To address the limitations of traditional transformer health index (HI)

such as high dependence on data integrity and inability to assist in diagnosing latent faults

this study proposes an HI prediction method based on dynamic Bayesian network (DBN). The method first constructs a DBN for characterising the health state indicator system of power transformers according to their structural composition and operational environmental conditions. Next

the entropy weight method and fuzzy membership function are employed to determine the prior probability tables between DBN nodes

whereas the logistic regression method is used to establish conditional probability tables. Finally

existing literature data and field measurement data are utilised to validate the proposed method. The results demonstrate that: (1) the developed DBN can not only effectively calculate the current HI of transformers but also predict HI values for different future time intervals and (2) it can evaluate the overall health state of transformers and identify potential faults in individual components based on node-specific HI values. Therefore

this method provides valuable support for operation and maintenance personnel in making informed decisions regarding maintenance strategies

batch retirement and replacement plans.