Transformer fault diagnosis method and system based on artificial intelligence, equipment and medium

Abstract

The invention provides a transformer fault diagnosis method and system based on artificial intelligence, equipment and a medium, and is applied to the technical field of intelligent operation and maintenance of power equipment. The method comprises the following steps: acquiring running state data and surface image data of a transformer; performing anomaly detection on the surface image data based on a target detection algorithm to obtain a surface anomaly category and a corresponding anomaly region of the transformer; performing fault structure association on the abnormal region corresponding to each surface anomaly category by using the transformer structure fault knowledge graph to obtain an associated fault structure and a corresponding association condition; transformer fault diagnosis is carried out based on the operation state data, the associated fault structure and the corresponding association condition, and a fault diagnosis result is obtained; the transformer structure fault knowledge graph is constructed based on the connection relationship among the component structures in the transformer and the fault association relationship among the component structures. According to the invention, the problems of high missing report and false alarm rate and high operation and maintenance difficulty in the transformer fault diagnosis process are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent operation and maintenance of power equipment, and in particular to an artificial intelligence-based transformer fault diagnosis method and system, equipment, and medium. Background Art

[0002] In power systems, transformers have long been crucial for power supply and energy conversion. However, their operation and maintenance often face numerous interference factors and harsh environmental conditions. Due to the significant randomness of load fluctuations, a single data dimension cannot fully capture the transformer's operating status. Therefore, real-time monitoring of transformer operation is necessary to promptly identify potential risks.

[0003] Existing solutions typically monitor a subset of operating parameters, triggering alarms when their values ​​exceed thresholds. However, this threshold-based monitoring approach cannot fully reflect more complex fault symptoms. Fixed criteria, especially in volatile environments and aging equipment, can lead to high rates of missed and false alarms. This increases the difficulty of inspection and maintenance, and can easily trigger costly incidents such as widespread power outages and equipment damage, posing a threat to power system reliability. Summary of the Invention

[0004] In order to overcome the problems of high missed alarm and false alarm rates and difficult operation and maintenance in the above-mentioned transformer fault diagnosis process, the present invention provides a transformer fault diagnosis method and system, equipment, and medium based on artificial intelligence.

[0005] In one aspect, the present invention provides a transformer fault diagnosis method based on artificial intelligence, comprising:

[0006] Acquire operating status data and surface image data of a transformer in operation; perform anomaly detection on the surface image data based on a target detection algorithm to obtain a surface anomaly category and a corresponding abnormal area of ​​the transformer;

[0007] The transformer structural fault knowledge graph is used to associate the fault structures of the abnormal areas corresponding to each surface abnormality category, and the associated fault structures and corresponding association situations are obtained;

[0008] Perform transformer fault diagnosis based on the operating status data, the associated fault structure, and the corresponding associated conditions to obtain a fault diagnosis result;

[0009] Among them, the transformer structural fault knowledge graph is constructed based on the connection relationship between the component structures in the transformer and the fault association relationship between the component structures.

[0010] Optionally, the surface abnormality category includes at least one of the following: chassis damage, discharge marks, surface contamination, and equipment tilt.

[0011] Optionally, before using the transformer structural fault knowledge graph to perform fault structure association on the abnormal areas corresponding to each surface abnormality category to obtain the associated fault structure and the corresponding association situation, the method further includes:

[0012] Construct entity nodes based on the equipment information and component structure of the transformer;

[0013] Based on the physical connection relationship and functional association relationship between each component structure, relationship edges are constructed between each entity node;

[0014] Based on the historical fault data of the transformer, a fault description is added to each entity node and the fault association relationship between the various component structures of the transformer is determined. Based on the fault association relationship between the various component structures of the transformer, a fault attribute is added to each relationship edge;

[0015] Based on each entity node and its fault description, each relationship edge and its fault attribute, a transformer structural fault knowledge graph is constructed.

[0016] Optionally, the use of the transformer structural fault knowledge graph to perform fault structure association on the abnormal areas corresponding to each surface abnormality category to obtain associated fault structures and corresponding association situations includes:

[0017] Based on the pre-trained word vector model, semantic matching is performed between each surface anomaly category and the fault description of each entity node in the transformer structure fault knowledge graph to determine candidate matching nodes;

[0018] Determining associated nodes associated with the candidate matching node based on each relationship edge in the transformer structure fault knowledge graph;

[0019] Performing position matching on the abnormal area corresponding to each surface abnormality category with the candidate matching node and the associated node, and taking the node with successful position matching as the associated fault structure;

[0020] The node fault description and fault association relationship corresponding to the associated fault structure are searched in the transformer structure fault knowledge graph as the associated situation corresponding to the associated fault structure.

