Ice accretion on overhead catenary systems presents a critical hazard to the operational safety of high-speed railways. This paper proposes an active anti-icing methodology utilizing circulating Joule heat generated via phase-shifting traction transformers. Specifically, by leveraging the voltage phase-shifting principle of traction transformers, a novel circulating current generation scheme is formulated. A collaborative control strategy, regulating both the transformer's phase angle and output voltage, is developed to create the specific electrical conditions required for Joule heating without disrupting normal operations. Furthermore, to address the engineering challenge of accurately determining the convective heat transfer coefficient (hc), a GPR-LSBoost estimation model is established. This model overcomes the limitations of traditional empirical formulas in complex environments. Finally, a weather-driven heat transfer control strategy is proposed. By taking real-time meteorological parameters as inputs, the system dynamically estimates hc via the GPR-LSBoost model and calculates the minimum required anti-icing current using the heat balance equation. This current serves as a reference to precisely regulate the transformer's turns ratio and phase angle. Results demonstrate that the GPR-LSBoost model achieves high accuracy with a coefficient of determination (R2) of 0.954. The proposed method effectively maintains the contact wire temperature above 0 degrees C, providing a stable and reliable solution for suppressing catenary icing.

A transformer is the core equipment in a power system. Winding faults are the main causes of sudden accidents, and it is important to obtain the winding status in a timely manner. This article studies the online frequency response analysis (FRAs) for transformer winding faults using the injection of high-frequency signals on the load side. First, the principle of the online method is proposed, and the feasibility is verified by the field experiments of a 500 kV transformer. Next, the online experimental platform considering the peripheral circuits is constructed, and the sensitivity differences of the different injection modes for the winding faults changes (type, degree, and location) are systematically studied. Finally, the categorical boosting (CatBoost) model is optimized using the whale migration algorithm (WMA) to construct a diagnostic model for online monitoring of transformer winding faults. The results show that the proposed high-frequency signal load-side injection method effectively achieves online FRAs of transformer winding faults; through monotonicity, linearity and sensitivity metrics, it was determined that different injection methods exhibit varying degrees of sensitivity to faults; the fault diagnosis model based on WMA-CatBoost achieves 100% accuracy in identifying the fault type, and maintains the accuracy of identifying the fault degree and location above 97% and 98%.

The outer sheath serves as the primary barrier for insulation and protection in a cable. The temperature detection method provides a direct reflection of cable faults, accurate understanding of the cable temperature distribution is a fundamental prerequisite for diagnosing cable issues. This paper first establishes and validates an electromagnetic-thermal coupled finite element (FEM) model of a 110 kV high-voltage (HV) cable installed in an air-filled cable trench. After validating the arc-shaped rectangular minimum bounding box (MBB) model against representative deformation geometries, the study systematically investigates the effects of deformation parameters (angle, depth, and length) on cable temperature distribution. Finally, based on the cable thermal model, the study explores the influence of deformation on the outer sheath's thermal resistance and proposes a correction to the IEC 60287 outer sheath thermal resistance formula. The results indicate that changes in the depth of deformation have the most significant impact on the minimum temperature of the deformed outer sheath. Furthermore, a noticeable linear relationship is observed between deformation angle and outer sheath thermal resistance. The modified outer sheath thermal resistance calculation formula exhibits a fitting error of within +/- 2.5%, enabling accurate calculation of deformed outer sheath thermal resistance for condition monitoring applications in cable trench installations.

Accurately determining the contamination level (CL) of onboard insulators is crucial for ensuring the safe and stable operation of high-speed trains (HSTs). This study proposes an evaluation method using leakage current (LC) monitoring and optimized symmetrized dot pattern (OSDP) images to accurately and automatically detect the CL of onboard insulators in HSTs. First, insulator samples and an LC measurement platform were prepared in the laboratory to obtain time-series LC signals under five different CLs. Then, the encoding principle of the OSDP was proposed, which involves integrating effective multimodal components derived from the decomposition of LC signals using the ensemble empirical mode decomposition (EEMD) algorithm into a polar coordinate system. This method effectively emphasizes the multiscale characteristics of LC signals, significantly enhancing data visualization. Finally, the OSDP images were integrated with four typical deep convolutional network (DCN) models to evaluate the CL of the insulators. The results indicate that OSDP images combined with the ResNet-18 model achieve the highest performance in CL evaluation, with a recognition accuracy of 98.5%, which is 3.8% higher than the accuracy of traditional SDP images. In addition, an online monitoring device for LC signals was developed to facilitate the application of the proposed technique to actual onboard insulators in HSTs.

