Page 192 - Decision Making Applications in Modern Power Systems

Adaptive estimation and tracking of power quality disturbances Chapter | 6 155

Thus the optimal feature selection techniques have been proposed to retain
the useful features and discard the redundant features.
To extract the features of disturbed signal, there are many feature extrac-
tion methods, such as WT [6],FT [7,8],ST [9,10], short-term FT, and
Hilbert transform (HT) [11,12], which were implemented, and then extracted
features are fed to pattern classifiers, such as artificial neural network (ANN)
[13,14], probabilistic neural network (PNN) [15], fuzzy logic [16], and sup-
port vector machine (SVM) [17], to classify PQ disturbances. There are
some advantages and disadvantages for each and every technique. In the pro-
posed work, empirical mode decomposition (EMD) with HT has been imple-
mented for feature extraction of the PQ disturbances. This is an inventive
technique in which the distorted signal is decomposed into number of intrin-
sic mode functions (IMFs). We can get instantaneous frequencies as well as
amplitudes of the signal by applying HT to the IMFs. For classification pur-
pose, ANN and PNN have been developed. Further, for better classification,
efficiency SVM has been implemented.

6.2 Methodologies for efficient estimation of power quality
disturbances by using adaptive filters

In this section, various methodologies for efficient estimation of PQ events
by using adaptive filters are discussed.

6.2.1 Signal model for power quality disturbances and harmonics
estimation

Efficient signal-processing architectures are used to design PQ estimation
models. State-space modeling in real and complex forms can be implemented
using state variables to estimate sag, swell, notch, and harmonic parameters.
State-space modeling is quite popular to implement Kalman filtering algo-
rithm for PQ estimation.

6.2.1.1 Signal model for power quality disturbances estimation
PQ disturbances, such as voltage sag, swell, notch, and momentary interrup-
tion [18,19], are related to time variation of signal amplitudes and can easily
be tracked from estimated state variables. The number of state variable in a
state vector depends on the nature of the PQ disturbances. The following
mathematical analysis describes the state-space models using three state
variables.
y k is the noisy observed signal generated by a sinusoid z k in the presence
of white Gaussian noise v k .

y k 5 z k 1 v k ð6:1Þ