Delivery service of O2O takeout platforms requests the high timeliness, which frequently results in explosive orders during peak period. Because the order must be completed delivery during committed service time, the orders merger and delivery routing optimization are essential to O2O takeout delivery service. Taking into consideration the amount of the different order locations, this paper studies the order delivery with the closest pickup distance principle in O2O takeout platform. Firstly, we use the K-means algorithm to cluster and merge the orders. Secondly, due to the strictly time constraints of real-time order, we propose the delivery routing optimization model of order cluster with soft time window, and which can be solved with simulated annealing algorithm. Finally, using the actual delivery data of O2O takeout platform, we further demonstrate the orders merger and delivery routing optimization mode we proposed with Python.