.png){kind=icon}
.png){kind=icon}
Return
Multi-Season Environmental Prediction in Caged Broiler Houses Using a Task-Adaptive GRU–Transformer Model
.png){kind=icon}
OA
AI
J
G
W
X
H
D
S
C
DOI：10.3390/agriculture16161726.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Environmental regulation in caged broiler houses requires strict control of temperature, relative humidity, and ventilation, as abnormal indoor conditions may adversely affect broiler health and productive performance. However, existing research has paid insufficient attention to environmental changes at multiple spatial locations within broiler houses, and some studies have focused solely on a single rearing cycle without seasonal differences in the indoor rearing environment. Therefore, this study proposes a GRU–Transformer prediction model based on task-adaptive fusion. By integrating data from multiple indoor temperature sensors, central relative humidity, outdoor environmental conditions, broiler age, and control signals from environmental control devices, the model predicts central temperature, central relative humidity, and temperatures at six specific locations within the house, namely the front, rear, left, right, upper, and lower positions. Additionally, the model calculates the temperature–humidity index (THI) using the predicted central temperature and central relative humidity. The model was trained using environmental data spanning multiple complete rearing cycles across four seasons and validated on four independent test sets. The validation results show that the proposed model achieved the lowest mean absolute error (MAE) in 10 out of 12 single-step tasks for central temperature, central relative humidity, and THI. Additionally, the model demonstrated relatively stable predictive performance in multi-point predictions. Furthermore, multi-step prediction tasks, ablation experiments, and random seed experiments were conducted to further verify the model’s cross-seasonal generalization capability, robustness, and tracking performance. In the future, the developed model can serve as a predictive basis to support intelligent environmental regulation using MPC (Model Predictive Control) or RL (Reinforcement Learning) frameworks.
Keywords:
broiler house
multi-point temperature prediction
GRU–Transformer
task-adaptive fusion
environment prediction
Journal
A
IF:
3.6
Papers:
204
Citations:
0
Organization
Cited Papers