https://orcid.org/0009-0006-6623-2519
Retraction
The PLOS One Editors retract this article [1] due to concerns about potential manipulation of the publication process. These concerns call into question the validity and provenance of the reported results. We regret that the issues were not identified prior to the article’s publication.
All authors either did not respond directly or could not be reached.
8 Apr 2026: The PLOS One Editors (2026) Retraction: Fast fault diagnosis of smart grid equipment based on deep neural network model based on knowledge graph. PLOS ONE 21(4): e0345825. https://doi.org/10.1371/journal.pone.0345825 View retraction
Figures
Abstract
The smart grid is on the basis of physical grid, introducing all kinds of advanced communications technology and form a new type of power grid. It can not only meet the demand of users and realize the optimal allocation of resources, but also improve the safety, economy and reliability of power supply, it has become a major trend in the future development of electric power industry. But on the other hand, the complex network architecture of smart grid and the application of various high-tech technologies have also greatly increased the probability of equipment failure and the difficulty of fault diagnosis, and timely discovery and diagnosis of problems in the operation of smart grid equipment has become a key measure to ensure the safety of power grid operation. From the current point of view, the existing smart grid equipment fault diagnosis technology has problems that the application program is more complex, and the fault diagnosis rate is generally not high, which greatly affects the efficiency of smart grid maintenance. Therefore, Based on this, this paper adopts the multimodal semantic model of deep learning and knowledge graph, and on the basis of the original target detection network YOLOv4 architecture, introduces knowledge graph to unify the characterization and storage of the input multimodal information, and innovatively combines the YOLOv4 target detection algorithm with the knowledge graph to establish a smart grid equipment fault diagnosis model. Experiments show that compared with the existing fault detection algorithms, the YOLOv4 algorithm constructed in this paper is more accurate, faster and easier to operate.
Citation: Jun L, Chenliang Z (2025) RETRACTED: Fast fault diagnosis of smart grid equipment based on deep neural network model based on knowledge graph. PLoS ONE 20(2): e0315143. https://doi.org/10.1371/journal.pone.0315143
Editor: Lei Zhang, Beijing Institute of Technology, CHINA
Received: August 19, 2024; Accepted: November 20, 2024; Published: February 14, 2025
Copyright: © 2025 Jun, Chenliang. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All relevant data are within the manuscript and its Supporting Information files.
Funding: The author(s) received no specific funding for this work.
Competing interests: The authors have declared that no competing interests exist.
1. Introduction
In recent years, social development has put forward higher requirements for the stability and intelligence of power grid operation. In this development background, smart grid came into being. The operation and maintenance process of smart grid equipment involves data information such as equipment, faults, real-time monitoring and maintenance, and these data are often interrelated. The traditional manual inspection and diagnosis work mode has been unable to fully meet the real-time and efficient smart grid equipment diagnosis requirements due to high labor cost, single diagnosis method and un-centralized data management [1]. On the other hand, with the increasing number of electric vehicles, the corresponding demand for charging piles is also increasing, electric vehicle charging, especially disorderly charging, will increase the peak load of the grid, exacerbate the peak and valley difference of the load, leading to frequency fluctuations in the power system, and exacerbate the frequency of grid equipment failure [2]. Therefore, how to effectively manage and apply the data is of great significance to improve the working efficiency of the operation and maintenance of smart grid equipment. In essence, this process is to extract and reuse fault knowledge [3]. At this stage, in the process of fault management of smart grid equipment, people usually only pay attention to the classification of faults and fault processing process, the recorded fault data is scattered, and the lack of more systematic data management, so a more effective method is needed to realize the management and application of fault data. In response to this demand, knowledge graph technology is introduced to model the fault knowledge of electric power equipment, which provides a new means of acquiring, storing, organizing, managing, updating and displaying knowledge information with complex relationships by taking advantage of the great advantages of knowledge graph technology in constructing knowledge networks and presenting knowledge associations [4].
