【作者】 李凯；

【导师】 袁新枚；

【作者基本信息】 吉林大学 ， 动力工程及工程热物理， 2020， 硕士

【摘要】 根据国际能源署2019的数据显示,建筑业占全球最终能源消耗的三分之一以上,占直接和间接二氧化碳排放总量的近40%,在建筑的最终能源使用用途中,供热负荷占了很大的比例,对建筑短期负荷进行预测是用于建筑能源管理优化很有潜力的方式,目前已经应用的建筑能耗预测方式是基于物理原理的方式,但构建建筑模型需要耗费大量的人力及时间,限制了这种方法大规模推广。给使用机器学习来进行建筑供热负荷预测留下巨大空间。为了获取用于建立机器学习建筑供热负荷预测模型的数据,使用EP(EnergyPlus)建立建筑能耗模型。利用平滑算法和统计方法处理建筑负荷数据的异常数据。通过Pearson相关系数法来选择与建筑供热负荷相关的变量,选取了负荷相关变量及其过去24小时数据作为原始训练数据集,使用统计方法从原始数据集中提取机器学习建筑供热负荷预测模型的输入特征,并使用无监督机器学习算法来提取特征探讨机器学习在负荷相关特征提取的潜力。分别使用由统计方法和无监督机器学习算法提取的输入特征建立基于机器学习的短期建筑供热负荷预测模型,学习算法为基于结构风险最小化及经验风险最小化具有更好泛化能力的支持向量机。使用R2系数发评级预测模型精度。更多还原

【Abstract】 According to data from the International Energy Agency in 2019,the construction industry accounts for more than the replacement of global final energy consumption,accounting for nearly 40% of total direct and indirect carbon dioxide emissions.In the final energy use of buildings,the heating load accounts for a large proportion The short-term load forecasting of buildings is a promising method for building energy management optimization.The currently used building forecasting methods are based on physical principles,but the construction of building models requires a lot of manpower and time,which limits the This method is popularized on a large scale.There is a huge space left for using machine learning to predict building heating loads.In order to obtain data about building machine learning building heating load prediction models,EnergyPlus was used to build the building model.The Pearson correlation coefficient method was used to select the variables related to the building heating load.The load-related variables and their past 24 hours data were selected as the original training data set,and statistical methods were used.Extract the input features of machine learning building heating load prediction model from the original data set,and use unsupervised machine learning algorithms to extract features.Discuss the potential of machine learning to extract load-related features.The input features extracted by statistical methods and unsupervised machine learning algorithms are used to establish a short-term building heating load prediction model based on machine learning.The learning algorithm has better generalization ability for structure-based risk minimization and empirical risk minimization Support Vector Machines.Use the R2 coefficient to publish the grade prediction model accuracy.更多还原