Optimal network planning of AC/DC hybrid microgrid based on clustering and multi-agent reinforcement learning

According to the characteristics of an AC/DC hybrid microgrid, this paper presents a new method for network planning of an AC/DC hybrid microgrid based on clustering partition and multi-agent reinforcement learning algorithm. The planning method is divided into three steps, namely, sub-microgrid partition, sub-network planning, and main network planning. On the basis of the principle of partition, a clustering partition model considering the location, type, and capacity of distributed generators (DG) and load is established, to divide this system into different AC and DC sub-microgrids. Then, the optimization model of the sub-network considering the annual cost of the converter and the optimization model of the main network considering the change of network loss of the sub-microgrid are proposed, which form a bilayer optimization model of the AC/DC hybrid microgrid. The optimal clustering partition is obtained by combining the k-means clustering algorithm with the particle swarm optimization algorithm, and considering the interaction between main network planning and sub-network planning, the network planning model is solved by the multi-agent reinforcement learning algorithm based on the Stackelberg equilibrium game. A numerical example is given to verify the validity and accuracy of the above models and algorithms.