A HOLISTIC INTEGRATION OF CONVENTIONAL AND MACHINE LEARNING TECHNIQUES TO ENHANCE THE ANALYSIS OF POWER TRANSFORMER HEALTH INDEX CONSIDERING DATA UNAVAILABILITY

• Main Author: Redha Arsya, Nanda
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/81771
• Institution: Institut Teknologi Bandung

The health index is an essential tool to evaluate the condition of power transformers. The health index is a method of determining the condition of equipment by considering several affecting parameters and is described with a certain score. The health index is also used by asset managers in carrying out maintenance planning and even to estimate the remaining life of transformers. Generally, there are two methodologies to evaluate the health index of power transformers: conventional methods and machine learning-based approaches. The main challenge in employing the health index is the lack of supporting data, which hinders its ability to accurately reflect the actual health condition of the transformers. While several studies on estimating the transformers health index using machine learning techniques can be found in the literatures, not much attention was given to the cumulative impact of multiple data unavailability on the calculation’s accuracy. This thesis methodology consists of 2 main steps. The first step focuses on health index analysis using complete data whereas the second step considers data unavailability. In the first step, this thesis presents a comprehensive analysis of health index predictions using 11 machine learning methods: specifically, 6 regression methods and 5 classification methods, which are validated with conventional or weighting and scoring method. Furthermore, in the second step, the thesis presents some strategies to overcome the unavailability of multiple data. Six methods which consist of 4 non-machine learning and 2 best-accuracy machine learning methods from the first step are presented and evaluated across 15 missing data scenarios to assess their effectiveness. Finally, the thesis presents an economic analysis to highlight the profitability associated with predicting missing data compared with the certainty level. The novelty of this research includes the more varied input data used, the combination of missing data based on experience and conditions in the field, the type of machine learning used, and then equipped with economic analysis. Moreover, the thesis provides a comprehensive view for asset owners, especially PLN to make informed decisions in case of data unavailability.