A MULTI-AGENT APPROACH FOR THE DYNAMIC OPTIMIZATION OF GREENHOUSE ENVIRONMENTS

The optimal control of climatic conditions in greenhouses is a complex and multifaceted problem. Many types and sources of knowledge can be used and this knowledge is in constant flux. A software system (GX) that allows the use of several different knowledge representation formalisms (rule-sets, mathematical models, fuzzy systems, neural networks, databases etc) was constructed and tested. It makes use of the object-oriented paradigm and of the blackboard formalism for problem resolution. It operates in real-time and has been used to implement flexible regulation strategies based on heuristics, models, user input and synthesized data. It supports the collaborative and concurrent work of several software "agents" and is thus resilient and incrementally upgradeable. The behavior of multi-agent regulation strategies is presented and compared to simpler, less dynamic strategies.