Measuring Supply Chain Resilience with Foundation Time-Series Models

Abstract

This study addresses the critical need for quantifying and assessing supply chain resilience by developing an advanced, data-driven measurement approach. We integrate Long Short-Term Memory (LSTM) networks with the Hybrid Foundation framework to construct a dynamic resilience index that captures both shock absorption and recovery capabilities across complex supply chains. The analysis leverages monthly operational data from 15 manufacturing enterprises collected between 2020 and 2024, encompassing key performance indicators such as order fulfillment rate, inventory turnover days, and transportation delay rate. The proposed model demonstrates exceptional predictive performance, achieving a mean squared error (MSE) of 0.0019 and a mean absolute percentage error (MAPE) of 4.32% in stability tests. Notably, it maintains high accuracy and robustness even under challenging conditions, including 5% missing data and ±10% noise perturbations, which reflects its resilience in real-world scenarios characterized by incomplete or uncertain information. By accurately simulating the temporal dynamics of supply chain disruptions and recoveries, the model provides a reliable and granular quantitative tool for resilience evaluation, supporting informed decision-making for supply chain managers. Furthermore, the methodology offers a scalable and adaptable framework applicable to diverse industrial contexts, enabling systematic monitoring, early-warning detection, and targeted strategic interventions to enhance overall operational stability and responsiveness. The findings underscore the value of combining deep learning with hybrid modeling techniques to achieve a nuanced understanding of supply chain behavior under both routine and stress conditions, contributing to the advancement of resilience analytics in contemporary supply chain management.