[0021] Optionally, before performing transformer fault diagnosis based on the operating status data and the associated fault structure and corresponding associated conditions to obtain a fault diagnosis result, the method further includes:

[0022] If the internal temperature of the transformer is detected to be abnormal, the associated fault structure and the corresponding associated conditions are corrected based on the temperature abnormality area.

[0023] Optionally, the correcting the associated fault structure and the corresponding associated situation based on the temperature abnormality area includes:

[0024] Screening out a target structure related to temperature from the associated fault structures based on the fault description of the associated fault structure;

[0025] Correcting the target structure and the fault impact range of the target structure based on the temperature anomaly area to obtain the temperature anomaly structure and the corresponding impact range;

[0026] The associated fault structure and the corresponding associated situation are updated based on the temperature anomaly structure and the corresponding impact range.

[0027] Optionally, the performing transformer fault diagnosis based on the operating status data and the associated fault structure and corresponding associated conditions to obtain a fault diagnosis result includes:

[0028] Searching a transformer historical fault case library based on the associated fault structure and corresponding associated conditions to determine a target fault case that meets the associated fault structure and corresponding associated conditions;

[0029] performing operational matching between the current situation and each fault case based on a difference between the operational status data and the operational parameters corresponding to each fault case in the target fault case;

[0030] The fault case that is successfully matched is used as the fault diagnosis result.

[0031] Optionally, the performing transformer fault diagnosis based on the operating status data and the associated fault structure and corresponding associated conditions to obtain a fault diagnosis result further includes:

[0032] If there is no fault case that is successfully matched, fault prediction is performed on the transformer based on the historical operating status data and environmental meteorological data of the target period to obtain the corresponding fault prediction result;

[0033] Based on the fault prediction result and the target fault case, a potential fault condition of the transformer is determined as the fault diagnosis result.

[0034] Optionally, determining a potential fault condition of the transformer as the fault diagnosis result based on the fault prediction result and the target fault case includes:

[0035] For each prediction result in the fault prediction results, determining a feature similarity of the prediction result based on a Euclidean distance between the prediction result and each fault case in the target fault case;

[0036] Determining a matching degree of the operating state of the prediction result based on a degree of difference between the operating state data corresponding to the prediction result and the operating parameters corresponding to each fault case in the target fault case;

[0037] Performing weighted fusion processing on the feature similarity and the running state matching degree of the prediction result to obtain a matching score of the prediction result;

[0038] Based on the matching score of each prediction result, a potential fault condition of the transformer is determined as the fault diagnosis result.

[0039] On the other hand, the present invention also provides a transformer fault diagnosis method system based on artificial intelligence, comprising:

[0040] An anomaly detection module is used to obtain operating status data and surface image data of the transformer in operation; perform anomaly detection on the surface image data based on the target detection algorithm to obtain the surface anomaly category of the transformer and the corresponding abnormal area;

[0041] An association module is used to use the transformer structural fault knowledge graph to perform fault structure association on the abnormal area corresponding to each surface abnormality category, and obtain the associated fault structure and corresponding association situation;

[0042] A fault diagnosis module, configured to perform transformer fault diagnosis based on the operating status data, the associated fault structure, and the corresponding associated conditions, and obtain a fault diagnosis result;

[0043] Among them, the transformer structural fault knowledge graph is constructed based on the connection relationship between the component structures in the transformer and the fault association relationship between the component structures.

[0044] Optionally, the surface abnormality category includes at least one of the following: chassis damage, discharge marks, surface contamination, and equipment tilt.

[0045] Optionally, the system further includes a knowledge graph construction module, which includes:

[0046] The node construction submodule is used to construct entity nodes based on the device information and component structure of the transformer;

[0047] The variable construction submodule is used to build relationship edges between various entity nodes based on the physical connection relationship and functional association relationship between various component structures;

[0048] The attribute adding submodule is used to add fault descriptions to each entity node based on the historical fault data of the transformer and determine the fault association relationship between the various component structures of the transformer, and add fault attributes to each relationship edge based on the fault association relationship between the various component structures of the transformer;

[0049] The graph construction submodule is used to construct a transformer structural fault knowledge graph based on each entity node and its fault description, each relationship edge and its fault attribute.

