Towards an adaptive model for greenhouse control

Application of advanced controllers in horticultural practice requires detailed models. Even highly sophisticated models require regular attention from the user due to changing circumstances like plant growth, changing material properties and modifications in greenhouse design and layout. Moreover, their calibration is data demanding and laborious. This study explores the suitability of the extended Kalman filter (EKF) for automatic, on-line estimation and adaptation of parameters in a physics-based greenhouse model. The method was tested with measured data recorded over a period of 1 year, and with a model that describes the air temperature and moisture content in the Watergy greenhouse. In order to keep the parameters estimation problem tractable, and to improve the local accuracy of the parameters, separate EKFs are applied to sub-systems, using observation data at the sub-system boundaries. The filter adequately adjusts parameter values, thus significantly improving the model fit as compared to simulations with no-varying parameters. It appears that the filter is robust with respect to sudden changes in the system; when a disturbance occurs, such as pruning of plants or emergency opening of the windows, the EKF changes the parameter values accordingly. The result suggests that the extended Kalman filter is, indeed, a suitable method to provide the required automatic adaptation to time-varying phenomena when modeling is impractical.