An Anticipative Order Reservation and Online Order Batching Algorithm Based on Machine Learning

For e-commerce warehouses, on-time delivery and less order picking cost are of great importance. Order batching is the critical operation issue both in manual picking system and robotic warehousing system. Few studies have considered the effect of future arrival orders on on-line order batching, however there may be potential benefits. In this paper, we propose a novel on-line order batching solution based on machine learning by considering the potential benefit of reserving some promising orders. In this new anticipative order reservation mechanism, when order arrives, we don’t batch all arrival orders and consciously reserve some orders to be batched in the following period which may obtain extra efficient improvement. We design a reserving algorithm to decide whether an order should be reserved in order pool or immediately released to next order batching stage. A regression model trained by AutoGluon is used to predict the similarity between a current order and future coming orders. Based on the mechanism, a complete algorithm was developed to solve on-line order batching problem, including batching, sequencing and all necessary process. Finally, we test the algorithm on the real data from an e-commerce warehouse and compare with fixed and variable time-window batching in the previous literature. The result shows using the reservation mechanism has higher performance in reducing order turnover time and picking cost.