COMPLEXITY AND CONTROL OF COMBUSTION PROCESSES IN INDUSTRY M. Annunziato 1 , I. Bertini 1 , A. Pannicelli 2 , S. Pizzuti 3 , L. Tsimring 4 1 ENEA, 2 CISI, 3 SOPIN, 4 UCSD ABSTRACT This paper reports some applications of artificial life environments for the development of solutions based on the evolutionary properties. We show how this approach is able to create complex structures and how they can be used to solve optimization problems and applied to the process control and optimization. 1 INTRODUCTION The ideas of evolution, complexity, intelligence and life reproduction have long been stimulating the collective thinking. Scientific approaches then become predominant on the formation of hypothesis and practices to answer to these basic questions. Research and development, carried out around mathematical and physical models of intelligence (Artificial Intelligence) and more recently of life itself (Artificial Life), are developing new tools and ideas for the solution of complex problems which require evolving structures. In problems ranging from traffic regulation to energy process control and optimization, the approaches that are based on model adaptation are not sufficient to solve the problem for long time. The not-controlled variables, the process aging, the irrational components caused by human intervention, the evolution of the process, in most of the cases require the change of the basic model or the objectives, or even the whole strategy. These difficulties expose the limitations of the systems based on the artificial intelligence or expert systems. In the expert systems, the intelligence of the human expert is formalized in a knowledge base and then transferred to the system. The artificial neural networks and fuzzy logic (Annunziato 1998) have been developed on the base of the emulation of the human reasoning (Hopfield 1983) and they have achieved large success in nonlinear modeling or control problems. If the knowledge base of the expert is not-optimal, the knowledge model is not accurate, the knowledge base or the neural network (in example the training data set) are not continuously updated following the process variations (continuous learning), then the system will not be able to drive/interpret the process for long time. Unfortunately, today it is quite clear that the idea of being able to transfer our intelligence to a machine is very difficult to realize in practice . To reach the goal of evolving structures, continuos learning of the system from the environment is necessary but not sufficient, and the ability of the system to change its internal structure is required. In short, we need information structures, which are able to evolve in parallel to the process we are modeling. Since late 70's a new branch of theory has been open in the evolutionary systems research: the genetic algorithms, starting from (Holland 1975) and developed in different directions (Goldberg 1989). In these approaches the algorithm structure is able to maximize an optimization function, or to optimize a winning strategy simulating some mechanisms of the genetic dynamics of chromosomes (reproduction, recombination, mutation, selection). These algorithms have been successful applied in many technological and engineering problems, in order to solve optimization (Oliver 1987) or design problems (Soddu 1993). The limitation of these approaches is that the internal structure of the information is generally static and defined/controlled by the author of the algorithm. In such a way these algorithms have been demonstrated to be very efficient to solve certain specific problems, but they are not really able to develop the necessary intelligence to evolve their internal structure. For instance they cannot produce a part of itself (autopoiesis (Maturana 1973)). A specific concept was introduced in 1960's and 1970's to take into account the evolving structures: the self-organization. This concept, introduced by (Ashby 1962), refers to the complex systems composed by a multitude of independent entities characterized by autonomous chaotic behavior. The self-organization is represented by the onset of a global organized structure (order) in the system. This concept has been studied by several authors (Prigogine 1971,1984) and (Kaufmann 1993) and applied to explain living and not-living systems. Lately, the self-organization concept has been adopted in many contexts of biological, physical and human sciences (fluidynamics, turbulence, laser theory (Haker 1977), social systems, economics, psychology) and it starts to be used for industrial applications (granular materials, optimization problems, process control and dynamics) through the modeling based on cellular automata [7,8]. The fusion of the concepts of genetics algorithms and the self-organization brought about the concept of the artificial life [9,10] started in the 80's.