SiR cold-shrink cable terminations are highly prone to defects during field assembly, which significantly compromises their service safety. Thus, this paper proposes a new detection method for cable termination assembly defects based on the measurement of microwave reflection curve. First, a method for 2D images conversion and fusion of microwave reflection curves, termed Gramian Angular Difference Fields-Adaptive Structured Low-Rank-Fusion Network (GADF-ASLR-FusionNet), is proposed, which reduces noise interference but also achieves complementary fusion of features. Second, an Inverted Residual-Coordinate Attention-ConvNeXt (IR-CA-ConvNext) model is constructed. It adopts IR blocks reduce computational complexity. CA is used to enhance key failure modes. ConvNeXt blocks are then stacked to progressively refine local textures into stronger global representations. The Asymmetric Loss function optimizes classification of hard-to-classify samples. Finally, a cable termination assembly defect detection method based on the above two approaches is proposed. Ablation study and comparison experiments demonstrate that the fused image outperforms a single GADF image in feature representation. The proposed model achieves a recognition accuracy of 98.33%, with an F1-score of 0.9835 and a multiclass AUC value of 0.9918. Its performance surpasses other models, confirming its superior performance and robustness in identifying cable termination assembly defects.

The 10 kV power cable constitutes a critical component in urban power grids, with fault diagnosis primarily comprising fault location and insulation damage assessment. Time-frequency domain reflectometry (TFDR) is widely employed for cable fault diagnosis. However, cross-term artifacts and inadequate feature extraction significantly impact the accuracy of both location and assessment processes. This paper presents a comprehensive approach to address these challenges. First, we optimize the Wigner-Ville distribution (WVD) using adaptive threshold segmentation to mitigate cross-term artifacts during TFDR data processing. Second, we implement a propagation-based amplitude compensation algorithm to eliminate distance-dependent attenuation, followed by extracting features from time-domain excitation responses and energy spectrograms using a three-stage progressive feature selection strategy. Finally, we propose a Transformer-gated multi-feature network (TG-MFN) framework that integrates convolutional neural networks, bidirectional long short-term memory (BiLSTM) networks, and Transformer-style adaptive attention gating mechanisms. This integration enables precise fault severity assessment through multi-scale feature extraction and dynamic feature importance adjustment. The results show that the proposed method achieves: 1) effective elimination of cross-term interference, with insulation damage fault localization errors controlled within 1.44%; and 2) precise severity assessment, with the maximum assessment error not exceeding two severity levels, enabling precise evaluation of insulation damage severity.

Transformer winding defects directly affect the safe and reliable power supply. This article conducts research on the online diagnosis of winding defects. Field trial for online data acquisition is first carried out, and the feasibility of acquiring frequency response analysis (FRA) data in real time is verified. Further, this article proposes a winding defect diagnosis method, which enhances spectral feature extraction by designing an attention clustering pooling (ACP) module and employs cuckoo catfish optimizer (CCO) to optimize the hyperparameters of the classifier. First, convolutional neural network is employed to extract multiscale local features from FRA signals. Subsequently, the ACP module employs multihead self-attention to capture cross-frequency global dependencies. By integrating attention-guided clustering downsampling and weighted feature aggregation, it achieves adaptive amplification and efficient compression of key frequency band information, thereby constructing a spectrum feature representation with high discriminative. On this basis, the CCO collaboratively optimizes the hyperparameters of the local cascade ensemble classifier to achieve multidimensional joint identification of winding defects. Finally, through online verification of axial and radial displacement defects on an experimental platform equipped with peripheral circuits, the proposed method achieved an overall diagnostic accuracy rate of 98.5%, fully validating its effectiveness.

Winding defect diagnosis is a key research focus in transformer detection. Deep learning has been widely applied in this field, but its performance is limited by insufficient defect samples and class imbalance. To tackle these problems, this paper proposes a MultiImageTime data generation model based on one-dimensional(1-D) Frequency Response Analysis(FRA) data, which enhances the generative model's ability to learn frequency response characteristics, thereby achieving high-discrimination data augmentation. Meanwhile, an adaptive Gramian Angular Field (AGAF) method is introduced to image the 1-D FRA data, adaptively focusing on critical defect frequency bands to reduce defect information loss. On this basis, a MultiImageTime-AGAF-ConvNeXtV2 framework for winding defect diagnosis is constructed, which fully exploits ConvNeXtV2’s efficient image feature extraction capability to achieve high-precision defect identification. Experimental results show that the framework performs excellently even with small samples, reaching a classification accuracy of 95.12%.