Knowledge graph has been widely used in intelligent search, in-depth question answering and social networking, as well as in finance, medical treatment and other fields. Linming and Guangtao [5] learned from the systematic ideas of RCM and PHM, took the product information model as the information organization mode, and built the knowledge graph by using the correlation between fault tree, system alarm and fault mode. Additionally, Lele et al. [6] proposed the establishment of knowledge graph for aircraft maintenance and maintenance, which can be used to query recommended steps of aircraft maintenance, effectively improving the storage and search ability of maintenance information. Later, Hao et al. [7] applied knowledge graph to the field of fault diagnosis and established a general knowledge ontology model in the field of fault analysis to realize dynamic fault information retrieval and improve the utilization rate and sharing degree of fault diagnosis knowledge by adopting ontology modeling method in semantic environment. The effective combination of knowledge graph technology and neural network algorithm can provide an efficient decision-making method for intelligent inspection. Ma et al. [8] proposed a joint learning framework that combines the generalization of interpretable rules from the knowledge graph with the construction of rule-guided neural recommendation models to facilitate the 2 modules to complement each other in generating valid and interpretable recommendations. Nie and Sun [9] proposes a text-enhanced knowledge graph embedding model to perform inferences about entities, relationships, and texts, which is well suited not only to modeling the interactions of their underlying features, but also to modeling paths between entities in the graph. In the research of power grid equipment fault detection, Liao et al. [10] proposed a semi-automatic knowledge graph construction technique and applied it to power grid equipment system. Based on the above organization and summary of the literature on knowledge map technology, it can be seen that the current application of knowledge map technology to the analysis of historical failure data can effectively solve problems. In larger, more complex structure in the electric power industry, considering the power equipment and intelligent information management system of the need of operations, a set of equipment information, fault information, real-time monitoring and maintenance of the information in the integration of fault knowledge base has become the important foundation. Irshad [11] studied several different classical image segmentation algorithms such as fuzzy C-mean segmentation, and by comparing their respective advantages and disadvantages in targeting infrared thermal images of power equipment, it was concluded that the region growing method is indeed superior in terms of infrared image segmentation. Dutta et al. [12] investigated a method to monitor the thermal state of power equipment, which was done by first investigating HSV color model, then using gradient-based edge detection to identify the equipment images, and finally using the improved OTSU algorithm to segment the images. Followed by recognizing the equipment images using gradient based edge detection, and finally segmenting the images using the improved OTSU algorithm, and proved the superiority of the algorithm in monitoring the thermal state characteristics of the equipment by performing experimental simulations on different infrared thermal images of the power equipment.
Judging from the literature review above, Knowledge graph-based deep learning neural networks (Long Short-Term Memory [LSTM], YOLOv4, and Convolutional Neural Networks [CNN]) for rapid fault diagnosis of smart grid devices encompasses a wide range of essential functions. Knowledge graphs are employed to depict the interconnected relationships among different components of the smart grid system, thereby facilitating the representation and integration of data. Subsequently, deep learning neural networks are utilized to integrate and process this diverse data, which includes sensor readings, equipment specifications, historical fault data, and maintenance logs. Deep learning models excel in the acquisition of relevant features from raw data through feature learning and extraction. In the context of smart grid fault diagnosis, these models can extract intricate patterns and relationships from the data represented in the knowledge graph. For example, they can identify correlations between specific equipment behaviors and particular fault types [13]. Furthermore, the primary purpose of the deep learning neural network is to classify and predict faults in smart grid equipment. By training on historical data that is annotated with fault labels, the model learns to associate certain patterns in the data with specific types of faults. As a result, it becomes proficient in accurately classifying and predicting faults in real-time or near real-time. This capability enables proactive maintenance measures and minimizes downtime in the smart grid system. In addition, the knowledge graph functions as a fundamental component for the integration of various sources of information that are pertinent to fault diagnosis. Deep learning models possess the capability to effectively merge information from different segments of the knowledge graph [14]. This allows them to leverage the extensive contextual information that is embedded within the graph structure, thereby enhancing the accuracy of fault diagnosis. Once implemented, the deep learning model continuously monitors the influx of data streams originating from the smart grid equipment. In the event that anomalies or patterns indicative of potential faults are detected, the model promptly generates real-time alerts or notifications. This enables operators or maintenance personnel to intervene in a timely manner. Moreover, it is important to note that smart grid systems are dynamic in nature and undergo evolution over time. Deep learning models possess the ability to adapt to changing operating conditions and evolving fault patterns by continuously learning from new data. This adaptability ensures that the fault diagnosis system remains effective and accurate, even as the smart grid environment evolves. By fulfilling these crucial roles, deep learning neural networks that are integrated with knowledge graphs play a pivotal role in facilitating swift and precise fault diagnosis in smart grid equipment. Ultimately, this contributes to the overall reliability and efficiency of the grid infrastructure [15].
