Semantic Web-Enhanced Reinforcement Learning Model for Urban Planning Optimization

Abstract

As urbanization accelerates, urban planning is essential for enhancing quality of life and sustainability. Current methods struggle with complex spatiotemporal data, limiting real-time feature capture and strategy adjustments. To address this, we propose the Semantic Web-Enhanced Reinforcement Learning-based Urban Planning Optimization Model (SWRL-UPOM). Integrating Semantic Web technologies with Spatio-Temporal Adaptive Multimodal Graph Convolutional Network (STAMFGCN) and Spatio-Temporal Gated Hierarchical Attention LSTM (STGHALSTM), SWRL-UPOM uses reinforcement learning to optimize strategies dynamically. STAMFGCN extracts complex inter-regional relationships from multimodal data, while STGHALSTM models and predicts spatiotemporal pollution evolution. Leveraging Semantic Web for structured data and reasoning, the RL framework iteratively updates strategies based on predicted pollution trends. Experiments show SWRL-UPOM outperforms traditional methods in pollution prediction, strategy optimization, and adaptability to dynamic changes.