According to the line loss calculation characteristics of distribution network, and the conditions of data missing and data abnormalities, the pretreatment model of line loss data pretreatment in distribution network based on group method of data handing (GMDH) algorithm is established, which can handle the abnormal data. The upper and lower limit of dependent variable and independent variable are determined based on k-nearest neighbors algorithm in the model, random interpolation and data packet processing model of all variables is established. The optimal GMDH model is searched and the missing value is calculated under iteration cycle. Experimental results show that the model has small calculations results error and fast computing speed, which can effectively and dynamically update missing line loss data, and optimize line loss calculation and analysis results.