Application of Neural Networks and multiple regression models in greenhouse climate estimation

Abstract

Artificial Neural Networks (ANNs) are biologically inspired computer programs designed to simulate the way in which the human brain processes information. After a comprehensive literature survey on the application of ANNs in greenhouses, this work describes the results of using ANNs to predict the roof temperature, inside air humidity, soil temperature and inside soil humidity (Tri, RHia, Tis, RHis), in a semi-solar greenhouse according to use some inside and outside parameters in the institute of renewable energy in East Azerbaijan province, Iran. For this purpose, a semi-solar greenhouse was designed and constructed for the first time in Iran. The model database selected beside on the main and important factors influence the four above variables inside the greenhouse. Neural estimation models were constructed with (Vo, Tia, Toa, Ir, Tis, RHia, Tri) as the inputs and (Tri, RHis, Tis, RHia) as the outputs. Optimal parameters for the network were selected via a trial and error procedure on the available data. Results showed that MLP (Multilayer Perceptron) algorithm with 4-6-1(4 inputs in first layer, 6 neurons in hidden layer and an output) and 4-9-1(4 inputs in first layer, 9 neurons in hidden layer and an output) topologies can predict inside soil and air humidity and inside roof and soil temperature with a low error (RMSE=0.25°C, 0.30%, 1.06°C and 0.25% for Tri, RHis, Tis and RHia), respectively. Also the results showed that regression model has a low error to predict Tri (RMSE=0.71°C) and high error to estimate Tis (2.71°C), respectively. In overall, the error for regression model to predict all 4 parameters (Tri, RHis, Tis, RHia) was about 2 times higher than MLP method. It is concluded that ANN represents a promising tool for predicting inside climate in a greenhouse and will be useful in automatic greenhouses. For practical application, however, the farmers should use metrological and experimental data for 12 months of the year to decrease the prediction error.