Combined Forecasting Model Based on Time Series Characteristics of Power Load Curve

Different load forecasting models have different forecasting accuracy for different types of loads. In order to further improve the forecasting accuracy, this paper proposed two combined forecasting models, SVM-gray system and SVMArima, based on the time series characteristics of power load curves. By analyzing the principles of the two traditional forecasting models, the grey system and Arima, it’s found that the dispersion coefficient of the historical data of the load to be forecasted and the average slope difference of the daily load to be forecasted have a greater impact on the forecast errors of the grey system and the Arima model, respectively. The prediction accuracy of the grey system increases with the decrease of the discrete coefficient of the historical data of the load to be predicted, and the prediction accuracy of the Arima model increases with the decrease of the average slope difference of the load to be predicted. Therefore, the two prediction models are combined with the SVM model while the prediction accuracy is high, that is, the daily load with a small discrete coefficient of the historical data of the load to be predicted is selected as the experimental data, and the grey system-SVM combined prediction model is used for prediction. The daily load with small average slope difference of the load to be predicted is selected as the experimental data, and the ARIMA-SVM combined prediction model is used for prediction. The combined model uses the variance-covariance method (MV) to determine the weight coefficient, and the prediction results of the single prediction model are weighted and averaged to obtain the predicted value of the combined model. Finally, the load data of Anchorage, Alaska is taken as an example to verify. While the discrete coefficient of the historical data of the load to be predicted is low, the prediction accuracy of the grey system-SVM combined prediction model is higher. While the average slope difference of the load to be predicted is low, the prediction accuracy of the SVM-Arima combined prediction model is higher, which verifies the effectiveness of the proposed method.