Intelligent layout and spatial planning of green urban buildings based on deep learning

The research proposes a novel DeepClusGAN, namely an intelligent combination of Deep Convolutional Generative Adversarial Network (DCGAN) and deep embedding cluster (DEC) to optimize urban building cluster layouts. DCGAN generates innovative and intelligent spatial layouts by learning patterns from existing urban residential building data sources and guarantees diversity and adherence to urban design principles. DEC analyzes the learned spatial features from the previous phase and performs clustering to identify functional zones and spatial relationships between metrics like proximity, compactness, and accessibility. The research maintains a spatial pattern and adjacency of building arrangements by optimizing objectives to meet urban planning goals, such as maximizing space usage, enhancing urban environmental growth, and promoting building architectural harmony. The research achieves an improved silhouette score, an adjusted rand index for validating the building cluster, and a 20.27% improvement in spatial diversity compared to existing models while maintaining an enhanced optimal building layout efficiency.