A simulation-based approach for optimized control of ventilation, lighting, and shading systems Matthias Schuss 1, Claus Pröglhöf 1, Kristina Orehounig 1, Mario Müller 2, Heinz Wascher 2, and Ardeshir Mahdavi 1 1 Department of Building Physics and Building Ecology, Vienna University of Technology, Austria 2 Hans Höllwart - Forschungszentrum für integrales Bauwesen AG, Stallhofen, Austria Abstract: This paper presents the implementation of a simulation-based predictive control approach in realistic test space. The method allows for the control of multiple devices in a holistic and predictive way to optimize the building's energy and environmental performance. Toward this end, a genetic algorithm is used to generate control options and evaluate these by using predicted trends produced with numeric simulations. Keywords: building simulation, predictive control, natural ventilation, passive cooling 1 Introduction Increased user needs as well as relatively low prices of energy and hardware result in heavily equipped buildings with dramatically increased energy consumption especially for space cooling even in moderate climate zones like Austria. Nevertheless, there have been also changes concerning ecological awareness and economical constraints that have encouraged new developments toward energy efficient building systems and controls. An integrative control approach that includes all relevant environmental systems and embodies a predictive capability would offer the potential to reduce buildings' energy demand. Moreover, such an integrated control strategy would facilitate the application of alternative indoor climate control options such as passive cooling and daylight utilization. Previous publications illustrated the potential to use natural ventilation and predictive building controls in existing buildings [1-4]. The main focus of the present contribution is the development of a novel predictive simulation-assisted control approach with the capability to further encourage the usage of these alternatives and optimize the overall building operation. 2 Method A real office building (in Styria, Austria) was selected as the framework for the implementation of the above mentioned new control approach. For the purposes of the implementation, two rooms in this office building were selected (see Figure 1 and 3) The building structure with a concrete skeleton (ceilings and staircases) and the lightweight internal and external walls are typical for new office buildings in Austria. This circumstance, combined with the glass and aluminum façade, results in a reduced useable thermal storage mass and thus aggravates the impact of solar gains. Seite 1 von 8
7. Internationale Energiewirtschaftstagung an der TU Wien IEWT 2011 Figure 1. Office Building with the two test rooms The two rooms are identical in terms of layout and are located in the first and second floor on the northwest corner of the building. To control the building systems, the offices were equipped with actuators for lighting, shading, and window operation. Indoor conditions in both rooms as well as the external climate (see Figure 4) were monitored with sensors for thermal and lighting parameters (see Figure 5). One room was controlled with an implementation of the proposed predictive simulation-based control method. The second room, which is identical in shape and dimensions, was used as a reference. Figure 2. Test room internal view Figure 3. Layout of the test room Figure 4. Weather station Figure 5. Sensors for internal conditions Seite 2 von 8
2.1 Predictive simulation based control approach This control approach uses the actual sensor values together with web-based weather forecast data [5] for parametric simulations in the thermal and lighting domains. The objective is to arrive at an optimum control decision based on a set of performance functions and indicators. The control method generates and evaluates alternative operation possibilities based on genetic algorithm and the multi-domain simulation results. Figure 6 illustrates the principle of this approach. This control procedure is executed on a regular basis (typically once every hour). Alternative states! Multi-domain simulation engine Predicted performance for the interval [ti, ti+n] Control systems state at t i+1 Control systems state at t i Optimum control systems state at t i+1 time Boundary conditions at t i Forecasting module Predicted boundary conditions for the interval [ti, ti+n] Preformance Indicators Figure 6. Illustration of the multi-domain simulation-assisted control method The implementation of the general control procedures was done in the matlab[6] environment and uses HAMbase [7] and Radiance [8] as simulation tools. Services for data monitoring, communication and the weather forecast were programmed in C and run independently. Monitoring data (internal and external sensor) together with the web-based weather forecasts data were stored in a SQLite database. 