Research on data-driven combined network reconfiguration and local control with smart inverter for voltage regulation problem

The combination of network reconfiguration (NR) and local control has become a promising strategy for voltage regulation arising from the high penetration of photovoltaics into the grid. However, it is challenging to implement in distribution systems due to the lack of infrastructure. Moreover, smart inverters (SI) serve as local controllers, typically in voltage regulation with flexible control but hard to design its rule with only local observation to operate feasibly and optimally under topology change. Hence, this paper proposes a data-driven combined NR and local control method with the SI to minimize network loss and maintain voltage security without excessive infrastructure. Firstly, the mixed-integer second-order cone programming (MISOCP) model considering NR and power flow (PF) are performed to obtain the dataset for SI and topology library. Then, the SI are trained with the physical constraint-based convolution neural network (PCNN) to support reactive power self-adaptively in topology variation. We test the proposed approach on IEEE 33 and 118 bus distribution systems in real-time control, and the results verify that our method further improves the economy and safety of the system with less infrastructure, and mitigate constraint violation on SI compared with other neural networkbased methods.