Integration of symbolic and connectionist AI techniques in the development of Decision Support Systems applied to biochemical processes
Davide Sottara · AMS Dottorato Institutional Doctoral Theses Repository (University of Bologna) · 2010
There exist complex systems, such as bio-chemical plants, which require a constant management to be kept in optimal operating conditions. To this end, automation is the only feasible option: the tasks involved, ranging from fault detection to diagnosis to control, can hardly be performed by traditional model-based controllers alone, but instead require the additional application of artificial intelligence-based techniques. AI is a vast field, covering many technologies which can be roughly classified in “Hard” or “Soft” techniques: the former typically use a symbolic representation of the data and elaborate it by logic reasoning; the latter, instead, process information at a sub-symbolic level, exploiting the interactions of many simple elaboration units hence they are also known as connectionist techniques. Since different techniques are more suitable for different problems, a complex management infrastructure is likely to include more than one module. These modules should be able to interact in order to exploit the mutual potentialities when processing the information coming from the managed system. Moreover, this information is likely to be imperfect vague and/or uncertain and/or incomplete so the modules should be able to deal with it appropriately. In this dissertation, we claim not only that taking imperfection into account is a necessary feature of an “intelligent” module interfaced to a real-world system, but also that extending formal logical reasoning with imperfection allows to obtain a deeper integration of “Hard” and “Soft” computing techniques than simply using them together, in cascade or in parallel, obtaining strongly hybrid modules which can be simpler and yet more robust than their pure counterparts. Moreover, it will be shown that such hybrid modules can be deployed effectively within an infrastructure which combines the concepts of service, agent and event in a natural way.