Abstract:  In view of problems that the electrical characteristics of power consumers have not been fully shown in short-term load forecasting and the accuracy of load forecasting is not enough, this paper proposes a new ‘layered-confluence’ model. Firstly, we layer the power consumers based on electrical characteristics, obtain the load characteristic curve of each layer with different electrical characteristics of power consumers, and take the load characteristic curve of layer as attribute factor to construct total load curve. Then, we train the model according to  real time load data and load characteristic curve of each layer confluence on different days. Finally, we implement the regression prediction. Taking the actual power load data of a region as an example, we forecast load based on the proposed prediction method. The results show that, the short-term power load forecasting method based on ‘layered-confluence’ model is superior to general regression forecasting method in 3 evaluation indices of mean absolute percentage error (MAPE), root-mean-square error (RMSE) and Pearsons correlation coefficient, which verifies the validity of the model. The ‘layered-confluence’ model has a good forecast effect through using different algorithm combination in ‘layered’ period and ‘confluence’ period, which verifies the robustness of the model. ‘Layered-confluence’ mode can improve the precision of load forecasting, which can provide a new idea for the short-term power load forecasting.

Key words:  short-term power load forecasting, electrical characteristic of power consumers, load characteristic curve of layer , ‘layered-confluence&rsquo, model