A method based on improved K-Means++ and Elman neural network is proposed to calculate line loss for the problem of difficult computation and low accuracy of theoretical line loss in low-voltage(LV) station region. First, the Pearson correlation coefficient approach is utilized to identify and validate the main feature indicators of the LV station region. Second, the improved K-Means++ algorithm is used to cluster line loss data sets in the LV station region to increase the accuracy of line loss computation. In addition, the PSO algorithm is introduced to optimize the Elman neural network. Finally, simulations of real data and comparisons to existing calculation methods show that the proposed method has high training efficiency and calculation accuracy. The maximum percentage error of the suggested method is 1.15%, which is lower than the 6.07% of the BP neural network method.