Partial Discharge Classification by using Artificial Neural Network

Tan, Tee Min (2017) Partial Discharge Classification by using Artificial Neural Network. Final Year Project (Bachelor), Tunku Abdul Rahman University College.

Text Tan Tee Min.pdf Restricted to Registered users only Download (1MB)

Abstract

Partial discharge (PD) measurement is a vital tool for assessing the insulation quality. Different types of cable defect will lead to different partial discharge pattern. The consequences of partial discharge that occurs on high voltage instruments and cables can be very severe and lead to eventual complete breakdown. Therefore, it is vital to have an early detection and measurement for partial discharge. In this project, five types of partial discharge data collected from five defected cable joints. Each type of data contains hundreds sets of sub-data with massive amount of charge (Q) and the relatively phase. In order to process huge numbers of data, some method had to apply for features extraction and selection. Linear Discriminant Analysis (LDA) has been chosen as the tool to extract features for this project as it performs well on dimensional reduction. Other than that, Artificial Neural Network (ANN) was used to carry out the classification process to create and train an artificial intelligence (AI) network. Configuration of creating a new ANN is important that will directly affect the output result. Thus, there are many different combinations of configuration which has to be statistically investigated to figure out the best setting for ANN to achieve the highest classification accuracy. Multiple times of configuration attempt is the most consuming time among other stages of this project due to big number of possibilities for ANN configuration settings. In this project, the best classification accuracy achieved at the end is 95.28% with configuration of 40%-training rate, 30%-validating rate, 30%-testing rate, 10 hidden neurons and 25 number loop.