基于建筑运行数据分布式建模的空调负荷预测方法研究

Cooling load prediction is the basis of the optimal control of air conditioning systems which is of great significance for energy conservation in public buildings. The thermal inertia of the internal thermal mass in buildings and the dynamic changes in equipment operating status and the number of people inside will have an important impact on the air conditioning load. The cooling load simulation by software, such as EnergyPlus and TRNSYS mostly adopts the equipment and people schedule recommended by the standards, which is quite different from the actual situation. While data-driven models often consider the total cooling load as a holistic research object, ignoring the impact of internal detail factors. Using the measured data to build models of each influencing factor is expected to improve the accuracy of cooling load forecasting results. Based on heat transfer theory and data mining method, the distributed modeling method is first proposed and the dynamic influence characteristics of key internal disturbances on cooling load will be studied in this project. The series-parallel hybrid RC model is constructed to characterize the heat storage and release characteristics of the internal thermal mass. Based on the sub-metering data, the windowed Fourier model is constructed to predict the real-time equipment load. Based on the measured personnel data, the support vector machine algorithm is used to construct the real-time people load forecasting. Finally, a set of cooling load forecasting method based on distributed modeling is formed, and the forecasting model is verified and evaluated by the measured data in real buildings. The research will reveal the variation characteristics of internal disturbances in actual buildings and their impact on cooling load, improve the accuracy of cooling load prediction and provides reliable technical support for the optimal control of air conditioning systems.

空调负荷预测是空调系统优化控制的基础，对公共建筑运行节能有重要意义。建筑内部蓄热体的蓄放热、设备和人员数量的动态变化都会对空调负荷产生重要影响。空调负荷模拟法多采用规范推荐的设备和人员时间表，这与建筑实际情况差异较大；数据驱动预测模型多将空调总负荷作为整体进行研究。根据实测数据对各影响因素分别建模有望提高负荷预测精度。本项目基于传热理论分析和数据挖掘建模，研究关键内扰对空调负荷的动态影响特性，采用分布式建模的方法，构建串并联混合RC模型，表征建筑内部蓄热体的蓄放热特性；在分项计量数据基础上构建加窗傅里叶模型，预测实时设备负荷；基于实测人员数据，选用支持向量机算法，构建实时人员负荷预测模型。最终形成一套基于分布式建模的空调负荷预测方法，并利用实测数据对预测模型进行验证和评价。该研究将揭示实际建筑内扰的变化规律及其对空调负荷的影响水平，提高空调负荷预测精度，为空调系统优化控制提供可靠技术保障。

建筑行业是节能减排的关键领域，负荷预测在建筑暖通空调系统节能中扮演着重要角色。负荷模拟软件有较好的精度，但需构建建筑几何外型，建筑几何模型一旦确定后往往无法更改，而且对复杂建筑模拟时耗时较长。数据挖掘方法也有其局限性，如受历史数据质量的限制。为解决上述问题，本研究提出分布式建模的构想，综合考虑气象条件、围护结构、照明、设备和人员等影响因素，选用轻量级梯度提升机（LightGBM）算法构建负荷预测模型并完成验证。为了充分反映内扰对负荷预测精度的影响，构建了空调系统末端设备能耗拆方法，解决末端设备能耗转化为建筑空调负荷的问题。为了实现对未来负荷的预测，解决了逐时太阳辐射预测问题。.主要成果如下：.（1）确定了负荷影响因素种类和取值范围，采用EnergyPlus软件构建了包含35种建筑几何外型，2870个建筑算例，2514万余条数据的模拟负荷数据库。.（2）基于模拟数据库和LightGBM算法，应用python编译，提取特征变量，完成重要度分析，构建负荷预测模型。冷负荷逐时误差为6.95%，偏差为4.31W/m2；热负荷逐时误差为7.09%，偏差为11.64W/m2。.（3）提出基于地外太阳辐射和有限气象数据的地表逐时太阳辐射简化预测方法，模型实测平均绝对误差为70.94W/㎡，均方根误差为108.39W/㎡。.（4）基于分类回归树算法建立了空调末端设备电耗拆分方法，在两栋建筑中进行验证性，工作日逐时误差小于5%。.（5）负荷预测模型在上海和北京两栋建筑中应用，冷负荷预测值与实测值对比，逐时误差为12.42%；与EnergyPlus模拟值对比，逐时误差值为8.79%；热负荷预测值与实测值对比，逐日误差为7.97%。.本课题构建模型的训练数据库庞大，气象参数、运行时间表可灵活调用，且给出了特征变量的重要度占比和适用范围。本方法一定程度上摆脱了传统模拟软件物理建模和建筑历史数据的限制，在建筑设计阶段和运行阶段都可适用，使用简便、计算速度快、精度良好。本研究成果可以推广到其他类型建筑，为其他建筑负荷预测研究提供参考。