In order to solve the problems of insufficient intuitiveness and low efficiency of existing power grid fault diagnosis, this paper uses knowledge graph technology and deep neural network algorithm to build a fault diagnosis model of smart grid equipment. The knowledge graph is applied to identify faults, and the YOLOv4 algorithm is used to improve the accuracy and efficiency of power grid equipment fault diagnosis. The power grid fault diagnosis method given in this paper has strong intuitiveness, and can improve the fault diagnosis efficiency and help maintenance personnel to make analysis and decision. To facilitate the understanding of the smart grid devices and fault semantics, this paper analysis substation as an example, this paper proposes a neural network model based on depth and knowledge map more modal semantic model. Model of overall frame as shown in Fig 1 below.
The research innovations and contributions of this paper are: in order to solve the problems of insufficient intuition and low efficiency of the existing grid fault diagnosis, this paper applies the knowledge graph technology and combines the deep neural network algorithm to construct a smart grid equipment fault diagnosis model. Through the application of knowledge graph to determine the faults, and the YOLOv4 algorithm is utilized to improve the accuracy and efficiency of grid equipment fault diagnosis. The grid fault diagnosis method given in this paper is highly intuitive and can provide help to improve the efficiency of fault diagnosis and help maintenance personnel to analyze and make decisions.
The research structure of this paper is as follows: Firstly, the general framework of multimodal semantic model based on deep neural network model and knowledge graph is given, and the implementation process of the model is introduced [11]. Then, the construction of knowledge graph is emphasized, the power equipment fault knowledge framework is built, the fault knowledge model is established, the concept, relationship and attribute are explained, the fault knowledge acquisition process is given, and the substation equipment fault knowledge graph is built on this basis. Then combined with the fast and accurate requirements of smart grid equipment fault diagnosis, the CSPn module is introduced to optimize the YOLOv4 model. Finally, the performance of the smart grid fault diagnosis model based on deep neural network model and knowledge graph is analyzed by means of experimental evaluation. The corresponding conclusions are drawn.
2. Knowledge graph construction
The smart grid equipment fault knowledge map constructed in this paper belongs to a typical domain knowledge map with clearer concepts and relationship patterns, which can help realize the continuous accumulation of knowledge on power equipment faults, and the key knowledge elements extracted from fault data collected by various means can be effectively integrated through standardized representations and relationships to continuously expand the scope of domain knowledge; and it can also provide basic elements for the intelligent application and visual display of knowledge. Intelligent application and visualization of knowledge provides the basic elements.
2.1. Fault knowledge framework
At present, the operation and maintenance of power equipment is developing in the direction of information and intelligence, and the fault knowledge of power equipment has an important application in fault analysis and fault handling, as well as the formulation and optimization of maintenance strategies and safeguard measures [16]. The fault knowledge of power equipment comes from the design knowledge, principle knowledge and experience knowledge, from which the fault analysis knowledge and fault handling knowledge are extracted. In addition, important fault knowledge exists in historical fault cases, which can be collected through fault reporting, and the processing of fault data can be utilized to obtain more macroscopic fault patterns and continuously update the fault knowledge. In order to better manage and apply, the fault knowledge needs to be stored in the specific fault knowledge base, and directly provide support for the fault diagnosis, maintenance strategy and safeguard measures of power equipment. The knowledge requirement of the power equipment fault knowledge architecture, that is, the knowledge graph under construction, is shown in Fig 2.
