Missing Data Reconstruction Method of Distribution Network based on RES-AT-UNET

When the power monitoring equipment in distribution network is affected by extreme events such as typhoon, thunderstorm and strong electromagnetic pulse, the measurement data is missing, the evaluation of equipment observability and availability cannot be realized effectively. The traditional data reconstruction method adopts linear interpolation method, which ignores the change rule of power system measurement data and context constraints, the reconstruction accuracy is very low. In this paper, a data missing value reconstruction method based on Residual Attention UNET(RES-AT-UNET) network is proposed. Considering the characteristics of distribution network and avoiding complex explicit modeling, the proposed method adopts the end-to-end model training method, which can still maintain the accuracy of data reconstruction in the case of missing large interval time series data. The experimental results show that the root mean square error of the data reconstructed by the proposed method is the smallest compared with the actual data, and the reconstruction model has strong applicability to the data under different missing rates.