A robust technique for overvoltages classification in power transformers

A fast and reliable technique for classifying overvoltage events across power transformers terminals is introduced in this paper. Currents at the secondary terminals of power transformer during various overvoltages are used to synthesis a modal current signal. A feature vector is extracted from the selected modal signal utilizing discrete wavelet transform. Finally the extracted feature vector is used to train an artificial neural network to differentiate between various overvoltage events occurring across the transformer premises. The results of this algorithm can be used to build an online model to help assessing the condition of power transformers, thus proper condition based maintenance can be scheduled. The proposed algorithm is also fast in the sense that it can differentiate between temporary and permanent overvoltages during the early transient stage, thus eliminating the need for any time delay in the overvoltage protection devices, and the overvoltage protection philosophy can be changed to become instantaneous rather than time-delayed protection. This technique is economical and simple; it doesn't require any special arrangements as it depends on the readily available measurements. Tests were conducted to validate the proposed algorithm and showed it to be robust and generic.