A spatiotemporal traffic flow prediction-driven decision-making framework for intelligent emergency lane opening

Dynamic emergency lane opening is an active traffic-management strategy in intelligent transportation systems that can temporarily increase freeway capacity during non-recurring congestion. This study develops a prediction-driven framework to translate sparse detector observations into operational emergency lane opening decisions. A dynamic extended conservation higher-order model reconstructs continuous traffic states between four monitoring points. The reconstructed fields and detector data feed a hybrid short-term predictor that combines gated recurrent units, time-frequency image representations, transformer encoders, and extreme gradient boosting. A rolling-horizon module identifies forecast congestion zones and shock-wave propagation and determines the opening location, activation time, duration, and recovery-based closure subject to roadway-safety constraints. Using surveillance data from a 5-km Changsheng Highway corridor, we evaluated the strategy through simulation against an optimized speed-threshold strategy. Relative to the closed initial state, mean speed increased by 28.10%, while saturation and delay rate decreased by 28.75% and 30.59%, respectively. Relative to the threshold strategy, the corresponding changes were +14.6%, −14.1%, and −14.8%. In simulated on-ramp and off-ramp scenarios, mean speed increased by 34.89% and 27.36% relative to the initial state of each scenario. The results indicate that combining physics-based state reconstruction with data-driven forecasting can support anticipatory lane opening decisions.