Machine Learning-Based Method for Detached Energy-Saving Residential Form Generation

Abstract

In recent years, machine learning has gradually been applied to building energy-saving designs to reduce the time consumption of the optimization screening stage. However, since most of the existing research scholars come from the fields of computers and engineering, the application of machine learning technology mostly involves complex programming as well as software in the field of engineering, which requires multiple software to be coupled to achieve. In view of the differences between disciplines and the high learning threshold, these theories are difficult to apply and promote in practical work in the field of architecture. In this regard, this paper focuses on the improvement of methods, based on the Grasshopper platform, proposes a detached energy-saving residential form generation design method and process, to explore the optimal energy-saving building form in a more concise and efficient way. Based on this new method, on the basis of verifying its feasibility through a residential building case, two machine learning algorithms, neural network (ANN) and support vector machine (SVM), are compared and studied, and the applicability of these two algorithms in different building performance indicators is further discussed. The results show that the ANN model has the highest accuracy and is more suitable for the prediction of building energy consumption; in view of the simple and fast operation of SVM, it is more suitable for comfort prediction with relatively low accuracy requirements. By combining the above two machine learning methods, work efficiency can be improved while satisfying the prediction of relevant performance indicators. This method can help architects quickly search for the best building energy-saving form design scheme in the scheme design stage and provide data support and information feedback for architects in design conception and deepening.