A Deep Learning Accelerated Heuristic for Truck Loading Optimization

The efficiency of inbound supply chains can be increased by consolidating shipments from suppliers to manufacturing plants. We address a key challenge in this area involving a combined truck loading and supplier to plant assignment problem. Determining whether a truck can be loaded with a particular set of items is challenging and time consuming, thus we introduce a deep neural network (DNN) that predicts the feasibility of packing a truck load. We integrate the DNN into an existing heuristic framework to tackle the combined problem of item to truck assignment and subsequent truck loading. The DNN is used to ignore truck packing problems that are likely infeasible, saving runtime by avoiding unproductive computation steps. We evaluate our approach on a real-world problem provided by Renault and show that our learning-based approach finds better solutions faster than the existing state-of-the-art heuristic, resulting in an improvement of the objective function value by 3.83% on average.