| 引用本文: | 余强,韩静娴,杨子梁,宋济东,杨德昌,齐海杰,于芃.基于改进双重压缩和激励与多头特征注意力机制的电-热负荷协同预测[J].电力自动化设备,2025,45(3): |
| YU Qiang,HAN Jingxian,YANG Ziliang,SONG Jidong,YANG Dechang,QI Haijie,YU Peng.Collaborative forecasting of electricity-thermal load based on improved dual squeeze and excitation and multi-head feature attention mechanism[J].Electric Power Automation Equipment,2025,45(3): |
|
| 摘要: |
| 综合能源系统中负荷多样且存在耦合，为提升负荷预测精度，提出一种基于改进双重注意力机制的分组卷积神经网络-门控循环单元短期电-热负荷协同预测模型。通过改进的压缩和激励注意力为各输入通道加权，再对其进行分组卷积；利用多头特征注意力对卷积结果进行赋权，并利用输入门控循环单元模型对负荷进行预测。算例仿真结果表明，所提模型的平均绝对百分比误差均低于3 %。 |
| 关键词: 综合能源系统 负荷预测 分组卷积神经网络 门控循环单元 改进的压缩和激励注意力机制 多头特征注意力机制 |
| DOI：10.16081/j.epae.202409030 |
| 分类号:TK01+9 |
| 基金项目:国家电网有限公司科技项目(SGSDDK00PDJS2250114)；国家自然科学基金资助项目(52377127) |
|
| Collaborative forecasting of electricity-thermal load based on improved dual squeeze and excitation and multi-head feature attention mechanism |
|
YU Qiang1, HAN Jingxian1, YANG Ziliang1, SONG Jidong1, YANG Dechang1, QI Haijie2, YU Peng3
|
|
1.College of Information and Electrical Engineering, China Agricultural University, Beijing 100083, China;2.State Grid Smart Grid Research Institute Co.,Ltd.,Beijing 102200, China;3.State Grid Shandong Electric Power Research Institute, Jinan 250003, China
|
| Abstract: |
| Loads are diverse and coupled in the integrated energy system, in order to improve the accuracy of load forecasting, a short-term electricity-thermal load collaborative forecasting model based on an improved dual attention mechanism and group convolutional neural network-gated recurrent unit is proposed. Each input channel is weighted by the improved squeeze and excitation attention mechanism, and grouped for the convolution. The convolution results are weighted by the multi-head feature attention mechanism and the load is forecasted by the input gated recurrent unit model. The example simulative results show that the mean absolute percentage errors of the proposed model are less than 3 %. |
| Key words: integrated energy system load forecasting group convolutional neural network gated recurrent unit improved squeeze and excitation attention mechanism multi-head feature attention mechanism |
