As an important mode of public transportation, urban rail transit needs to meet passenger travel demand while reducing energy consumption costs. To meet the dual goals of passenger travel efficiency and enterprise energy cost control, the constraints of train departure interval, passenger capacity and passenger flow was considered, and a train operation optimization model was constructed based on passenger demand and enterprise interests. For this optimization model, a convolutional neural network enhanced genetic algorithm was proposed. The calculation results of the case show that the convolutional neural network enhanced genetic algorithm is faster than the traditional genetic algorithm, and it is less likely to fall into local optimal. Using the model and algorithm proposed, the travel time of optimized passengers is reduced by 10.21% and the energy consumption after optimization is reduced by 5.13%. The result provides an effective theoretical basis for the optimization of urban rail transit schedules, which can play a role in reducing the energy consumption cost of enterprises and reducing the travel time of passengers and has important significance in improving the efficiency of transportation system and saving resource consumption.