Resilience evaluation and adaptive recovery decision model of logistics network based on multimodal data fusion
Abstract
This paper proposes integrated framework that combines established multimodal data fusion, graph learning, and reinforcement learning techniques for logistics network resilience evaluation and adaptive recovery decision-making in uncertain environments. The model employs the visual-language pre-training framework to deeply integrate multimodal data and construct global semantic feature representation. For resilience evaluation, a hybrid architecture combining graph convolutional network and Transformer is designed to generate accurate network resilience scores by jointly capturing topological structures and spatiotemporal dependencies. For recovery decisions, deep reinforcement learning integrated with meta-learning enables rapid adaptation and policy optimization under dynamic disruption conditions. Experimental results on the Smart Logistics Supply Chain Dataset demonstrate that the proposed model achieves a Mean Squared Error (MSE) of 0.074 and an F1-Score of 0.905, outperforming all baselines. In recovery decision tasks, the average recovery time is reduced by 23.9% and the recovery is decreased by 39.2% compared with conventional methods. Ablation studies further confirm the essential contributions of the Transformer module and the spatiotemporal encoder. These findings indicate that the proposed framework offers a promising solution for enhancing logistics network resilience in practice. Future work will extend the model to cross-border multimodal transportation scenarios and incorporate carbon emission constraints to support low-carbon recovery strategies.