This study considers the train timetable as the problem object to investigate the station blockage caused by emergencies. Based on the evolutionary algorithm framework, a real-time rescheduling approach using problem-specific knowledge is proposed. This approach ensures the safety and efficiency of high-speed rail operations and the satisfaction and comfort of passengers by minimizing the total train delay. First, a permutation encoding method is developed to reduce invalid searches in the search space based on the rescheduling strategy of reordering train departure sequences. Next, according to the train operation tracking mode, called tracking interval control, a heuristic decoding method is designed to eliminate all constraints of train operation to improve efficiency. Finally, the dispatcher's experience in adjusting the train timetable is used as the problem-specific knowledge to initialize the initial population of evolutionary computing. A heuristic population initialization method based on problem-specific knowledge is employed to speed up the algorithm convergence in the early stage and improve solution equality. Furthermore, the Beijing–Tianjin high-speed railway line is used as an example. Nine typical scenarios with different blockage durations of 20–150 min under the station blockage are installed at the Beijing South station. The strengthened elitist genetic algorithm (SEGA) and differential evolution (DE) are selected to perform the simulation using different combinations of the real-integer or permutation encoding and random or heuristic population initialization, respectively. The simulation results indicate that, compared with the real-integer encoding that cannot obtain feasible solutions, the two evolutionary algorithms can provide the rescheduling solution with a smaller total train delay in an average time of 9 s after using the permutation encoding. Besides, the results of the two evolutionary algorithms improved by the heuristic population initialization can quickly converge to a quasi-optimal solution. Finally, the SEGA in the permutation encoding with the heuristic population initialization is chosen as the optimal improved evolutionary algorithm. This improved evolutionary algorithm can provide quasi-optimal solutions in 20 s in the seven scenarios where the CPLEX cannot provide optimal solutions in 10 min.