融合CNN与BiLSTM模型的短期电能负荷预测  

Short-term Power Load Forecasting Based on CNN and BiLSTM Model

作　　者：杨桂松[1] 高炳涛 何杏宇 YANG Guisong;GAO Bingtao;HE Xingyu(School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China;College of Communication and Art Design,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200093 [2]上海理工大学出版印刷与艺术设计学院,上海200093

出　　处：《小型微型计算机系统》2024年第9期2253-2260,共8页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61602305,61802257)资助;上海市自然科学基金项目(18ZR1426000,19ZR1477600)资助;南通市科技局社会民生计划项目(MS12021060)资助.

摘　　要：针对卷积神经网络(CNN)在捕捉预测序列间历史相关性方面的不足以及在变量复杂情况下出现的无法精准提取预测关键信息的问题,提出一种将双向长短期记忆网络(BiLSTM)与卷积神经网络结合的CNN-BiLSTM模型.首先,采用数据预处理方法保证数据的正确性和完整性,并对数据进行分析以探究多变量之间的相关性;其次,通过CNN与L1正则化对多维输入特征进行特征筛选,选取与预测相关的重要性特征向量;最后,使用BiLSTM对CNN输出的关键特征信息进行保存,形成向量与预测序列,并通过分析时序特征的潜在特点,提取用户的内在消费模式.实验比较了该模型与其他时序模型在不同时间分辨率下的预测效果,实验结果表明,CNN-BiLSTM模型在不同的回望时间间隔下表现出了最佳的预测性能,能够实现更好的短期负荷预测.In view of the defect that CNN(Convolutional Neural Network)has in capturing the historical correlation between forecasting sequences and extracting key prediction information accurately when variables are complex,this paper proposed a hybrid prediction model called CNN-BiLSTM that combined BiLSTM(Bidirectional Long Short-Term Memory Network)and CNN.Firstly,the data preprocessing method is used to ensure the correctness and integrity of the data,and the data is analyzed to explore the correlation between multiple variables;Secondly,the multi-dimensional input features are filtered by CNN and L1 regularization,and the importance feature vectors related to prediction are selected;Finally,BiLSTM is used to save the key feature information output by CNN to form a vector and prediction sequence,and the user′s internal consumption pattern is extracted by analyzing the potential characteristics of time series features.The prediction results of this model and other time series models at different time resolutions are compared.The experimental results showed that the CNN-BiLSTM model has the best prediction performance under different look back time intervals,which achieved better short-term load forecasting.

关 键 词：卷积神经网络 双向长短期记忆网络 特征筛选 CNN-BiLSTM模型 短期负荷预测 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]