【作者】 张强；

【导师】 丁研；

【作者基本信息】 天津大学 ， 供热、供燃气、通风及空调工程， 2018， 硕士

【摘要】 当前,建筑暖通空调系统(HVAC)有非常大的能耗,通过对冷、热负荷的准确预测,可帮助运行维护人员提前了解建筑用能需求,提高系统运行效率。本文以天津市某栋办公建筑实测数据为基础,提出一套负荷预测模型建立方法,且主要研究模型输入参数的优化过程,并在此基础上体现了负荷预测的节能效果。负荷预测模型包括短期负荷预测和超短期负荷预测。对于短期负荷预测模型,本文首先利用相关性分析初步筛选影响因子,筛选对象为室外各类变量的实时监测数据,考虑到建筑的热惰性,室外变量历史时刻值的影响也被考虑。进而利用主成分分析处理所得影响因子,获得模型输入参数。最终将所得输入参数代入人工神经网络和支持向量回归模型,建立负荷预测模型。针对超短期预测模型,会同时考虑室内、外各类变量及其历史时刻值的影响;在选择影响因子前,首先通过小波分解对负荷信号进行多频段分解,然后类似于短期负荷预测在多频段建模,最终利用小波重构获得预测值。利用实测数据对建模方法进行案例分析,结果表明短期和超短期热负荷预测模型的R~2值分别能够达到68.1%和94.0%,短期和超短期冷负荷预测模型的R~2值分别能够达到71.3%和83.9%。对比多种输入参数选择方法,本文所提出的基于小波分解和重构、相关性分析和主成分分析的输入参数优化方法能够有效提升预测精度。分析室内外变量对冷、热负荷预测模型的影响程度,结果表明热负荷主要受到室外变量的影响,仅用室外变量作为输入参数即可保证很高的预测精度;而对于冷负荷,室内变量的影响大于室外变量,但仅有室内变量作为输入参数时精度有限,室外变量的作用不可忽视。本文利用负荷预测模型提前获得建筑的用能需求,进而指导运行调节,有很大的节能效果。更多还原

【Abstract】 At present,the building HVAC system has a very large energy consumption.The accurate prediction of the cooling and heating load can help the building managers to know the energy demand in advance and improve the operation efficiency.Based on the measured data of an office building in Tianjin,the study presents a method for establishing building load prediction model,and mainly focus on the model inputs optimization.And then,the study reflects the energy saving effect of building load prediction.Two type of prediction models,which include short-term prediction and ultra-short-term prediction,are analyzed.For the short-term prediction model,this paper uses correlation analysis to select the influence factors preliminarily based on the real-time exterior variables firstly.Considering the thermal inertia of the building envelope,the influences of historic hours’exterior variables are also analyzed.Then,the principal component analysis is used to process the influence factors and the model inputs are obtained.Finally,the artificial neural network and support vector regression prediction models are established using model inputs,respectively.For the ultra-short-term prediction model,the interior variables will also be analyzed.Before selecting the influence factors,the load signals should be processed into multi-frequency-band signals by the wavelet decomposition,and then,the models are established on different frequency bands.The predicted load can be obtained through wavelet reconstruction.Using the measured data of case building to verify the proposed modeling methods,the prediction results show that the R~2 of short-term and ultra-short-term heating load prediction model can reach 68.1%and 94.0%respectively,the R~2 of short-term and ultra-short-term cooling load prediction model are 71.3%and 83.9%respectively.Comparing the various model inputs selection methods,the results show that the superiority of the model inputs optimization method proposed in this study,which is based on wavelet decomposition and reconstruction,correlation analysis,and principal component analysis.The importance of the interior and exterior variables that influencing the building load is compared.The conclusions show that the building heating load is mainly influenced by exterior variables,and only the exterior variables processed for model inputs can obtain a high prediction accuracy;for the cooling load prediction,the influence of interior variables is more important than that of exterior variables,however,only the interior variables processed for the model inputs cannot ensure a satisfactory prediction accuracy,the exterior variables should not be ignored when modelling.This study uses load prediction models to obtain the building energy demand ahead of time,and then guide the operation adjustment,which has great energy saving effect.更多还原