EXPLORING THE USE OF VARIABLE MAPPING FOR OPTIMISING URBAN MORPHOLOGIES Christoph Waibel 1,2 , Ralph Evins 1,2,* 1 Urban Energy Systems Laboratory, Empa, Swiss Federal Laboratories for Materials Science and Technology, Überlandstrasse 129, 8600 Dübendorf, Switzerland. 2 Chair of Building Physics, Swiss Federal Institute of Technology, ETH Zürich, Stefano-Franscini-Platz 5, 8093 Zürich, Switzerland. * ralph.evins@empa.ch ABSTRACT A geometrical optimisation process is applied to an urban district of 12 buildings, using detailed energy simulations to evaluate their performance. The objectives were total floor area, lighting energy demand, and space conditioning energy demand (heating + cooling). The geometric modelling, energy simulation and optimisation framework employed is based on the Rhinoceros / Grasshopper platform and associated plugins (ArchSim as a link to EnergyPlus; Octopus for multi-objective optimisation). A novel variable representation is investigated in order to phrase the problem in a way which is both flexible and computationally achievable. This was compared to a standard approach in which all building parameters are optimised, and to cases in which a single building is optimised. Results show that the multi-building optimisation approaches are outperformed by both single-building optimisation and manual adjustment. INTRODUCTION Overview It is now practical to apply building energy simulation tools at larger scales encompassing many buildings, allowing their use to inform urban-level planning and design. This is due to improved efficiency of models and more readily-available computational resources. The urban context in which buildings exist defines many aspects of their behaviour, but energy-related issues are rarely considered in detail when planning new districts. The form of buildings is also a challenging area for low-energy design, and is closely coupled with the urban-scale issues as surrounding buildings can have a significant impact on the performance of a given building form. Previous work Many applications of optimisation algorithms to building energy related problems have been reviewed by Evins (2013). Detailed optimisations of building energy use tend to focus on single buildings, and if geometry is included it is via fixed sets of parameters. For example McKinstray et al. (2015) optimised the span, wall height and ridge location as well as glazing ratios and thicknesses. They focussed on the use of neural networks as a means of reducing the computational time of the optimisation process. In the architectural domain, there are many approaches to form optimisation, which can be broadly split into direct representations, where the variables (the genotype) are explicitly converted to a building form (the phenotype), and indirect approaches, where the mapping of genotype to phenotype can vary as part of the optimisation process. The latter gives much greater freedom in exploring the design space, but can lead to problems due to the unconstrained nature of the space and the difficulty of exploring it efficiently. An example of an indirect representation applied to building form optimisation is given in (Evins et al., 2014). This work employs a direct representation in that it remains constant during the optimisation, but attempts to find powerful combinations of representations that allow a broad range of forms to be explored efficiently. Kämpf and Robinson (2009) were the first to apply evolutionary optimisation to multiple buildings when they coupled the urban energy simulation tool CitySim with a hybrid evolutionary algorithm CMA- ES/HDE to vary insulation values and glazing ratios. They extended this to geometric aspects, using Radiance to optimise the solar gains to building and urban forms. (Kämpf et al., 2010) optimised the roof heights and angles of 25 buildings for three urban layouts. (Kämpf and Robinson, 2010) addressed three cases: the heights of 25 roofs, the heights of 31 triangulation vertices defining the roofs of several buildings, and 25 control points governing a 2D Fourier surface. Vermeulen et al. (2013) used CitySim and CMA-ES/HDE to optimise the heights of 16 buildings that form a city block as well as the glazing ratios for the four orientations (which was constant for all buildings). The objective was heating demand, and constraints governed the total built volume. Vermeulen et al. (2015) examined the maximisation of solar gain by varying the height of 25 buildings and subsequently the height, rotation, location and scaling of 9 buildings. Yi and Malkawi (2009) applied a method based on hierarchical geometrical relationships (‘agent-based geometry control’) to building form optimisation. Proceedings of BS2015: 14th Conference of International Building Performance Simulation Association, Hyderabad, India, Dec. 7-9, 2015. - 1837 -