Hybrid Machine Learning Approach for Microclimate Prediction in Equipment-Operated Open Ventilated Greenhouse

A proposed hybrid model did the assessment on microclimate prediction with high fluctuations of indoor temperature and humidity in equipment-operated open-ventilated greenhouses. Microclimate prediction for those greenhouses is required in order to optimize equipment operation effectiveness for the next step. The proposed model is structured at three levels and combined with three heterogeneous models: multilinear regression, gradient-boosting regression, and artificial neural networks. Accuracy performance was better than only using a single model, except for the artificial neural network in temperature prediction, and the model was also compared with multivariate LSTM (long short-term memory) as well. The experiment setup configured different settings for the fans, water spray top, and sides of the three greenhouses. A remote access control system has already been implemented as a semi-automatic system for the configuration of the equipment operations in all of those greenhouses. The hybrid model achieved RMSE = 0.31 and R
²
= 0.980 for temperature and RMSE = 1.07 and R
²
= 0.96 for humidity.