Series DC Arc Detection of Photovoltaic System Based on Ensemble Learning

Aiming at the frequent occurrence of photovoltaic fires caused by electric arcs, an arc detection algorithm based on multi-feature fusion is proposed in this paper, so as to detect electric arcs in time and prevent further expansion of accident losses. To obtain more perfect data, a simulation platform of photovoltaic power generation system based on the Cassie arc model is established in Simulink, and a variety of typical cases are tested with the platform. After that, the original data is pre-processed via singular value decomposition for extracting multiple feature vectors in the time domain, frequency domain, and time-frequency domain. Then, the obtained multiple feature vectors are used to train the Extra Trees Classifier model to detect the arc accurately and quickly. The test results show that the accuracy of the trained model on the verification set is as high as 92%, but there is a problem of eigenvector redundancy. To further improve the efficiency of the arc detection model, the non-negative matrix decomposition is used to reduce the dimensionality of the data. However, the results revealed a notable decrease in model accuracy despite a reduction of 58.3% in the training data volume. Therefore, ensemble learning is adopted to improve the data mining ability of the model. The results show that the final model can achieve nearly 90% accuracy in the verification set.