A Machine Learning Model Based on GRU and LSTM to Predict the Environmental Parameters in a Layer House, Taking CO₂...

Area of Science:

• Agricultural Engineering
• Environmental Science
• Machine Learning

Background:

• Elevated carbon dioxide (CO2) in poultry houses impairs layer health and productivity.
• Maintaining optimal CO2 levels is crucial for preventing chronic CO2 poisoning and ensuring animal welfare.
• Predictive modeling of CO2 concentrations is essential for proactive environmental control in large-scale layer farms.

Purpose of the Study:

• To develop and evaluate a CO2 prediction model for layer houses using machine learning.
• To assess the performance of Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM) networks for CO2 forecasting.
• To identify key environmental factors influencing CO2 dynamics in poultry environments.

Main Methods:

• Utilized temperature, humidity, and CO2 data collected from an experimental layer house (June-July 2023).
• Applied data pre-processing techniques, including standardization, to 22,000 time-series data points.
• Constructed and trained GRU and LSTM models, evaluating their predictive accuracy on a test set.

Main Results:

• Both GRU and LSTM models demonstrated good generalization, stability, and high prediction accuracy for CO2 levels.
• LSTM models exhibited higher stability but lower prediction accuracy and speed compared to GRU models.
• GRU models achieved a Mean Absolute Error (MAE) of 70.8077–126.7029 ppm with prediction times of 16–24 ms for 15,000–17,000 data points.

Conclusions:

• Machine learning models, particularly GRU and LSTM, are effective for predicting CO2 concentrations in layer houses.
• Accurate CO2 prediction enables timely ventilation adjustments, crucial for maintaining optimal poultry housing conditions.
• The study highlights the potential of AI-driven environmental monitoring to enhance poultry health and production efficiency.