Improving support for greenhouse climate management : an exploration of a knowledge-based system

<p>This thesis discusses automated support for operational management of greenhouse crops and proposes a knowledge-based system to support the grower in his operational management task.</p><p>Operational management is defined as the day-to-day decision making processes which directly or indirectly lead to activities that influence the growth and development of the crop. To improve automated support for operational management, the growers operational decision making has been analyzed in the light of theory related to problem solving. The analysis of the task environment has resulted in a model of the decision processes within operational crop production management. This model has been based on the intelligence - design - choice cycle of Simon (1997). During the design and choice phases of this model the grower has to convert his observations at the crop and environment level into actions that can be implemented at the control level. Since this conversion is considered a complex and knowledge intensive task, a knowledge-based system is proposed to support the grower. The main idea behind the approach is to allow the grower to tell the system what objectives it must pursue and have the system deduce the required device settings at the control level. As these objectives may be situated at the crop, environment and control level, both domain knowledge as well as a suitable inference mechanism is required to realize such an approach.</p><p>Analysis of the knowledge in the domain of crop production shows that this knowledge is, or can be made available. Regarding the latter, the Blossom-end Rot example shows that knowledge can be made available in a suitable format.</p><p>With respect to the inference mechanism past approaches have been surveyed. Based on the results of this survey, the characteristics of the inference problem, and the attributes of the domain knowledge, it has been concluded that constraint reasoning fits the requirements best.</p><p>Simulation experiments with a prototype implementation show that the constraint reasoning can indeed be used as inference mechanism, however it is argued that the amount of work needed to realize and implementation in practice is formidable.</p&gt