Data Management And Knowledge Discovery In Fuel Management Of Nulcear Reactor For Third Qinshan Nuclear Power Cooperation

The CANDU nuclear reactor in Third Qinshan Nuclear Power Company (TQNPC) is refueled while not stopping the operation. It has a very complex controlling and safety protection system. While the reactor is in operation, the refueling engineers have to conduct fuel management. The fuel management is very important for the stable and safe operation of the reactor. Currently, there are two problems in TQNPC. One problem is about the fuel management system. It is outdated and can not organize the data orderly. This has hindered the improvement of the effect and efficiency of fuel management, thus, current system does no longer meet the requirement of modern management. The second problem is that there is no effective mechanism for selecting channels to be refueled. Current method can not provide effective decision support for refueling engineers. The experienced refueling engineers have to take a lot of time to compute and make the decision which channel should be refueled.Rough set theory is a new fruit in the international research field of artificial technology. It is a theory to express, learn and generalize the information which is incomplete, inaccurate and uncertain. It not only can classify and reduce information, but also can mining out the rules. Compared with other data mining technology, the major strength of Rough Set Theory is that it does not need any prepared or additional information, and it is also a powerful data analysis tool. Presently, the Rough Set Theory's application in nuclear field is mainly focus on how to apply the theory in the fault diagnosis to deal with simple information table of fault diagnosis. However, the research on the application in fuel management is seldom.To solve the problems in the fuel management of TQNPC, firstly the mature program developing tools were used to design and develop the TQNPC refueling expert system. Then taking advantages of the merit and strong ability of knowledge capture of Rough Set Theory, the orientation research on the application of the theory in the reactor fuel management was conducted. The main work of the paper has been done as follows:1 To solve the problems of data management for TQNPC, an refueling expert system was designed and developed. After careful investigation, system feasibility analysis, demand analysis, the system structure and function modules are designed. Then each module was analyzed and designed explicitly. Finally, the software Visual Basic 6.0 and SQL Server 2000 was utilized to implementation the system.2 An information entropy-based discretization algorithm for decision table was proposed. Firstly, the change amount of conditional entropy was defined when a cut point was added to the decision table and treated as a measurement of the cut point's important degree. On the basis of that, the important degree was taken as heuristic information to select cut points until the classification ability of the new decision table equaled the primitive one. The algorithm is feasible for both consistent and inconsistent decision table and the consistence of classification is ensured. Finally, an example of stream turbine failure diagnosis was applied to test the algorithm. The experiment shows that the algorithm is effective and practical.3 An attribute reduction algorithm was proposed based on extended information entropy. A core attributes computation algorithm of decision table was proposed in the light of an extended information view. And a from-bottom-to-top decision table attribute reduction algorithm which took the significance of the attribute as the heuristic information was designed. The new algorithm"EIEAAR"can deal with both the consistent and inconsistent decision tables, and integrate the core attribute computation and non-core attribute reduction in a whole. At last, the complexity of the algorithm was analyzed and two kinds of examples were taken to test the validity of the algorithm. The experiment showed that the algorithm was valid.4 The application of Rough Set Theory in fuel management of CANDU reactor was analyzed and researched. Fistly, a framework and method of fuel management knowledge acquisition based on Rough Set was presented. Then the software Matalab was utilized to program the discretization algorithm, attribute reduction algorithm and value reduction algorithm. Furthermore, refueling knowledge was extracted by applying the key technology of Rough Set Theory from reactor historic data. Finally, the knowledge was applied in the practical channel selection in combination with fuzzy reasoning technology and the application effect was verified.