[0050] Optionally, the association module includes:

[0051] A node matching submodule is used to perform semantic matching between each surface anomaly category and the fault description of each entity node in the transformer structure fault knowledge graph based on a pre-trained word vector model to determine candidate matching nodes; and to determine associated nodes associated with the candidate matching nodes based on each relationship edge in the transformer structure fault knowledge graph;

[0052] A position matching submodule is used to perform position matching on the abnormal area corresponding to each surface abnormality category with the candidate matching node and the associated node, and use the node with successful position matching as the associated fault structure;

[0053] The search submodule is used to search the transformer structure fault knowledge graph for the node fault description and fault association relationship corresponding to the associated fault structure as the associated situation corresponding to the associated fault structure.

[0054] Optionally, the system further includes:

[0055] The relationship correction module is used to correct the associated fault structure and corresponding associated conditions based on the temperature abnormality area if the internal temperature of the transformer is monitored to be abnormal.

[0056] Optionally, the relationship correction module includes:

[0057] a structure screening submodule, configured to screen out temperature-related target structures from the associated fault structures based on the fault descriptions of the associated fault structures;

[0058] a correction submodule, configured to correct the target structure and the fault impact range of the target structure based on the temperature anomaly area, to obtain the temperature anomaly structure and the corresponding impact range;

[0059] An updating submodule is configured to update the associated fault structure and the corresponding associated situation based on the temperature anomaly structure and the corresponding impact range.

[0060] Optionally, the fault diagnosis module includes:

[0061] A case search submodule is used to search the transformer historical fault case library based on the associated fault structure and the corresponding associated conditions, and determine a target fault case that meets the associated fault structure and the corresponding associated conditions;

[0062] The operation matching submodule is used to perform operation matching between the current situation and each fault case based on the difference between the operation status data and the operation parameters corresponding to each fault case in the target fault case; and the fault case with successful operation matching is used as the fault diagnosis result.

[0063] Optionally, the fault diagnosis module further includes:

[0064] A fault prediction submodule is used to predict the fault of the transformer based on the historical operating status data and environmental meteorological data of the target period if there is no fault case that is successfully matched;

[0065] The diagnosis submodule is configured to determine a potential fault condition of the transformer as the fault diagnosis result based on the fault prediction result and the target fault case.

[0066] Optionally, the diagnosis submodule is specifically used to:

[0067] For each prediction result in the fault prediction results, determining a feature similarity of the prediction result based on a Euclidean distance between the prediction result and each fault case in the target fault case;

[0068] Determining a matching degree of the operating state of the prediction result based on a degree of difference between the operating state data corresponding to the prediction result and the operating parameters corresponding to each fault case in the target fault case;

[0069] Performing weighted fusion processing on the feature similarity and the running state matching degree of the prediction result to obtain a matching score of the prediction result;

[0070] Based on the matching score of each prediction result, a potential fault condition of the transformer is determined as the fault diagnosis result.

[0071] On the other hand, the present invention also provides an electronic device, comprising: at least one processor and a memory; the memory and the processor are connected via a bus;

[0072] The memory is used to store one or more programs;

[0073] When the one or more programs are executed by the at least one processor, any one of the above methods is implemented.

[0074] On the other hand, the present invention further provides a readable storage medium having an execution program stored thereon, and when the execution program is executed, any one of the methods described above is implemented.

[0075] Compared with the prior art, the present invention has the following beneficial effects:

[0076] The present invention provides an artificial intelligence-based transformer fault diagnosis method and system. By performing target detection on surface image data, the transformer's appearance anomaly category and location are determined. Then, a transformer structural fault knowledge graph constructed based on the connection relationships between the transformer's component structures and the fault association relationships between the component structures is used to perform correlation retrieval on the appearance anomaly category and location in both structural and fault dimensions. This method determines the associated fault structures and specific association situations that may result from the appearance anomaly. By combining this with an artificial intelligence algorithm through a step-by-step correlation from the outside in, various potential fault risks that may arise in transformers under complex environments can be promptly discovered, thereby reducing the rate of missed fault reports. Furthermore, the method combines operating status data to further identify high-risk potential fault structures within the associated fault structures, thereby improving the accuracy of fault diagnosis, reducing the false alarm rate, and enhancing the reliability of the power system. The present invention utilizes artificial intelligence to achieve real-time online fault diagnosis of transformers, enabling the timely detection of transformer faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 A schematic flow chart of a transformer fault diagnosis method based on artificial intelligence according to the present invention;

[0078] Figure 2 This is a structural block diagram of an electric power device of the present invention. DETAILED DESCRIPTION

[0079] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0080] Example 1

[0081] The present invention provides a transformer fault diagnosis method based on artificial intelligence, the schematic diagram of which is shown as follows: Figure 1 As shown, the method includes:

[0082] Step S110, obtaining operating status data and surface image data of a transformer in operation; performing anomaly detection on the surface image data based on a target detection algorithm to obtain a surface anomaly category and a corresponding abnormal area of ​​the transformer;

[0083] Step S120, using the transformer structural fault knowledge graph, performing fault structure association on the abnormal area corresponding to each surface abnormality category to obtain associated fault structures and corresponding association situations;

[0084] Step S130 , performing transformer fault diagnosis based on the operating status data, the associated fault structure, and the corresponding associated conditions to obtain a fault diagnosis result.

