Related Experiment Video
Updated: Jun 13, 2026
07:08
Estimate the Cognitive Load Using Electrocardiographic Measure: A Human-AI Collaborative Task
Published on: December 5, 2025
Graph-Enhanced Management-Context-Aware Multi-Step Forecasting of Hourly Sensor-Derived Physiological and Behavioral
1School of Computer Science and Artificial Intelligence, Northeast Forestry University, Harbin 150040, China.
Animals : an Open Access Journal From MDPI
|June 12, 2026
Summary
Forecasting sheep behavior using the GCL-Sheep model improves precision farming. This management-aware model accurately predicts animal states, aiding farm management and welfare decisions.
Area of Science:
- Animal Science
- Machine Learning
- Precision Agriculture
Background:
- Animal state indicators are crucial for precision sheep farming.
- Sheep behavior is influenced by environmental, historical, and management factors.
- Accurate forecasting of animal states requires context-aware models.
Purpose of the Study:
- To develop a management-context-aware model for multi-step forecasting of sheep behavior and physiological states.
- To evaluate the model's performance in predicting active duration, rumination, feeding, intense exercise, and body temperature.
- To assess the model's applicability across different farm environments and management contexts.
Main Methods:
- Developed Graph-Enhanced Contextual Long-Range Forecasting for Sheep Farming (GCL-Sheep).
- Utilized monitoring records from 115 Hu sheep across two farms and three barns.
- Integrated barn environmental, individual physiological/behavioral, and management-operation data into hourly sequences.
- Employed Cross-Variable Graph Construction, hierarchical management-context prefixes, and long-context temporal modeling.
Main Results:
- GCL-Sheep significantly reduced forecasting errors (MAE by 20.0%, RMSE by 19.2%) for active duration at a 12h horizon.
- Improved the coefficient of determination by 0.079 compared to the best baseline in in-domain forecasting.
- Achieved an average coefficient of determination of 0.792 in Leave-One-Domain-Out evaluation, with further improvements via few-shot fine-tuning.
- Identified a 96h historical window as optimal for balancing accuracy and temporal coverage.
Conclusions:
- GCL-Sheep demonstrates promising retrospective multi-step forecasting performance for sheep.
- Sensor-based animal-state forecasting can provide decision support for farm management, including inspection scheduling and environmental control.
- Prospective field validation is needed to confirm the effects of welfare-threshold-based early warnings and interventions.