Feature-Analyse zur Klassifikation von Teilentladungen bei Gleichspannungsbeanspruchung mit Machine Learning

For a safe supply of electrical energy, it is essential to monitor the electrical equipment of the power supply network and to determine its condition. One of the most important of these diagnostic methods is the partial discharge measurement. This measuring method and the interpretation of the measured data is internationally recognized and established in the field of AC voltage technology. Due to current changes in the field of energy supply, a trend away from the classic transmission with alternating voltage towards transmission with direct voltage can be observed. These changes makes it necessary to implement the established measuring methods also for direct voltage. However, many of the interpretation methods cannot be adopted directly, which is why new methods have to be developed. The interpretation can be automated with the help of machine learning, which has many advantages, especially for continuous monitoring. Such systems are already available for AC voltage but not yet for DC voltage. The most important steps of machine learning are feature extraction and analysis. Features are extracted from the raw data, which are then used to classify and interpret the measurement data. These features must be significant and the different partial discharge defects must be distinguishable from each other with these features.