Analyzing the relative effectiveness of artificial intelligence algorithms in the context of adaptive facades

Adaptive facades are an effective tool for increasing energy efficiency in buildings, as well as preventing overheating inside the building and ensuring user comfort. AI algorithms are often used for performance-based optimization of adaptive facades, but there is a need for a multi-objective optimization method that integrates various meta-models and optimization algorithms. The aim of this study is to compare the performance of different AI optimization algorithms for evaluating the energy efficiency and daylight performance of adaptive facade systems. For this purpose, a multi-stage and systematic method has been developed. The parametric modeling and simulation of the BESTEST building geometry were performed in the Rhino/Grasshopper environment to provide data for the optimization process. Data was generated through simulations, and effective parameters for the optimization process were determined using different feature selection algorithms. With the data provided, the performance of single and hybrid meta-models belonging to machine learning methods was compared. The ANN-CatBoost meta-model, which had the lowest error rate and highest reliability, was used as the objective function for the optimization algorithms. When the performance of the algorithms was examined in terms of computation time, SA was the fastest algorithm at 3.68 minutes, while EP was the slowest at 277 minutes. The SA algorithm can be used in the early design stages of adaptive facades, but it showed the worst performance in terms of solution quality. The SMA algorithm was the most effective in terms of energy performance (6397.24 kWh), while the DE algorithm was the most effective in terms of useful daylighting performance (93.82%). In the overall performance comparison, DE, SMA, MFO, and EP were the algorithms with the highest performance. These algorithms achieved very low fitness values in all iterations, producing the most effective solutions to the multi-objective problem. Furthermore, it is observed that the performance of the algorithms changes significantly when dynamic weighting is applied to the optimization objectives. In future studies on the optimization of adaptive facades, comparing the performance of the optimization algorithms used in this study with different design scenarios will contribute to research on this topic.