ANN approach to direct and decoupled inverse modelling of inside climate in agricultural greenhouses

摘要

Greenhouses are classified as complex multivariable non stationary processes. Indeed, the modelled parameters are nonlinear coupled and strongly influenced by outside weather. The complexity of modelling and greenhouses control inside climate consists on the diversity of phenomena that influences its evolution. It is so difficult to formulate a model emulating the real dynamic of the system. In This paper Neural Networks are combined with a decoupling method to build models of greenhouses. A Multilayer Feed-Forward Neural Network has been trained to learn both inverse and direct dynamics. To adapt the parameters of neural models, the back propagation algorithm is used. Basing on experimental measures collected in successive days and the gradient descent method, models of different variables are trained and validated successfully. The models formulation and the obtained results are reported comparing with the classical Least Squares method.