Optimal CO2 control in a greenhouse modeled with neural networks

CO, enrichment in warm climates requires a delicate balance between the need to ventilate and the desire to enrich. Model-based optimization can achieve this balance, but requires reliable models of the greenhouse environment and of the crop response. This study assumes that the crop response is known, and focuses on the greenhouse model. Neural network greenhouse models were trained using data collected over two summer months in a small greenhouse. The models were reduced to minimum size, by predicting separately the temperature and CO2 concentration, and by eliminating any unessential input. The resulting models not only fit the data well, they also seem qualitatively correct, and produce reasonable optimization results. Using these models, the effect of evaporative cooling on extending the enrichment duration is demonstrated. (C) 1998 Elsevier Science B.V. All rights reserved.