Time-series risk forecasting of airport conflict hotspot with SA-GRU

Airport surface operations increasingly confront collision risks from intricate layouts, vehicle-aircraft interactions, and dense mixed traffic flows. This study develops a predictive framework for conflict hotspot identification by integrating topological, network vulnerability, and traffic complexity metrics into a composite risk evaluation system. A hybrid method combining composite weighting and improved TOPSIS first identifies latent hotspots through node-level risk assessments. Temporal risk patterns are then extracted via principal component analysis of hotspot features, with future risk trajectories predicted using a GRU network enhanced by self-attention mechanisms. Validated through Shenzhen Bao'an International Airport simulations, the proposed SA-GRU model reduces RMSE by 9.14–11.55 % against benchmark models (HA/ARIMA/SVR/LSTM/GRU). Analysis reveals significant spatiotemporal variations in hotspot risks, where daily trends show similar risk fluctuation patterns across zones but differ substantially in intensity. High-risk areas dynamically shift across operational phases, emphasizing the necessity of time-sensitive predictions. The framework enables proactive identification of critical conflict zones through predictive risk monitoring, demonstrating practical potential for optimizing airport surface management. By translating multidimensional operational data into actionable safety insights, this methodology supports intelligent decision-making for collision prevention and resource allocation in complex aviation environments, while remaining adaptable to diverse airport configurations.