Article Preview
TopIntroduction
With the continuous acceleration of global urbanization, the high-density aggregation of urban populations and industrial activities has led to increasingly complex and diversified issues in resource allocation, traffic organization, cultural activity planning, and environmental governance (Alem & Kumar, 2022; Puchol-Salort et al., 2021). How to reasonably plan urban regional layouts to meet the dual needs of urban operational efficiency and social activity demands, while effectively reducing environmental pollution and resource consumption, has become a critical challenge that urgently needs to be addressed. Traditional urban planning heavily relies on empirical judgment or methods based on simple statistical models. For example, traffic flow evaluations are often calculated using averaged data from fixed time periods, making it difficult to adapt to dynamic peak-time variations and resulting in delayed planning decisions (Laurini, 2018; Liu & Ye, 2023; Sharma et al. 2023). At present, with rapid advancements in data acquisition and computational technologies, research paths that utilize deep learning and artificial intelligence (AI) methods to model complex spatiotemporal data to support urban planning decisions are attracting increasing attention (Dowlatshahi et al., 2021; Gokasar et al., 2023; Khalil et al., 2021; Wang et al., 2022).
Current studies have made initial progress in leveraging big data and AI for urban planning analysis, as represented in Figure 1. For instance, some studies attempted to use geographic information system data (Yu et al., 2021), remote sensing images (Bai et al., 2022), traffic flow data, and pollution indicators for correlation analysis (Orieno et al., 2024), aiming to uncover potential spatial patterns and temporal features to help formulate more targeted planning strategies. However, these methods often have several shortcomings. First, many studies remain focused on using statistical models or traditional machine learning algorithms, which have limited data feature extraction and modeling capacity, making it challenging to fully capture deep structures and spatiotemporal dynamics in high-dimensional heterogeneous data (Zheng et al., 2024). Second, some deep learning-based studies focus solely on single data source predictions, lacking robust capabilities for integrating and adaptively mining multimodal data in the spatiotemporal dimension (Wang et al., 2022). Third, most studies lack dynamic decision-making and feedback optimization mechanisms, rendering them unable to actively adjust regional planning strategies based on prediction results to achieve closed-loop optimization and continuous improvement (Juárez-Varón & Juárez-Varón, 2024; Villalonga et al., 2021). Therefore, this study argues that a new model is needed—one that integrates multisource heterogeneous data, possesses strong spatiotemporal feature extraction capabilities, and can dynamically optimize planning strategies within a reinforcement learning (RL) framework.
Figure 1. Evaluating urban planning analysis methods
This article proposes a semantic web-enhanced RL-based urban planning optimization model (SWRL-UPOM), to optimize urban planning through precise prediction of spatiotemporal dynamics and iterative strategy refinement.