The invention discloses a power transformer fault deduction method, which fuses the substance element augmented extension association mining and fuzzy set pair analysis, improves the influence of thesample set imbalance on the correctness of transformer state feature association extraction, reduces the influence of subjective dependence of a constant weight calculation process on transformer fault diagnosis and state evaluation effectiveness, provides a clearer and more objective theoretical guidance for the field operators to clearly quantify the internal association between the power transformer fault symptom and the fault type, provides the reliable theoretical and technical supports for the operators to select operation and maintenance overhaul means, realize power transformer fault backtracking and improve fault positioning analysis efficiency, and provides the solid technical support for the further online real-time deduction and the rapid hidden danger positioning of the powertransformer fault based on the virtual reality technology, and guarantees the safe and stable operation of a power system.