To optimize emergency response and address the inefficiencies observed in traditional dispatch models, a study was conducted to develop a scheduling method that considers a range of emergency dispatch modes. The objective was to evacuate stranded passengers at disrupted stations using emergency buses while improving transportation efficiency. The dispatch model incorporated key factors such as the dispatching capacity of the emergency parking lot, vehicle capacity, and vehicle rescue time. The study developed a multi-objective combination scheduling optimization model utilizing a fast non-dominated sorting genetic algorithm (NSGA-II) to reduce transportation costs and the average passenger delay. The algorithm was improved to enhance the diversity and performance of the population, and a Pareto distribution optimization solution was ultimately obtained. The Nanchang Rail Transit Line 1 was used as an example to solve both the emergency bus combination dispatch plan and the single dispatch plan separately. Results revealed that the proposed combination dispatch plan reduced passenger delay time by 20.48% and transportation costs by 16.96% compared to traditional single dispatch plans. Additionally, the improved NSGA-II algorithm further reduced passenger delay time and transportation costs by 4.50% and 3.59%, respectively. Sensitivity analysis showed that fleet size negatively correlated with the average delay time of stranded passengers and positively correlated with emergency bus transportation costs, depending on the demand for transporting stranded passengers.