A combinatorial approach based on wavelet transform and Hidden Markov Models in differential relaying of power transformers

This paper proposed a new classification method based on hidden Markov models (HMM) to discriminate between magnetizing inrush current and internal faults in transformers. HMM is a pattern recognizer which can classify signals based on waveform and characteristics. Training data-set is achieved by application of k-means clustering approach. First some principle characteristic of the signals is extracted to reduce the computational time of HMM training procedure and enhance discrimination accuracy, and then two HMM blocks is trained for these two kinds of signals. This procedure makes the scheme insensitive to irrelevant disturbances such as CT saturation. Based on the proposed algorithm a high speed differential relaying could be performed in a quarter of a cycle. Small computation requirement is the advantage of this method which can be applied for real time applications. The suitable performance of this method is demonstrated by simulation of different faults and switching conditions on a power transformer. Since the discrimination method is done with probabilistic characteristics of signals without application of any deterministic index, more reliable and accurate classification is achieved. It provides a high operating sensitivity for internal faults and remains stable for inrush currents of the power transformers.