order to avoid the power loss caused by the frequent actions of inter-linking converter, a hierarchical control strategy is proposed, which includes power autonomy model and power interaction model, and the changing between system operating models is designed reasonably. Then the goal of mutual power support based on the conditions of the sub-grids and the state of charge (SOC) of the battery is proposed. Meanwhile, a power cooperation control algorithm, which gives consideration to both AC frequency and DC voltage, is designed to achieve the goal. A simulation model is built in PSCAD/EMTDC. The simulation results show that, with the mutual power support strategy, the AC and DC sub-grids can bear the power fluctuations of the system according to their own conditions. And reasonable SOC of the battery can be maintained.

In order to improve the frequency anti-interference of the island microgrid, a coordinated frequency control method based on reinforcement learning for island microgrid is proposed. The proposed control method performs Q-learning on the basis of the frequency deviation of the microgrid, and dynamically adjusts the droop control parameters of multiple distributed power sources to change their output power, to realize multi-source coordinated active frequency control in the microgrid. Firstly, the principle of Q-learning algorithm is introduced. Secondly, a frequency recovery control method based on Q-learning is proposed, and a controller based on the Q-learning algorithm is set up. The Q-learning algorithm is used to dynamically correct the droop parameters and coordinate multiple distributed power sources in the microgrid for frequency recovery control. Finally, MATLAB is used to establish a typical microgrid simulation model, and on the basis of S-function, a self-defined reinforcement learning controller is established to verify the effectiveness and adaptability of the proposed method from the aspects of the Q-learning training process and frequency control response characteristics.