Multi-Step Apparent Temperature Prediction in Broiler Houses Using a Hybrid SE-TCN-Transformer Model with Kalman...

Area of Science:

• Agricultural Engineering
• Environmental Science
• Data Science

Background:

• Intensive broiler production faces challenges from rapid environmental changes causing heat stress.
• Apparent temperature (AT), a composite index, is vital for predictive climate control in poultry houses.
• Effective heat stress management is critical for broiler welfare and economic performance.

Purpose of the Study:

• To develop and validate a multi-step apparent temperature forecasting model for broiler production environments.
• To enhance the accuracy and robustness of thermal environment prediction for proactive climate control.
• To provide a reliable tool for intelligent ventilation and heat stress mitigation strategies.

Main Methods:

• Developed a hybrid SE-TCN-Transformer architecture incorporating Kalman filtering for AT forecasting.
• Utilized multi-source time-series data from a commercial broiler house for model training.
• Evaluated the model's performance against benchmark models (LSTM, GRU, Autoformer, Informer) at various prediction horizons (5, 15, 30 min).

Main Results:

• The proposed hybrid model demonstrated substantially lower prediction errors compared to benchmark models.
• Achieved higher determination coefficients, indicating superior accuracy in AT forecasting.
• The integration of SE attention, Transformer, and Kalman smoothing significantly improved model robustness.

Conclusions:

• The developed model offers a precise and reliable tool for intelligent ventilation and heat stress management in broiler production.
• Findings provide scientific insights into multi-step thermal environment prediction using hybrid deep learning architectures.
• The study offers practical guidance for optimizing broiler welfare and production performance through advanced climate control.