The data mining method is limited to be applied in the distribution network fault diagnosis because of the unbalanced fault sample problem. Aiming at this problem, combining the modified complete ensemble empirical mode decomposition with adaptive noise (MCEEMDAN) and conditional generative adversarial network (CGAN), a fault identification method for the distribution network was proposed. At first, the CEEMDAN was modified by the partial mean of multi-scale permutation entropy. The MCEEMDAN may decompose the electric signal into a series of intrinsic mode functions. The raw time-domain signal can be transformed into the two-dimensional gray-level image by the pseudo-color coding of the intrinsic mode functions. Then, the fault gray image can be labeled and put into CGAN to generate a large number of new samples to achieve data augmentation. In order to improve the quality of the generated samples, the least square loss function is introduced into the original CGAN network to make the generated samples close to the raw samples. Finally, the convolutional neural network (CNN) is used to mine the fault features autonomously. The Softmax classifier is used to achieve distribution network fault classification. The experiments show that the proposed method can effectively learn the distribution characteristics of the original sample. Furthermore, the fault recognition accuracy can be effectively improved. The proposed method has good stability, fast convergence speed, and high precision, and it can effectively complete the fault identification of the distribution network.

As the non-solid-earthed network fails, the zero-sequence current of each line is highly non-stationary, and the noise component is serious. This paper proposes a fault line selection method based on modified complementary ensemble empirical mode decomposition (MCEEMD) and the Duffing system. Here, based on generalized composite multiscale permutation entropy (GCMPE) and support vector machine (SVM) for signal randomness detection, the complementary ensemble empirical mode decomposition is modified. The MCEEMD algorithm has good adaptability, and it can restrain the modal aliasing of empirical mode decomposition (EMD) at a certain level. The Duffing system is highly sensitive when the frequency of the external force signal is the same as that of the internal force signal. For automatically identifying chaotic characteristics, by using the texture features of the phase diagram, the method can quickly obtain the numerical criterion of the chaotic nature. Firstly, the zero-sequence current is decomposed into a series of intrinsic mode functions (IMF) to complete the first noise-reduction. Then an optimized smooth denoising model is established to select optimal IMF for signal reconstruction, which can complete the second noise-reduction. Finally, the reconstructed signal is put into the Duffing system. The trisection symmetry phase estimation is used to determine the relative phase of the detection signal. The faulty line in the non-solid-earthed network is selected with the diagram outputted by the Duffing system.

In the case of a single‐phase grounding fault in the distribution network, the transient zero‐sequence current (TZSC) tends to be non‐linear and non‐stationary. The faulty line selection is relatively difficult. The distributed power access further brings many difficulties to faulty line selection. This work proposes a novel method of faulty line selection using spatial image generation and deep learning. At first, the optimal smooth denoising model can be used to smooth the zero‐sequence current for each line, reducing the external environment electromagnetic interference. Then, the treated zero‐sequence current is mapped into the colorful floral hoop image by using symmetrized Hilbert transform pattern (SHTP). The SHTP transforms the one‐dimensional time domain signal into the two‐dimensional space domain image, enhancing invisible information and obtaining more abundant feature information. Finally, the deep features of the SHTP floral hoop image are extracted by means of deep learning method. In order to improve the faulty line selection universality, a mixed sample library containing three different topologies is established, including the 10 kV radial distribution network, IEEE‐13 node model, IEEE‐34 node model and StarSim platform. The comparisons show that the proposed method has a more noticeable visualization effect on fault features, higher classification precision rate, and better anti‐noise performance.

In order to extract the fault characteristics of the distribution power system automatically, a new method for fault identification based on parameter optimized variational mode decomposition (VMD) and convolutional neural network (CNN) is proposed. First, the VMD is used to decompose electric signals of the main transformer, including the current of the low‐voltage inlet line, bus voltage, and so on. The obtained intrinsic mode functions should be arranged in sequence from low to high according to the central frequency. The VMD can obtain the time‐frequency matrices. In order to decide the decomposition layers of the VMD adaptively, the partial mean of multi‐scale permutation entropy is used to optimize it. Then, the time‐frequency matrixes can be converted into pixel matrices by pseudo‐colour coding. The pixel matrices which are appropriately compressed can be used as input of CNN. The CNN can autonomously extract the eigenvectors of fault. Finally, the classification algorithms of support vector machine (SVM), naive Bayesian classifier (NBC), extreme learning machine (ELM), and random forest (RF) are used to train and test the feature vectors extracted by CNN to verify the effectiveness of the method. The experiments show that the proposed method has excellent stability, fast convergence speed, and high precision. The technique can effectively complete the fault identification of the distribution network.

This research proposes a face detection algorithm named LighterFace, which is aimed at enhancing detection speed to meet the demands of real-time community applications. Two pre-trained convolutional neural networks are combined, namely Cross Stage Partial Network (CSPNet), and ShuffleNetv2. Connecting the optimized network with Global Attention Mechanism (GAMAttention) extends the model to compensate for the accuracy loss caused by optimizing the network structure. Additionally, the learning rate of the detection model is dynamically updated using the cosine annealing method, which enhances the convergence speed of the model during training. This paper analyzes the training of the LighterFace model on the WiderFace dataset and a custom community dataset, aiming to classify faces in real-life community settings. Compared to the mainstream YOLOv5 model, LighterFace demonstrates a significant reduction in computational demands by 85.4% while achieving a 66.3% increase in detection speed and attaining a 90.6% accuracy in face detection. It is worth noting that LighterFace generates high-quality cropped face images, providing valuable inputs for subsequent face recognition models such as DeepID. Additionally, the LighterFace model is specifically designed to run on edge devices with lower computational capabilities. Its real-time performance on a Raspberry Pi 3B+ validates the results.