Modelling of greenhouse climate parameters with artificial neural network and multivariate adaptive regression splines approach

Abstract

In this study, it is aimed to model some greenhouse climate parameters by using two different prediction tools based on machine learning and artificial intelligence. For this purpose, in the first part of the study, the data set which consists of indoor and outdoor measurements taken from 7 different points of the greenhouse for 12 months from a region with terrestrial climate was adapted for modeling. In the second part, the functional relationship between input (independent) and output (dependent) variables was examined by artificial neural network (ANN) and multivariate adaptive regression splines (MARS) methods. In the third part, the models were evaluated with performance criteria and the best estimation model is selected. Comparison of ANN and MARS models indicated that MARS performs better than ANN with lesser values of MAPE (mean absolute percentage error), RMSE (root mean square error) and MAD (mean absolute deviation), and slightly higher value of R2 (coefficient of determination) in order to predict mean temperature (Tmcan, °C) and relative humidity (RHmcan, %). Based on these findings, it was observed that MARS method could provide a more detailed modeling as an alternative to ANN in developing comprehensive greenhouse climate mechanization. © 2019 Elsevier B.V., All rights reserved.