| Knowledge representation is the basis of artificial intelligence. Many models have been developed to represent domain knowledge. Some analyzing and reasoning techniques have been proposed to handle a specified kind of knowledge in order to get new information. Among all of them uncertainty knowledge representation and reasoning is the most important but very difficult. Uncertainty knowledge representation can be divided into two parts. One of them is knowledge representation based on probability theory such as belief network model, dynamic causality diagram model, markov network model, the method used in PROSPECTOR specialist system and so on. Another kind is knowledge representation not based on probability theory such as the method of certainty factor used in MYCIN specialist system, fuzzy logic model, evidential reasoning model and so force. Among of uncertainty knowledge representation and reasoning models based on probability theory, belief network model is representative, due to its rigorousness and consistence in theory, the efficient computation mechanism and intuitive graphical expression of knowledge. But it has some difficulties on knowledge representation and reasoning. The causality diagram model based on belief network overcomes some shortages of belief network, through applying Boolean logic. So it is useful in industry application. The theory of causality diagram and the fault diagnosis methods of complexity system based on it have been researched in detail in this paper. After introducing the causality model and summarizing conventional reasoning algorithm, a new reasoning method of causality diagram has been presented, to improve the deficiency of logic operation complexity and computation complexity. This method translates causality diagram into some causality trees by normalization and standardization. A new computation method has been brought forth, by using the cut sets matrix in former non-intersect causality according to the former non-intersect mind. It can lower the "NP" difficulty and raise the computation velocity in causality diagram reasoning. So it is very useful to experiment application such as fault diagnose. The multiply value causality diagram has a difficulty which does not satisfy with probability theory rigorousness. The difficulty and its reason have been discussed. Simultaneously, we analyses the other difficulty derived from the front difficulty, i.e. the reasoning result may be error when themultiple value causality diagram was applied to experiment problem. Due to overcome the difficulty in multiply value causality diagram,a reasoning algorithm has been presented. The multiply value causality diagram methodology has been supplemental defined, so it can be compatible with a single value causality diagram. A causality diagram on fault cause of a steam generator in nuclear power plant has been established. After defining the possibility of an event variable, we gave its precise computation method and approximate computation method. The computation step which transforms a multiply causality diagram into a single causality diagram and the computation formula of linkage event had given. Finally, the method updating belief after accepting the evidences has been given. The reasoning process has been illustrated through an example, the result coincides with the fact.To improve the reasoning algorithm in multiply value causality diagram, which could not deal with the fuzzy case, a fuzzy reasoning algorithm was presented. It extended the definition of the multiply value causality diagram with fuzzy. The fuzzy mapping relation between every event variable and every reader variable was made. So, the fuzzy knowledge can be expressed with membership functions. The figure distribution and the construction steps of membership functions were given. A dummy fuzzy state of event variable was defined, which maps to the reader variable, and transforms the fuzzy reasoning of reader variable into the non-fuzzy reasoning of fuzzy state. The calculation method of...
|
PDF Full Text Request