[论文解读] Data Driven Energy Efficiency in Buildings
本文提出了一种数据驱动的框架,通过五个核心支柱提升建筑能效:最优仪表化、子系统互联、推断决策、人员参与以及智能运行。该研究将非侵入式负载监测(NILM)定位为贯穿所有五个方面的核心应用,展示了如何利用高分辨率智能电表和传感器数据实现实时能耗分解、故障检测与需求优化。
Buildings across the world contribute significantly to the overall energy consumption and are thus stakeholders in grid operations. Towards the development of a smart grid, utilities and governments across the world are encouraging smart meter deployments. High resolution (often at every 15 minutes) data from these smart meters can be used to understand and optimize energy consumptions in buildings. In addition to smart meters, buildings are also increasingly managed with Building Management Systems (BMS) which control different sub-systems such as lighting and heating, ventilation, and air conditioning (HVAC). With the advent of these smart meters, increased usage of BMS and easy availability and widespread installation of ambient sensors, there is a deluge of building energy data. This data has been leveraged for a variety of applications such as demand response, appliance fault detection and optimizing HVAC schedules. Beyond the traditional use of such data sets, they can be put to effective use towards making buildings smarter and hence driving every possible bit of energy efficiency. Effective use of this data entails several critical areas from sensing to decision making and participatory involvement of occupants. Picking from wide literature in building energy efficiency, we identify five crust areas (also referred to as 5 Is) for realizing data driven energy efficiency in buildings : i) instrument optimally; ii) interconnect sub-systems; iii) inferred decision making; iv) involve occupants and v) intelligent operations. We classify prior work as per these 5 Is and dis-cuss challenges, opportunities and applications across them. Building upon these 5 Is we discuss a well studied problem in building energy efficiency -non-intrusive load monitoring (NILM) and how research in this area spans across the 5 Is.
研究动机与目标
- 通过利用智能电表、BMS和环境传感器的高分辨率数据,应对建筑能耗持续增长的问题。
- 识别从建筑数据实现能效优化的关键挑战,包括数据稀疏性、系统孤立性以及缺乏人员参与。
- 提出一个统一框架——'五大支柱(5 Is)'——以指导建筑中数据驱动能效研究与部署。
- 通过非侵入式负载监测(NILM)作为代表性案例研究,展示该框架的适用性。
- 强调需要采用无监督、计算高效且考虑人员因素的方法,以实现能效解决方案的规模化。
提出的方法
- 提出'五大支柱'框架:最优仪表化、子系统互联、推断决策、人员参与以及智能运行。
- 以15分钟间隔的高分辨率智能电表数据以及来自电流互感器(CTs)或电器传感器的分支电表数据作为主要输入。
- 应用信号处理与机器学习技术,如组合优化、边缘检测和因子隐马尔可夫模型,用于NILM。
- 将环境传感器(如温度、人员占用)的多模态数据与电气数据集成,以提高分解精度。
- 通过项目化能耗反馈和可操作洞察,建立反馈回路以促进人员参与。
- 推动从集中式、监督式离线学习向实时、自动化控制系统的转变,以实现动态能耗优化。
实验结果
研究问题
- RQ1如何对智能电表、BMS和环境传感器的建筑能耗数据进行最优仪表化,以在成本与精度之间取得平衡?
- RQ2如何通过互联孤立的建筑子系统(如暖通空调、照明、安防)来提升能效表现?
- RQ3从分解后的负载数据中推断决策,如何实现故障检测、异常报警和节能建议?
- RQ4人员行为与反馈机制在降低建筑整体能耗方面发挥何种作用?
- RQ5基于NILM输出实现实时、智能控制系统的实际技术与计算挑战是什么?
主要发现
- 五大支柱框架——仪表化、互联、推断、人员参与和智能运行——为建筑中的数据驱动能效提供了全面的结构化指导。
- 非侵入式负载监测(NILM)贯穿全部五个支柱,展示了如何从聚合功率数据中推断出电器级别的能耗使用情况。
- 监督式NILM方法依赖于分支电表的真值数据,导致对昂贵仪表化的依赖,从而凸显了无监督学习方法的必要性。
- 计算开销大的NILM算法通常排除低功耗电器,限制了其在现实场景中的可扩展性与实际影响。
- 将电气数据与环境传感器数据(如人员占用、温度)互联,可提高分解精度,并实现上下文感知的能效优化。
- 实证研究表明,向人员提供分项能耗反馈可有效降低用电量,凸显了参与式能效管理的重要性。
更好的研究,从现在开始
从阅读论文到最终审阅,大幅缩短您的研究时间。
无需绑定信用卡
本解读由 AI 生成,并经人工编辑审核。