•  
  •  
 

Abstract

Given the rise in building energy consumption and demand worldwide, energy inefficiency detection has become extremely important. A significant portion of the energy used in commercial buildings is wasted as a result of poor maintenance, degradation or improperly controlled equipment. Most facilities employ sensors to track energy consumption across multiple buildings. Smart fault detection and diagnostic systems use various anomaly detection techniques to discover point anomalies in consumption. While these systems work reasonably well in detecting equipment anomalies over short-term intervals, further exploration is needed in finding methods that consider long-term consumption to detect anomalous buildings. This paper presents a novel approach for a multi-building campus to rank and visualize the long-term volatility of building consumption. This allows for the optimal allocation of limited time and resources for the detection and resolution of energy waste. The proposed method first classifies daily consumption into 5 classes using an ensemble learner and then calculates the information entropy on the resulting classification set to determine volatility. The ensemble learner receives input from a K-Nearest Neighbor classifier, a Random Forest classifier and an Artificial Neural Network. In general, buildings are expected to keep the same energy profile over time, all else being equal. Buildings that frequently change energy profiles are ranked and flagged by the system for review, which would call for the next step to reduce waste and costs and to increase the sustainability of buildings. Data on energy consumption for 132 buildings is obtained from energy management at the Georgia Institute of Technology. Experimental results show the effectiveness of the proposed approach.

Download

Share

COinS