As can be seen from this figure, design knowledge, experience knowledge and other data are first processed through the fault knowledge analysis, and then imported into the model data center. In the actual application process, the system can match the grid fault types by retrieving the data information in the data center, and then record the detection result data, and finally the management decision module can update the data information through the regular update of the data information, and transmit it to the relevant knowledge base, forming a complete closed loop.
2.2. Fault knowledge modeling
Knowledge map is the basic unit of "entity-relationship—entity", so the core of knowledge map construction work is extracted from large amounts of unstructured data entities and entity relationship. Entity is the atomic element of information in knowledge map, it is abstract and form the concept of specific kind of entity within the territory [17]. Each entity may have different attributes. Conceptual design is about identifying the concepts and their attributes included in the domain based on the domain knowledge requirements. Relationship is to reflect the connection that exists between two entities. For a particular field, the relationship between different types of entities may have different knowledge modeling to determine the type of entities and relationships in the knowledge network, namely the concept and the relational schema design.
Therefore, the knowledge model in this paper includes 3 tuples: concept, relationship and attribute. The specific meanings are shown in Tables 1–3. The concept in the fault knowledge graph refers to different entity categories required by the knowledge base, such as: equipment of different levels, fault mode data, maintenance data and monitoring data, etc [18]. Model relationship refers to the connection between each entity, for example: fault impact relationship, fault cause relationship and subordinate relationship, etc.; attribute is the entity has its own attributes, which can be distinguished from the characteristics of other entities, different entities have different attributes, for example: the fault mode has attributes including the number, name, and type, etc., and each attribute of the entity has its own determined attribute value.
2.3. Knowledge acquisition
The key knowledge elements of power equipment fault extraction need to identify the key knowledge elements associated with a given fault from the fault description text, such as fault equipment, fault mode, fault cause, fault impact and maintenance measures [19]. The fault mode and its related entity information and relationship information are extracted from the FMEA data table obtained by Failure Modes and Effects Analysis (FMEA) carried out on some systems, mainly from the fault mode library, and historical fault cases, as shown in Fig 3. The process of knowledge acquisition also involves the standardized processing of data, and it is necessary to formulate a set of perfect standards to standardize the data type, data format and data content description.
As can be seen from this figure, the fault knowledge is mainly derived from the failure mode library, the FMEA data table obtained from the FMEA (Failure Modes and Effects Analysis) carried out on part of the system, and the fault modes and their related entity and relationship information extracted from the historical fault cases [20]. The data such as failure reports and historical failure cases will be sent to the data center after knowledge extraction, and then these data will be standardized and feature extraction will be carried out by using association rules to finally build up the equipment failure knowledge base.
2.4. Knowledge graph construction
In order to achieve the representation and storage of substation operation scene information, the difficulty lies in how to encode the information knowledge of data including image, video and text in a unified form. Since the data of substation equipment fault knowledge graph is domain specific, the top-down knowledge graph construction technology is used to complete the construction of the ontology model and data layer [21]. Collected from the scene of the transformer substation and the Internet network in the text, images, video and audio data, choose the information to join to the knowledge base with high degree of confidence, describe the concept of substation equipment and their mutual relations. Then the graph database is used to store the data in a networked mode, to realize the construction of knowledge graph, so that the knowledge graph query can be associated like a human being, and become the key to enhance the intelligence of the model and automatic fault diagnosis. The process of the construction of the substation equipment fault knowledge map as shown in Fig 4.
