With the wide participation of flexible load resources, the change of line losses in active distribution network was more complicated due to the influence of uncertainty and intermittence, which introduced great challenges to line losses prediction. In order to accurately predict the line losses of active distribution network, a line losses prediction method based on improved Stacking ensemble model was proposed. Firstly, considering the different operating characteristics affecting line losses after flexible load resources were connected to the distribution network, a characteristic index system of line losses was established. Then, the key indexes of line losses prediction for active distribution network were extracted by improved grey correlation analysis method. Finally, the Stacking ensemble prediction model for line losses of active distribution network was established. The improved grid search was used to determine the optimal hyperparameters combination of the model, and the precision weighted improved K-fold cross-validation method was used to accurately judge the prediction performance of the model. The superiority of the proposed method was verified by simulating on a line of 10 kV distribution network in Jiangsu Province.