Graph Neural Network Supply Chain Prediction 2026
Graph Neural Networks for Supply Chain Prediction
GNNs are emerging as a core AI architecture for modeling global supply chains as graphs — enabling demand forecasting, disruption early warning, and hidden link prediction. This report maps the patent and literature landscape from 2016 to 2026.
GNNs Transform Supply Chain Modeling
GNN-based supply chain prediction treats the supply network as a graph: suppliers, manufacturers, distributors, and retailers are modeled as nodes, while procurement relationships, shipment flows, and financial transactions form edges. This graph representation allows GNN models to propagate information across network neighborhoods, capturing both local relationships and global structural properties simultaneously.
The field spans four distinct sub-problems within this dataset: supply chain visibility and link prediction, demand and inventory forecasting, digital twin construction, and risk and disruption early warning. These tasks share a common GNN substrate — heterogeneous graph architectures, dynamic temporal GNNs, graph attention networks (GATs), and graph convolutional networks (GCNs).
Publication dates among retrieved results span 2016 to 2026, revealing a clear three-phase trajectory: a foundational phase relying on classical ML (2016–2019), a development phase of rapid GNN methodology expansion (2020–2022), and a commercialization and scaling phase featuring enterprise-grade deployments by Walmart, Siemens, JPMorgan Chase, Microsoft, and Daybreak AI (2023–2026).
In this dataset, Siemens Aktiengesellschaft leads with 3 supply-chain-relevant GNN filings, followed by JPMorgan Chase Bank, Sichuan University, Beijing Institute of Technology, and Target Brands each with 2 filings in retrieved records. The US and CN together account for the majority of filings in this dataset, though the landscape is distributed across multiple geographies.
Filing Trends and Technology Cluster Distribution
Among retrieved records, GNN supply chain patents accelerated sharply after 2022, with the 2023–2026 window featuring enterprise-grade filings from Walmart, Siemens, JPMorgan Chase, and Daybreak AI. Four primary technology clusters are identifiable in this dataset: heterogeneous GNN digital twins, link prediction, risk and disruption identification, and temporal/dynamic GNN forecasting.
GNN Supply Chain Patent Filing Count by Technology Cluster (Dataset Snapshot)
In this dataset, supply chain link prediction and risk/disruption identification together account for the largest share of GNN-specific filings, followed by heterogeneous GNN digital twin construction and temporal/dynamic GNN architectures.
↗ Click bars to exploreGNN Supply Chain Patent Filings by Phase and Jurisdiction (Dataset Snapshot)
In this dataset, the 2023–2026 commercialization phase shows the highest concentration of enterprise GNN supply chain filings, with US and CN jurisdictions contributing the most records across all three phases.
↗ Click bars to exploreKey Application Domains for GNN Supply Chain Prediction
GNN-based supply chain prediction has been applied across retail, automotive and manufacturing, financial services, energy and utilities, pharmaceutical and agri-food, and ICT sectors — each with distinct graph topology requirements and prediction targets drawn from the retrieved patent and literature records.
Retail and Consumer Goods
Walmart Apollo’s 2025 US patent describes a heterogeneous GNN-based digital twin capturing structural and dynamic aspects of fast-moving consumer goods supply chains, integrating real-time data feeds for continuous state updates. Daybreak AI’s 2025 US patent trains a GNN on node features (demand information, planned shipments) and edge-level features (inter-node shipments), outputting shipment event delta probabilities and quantity scalers. Target Brands filed supply chain replenishment simulation patents in 2021 and 2024, modeling inventory at per-item, per-node, per-epoch granularity.
Digital TwinAutomotive and Manufacturing
The 2021 literature study on hidden link prediction applied GNNs to a real automotive supplier network, detecting unknown procurement interdependencies and outperforming prior heuristics. A 2020 literature study applied Node2Vec and DeepWalk embeddings to manufacturing material flow networks undergoing cloud-manufacturing-driven reconfiguration. Siemens’ two 2025 EP patents on critical path identification and supplier risk scoring are framed for industrial plants with multi-tier supplier dependencies.
Link PredictionFinancial Services and Fintech
JPMorgan Chase Bank filed two US patents (2024 and 2026) applying GNN with Laplacian Eigenmap embeddings and pairwise ranking loss to detect competitors and rank supply chain entities by Euclidean distance in embedding space. Zhejiang MYbank (Ant Group’s rural finance arm) filed a 2025 CN patent combining spatio-temporal knowledge graphs with LLM integration, targeting financial risk assessment of enterprise supply relationships. Korean firm Rakaive filed a 2025 KR patent on supply chain knowledge graph GNN for enterprise relationship scoring and credit risk assessment.
