Reinforcement-Learning-Based Local Search Approach to Integrated Order Batching: Driving Growth for Logistics and Retail
Lijie Zhou, ChengRan Lin, Zhengcai Cao
Abstract
As an important part of Industry 4.0, a smart warehouse can offer smart tips and operational constraints for users. Improving its work efficiency is a promising growth driver for logistics companies and retailers. Therefore, a reinforcement-learning-based adaptive iterated local search (RAILS) approach is proposed to improve order-picking efficiency for a smart warehouse. A batching algorithm is proposed to deal with fluctuating orders efficiently and quickly obtain a high-quality initial solution. It can speed up the search for near-optimal solutions by extracting and using the features of the orders. Then, a perturbation mechanism is designed based on reinforcement learning that can adaptively select the perturbation type and determine the perturbation strength instead of a random way. Experimental results demonstrate that the proposed approach outperforms several existing ones, and its superiority becomes more significant as problems scale up.