As can be seen from this figure, its specific construction process is:
- ① Text, image, video, audio and other data collected from the substation scene and the interconnection network;
- ② Organizing and analyzing these data information and defining their attribute relationships and levels;
- ③ Categorizing and organizing the defined and completed data;
- ④ Further processing of the data to eliminate ambiguities between entity relationships;
- ⑤ Extracting the entities and relationships of the data information;
- ⑥ Selecting the information with higher confidence level and adding it to the knowledge base for describing the concepts of substation equipment and their interrelationships;
- ⑦ Use the graph database to store the data in a networked mode to realize the construction of the knowledge graph.
The construction of ontology rules is an important task in the construction of substation equipment fault knowledge graph [22]. The construction of substation equipment ontology includes the steps such as the definition of concept category hierarchy and concept attribute relation. The concept classification of the ontology is mainly to classify and define the types of equipment faults. According to its internal elements, it can be divided into the following categories: equipment class, component class, fault cause class, suggestion and measure class. The definition of concept attribute relations can make the ontology more detailed, and then form a classification hierarchy system with a good structure [23]. Each fault class is composed of equipment, components, fault causes, suggestions and measures, and can be abstractly described into an entity and entity state form, thus forming a conceptual framework of substation equipment with accurate definition and clear structure.
3. YOLOv4 detection model
YOLO detection network is a regression function of the detection algorithm, the biggest advantage is the detection of fast, has been widely used in many areas of detection. YOLOv4 in YOLOv3 based on the integration of a variety of deep learning techniques commonly used to achieve a good balance between the detection speed and accuracy. YOLOv4’s birth, heralding the deep learning algorithms in the field of target detection and made further development. YOLOv4 has a strong real-time performance, and its main components are: Backbone, which is used to extract features from the input data; Neck, which fuses the extracted features of various sizes; and Detection Head, which is usually used to complete the target classification as well as regression operations [24]. Combined with the fast and accurate requirements for the fault diagnosis of smart grid equipment, this paper optimizes for the YOLOv4 model and introduces the CSPn module, which is used to deeply estimate and match the CBM module of the YOLOv4 model. The model structure is shown in Fig 5 below.
YOLOv4 is mainly composed of five basic components. (1)CBM: This module is the smallest unit in the whole network, and its components include Conv, BN and Mish. (2)CBL: Similar to CBM, but the modified activation function is LeakyReLU. (3) ResUnit: Residual module in YOLOv4, using ResUnit you can build a deeper network. (4)CSPn: Thanks to the excellent overall structure of CSPNet, CSPn is composed of n residual modules and convolutional layers for Concat operations [20]. (5)SPP: The maximum pooling method of 13×13, 9×9, 5×5, 1×1 is adopted to realize the fusion of features of different sizes. When the input image size is 416×416, the feature map of 13×13 size is obtained after 5 CSP modules [25].
Compared to previous network models, YOLOv4 applies the best optimization methods in the CNN field on MAP (mean average precision) and FPS (frames per second) to achieve the optimal balance.YOLOv4 loss function by CIOU (completely intersection calculation over and set) error as the regression prediction error [26]. Loss function can be divided into the return box Lloc prediction error, error in lci processing, incredible Lconf and classification error. The expression of the loss function is:
(1)
The regression box measurement error is:
(2)
(3)
(4)
In the formula, IOU (A,B) is the intersection ratio of predicted box A and true box B; ρ2(Actr,Bctr) is the Euclidean distance between the center point of the predicted box and the real box; m is a true box that contains both predict box and diagonal distance minimum enclosed area; wgt is the width of the actual boxx; hgt is height of the actual box; w is Predict wide box; h is Predict high box.
The confidence error is calculated by:
(5)
In the formula, is the predicted boundary box contains the target;
is the predicted boundary box does not contain targets;
is the predicted confidence;
is actual confidence; λnobj is self-set parameter value [21].
The classification error is calculated by:
(6)
In the formula,c is the type of target to be detected; is the actual probability that the target in the cell belongs to category c;
is prediction probability.
4. Trial evaluation
4.1. Test conditions
This experiment simulation software and hardware environment is configured to: Windows 10 operating system, Pycharm programming environment, NVDIA TITAN RTX graphics, Pytorch1.6, Python3.6, CUDA10.1 and secondary effects [27]. Part of the network parameters in the experiment are described in Table 4 below.
