基于Ａｔｔｅｎｔｉｏｎ.ＬＳＴＭ与多模型集成的短期负荷预测方法

作者：Zhu Jizhong; Miao Yuwang; Dong Zhaoyang; Dong Hanjiang; Chen Ziyu; Li Shenglin

来源：Electric Power Engineering Technology, 2023, 42(5): 138-147.

DOI：10.12158/j.2096-3203.2023.05.016

• attention mechanism
• long short-term memory network
• mean coding
• power system
• short-term load forecasting
• Stacking ensemble learning

Lately, deep learning techniques have developed rapidly. To address the problem of the poor processing capability of discrete data and the problem of poor generalization in short-term load forecasting tasks, a load forecasting method based on a long short-term memory network with attention mechanism (Attention-LSTM) and Stacking multi-model integration is proposed, which can take into account their respective advantages. Firstly, the discrete features are processed by mean coding, and then Attention-LSTM is used to extract the features of the load data. Next, the processed data are input into the multi-model integrated prediction model based on Stacking. The input features are analyzed and processed through three basic learners. Finally,the prediction is completed through the meta-learner. Actual load data from two datasets are used for analysis in the case study, and the load data in the two datasets are predicted separately, and compared with gated recurrent units, light gradient boosting machine and support vector machine. The simulation results show that the prediction accuracy based on the proposed method in the two datasets can exceed 98% in terms of accuracy, which is higher than the other three methods. ? 2023 Editorial Department of Electric Power Engineering Technology.

• 单位
  华南理工大学