In the realm of urban transportation, metro systems serve as crucial and sustainable means of public transit. However, their substantial energy consumption poses a challenge to the goal of sustainability. Disturbances such as delays and passenger flow changes can further exacerbate this issue by negatively affecting energy efficiency in metro systems. To tackle this problem, we propose a policy-based reinforcement learning approach that reschedules the metro timetable and optimizes energy efficiency in metro systems under disturbances by adjusting the dwell time and cruise speed of trains. Our experiments conducted in a simulation environment demonstrate the superiority of our method over baseline methods, achieving a traction energy consumption reduction of up to 10.9% and an increase in regenerative braking energy utilization of up to 47.9%. This study provides an effective solution to the energy-saving problem of urban rail transit.