Order sequencing in automated storage and retrieval systems: A neural network-assisted simulation-optimisation approach

Automated warehouses are critical to modern supply chains. One type of automation is the automated storage and retrieval system (AS/RS), which improves efficiency, reduces operational costs, and enhances inventory control. Among several optimisation problems in AS/RS, this paper focuses on the order sequencing problem (OSP) in a multi-level shuttle system, which involves finding the optimal sequence of a given number of picking orders to minimise the makespan. Because the objective function of the OSP is difficult to formulate mathematically, stochastic simulation is often used for evaluation, despite its potentially high computational cost. To address this challenge, a neural network-based metamodel (or surrogate model) is proposed to approximate the objective function. The predictive capabilities of recurrent neural networks (RNN), long short-term memory networks (LSTM), and gated recurrent units (GRU) were evaluated, with GRU identified as the most effective in predicting the makespan required to complete the order set. The GRU model was then embedded into an improved particle swarm optimisation (PSO) algorithm designed to solve the OSP. Experimental results demonstrated the effectiveness of the GRU-based PSO approach, showing competitive performance against established methods such as first-in-first-out, random search, multi-start simulated annealing, genetic algorithm, and standard PSO. • Order sequencing problem in automated storage and retrieval systems. • Evaluation of recurrent neural network-based metamodels to predict the makespan. • Computational performance comparison with a discrete event simulation model. • Integration of the best metamodel with a metaheuristic to solve the problem. • Comparison of the proposed metaheuristic with state-of-the-art approaches.