Energy conscious automated design of building faades using genetic algorithms

Aris Tsangrassoulis1, Vassilis Geros2, Vassilis Bourdakis3 1 Department of Architecture, University of Thessaly, Greece 2 Group of Building Environmental Studies, University of Athens, Greece 3 Department of Planning and Regional Development, University of Thessaly, Greece [email protected]; [email protected]; [email protected]

Various European Directives have been issued concerning the energy efficiency of the buildings. The target is the achievement of a near optimum energy efficient environment while at the same time satisfying occupant needs throughout the year within an integrated holistic or whole building framework. A variety of different antagonistic parameters should be balanced such as window size and glazing transmittance or daylighting and shading and in most cases this requires an examination of various scenarios. Thus the design of building envelopes should address a careful balance between internal requirements and loads, the materials and properties of the faade and the external environment. Nowadays, the available tools for faade design-in terms of buildings energy efficiency- are inappropriate for interactive or creative use. In this paper we examine the development of a genetic algorithm which is capable to optimise the opening areas, glazing properties and shading configurations on the basis of minimum energy consumption- and then to design automatically simplistic alternative scenarios of the building faade. Keywords: Genetic algorithms; building energy efficiency Introduction
Todays a growing concern of energy usage in buildings has lead to a huge interest in alternative energy production and most of all, energy conservation. Thus, the design of facades in commercial buildings and the associated heat transfer though is quite crucial for the energy performance of the whole building. However techniques to reduce possible energy penalties can be quite costly, therefore it is extremely important to accurately calculate the energy impact of these techniques.
898 eCAADe 24 - session 20: generative design systems

In Europe, buildings represent almost 40% of its global energy consumption. The use of passive solar techniques can contribute to reduce significantly the total energy consumption. These techniques, aiming at increasing direct gains and daylight during the winter period, should make use of shading devices or window size to avoid overheating during the summer period while at the same time daylighting levels should be maintained. This can be quite tedious, since various and very often contradictory- parameters should be balanced. Although the use of building energy simulation

tools (e.g. ESP-r Clarke, 2001 TRNSYS, TAS, DOE-2, etc.) can be used to optimise the faade design in energy terms, such an approach can be quite expensive and time consuming as well. Thus estimation of glazing size-properties and shading can significantly improve if some kind of numerical optimisation is used. Such an approach has already been investigated and optimisation codes using the above mentioned simulation engines are available (GENOPT, Wetter, 2001). One of the main disadvantages is the complexity of the interface. Practical methods to estimate the ideal window size (Backer et al, 1988) or the design of shading device have been developed but information, these tools offer to the designer, is not very useful, and in extreme cases it is sometimes even inaccurate and misleading. The lack of simplified evaluation tool, capable on providing information on the suitability of a window size and its potential to save energy, is considered as one major reason for the reluctance of building professionals in incorporating some kind of energy optimisation in their design. It is the belief that significant benefits are available in introducing an optimisation scheme at the earliest design stages leading to improved design entities.

Method
Figure 1 Top view on floor

nances for a point (or on average) in the interior. This is difficult to be performed on hourly basis for the typical meteorological year. Thus a more simplified procedure is followed, using a monthly cumulative frequency distribution of the exterior horizontal illuminances. By multiplying with the average daylight factor (BRE Digest 309, 1986), interior cumulative frequency distribution can be estimated. Although the method is simple, the multiplication implies that the sky is considered as overcast all year long thus the result somehow represents a worst-case scenario. Of course sunlight is excluded. These tools can be used as a pre-design tool to assess the potential of the design, avoiding lengthy input process and time-consuming calculation. The building is a middle-size office building with office modules aligned on two faades, separated by a central corridor, with staircase/service spaces at both ends of the building. Dimensions and the data that were used are given in the next tables and figures. The windows are located on the south faade of the building. The optical properties of the elements inside the office room are presented in Table 1. Since the specification (in our case room dimensions and walls thermal and optical properties) have

The design problem concerns the estimation of window size, glazing thermal and optical properties and shading factor for a buildings typical office module. The tools that have been used for the estimation of its energy consumption was Summer Building a tool that uses Balance Point Temperature to estimate monthly heating and cooling loads (Santamouris et al, 1995). The simplest method to predict energy savings due to daylight is to assign a single annual figure for lighting energy use. This is accomplished by estimating a cumulative frequency distribution of illumi

Figure 2 Vertical cross section of the office room

 

The Transient Energy System Simulation Tool, http://www. trnsys.com/ Thermal Analysis Software, http://www.harvardthermal.com/. widely used and accepted building energy analysis program, http://www.doe2.com/

session 20: generative design systems - eCAADe 24

899

Elements of the room Ceiling Walls Floor Door

Reflectance 0.75 0.50 0.30 0.40

Transmittance -

been set, the energy consumption for heating, cooling lighting is evaluated. If this specification is not satisfied, the initial design is modified and re-evaluated. The whole procedure is repeated until the design fulfils the initial demands. This is more or less a trail & error method and despite its simplicity, processing time can be huge since all possible variations have to be examined. A better methodology is to use a genetic algorithm approach (Davis, 1991). Genetic algorithms are search algorithms based on the mechanics of natural selection and are very well suited to solving problems involving very large search spaces. Each possible solution to the design problem is encoded as a string of numbers which is called a chromosome. Each chromosome is consisted of a number of genes which represent the design parameters. In this paper the genetic algorithm (GA) that have been used is an implementation of the simple GA described by Goldberg (1989).

