Safe Urban Traffic Control via Uncertainty-Aware Conformal Prediction and World-Model Reinforcement Learning

Urban traffic management demands systems that simultaneously predict future conditions, detect anomalies, and take safe corrective actions while providing reliability guarantees. We present STREAM-RL, a unified framework with three novel algorithmic contributions: PU-GAT+, an Uncertainty-Guided Adaptive Conformal Forecaster; CRFN-BY, a Conformal Residual Flow Network; and LyCon-WRL+, an Uncertainty-Guided Safe World-Model RL agent with Lyapunov stability certificates. This is the first framework to propagate calibrated uncertainty from forecasting through anomaly detection to safe policy learning with end-to-end theoretical guarantees. Experiments on real-world traffic trajectory data demonstrate high coverage efficiency, controlled False Discovery Rate, and improved safety rate compared to standard PPO.