| 引用本文: | 崔杨,朱晗,王议坚,张璐,李扬.基于CNN-SAEDN-Res的短期电力负荷预测方法[J].电力自动化设备,2024,44(4):164-170 |
| CUI Yang,ZHU Han,WANG Yijian,ZHANG Lu,LI Yang.Short-term power load forecasting method based on CNN-SAEDN-Res[J].Electric Power Automation Equipment,2024,44(4):164-170 |
|
| 摘要: |
| 基于深度学习的序列模型难以处理混有非时序因素的负荷数据，这导致预测精度不足。提出一种基于卷积神经网络(CNN)、自注意力编码解码网络(SAEDN)和残差优化(Res)的短期电力负荷预测方法。特征提取模块由二维卷积神经网络组成，用于挖掘数据间的局部相关性，获取高维特征。初始负荷预测模块由自注意力编码解码网络和前馈神经网络构成，利用自注意力机制对高维特征进行自注意力编码，获取数据间的全局相关性，从而模型能根据数据间的耦合关系保留混有非时序因素数据中的重要信息，通过解码模块进行自注意力解码，并利用前馈神经网络回归初始负荷。引入残差机制构建负荷优化模块，生成负荷残差，优化初始负荷。算例结果表明，所提方法在预测精度和预测稳定性方面具有优势。 |
| 关键词: 短期电力负荷预测 卷积神经网络 自注意力机制 残差机制 负荷优化 |
| DOI：10.16081/j.epae.202308018 |
| 分类号:TM73 |
| 基金项目:现代电力系统仿真控制与绿色电能新技术教育部重点实验室开放课题(MPSS2021-09)；吉林省自然科学基金资助项目(YDZJ202101ZYTS149) |
|
| Short-term power load forecasting method based on CNN-SAEDN-Res |
|
CUI Yang1, ZHU Han1, WANG Yijian1, ZHANG Lu2, LI Yang1
|
|
1.Key Laboratory of Modern Power System Simulation and Control & Renewable Energy Technology, Ministry of Education, Northeast Electric Power University, Jilin 132012, China;2.College of Information and Electrical Engineering, China Agricultural University, Beijing 100083, China
|
| Abstract: |
| The sequence model based on deep learning is difficult to deal with load data mixed with non-temporal factors, which leads to insufficient forecasting precision. A short-term power load forecasting method based on convolutional neural network(CNN),self-attention encoder-decoder network(SAEDN) and residual-refinement(Res) is proposed. The feature extraction module is composed of a two-dimensional convolutional neural network, which is used to mine the local correlation between data and obtain the high-dimensional features. The initial load forecasting module is composed of a self-attention encoder-decoder network and a feedforward neural network. The self-attention mechanism is used for self-attention encoding of high-dimensional features, the global correlation between data is obtain, thus the model is able to retain important infor-mation in the data mixed with non-temporal factors according to the coupling relationship between data. The self-attention decoding is performed by the decoding module, and the feedforward neural network is used for the initial load regression. The residual mechanism is introduced to build the load optimization module, the residual load is generated, and the initial load is optimized. The example results show that the proposed method has advantages in terms of forecasting accuracy and forecasting stability. |
| Key words: short-term power load forecasting convolutional neural network self-attention mechanism residual mechanism load optimization |
