基于神经网络的地铁车站冷负荷预测对比研究  

Research on Cooling Load Prediction of Subway Stations Based on Backpropagation Neural Network and Convolutional Neural Network

作　　者：刘舸争 李舒宏[2] 胡远洋 李新美 李媛 于德涌 王珩 刘守超 陈诚 Liu Gezheng;Li Shuhong;Hu Yuanyang;Li Xinmei;Li Yuan;Yu Deyong;Wang Heng;Liu Shouchao;Chen Cheng(Shenzhen Metro Construction Group Co.,Shenzhen,518000;School of Energy and Environment,Southeast University,Nanjing,210096;Nanjing Fuca Automation Technology Co.,Nanjing,210046)

机构地区：[1]深圳地铁建设集团有限公司,深圳518000 [2]东南大学能源与环境学院,南京210096 [3]南京福加自动化科技有限公司,南京210046

出　　处：《制冷与空调（四川）》2024年第3期299-304,350,共7页Refrigeration and Air Conditioning

摘　　要：冷负荷预测是建筑节能的基础,然而输入的不同会影响神经网络的预测精度,且对新建筑进行负荷预测时需要数据的积累。为得到地铁站冷负荷预测的最佳输入并评估数据库迁移预测的可行性,基于实测数据,以常用的时间变量、气象变量和历史负荷作为待选输入,比较了反向传播神经网络和卷积神经网络在不同输入下及数据库逐步更新时的预测精度。结果表明:最佳输入变量与冷负荷的皮尔逊相关系数需大于0.5;另一方面,同类型建筑在预测初期可以基于数据库逐步替换实现预测,预测精度随着数据库的更新逐渐提升,且卷积神经网络表现出更好的预测表现。Cooling load prediction forms the foundation of building energy conservation.However,the variation in input variables significantly influences the predictive accuracy of neural networks.Furthermore,predicting loads for new buildings necessitates the accumulation of data.This study aims to determine the optimal inputs for predicting subway station cooling loads and evaluate the feasibility of database migration predictions.Leveraging empirical data and considering commonly used time variable,meteorological factors,and historical loads as potential inputs,this research compares the predictive accuracy of Backpropagation Neural Network(BPNN)and Convolutional Neural Network(CNN)under different inputs and during progressive database updates.Results indicate that the optimal input variables for cooling load prediction should exhibit Pearson correlation coefficient with the load greater than 0.5.Moreover,for similar building types,initial predictions can be made by gradually replacing the database,showcasing an enhancement in predictive accuracy with each update.Notably,the CNN demonstrate superior prediction performance throughout this process.

关 键 词：负荷预测 地铁站 神经网络 输入组合 数据库更新 

分 类 号：TU831.2[建筑科学—供热、供燃气、通风及空调工程]