ANOMALY IDENTIFICATION ON GAS-INSULATED SWITCHGEAR USING ARTIFICIAL NEURAL NETWORK

• Main Author: A. Suffaturrachman, Zuhri
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/86926
• Institution: Institut Teknologi Bandung

As a provider of electricity system in Indonesia, PT PLN (Persero) keep doing innovation to deliver reliable and efficient electricity. Amid growing electricity consumption demands and the need to improve operational efficiency, PLN has implemented gas-insulated switchgear (GIS) on its transmission system. GIS is a technology which is claimed to be maintenance-free, requiring only routine physical inspections and scheduled overhauls determined by the manufacturer. However, anomalies occur to indicate undetected deterioration that lead to system disruptions. Therefore, it is essential to identify factors causing deterioration and the potential for anomalies. In this study, a machine learning-based method is introduced to predict anomalies in GIS. Historical data including dielectric quality and characteristics of each GIS compartment, are extracted using artificial neural network (ANN) algorithm. The combination of gas quality measurements, serving as insulation, and GIS compartment characteristics is classified to produce outputs indicating normal or anomalous conditions. Using 3,391 rows of data sourced from the East Java and Bali regions, the effectiveness of the method is demonstrated through sample testing, method comparisons, and cross-validation with data from other regions. This study output identifies the best modeling hierarchy and generates valid anomaly predictions.