This paper presents an adaptive optimization framework for dynamic rail transit network operations with use of a rollout surrogate-approximate dynamic programming method. The optimization algorithm derives coordinated decisions of service schedules and train unit deployment with respect to prevailing passenger demand. Considering the computational effectiveness needed for real-time applications, a state-dependent surrogate function is incorporated to approximate the costs associated with operational decisions over future stages. The surrogate approximation is updated iteratively via a temporal difference learning process with feeding of observations made from the transit network. The proposed framework is implemented and tested on a real-world scenario in Hong Kong Light Rail Transit (LRT) network. The results reveal that the proposed framework is able to reduce significantly the total passengers’ waiting times over existing plans with reasonable computational time via use of the surrogate approximation. This suggests the potential of the proposed optimizer for real time applications in large-scale rail transit networks.