According to the statistical distribution characteristics of the actual electric energy data, and considering the treatment effect mean substitution method of energy loss data is usually not satisfactory，The estimation error of LOWESS model is limited by its given window width and fitting order, It is proposed a LOWESS regression model of the electric energy data deletion optimization based on prediction error minimization parameters automatic processing method. By comparing the fixed window and order number in the non stationarity of the electric energy data on the prediction effect and study parameters optimization LOWESS model accuracy, adaptability and comparative advantage. Through the actual data validation, the model can adapt to different data distribution of electric energy data, different loss ratio and so on, the prediction accuracy is high, it has certain practical reference value.
