With increasing urban car ownership, parking demand has surged, yet limited space often makes nearby spots unavailable. Although real-time, cloud-based applications and historical-data-based predictions exist, they frequently overlook travel distance and time, causing mismatches in availability. Some studies employ heuristic or reinforcement learning for route guidance but neglect factors like agent-to-parking distance, probability variations, and geographic topology. This study introduces a utility-based model integrating these elements, using a Spatio-Temporal Graph Neural Network (ST-GNN) to capture road-parking correlations. Compared with Agent57 and DQN, the model, simulated in SUMO, adapts to traffic and demand fluctuations, improving parking success and efficiency.
Jang et al. (Mon,) studied this question.