Modeling of a greenhouse temperature : comparison between multimodel and neural approaches

The aim of this paper is to discuss different possibilities of modeling the temperature evolution inside an agricultural greenhouse. Our analysis is exploring a wide data measurements representing different profiles of inputs, outputs and disturbances. Our study particularly considers the situation where the control of the greenhouse temperature needs the use of a heating system. Two approaches have been considered in the present work: The first one is based on the multimodel techniques which reduce the system complexity by using a set of several simple linear models. In the second approach we use the feed forward neural networks that allows elaborating a single black box model that can spread out to another variable and command. Finally we have made a synthesis resting on the comparison between these two approaches by testing the performance of each one with the same data measurements. This work shows that both approaches give satisfactory results, multimodel representation is more suitable to obtain different variables describing the internal state system and to develop appropriate controller. The neural network model is rather a black box model but it is easier to obtain and could well be used to simulate different output variables at the same time (temperature and humidity in our case)