A methodology for identification of greenhouse humidity based on Recurrent Radial Basis Functions

Abstract

The main purpose of the agricultural greenhouses is to improve the environmental conditions of the crop. For this reason these greenhouses are generally equipped with climate control systems, in order to regulate the effect of environmental factors such as the air temperature, humidity, CO2 and solar radiation. A critical issue in the design of a controller that adapts to different conditions is the identification of the behaviour of the greenhouse, thus in this paper we explore the application of Radial Basis Function neural networks to identify and forecast the dynamics of the humidity, as a function of other climatic parameters. The performance of the proposed methodology was assessed on an experimental greenhouse, where outside and inside climatic data were recorded over the period between 11th and 18th December 2007. The main climatic parameters considered to estimate the inside humidity were outside temperature, outside humidity, inside temperature and command. After a set of preliminary experiments, an appropriate preprocessing was determined. The experiments with the standard network showed a limited performance, so a Recurrent Radial Basis Function neural network was introduced. A comparison of measured and simulated data showed that the models were able to identify and forecast inside greenhouse
conditions reasonably well. Besides, the recurrent model presented consistently superior performance, hence the usage of recurrent networks is advocated for identification tasks.