CAREER:Energy Management for Smart Residential Environments through Human-in-the-loop Algorithm Design

虽然在考虑网络和物理特性的电网控制方面取得了实质性进展，但在智能电网研究的整合方面存在差距，因为它与人类行为相结合-特别是与能源管理系统的互动。例如，住宅能源消耗在过去几十年中迅速增加，特别是在美国，2015年消耗了2.6万亿千瓦时，预计到2040年将再增加13.5%。已经进行了诸如需求响应的研究工作，以减少这种消耗，特别是在智能住宅环境中。需求响应等概念在很大程度上忽视了人类行为和感知的复杂性，社会科学领域的最新研究和最近的经验对这种方法的有效性提出了挑战，在某些情况下导致了对这些概念的放弃和回避。该提案的目的是通过设计专门考虑用户行为，感知和心理过程的新颖算法，机器学习模型和优化技术来克服与最先进的能源管理系统相关的限制。这种革命性的方法将释放智能住宅环境在降低住宅能耗方面的全部潜力，并有可能改变人们设计、实施和使用能源管理系统的方式。该项目还支持创新的教育活动，如课程，真实的时间演示，编码挑战和高中生的研究经验。PI还将带领一批学生参加以多样性为导向的Grace Hopper会议，并为西班牙裔小学生举办研讨会。最后，我们将设计一个新的关于网络-物理-人类系统的课程，一些研究生和本科生将参与研究活动。拟议的研究结合了新颖的算法，机器学习和优化解决方案，考虑了以前未研究过的人类行为，感知和心理过程。具体来说，为了实现细粒度的能源监测，我们提出了新的基于流的智能插座的电器识别算法。这些算法学习设备消耗签名和用户与系统的互动，以优化学习过程。此外，节能优化策略的设计，通过考虑用户的感知，通过社会行为幸福模型。这些模型通过基于回归图、插值和回归的新型机器学习算法学习和改进，并使用智能手机提供的用户反馈。此外，我们还开发了在配备可再生能源发电的智能住宅环境中进行能量交换的优化算法。这些算法通过考虑和学习用户在能量交换过程中的可用性和偏好来匹配用户的需求和生产。拟议的研究通过真实的试验台和基于真实的痕迹的大规模模拟进行验证。该奖项反映了NSF的法定使命，并被认为值得通过使用基金会的智力价值和更广泛的影响审查标准进行评估来支持。

While substantial progress has been made in the control of electric grid considering the cyber and physical characteristics, there has been a gap in the integration of smart grid research as it integrates with human behavior -- especially in interactions with energy management systems. For example residential energy consumption has been rapidly increasing during the last decades, especially in the U.S. where 2.6 trillion kilowatt-hours were consumed during 2015, and an additional 13.5% increase is expected by 2040 . Research efforts such as demand response have been made to reduce this consumption especially in smart residential environments. Concepts such as demand response have largely overlooked the complexity of human behaviors and perceptions, and recent research in the social-science domain and recent experience has challenged the effectiveness of this approach and in some instances led to an abandonment and avoidance of such concepts. The objective of this proposal is to overcome the limitations associated with state-of-the-art energy management systems by designing novel algorithms, machine learning models, and optimization techniques that specifically consider user behaviors, perceptions, and psychological processes. This revolutionary approach will unleash the full potential of smart residential environments in reducing residential energy consumption and has the potential to transform the way in which energy management systems are designed, implemented, and used by people. This project also supports innovative educational activities such as classes, real time demonstrations, coding challenges, and research experiences for high school students. The PI will also lead a cohort of students to the diversity-oriented Grace Hopper conference and teach seminars for Hispanic elementary students. Finally, a new class on Cyber-Physical-Human System will be designed and several graduate and undergraduate students will participate in the research activities.The proposed research combines novel algorithmic, machine learning, and optimization solutions that consider previously un-examined human behaviors, perceptions, and psychological processes. Specifically, in order to enable fine grained energy monitoring, we propose novel stream-based appliance recognition algorithms for smart outlets. These algorithms learn the appliance consumption signatures and the user engagement with the system to optimize the learning process. In addition, energy saving optimization strategies are designed by considering the user perception through social-behavioral well-being models. These models learned and refined through novel machine learning algorithms based on regressograms, interpolation, and regression using user feedback provided through a smartphone. In addition, we develop optimization algorithms for energy exchange in the context of smart residential environments equipped with renewable energy generation. These algorithms match the users' demand and production, by considering and learning also their availability and preferences in the energy exchange process. The proposed research is validated through real testbeds and large-scale simulations based on real traces.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.