基于数据挖掘的住宅建筑用户行为及相应能耗预测模型研究

Occupant behavior has a significant impact on residential building energy consumption. An accurate prediction of occupant behavior as well as its corresponding energy consumption can substantially improve the accuracy of building energy simulation. This will further assist architects and building designers in designing and realizing low/zero energy residential buildings. Also, it is crucial to the improvement of building energy performance and the decrease in environment influence. In this research, the Association Rule Mining technique of Data Mining will be employed in order to identify the associations between occupant behavior and its influencing factors, and further identify the input parameters of occupant behavior prediction models. The Clustering technique of Data Mining will be employed in order to classify households into different groups with the purpose of modeling occupant behavior separately. This can help to improve the accuracy of occupant behavior prediction models. Based on the above analysis, an occupant behavior prediction model for different households will be established by using the Decision Tree technique of Data Mining with confidence intervals given. A prediction model of energy consumption caused by occupant behavior will be established by using the Regression Tree technique of Data Mining. This research will provide suitable prediction models of occupant behavior as well as corresponding energy consumption for building simulation. Based on the established models, energy consumption analysis can be conducted for optimization design, efficient operation, renovation and new technologies utilization of residential buildings. Consequently, a set of guidelines on residential building energy conservation can be determined which can further help to formulate relevant policies.

用户行为对住宅建筑能耗有很大影响。准确预测用户行为及其相应能耗可显著提高住宅建筑能耗模拟精度，进一步实现低/零能耗住宅建筑，对提高建筑能效和减少环境影响具有重要意义。本课题提出采用数据挖掘方法中关联规则挖掘技术确定用户行为影响因素与用户行为的关联度，并辨识用户行为预测模型的输入参数；采用数据挖掘中聚类技术对住宅用户进行合理分类以分别建模，提高模型预测准确性。在此基础上，建立基于数据挖掘中决策树技术的以住户为单位的用户行为预测模型并计算置信区间；建立基于数据挖掘中分类树技术的用户行为相应能耗预测模型。本项目的研究可为建筑模拟分析提供预测用户行为及其对住宅建筑能耗影响的合适模型，以进一步进行住宅建筑优化设计、高效运行、节能改造及新技术采用的能耗分析，从而为住宅建筑节能提供理论和应用指导以及政策制定依据。

研究表明在建筑能耗模拟过程中，建筑人行为是造成模拟能耗与实际能耗往往显著不同的关键因素之一。对建筑人行为的深入研究和准确模型预测可有效提高建筑能耗模拟精度，同时有助于建筑性能评估和建筑运行优化控制。.研究团队成员首先对建筑人行为研究和数据挖掘应用进行了文献回顾，在此基础上明确了本项目核心研究内容，包括：模型输入参数辨识方法研究、住户分类方法与建筑能耗基准确定及行为节能评价方法研究、用户行为模型预测研究、用户行为相应能耗预测及节能潜力评估研究等，并建立了六个相应数据库。.针对上述研究内容，建立了基于关联规则挖掘技术的用户行为及影响参数关联度确定方法，可辨识出与用户行为关联度高的参数作为模型预测参数。建立了基于聚类技术的住户分类方法，可通过同时考虑不同重要参数影响进行合理分类，并进一步建立了建筑能耗基准确定及行为节能评价方法。基于上述研究，采用随机森林、马尔科夫链和基于主体建模方法建立了用户在室行为预测模型，并对马尔科夫链模型时间精度进行了研究，结果表明选取与状态转换概率矩阵变化步长相同的模拟步长能达到更高预测精确度。为进一步提高用户行为映射至建筑能耗的准确性，以湖南省住宅建筑待机行为和能耗为例，深入研究了不同住户之间设备及行为的差异性，并开发了不同节能模式下待机能耗节能潜力评估模型。此外，为深刻理解热适应性行为，对其关键影响因素热舒适性进行了研究，揭示了人体热舒适在停留时间和不同因素过激效应共同作用下的变化规律，并藉此确定了暂留区室内环境设计参数。.上述工作为建筑人行为研究提供了新的方法和工具，有助于在克服现有研究不足的基础上，建立满足精度要求的建筑人行为预测模型并准确映射至建筑能耗上，通过将其有效耦合应用于建筑能耗模拟软件和建筑运行控制系统中，可帮助相关人员在建筑设计、运行和改造等不同环节中提高建筑能效，从而实现建筑节能。