A temperature-based approach to detect abnormal building energy consumption

Highlights

• We developed DET-Toa method to detect abnormal building energy consumption fault.

• The method successfully detected 19 faults in the 20 simulation test cases.

• The DET-Toa method has better performance than the DET-Date method in the test.

• The DET-Toa method uses outside air temperature dependence in its detection metric.

• The smallest faults that could be identified by the DET-Toa method are summarized.

Abstract

This paper introduces a temperature-based approach – called the Days Exceeding Threshold-Toa (DET-Toa) method to detect persisting small increase or decrease in the normal building energy consumption. This method identifies an abnormal energy consumption fault when the deviation between the measured and simulated consumption is greater than one standard deviation of the residuals in the baseline period and persists for at least 20 days which are consecutive when ordered according to increasing or decreasing outside air temperature. The fault detection capability of the method is evaluated with simulation tests with two on-campus buildings. Ten synthetic control changes were assumed to happen and lasted for one year for each building. In the test, the DET-Toa method showed superior capacity for detecting abnormal building energy consumption compared with the DET-Date method. It successfully detected 19 synthetic control changes in the 20 simulation test cases. The reason for the better performance of the DET-Toa method is discussed and demonstrated in an example case. The smallest faults that could be identified by the DET-Toa method in the two analyzed buildings and their related energy consumption impact statistics are also provided in the paper.

Introduction

Building commissioning services have proven to be successful in saving building energy consumption [1]. The Energy Systems Laboratory at Texas A&M University began Continuous Commissioning^® (CC^®) in 1992. The CC^® process has produced an average energy savings of about 20% without significant capital investment in over 150 large buildings in which it has been implemented [2]. Though commissioning services are effective in reducing building energy consumption, the optimal energy performance obtained by commissioning may subsequently degrade. Faults in HVAC systems can increase HVAC energy consumption by 30% [3]. A study of the persistence of savings in 10 university buildings over a period of 12 years found that on an average, $1000/year of savings in heating or cooling would decrease to $750/year in 3–5 years if there was no follow-up effort to maintain the savings [4]. Fault detection which can alert operators early after the onset of significant increases in consumption would have great value.

Fault detection in buildings has been widely investigated in recent decades. There are two fundamental approaches to fault detection in buildings: a component level approach and a whole building approach. Component level fault detection focuses on faults in individual systems, e.g., air-handling units (AHU) [5], [6], variable-air-volume (VAV) boxes [7], [8], chillers [9], [10] or boilers [11], [12]. Whole building fault detection compares building energy consumption (e.g., electricity, gas, chilled water) to assess whether or not a building and its systems operate efficiently [13]. Whole building fault detection needs much less data compared to fault detection at the component level, and thus saves time and effort. Representative previous studies about whole building fault detection are reviewed below.

Dodier and Kreider [14] illustrated a method using an Energy Consumption Index, the ratio of measured energy consumption to predicted energy consumption as determined from a neural network, to determine if the energy consumption was above normal, normal or below normal. Seem [15] and Jankkula and Cook [16] described methods using outlier detection for detecting abnormal energy consumption in buildings. Sun et al. [17] developed an approach for fault detection based on hourly energy consumption of HVAC system, integrating a gray-box model, statistical process control technique and Kalman filtering method. Lee et al. [18] examined the use of the ASHRAE Simplified Energy Analysis Procedure (SEAP) [19] for fault detection at the whole-building level. Visual comparison with measured post-commissioning data is used to detect significant deviations from expected performance. Curtin [20] developed a prototype Automated Building Commissioning Analysis Tool (ABCAT) following the approach of Lee et al. [18]. A “Cumulative Cost Difference” plot is the primary fault detection metric. Bynum et al. [21] presented the application of ABCAT in retrospective and live test cases.

Lin et al. [22] presented a statistical method using a Days Exceeding Threshold plot to detect persisting small increase or decrease in the normal energy consumption. This method is referred to as the Days Exceeding Threshold-Date (DET-Date) method in this paper. In this method, a calibrated simulation model is used to predict the normal building energy performance. An abnormal energy consumption fault is identified if the deviation between the measured and simulated consumption is greater than one standard deviation of the residuals between measured and simulated consumption in the baseline period and persists for at least 20 consecutive days.

Building cooling and heating consumption variations due to HVAC operational changes are temperature-dependent. Claridge et al. [23] addressed “characteristic signatures” using for calibration. A characteristic signature is a normalized plot of cooling and heating consumption deviation caused by a specified change in HVAC system operation as a function of outside air temperature. For each operational change in a specified system, the characteristic signature has a characteristic shape as a function of outside air temperature. As an example, Fig. 1 shows the outside airflow characteristic signatures for a dual-duct variable airflow volume system for the case where the outside airflow ratio X_oa (outside airflow volume/maximum designed airflow volume of the system) increases 6%. The plot on the left shows that the chilled water (CHW) characteristic signature shows a consumption decrease of about −10% at 2 °C and the consumption gradually increases to 15% over the range of 2–32 °C. The plot on the right shows that the hot water (HW) characteristic signature gradually decreases from 5% to 0% from 2 °C to 20 °C and stays at 0% from 20 °C to 32 °C. The change in heating or cooling consumption is expressed on the Y axis as a percent of the maximum baseline cooling or heating consumption, respectively for a cooling or heating characteristic signature.

Since building cooling and heating consumption variations due to HVAC operational changes are temperature-dependent, it would be beneficial if we take temperature dependence into account when setting up the fault detection metric. The proposed DET-Toa method identifies abnormal building energy consumption based on this concept. Toa stands for the daily average outside air temperature.

The DET-Toa method is introduced and demonstrated using simulation tests on two buildings. The fault detection results of the DET-Toa method are compared with those of the DET-Date method. Finally, the smallest faults that could be identified by the DET-Toa method in the two buildings analyzed are determined and their related energy consumption impact statistics are provided.