The internal winding of converter transformer in operation is subjected to the combined action of AC and DC voltages

which leads to the complicated vibration response of vibration source. In order to pay attention to the mechanical state variation of converter transformer under long term vibration

an on-line monitoring system of vibration is firstly developed based on the vibration mechanism and characteristic of the winding of converter transformer in this paper. The long term live monitoring of the vibration signal of one converter transformer at site is performed

the proportion of wavelet energy of vibration signal of converter transformer at a converter station is extracted. At last

the mechanical state intelligent diagnosis technology which is suitable for converter transformer at site is proposed with the time series consisting of such vibration characteristic as the proportion of wavelet energy as the data sample and in combination with time series. The analysis results show that the the vibration on-line monitoring system at site can be able to perform long term monitoring of converter transformer and the characteristic of the proportion of wavelet energy can

to a certain extent

reflect the operation and internal mechanical mechanical state of converter transformer. It is concluded that the higher similarity between the predicted and expected value is reached through the treatment of the proportion of wavelet energy by the time series prediction model

which provides a more timely and effective technical way for state assessment and potential defect diagnosis of the converter transformer.