NEURAL-NETWORK MODELS OF THE GREENHOUSE CLIMATE

Neural network (NN) models of the greenhouse climate are shown to be potentially useful for the following tasks: as models for optimal environmental control, and as a screening tool in preparation for developing physical models. The main advantages of NN models in the present context are that they do not require explicit evaluation of transfer coefficients, and need no model formulation. The main disadvantage is that they cannot be used for design purposes. NN models were trained with experimental data from research greenhouses in Avignon, France and Silsoe, UK, and it was found that these produced good predictions of the inside environment, given the outside conditions and the operation of the control equipment. Surprisingly, changes of leaf area index in a tomato greenhouse did not significantly affect the inside environment. The results support previous findings that wind direction and temperature elevation are not important predictors of summer ventilation. The results also suggest that ventilation rate can be predicted somewhat better from outside weather conditions than from temperature elevation. © 1994 Silsoe Research Institute.