Study on Fault Detection by Pattern Analysis in Data Driven Energy Management

Part 1-Application to Fault Detection for Energy Consumption of Lighting

Abstract

This study investigates a the mechanical detection method based on pattern analysis to understand the operation situation of a building and detect equipment faults and improper operation for utilizing BEMS data to energy management. First, the mechanism of the proposed method generates consumption patterns automatically from operation data using clustering analysis. Next, test data are classified into similar patterns and detected as anomalies based on the distance. The lighting power consumption data of an office were used for detection, and the detection accuracy was compared and evaluated when using only power consumption and when using indoor illuminance, outdoor solar radiation, and occupancy information as auxiliary data. Moreover, a simplified method for removal of anomaly data from training data in clustering was proposed, and its effectiveness was confirmed from a verification using real operation data.