Detecting thermal anomalies in buildings using frequency and temporal domains analysis

Abstract

Identifying and modifying areas in a building where thermal anomalies occur will help to maintain a comfortable indoor thermal environment while reducing energy waste. In this study, a Fourier transform-based method was developed to detect places with thermal anomalies by considering the periodic characteristics of the temperature time series. Results were compared with anomalies detected by a support vector regression-based method. Both methods were applied to six days of temperature data obtained from sensors placed in various locations in two office buildings, each building having 60 sensors placed within them during the test period. Results showed that both Fourier transform and support vector regression based methods are capable of identifying abnormal time series without prior knowledge of the series. This developed Fourier transform-based anomaly detection method can be used effectively for anomaly detection in time series with periodic characteristics, which will make it easier to identify poorly-performing areas within an internal building environment using non-permanent sensors.