Load Forecasting Combining Phase Space Reconstruction and Stacking Ensemble Learning
Hui Hou,Chao Liu,Qing Wang,Bo Zhao,Leiqi Zhang,Xixiu Wu,Changjun Xie
DOI: https://doi.org/10.1109/tia.2022.3225516
IF: 4.079
2023-03-01
IEEE Transactions on Industry Applications
Abstract:Electric power load forecasting used to rely on some key information such as weather or socioeconomic development data. However, it is usually difficult to obtain accurate data on all the contributing factors. A rolling load prediction method based on phase space reconstruction of chaos theory and stacking ensemble learning is proposed. First, we adopted phase space reconstruction method to process the historical load series to avoid collecting some key data. The rolling forecasting method is used to predict a forecast length in advance, and the test data of learning model is obtained by shifting a forecast length of training data. And then we build a stacking ensemble learning model for load forecasting giving full play to the advantage of various algorithms such as Random Forest, Adaptive Boosting, Gradient Boosting Regression, Decision Tree, eXtreme Gradient Boosting and Long Short Term Memory, etc. Finally, the PJM power grid data set of the United States is used to verify the effectiveness of the proposed method, and the performance of the ensemble model is evaluated. Results show that comparing with single model forecasting, the proposed stacking ensemble learning method based on multi model fusion has higher forecasting accuracy.
engineering, electrical & electronic, multidisciplinary