Reliable pantograph-catenary current collection is essential for safe railway operation, while catenary icing severely threatens operational safety and performance. However, accurate icing perception remains challenging due to complex thermodynamic mechanisms and limited monitoring data. To address this problem, this study proposes a physics-informed hybrid deep learning framework, CNN-SimAM-PINN, for estimating icing distribution along the catenary. A location-based distributed parametric traction load model is first developed to quantify the effect of traction-load-induced Joule heating on icing evolution. A dual physics-informed strategy is then introduced by reconstructing temporal load features at the input level and embedding physical constraints in the training loss. The parameter-free SimAM attention mechanism is further used to enhance feature extraction from micrometeorological and traction load data without increasing model complexity. Experimental results show that the proposed method achieves high estimation accuracy and robustness, with an R2 of 0.9953, outperforming conventional data-driven models. The framework provides a cost-effective approach for icing perception in data-scarce scenarios and can support icing prediction and early warning, as well as anti-icing and deicing maintenance.

Autotransformer is one of the core electromechanical equipment in high-speed railway, of which winding mechanical deformation is the primary cause of operation failure. Frequency response analysis (FRA) is an effective method for detecting winding mechanical deformation. For improving generalization capability of detection method based on FRA, this paper studies a discrepancy analysis based on deep learning and result interpretation method for FRA curves. Firstly, a differential signal extraction network is developed to capture the discrepancy information between healthy and faulty curves, improving the effective utilization of FRA curve data. Subsequently, a combination of Multi Kernel Maximum Mean Discrepancy (MK-MMD) and adversarial training is incorporated into the network to guide the feature extractor in learning domain invariant features. Finally, the differential adversarial transfer network (DATN) diagnostic algorithm is proposed to achieve identification of winding mechanical deformation faults. Experimental results demonstrate that the DATN algorithm achieves an accuracy of 96% in the source domain and 90% in the target domain.

Winding deformation poses a significant threat to power transformers and may further lead to serious insulation failure. Oscillating wave analysis (OWA) is a winding detection method with the advantages of high testing efficiency and strong anti-interference capability. To extract more useful information from the transformer winding oscillating wave response curve, this paper proposes an OWA-FMD signal processing method based on traditional Feature Mode Decomposition (FMD). To verify the effectiveness of OWA-FMD and improve the accuracy of the fault diagnosis model, the Hippopotamus Optimization algorithm (HO) is used to optimize CatBoost, and an HO-CatBoost model for transformer winding fault diagnosis is established. Model training and result analysis are then carried out using the data obtained from the experimental platform. The experimental results show that OWA-FMD can effectively improve the fault characterization ability of oscillating wave signals. The diagnostic accuracy of the data processed by OWA-FMD reaches 97.42 %, while that of the unprocessed data is 67.05 %. In addition, after HO optimization, the diagnostic accuracy of CatBoost increases from 84.57 % to 97.42 % using the same OWA-FMD processed data. Finally, the proposed methods are applied to an actual engineering case, and the diagnosis results are consistent with the transformer dismantling inspection results.

Winding failures are recognized as one of the primary causes of transformer accidents, making effective monitoring of winding conditions crucial. A study on autotransformer (AT) winding faults diagnosis is conducted through the following procedure. Firstly, an experimental platform is established to simulate typical single and combined winding faults in autotransformers, through which frequency responses under various fault conditions are tested. Subsequently, a fast vector matching method is employed to fit transfer functions of winding systems under normal and faulty states, from which zero point distribution diagrams in polar coordinates are derived. Then, the gray level difference statistical (GLDS) features and gray-gradient co-occurrence matrix (GGCM) features are extracted from the zero point distribution diagrams, and the particle swarm optimization (PSO)-random forest (RF) algorithm is combined to realize the classification of faulty windings and fault types. Finally, the proposed method is validated using actual autotransformer fault cases. The results show that the zero point distributions in polar coordinates obtained by fast vector fitting can capture the subtle differences in the original frequency response curves by combining amplitude-frequency and phase-frequency information. Compared with optimization algorithms such as cuckoo search and genetic algorithm, the PSO-RF algorithm maintains an accuracy rate consistently exceeding 93% in identifying winding faults and fault types of autotransformers. The analysis results of the proposed method are consistent with the tank lifting inspection results in real autotransformer fault cases.

