Disturbance Classification Using Hidden

Therefore, Hidden Markov Models (HMMs) are introduced in [10] to classify between fast transient phenomena by using wavelet packet coefficients. Since the authors [10] used wavelet packet coefficients as the HMMs observation sequences, HMMs have not been able to classify slow phenomena such as sag because a very long observation sequence is required for slow disturbances. To overcome this limitation, power quality disturbance classification utilizing Discrete Density Hidden Markov Models combined with Vector Quantization (VQ) is proposed in this paper. The proposed method can classify between fast and slow phenomena and it is computationally efficient. To achieve this goal, a sliding window of either Wavelet Transform (WT) or Fourier transforms (FT) is applied to the recorded voltage signal. Then, the VQ is utilized to convert the continuous observation (FFT or DWT vectors) into a discrete observation sequence. The labels produced by VQ step are utilized as an input observation to HMMs. This results in a short observation sequence, which can be used to distinguish between slow and fast phenomena efficiently. In the training stage, an HMM is constructed for each phenomenon under consideration. In the classification (testing) stage, these models are used to produce a ranked list of the most probable power quality disturbances associated with the signal under consideration. Moreover, with the inherent scalability of the HMMs, models representing new phenomena can be constructed without the need to retrain the already existing models. The architecture and methodology of the proposed system is discussed in Section II. The WT and FT that are used in feature extraction are presented in Section III. Section IV details the implementation of the proposed HMM. The VQ algorithm is described in Section V. The supporting results are given in Section VI, and Section VII concludes the paper.