Abstract: Many advanced applications in the power grid are based on load data. However, in the process of data collection, transmission and storage, there will inevitably meet with data missing, which may bring bad effect to data analysis. In allusion to the above problem, this paper proposes a missing data completion strategy combining load curve similarity and singular value threshold algorithm. Firstly, through analyzing the distribution of singular values in electric data matrix, a low rank characteristic is revealed. On the basis of that, considering the differences between power consumption curves, a data recovery method of pre-filling, curve similarity classification and secondary filling is proposed. After the pre-filling, for each power curve to be recovered, the k curves which are most similar to it are found on the basis of the improved similarity index to form the data matrix, and then the low-rank matrix completion is applied again. Through the above steps, the missing recovery is supposed to be improved. The load data from the power grid in north China is used for test, and the method proposed is compared with the classical interpolation method to verify the effectiveness. Results prove that the combination of load curve similarity and singular value thresholding (SVT) can effectively complete the missing data and provide strong support for load data repair.

Key words: load data, missing data, curve similarity, low-rank matrix completion