Development and Application of an Inverse Building Model for Demand Response in Small Commercial Buildings

This paper describes development of an inverse building model and its application in studying the performance of a demand-limiting (DL) control strategy. The demand-limiting strategy involves precooling (PC) a building during unoccupied times, maintaining the zone temperature setpoints at the lower limit of comfort during off-peak, occupied periods, and then limiting the peak cooling rate to a target for on-peak, occupied times that keeps zone conditions within comfort limits. Data from the Iowa Energy Center (IEC), which is typical of small commercial buildings, were used to train an inverse model that was then employed as a tool to evaluate the potential for peak load reduction through control of building thermal mass. The potential for demand limiting was investigated through parametric analysis compared with night-setup (NS) control. INTRODUCTION In small commercial buildings, the control of dedicated HVAC equipment is achieved through the use of a thermostat, which allows the setting of zone temperature set points. Most HVAC systems employ a strategy called night-setup control. Under night-setup control, the air temperatures are in the middle of the comfort range during occupied times and the cooling system is typically inactive during unoccupied periods. Recently there has been increased interest in the development of setpoint adjustment strategies that utilize building thermal mass to perform load shifting and peak load reduction (Braun (1990), Kintner-Meyer and Emery (1995), Keeney and Braun (1996), Keeney and Braun (1997), Braun et al. (2001), Chaturvedi and Braun (2002), Henze et al. (2004)). These strategies typically involve precooling during unoccupied periods and raising of setpoints during occupied, on-peak periods. Transient building models are useful in the development and evaluation of strategies that take advantage of building thermal mass. Building models can be categorized as forward or inverse models. Forward models use a physical description of the building, whereas inverse models are trained using on-site measurements. Forward models are necessary for design, whereas inverse models can be useful for developing site-specific control strategies. There has been a lot of work on the development of simplified inverse building models. Recently, Chaturvedi and Braun (2002) developed an inverse gray-box building model that was applied by Braun et al. (2002) to develop and evaluate load-shifting strategies for a large commercial building. The model uses a simplified physical representation with a thermal network of resistors and capacitors. The network is transformed to a transfer function representation and used to predict the cooling loads. The resistors and capacitors are learned using a training algorithm that minimizes model prediction errors for the overall cooling load. The model was tested using data from a field site in Chicago and was able to predict cooling loads to within about 9% for a four-week test period. Gouda et al. (2002) also suggested a lumped parameter building model based on a thermal network. Parameters for the wall models were estimated by nonlinear constrained optimization using Kuhn-Tucker equations. Free-floating simulation results for internal air temperatures for low and high capacity rooms based on 2-order walls were compared with measured data. The work described in the current paper involved further development, evaluation, and application of the inverse building model presented by Chaturvedi and Braun (2002). An inverse model was developed and applied to the