A new condition assessment method for distribution transformers based on operation data and record text mining technique

In power grid, operating and maintaining engineers have recorded plenty of operation data and fault or defect texts of distribution transformers, which contain asset health information. However, less information is effectively exploited in devices condition assessment. Aiming at improving the decision-making for condition based maintenance, a practical condition assessment method based on multi-source information fusion for distribution transformers is proposed in this paper. Firstly, an assessment index system based on real time and statistical operation data of transformers is introduced. Secondly, through HMM-based text preprocessing and the machine learning of relativity, the text mining technique realized the key information extraction from transformer's fault and defect elimination record texts for condition assessment. Finally, after synthesizing the subjective initial index weight, equipment's aging and the “Short Board Effect”, the optimized condition assessment model based on multi-source information fusion for distribution transformer is established. An operating transformer example reveals that this model is effective and accurate, comparing with the other available models. It shows that with this method, the transformer's accurate health status and evolution trend could be reflected much more timely and rigorously.