Building Shape Optimization Using Neural Network and Genetic Algorithm Approach

摘要

This paper presents a connectionist approach using a neural network (NN) and genetic algorithm (GA) to optimize the selection of office building shape. The optimization takes into account both energy and construction costs. In the analysis, the footprint dimensions of the office building are optimized while its volume and height are assumed to be predefined. Various input parameters are considered in the optimization analysis, including climate, window-to-wall ratio, type of glazing, and wall or roof insulation. Results from a whole-building simulation tool are utilized to train and test the NN-GA-based optimization approach. The analysis indicates that the hybrid NN-GA approach offers a robust and efficient method of optimizing the shape selection for office buildings when Bayesian neural networks are utilized instead of conventional neural networks.