双路卷积神经网络和序列到序列的多步短期负荷预测  

Dual-branch CNN and Seq2Seq for Multi-step Short-term Load Forecasting

作　　者：袁建华[1,2] 徐杰 蒋文军 李洪强 YUAN Jianhua;XU Jie;JIANG Wenjun;LI Hongqiang(College of Electrical Engineering and New Energy,China Three Gorges University,Yichang 443002,China;Hubei Provincial Key Laboratory for Operation and Control of Cascaded Hydropower Station,Yichang 443002,China)

机构地区：[1]三峡大学电气与新能源学院,宜昌443002 [2]梯级水电站运行与控制湖北省重点实验室,宜昌443002

出　　处：《电力系统及其自动化学报》2024年第9期96-104,共9页Proceedings of the CSU-EPSA

基　　金：煤燃烧国家重点实验室开放基金资助项目(FSKLCCA1607)。

摘　　要：为提高多步短期电力负荷预测精度,满足实际应用中对于不同时间长度的预测需求,提出一种双路卷积神经网络和序列到序列相结合的多步预测方法。首先,构造双支路并行结构的特征提取网络双路卷积神经网络对负荷输入数据进行不同尺度的深层次特征提取;其次,分别以双向门控循环单元和长短期记忆网络作为编码器和解码器构建序列到序列网络,利用编码器对双路卷积神经网络的输出特征进行编码,并引入注意力机制完成输入数据到动态变化的中间向量的信息转换;最后通过解码器解码实现未来多个时刻的负荷序列输出。实验结果表明,所提方法较其他方法具有更优的多步预测性能。To improve the forecasting accuracy of multi-step short-term power load and meet the forecasting demand un-der different time steps in practical applications,a multi-step forecasting method combining dual-branch convolutional neural networks(DCNN)and sequence-to-sequence(Seq2Seq)is proposed.First,the feature extraction network DCNN with a dual-branch parallel structure is constructed to extract the deep feature of load input data on different scales.Sec-ond,Seq2Seq is constructed using a bidirectional gated recurrent unit and a long short-term memory network as the en-coder and decoder,respectively.The encoder encodes the output features of DCNN,and the attention mechanism is in-troduced to complete the information conversion from input data to dynamically changing intermediate vectors.Finally,the load sequence output at multiple moments in the future is realized by decoding using the decoder.Experimental re-sults show that the proposed method has a better multi-step prediction performance than other methods.

关 键 词：负荷预测 多步预测 序列到序列 深度学习 注意力机制 

分 类 号：TM715[电气工程—电力系统及自动化]