2.2 Performance evaluation based on indicators The well-balanced operation of all relevant devices is the main focus of the predictive simulation-assisted control method. Therefore, performance functions and indicators were defined as follows and used to evaluate the multi-domain simulation results. The overall performance indicator i (Equation 1) is the weighted sum of all indicators i x. The value of each indicator and the sum of the weighting factors w x is in the range of 0 to 1. Hence i must be in the same range. The ranking of the alternative control possibilities is done by maximum to minimum sorting. Exactly two indicators were used for this first implementation. One was related to the air temperature and the second related to mean storage temperature. These two indicators were equally weighted. i = " i x! w!air = i!air! w!air + i!storage! w!storage = 1 2!i! air + 1 2!i! storage (1) To derive the single indicator values, a negative exponential estimation based on related interval deviations was used (Equation 2 and 3) Seite 3 von 8
i!air =1! e!c.d! air (2) i!storage =1! e!c.d! storage (3) The related time deviations are calculated as a sum of all discrete time deviations in the forecast interval in case of storage temperature indicator (Equation 4 and 6) or only the occupancy hours of the forecast interval in case of air temperatures (Equation 5 and 7) Figure 7 and 8 illustrate this. t i +n t i +n d!storage =! m!storage (t) (4) t=t i d!air =! m!air (t)" g(t) with g(t) presenting the occupancy (5) t=t i #% m!storage (t) =! storage(t)!! storagemax if! storage (t) >! storagemax $ &% 0 if! storage (t) "! storagemax #! air min!! air (t) if! air (t) <! air min % m!air (t) = $ 0 if! air min "! air (t) "! air max % &! air (t)! p! air max if! air (t) >! air max (6) (7)! air Prediction interval! storage Prediction interval! air max! storage max! air min max Occupancy t t t i t i +n Figure 7. Deviation calculation for the air temperature performance indicator. t i t i +n Figure 8. Deviation calculation for the storage mass temperature performance indicator. 2.3 Alternative control schedule generation The predictive control method needs a set of alternative operation states in terms of the relevant device control schedules. These schedules have to be produced to run the multidomain simulations. Using all possible combinations over the whole forecast interval would end up in an explosion of possibilities. A genetic algorithm seems more appropriate to handle this challenge. A number of default operation schedules are used together with randomized schedules as the initial setup. Needed state definitions and device attributes are stored in a predefined data structure to generate the schedules automatically. Based on the first generation simulation the best-ranked schedules were selected to generate new child Seite 4 von 8
schedules in a random multipoint crossover reproduction process (Figure 9). For this purpose, the high-ranked schedules are crossed with themselves as well as with additional randomly selected schedules dealing as parent elements. The selection of the fittest alternative is done by the use of performance indicators (discussed before). Figure 10 illustrates this genetic approach. Individuals: Generation: 1 1 2 3 4 n Random bit Pattern Parent schedule 1 Parent schedule 2 child schedule Multi domain Simulation Best fitness ranking Creation of next generation m=2 + = Individuals: Generation: m 1 2 3 4 n Figure 9. Illustration of the genetic multipoint crossover reproduction Multi domain Simulation Best fitness ranking Creation of next generation m=m+1 Best performing schedule Figure 10. Illustration of the genetic generation of the desired operation schedules 3 Results The proposed control method was implemented in a realistic test setup and operated in the course of a long-term test during two successive summer periods. Preliminary results are presented in the following. Figure 11 illustrates, as an example, the operation of the proposed control method for a day in August 2010. The predicted room temperature trends for all genetically produced scenarios are plotted. On the left half of the Figure, the measured room air temperature (light grey) is shown together with the outside temperature (red). On the right side, the predicted temperature for all scenarios is plotted in light grey. The best performing one is marked in black. The future device schedule is presented for the best performing scenario. This operational cycle is repeated on a regular basis (typically every hour). The derived optimum states of the devices are enacted accordingly. To address the performance of the control strategy in terms of thermal comfort, collected data were analyzed. Figures 12 shows, Seite 5 von 8
