Application of Clustering Techniques to Energy Data to Enhance Analysts' Productivity

摘要

Honeywell Atrium~(TM) collects and analyzes its customers' utility data to study building operations, energy usage, equipment performance and diagnostics for the purpose of reducing customers' operational costs. Each Atrium~(TM) analyst is responsible for supporting several hundred buildings and supporting customers with recommendations for cost reduction. This makes it necessary to develop tools that will direct analysts towards problem facilities, rather then manually scanning all available data. In this paper we present two different applications of clustering analysis to discover anomalous data. The first application identifies facilities that operate with unusual electrical load profiles, based on characteristics of multiple facilities' electric operational data. The second application identifies days exhibiting abnormal operation within individual facilities. Our experiments using data sets derived from energy consumption profiles of a major retailer show that clustering analysis is a useful tool in guiding analysts to identify anomalous buildings and daily profiles and enhancing analysts' productivity.