Building energy modeling (BEM) using clustering algorithms and semi-supervised machine learning approaches

Energy efficiency is a critical element of building energy conservation. Energy Information Administration (EIA) and International Electrotechnical Commission (IEC) estimated that over 6% of electrical energy was lost during transmission and distribution. Sensing and tracking technologies, and data-mining offer new windows to better understanding these losses in real-time. Recent developments in energy optimization computational methods also allow engineers to better characterize energy consumption load profiles. The paper focuses on developing new and robust data-mining techniques to explore large and complex data generated by sensing and tracking technologies. These techniques would potentially offer new avenues to understand and prevent energy losses during transmission. The paper presents two new concepts: First, a set of clustering algorithms that model the supply-demand characterization of four different substations clusters, and second, a semi-supervised machine learning and clustering technique are developed to optimize the losses and automate the process of identifying loss factors contributing to the total loss. This three-step process uses real-time data from buildings and the substations that supply electricity to the buildings to develop the proposed technique. The preliminary findings of this paper help the utility service providers to understand the energy supply-demand requirements.