Why Occupancy-Responsive Adaptive Thermostats Do Not Always Save - and the Limits for When They Should

So-called ‘smart thermostats’ are beginning to fill the gap left in efficiency programs after researchers and policy makers discovered that in practice, simple programmable thermostats do not guarantee energy savings. As a result, EPA ended EnergyStar certification of programmable thermostats in 2010. Many recent pilots for communicating thermostats, occupancy-responsive thermostats, and adaptive control schemes have shown significant annual HVAC savings on the order of 10-20%. However, the form and function for technologies in this space vary widely. Some controls merely allow for remote management (e.g., web-based setpoint scheduling or smart-phone interface and control), while other devices monitor occupancy and automatically adjust set-points when a space is vacant. Still other technologies automatically adapt to user behaviors and preferences in order to anticipate changes and adjust HVAC operation. These differences have different savings implications. Further, the application into which any of these technologies is installed also impacts savings potential. The study focuses particularly on a series of pilot evaluations conducted with one occupancy-responsive adaptive thermostat system that resulted in very little energy savings during normal operation in university residence halls. These results came as a great surprise to the research team, especially since the HVAC system run-time for vacant zones was reduced to nearly zero in the buildings. The detailed evaluation of this case forms a conceptual basis for explanation of the limitations for smart thermostat devices. The research shows that considerable savings can be had in certain instances, but that the impact is sensitive to technology and application. The study also reviews previous research on the technology and recommends methodological improvements for future studies. Introduction & Technology Overview ‘Smart thermostats’ are characterized generally by their communicating capabilities, including web and mobile user interface options, as well as networked control that allows for instantaneous management of multiple thermostats in a facility. Smart thermostats may include occupancy responsive control, adaptive or learning functionality, demand response capability, fault detection and diagnostics, and runtime optimization features. This promises general improvement to programmed setpoint scheduling, as well as automated schedule and setpoint optimization. However, amidst the range of new and emerging opportunities in this space, it is not clear which technology features actually provide energy savings, which improve level of service, which enhance usability, or which are actually of little technical value 1 (Lopes et alt. 2010, Meier et alt. 2010, Peffer et alt. 2011, Pistochini et alt. 2008, Woolley et alt. 2012). Building from programmable ‘setback’ thermostats and modern lighting controls, occupancy-responsive thermostats adjust operation for heating, cooling, and ventilation when a space is vacant (Gupta, Intille, and Larson 2009, Lu et al, 2010). Most occupancy responsive thermostats do this by shifting the occupied temperature set point to a setback, which allows the room temperature to drift and should result in reduced runtime for heating 2 and cooling equipment. In certain applications it may also reduce energy use related to ventilation. Generally, this adjustment is intended to capture energy savings when no occupants are detected while also maintaining level of service (thermal comfort, indoor air quality, sense of control) during occupied periods. However, understanding the transition from the unoccupied to occupied state is critical for predicting energy savings. When the set-point is restored, additional energy must be expended for a period of time to recover from the setback. For example, if the setback and temperature drift occur during a hot afternoon, and recovery is in the evening, the energy saved during the setback will be greater than energy needed for recovery. There are also conditions for which the energy for recovery exceeds the energy saved during the setback period. Setback may also create periods of unsatisfactory thermal comfort for occupants (Manning et al, 2007). Adaptive controls automatically change operating parameters according to learned and predicted factors. These systems adapt over time according to measured responses. They can learn about system physical characteristics (cooling capacity, temperature response time, etc) and user schedules and preferences (e.g., Nest, EcoFactor) in order to predict appropriate setback periods and ranges. These features can save energy, improve thermal comfort and/or improve convenience and user experience. For example, EcoFactor will automate schedule programming. These learning algorithms can be integrated with features that respond to occupant proximity (eg: Allure Energy), or that predict occupant comfort according to user feedback and measured and forecast outdoor temperature (eg: as per ASHRAE 55 Adaptive Thermal Comfort). This study focuses on one occupancy-responsive adaptive thermostat technology that learns about system response capabilities and automatically programs a setback for vacant periods to ensure a timely recovery to the comfort setpoint when a room is again occupied. Field Study Methodology The authors collaborated with Student Housing and the Energy Management Office at the University of California, Davis to monitor and analyze field performance of an occupancyresponsive adaptive thermostat technology by Telkonet. A previous paper outlines a preliminary study (Woolley et al, 2012). The SS6000 Energy Management Thermostat is the center of the Telkonet EcoInsight system, which includes a ZigBEE mesh network, gateway, and centralized web-based user interface. Each thermostat has an on-board (or remote wireless) infrared motion detector. Additionally, the system incorporates an on-board light sensor and logic to distinguish between vacancy and a nighttime condition where occupants are sleeping. 1 Another factor in success of any technology is the appropriate match of application; in previous work, we suggested that occupancy responsive thermostats should be cost-effective in dormitory settings because of high occupancy and low predictability (Woolley et al 2012). 2 Studies performed in the 1970s, based on models of energy flows through a house, suggested that on average a daily eight-hour nighttime setback could bring approximately 1% reduction in natural gas consumption for each degree Fahrenheit offset (Nelson et al 1978). Telkonet applies a learning algorithm called Recovery Time that continually adapts the setback temperature for unoccupied periods so that a room can recover within an acceptable period upon the occupant’s return. The thermostat learns how quickly the associated mechanical system is able to respond, allowing the room temperature to drift only so far that it can still return to the occupied set-point within the allotted time. During occupied periods, users are allowed temperature control, although facility managers may limit the selectable set-point bandwidth to avoid excessive heating or cooling by residents. In this study, the Telkonet system was installed in four dormitory buildings, each a 5story concrete and steel structure originally constructed in 1965. The residence halls each have 110 rooms and various common spaces, such as corridors, meeting rooms, laundry rooms and bathrooms (about 4,000 m 2 ). Rooms occupy about 50% of the total floor area. These buildings are equipped with two-pipe, three-speed fan coil systems for heating and cooling, so all rooms are restricted to either cooling or heating during any given period. Chilled water and steam supplied from the campus central plant are generally switched only once per season. Each room has a fan coil that is controlled by a thermostat in the room. Fresh air ventilation is provided to each room by infiltration and through operable windows. Corridor spaces have separate controls. Telkonet occupancy-responsive thermostats were installed throughout Bixby Hall in September 2011. Installation in Malcolm, Gilmore and Ryerson followed in May, June and July 2012 respectively. In all cases, Telkonet thermostats replaced unrestricted manual thermostats. We collected data in cooling seasons and during periods of high and low occupancy corresponding to the academic quarter, and summer conference housing periods. Data included whole building chilled water energy consumption, outside air temperature, occupancy, thermostat state, active set-point temperature (or set-back temperature), room temperature, and fan coil run time in every room. Since historical whole-building chilled water energy consumption data was only available for Ryerson and Gilmore, data from cooling season performance in September – October 2012 (post-installation) were compared against chilled water energy consumption data from April – May 2012 (pre-installation). Further, from April 2012 to February 2013, the thermostats in Gilmore and Malcolm were switched between an occupancy-responsive mode and a conventional operating mode in alternating weeks (ON-OFF). This allowed for comparison both in academic and non-academic periods (Table 1). Table 1: Data periods utilized for study