Revision as of 10:00, 18 March 2026

Abstract

As global environmental challenges intensify, circular economy (CE) has emerged as a critical pathway for sustainable development. This study proposes a deep learning-based CE supply chain network design framework that optimizes resource allocation, reduces waste, and improves sustainability. The framework employs graph convolutional networks, long shortterm memory networks, and multi-head attention mechanisms to capture topological, temporal, and multi-dimensional supply chain features. An improved NSGA-III algorithm achieves coordinated balance among economic, environmental, social, and circularity objectives. A comprehensive sustainability evaluation system provides quantitative assessment tools. Experimental validation using real data from 15 enterprises across five industries shows the deep learning model achieves 89.2% prediction accuracy on test sets, representing 16.1% improvement over baselines and 67.9% improvement in computational efficiency. The optimized network achieves 32.4% waste reduction, 28.7% resource efficiency improvement, 25.3% cost reduction, 68.5% material circulation rate, and 89.2% network efficiency. This research contributes to theoretical understanding and provides practical guidance for manufacturing enterprises’ transition to CE, supporting sustainable development goals.OPEN ACCESS Received: 27/09/2025 Accepted: 17/11/2025 Published: 20/03/2026

Document