SmartEMS: Applying machine learning in building energy management systems

SmartEMS项目将开发适用于非住宅建筑能源管理系统的数据分析和机器学习方法，并在实际建筑中进行测试。最近机器学习能力的巨大进步使得训练和部署这样的算法来解决实际挑战成为可能。建筑物中复杂的能源系统面临着许多这样的挑战，从集点优化到预测控制。SES Inc.拥有充分利用这一优势的能力和客户群。有两种核心方法：从数据中离线学习和实时预测控制。前者将通过分析业务数据确定潜在趋势和关注领域；后者将开发和训练机器学习控制器，以改善基于天气预报的操作。两者结合起来可以提供一个高度灵活，强大的解决方案。SmartEMS将使用开源系统和协议（voltron, BACnet），这些系统和协议已被SES公司在以前的项目中成功使用。将使用基于Python的开源最先进的机器学习库（scikit-learn, TensorFlow），以及其他基于Python的数据分析和可视化库。远程访问接口硬件（裸机；树莓派）将部署在3个测试大楼（一个大学校园和SES Inc.的两个客户）。CANARIE的云计算将用于该过程的计算密集型部分。产出将是基于最新学术研究的商业可部署解决方案；其中的一部分也将作为开源发布。这将成为持续合作的基础。这个概念在改善商业建筑的能源使用、排放和舒适度方面具有巨大的潜力。

The SmartEMS project will develop data analysis and machine learning approaches suitable for incorporationin non-residential building energy management systems, and test them in real buildings. Dramatic recentimprovements in the power of machine learning have made it possible to train and deploy such algorithms tosolve practical challenges. Complex energy systems in buildings present many such challenges, from set-pointoptimization to predictive control. SES Inc. has the capabilities and client-base to take advantage of this.There are two core approaches: offline learning from data, and real-time predictive control. The former willidentify underlying trends and areas of concern by analysis of operational data; the latter will develop and trainmachine learning controllers that will improve operation based on weather predictions. The two together candeliver a highly flexible, robust solution.SmartEMS will use open-source systems and protocols (VOLTTRON, BACnet) that have been successfullyused by SES Inc. on previous projects. Open-source state of the art machine learning libraries based in Python(scikit-learn, TensorFlow) will be used, along with other Python-based data analysis and visualisation libraries.Remotely accessible interface hardware (bare-bones PCs; Raspberry Pi) will be deployed in 3 test buildings(one university campus and two clients of SES Inc.). Cloud computing from CANARIE will be used forcomputationally intensive parts of the process.The output will be a commercially deployable solution based on the latest academic research; parts of this willalso be released as open-source. This will form the basis for an ongoing collaboration. The concept hassignificant potential to improve energy use, emissions and comfort in commercial buildings.