Adaptive Environmental Control System for Large-Scale Poultry Houses Based on Multif-LSTM

The environmental conditions in large-scale, intensive poultry farming systems require high precision, and accurate prediction of environmental factors is critical for effective control.Existing control methods generally focus on the prediction and control of individual environmental factors without considering the interdependencies among these factors, leading to low prediction and control accuracy.To address the complex nature of the environmental system in poultry houses, characterised by multi-factor dependencies, an adaptive environmental control system based on Multi-feature Long Short-Term Memory (Multif-LSTM) was proposed.The Multif-LSTM model within the system calculates the dependencies between environmental factors using correlation coefficients and establishes a multi-input, multi-output neural network architecture.External climate factors are also incorporated during the input phase.Experimental comparisons conducted in a duck house environment, with Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) models, show that the Multif-LSTM model outperforms others in terms of prediction accuracy.For NH 3 concentration, the Root Mean Square Error (RMSE), the Mean Absolute Percentage Error (MAPE), and Coefficient of Determination (R 2 ) values are 1.34, 8.3, and 0.55, respectively; for temperature, they are 0.29, 2.83, and 0.98; and for relative humidity, they are 1.73, 2.46, and 0.95, respectively.Compared to the average performance of the RNN and LSTM models, the RMSE is reduced by 2.5, MAPE by 4.6, and R 2 increased by 0.32.The results demonstrate that the Multif-LSTM model achieves higher prediction accuracy and is suitable for high-precision adaptive environmental control in poultry houses.