| 引用本文: | 吉兴全,曾若梅,张玉敏,等.基于注意力机制的CNN-LSTM短期电价预测[J].电力系统保护与控制,2022,50(17):125-132.[点击复制] |
| JI Xingquan,ZENG Ruomei,ZHANG Yumin,et al.CNN-LSTM short-term electricity price prediction based on an attention mechanism[J].Power System Protection and Control,2022,50(17):125-132[点击复制] |
|
| 摘要: |
| 短期电价预测结果的准确性对存在多元化竞争格局的电力市场具有重要意义。为提高在电价跳跃点和尖峰点的预测精度及预测效率，针对多因素融合影响的电价序列与其影响因素间隐含的非线性关系，提出了一种基于ATT-CNN-LSTM的短期电价预测方法。首先，采用灰色关联度分析法分析负荷因素与电价之间的关联程度，筛选出关联度较高的数据作为最优模型输入。其次，通过注意力机制(Attention, ATT)自适应分配输入数据的权重，以权重大小区分强弱特征数据。再利用卷积神经网络(Convolution Neural Networks, CNN)对数据集进行二次特征提取及降维处理，优化输入长短期记忆神经网络(Long Short-Term Memory, LSTM)中的数据，从而提升LSTM网络的预测精度与训练速度。对澳大利亚电力市场的实测数据进行算例分析，通过与其他主流算法对比，验证了所提方法具有更高的预测精度和计算效率。 |
| 关键词: 注意力机制 卷积神经网络 长短期记忆神经网络 电价预测 灰色关联度分析 |
| DOI：DOI: 10.19783/j.cnki.pspc.211472 |
| 投稿时间：2021-11-01修订日期：2022-04-18 |
| 基金项目:国家自然科学基金青年项目资助(52107111)；山东省自然科学基金青年项目资助(ZR2021QE117)；青岛西海岸新区2020年科技项目资助(源头创新专项)(2020-92) |
|
| CNN-LSTM short-term electricity price prediction based on an attention mechanism |
| JI Xingquan,ZENG Ruomei,ZHANG Yumin,SONG Feng,SUN Pengkai,ZHAO Guohang |
| (1. College of Electrical Engineering and Automation, Shandong University of Science and Technology, Qingdao 266590, China;
2. Yantai Power Supply Company, State Grid Shandong Electric Power Company, Yantai 264000, China) |
| Abstract: |
| The accuracy of short-term electricity price forecasts is of great significance to the electricity market with a diversified competitive landscape. To improve prediction accuracy and efficiency at the jump and peak points of electricity price, a short-term electricity price prediction method based on ATT-CNN-LSTM is proposed for the implied nonlinear relationship between the electricity price series influenced by the fusion of multiple direct and influencing factors. First, the grey correlation degree analysis method is used to analyze the correlation degree between load factors and electricity prices, and the data with a higher correlation degree is selected as the optimal model input. Secondly, the weight of the input data is adaptively allocated through the attention mechanism (ATT), and the strong and weak feature data are distinguished by the weights. Then, a convolution neural network (CNN) is used to perform secondary feature extraction and dimensionality reduction of the data set to optimize the data input into the long short-term memory (LSTM) network, thereby improving the prediction accuracy and training speed of the LSTM network. The actual measurement data of the Australian electricity market is used for a case study, and the comparison with other mainstream algorithms verifies that the proposed method has higher prediction accuracy and computational efficiency. |
| Key words: attention mechanism convolutional neural network long-short term memory neural network electricity price forecast grey relation analysis |