4.2. Establishment data set
This paper chooses to use the electrical class 110 dataset as the data source, which contains more than 20,000 sets of data, involving 17 parameters, including three-phase current, voltage, ambient temperature, etc., which can be used for power grid fault diagnosis and application research of machine science. In this data set, 1436 pieces of equipment fault data are randomly selected, and then the acquired image data is manually screened [28, 29]. After screening, 824 high-quality images are selected as the initial data set, and the size of these images is uniformly processed as 416×416 pixels. After the image data is acquired, it is necessary to annotate all kinds of images by the labelImg tool, convert the images in the data set into PASCALVOC format, and generate XML files. The image dataset was randomly divided into 3296 images as the training set, 412 images as the validation set, and 412 images as the test set in the ratio of 8:1:1.
4.3. Experimental indicators of target detection
The results of evaluation criteria include precision, detection speed FPS, recall rate P, R and the average accuracy of PmA. Expressions are as follows:
(7)
(8)
(9)
(10)
In the formula, Type of TP was right in the class is divided into class is the number of;FP negative class number off by mistake;FN was right in the number of such false negative class;PA is the average accuracy and the performance index of each sample classifier.N (class) as the total number of samples [30].
4.4. Analysis of results
Here we take three cases of transformer equipment faults as an example to check the performance of the smart grid equipment fault detection model constructed in this paper. First of all, visual detection technology is firstly adopted as a preliminary diagnostic means, utilizing high-definition camera and image processing technology to conduct real-time monitoring of the appearance and operation status of the equipment in order to obtain the preliminary detection results of the equipment. After obtaining the preliminary detection results, these results are used as inputs for querying the knowledge graph. By matching the preliminary detection results with the entities in the knowledge graph, the knowledge entries related to the current fault can be quickly localized. Next, the graph database query language Cypher is utilized to perform the query operation. The graph database will search in the knowledge graph according to the conditions in the query statement and return the diagnostic analysis results that meet the conditions. Finally, the diagnostic analysis results based on the knowledge graph are obtained by executing the Cypher query. The resulting detailed diagnostic analysis records are shown in Table 5.
By comparing and analyzing the diagnostic results of this paper’s model with the input case, it can be seen that the model detection results of this paper are accurate, consistent with the cause of the case failure, and can give practical repair measures in combination with the cause of the failure. The experimental diagram is shown in Fig 6.
In order to further validate the performance of the smart grid equipment fault detection model constructed in this paper, the performance of this algorithm is simulated and analyzed in the substation inspection environment together with the existing models constructed based on the Faster R-CNN algorithm, HSD (hierarchical corporative multi-agent reinforcement learning with skill discovery) algorithm and EFGRnet (guided refinement network) algorithm, and EFGRnet (guided refinement network) algorithm in substation inspection environment. Faster R-CNN is a popular target detection algorithm, which combines the advantages of Fast R-CNN, a regional proposal network, and achieves efficient target detection. detection. In substation inspection, Faster R-CNN can be used to detect external defects of substation equipment, such as metal corrosion, oil leakage from equipment, etc. The HSD algorithm has high detection accuracy, so it is also often used for substation equipment fault detection. The EFGRnet algorithm has a unique network structure and optimization strategy. The performance comparison results of the four different models are shown in Fig 7. Datasets on Table 7 were shown in S1 Table (see Supporting Information).
It can be seen that in terms of detection speed, YOLOv4 (this paper’s algorithm) has the highest detection speed, reached 54, compared with the detection speed of the lowest Faster R-CNN algorithm is 44 higher, can meet the needs of substation inspection work. In terms of target detection recall, the difference between YOLOv4 (this paper’s algorithm) and other algorithms is not very obvious, 68.2%, which is at a relatively excellent level, and in terms of detection precision, this paper’s algorithm is 43.5%, which is still the best performance among the four algorithms. Therefore, it can be said that compared with the existing fault detection algorithms, the YOLOv4 algorithm constructed in this paper has good performance.
