基于LSTM的变频太阳能-空气源热泵系统逐时负荷预测研究  

Design of variable frequency light-heat-air source heat pump system based on particle swarm optimization

作　　者：胡洋 程志江[1] 崔澜 Hu Yang;Cheng Zhijiang;Cui Lan(School of Electrical Engineering,Xinjiang University,Urumqi 830047,China)

机构地区：[1]新疆大学电气工程学院,新疆乌鲁木齐830047

出　　处：《可再生能源》2022年第7期866-873,共8页Renewable Energy Resources

基　　金：国家自然科学基金(51567022);国家自然科学基金(51667020);面向南疆地区的多能互补供能系统关键技术与示范(2018AA007)。

摘　　要：文章以建筑中可再生能源系统为研究对象,利用长短期记忆(Long Short-Term Memory,LSTM)神经网络建立变频太阳能-空气源热泵(Variable Frequency Solar Air Source Heat Pump,VFAP)系统,在乌鲁木齐市气象条件下选取一个6层办公楼进行分析。首先,在实测数据校验模型的基础上,基于TRNSYS软件搭建VFAP系统,以一个供暖季为研究周期,获取VFAP系统的室外参数和过程运行数据。其次,利用灰色关联度分析(Grey Relation Analysis,GRA)计算各特征与供暖负荷的灰色关联度,并利用局部保留投影算法(Locality Preserving Projection,LPP)进行数据降维,得到VFAP系统的最优预测向量。最后,通过选择合理的网络参数,提出基于LSTM神经网络的VFAP系统的短期负荷预测模型,并与现有预测模型相对比。结果表明,LSTM神经网络对VFAP系统的负荷预测具有较好的识别效果,该精度要优于传统的神经网络预测模型,具有潜在的应用价值。Taking the building renewable energy system as the research object,a 6-story office building under weather conditions in Urumqi is selected for analysis,and the Long Short-Term Memory(LSTM)neural network is used to establish a variable frequency solar-air source heat pump(VFAP,Variable frequency).solar air source heat pump system.First,based on the verification model of measured data,a VFAP system is built based on TRNSYS software,and the outdoor parameters and process operation data of the VFAP system are obtained with a heating season as the research cycle.Second,use the Grey Relation Analysis(GRA)method to analyze the gray correlation between each feature and the heating load;and use the Locality Preserving Projection(LPP)to perform data dimensionality reduction to obtain the optimal VFAP system Forecast vector.Finally,by selecting reasonable network parameters,a short-term load forecasting model of the VFAP system based on the LSTM neural network is proposed,and compared with the existing forecasting model;the results show that the LSTM neural network has a better recognition effect on the load forecast of the VFAP system,The accuracy is better than the traditional neural network prediction model and has potential application value.

关 键 词：长短期记忆神经网络 空气源热泵 灰色关联度分析 局部保留投影 负荷预测 

分 类 号：TK511[动力工程及工程热物理—热能工程] TP183[自动化与计算机技术—控制理论与控制工程]