Copyright © IFAC Large Scale Systems, Rio Patras, Greece, 1998 A KNOWLEDGE BASED SCADA SYSTEM FOR GREENHOUSE AUTOMATION A. Anastasiou, N. Sigrimis, N.Rerras Dept, of Agricultural Engineering Agricultural University of Athens, Athens 11855, Greece e-mail: ns@auadec.aua.gr Abstract : The development of an open Knowledge Based System in the form of tasks and subtasks, provides an elegant way of rapid program development New rules can be added at any time and new control scenarios may be implemented by adding new tasks, Fuzzy decisions and fuzzy controllers at the supervisory level provide adaptive reference (set-point) generators, which are key elements in optimal greenhouse controL At the low level, the adaptive reference generators enable the realization of optimality criteria in real-time, based on maximizing the "gain" when possible, and minimizing the "expenditure" based on a sequential estimate of plant's reserves, The system is also equipped with an on-line process optimization tool, which was built based on a modified search algorithm with accelerated learning, Each process when entered under the optimizing monitor, is performance driven, conducts real experiments on the site and uses a modified descent method to maximize performance, An adaptive system guides the selection of control parameter values on-line, in a process searching for the global minimum of the cost function" The system can operate in "central mode", "remote control" or "hybrid mode", The overall system consists of many different component processes, with strong or week coupling, which must be viewed and served hierarchically, The overall system to be optimized is a very complex large scale system, with many parallel processes and events, and needs a powerful SCADA system, with control and management components, This paper presents the general features of the system, how it can be set-up to achieve specific goals and discusses the features of the KBS subsystem. Copyright © 1998 IFAC Keywords : optimization, intelligence, process control, knowledge based system 1, INTRODUCTION Researchers report an overwhelming satisfaction in implementing pioneer expert systems for the energy and general management of greenhouses or other agricultural processes, Nevertheless the number of actual implementations of such systems in real practice remains small compared to the number of systems successfully demonstrated in prototype form, This discrepancy suggests that there are concerns about the cost and difficulty of integrating expert systems into a Greenhouse Management System (GMS), Applications must address this problem before the technology finds wider acceptance, 561 Expert control systems solve problems in a heuristic way, they interact with the user easily, gain experience from past records and can give answers to nonlinear, or imprecisely stated, problems, They combine the advantages of adaptive controllers (Astrom, 1988) with the ability to explore and reason about selection of control strategies and assess control performance, Part of this effort was to determine why integrating an expert system or other AI applications in a GMS is more complex than adding a conventional application, Based on such an analysis the interfaces needed to achieve an effective integration were defined, This groundwork led to the development of a software environment (MACQU) which hosts a native expert shell, OPEN for developing applications and supporting the necessary