In recent years online supermarkets such as JD.com have developed greatly in the e-commerce environment. Order splitting has been a common phenomenon in practical operations due to two main features of the online supermarket： one is the multi-warehouse network in one big city， the other is the vast multi-item customer orders. Serious order splitting problem intrigues some challenges such as high delivery disturbance， high fulfillment cost. The complexity of this problem is that the solution space will be large because of the explosive increased number of splitting solutions with the increase in the number of orders and the number of items in each order. Synthetically applying the theories of combinatorial optimization and clustering analysis， the order splitting optimization problem is investigated to minimize the order splitting fulfillment cost. For the sake of decreasing the solution space， the product link-based hybrid heuristic large neighborhood search algorithm （PLBH-LNS） is proposed. The effectiveness and practicality of the PLBH-LNS algorithm are verified using a series of generated small-scale datasets and a real large-scale dataset from an online retailer in China. The results show that， compared with the myopic order splitting strategy in practical operations， the PLBH-LNS algorithm could decrease the average order fulfillment cost by 22.48%. A new method and idea for the online supermarket order splitting in e-commerce is provided， which also supports the order fulfillment decision for decreasing the order splitting.