A Neural Network Model to Control Greenhouse Environment

Abstract

This research was developed in a greenhouse located in Mexico, in which there are big variations in temperature and relative humidity, generating production losses. Consequently a good greenhouse control tool was necessary to keep these variables inside of the optimal levels. Black box models have been applied in this greenhouse to predict temperature and relative humidity, however they fail in relative humidity predictions because of non linear relationships in the variables. Therefore an Artificial Neural Network (ANN) was implemented because it excel at uncovering patterns or relationships in data and it is also a powerful non-linear estimator. A total number of 14,490 data patterns were available 50% for training, 25% for verification, and 25% for testing. The ANN developed demonstrates a highly accurate estimation for both variables which can be used to forecast the conditions inside of the greenhouse and consequently take actions ahead of time, avoiding economical losses.