Power Short-term Load Prediction Based on Fusion Model

Power short-term load consumption has the characteristics of nonlinearity and high randomness. The existing single machine learning model is difficult to make accurate prediction of short-term load consumption of electricity. A short-term power load prediction method based on fusion model is developed. Based on the idea of stacking fusion model, Extreme Gradient Boosting model, Support Vector Regression model and Long Short-term Memory network are fused to extract the deep features of the data as much as possible. XGBoost model can automatically learn the splitting direction for missing samples. SVR model has strong generalization ability. LSTM model has long-term memory function and has certain advantages in sequence modeling. The experimental results show that the fusion model inherits the advantages and characteristics of the three algorithms. Compared with the single model, the fusion model reduces MAPE, MAE and MSE which verifies the effectiveness of the proposed algorithm for power short-term load consumption.