Advanced cooling system optimization algorithm development for building energy management

随着对温室气体排放的担忧增加以及未来能源成本的不确定性，提高建筑能效的需求正在增加。对建筑系统的监测被广泛认为是通过测量和验证实际性能来提高和保持建筑能源效率的重要的第一步。一旦这些监测系统到位，收集到的数据可以为一些能源管理活动提供信息，例如重新调试和微调建筑物控制。然而，尽管开发人工智能（AI）支持的控制系统的研究程度很高，但商业系统通常仅限于更简单的算法，并且需要大量时间来部署和定制这些系统以适应特定的建筑物，以及季节性调整。因此，使用先进的人工智能方法开发自动控制调谐以允许在线学习和季节性调整是非常可取的。****这个项目专注于一种新的方法，将强化学习应用到现有的建筑控制能源管理算法中。这些算法根据过去的性能和最优系统特性修改温度和设备控制点。这种方法将大大减少新建筑的部署时间，减少或消除季节性优化转换的人为干预，从而允许将该技术快速扩展到建筑组合级别。****在这个项目中，我们将开发两代新算法，以行业合作伙伴开发的现有模型为基础，将上述在线学习纳入其中。通过与现有设施合作，算法将在试点项目中进行测试和部署，以确认实际建筑物的实际节能效果。该项目的长期目标是提高对如何优化整个系统的理解，从而允许集成额外的功能，如现场存储和可再生能源发电。

The need for improved energy efficiency in buildings is increasing as concerns over GHG emissions rise and uncertainty over future energy costs. Monitoring of building systems is widely recognized as an important first step in improving and maintaining building energy efficiency by measuring and verifying actual performance. Once these monitoring systems are in place, the data collected can inform several energy management activities such as re-commissioning and fine-tuning of building controls. However, despite the high degree of research to develop artificial intelligence (AI)-supported controls systems, commercial systems are typically limited to simpler algorithms and require significant time to deploy and tailor these systems to a particular building, along with seasonal adjustment. Therefore, the development of automated control tuning using advanced AI approaches to permit online learning and seasonal adjustment is extremely desirable. ****This project focuses on a novel approach to implementing reinforcement learning into existing energy management algorithms for building controls. These algorithms modify temperature and equipment controls points based on past performance and optimal system characteristics. This approach will substantially decrease the deployment time for new buildings, reduce or eliminate human intervention for seasonal optimization changeover, and thus permit rapid scaling of this technology to the building portfolio level. ****In this project, we will develop two new generations of algorithms, building on existing models developed by the industry partner to incorporate the online learning described above. By working with existing facilities, the algorithms will be tested and deployed on pilot projects to confirm actual energy savings in real buildings. The long-term goal of this project is to improve the understanding of how the overall system can be optimized, thus permitting the integration of additional functionality such as on-site storage and renewable energy generation.