CAREER: A Scalable Occupant-Driven Energy Optimization System for Commercial Buildings

该项目寻求从根本上改变人们的日常生活方式，迈向一个对环境更负责任和更可持续的未来，并有可能在减少我们对化石燃料的依赖和应对气候变化方面发挥重要作用。这项研究将通过提供每项行动的能源成本的实时可见性，帮助个人减少在室内商业建筑环境中的个人能源消耗。这将通过为居住者和建筑物管理人员提供具有可量化节约的可行建议以及洞察力，从而实现商业建筑的能源问责，以便他们能够及时采取行动。这些小小的节省可以加在一起：即使商业建筑能耗降低1%，也可以为国家每年节省15亿美元。该项目将有助于推进建筑规模系统、物联网、可穿戴系统和推荐系统的交叉研究。针对嵌入式系统、移动计算、物联网和深度强化学习开发的课程模块将用于培养本科生和研究生。这项提案旨在通过研究每个居住者独特和个性化的能源使用情况(或称“能源足迹”)，并为居住者提供可操作和可衡量的节能建议，显著降低商业建筑的能源消耗。该系统由多个组件组成，包括实时感知和驱动、建筑能耗监测、室内定位、大规模时间序列数据分析和推荐系统。该项目分为三个研究方向：(1)开发一个商业建筑的数字孪生模型来模拟人类行为的节能效果，并研究高效的算法来计算每个居住者在共享环境中的“能源足迹”。(2)推进深度强化学习中的知识，设计推荐系统，发现最具节能潜力的动作，同时适应用户偏好。(3)探索有效的反馈机制和激励机制，鼓励节能行为，提高推荐质量，提高推荐接受率。研究成果，包括数据集、嵌入式硬件设计、软件代码、模拟器、智能手机应用代码、报告、演示文稿和论文，将通过研究小组网站和GitHub与研究社区共享。为了确保隐私，数据集中的任何个人身份信息都将被删除。为了鼓励其他人进行部署和实验，记录在案的开源软件和硬件将在https://github.com/Columbia-ICSL/EnergyFootprinting的GitHub存储库中发布并维护至少五年。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

This project seeks to fundamentally change how people live their everyday lives towards a more environmentally responsible and sustainable future and has the potential to play a major role in reducing our reliance on fossil fuels and combating climate change. This research will help enable individuals to reduce their personal energy usage in indoor commercial building settings by providing real-time visibility into the energy cost of each action. This will enable energy-accountability in commercial buildings by providing actionable recommendations with quantifiable savings, as well as insights for both occupants and building managers so that they can act in a timely manner. These small savings can add up: even a small 1% reduction in commercial building energy consumption translates to 1.5 billion dollars of annual savings for the nation. This project will help advance research at the intersection of building-scale systems, Internet-of-Things, wearable systems, and recommender systems. Course modules developed on embedded systems, mobile computing, Internet-of-Things, and deep reinforcement learning will be used to train undergraduate and graduate students.This proposal aims to significantly reduce energy consumption in commercial buildings by examining each occupant’s unique and individualized energy usage, or “energy footprint”, and providing occupants with actionable and measurable energy-saving recommendations. The proposed system comprises of several components, including real-time sensing and actuation, building energy monitoring, indoor localization, large-scale time-series data analytics, and recommender systems. This project is organized into three research thrusts: (1) develop a digital twin model of a commercial building to simulate energy savings from human-driven actions, and research efficient algorithms for computing each occupant’s “energy footprint” in shared environments. (2) advance knowledge in deep reinforcement learning and design a recommender system to discover actions that have the best potential for saving energy while adapting to user preferences. (3) investigate effective feedback mechanisms and incentive schemes to encourage energy saving behaviors, increase recommendation quality, and improve recommendation acceptance rate.Research results, including datasets, embedded hardware designs, software code, simulators, smartphone application code, reports, presentations, and papers will be shared with the research community through the research group website and GitHub. To ensure privacy, any personally identifiable information in datasets will be removed. To encourage deployment and experimentation by others, documented open-source software and hardware will be release and maintain in the project’s GitHub repository at https://github.com/Columbia-ICSL/EnergyFootprinting for at least five years.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.