基于时序数据逆向解析的建筑耗热量热流组分定向提取方法

Building cooling and heating load consists of several heat components. Quantitative analysis of the components is essential to the energy saving design on building energy efficiency and the optimal operation of HAVC systems. Due to the difficulties of direct measurements on the actual heat components, thermal performance evaluation on the building elements is mainly based on theoretical analysis and simulation methods, which fail to show the actual performances by practical operation data. The building heat loss data which are continually monitored with certain sampling frequency can be considered as a a time series signal, that is also the same with the heat components. As each heat component generated by different disturbance signal and different thermal process, the fluctuation characteristics of each time series signals also vary with each other, like the heat flux component though building envelope tends to low-frequency and sine waves while the component generated by solar radiation appears as high-frequency and impulse waves. The variation among fluctuation characteristics provides possibilities for the disaggregation of time series data and the directional extraction of heat flux components with blind source separation theory. In this application, basic morphological structures of different disturbance signals are studied to expand the thermal process response mechanism adaptively and get acquaintance of the composition of morphological elements in each heat flux component signal, then basis functions or their combination are determined to sparsely represent the various morphological components, which can further differentiate each heat flux component. Accordingly, disaggregation algorithm of building heat loss can be proposed to analyze the heat flux composition, and corresponding experiments are designed to validate and modify the theoretical method.

建筑冷热负荷由多个热量组分叠加而成，解析负荷组分和构成是建筑节能设计和运行优化的基础。由于实际热量组分不易直接测量，对建筑部件节能效果的评判，大多借助理论仿真手段，难以使用运行数据反映实际表现。连续监测的建筑负荷可视为一组热流时序信号，其各热量组分亦是如此。由于扰动和热过程路径各不相同，各热流信号的波动形态也差异明显。墙体热流类正弦波动，日射得热呈脉冲波动，而室内得热高频随机性更强。信号处理领域的“盲源分离”思想，是利用源信号形态差异，从可观测的混迭信号中逆向解析出未知源信号。负荷组分的解析需求契合盲源分离应用范畴。申请书拟从负荷扰动信号的基本形态结构入手，分析建筑热过程对其影响，进而掌握各热流组分信号的形态结构成分。通过差异特征提取，选用适当基函数或组合实现热流组分的各自稀疏化表示，进一步差异化热流信号形态，以此实现耗热量监测数据的逆向解析，提取热流组分，并设计实验，验证和修正理论方法。

建筑能耗由多个热量组分叠加而成，获取建筑能耗的分项组分，对于掌握建筑各部件在实际中的作用和表现，优化建筑设计和运行有着重要的现实意义。.本课题首先对建筑耗热量分项热流组分的基本形态结构进行分析，基于稀疏编码的思想采用字典学习算法自适应地获取各组分的形态结构元素-基函数。进一步的，分析各组分在总耗热量中的分布特点。结合组分的形态特征和分布特点，提出一种运用稀疏自编码器和混沌优化算法的建筑耗热量时序信号逆向解析方法。最后，分别在模拟数据和实验数据中对方法进行了可行性验证。主要研究成果如下：.1. 根据建筑耗热量成分的形态特征，将建筑夏季冷负荷拆分成四项组分，包括温差组分、日射组分、新风组分和内扰组分；冬季热负荷拆分成三项组分，包括温差组分、日射组分、内扰组分。.2. 对比了非负K-SVD字典学习算法及稀疏自编码器对建筑耗热量分项组分形态特征提取的效果，实现不同组分形态的描述和量化。.3. 分析了不同类型建筑中各热流组分在总耗热量中的分布特点。.4. 提出了一种结合稀疏自编码器和混沌优化算法的建筑耗热量时序数据逆向解析方法。基于模拟的典型办公建筑案例对方法进行可行性分析：夏季工况下，各组分的平均解析精度CV-RMSE为41%；冬季工况下，各组分的平均解析精度CV-RMSE为35%。.5. 分析了各分项组分贡献比重和形态结构的差异程度对建筑耗热量逆向解析精度的影响。.6. 分析了分项热流组分字典学习样本生成过程中参数的设置对所提出的建筑耗热量时序数据逆向解析方法的影响，并给出影响较大的参数类别。.7. 构建建筑实验模型，在实验条件下对所提出的建筑耗热量时序数据逆向解析算法进行了可行性验证。