With the rapid advancement of smart grid construction and the increasing maturity of big data and artificial intelligence technologies in recent years, the exponential growth of grid operation measurement data has provided effective support for accurate prediction of distribution grid line loss. However, the traditional prediction method based on physical mechanisms cannot effectively analyze the non-linear hidden connection between the multiple sources of the data and the transmission line loss of the distribution power grid. Thus, this article proposes a multi-time granularity prediction way for the transmission line loss of the distribution power grid based on the maximal information coefficient (MIC) and deep learning, which quantitatively describes the correlation between input features and output features. Compared with linear correlation analysis, the feature selection method based on MIC is more comprehensive and can explore the nonlinear correlation between input features and output features. And based on the selected input feature set, combined with the deep LSTM neural network, a multi-time granularity prediction pattern of the transmission line loss is constructed, and the prediction accuracy is further improved. Finally, the proposed prediction method is validated by combining the real line loss data.

the maximal information coefficient; deep LSTM neural network; line loss prediction