A Multi-Mode Data-Driven Volt/Var Control Strategy With Conservation Voltage Reduction in Active Distribution Networks

This paper proposes a two-stage multi-mode data-driven Volt/Var optimization-based conservation voltage reduction (VVO/CVR) strategy to reduce total energy consumption while mitigating fast voltage violations in distribution networks. In the first stage, optimal power flow (OPF) is performed in the central controller (CC) to obtain day-ahead dispatch results of on-load tap changers (OLTC) and capacitor banks (CBs) based forecasting data of PV and load. In the second stage, a multi-mode data-driven local control strategy is designed in real-time Volt/Var control (VVC) with three operation modes, which are energy saving mode, reactive power reservation mode and security operation mode respectively. The coordination and transition among different control modes allow system operators to achieve both energy savings and voltage security under various conditions. The flexible reactive power margin of PVs can also be adaptively improved through cooperation with CBs. Moreover, a data-driven deep convolution neural network (CNN) is developed and trained as a local controller (LC) to regulate the reactive power of PVs based on local measurements. The influence of input selections on training losses is analyzed in detail, and a clustering-based data preprocessing method is also proposed to simplify the training process without compromising the training performance. The proposed approach is tested on the IEEE 33-bus distribution system and simulation results verify the effectiveness both in saving energy and addressing voltage problems.