With the continuous increase of transformer capacity in power system

the influence due to looseness defect of transformer winding is becoming more and more serious

and it is therefore necessary to perform fault diagnosis. In view of looseness defect diagnosis of transformer winding by the use of vibration signal

the feature extraction method of transformer vibration signal based on variational mode decomposition（VMD）permutation entropy（PE）and transformer winding looseness defect diagnosis method based on support vector machine（SVM）optimized by beetle antennae search（BAS）are proposed. Firstly

one set of practical 110 kV transformer is set with different looseness states to collect vibration signals under normal and different degrees of looseness;Then

the variational mode decomposition combined with permutation entropy is used for feature extraction of looseness defects of transformer winding. After that

the support vector machine optimized by beetle antennae search algorithm is used for pattern recognition of winding looseness. Finally

the method is compared with the the traditional method. The results show that VMD decomposition effect is better than presently mainly-used empirical mode decomposition

and the permutation entropy fault feature effect is better than the time series complexity calculation index of multi-scale permutation entropy and approximate entropy. The BAS-SVM operation time and diagnosis accuracy rate are better than such optimization algorithms as grid search optimization SVM and genetic algorithm optimization SVM. The accuracy of the fault diagnosis of the transformer winding looseness tested at 60%～110% rated current by the proposed method is above 98.7%.