蓄能空调需求响应时段负荷和储释能时长预测  

Prediction of Load of Air-conditioning System and Energy Storage/Release Time of Tank During Demand Response Period

作　　者：任效效 孟庆龙 李洋 熊成燕 奚源 REN Xiao-xiao;MENG Qing-long;LI Yang;XIONG Cheng-yan;XI Yuan(School of Civil Engineering,Chang'an University,Xi'an 710061,China;School of Energy and Power Engineering,Xi'an Jiaotong University,Xi'an 710049,China)

机构地区：[1]长安大学建筑工程学院,西安710061 [2]西安交通大学能源与动力工程学院,西安710049

出　　处：《建筑节能（中英文）》2021年第9期95-104,共10页Building Energy Efficiency

基　　金：陕西省重点研发计划项目(2020NY-204);山东省可再生能源建筑应用技术重点实验室开放课题(JDZDS02)。

摘　　要：空调参与需求响应能够提高电力系统的稳定性。考虑主动储能的需求响应是通过准确预测蓄能罐的储能、释能时长以及系统的运行负荷来保障策略的合理性和高效性。为此,搭建TRNSYS仿真实验平台获取数据,采用相关性分析和主成分分析,选择输入变量并对其降维,比较5种机器学习算法(BP神经网络、RBF神经网络、广义回归神经网络、Elman神经网络和支持向量回归)对空调系统未来1 h和24 h的静态负荷预测。选择Elman神经网络预测蓄能罐的储、释能时长并利用改进的粒子群优化算法进一步优化,对未来1 h和24 h负荷进行滚动预测。结果表明:相关性分析+主成分分析能提高模型的预测精度,Elman神经网络预测精度最高,利用改进的粒子群算法优化后,该模型对未来1 h和24 h负荷预测的拟合优度R^(2)值分别从0.790和0.972提高到0.845和0.975;利用Elman神经网络预测储、释能时长R^(2)值分别为0.993和0.984。Air-conditioning participating in demand response(DR)can improve the stability of power system.Considering the demand response of active energy storage is to ensure the rationality and efficiency of the strategy by accurately predicting the energy storage of the storage tank,the duration of energy release and the operating load of the system.Therefore,a TRNSYS simulation experiment platform is built to acquire data,select input data with correlation analysis and principal component analysis.Meanwhile,the static load forecasting of air-conditioning system in the next 1 h and 24 h is compared with five machine learning algorithms,such as BP neural network,RBF neural network,generalized regression neural network,Elman neural network and support vector regression.The best one is selected to predict the storage and release time of the energy storage tank,and improved by particle swarm optimization(PSO)algorithm that is used to rolling forecast the load in the next 1 h and 24 h.The results show that:correlation analysis+principal component analysis can improve the prediction accuracy of the model;Elman neural network has the highest accuracy.After optimized by the improved PSO,the goodness of fit(R^(2))of the model for the future 1 h and 24 h load forecasting is improved from 0.790 and 0.972 to 0.845 and 0.975 respectively,and the R^(2) values of predicted energy storage and release time of the tank by Elman neural network are 0.993 and 0.997 respectively.

关 键 词：空调 需求响应 TRNSYS 神经网络 主动储能 支持向量回归 

分 类 号：TU831.3[建筑科学—供热、供燃气、通风及空调工程]