High-speed train operation scheduling requires high accuracy and timeliness to ensure safe and efficient railway operations. This study proposes a multi-agent deep reinforcement learning (MADRL) framework for intelligent train operation adjustment, formulating the adjustment process as a continuous and cooperative decision-making problem involving both station dwell time and inter-station running time. Two agents, based on the Deep Deterministic Policy Gradient (DDPG) algorithm, are designed: one controls dwell time at stations, and the other adjusts inter-station running time. The agents interact asynchronously within a shared simulation environment to coordinate adjustment strategies, minimize total delay, and maintain timetable stability. The framework is validated using real-world operation data from the Wuhan–Guangzhou section of the Beijing–Guangzhou High-Speed Railway. Simulation results show that the MADRL-based approach outperforms traditional optimization methods, including the Genetic Algorithm (GA), First-Come-First-Served (FCFS), and First-Planned-First-Served (FPFS). The proposed method provides a promising and intelligent solution for real-time high-speed railway operation adjustment.