A prediction–optimization–decision framework for logistics network design under demand and disruption uncertainty

Aiming at the problem of the separation of logistics network design prediction and optimization under the double uncertainty of dynamic demand and disruption risk, this paper proposes a prediction-optimization-decision integration framework with deep integration of machine learning and system optimization. The framework adopts a three-layer architecture: the prediction layer constructs a hybrid model combining spatial-temporal graph neural network and multi-head attention mechanism and uses deep ensembles and quantile regression to quantify epistemic and aleatoric uncertainty. The risk layer identifies disruption risks through attention-based multi-label classification and deep support vector data description and generates a representative risk scenario set combining multi-layer network propagation model and importance sampling. The optimization layer establishes a multi-period mixed integer programming model and introduces resilience service level constraints and conditional value-at-risk measurement to realize the synergy between cost and resilience. In order to solve large-scale problems, a learning-enhanced large-scale neighborhood search algorithm is designed, and reinforcement learning agents are embedded in the operator selection process. The experimental results show that the MAE of the proposed prediction model is reduced by 8.2% to 15.6% compared with Transformer in each step. The average F1 of risk classification reached 0.847, 6.2% higher than that of XGBoost. The total cost of the optimization algorithm is 13.4% lower than that of stochastic programming on large-scale instances, and the optimality gap reaches 0.8% within 600 s. The total cost of the framework in 12 months decreased by 16.9%, the service level improved by 5.6% points, and the interruption recovery time shortened by 62%.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2026-11-21
DOI
https://doi.org/10.1038/s41598-026-81659-8
Primary Topic
Supply Chain Resilience and Risk Management
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A prediction–optimization–decision framework for logistics network design under demand and disruption uncertainty

Piyapatr Busababodhin, Tossapol Phoophiwfa, Guoqing Chen, Tianwen Zhao et al.
Scientific Reports
Supply Chain Resilience and Risk Management
article

A prediction–optimization–decision framework for logistics network design under demand and disruption uncertainty

Piyapatr Busababodhin, Tossapol Phoophiwfa, Guoqing Chen, Tianwen Zhao, Lubing Li, Yarui Gu, Nanjun He
article en

Abstract

Aiming at the problem of the separation of logistics network design prediction and optimization under the double uncertainty of dynamic demand and disruption risk, this paper proposes a prediction-optimization-decision integration framework with deep integration of machine learning and system optimization. The framework adopts a three-layer architecture: the prediction layer constructs a hybrid model combining spatial-temporal graph neural network and multi-head attention mechanism and uses deep ensembles and quantile regression to quantify epistemic and aleatoric uncertainty. The risk layer identifies disruption risks through attention-based multi-label classification and deep support vector data description and generates a representative risk scenario set combining multi-layer network propagation model and importance sampling. The optimization layer establishes a multi-period mixed integer programming model and introduces resilience service level constraints and conditional value-at-risk measurement to realize the synergy between cost and resilience. In order to solve large-scale problems, a learning-enhanced large-scale neighborhood search algorithm is designed, and reinforcement learning agents are embedded in the operator selection process. The experimental results show that the MAE of the proposed prediction model is reduced by 8.2% to 15.6% compared with Transformer in each step. The average F1 of risk classification reached 0.847, 6.2% higher than that of XGBoost. The total cost of the optimization algorithm is 13.4% lower than that of stochastic programming on large-scale instances, and the optimality gap reaches 0.8% within 600 s. The total cost of the framework in 12 months decreased by 16.9%, the service level improved by 5.6% points, and the interruption recovery time shortened by 62%.

Scientific Reports
Mahasarakham University (TH), Chengdu Medical College (CN), Daegu Catholic University (KR), Chengdu Jincheng College (CN), Northwest A&F University (CN)
Supply Chain Resilience and Risk Management
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon