Multi-objective sustainability optimization in modern supply chain networks: A hybrid approach with federated learning and graph neural networks

Abstract

Modern supply chain systems are becoming layered and discontinuous, so sustainable performance is key to efficiency and decreased negative impacts on the environment. Even today, this goal is not easy to address because supply chain data is economic, logistic, and environmental data all in one. Conventional approaches to tackle these hurdles are inadequate since they are rigid in managing varying and scattered data types, which contributes to ineffective results in sustainability optimization. Existing approaches mostly involve the use of several methods, which include regression models, gradient boosting methods/techniques, and recurrent neural networks to study various facets of supply chain data. The main objective of this research is to optimize supply chain management using a new hybrid framework by combining Federated Learning with Graph Neural Networks. The study shall address the problems of supply chain prediction, including demand and supply fluctuation predictions while maintaining data privacy through decentralized learning. The framework aims to improve the accuracy of supply forecasts and the management of inventory within multinational supply chains by incorporating these sophisticated techniques. This paper presents the design of an FL-GNN model, assesses its performance using real-world data, and discusses its applicability for real-time supply chain optimization. Moreover, increasing the effectiveness of demonstrating and achieving multifaceted objectives, including cost minimization, increased efficiency, and a decreased negative impact on the environment, is the Fed-GNN model. The proposed Fed-GNN model demonstrates superior performance with a Mean Squared Error of 0.198, a Root Mean Squared Error of 0.560, and a Standard Deviation of 0.590. These metrics highlight the model’s accuracy and consistency, validating its effectiveness in optimizing supply chain sustainability.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 24 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.