5. Conclusion
In summary, based on YOLOv4 architecture, this paper introduces knowledge graph to build a fault diagnosis model for smart grid equipment. The detailed process of knowledge graph construction is given, and the CSPn module is introduced to optimize the YOLOv4 model, so as to match with the CBM module, so as to meet the fast and accurate requirements of smart grid equipment fault diagnosis. Using the experimental evaluation method, the detection model of this paper is verified by taking 3 transformer faults as examples. Experimental results show that compared with Faster R-CNN model, EFGRnet algorithm model and HSD algorithm model, the YOLOv4 algorithm constructed in this paper has excellent performance compared with existing detection models in terms of detection speed, detection accuracy, target detection recall rate and other aspects.
On the whole, the smart grid fault detection model constructed in this paper has reached the established goals and can meet the requirements of equipment inspection. However, the knowledge graph constructed in this paper is based on substation maintenance records and images, and the detection model used is also used for fault detection on the premise of learning from existing maintenance records. If there is less data information about a certain type of fault, it is difficult for the system to accurately diagnose it. In the future work, it will continue to enhance its diagnostic capabilities, improve the functions of pickup detection and infrared detection, fully realize the application of four diagnostic methods to diagnose potential problems of substation equipment, and improve the learning ability and diagnostic ability of the model to solve the problem of high dependence on historical fault data.
Supporting information
S1 Table. Datasets related to algorithms performance.
https://doi.org/10.1371/journal.pone.0315143.s001
(XLSX)
References
- 1. Li Y, He H, Chen Y, et al. A cloud-based eco-driving solution for autonomous hybrid electric bus rapid transit in cooperative vehicle-infrastructure systems: A dynamic programming approach. Green Energy and Intelligent Transportation, 2023, 2(6): 100122.
- 2. Meng F, Yang S, Wang J, et al. Creating knowledge graph of electric power equipment faults based on BERT–BiLSTM–CRF model. Journal of Electrical Engineering & Technology, 2022, 17(4): 2507–2516.
- 3. Yiyang Liu, Fei Li, Qingbo Guan. Power Equipment Fault Diagnosis Method Based on Energy Spectrogram and Deep Learning. Sensors, 2022, 22 (19): 7330–7330. pmid:36236431
- 4.
Yin T, Lu N. Knowledge graph model of power grid for human-machine mutual understanding. Chinese Automation Congress (CAC). IEEE, 2020: 6165–6169.
- 5. Linming Du, Guangtao Liu. Application of Hybrid Fault diagnosis and analysis method of RCM and PHM in Locomotive Remote operation and maintenance. China High and New Technology, 2019 (7): 103–106.
- 6. Lele Li, Yiwei Wang, Chao Ding. Application of Knowledge Graph for Aircraft Repair and Maintenance. Internal Combustion Engine and Accessories, 2019 (23): 147–148.
- 7.
Hao M, Yan Y, Sun J, et al. Research on Power Fault Inference Diagnosis Method based on Knowledge Map. Proceedings of the 5th International Conference on Information Technologies and Electrical Engineering. 2022: 283–287.
- 8.
MA W Z, ZHANG M, CAO Y, et al. Jointly learning explaina blerules for recommendation with knowledge graph. The World Wide Web Conference on—WWW ƴ19. New York: ACM Press, 2019.
- 9. NIE B L, SUN S Q. Knowledge graph embedding via reasoning over entities, relations, and text. Future Generation Computer Systems, 2019, 91: 426–433.
- 10. Liao W, Yang D, Wang Y, et al. Fault diagnosis of power transformers using graph convolutional network. CSEE Journal of Power and Energy Systems, 2020, 7(2): 241–249.
- 11.
Irshad, Z.J. Performance comparison of image segmentation techniques for infrared images. 2015 Annual IEEE India Conference, 2015, 12: 1–5.
