AMBD: Attention Based Multi-Block Deep Learning Model for Warehouse Dwell Time Prediction

Warehouse dwell time consists of working time and waiting time of trucks that have loading tasks in a warehouse. Warehouse dwell time prediction plays a crucial role for improving the truck scheduling strategies as well as the truck drivers’ experiences, and further proliferating the efficiency of warehouse logistics. Queuing theory based time prediction methods mainly focus on some queuing events with regularity to conform. However, the warehouse queuing system has a more complex structure. Specifically, the dwell time of any truck depends on the dwell time of its preceding trucks and the loading ability of warehouse. While warehouse loading ability keeps dynamically changing due to several factors like the capacity of the production line and loading device failures. This greatly increases the difficulty of warehouse dwell time predicting. In this paper, we first put forward the definition of block to represent the loading task statuses of different trucks. On the basis of that, we propose a deep learning based multi-block dwell time prediction model, called AMBD. It incorporates the loading ability of warehouse and the execution process of loading tasks of preceding trucks in the queue. Moreover, to proliferate the precision of warehouse dwell time prediction, we introduce attention mechanism to extract strong correlations among the trucks’ dwell time. Experimental results on real-world steel logistics data demonstrate the efficacy of our proposed models.