基于ConvLSTM-LSTM的短期负荷预测方法  

Short⁃term load forecasting method based on ConvLSTM-LSTM

作　　者：随春光 张玲华[1,2] SUI Chunguang;ZHANG Linghua(School of Communication and Information Engineering,Nanjing University of Posts and Telecommunications,Nanjing 210003,China;Jiangsu Engineering Research Center of Communication and Network Technology,Nanjing University of Posts and Telecommunications,Nanjing 210003,China)

机构地区：[1]南京邮电大学通信与信息工程学院,江苏南京210003 [2]南京邮电大学江苏省通信与网络技术工程研究中心,江苏南京210003

出　　处：《电子设计工程》2024年第10期54-58,共5页Electronic Design Engineering

基　　金：湖北省重点实验室开放基金(HBSEES202113)。

摘　　要：长短时记忆(LSTM)网络和结合卷积神经网络(CNN)的CNN-LSTM预测模型由于其网络模型本身的缺陷,限制了预测精度的提高。针对以上问题,提出了一种结合卷积长短时记忆(ConvL⁃STM)网络的ConvLSTM-LSTM负荷预测模型。利用ConvLSTM网络充分提取时序特征,将提取到的信息输入到LSTM网络中进行进一步的选择性记忆,并输出预测结果。将该模型与CNN-LSTM网络模型、LSTM网络模型、以及门控循环单元(GRU)网络模型进行了对比,结果显示所构建的Con⁃vLSTM-LSTM模型的预测效果均优于对比模型,在精度评价指标平均绝对百分比误差(MAPE)上,分别减小了1.10%、1.54%、1.91%。Long⁃short Time Memory(LSTM)network and Convolutional Neural Network(CNN)combined with CNN-LSTM prediction model limit the improvement of prediction accuracy due to the defects of their network models.To solve the above problems,a ConvLSTM-LSTM load forecasting model combined with Convolutional Long⁃Short Time Memory(ConvLSTM)network is proposed.Use the ConvLSTM network to fully extract the time series features,input the extracted information into the LSTM network for further selective memory,and output the prediction results.The comparison between the model and CNN-LSTM network model,LSTM network model,and GRU network model shows that the prediction effect of the ConvLSTM-LSTM model is better than the comparison model,and the Mean Absolute Percentage Error(MAPE)of the quantitative accuracy evaluation index is reduced by 1.10%,1.54%,and 1.91%respectively.

关 键 词：短期负荷预测 长短时记忆网络 卷积长短时记忆网络 组合预测模型 时序预测 

分 类 号：TN919.5[电子电信—通信与信息系统]