Winding fault detection is critical for maintaining the operational stability of power transformers. This study presents an approach for fault type identification and spatial localization utilizing frequency response analysis (FRA) data. First, an Integrated Gramian Angular Field (IGAF) coding scheme is proposed to transform one-dimensional FRA sequences into compact two-dimensional visual representations. By integrating unique information regions from the Difference Field and Summation Field, this approach eliminates symmetric data redundancy while enhancing feature information from fault-induced spectral deviations. Subsequently, an activation-free "star operation" mechanism processes IGAF features, mapping low-dimensional textures into a high-dimensional implicit feature space. This mechanism effectively captures fine-grained fault patterns, avoiding the information loss common in clipping activation functions. Finally, the StarNet-S1 architecture serves as the diagnostic backbone, performing efficient feature modeling while accomplishing both fault classification and localization. Experimental results validate the effectiveness of the proposed framework, achieving a classification accuracy of 96.05% and a localization accuracy of 96.73%. Requiring only 2.68 million parameters and delivering a single-sample inference latency of just 1.03 ms, the proposed framework enables efficient real-time diagnostics on resource-constrained hardware.

Accurate and effective diagnosis of transformer winding conditions is crucial for ensuring their safe and reliable operation. This article proposes a transformer-winding fault detection method based on oscillating wave analysis (OWA). First, a transformer-winding fault simulation platform was established to obtain OWA curves. Second, we propose an oscillation wave alignment method based on shape context, which can systematically extract Morpho-offset features to more comprehensively quantify the offsets of curves. The analysis results indicate that the Morpho-offset feature has a good clustering ability for fault states. Then, to enhance the extreme gradient boosting (XGBoost) algorithm's ability to identify winding faults, a transformer-winding fault diagnosis algorithm based on an XGBoost classifier optimized by the crown porcupine algorithm is proposed. Finally, the Morpho-offset features and diagnostic algorithm are integrated to identify the type, location, and degree of winding faults. The results demonstrate that the proposed method achieves an accuracy of 99.45% in identifying winding faults, with an 11.89% improvement over the unoptimized XGBoost.

On-board insulators are key components of the electric multiple units (EMUs) power system, and accurate evaluation of their contamination level (CL) is of great significance in preventing contamination flashover accidents and ensuring the safe operation of equipment. Current methods rely on manual experience, which is inefficient and prone to errors. This article proposes a novel method for evaluating the CL of on-board insulators in EMUs, aimed at improving detection efficiency and adaptability to complex field environments. The proposed approach utilizes the Gramian angular field-visual saliency map-weighted least square optimization (GAF-VSM-WLSO) framework combined with a depthwise separable convolution-convolutional block attention module-ResNet18 (DSC-CBAM-ResNet18) network. Initially, the leakage current (LC) signal is transformed into a 2-D color fusion GAF image through the above-mentioned composite framework, which can capture both the global structure and local details. A composite model is then constructed, enhancing feature extraction by introducing the DSC module to reduce computational complexity and the CBAM attention mechanism to highlight key features. The model is optimized with the focal loss (FL) function to improve classification accuracy for difficult samples. Ablation study and comparison experiments demonstrate that the fusion image outperforms normal 2-D images in feature representation. The proposed model achieves 98.67% recognition accuracy, with a $0.9867 F1 score and a 0.9968 multiclass AUC value, outperforming other models and confirming its superior performance and robustness in CL evaluation of insulators.

Vehicle cable terminals (VCTs) are crucial for ensuring a continuous and reliable power supply to high-speed trains (HSTs). Ultrasonic testing is a promising technique for diagnosing internal defects in VCTs. However, the non-intuitive features and limited distinguishability of time-domain ultrasonic signals (USs) hinder the accurate identification of defect types. To address this issue, we first proposed an improved Gramian angle field (IGAF) method to convert US data into 2-D images. This approach integrates the effective features of the traditional Gramian angle summation field (GASF) and Gramian angle difference field (GADF) into a single image, thereby enhancing data representation and eliminating redundant information caused by pattern symmetry in conventional methods. We then developed a model named CBAM-EfficientNetB0, which integrates convolutional block attention modules (CBAMs) to enhance feature extraction and representation capabilities. Finally, we proposed a combined method that utilizes IGAF images generated from US data alongside the CBAM-EfficientNetB0 model for the accurate diagnosis of internal defects in VCTs. The results indicate that the proposed combined method achieves a defect diagnosis accuracy of 95.8%, which is 5.6% and 4.3% higher than the GASF and GADF methods, respectively, and surpasses the original EfficientNetB0 by 1.3%. Furthermore, we developed a portable US measurement device, facilitating the application of the proposed technique for intelligent detection in practical scenarios.

