The business pattern of takeout is very complex and there are shot of references focusing on the optimization of takeout distribution. In view of this, a time penalty cost is introduced for deliveries outside a specified time window, which is based on the vehicle routing problem with simultaneous pickup and delivery. Then, the objective function is set as the sum of the incremental costs of takeout distribution, which includes the fixed distribution costs as well as the additional distribution cost and time penalty cost of new orders. As the random parameters have an impact on the computational complexity of this problem, a specified distribution area is designed, although orders that have not yet been completed remain with the original vehicle while new orders are dispatched. Then the time penalty cost is used as a variable cost to modify the objective function, and the time window is removed to reduce the mathematical complexity. We employ k-means clustering to group “seller–customer” items and a genetic algorithm to identify the optimal routes for seller–customer pairs in the same cluster. Finally, random simulation algorithm is used to generate a dynamic order test dataset, and R language is used to test the effectiveness of the algorithm.