- 12.
Dutta T, Sil J, Chottopadhyay P. Condition monitoring of electrical equipment using thermal image processing. IEEE First International Conference on Control, Measurement and Instrumentation.IEEE, 2016: 311–315.
- 13.
Sarathkumar D, Srinivasan M, Stonier A A, et al. A technical review on classification of various faults in smart grid systems. IOP conference series: Materials science and engineering. IOP Publishing, 2021, 1055(1): 012152.
- 14. Wang J, Wang X, Ma C. A survey on the development status and application prospects of knowledge graph in smart grids. IET Generation, Transmission & Distribution, 2021, 15(3): 383–407.
- 15. Huang H, Hong Z, Zhou H. Knowledge graph construction and application of power grid equipment. Mathematical Problems in Engineering, 2020: 1–10.
- 16.
Yun F, Feng Z, Baofeng L. Research on intelligent fault diagnosis of power acquisition based on knowledge graph. 2019 3rd International Conference on Electronic Information Technology and Computer Engineering (EITCE). IEEE, 2019: 1737–1740.
- 17. Dorji S, Stonier A A, Peter G, et al. An Extensive Critique on Smart Grid Technologies: Recent Advancements, Key Challenges, and Future Directions. Technologies, 2023, 11(3): 81.
- 18. De Almeida L F F, Pereira L A M, Sodré A C, et al. Control networks and smart grid teleprotection: Key aspects, technologies, protocols, and case-studies. IEEE Access, 2020, 8: 174049–174079.
- 19. Martinez-Ríos E A, Ponce-Cruz P, Molina A. A holistic educational platform for the study of the smart grid. International Journal on Interactive Design and Manufacturing, 2022, 16(3): 841–861.
- 20. Srivastava I, Bhat S, Singh A R. Smart grid communication: Recent trends and challenges. Next Generation Smart Grids: Modeling, Control and Optimization, 2022: 49–75.
- 21. Caporuscio M, Flammini F, Khakpour N, et al. Smart-troubleshooting connected devices: Concept, challenges and opportunities. Future Generation Computer Systems, 2020, 111: 681–697.
- 22.
Favre-Perrod P, Auberson E, Bernasconi C, et al. Reduction of fault location time and troubleshooting time in low voltage distribution networks by analysis of smart meter PLC communication data traffic and alarm messages. CIRED 2021-The 26th International Conference and Exhibition on Electricity Distribution. IET, 2021, 2021: 2959–2963.
- 23. Dileep G. A survey on smart grid technologies and applications. Renewable energy, 2020, 146: 2589–2625.
- 24. Pani B B, Giri N C, Nayak S R, et al. Fault detection and troubleshooting in a PV grid-tied inverter. Indian Journal of Science and Technology, 2021, 14(22): 1829–1838.
- 25.
Gupta P, Kandari R, Kumar A. An introduction to the smart grid-I. Advances in Smart Grid Power System. Academic Press, 2021: 1–31.
- 26.
Saadat S, Bahizad S, Ahmed T, et al. Smart Grid and Cybersecurity Challenges. 5th IEEE Workshop on the Electronic Grid. IEEE, 2020: 1–8.
- 27.
Trujillo C, Zambrano C, Celeita D, et al. An advanced laboratory for research and training of smart grid control and protection schemes. IEEE/IAS 56th Industrial and Commercial Power Systems Technical Conference (I&CPS). IEEE, 2020: 1–7.
- 28. Abrahamsen F E, Ai Y, Cheffena M. Communication technologies for smart grid: A comprehensive survey. Sensors, 2021, 21(23): 8087. pmid:34884092
- 29. Ourahou M, Ayrir W, Hassouni B E L, et al. Review on smart grid control and reliability in presence of renewable energies: Challenges and prospects. Mathematics and computers in simulation, 2020, 167: 19–31.
- 30. Demertzis K, Tsiknas K, Taketzis D, et al. Communication network standards for smart grid infrastructures. Network, 2021, 1(2): 132–145.