Bayesian networks for greenhouse temperature control

Greenhouse crop production is directly influenced by climate conditions. A Bayesian network is introduced in this paper aimed at achieving adequate inside climate conditions (mainly temperature and humidity) by acting on actuators based on the value of different state variables and disturbances acting on the system. The system is built and tested using data gathered from a real greenhouse under closed-loop control (where several controllers as gain scheduling ones are used), but where growers can also perform control actions independent on the automatic control system. The Bayesian Network has demonstrated to provide a good approximation of a control signal based on previous manual and control actions implemented in the same system (based on predefined setpoints), as well as on the environmental conditions. The results thus show the performance and applicability of Bayesian networks within climate control framework.