Open AccessMultifeature pool importance fusion based GBDT (MPIF-GBDT) for short-term electricity load prediction
Author(s) -
Shengwei Lv,
Gang Liu,
Bin Xue
Publication year - 2021
Publication title -
iop conference series. earth and environmental science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.179
H-Index - 26
eISSN - 1755-1307
pISSN - 1755-1315
DOI - 10.1088/1755-1315/702/1/012012
Subject(s) - feature selection , benchmark (surveying) , feature (linguistics) , computer science , selection (genetic algorithm) , artificial intelligence , term (time) , pattern recognition (psychology) , minimum redundancy feature selection , machine learning , data mining , linguistics , philosophy , physics , geodesy , quantum mechanics , geography
Feature selection is one of the key factors in predicting. Different feature selection algorithms have their unique preferences for elemental analysis of the data. This results in failing to determine the optimal features when a dataset goes through different feature selection algorithms to get different pools of input features, which in turn affects the prediction quality. To address this problem, the method integrates and fuses the feature importance values of two different feature selection methods. Then the input feature pools are optimized and filtered for the prediction model. Finally, the multifeature pool importance fusion based GBDT (MPIF-GBDT) is developed, which integrates the different feature selection methods and predicts the short-term power load in combination with the gradient boosting decision tree algorithm. In this paper, the tree model feature selection and the Recursive Feature Elimination (RFE) are chosen as feature selection methods. The experimental results show that MPIF-GBDT can significantly improve the accuracy of the prediction compared with the benchmark model.