Neural models for an intelligent greenhouse - the heating

High quality greenhouse control requires accurate modeling of the greenhouse as a thermal system along with all the influences affecting it. A decomposed model is the only way to tackle the complexity of such a system. A very important module of the decomposition is the heating system, due to its high impact on the overall financial cost of the greenhouse. This paper inspects the theoretical limits of heating modeling considering the stochastic circumstances present in the data measured in an industrial greenhouse. After that various models of different complexity and structure are examined. The best performance is produced by the usage of two neural networks separately for the warming and cooling heating pipe process.