Multi-objective Optimisation of the Configuration and Control of a Double-skin Facade

We present a new approach to the optimisation of Double-Skin Facades (DSFs). Parameters defined possible geometries, shading devices, openings and ventilation paths, as well as control schedules for their operation. A genetic algorithm was used to discover the best configuration and control strategies for a given scenario from scratch, rather than using a particular configuration type. The algorithm performed a thermal and air-flow simulation of each proposed solution using EnergyPlus. The optimisation process has been illustrated with a case study. In addition, the process has been applied to a range of use types and the results examined graphically to derive innovative design guidelines (a process known as “innovization”). INTRODUCTION Double-Skin Facades Double-Skin Facades (DSFs) may be beneficial in reducing the energy used in buildings for heating and / or cooling. A DSF consists of two glazing layers with an air space between. The air space may be sealed or may be ventilated in a range of configurations. For example, if air is drawn into the room through the facade, it will be pre-heated by solar gain to the air space; this should reduce the heating load. Alternatively hot air from the room may be exhausted via the air space, where it will receive the solar gain that would otherwise have been directly transmitted to the room; this should reduce the cooling load. DSFs are an expensive addition to a building, and must be justified by improved performance. Because of the complex nature of their operation, careful simulation is required to assess their benefits. Performance is dependent upon the geometric configuration of the facade, the operational mode governing air flow, and the control system used to activate different modes. This presents a complex engineering challenge. This paper aims to use an optimisation algorithm to derive the best configuration and control for a DSF for a given scenario, to aid in the resolution of this design problem. Multi-objective optimisation Computational optimisation is a rapidly emerging discipline for aiding engineering design. Multiobjective optimisation is particularly useful as it involves the consideration of several objectives simultaneously, with no weightings or aggregations, allowing the robust resolution of complex trade-offs between conflicting objectives. This involves finding the non-dominatedor Pareto-front, a set of points in the objective space for which no point performs better in all objectives (see Figure 1). Figure 1: An example of a Pareto front for the minimisation of two objectives, one on each axis. Triangles are members of the Pareto front; dots are not. For each point in the front there is no other point which performs better, e.g. for the highlighted point there is no point within the shaded area. Genetic algorithms are a common means of achieving multi-objective optimisation. The principle follows that of Darwinian evolution: a population of possible solutions is maintained, with the “fittest” allowed to progress to the subsequent generation. Fitness for multi-objective problems attempts to quantify distance from the non-dominated front. Solutions are altered by crossover (splicing variable values with another solution) and by mutation (changing variable values randomly). The algorithm used in this work is the Non-dominated Sorting Genetic Algorithm (NSGAII) of Deb et al. (2002). This algorithm selects solutions for subsequent generations based firstly on non-domination rank1, and secondly by crowd1Rank 1 solutions are the non-dominated front. These are removed and domination is recalculated to form a new front, which Proceedings of Building Simulation 2011: 12th Conference of International Building Performance Simulation Association, Sydney, 14-16 November.