Fuzzy Inference Network Fault Diagnosis

As an important part of network management, fault management must have the abilities to detect the faults occurred in the network, and then confirm them, at last eliminate them. Because modern network system have the characters of hug scale and complex structure which also is difference, diverse and distribute, these will bring expensive loss when a fault occurs even just a tiny fault. And in a real network, a fault will bring more than one alarm, in order to find out the root alarms we need alarms correlation analysis.Expert system based rule is one of the most popular correlation analysis methods now because of its intelligence. However, the performance of expert system depends on the knowledge and experience of field experts, which always are not precise and complete, and always change a lot. So alarms correlation analysis using expert system has the subjective and man-made elements, and which can't be described and understand using traditional Boolean logic. As we know, a fault will bring more than one alarm, also an alarm can be aroused by different faults, the relationship between faults and alarms is not accurate corresponding but fuzzy.Combined fuzzy theory and fuzzy reasoning technology, this thesis design a knowledge representation model transforming the network alarm to fuzzed network alarm considering the subjective element of experts and the fuzzy corresponding between faults and alarms to get the fuzzy system input consistent with fuzzy reasoning, and use fuzzy reasoning technology to get the root alarms.In the first part, this thesis firstly analyze the uncertain and fuzzy characters of network and the meanings of each field use by network alarm, and then define a fuzzy judge language set for network alarm and fuzzy each information field using the judge language set as domain set, this fuzzy set can be considered as the relationship between the field and root network alarm, so we can use fuzzy integrate judge method to transform network alarm to fuzzy network alarm, which can be considered as a relationship between network alarm and network root alarm.In the second part we use production rules representation method and OO rules representation method to represent the fuzzy rules using in the fuzzy reasoning in this thesis after we got the fuzzy network alarm, and then according to the character of fuzzy reasoning, bring forward the confliction resolution stratagem using fuzzy matching method and the fuzzy control stratagem after analysis the traditional control stratagem.In the third part, this thesis tries to get the result using fuzzy reasoning and explain it. It's easy to explain the results because the fuzzy network alarms is based on the fuzzy language set, which reflect the relationship between an alarm and the root alarm. Considering the lack of fuzzy rules, the result may be a set of fuzzy network alarms; however, we can give useful suggestion to user by fuzzy distributing method.In the last part, we simulate the whole process discussing in the thesis using R0, Mamdani, Lukasiewicz inference engines and Hamming, Euclid, radix matching operators, and find that R0 inference engine and Hamming matching operator is much better the each other combination. And using this combination we validate the correctness of this fuzzy reasoning system.