A fault classification method for medium voltage networks with a high penetration of photovoltaic systems using artificial neural networks
Conference
Hoang, TT, Duong, QB, Tran, QT et al. (2020). A fault classification method for medium voltage networks with a high penetration of photovoltaic systems using artificial neural networks
. IEEE POWER AND ENERGY SOCIETY GENERAL MEETING 2010, 2020-August 10.1109/PESGM41954.2020.9281856
Hoang, TT, Duong, QB, Tran, QT et al. (2020). A fault classification method for medium voltage networks with a high penetration of photovoltaic systems using artificial neural networks
. IEEE POWER AND ENERGY SOCIETY GENERAL MEETING 2010, 2020-August 10.1109/PESGM41954.2020.9281856
With the rapid advancement of power electronic technologies and the reduction of photovoltaic cell price, the share of solar energy in the total power production has been booming recently. On the one hand, the increase in the amount of power delivered by solar energy can be beneficial in many economic and environmental aspects. On the other hand, this can cause various technical challenges to network operators. One of these issues is related to classifying faults located in distribution networks with high penetration of photovoltaic systems. Although many studies have paid significant attention to developing new algorithms applicable for a more active today distribution networks, there is still space for other improvements. Hence, after reviewing state-of-the-art researches, this paper was intended to develop a fault classification that is based on artificial neural networks. In particular, a technique so-called Multiplayer Perceptron Classifier was selected for the proposed algorithm. First, the authors generated a data set for the study by modeling and simulating a real distribution network with practical parameters provided by a local utility in the environment software PowerFactory/DigSILENT. Multiple fault scenarios were simulated. Second, a part of the generated data collection was used for network learning. Finally, the performance of the proposed methodology was demonstrated via testing on the remaining number of generated data.
