An Active Learning Local Control Method for Optimal Power Flow in Low Voltage Distribution Networks Considering Missing Data

The high penetration of renewable energy sources, especially solar photovoltaic, poses a significant challenge in distribution networks. Data-driven local control is an effective and budgeted way to ensure reliable distribution operation. However, this mode will face computationally expensive and ineffective problems with extensive historical data in the same operational period. In addition, the phenomenon of missing data will worsen due to the errors of measurement instruments. Therefore, an active learning local control method is proposed to select samples with diversity to improve the efficiency of the control scheme and maintain the performance designed by the original samples under the missing condition. Firstly, an optimal power flow model in a low-voltage distribution network is constructed considering the neutral line's impact. Then, the historical data containing missing values are processed by an imputation method, and an active learning method based on a greedy algorithm is introduced to select diverse samples, which speeds up the offline process of local control. Finally, we formulate the operation rules of the photovoltaic inverter and energy storage systems, which work as local devices in real-time control. The simulation results show that the proposed method realizes safe operation, saves the required time in the training stage, and achieves nearly approximate performance compared to the scheme designed by the original samples. Furthermore, this paper investigates the impact of different rates of missing data on local control and presents the proposed method to achieve the security and cost-effectiveness of the system under any missing condition.