Financial Risk AIEnergy, Utilities, and ICT
State Grid Hubei Electric Power Co. applied dynamic GNN link prediction to emergency supply chain networks for power materials (CN, 2024), using both temporal state features and graph structure features to predict future emergency supply behaviors under disruption. Beijing Institute of Technology’s 2020 CN patent on ICT supply chain product share trend prediction uses Hawkes-process-augmented temporal graph embeddings to model ICT vendor supply relationships in semiconductor and electronics sectors. Microsoft’s 2023 WO patent integrates agricultural IoT sensor data and energy grid scenarios into a reinforcement learning supply chain graph simulation framework.
Dynamic GNN ForecastingLeading Patent Assignees in GNN Supply Chain — Dataset Snapshot
In this dataset, Siemens Aktiengesellschaft is the most active single entity with 3 supply-chain-relevant GNN filings across EP, WO, and US jurisdictions. JPMorgan Chase Bank and Sichuan University each hold 2 filings in retrieved records, while Walmart Apollo, Daybreak AI, Beijing Institute of Technology, and Microsoft each contribute distinct GNN supply chain patents filed between 2020 and 2026.
Top Assignees by GNN Supply Chain Patent Filings in Retrieved Records (Dataset Snapshot)
↗ Click bars to exploreSiemens Aktiengesellschaft
Siemens holds 3 GNN supply chain filings in retrieved records spanning EP, WO, and US jurisdictions filed between 2023 and 2025, making it the most active single entity in industrial supply chain graph-learning patents in this dataset. Key filings include “Automatically identifying critical objects supplied to an industrial plant” (EP, 2025), which uses a GNN classifier with centrality-scored supplier knowledge graph embeddings to output per-node risk scores, and “Method and system for automated critical path identification for supply chain management” (EP and WO, 2025), which combines a graph-to-text encoder with an LLM to identify critical supply paths. A third filing covers graph-driven production process monitoring (WO, 2023).
Germany — DE / EP / WOJPMorgan Chase Bank
JPMorgan Chase Bank holds 2 GNN supply chain filings in retrieved records, both filed in the US in 2024 and 2026. Both patents apply GNN with Laplacian Eigenmap embeddings and pairwise ranking loss to detect and rank competitor companies of supply chain entities by Euclidean distance in embedding space — treating supply chain relationships as a node classification and ranking problem. This represents a financial services crossover into supply chain competitive intelligence, distinct from operational forecasting approaches found elsewhere in this dataset.
United StatesFive Directional Signals Shaping GNN Supply Chain Prediction Through 2026
Based on the most recent filings (2024–2026) in this dataset, five directional signals are shaping the GNN supply chain prediction field: LLM and knowledge graph hybrid architectures, temporal dynamic GNNs for real-time tracking, GAT with community-aware features, financial risk applications, and hardware-accelerated multi-model forecasting.
LLM + Knowledge Graph + GNN Hybrid Architectures
The 2025 filings from both Siemens (WO, 2025) and Zhejiang MYbank (CN, 2025) integrate large language models with supply chain knowledge graphs for relationship and path prediction. Siemens’ method combines a graph-to-text encoder with an LLM to identify critical supply paths in natural-language-interpretable form. Zhejiang MYbank’s approach targets financial risk assessment of enterprise supply relationships using spatio-temporal knowledge graphs combined with LLM generalization.
Temporal and Dynamic GNN for Real-Time State Tracking
Sichuan University’s 2026 updated CN patent explicitly handles temporal graph sequences with spatial and temporal feature aggregation, addressing the insufficiency of traditional ARIMA and moving average forecasting methods. State Grid Hubei Electric Power Co.’s 2024 CN patent applies dynamic network link prediction to emergency supply chain networks, using both temporal state and graph structure features to predict future emergency supply behaviors. Both filings signal an architecture shift from static snapshot GNNs to continuous-time or multi-snapshot dynamic models.
