Development of machine learning-based algorithm to determine the condition in transformer oil
One very popular and useful electric device in daily life is a transformer, and it is one of the greatest components of the power network system. The main fault of these transformers can purpose considerable damage. This not only disrupts other functions of the power supply, rather than caused...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English English English |
| Published: |
2021
|
| Subjects: | |
| Online Access: | http://eprints.uthm.edu.my/6955/1/24p%20HUSSEIN%20HASAN%20MOHSEN%20AL-KATHERI.pdf http://eprints.uthm.edu.my/6955/2/HUSSEIN%20HASAN%20MOHSEN%20AL-KATHERI%20COPYRIGHT%20DECLARATION.pdf http://eprints.uthm.edu.my/6955/3/HUSSEIN%20HASAN%20MOHSEN%20AL-KATHERI%20WATERMARK.pdf http://eprints.uthm.edu.my/6955/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Universiti Tun Hussein Onn Malaysia |
| Language: | English English English |
| Summary: | One very popular and useful electric device in daily life is a transformer, and it is one
of the greatest components of the power network system. The main fault of these
transformers can purpose considerable damage. This not only disrupts other functions
of the power supply, rather than caused very large losses. The interpretation of
dissolved gas analysis (DGA) is used to detect incipient faults in transformer oil. This
paper aims to develop a model for taking into consideration the results obtained from
DGA to investigate the condition of transformer oil fault. Machine Learning (ML)
algorithm have been utilized to detect the fault more accurate. Classification learning
app used to train DGA data divided into three categories fault, Not determined (N/D)
and stray gassing. Three different types of ML algorithm have achieved high accuracy
of 93.0%, 95.4% and 97.7% support vector machine (SVM), Naïve Bayes algorithm
(NB), K-nearest neighbour (KNN) respectively. Graphical User Interface (GUI) has
overall the system by testing and verified with many different user data and performed
a correct classification. |
|---|