IEEE Power Electronics Magazine - December 2020 - 26

waves, the space vectors are circles of fixed diameters. For
overvoltage and overcurrent conditions, the respective circle
diameters will have proportional increase. Similarly, for
decrease in voltage and current, the respective circle diameter will decrease proportionally. Therefore, the quantitative
deviations from the rated values can be determined by measurement of the circle diameter. For system imbalance or
harmonics in the waves, the space vectors will be distorted
characteristically. For every abnormal condition of the system, the corresponding space vectors will have some distortion, the signature of which can be analyzed to identify
the fault or any abnormal condition quantitively. The space
vector patterns are displayed on optical grids with m # n
matrix, and the characteristic logic signals 1's (shaded) and
0's (unshaded) can be generated similar to optical character
recognition (OCR) system used by NNW [10].These matrices
of the characteristic logic signals are impressed at the input
of a feedforward neural network, as shown in Figure 8. The
NNW is initially trained with example data patterns by backpropagation technique to identify the corresponding fault or
any abnormal condition in the system. Some of the faults are
indicated in the figure. Based on these signals, the network
can generate warning signals, be protected by selective tripping or fault-tolerant system control can be designed.

direction so that the output data pattern matches with the de--
sired output pattern (pattern matching or pattern recognition) [9].

Neural Network based Fault Pattern Identification
of Smart Grid Subsystem
The applications of NNW in power electronics include PWM
(SHE and SVM) implementation, multi-dimensional look-up
table generation, waveform processing, delayless filtering,
estimation for distorted waves, FFT analysis, adaptive control of drives, vector control, feedback signal estimation of
drives, on-line system diagnostics, fault-tolerant control,
neuro-fuzzy based vector and DTC control, system health
monitoring, etc. We will discuss here briefly the technique of
fault identification of smart grid subsystem by neural mapping technique [4].
This application is concerned with identification of fault
and power quality problems in a smart grid subsystem (or
microgrid) as indicted in Figure 8. Three-phase voltage and
current wave signals of a power system give the footprint of
different types of faults in a system. The voltage and current
wave signals are sensed and their space vectors are computed. These are then analyzed to identify different types of
faults in the system. For example, in a healthy harmonic-free
balanced system with the rated sinusoidal voltage and current

Bias Source
+1
Hidden
Layer (b)
Link
4
Input
Layer (a)
X1

W14

Output
Layer (c)

W49
V5
9

V6
Input

Output

6
Descaling or
Denormalization

V7
X3
Scaling or
Normalization

3
Node
W38

V8
8

Weight
W8 10
Y1

Weight
Adjustment
Algorithm

Error
Z1
FIG 7 Feedforward NNW structure (3-5-2) showing backpropagation training.

IEEE POWER ELECTRONICS MAGAZINE

z December 2020

Actual
Output
Pattern

Desired
Output
Pattern

IEEE Power Electronics Magazine - December 2020

Table of Contents for the Digital Edition of IEEE Power Electronics Magazine - December 2020