Static GNN vs. Dynamic/Temporal GNN for Supply Chain Prediction
Click any row to explore further.
| Dimension | Static GNN (Snapshot-Based) | Dynamic / Temporal GNN |
|---|---|---|
| Graph representation | Fixed graph topology at a single point in time; node and edge features do not update between inference calls | Temporal graph sequences with evolving topology; multi-snapshot or continuous-time models track edge additions and deletions |
| Primary prediction task | Link prediction (inferring hidden supplier relationships), node classification (risk scoring) | Demand forecasting, emergency supply behavior prediction, real-time disruption early warning |
| Representative filings | Siemens EP 2025 (critical object identification); JPMorgan Chase US 2024/2026 (competitor detection); CP-GAT CN 2025 (University of Science and Technology Beijing) | Sichuan University CN 2026 (temporal graph sequences); State Grid Hubei CN 2024 (dynamic link prediction for emergency supply); GCN-GAN Literature 2019 |
| Architecture approach | GCN, GAT with multi-head attention, knowledge graph embeddings, Laplacian Eigenmap embeddings | GCN-LSTM-GAN hybrids, spatial and temporal feature aggregation stages, dynamic network link prediction |
| LLM integration | Siemens WO 2025 combines graph-to-text encoder with LLM for critical path identification in static knowledge graph | Zhejiang MYbank CN 2025 combines spatio-temporal knowledge graph with LLM for financial risk assessment of enterprise supply relationships |
| Jurisdictions observed | EP, US, CN — primarily enterprise and academic filings in Germany, United States, and China | CN, KR, WO — primarily academic and state-enterprise filings in China, plus Korean and international filings |
| Maturity signal | Enterprise-grade deployments (Walmart Apollo 2025, JPMorgan 2026) indicate production readiness for static graph inference | 2026 updates from Sichuan University and 2024 State Grid filings indicate active research-to-deployment transition for temporal models |
Frequently Asked Questions: GNN Supply Chain Prediction Patents
GNN-based supply chain prediction treats the supply network as a graph where suppliers, manufacturers, distributors, and retailers are modeled as nodes, while procurement relationships, shipment flows, and financial transactions form edges. GNN models propagate information across network neighborhoods, capturing both local relationships such as a specific supplier-buyer pair and global structural properties such as multi-tier supplier concentration risk simultaneously.
Among retrieved records, key enterprise filers include Siemens Aktiengesellschaft (3 filings, EP/WO/US), JPMorgan Chase Bank (2 filings, US), Walmart Apollo LLC (1 filing, US, 2025), Daybreak AI Inc. (1 filing, US, 2025), and Microsoft Technology Licensing LLC (1 filing, WO, 2023). Academic filers include Sichuan University (2 filings, CN), Beijing Institute of Technology (2 filings, CN), and University of Science and Technology Beijing (1 filing, CN, 2025).
Based on retrieved records, the four main clusters are: (1) Heterogeneous Graph Neural Networks for Digital Twin Construction, exemplified by Walmart Apollo’s 2025 US patent; (2) Supply Chain Link Prediction and Relationship Inference, exemplified by the 2021 automotive GNN hidden link prediction study and JPMorgan Chase’s competitor detection patents; (3) GNN-Based Risk, Disruption, and Critical Path Identification, exemplified by Siemens’ two 2025 EP patents; and (4) Temporal and Dynamic GNN Architectures for Forecasting, exemplified by Sichuan University’s 2026 CN update and State Grid Hubei’s 2024 CN patent.
Two 2025 filings signal LLM plus GNN convergence. Siemens’ WO 2025 patent on automated critical path identification combines a knowledge graph graph-to-text encoder with a Large Language Model to identify critical paths in natural-language-interpretable form, with nodes representing suppliers, branches, countries, products, and materials. Zhejiang MYbank’s CN 2025 patent combines spatio-temporal knowledge graphs with LLM integration, targeting financial risk assessment of enterprise supply relationships.
Dr. Indranil Mutsuddi’s 2026 IN patent describes a machine learning-based intelligent supply chain system explicitly targeting hardware-accelerated neural processing units (NPUs) combining LSTM, temporal convolutional networks (TCN), and gradient boosting for supply chain demand forecasting. This hybrid approach is reported to reduce mean absolute percentage error by 28–32% relative to ARIMA baselines.
Among retrieved patent results, the US is the dominant jurisdiction for enterprise-grade GNN supply chain patents, with filings from Walmart, Daybreak AI, JPMorgan Chase, Microsoft, and Target. China is highly active in academic-origin patents from Sichuan University, Beijing Institute of Technology, University of Science and Technology Beijing, Zhejiang MYbank, and State Grid Hubei. EP and WO filings are led by Siemens. India and Korea each have early-stage filings in this dataset.
Data and insights on this page are based on a limited patent and literature dataset and are for reference only. Figures may not represent the complete technology landscape.