Skip to main content

Table 11 Summary of classification techniques

From: A systematic review of real-time detection and classification of power quality disturbances

Type of technique

Advantages

Disadvantages

Application

Multiple PQDs

Sags

Notches

Refs

%

Refs

%

Refs

%

Handcrafted

Simple application

Physical interpretation

Fast operation

Inaccuracy

Expert knowledge needed

Time-consuming design

[33]

1

[137, 148, 153, 163, 172]

9

–

0

Probabilistic

Simple application

Sound theoretical foundation

Physical interpretation

Difficult modeling and implementation

[34, 36, 40, 41, 44, 70, 102]

7

[133, 135, 139, 141]

7

–

0

Shallow ANN

Flexibility

Detailed knowledge of the phenomena is not required

Able to solve nonlinear functions

Time-consuming training

Extensive data for training

Handcrafted feature extraction

[37,38,39, 44, 46,47,48, 52, 53, 56, 57, 62, 64, 65, 71, 79, 81,82,83,84,85, 89, 90, 93,94,95, 100, 102, 106,107,108, 113, 123]

33

[143, 147, 168, 169, 176,177,178]

13

–

0

Deep ANN

Flexibility

Detailed knowledge of phenomena is not required

Able to solve nonlinear functions

Automatic feature extraction

The very high computational burden for training

Extensive computations during operation hinder real-time applications

[103, 111, 116, 119, 121, 124, 125, 129,130,131]

10

[158, 159, 164, 165, 167, 171, 175, 179, 180, 182, 183]

20

–

0

Decision tree

Very simple application

Efficiency for real-time application

Robustness to outliers and noisy data

Complex decision trees can be difficult to understand

Complexity increases exponentially with the size of the tree

[42, 49, 66, 70, 72, 75, 77, 79, 80, 85, 87, 89, 91, 96,97,98,99, 101, 102, 104,105,106, 109, 112, 114, 117, 122, 126, 128]

30

[140, 151, 152, 185]

7

[200]

4

SVM

Sound theoretical foundation

Only a dozen examples for training are required

Computational inefficiency

Low scalability

[47, 50, 51, 59, 60, 63, 69, 74, 76, 86, 88, 92, 98, 101, 102, 110, 114, 115, 127]

19

[149, 150, 168, 170, 177, 185]

11

[211]

4

k-NN

Very simple application

Ease understanding

Reduced accuracy

Sensitive to the choice of k

[55, 65, 102, 114]

4

[171, 185]

4

–

0

Fuzzy logic

Better representation of expert knowledge

Physical interpretation of events

Robustness to noisy data

Reduced accuracy

No systematic

It depends on human knowledge

A lot of testing is necessary

[43, 45, 48, 54, 61, 64, 67, 68, 96, 120]

10

[132]

2

[189]

4