Dynamic worker allocation in Seru production systems with actor–critic and pointer networks

Abstract

Following the rapid evolution of manufacturing industries, customer demands may change dramatically, which challenges the conventional production systems. Seru production system (SPS) is a key to deal with uncertain varieties and fluctuating volumes. In dynamic scenarios, orders with uncertain demands arrive over time. For each arriving order, appropriate workers should be allocated to assemble it. This study investigates the dynamic worker allocation problem with the objective of maximizing the revenue obtained by the SPS. To tackle this problem, a novel algorithm that integrates actor–critic and pointer networks is proposed. The global-and-local attention mechanism and twin focus encoders are particularly designed to address the dynamic and uncertain properties of the problem. The algorithm is compared to three approaches, including the standard actor–critic algorithm, proximal policy optimization algorithm, and the approximation algorithm with the best approximation ratio, in different scenarios, i.e., small, medium, and large factories. The proposed algorithm outperforms the standard actor–critic approach and proximal policy optimization algorithm, showing performance gaps ranging from 7.23% to 37.44%. It also outperforms the approximation algorithm, with gaps between 56.73% and 96.94%. Numerical results of the three scenarios show that the proposed algorithm is more efficient and effective in handling uncertainty and dynamics, making it a promising solution for real-world manufacturing production systems.