Transformers, serving as critical components in power systems, are predominantly affected by winding faults that compromise their operational safety and reliability. Frequency Response Analysis (FRA) has emerged as the prevailing methodology for the status assessment of transformer windings in contemporary power engineering practice. To mitigate the accuracy limitations of single-classifier approaches in winding status assessment, this paper proposes a differentiated M-training classification algorithm based on White Shark Optimization (WSO). The principal contributions are threefold: First, building upon the fundamental principles of the M-training algorithm, we establish a classification model incorporating diversified classifiers. For each base classifier, a parameter optimization method leveraging WSO is developed to enhance diagnostic precision. Second, an experimental platform for transformer fault simulation is constructed, capable of replicating various fault types with programmable severity levels. Through controlled experiments, frequency response curves and associated characteristic parameters are systematically acquired under diverse winding statuses. Finally, the model undergoes comprehensive training and validation using experimental datasets, and the model is verified and analyzed by the actual transformer test results. The experimental findings demonstrate that implementing WSO for base classifier optimization enhances the M-training algorithm’s diagnostic precision by 8.92% in fault-type identification and 8.17% in severity-level recognition. The proposed differentiated M-training architecture achieves classification accuracies of 98.33% for fault-type discrimination and 97.17% for severity quantification, representing statistically significant improvements over standalone classifiers.

Given the short daily maintenance time for electric multiple units (EMUs), it is essential to improve the diagnosis efficiency of the shielded cables in EMUs. Broadband impedance spectroscopy (BIS) is a promising technique for diagnosing these cables. However, the localization algorithms based on Fourier transform (FT) currently utilized in BIS have drawbacks such as the positioning resolution being strongly affected by the frequency bandwidth and test noise. Additionally, fault type identification in BIS faces challenges, including insufficient feature parameters, identification accuracy affected by tester experience, and demanding test requirements. In this article, we propose an efficient localization and type identification method to overcome the above problems. We employ the total least squares estimation of signal parameters via rotational invariance technique (TLS-ESPRIT) to improve positioning resolution and reduce positioning interference, and we extract image features from 3D-BIS data to distinguish between hard, low-resistance, and high-resistance faults. Furthermore, we propose a resonance counts calculation algorithm to further distinguish between the open-circuit and short-circuit faults. We compare positioning results for our method with traditional localization methods to demonstrate the utility of the approach, and we present laboratory data and field experimental results for EMUs to validate the proposed method.

Defect diagnosis for vehicle cable terminals (VCTs) is essential for the safe and stable operation of electric multiple units (EMUs). Because online monitoring of partial discharge (PD) in VCT is not feasible, there is a scarcity of field PD data and an imbalanced data distribution. To address this issue, we first compared and analyzed the effectiveness of three typical improved generative adversarial networks (GANs) in generating PD data. Then, we developed an improved ResNet18 model that incorporates convolutional block attention modules (CBAMs) to improve feature extraction and representation capabilities. Finally, we proposed a combined approach that utilizes data augmentation alongside an improved ResNet model to accurately diagnose internal defects in VCT. The results indicate that the deep convolutional GAN (DCGAN) achieves the highest performance regarding image structure, texture, and diversity of generated samples, effectively reducing the cost of on-site PD data acquisition. Furthermore, the improved ResNet18 model outperforms the traditional ResNet18 model across various performance metrics, including accuracy, recall, ${F}1$ -score, and G-mean, on both small sample and augmented datasets, surpassing other prominent deep learning (DL) models such as AlexNet and VGG-19. The proposed combined diagnostic method increases the accuracy of defect recognition for internal defects in VCT from 86.8% to 98.5%, effectively addressing the problems of low accuracy and limited generalization in diagnostic models caused by the scarcity of field PD data.