A Model Predictive Control Framework for Urban Rail Transit Signal Optimization with Time-Series Passenger Flow Forecasting

Abstract

Urban rail transit systems demand ever-improving responsiveness to fluctuating passenger flows. In this study, we use a 10-day dataset of City A subway passenger counts, headways, and flow intensities (N=10 days, Nov 29–Dec 8 2024). We first apply two time-series forecasting techniques—ARIMA (2,0,0) and single-pass exponential smoothing (α=0.95)—to predict short-term passenger demand. Based on these predictions, we formulate a constrained Model Predictive Control (MPC) problem that simultaneously minimizes average train delay, enforces block‐section safety headways, and adapts signal timings across a prototype corridor. Simulations implemented in Python—with vehicle dynamics, signal phase duration limits, and safety constraints explicitly modeled—show that the MPC strategy reduces average delay from 25 s to 10 s (60 % reduction) compared to fixed-timing baselines. We quantify trade-offs among prediction horizon, computational load (solved via rolling-horizon quadratic programming), and control performance, and clearly demonstrate our contributions in integrating demand forecasting into a real-time MPC framework for rail signal priority.