DC Series Arc Fault Detection Method for Photovoltaic Array Based on Random Forest

• Published in: 2024 IEEE 2nd International Conference on Control, Electronics and Computer Technology (ICCECT) pp. 138 - 143
• Main Authors: Li, Weijun, Li, Tianyou, Su, Jun
• Format: Conference Proceeding
• Language: English
• Published: IEEE 26.04.2024
• Subjects: Arc detection; Data models; DC arc; Fault detection; Fault diagnosis; Feature extraction; Inverters; Photovoltaic array; Photovoltaic systems; Random forest; Training
• DOI: 10.1109/ICCECT60629.2024.10545829

With the continuous development of the domestic photovoltaic industry and the increasing installed capacity of photovoltaics, fires occur frequently in photovoltaic power stations. The cause of fires is often caused by DC arc faults. Therefore, a reliable DC arc fault detection method for photovoltaic arrays is of great significance for ensuring the safe and reliable operation of photovoltaic power generation systems. The photovoltaic array is greatly affected by environmental factors, the output data is unstable, and the amplitude change of the fault arc data is also uncertain, which leads to the misjudgment of the DC arc detection method. Aiming at the DC series arc fault of photovoltaic array, this paper proposes a detection method of DC series arc fault of photovoltaic array based on random forest. Based on the Simlink platform, a simulation model of photovoltaic series arc fault is built. The simulation model is used to obtain the series arc fault data, and the difference between the normal string and the series arc fault string is compared. The DC series arc fault data is analyzed by statistical method and the feature quantity is extracted as the training set. The feature quantity is used to train the random forest model to establish the photovoltaic series fault diagnosis model. The string voltage data of each group collected by the inverter before and after the series arc fault of a photovoltaic project in a new energy enterprise in Xiamen are used for diagnosis. The inverter string data of the actual series arc fault verify that the method can effectively identify the DC series arc fault, and is less affected by environmental factors. The stability of the detection method is good.