Background

BACTool focuses on the exploration of so-called "non-standardized" control solutions, i.e. solutions where the control is being tailored to the given building, combination of automated subsystems and user requirements by means of corresponding programs that govern the behavior and interplay of the individual subsystems. Building automation systems with programmable controllers are typically used for that purpose. Traditional programs are susbsumized under the term Rule-Based Control (RBC). A promissing alternative that is currently undergoing intensive research is co-called Model Predictive Control (MPC). A brief introduction to RBC and MPC is given below. A more in-depth discussion of the two approaches and a comprehensive list of criteria for the evaluation of non-standardized control solutions is given in [1] (in particular see the Introduction therein).

All data presented in the BACTool are simulation results. They were produced using the data and methods described in [1]. The simulations were based on a 12th order multiple-input-multiple-output bilinear model of the coupled thermal, light and air quality dynamics of a single room or building zone. The model is used, firstly, as a "plant model" to simulate the building zone's response to different control algorithms. Secondly, the same model is also used as a "controller model" for Model Predictive Control (see below). Details on the model can be found in [2] and [5]. Each simulation covered one year and employed a time step of one hour. The used weather data sets are documented in [6]. Statistical analyses dealing with trends and patterns across a large number of simulations are provided in [7],[8],[9],[10]. A more detailed analysis of selected cases can be found in [8].

Control Algorithms

Rule-Based Control (RBC) determines the control inputs based on a series of rules of the kind "if condition then action". The conditions and actions typically involve numerical parameters (e.g., threshold values), the so-called control parameters. In BACTool they are determined from building parameters based on carefully derived, automated calculation procedures. The used RBC algorithms consist of a high-level and a low-level part. The high-level part yields operating modes that determine the "low-cost" actions (blind positioning, free cooling operation, and energy recovery operation). The low-level part determines, firstly, the control actions for the "low-cost" action aggregates. In a second step it calculates the remaining control outputs for "high-cost" actions such as active heating or cooling, and mechanical ventilation. The various RBC variants differ only in their high-level parts. For further information see [3]

Model Predictive Control (MPC) relies upon a model of the building that is used together with predictions of relevant disturbances (e.g., weather, internal gains) to predict the system's future evolution. At the beginning of each time step (e.g., every hour) MPC computes the "best possible" sequence of control actions that minimize a cost function (e.g., total energy demand) over a given prediction horizon (e.g., a few days) while respecting comfort (e.g., illumination levels, room temperature range) and any other (e.g., maximum power demand) constraints. The control actions identified for the very first time step within the prediction horizon are then applied to the system, and the whole procedure is repeated at the beginning of the next time step. This "receding horizon" approach ensures that the control plan is continuously updated using the newest information on the building's state, thus allowing to account for model inaccuracies or any unknown disturbances that have meanwhile acted on the building. See also [4].

The Performance Bound (PB) is a theoretical value that presents the lowest achievable control cost (in terms of energy or money) for a given building, cost function, disturbances (weather, internal gains) and set of comfort requirements. The PB can be estimated by applying Model Predictive Cnotrol (see below) over a representative period (e.g., one year) assuming a perfect building model, and perfect knowledge of all future disturbances. Knowledge of the PB makes it possible to compare different design variants for a given system net of any effects related to control.

Theoretical energy savings potential: By definition the PB presents a theoretical number that can not be beaten by any real controller. The difference between a real controller's energy usage and the PB gives a measure of the maximum achievable improvement for that controller. Nothing can be said about to what extent the potential can be exploited by a feasible control. However, the size of the potential indicates for what applications further control strategy development may be promising.

• PB: The PB calculation in BACTool uses a perfect MPC model and perfectly known disturbances. The optimal sequence of control inputs is therefore determined only once every T_OL = 48 h (2 d) using a prediction horizon of T_H = 144 h (6 d). The subscript "OL" stands for "open loop" and this refers to the fact that the control inputs during the 48 hours following an optimization are precisely the ones delivered by this optimization, i.e. during these 48 hours the control inputs are directly applied to the plant model without any feedback to the controller. Further information can be found in [7].
• RBC-A: A typical, broadly applied, non-predictive control strategy. Inputs for control are current measurements of room temperature, outside air temperature, external heat gains, and the occupancy state. Three blind transmission values are considered: fully open, fully closed and shading transmission. Blinds are repositioned depending on threshold crossings for solar gains. For simulation this behavior is approximated by setting the blinds for a given time step based on the time step's average solar gains. This control strategy correponds to the strategy "RBC-1" reported in [3].
• RBC-B: A novel, non-predictive control strategy. The controller inputs are the same as for RBC-A; in addition are used historical heat and cold demand signals, and historical room temperature data. Blind transmission varies continuously between a minimum (blinds fully closed) and maximum (blinds fully opened) value. Blinds are repositioned once per hour (once per control step), based on the historical signals and data. See control strategy "RBC-4" in [3].

