Optimization-Driven Resilience: A Foundation Time-Series and Adaptive Robust Paradigm for Dynamic Logistics Scheduling

Abstract

The inherent vulnerability of global supply chains to multi-scale disruptions demands an entry into more sophisticated paradigms that bridge predictive intelligence with prescriptive optimization. This paper explores a novel framework integrating foundation time-series models with adaptive robust optimization to address the dynamic vehicle routing problem under highly volatile, time-varying demands. Recognizing that traditional stochastic or fuzzy frameworks often fail to capture unprecedented black-swan events, we leverage the zero-shot generalization capabilities of a foundation time-series architecture to forecast upstream supply shocks and downstream demand fluctuations simultaneously. However, migrating these massive model outputs into operational logistics scheduling revealed non-trivial misalignments regarding temporal granularity and error propagation. To accommodate these data-driven uncertainties, we construct an adaptive robust optimization model where the uncertainty sets are dynamically updated by the foundation model’s probabilistic intervals. Rather than assuming a linear, frictionless execution, our simulated empirical evaluations on empirical datasets indicate that while the proposed method significantly mitigates total routing costs under extreme scenarios, its performance gains are to some extent bounded by the computational latency of the foundation model and potential data-drifts in highly non-stationary environments. The findings suggest that while foundation models offer plausible pathways for enhanced resilience, a complete substitution of classical operational heuristics remains premature, thereby necessitating further research into hybrid, low-latency decision architectures.