Neural Network Model for Greenhouse Microclimate Predictions

Abstract

Food production and energy consumption are two important factors when assessing greenhouse systems. The first must respond, both quantitatively and qualitatively, to the needs of the population, whereas the latter must be kept as low as possible. As a result, to properly control these two essential aspects, the appropriate greenhouse environment should be maintained using a computational decision support system (DSS), which will be especially adaptable to changes in the characteristics of the external environment. A multilayer perceptron neural network (MLP-NN) was designed to model the internal temperature and relative humidity of an agricultural greenhouse. The specific NN uses Levenberg–Marquardt backpropagation as a training algorithm; the input variables are the external temperature and relative humidity, wind speed, and solar irradiance, as well as the internal temperature and relative humidity, up to three timesteps before the modeled timestep. The maximum errors of the modeled temperature and relative humidity are 0.877 K and 2.838%, respectively, whereas the coefficients of determination are 0.999 for both parameters. A model with a low maximum error in predictions will enable a DSS to provide the appropriate commands to the greenhouse actuators to maintain the internal conditions at the desired levels for cultivation with the minimum possible energy consumption.