Note, the set of subsystems actually controlled by a given control algorithm depends on the currently chosen HVAC System variant.

Sites

Internal Gains

The occupancy density of the building (a number ranging from 0-100%) was used as the key quantity to determine the internal heat gains from persons and equipment, plus CO₂ production. Considered were two internal gains levels based on the Swiss standard SIA 2024 [11].

Diurnal and weekly variations in internal gains were obtained from the Swiss standard SIA 2024 [11] for cellular offices. During weekends no persons were assumed to be present and the person gains were set to zero. The equipment gains were set to the weekdays' night-time value.

Thermal Comfort

The used minimum and maximum room temperature set points for heating and cooling were similar to the definitions in SIA 382/1 [12].

The actual range at a given point in time was determined as a function of the exponentially weighted running mean of the past measured outside air temperature values. The running mean was calculated in a similar manner as described in EN 15251 [13]. The comfort settings were applied 24 hours a day and 7 days a week.

References

1. Gyalistras, D. & Gwerder, M. (Eds.) (2010). Use of weather and occupancy forecasts for optimal building climate control (OptiControl): Two years progress report. Terrestrial Systems Ecology ETH Zurich, Switzerland and Building Technologies Division, Siemens Switzerland Ltd., Zug, Switzerland, 158 pp, Appendices. ISBN 978-3-909386-37-6.
2. Lehmann, B., Wirth, K., Dorer, V., Frank, Th. & Gwerder, M. (2010a). Control problem and experimental set-up. In [1], Chapter 2, pp 15–28.
3. Gwerder, M., Tödtli, J. & Gyalistras, D. (2010). Rule-based control strategies. In [1], Chapter 3, pp 29–42.
4. Oldewurtel, F., Jones, C.N., Parisio, A. & Morari, M. (2010). Model predictive control strategies. In [1], Chapter 4, pp 43–58.
5. Lehmann, B., Wirth, K., Carl, S., Dorer, V., Frank, Th. & Gwerder, M. (2010b). Modeling of buildings and building systems. In [1], Chapter 5, pp 59–66.
6. Stauch, V., Schubiger, F. & Steiner, P. (2010). Local weather forecasts and observations. In [1], Chapter 6, pp 67–78.
7. Gyalistras, D., Lehmann, B., Wirth, K., Gwerder, M., Oldewurtel, F. & Stauch, V. (2010). Performance bounds and potential assessment. In [1], Chapter 7, pp 79–106.
8. Gyalistras, D., Wirth, K. & Lehmann, B. (2010). Analysis of savings potentials and peak electricity demand. In [1], Chapter 8, pp 107–134.
9. Gyalistras, D., Gwerder, M., Oldewurtel, F., Jones, C.N., Morari, M., Lehmann, B., Wirth, K., & Stauch, V. (2010). Analysis of energy savings potentials for Integrated Room Automation. Paper presented at the 10th REHVA World Congress Clima 2010, 9-12 May 2010, Antalya, Turkey, 8pp.
10. Gwerder, M., Gyalistras, D., Oldewurtel, F., Lehmann, B., Wirth, K., Stauch, V. & Tödtli, J. (2010). Potential assessment of rule-based control for Integrated Room Automation. Paper presented at the 10th REHVA World Congress Clima 2010, 9-12 May 2010, Antalya, Turkey, 8pp.
11. SIA 2024 (2006). Standard-Nutzungsbedingungen für die Energie- und Gebäudetechnik, Raumnutzungen "Einzel-, Gruppenbüro" und "Grossraumbüro", pp. 34-37.
12. SIA 382/1 (2007). Lüftungs- und Klimaanlagen – Allgemeine Grundlagen und Anforderungen.
13. Standard EN 15251 (2007). Eingangsparameter für das Raumklima zur Auslegung und Bewertung der Energieeffizienz von Gebäuden – Raumluftqualität, Temperatur, Licht und Akustik.