A DC Arc Fault Diagnosis Method for Photovoltaic Systems Based on Convolutional Neural Networks and Attention Mechanisms
Annotatsiya
Due to the strong concealment, obvious non-stationarity and high susceptibility to electromagnetic interference of DC arc faults in photovoltaic systems, existing diagnostic methods are difficult to effectively balance diagnostic accuracy, real-time response speed and anti-interference capability. Therefore, this paper proposes a CNN-SE-Net fusion diagnostic method for DC arc faults. This method first optimizes the original input signal through wavelet threshold denoising and sample enhancement preprocessing to reduce noise interference and improve data quality. Then it uses a lightweight CNN to automatically extract deep time-domain and frequency-domain features from the processed signal. By embedding the SE-Net attention mechanism, the model can adaptively strengthen the weight of key feature channels and suppress useless information. The verification based on the AI-generated multi-condition simulation dataset shows that the diagnostic accuracy of the proposed method reaches 99.3%, and the average diagnostic time is only 45 ms. The comprehensive performance is significantly superior to traditional threshold methods, SVM classifiers and single CNN models. This research provides a stable and reliable technical solution for online precise diagnosis of DC arc faults in photovoltaic systems, and has important practical significance and engineering application value.
Hali tarjima qilinmagan