Research And Implementation Of CBR Based Fault Diagnosis And Prediction For Power Transformers

Power transformer is an important part of power system, the operation state areimportant to the power system’s security, stability and economic operation. Therefore,its diagnosis for improving operational reliability is essential.Case-based reasoning technology is a new method of solving the field of artificialintelligence’s problem. Applied to fault diagnosis, it can improve the fault diagnosis ofpower transformers. With the power transformer systems become large, complex,systematic and automated, it makes a higher requirement for fault analysis andpositioning. Therefore, the study of fault diagnosis based on case-based reasoningtechnology is very important to improve the skills of maintenance personnel, shorten thetraining period and maintenance cost savings.In this paper, reference to "preventive test code for electrical equipment"(hereinafter referred to as "Regulations")",electrical equipment maintenance andtroubleshooting Quick Reference" and the IEC three-ratio method, based on case-basedreasoning, I designed the fault diagnosis system with diagnosis and prediction ability.For the problem of easy to misjudge the fault near threshold, I realized the uncertainreasoning by fuzzy mathematics; based on the analysis of case contents andrepresentation, I use a case representation combining object-oriented technology anddatabase technology. Introduce the concept of multi-level case database and establish ahierarchical model of failure case. I design the structure of case base in detail; it isdivided into three part containing system-level case database, fact-level case databaseand specific sub-case database. In this paper, I make a comprehensive study anddiscussion about the key technology, including case retrieval, case changes, and casestudy. Among them, case retrieval is the core of the system; the system uses the nearestneighbor hierarchical search strategy to search the aim case. This strategy combinesthree levels of the organizational structure of the case library, it reduces the number ofcase retrieval and improves the efficiency of retrieval: using multi-level hierarchicalindex for the case library and constructing the various sub-case database; in eachsub-case database, we search for the biggest similar case to the aim case. As forsimilarity computing, based on the normalized fault characteristic value, we use thenearest neighbor method. Because the algorithm’s complexity and poor objectivity ofthe case feature weight, we use rough set to get the weight and improve the weightedalgorithm. Reference to "preventive test code for electrical equipment", I give the method of failure prediction based on the tabular CUSUM.Finally, in the Window XP operating system, I develop the case-based reasoningfault diagnosis system with the help of Microsoft SQL Server2000and ADO databasetechnology, and make a authentication of the reasoning algorithms and operation of thesystem. I analyze the advantages and disadvantages of the system. In the end of thispaper, I give an outlook about the future work.