Research on CNN-BiLSTM Short-term Power Load Forecasting Model Based on Attention Mechanism and ResNet

WANG Lize1,2, XIE Dong1,2*, ZHOU Lifeng1,2, WANG Hanqing1,2

(1.School of Civil Engineering, University of South China, Hengyang, Hunan 421001, China;2.Hunan Engineering Laboratory of Building Environmental Control Technology, University of South China, Hengyang, Hunan 421001, China)

Abstract：Short term power load forecasting is beneficial to the efficient operation of power system and is of great significance to the effective dispatching of power market. Short term power load is affected by many factors, with large volatility and strong randomness, which makes its prediction accuracy low. It is difficult for BiLSTM and CNN to extract enough information from short-term load series. This paper proposes a CNN BiLSTM short-term load forecasting method combining Attention and ResNet. First, use the benchmark model BiLSTM and CNN to extract information from the input features, and use the Attention mechanism to highlight the extracted key information. Finally, ResNet creates a residual layer to fully learn the temporal features. Experiments on an open dataset show that the of this method reaches 2.80%, and the reaches 2.15. Compared with the prediction results of five commonly used models, the accuracy and effectiveness of the proposed model are verified.

key words：short-term load forecasting;convolutional neural network;bi-directional long short-term memory;attention mechanism;residual network

中图分类号：TM714

文献标志码：A

文章编号：1673-0062(2023)01-0033-07

收稿日期：2022-10-03

基金项目：国家自然科学基金资助项目(U1867221);湖南省教育厅科学研究项目(19C1568)

作者简介：王立则(1996—),女,硕士研究生,主要从事负荷预测等方面的研究。E-mail:[email protected]。*通信作者:谢 东(1978—),男,教授,博士,主要从事核通风智能控制与空气净化方面的研究。E-mail:[email protected]

DOI：10.19431/j.cnki.1673-0062.2023.01.005