IMPLEMENTATION OF COGNITIVE ARTIFICIAL INTELLIGENCE FOR DISSOLVED GAS ANALYSIS (DGA) INTERPRETATION IN TRANSFORMERS

• Main Author: Nurul Yaqin, Elko
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/55641
• Institution: Institut Teknologi Bandung

Dissolved Gas Analysis (DGA) is a method that can detect transformer faults based on the dissolved gas content in transformer oil. Some of the methods commonly used for DGA interpretation includes: Total Dissolved Combustible Gases (TDCG), Key Gases, Duval Triangle Method, Roger's Ratio Method (RRM), and Doernenburg Ratio Method (DRM). In recent years, Artificial Intelligence (AI) methods such as Fuzzy Inference Systems (FIS), Artificial Neural Networks (ANN), and Genetic Algorithm (GA) have been applied for DGA interpretation. Today, there is a new method that works based on Knowledge Growing System (KGS), it's called Cognitive Artificial Intelligence (CAI). Previous studies (Octavianus, 2018), have applied the CAI based on DRM method for the DGA interpretation. However, in his research, there were several errors in the use of ratios, data input, and "not significant" conditions in the grouping process. This paper improves that. In addition, the previous study used a dataset labeled IEC TC 10. This study retested the same data and also tested secondary data referring to 6 Paper, primary data referring to 6 Transformers data from GI PT PLN (Persero) Transmission Main Unit West Java (UITJBB), and a small amount of data with reference to 2 Paper and 2 Transformer data from the GI PT PLN (Persero) UITJBB. The results show in this study that the accuracy of CAI decreased from previous studies. The accuracy in this study decreased from 98.29% to 94.87% for CAI 0 and 95.73% for CAI 1. This study also tested other methods such as FIS with an accuracy of 94.02%, Duval Triangle 87.18%, RRM 0.68%, and DRM 55.56% for the same data. Thus, the implementation of CAI can increase DRM accuracy from 55.56% to 94.87% for CAI 0 and 95.73% for CAI 1. The results of the analysis on the IEC TC 10 dataset, secondary data (6 journals), primary data (6 PLN GI transformer) and the small amount of data show that CAI can give the highest accuracy. Thus, these methods are not limited to the amount of data and this method can be implemented on small amounts of data.