Short-term power load forecasting based on SKDR hybrid model

Load forecasting is an important index to ensure the stable operation of power system. In recent years, load forecasting methods based on machine learning algorithms have received extensive attention. However, for such a complex problem, the traditional load forecasting method based on machine learning cannot solve the problem efficiently and accurately. Therefore, the ensemble learning method has gradually entered the field of view of researchers. Among them, stacking methods based on heterogeneous learners have received less attention. To that end, support vector machine (SVR), K-nearest neighbor (KNN) and decision tree (DT) are used as the base learners, and random forest (RF) is used as the meta-learner to construct a novel Stacking ensemble learning model (SKDR) in this study. Besides, due to hyperparameters are essential elements affecting the predicted result, the sparrow search optimization algorithm is introduced to obtain the optimal combination of hyperparameters. The effectiveness and advancement of SKDR is validated on a real-world dataset. Experimental results showed that compared with traditional methods, the proposed method could provide competitive prediction results, that is R2 = 0.984/0.987, RMSE = 1.315/1.253, MAPE = 0.146/0.163, this illustrates the SKDR's potential in terms of load forecasting. The performance of SKDR is also verified on the open-source dataset.