Related Experiment Videos
SC-DGLA: constraint-aware pallet demand forecasting with dynamic graph and learnable lag alignment.
Bin Ye1, Junhong Zheng2, Linfei Zhu1
1China Tobacco Zhejiang Industrial Co., Ltd, Hangzhou, 310008, China.
Scientific Reports
|May 8, 2026
Summary
Accurate pallet demand forecasting is crucial for warehouses. SC-DGLA, a new framework, improves forecast feasibility and reduces errors by aligning data and enforcing constraints.
Area of Science:
- Supply Chain Management
- Operations Research
- Data Science
Background:
- Accurate pallet demand forecasting is vital for efficient multi-echelon warehouse operations.
- Existing forecasting methods struggle with temporal signal misalignment and operational constraints.
Purpose of the Study:
- To develop a constraint-aware forecasting framework for improved pallet demand prediction.
- To address temporal misalignment and enforce operational feasibility in supply chain forecasts.
Main Methods:
- Dynamic graph learning to model evolving network structures.
- Conditional learnable lag alignment (LLA) for signal synchronization.
- Constraint-aware training with projection-based decoding for feasible outputs.
Main Results:
- SC-DGLA framework achieves high forecast feasibility (92.8%).
- Significantly reduces shortage (3.1%) and overcapacity (2.7%) rates.
- Maintains forecasting accuracy comparable to existing baselines.
Conclusions:
- SC-DGLA offers practical and operationally feasible forecasting support for warehouse planning.
- The framework enhances decision-making by providing reliable and constraint-compliant demand predictions.
