Multimodal Spatio-Temporal Traffic Modeling with TF-STGCN and Bi-Level Optimization for Tide-Influenced Coastal Tourism Zones

Addressing the severe imbalance between overloaded terrestrial traffic and underutilized maritime areas in Tide-Influenced Coastal Tourism Zones, this study proposes an intelligent transportation optimization framework based on a Dynamic Digital Twin. The framework successfully constructs a closed-loop system encompassing “perception-prediction-optimization-scheduling”, effectively overcoming core limitations in traditional management such as data fragmentation, insufficient consideration of dynamic factors, and imbalanced resource allocation. By integrating 12 heterogeneous data categories, the research introduces an innovative TF-STGCN model that significantly reduces the Mean Absolute Percentage Error (MAPE) for holiday-peak passenger flow forecasting, from 1 7. 3% to 9.8%. Furthermore, by constructing a Multi-Objective Optimization model, a notable 28.6% reduction in shuttle delay was achieved. Simultaneously, applying a Bi-Level Optimization scheduling model improved overall transfer efficiency by 37.5%. Following system integration, real-time scheduling response time was shortened to under 3 minutes, the terrestrial overload coefficient decreased to 1.9, and the share of maritime traffic increased to 7.9%. This establishes a quantifiable and replicable new paradigm for coastal eco-tourism governance.