Neural Spectral Clustering Based Voltage Area Partition of Active Distribution Systems

As distribution systems evolve to accommodate large-scale renewable energy sources, maintaining voltage stability becomes increasingly challenging. Network partitioning plays a pivotal role in voltage control tasks, especially in active distribution systems (ADSs). By partitioning the network into manageable small sub-networks, i.e., voltage area partition (VAP), fine-grained, decentralized, and coordinated voltage control can be realized, which prevents over-voltage or under-voltage issues and facilitates the integration and absorption of renewable energies. However, because of the weak ability to extract complicated voltage relationships, existing naive graph clustering VAP methods are likely to suffer a performance bottleneck in voltage cohesiveness for large-sized distribution networks. Therefore, this paper proposes a neural spectral clustering-based VAP method for ADSs. Specifically, a network partition problem is solved by clustering a neural spectral mapping of multi-phase voltage coupling features. Theoretical and experimental results show that the proposed method can partition the network with voltage cohesiveness higher than that of the standard spectral clustering method while bringing certain advantages in computational efficiency.