Given the increasing deployment of photovoltaic (PV) power generation systems worldwide, ensuring the safe and reliable operation of these systems is of paramount importance. Among the various safety concerns, DC series arc faults have emerged as a significant threat to the stability and performance of PV systems. These faults, often the result of disconnections or substandard connections within the PV circuit, can potentially result in fires and system failures if not identified promptly. The objective of this study is to accurately identify DC series arc faults in PV systems, which would greatly enhance the safety measures and operational reliability of these systems. In addition, the study investigates the characteristics of these faults under different experimental conditions, focusing on the impact of electrode shapes and current intensity on fault characteristics. Finally, a robust model for detecting such faults is developed based on extracted time-domain and frequency-domain features of fault signals.

A series of experiments were conducted to simulate DC series arc faults in PV systems. The experiments utilized electrodes of various shapes to study the effect of electrode design on arc characteristics and were performed under various current intensities to simulate real-world operating conditions of PV systems. For each fault condition, the voltage-current characteristics of the arcs were recorded to gain an initial understanding of the fault dynamics. The data were then analyzed using both time-domain and frequency-domain methods. Specifically, the Fast Fourier Transform (FFT) was used to convert time-domain fault signals into the frequency domain for further feature extraction. Mathematical statistics techniques were also applied to analyze the spectral energy distribution across different frequency bands, with a particular focus on the 0—50 kHz frequency range, which has been identified as critical for distinguishing different arc fault signatures. Based on these extracted features, a hybrid model combining FFT and deep learning techniques was developed. This model integrates the FFT with a 1D Convolutional Neural Network (1DCNN) and a Long Short-Term Memory (LSTM) network. This architecture identifies arc fault types based on their frequency-domain characteristics.

The experimental results revealed that the frequency characteristics of DC series arc faults highly depend on the shape of the electrodes and the current intensity. Specifically, faults generated by electrodes of different shapes exhibited distinct features in the frequency domain, with significant variations observed in the spectral energy distribution within the 0—50 kHz frequency range. These results imply that electrode shape plays a significant role in determining the frequency signature of arc faults, which can be used for fault identification purposes. The FFT-based feature extraction technique successfully isolated the most relevant frequency components indicative of arc faults. The FFT-1DCNN-LSTM model was then trained using these features and achieved an accuracy rate of 99.87% in correctly classifying arc faults generated by different electrode shapes. This result demonstrates the model's robustness and potential for real-world applications, as it can effectively differentiate various fault scenarios in PV systems. Furthermore, the model's high accuracy indicates its potential for the early detection of arc faults, which can significantly improve the safety and reliability of PV systems.

This study introduces an effective and innovative method for detecting DC series arc faults in PV systems. By analyzing fault characteristics under different electrode shapes and current intensities, the study provides valuable insights into the role of these parameters in detection. The FFT-based feature extraction method, combined with an advanced deep learning model (FFT-1DCNN-LSTM), achieves exceptional performance in accurately identifying arc faults. The model's high classification accuracy highlights its potential for practical deployment in real PV systems, where early arc fault detection is critical for preventing potential hazards such as fires. The study's findings contribute to ongoing efforts to enhance system safety and provide a reliable technical framework for arc fault detection and mitigation. Future research may focus on refining the model by including additional fault scenarios and exploring its scalability to larger, more complex PV system configurations.

Neural network models have shown strong performance in fault arc detection. However, these models often relied on fragmented, single-modality features—such as time-domain, frequency-domain, or time-frequency representations of one-dimensional time series—making it difficult to capture transient high-frequency oscillations at the microsecond level. This resulted in the loss of critical detail, limiting the ability to predict arc occurrence precursors and weakening emergency response. To address this issue, this paper proposed a fault arc detection method based on multi-modality feature fusion.

An experimental circuit simulating parallel breakdown arc induced by overcurrent was built, with 75 effective working conditions designed and over 100, 000 data points collected per scenario. Based on the typical characteristics of pre-fault and fault waveforms, 5, 456 one-dimensional time-series signal samples were constructed. Five conversion methods—Markov transition field (MTF), recurrence plot (RP), Gramian angular field (GAF), short-time Fourier transform (STFT), and continuous wavelet transform (CWT)—were used to convert the transient current signals into time-frequency-space feature maps (TFS-Maps). Each mapping method involved multiple parameters, and their effectiveness in feature extraction varied, necessitating the selection of optimal settings. For MTF, parameters such as the number of bins, interval division strategy, and color mapping scheme were chosen. For GAF, the visualization results of the summation and difference angular fields were compared. For STFT, window lengths of 16, 32, and 64 were tested. For CWT, the wavelet basis, scale, center frequency, and bandwidth were optimized. For RP, the signal's standard deviation was used. The resulting multi-modality dataset—containing original signals and their corresponding TFS-Maps—was split into training, validation, and test sets in a 7∶2∶1 ratio. A gated recurrent unit (GRU) was used to model sequence dependencies in the original signals. A Swin transformer integrated with the convolutional block attention module (Swin Transformer-CBAM) was applied to highlight key regions within the TFS-Maps. The outputs from GRU and Swin Transformer-CBAM were fused via cross-modality concatenation to perform arc detection. Accuracy, precision, recall, F1-score, and comprehensive evaluation visualization graphs were used to assess the algorithm's performance.

The experimental results showed that (1) among various TFS-Maps, GADF achieved the highest performance, with 98.07% accuracy, 97.52% F1-score, and 98.01% recall; 2) Swin Transformer-CBAM outperformed the convolutional neural network, with improvements of 0.37% in accuracy, 0.17% in F1-score, and an increase in recall from 97.67% to 98.01%; and (3) the confusion matrix indicated very few misclassifications, with over 98% agreement between predicted and actual labels.

Time-frequency imaging enhanced sensitivity to high-frequency transient features. The attention mechanism effectively captured fault arc features by focusing on critical frequency bands and time-domain segments. The proposed detection method met expectations, improved detection efficiency, and provided a more reliable technical solution for identifying parallel breakdown arcs induced by overcurrent.