Research on predicting microclimate in pig house based on machine learning algorithms

Temperature, humidity, ammonia, and carbon dioxide have a significant impact on pig production, so this article studies them as a microclimate in a pig house. Predict the microclimate of the pig house and select the algorithm with the best prediction accuracy for the microclimate in the pig house. Most studies on predicting temperature and humidity, ammonia, and carbon dioxide in pig pens only use a single algorithm for prediction, without comparing multiple algorithms on the same dataset to select the algorithm with the highest prediction accuracy. To solve this problem, seven algorithm models based on GRU, LSTM, BP neural network, XGBoost, SVM, Linear regression, and Random forest were constructed. Each model consists of four modules, namely the correlation factor screening module, data preprocessing module, data normalization module, and training and prediction module. The core module is the training and prediction module of the algorithm. The seven algorithm models use corresponding built-in algorithm layers in TensorFlow to learn from historical data, find relationships between data, and then make predictions for microclimates at the next moment. The experimental results indicate that in the microclimate of the pig house mentioned in this article, all four types of environmental factors achieved the best predictive performance in the model based on linear regression algorithm.