In order to use transformer vibration signals to diagnose faults under the fluctuation of operating conditions

the vibration signals of winding in looseness and normal states under different load currents as well as the vibration signals of iron core in looseness and normal states under different no-load voltages are collected

and their characteristic quantities in time domain

frequency domain and information domain are extracted. The evaluation method of distinguishing ability of vibration signal feature quantity is established

and a linear discriminative analysis（LDA） projection algorithm is introduced to process the characteristic quantity. The support vector machine（SVM） classifier is used to verify the fault recognition effect of the feature quantity. The results show that

under the fluctuation of operating conditions

most of characteristic variables have their values overlapped in normal and fault conditions

and the ability to distinguish faults is weak due to the fluctuation of load current and no-load voltage. The method of distinguishing ability established in this paper can be adopted to screen out the characteristic quantities with strong fault distinguishing ability under the fluctuation of operating conditions. The LDA algorithm can be adopted to deal with the characteristic quantity with overlapping and filter out the influence of operation condition. When using the vibration data of rated working condition to identify the fault of transformer under the fluctuation of operating conditions

the accuracy of the characteristic quantity obtained by screening and projection processing is higher than that of the traditional feature quantity

which can diagnose transformer faults under the fluctuation of operating conditions by using the vibration signal. The method proposed in this paper can provide a solution for transformer fault diagnosis by using vibration signals under the fluctuation of operating conditions

such as three-phase unbalance

variable power factor

harmonic and other changing conditions.