Notes

Power quality (PQ) issue has attained considerable attention in the last decade due to increasing utilization of highly sensitive electronics equipment and/or microprocessor based controlled loads, such as in computers, communication and consumer electronics etc. moreover, they can appear simultaneously as there are multiple sources of different disturbances.
Any disturbance manifested
in voltage, current or frequency deviations that result in failure or misoperation
of electronic equipment can be termed as an electric power
quality problem.
Identification and classification of various power quality disturbances are keys to ensuring high-quality electrical energy.
PQ disturbances cover a broad frequency range with significantly different magnitude variations and can be non-stationary and transitory in nature, thus, accurate and advanced tools and techniques are required to identify and classify these events/disturbances. This paper proposes a detection and classification technique for several power quality disturbances……..
This forms a feature vector that is fed to the input nodes of probabilistic neural network which classifies the power quality disturbances. To validate the efficiency and preciseness of the proposed method the simulation results are analyzed.
The power quality of the electric power has become an important issue for the electric
utilities and their customers. In order to improve the quality of power, electric utilities
continuously monitor power delivered at customer sites. Thus automatic classification of
distribution line disturbances is highly desirable. The detection and classification of the
power quality (PQ) disturbances in power systems are important tasks in monitoring and
protection of power system network. Most of the disturbances are non-stationary and
transitory in nature hence it requires advanced tools and techniques for the analysis of PQ
disturbances. In this work a hybrid technique is used for characterizing PQ disturbances using
wavelet transform and fuzzy logic. A no of PQ events are generated and decomposed using
wavelet decomposition algorithm of wavelet transform for accurate detection of disturbances.
It is also observed that when the PQ disturbances are contaminated with noise the detection
becomes difficult and the feature vectors to be extracted will contain a high percentage of
noise which may degrade the classification accuracy. Hence a Wavelet based de-noising
technique is proposed in this work before feature extraction process. Two very distinct
features common to all PQ disturbances like Energy and Total Harmonic Distortion (THD)
are extracted using discrete wavelet transform and is fed as inputs to the fuzzy expert system
for accurate detection and classification of various PQ disturbances. The fuzzy expert system
not only classifies the PQ disturbances but also indicates whether the disturbance is pure or
contains harmonics. A neural network based Power Quality Disturbance (PQD) detection
system is also modeled implementing Multilayer Feedforward Neural Network (MFNN).

The evaluation of electrical power systems during recent
periods increases the interest in Power Quality (PQ). The increasing
utilization of non-linear and sensitive loads, for instance the power
supplies in computers, communication and consumer electronics leads to
gradual deterioration of the Power Quality. Any disturbance manifested
in voltage, current or frequency deviations that result in failure or misoperation
of electronic equipment can be termed as an electric power
quality problem.
Power quality investigations often require monitoring to
identify the exact problem. Power quality monitoring is the process of
gathering, analyzing and interpreting raw measurement data into useful
information. The process of gathering data is usually carried out by
continuous measurement of voltage and current using power monitoring
recorders over an extended period. But, these recorders lack the ability to
distinguish between events. The user has to categorize the collected
events for further analysis manually. There is usually a high volume of
data to be processed and classified. This makes it very tedious and time
iv
consuming to interpret the data. It is highly desirable if the analysis is
automated.
This work is focused on the development of signal processing
techniques and Artificial Neural Networks (ANN) for the automatic
detection and classification of disturbance waveforms.
Signal processing techniques are applied to analyze the
changes in current and voltage waveforms and to extract various features
from the disturbance signals. In this thesis, Wavelet Transform based
energy distribution and S-Transform contour based techniques are
applied to analyze and detect the disturbances in time-frequency domain.
Amplitude-frequency estimation methods namely Hilbert Transform and
Hilbert-Huang Transform are used for detecting the disturbances.
Suitable signal processing techniques are proposed to detect the
disturbances under noisy environment.
The ANN-based approaches are used for the automatic
classification of disturbance waveforms. A set of feed forward neural
networks trained by back propagation algorithm is developed to classify
the disturbances. With feed forward neural networks, any continuous
function can be approximated to within an arbitrary accuracy by
v
carefully choosing the parameters in the network provided the network
structure is sufficiently large. But the shortcoming of the network is that,
it takes a long time for training. Also, the outliers cannot be detected by
this network. This thesis proposes Radial Basis Function (RBF) network
for disturbance classification. The RBF networks take less time for
training and distance-based activation function used in the hidden nodes
gives the ability to detect the outliers during classification.
Dimensionality reduction is necessary to make the neural
network approach suitable for the automatic disturbance recognition.
While developing the neural network, by selecting only the relevant
attributes of the data as input features, the higher performance is
expected with smaller computational effort.
In this work, the features of the disturbance waveforms are
extracted by applying standard statistical techniques to the transformed
signal and a novel entropy based feature selection technique is used to
select the most useful features of the network. These features are given
as input to the ANN for further classification. During the training stage,
the network captures the relation between the input features and the
output class. After training, the network is tested with the test data to
vi