Imagine having a crystal ball to predict the best conditions for your pigs. Well, scientists have created something similar: a smart model that uses deep learning to forecast the environment in pig houses. This model combines several advanced techniques, including Bayesian optimization, squeeze and excitation blocks, convolutional neural networks, and gated recurrent units. The goal? To improve the accuracy of predictions and boost animal welfare.

Let's break it down. The model uses Bayesian optimization to tweak its settings, like the number of GRUs, initial learning rate, and regularization factor. Environmental data is fed into the system, where it's analyzed by squeeze and excitation blocks combined with convolutional neural networks. This combo helps extract important features from the data and ignore the less important stuff.

Next, the extracted features are fed into a GRU network, which captures long-term patterns in the data. This information is then used to predict future values, such as temperature, humidity, CO₂, and NH₃ concentrations.

But how well does it work? Comparative experiments showed that this model outperformed others like CNN-LSTM, CNN-BiLSTM, and CNN-GRU. It scored higher on key metrics like the coefficient of determination (R²) and lower on errors like mean absolute error (MSE) and mean absolute percentage error (MAPE). Especially impressive was its prediction of ammonia, hitting an R² of 0. 9883, MSE of 0. 03243, and MAPE of 0. 01536.

These results show that this model is not just good, but really, really good at predicting the pig house environment. It's like having a super-smart assistant helping farmers make better decisions for their animals.

Smart Farming: Predicting Pig House Environments with AI

Actions