A methodology for the optimization of building energy, thermal, and visual performance

Buildings are under the scope of environmentalists since they are the biggest energy consumers and polluters. Building performance could be greatly improved thanks to optimization. Yet, optimizing for different aspects of a building's performance is a conflicting process and building designers have to rely on their experience to make decisions. The present work proposes a method to assess the optimal configuration for a building in terms of energy and indoor environment performances. The method relies on the good performance of Genetic Algorithms (GA) for complex optimization problems. However, GAs require extensive computations. Artificial Neural Networks (ANN) were used to alleviate the computational burden. The main concern has been to make this method as universal and easy to use as possible, resorting to widely used tools only. The method was first successfully tested on a small-scale, four-room section of an office building and on a full-scale school. In both cases, the ANN model performed well with prediction errors in the order of 5%. Finding a better design for the school building was rather difficult since the building performed well already, but thermal comfort could be improved without increasing the energy demand or decreasing visual comfort. The limits of the method were tested by playing with the number of inputs and outputs. The ANN performed well though its performance decreased as the number of design parameters increased. The limits of the method were established regarding the performance of the ANN and the number of cases required to train and validate the AN