could in principle- included in the design parameters. However solar and visible transmittance are not independed variables. Thus it was decided that only solar transmittance should be used and then by using solar transmittance and U-value a glazing type is selected through a list of 20 glazing types and hence the visible transmittance . The fitness function which measures the quality of the performance of a solution and which the GA is trying to maximize is defined as follows: FF=1/ (a*CL+ b*HL+c*LL) Where CL, HL and LL are the annual cooling, heating and lighting load respectively (KWh/m2). These loads should be modified in order to take into account heating and cooling system efficiencies and the associated cost and this is performed through the use of the parameters a, b and c. Since these parameters can affect the fitness function, the relative difference in oil (for heating) and electricity (for cooling and lighting) prices can also affect the final solution. The choice of population size, probability of mutation and crossover is depended on the characteristics of the optimisation problem. For these kind of problems bibliography suggests population sizes to range from 5 to 100, crossover probability from 0.3 to 1 and the mutation probability from 0.001 to 0.05 (Wetter et al, 2003).

Table 1 Optical properties of the elements inside the office module

Implementation of the GA
Table 2 presents the design parameters, their lower and upper bound and their step size used in the optimisation. These variables are included in the design representation are those that give all the required information so that the performance of the specific faade design. Since daylighting is taken into account visible transmittance is also required and this
Lower bound Window Length Window height Glazing Solar Transmittance Glazing U-value Overhang Width 0.1 0.1 0.1 2 0 Upper bound 5m 3m 0.7 5 W/m2K 1.5 m Step size 0.07 0.04 0.01 0.02 0.04

Results
Genetic algorithms are stochastic processes and tend to be sensitive in the quality of the initial randomly produced-population. Several runs of the genetic algorithm were performed while the evolution of the best (maximum) individual (best fitness function) and the entire population (average fitness value) were recorder. Some results are presented below. Average population values for the parameters that affect the fitness function are presented in the figure 4. while Figure 5 shows the evolution of heating, cooling and lighting specific consumption (kWh/m2).

Table 2 Lower, upper bound and step size of the design parameters.

900

eCAADe 24 - session 20: generative design systems

Figure 3 Evolution of average and maximum value of the fitness function a) Generations 100, population 20, one-point crossover probability 0.1, mutation prob. 0.001 b) Generations 100, population 20, one-point crossover probability 0.3, mutation prob. 0.001 c) Generations 100, population 20, one-point crossover probability 0.7, mutation prob. 0.001

After the tuning of the GA the results where passed to RADIANCE for the visualization of the faade. This is presented in Figure 6.

Conclusions
The paper demonstrated how genetic algorithms combined with simplistic calculation tools for estimating energy consumption can applied to the

Figure 4 Evolution of the generation average values of the five design parameters

Figure 5 Heating, cooling and lighting annual consumption per generation.

Figure 6 Initial and final faade designs

session 20: generative design systems - eCAADe 24

901

initial phase design of a buildings faade by estimating: Window thermal and optical properties Window size Shading Trial and error design approaches become less useful as the design complexity increases thus a more generalized method is needed. Genetic algorithms as search engines can cover quickly and effectively the design space offering a family of design solutions that fulfils the design criteria.

References
Clarke J.: 2001, Energy Simulation in Buiding Design, Butterworth-Heinemann. Wetter M.: 2001, Genopt- A generic optimisation Program Proc. Of the 7-th IBPSA Conference, Vol. 1, Rio de Janeiro, Brazil, pp. 601-608. Backer N., Hoch D.: 1988, A design tool which combines the energy value of daylight and the thermal value of solar gain, Energy and Buildings for Temparate Climates, PLEA 88, Proc. Int. Conf., Santamouris M., Geros V., Klitsikas N., Argyriou A.: 1995, Summer Building, a computer tool for passive cooling application, Proc. Of International Symposium on Passive Cooling of buildings, Athens. BRE Digest 309 : 1986, Estimating Daylight in buildings: Part 1 and 2, Building Research Establishment Digest, Watford, England. Davis L. (ed): 1991, Handbook of genetic algorithms, Van Nostrand Reinhold, New York. Goldberg D.: 1989, Genetic Algorithm in Search, Optimisation and Machine learning, Adisson-Wesley. Wetter M., Wright J.: 2003, Comparison of a generalized pattern search and a genetic algorithm optimisation method, Proc. 8-th International IBPSA Conference, Eindhoven, Netherlands. Ward Larson G. & Shakespeare R.: 1998, Rendering with Radiance. The Art and Science of Lighting Visualization. Morgan Kaufmann Publishers. San Francisco.

902

eCAADe 24 - session 20: generative design systems