A deep neural network-based green space design optimization framework for smart cities.

Area of Science:

• Urban Planning and Ecology
• Artificial Intelligence in Environmental Science

Background:

• Rapid urbanization in China diminishes green spaces, impacting environmental sustainability and urban livability.
• Traditional green space design methods lack scalability, responsiveness, and data integration for efficient, context-aware planning.

Purpose of the Study:

• To propose a Deep Neural Network-based Green Space Design Optimization Framework (DNN-GSOF) for optimizing urban green space layouts in smart cities.
• To enhance the ecological efficiency, regulation compliance, and overall livability of urban environments through AI-driven design.

Main Methods:

• Utilized a modified U-Net Convolutional Neural Network (CNN) to process satellite imagery, environmental sensor data, and infrastructure maps for identifying optimal greening zones.
• Employed a constrained deep convolutional generative adversarial network to generate diverse, regulation-compliant urban green space layout proposals.
• Implemented a hybrid DenseNet-based multilayer perceptron model for quantitative performance evaluation across ecological, social, and economic dimensions.

Main Results:

• The DNN-GSOF model achieved 13.2% better layout overlap accuracy and a 17.2% higher compliance rate compared to other models.
• Demonstrated a 44.8% better FID score, indicating improved generation quality and diversity of green space layouts.
• Accelerated inference by 29.2% and reduced mean absolute error for zoning compliance (40.95%) and green space ratio (43.14%).

Conclusions:

• The DNN-GSOF framework efficiently generates accurate and regulation-compliant urban green space layouts.
• Offers practical applications for urban planners and policymakers to improve ecological sustainability and livability in rapidly developing cities.