Design and implementation of a hybrid deep reinforcement learning-evolutionary algorithm (DRL-EA) for dynamic optimization of large-scale intelligent supply chain networks.
Design and implementation of a hybrid deep reinforcement learning-evolutionary algorithm (DRL-EA) for dynamic optimization of large-scale intelligent supply chain networks.
Yaldamesgar1
1) M.Sc. in Software Engineering, Islamic Azad University, Ayatollah Amoli Branch
IOI Code
:
AAHD-ABEBBA
Publication :
National Conference on Information Technology, Nanotechnology, Artificial Intelligence and Technological Futures Studies(itnaf.ir)
Abstract :
Abstract
Modern supply chain networks face challenges of dynamic complexity, large scale, and uncertainty, which render classical optimization methods inefficient. This paper proposes a novel hybrid algorithm called Deep Reinforcement Learning-Evolutionary Algorithm (DRL-EA) based on a fuzzy multi-objective mathematical model (F-MOP). The algorithm consists of two main units: a PPO-based deep reinforcement learning unit for rapid exploration of the search space, and an improved MOEA/D-based evolutionary algorithm with adaptive operators for precise exploitation. Performance evaluation on real data from the automotive industry (including 5,000 SKUs and 200 suppliers) shows a 32.5% improvement in Hypervolume, a 41% reduction in convergence time, and over 95% reliability under disruption scenarios. Sensitivity analysis and comparisons with NSGA-III, MOPSO, and baseline DRL confirm the superiority of the proposed method.
Keywords : intelligent supply chain deep reinforcement learning multi objective evolutionary algorithm fuzzy multi objective optimization large scale