Enhanced Techniques for Investigating and Locating Partial Discharge Sources using Machine learning method with Self-Organizing Maps, and Back-propagation Neural Networks

Abstract

As the need for effective and efficient power systems grows, so does the need for more advanced methods to find and study partial discharge (PD) sources, which can be a sign that high-voltage equipment's protection is breaking down. This study suggests using Support Vector Machines (SVMs), Random Forests (RF), Self-Organizing Maps (SOM), and Back-propagation Neural Networks (BPNN) together to make PD cause discovery more accurate and useful. SVMs are very good at sorting complicated patterns into groups. They provide a strong framework for telling PD events apart from background noise. By mixing several decision trees, RF, which is known for its ability to learn in groups, helps make generalization better. A strong autonomous learning method called SOM helps group and show how PD sources are spread out in space. A popular artificial neural network design called BPNN is used because it can model complicated relationships and change to trends that don't follow a straight line. Putting these methods together in a way that makes the best use of their individual strengths and weaknesses creates a complete and reliable framework for PD investigations. Using the combined knowledge of these advanced machine learning algorithms, the suggested method can correctly find and spot PD sources, which will eventually make high-voltage systems work better and be more reliable. Comprehensive models and testing validations show that the proposed method works, showing that it could be used in real life for power system upkeep and diagnosis. This study is a big step forward in improving the most up-to-date methods for finding and detecting PD. It will help build stronger and longer-lasting power grids.