The increasing integration of distributed generation systems into the grid to fulfil energy needs highlights the importance of addressing islanding cases, which pose a significant threat to grid stability. This paper focuses on developing a method for identifying islanding incidents to detect grid irregularities effectively. The methodology includes generating datasets that correspond to various grid anomalies and then utilizing a machine learning classifier for training. To support this process, a 16 kW $3 - \phi$ grid-tied photovoltaic (PV) system is modelled and operated in a Typhoon hardware-in-loop (HIL) setting to collect diverse datasets representing different grid anomaly scenarios. Following this, the data is utilized to train a k-nearest neighbor (kNN) classifier to establish the islanding detection system. The trained classifier demonstrates an accuracy of 98.58% and has the ability to classify at a rate of around 1400 observations per second.
Enhanced Islanding Detection for Grid-Connected PV Systems: A Machine Learning Approach
2024-07-31
1124569 byte
Conference paper
Electronic Resource
English
Islanding Detection of Grid-Connected Photovoltaic Systems Using Active Disturbance-Based Techniques
Springer Verlag | 2021
|