A transformer condition assessment framework based on data mining

The framework of an assessment system on transformers´ condition is proposed in this paper through mainly using data mining techniques. Moreover, a warehouse is used to collect transformers´ testing data, and a multi-agent system is used to design the framework of the software. The present framework is open and flexible, so the objective system is easy to be developed and maintained. The system can support transformers´ condition-based maintenance to reduce electric utility´s cost. The condition of a transformer depends on its design, present and historical data relating to its installation environment, load amounts, being switched number and so on. Usually the off-line testing results, operational data, fault records and weather conditions have been stored in different systems, so finding an effective method to utilize all this information for condition assessment is difficult. Therefore, a data warehouse has been used to integrate all of the above data, and some data mining techniques have been used to find the pattern and trend of the condition of a transformer. Then whether it is healthy can be determined. In order to make the system open and flexible, open agent architecture (OAA) is employed to compose the multi-agent system. Seven application agents are designed to evaluate transformers´ conditions synthetically. The Grey correlation method, grey theory prediction model GM(1,1), Bayesian network classifier and Bayesian network are employed in the agents.