Technical University of Munich, Department of Computer
MQM Paper: [Struss 93a]
(Multiple-)Model-based Diagnosis Applied to Power Networks.
In: Proceedings of the Second European Control Conference, Groningen, June 28 - Jul. 1, 1993.
Differential equations, transfer functions, and numerical methods and systems for handling them form a good basis for modeling different kinds of technical systems and are commonly used for designing automation and control systems. However, the operator, say, in a control center of a power transmission network is able to localize a short circuit in the network and to restore its functionality without applying them, even without knowing they exist. They would not be of any help, because the information available to him is not of the numerical nature required by such models and methods. There are no exact measurements of current or voltage reported to him, but simply information indicating an overcurrent in the network. The operator exploits knowledge about the structure of the network and a qualitative model of its behavior and the current scenario. Obviously, there exist different models of physical systems and phenomena, and they vary in degree of detail, validity, and utility for different purposes. Any attempt to develop computer systems that support this kind of diagnostic task have to take this into account, as well.
Research on a second generation of knowledge-based systems in AI offers a model-based approach to automated diagnosis ([Hamscher et al. 92]). In this approach, a model is not necessarily an analytical or numerical model. It can be a rather abstract and qualitative model, and even a model that deliberately ignores some of the physical phenomena known to exist in reality, a simplified or "wrong" model.
In this paper, we
We explain this application and its requirements in the following section. Next we briefly discuss the objectives and foundations of model-based diagnostic techniques in contrast to first generation systems, showing that they provide some necessary conditions for satisfying these application requirements. In section 4, we describe the network diagnosis system DPNet in more detail, demonstrating, in particular, how the system
This description is based on work carried out by the author and his former group at Siemens Corporate R&D (Munich) and, jointly with CISE (Milan), in the ESPRIT project ARTIST. Section 5 presents some advances of the basic theory and system that encorporate