as an example, the thermal conditions in both office spaces in terms of psychometric charts for the month of August (red dots represent mean hourly values, green polygons the applicable thermal comfort zone according to the adaptive thermal comfort theory). This Figure includes data for the working hours only (8:00 to 17:00). These results point to an improvement of the thermal comfort conditions in the test rooms, which, in comparison to the reference room, is around 35% less outside thermal comfort zone. This circumstance can be also seen in the related PMV and PDD results. The cumulated PPD percentage of working hours in August 2010 is presented in Figure 13. A PMV boxplot for this period is shown in Figure 14. Figure 11. Multiple predictions and the actual course of indoor temperature in the test room for a day in August 2010 20 R2OG 08 2010 20 R1OG 08 2010 18 out of range: 95.39 %! air max : 31.41 C! air min : 19.24 C 18 out of range: 60.83 %! air max : 30.97 C! air min : 16.94 C 16 16 14 100% 14 100% Specific Humidity [g/kg] 12 10 8 6 90% 80% 70% 60% 50% 40% Specific Humidity [g/kg] 12 10 8 6 90% 80% 70% 60% 50% 40% 4 30% 4 30% 2 20% 10% 2 20% 10% 0 15 20 25 30 35 Dry Bulb Temperature [ºC] 0 15 20 25 30 35 Dry Bulb Temperature [ºC] Figure 12. Test rooms' temperature and Humidity of working hours with the related comfort zone in August 2010 (Left: reference room; Right: test room) Seite 6 von 8
100 90 PPD August 2010 R1OG R2OG percentage of working hours with lower PPD 80 70 60 50 40 30 20 10 0 10 20 30 40 50 60 70 80 90 100 PPD Figure 13. PPD values for the working hours in August 2010 PMV August 2010 2 1.5 1 0.5 0 PMV 0.5 1 1.5 2 2.5 3 R1OG R2OG Figure 14. PPD values for working hours in August 2010 4 Conclusion The results of the long-term test showed the potential of the proposed predictive control. However, there is potential for improvements. A first analysis of the results indicated the importance of weather forecast quality. The quality of temperature predictions was fairly good. But the prediction of solar radiation needs improvement. Overall, the implementation demonstrated that the proposed approach can be realized in existing buildings and be integrated with the legacy automation systems. 5 Acknowledgments The research presented in this paper is supported in part by a fund from FFG "Naturally Cool" (Project-Nr: 817575) and supported by the K-Project "Multifunctional Plug & Play Façade" (Project-Nr: 815075) and the Hans Höllwart - Forschungszentrum für integrales Bauwesen AG (CEO, Mario J.Müller). Seite 7 von 8
References [1] Mahdavi, A., Pröglhöf, C. 2004. Natural ventilation in buildings Toward an integrated control approach, Proceedings of the 35th Congress on Heating, Refrigerating and Air-Conditioning, Belgrade, Serbia, pp. 93-102. [2] Mahdavi, A., Pröglhöf, C. 2005. A model-based method for the integration of natural ventilation in indoor climate systems operation, Proceedings of the 9th International IBPSA Conference, Building Simulation 2005, Montreal, Canada, pp. 685-692. [3] Mahdavi, A. 2008. Predictive simulation-based lighting and shading systems control in buildings, Building Simulation, an International Journal, Springer, Volume 1, Number 1, ISSN 1996-3599, pp. 25 35 [4] Schuss, M., Pröglhöf, C., Orehounig, K., Mahdavi, A. 2010. A case study of model-based ventilation and shading controls in buildings, Proceedings of the 10th Rehva World Congress, Sahin N. (ed.), Antalya, Turkey. [5] Weather.com, 2010. National and Local Weather Forecast, Hurricane, Radar and Report, The Weather Channel Interactive, Inc., http://www.weather.com/. [6] MATLAB, 2010. MATLAB Release 2010a, The MathWorks, Inc., http://www.mathworks.com. [7] van Schijndel, A.W.M. 2007 Integrated heat air and moisture modeling and simulation, PhD thesis, Eindhoven University of Technology, http://alexandria.tue.nl/extra2/200612401.pdf or http://sts.bwk.tue.nl/hamlab accessed June 2010. [8] Radiance, 2010. Radiance Synthetic imaging system Version 4, University of California, http://radsite.lbl.gov/radiance/. Seite 8 von 8