Eventos Anais de eventos
COBEM 2019
25th International Congress of Mechanical Engineering
Prediction of Environment Parameters Inside a Greenhouse Using an LSTM Model
Submission Author:
Luísa Castello Branco de Sá , MG
Co-Authors:
Luísa Castello Branco de Sá, Antônio Maia, Thiago Cardoso, Ricardo Mortara Batistic
Presenter: Luísa Castello Branco de Sá
doi://10.26678/ABCM.COBEM2019.COB2019-0123
Abstract
Greenhouses production depends on controlling its environmental conditions so that they are suitable for vegetables growth. In this context, greenhouses temperature and humidity forecast models have been developed to help farmers mitigate production losses, as well as to support the development of optimal climate control strategies. Among these models are the ones based on Artificial Neural Networks (ANNs), which have the ability to capture linear and non-linear relationships of temporal series. The current literature is mostly focused on Multilayer Perceptrons (MLP), although the architecture of Recurrent Neural Networks (RNNs) is specific to handle sequential data and forecast models. Therefore, the present study aims to develop a temperature and humidity forecast model of a greenhouse, with a type of RNN denominated Long Short-Term Memory (LSTM). Environmental data from a commercial greenhouse in Belo Horizonte, Brazil, was collected with six internal and one external meteorological station and used to train the network. Performances of the LSTM model and a baseline multiple linear regression (MLR) model were compared, and the predictions of the LSTM method were better according to smaller values of RMSE and MAPE. On temperature test data, LSTM average RMSE and MAPE scores were respectively 1.232ºC and 3.349%, while MLR average scores were respectively 2.354ºC and 6.759%. On humidity test data, LSTM achieved average RMSE and MAPE of 6.051% and 7.442%, while MLR RMSE and MAPE scores were 10.313% and 12.676%.
Keywords
Artificial neural networks (ANN), Long Short-Term Memory Networks, Multilayer Perceptron, Greenhouse
