Data-based affinity analysis of power transformer defects with adaptive frequent itemset mining algorithm

In worldwide power market, intelligent assessment techniques for power transformers are continuously necessitated to deal with the issues of aging apparatus, upgraded power networks, market deregulation, etc. Considering the disadvantage of subjective judgement with regard to the current assessment systems, data-based analysis approaches (e.g. big data and machine learning techniques) are intended to apply into condition assessment of power apparatus. The database related to power transformers needs multistage processes. The dataset thereof, which relates to familial defects, is considered containing valuable information for condition assessment of power transformers. Affinity analysis can help to mine this type of information. To perform the affinity analysis, this study presents a novel adaptive frequent itemset mining algorithm which is developed based on Apriori algorithm. This algorithm is proposed to meet the specific requirements of mining the familial defect information, thereby supporting an intelligent condition assessment system of power transformers. 11,266 real records from in-service power transformers are used to implement the affinity analysis. The results can quantitatively indicate the affinity among manufacturers, operating years, voltage grades of transformers and their defective parts.