[0085] In this example embodiment, the operating status data may include multi-dimensional operating data. For example, the operating status data may include input / output voltage, load current, output power, temperature, pressure, and the like. The transformer may be a pole-mounted transformer, and the operating status data may be collected using sensors installed on the transformer body and its surroundings. For example, the sensors may include, but are not limited to, voltage sensors, current sensors, power meters, temperature sensors, and pressure sensors. The voltage sensor measures the voltage at the transformer's input and output terminals, the current sensor records changes in load current, the power meter calculates power factor and energy consumption, the temperature sensor monitors transformer oil temperature and external ambient temperature, and the pressure sensor measures the transformer's internal pressure. These sensors may collect data at a preset sampling frequency and transmit the data to a data processing platform via wireless communication (e.g., LoRa, NB-IoT) or wired communication (e.g., Modbus, CAN bus) for fault diagnosis. Surface image data may be collected using cameras deployed around the transformer. These cameras may include fixed high-definition cameras, infrared cameras, or drone inspection equipment. Fixed high-definition cameras are used to capture the external state of the transformer, detecting problems such as casing damage and contamination accumulation. Unmanned aerial vehicle inspection equipment is used to periodically capture multiple transformers from an aerial perspective, providing global monitoring from a high-altitude perspective. The captured image / video data can be stored at set intervals and can also be preliminarily screened using edge computing devices to remove invalid or redundant data. Surface anomaly categories include at least one of the following: chassis damage, discharge marks, surface contamination, and equipment tilt. The abnormal area corresponding to each surface anomaly category can be the location of the anomaly. The transformer structural fault knowledge graph is constructed based on the connectivity relationships and fault associations between the various component structures within the transformer. For example, the transformer structural fault knowledge graph can be constructed based on historical transformer fault data and a three-dimensional structural model / drawing. Associated fault structures and corresponding associated conditions can be matched against corresponding operating status data to determine the final fault diagnosis result. For example, if damage to the casing of a certain transformer model is detected, along with a sudden increase in internal pressure and / or abnormal voltage, current, or temperature, the transformer may be diagnosed with a short circuit fault. For anomaly detection results, the model outputs the specific anomaly type, location coordinates, and confidence score for subsequent analysis. By gradually correlating the results from the outside in, the present invention can promptly identify various potential fault risks that may arise in transformers under complex environments, thereby reducing the rate of missed fault reports. Furthermore, by combining operational status data, high-risk fault structures within the correlated fault structure can be further identified, thereby improving the accuracy of fault diagnosis, reducing the false alarm rate, and enhancing the reliability of the power system.

[0086] In some embodiments, to ensure data standardization, all acquired data (operational status data and surface image data) must be structured. Operational status data is stored in a time series format using a unified data format, such as JSON, CSV, or a database table, to facilitate subsequent rapid retrieval and correlation analysis. Surface image data can be archived according to timestamps and compressed using a standard compression encoding format (e.g., JPEG, H.264) before transmission to reduce the burden of data storage and transmission.

[0087] Data from different data sources (operating status data and surface image data) is fused through techniques such as time series alignment and feature normalization to ensure comparability and consistency between the various data types, thereby improving the accuracy of subsequent analysis. First, effective data fusion requires a unified time base. Different types of data originate from different acquisition devices, and the time intervals between data collection may be inconsistent. For example, operating data is typically recorded in seconds or minutes, while image / video data is generated based on the monitoring system's frame rate or trigger mechanism. Therefore, before data fusion, all data must be timestamped. Linear interpolation or nearest neighbor matching methods are used to unify data from different sources into a common time series framework to ensure temporal consistency. Second, feature normalization is performed on the different data types. After completing time series alignment and feature normalization, data fusion techniques are used to establish correlations. Specifically, the appearance anomaly detection results, as key event data, are matched with the operating status data at the corresponding time point. For example, if damage to the transformer casing is detected at a certain moment, operating parameters such as current, voltage, and temperature at that moment need to be queried. The fused multi-source data is stored as a structured dataset. Each data record includes fields such as timestamp, operating status, and the type and location of the appearance anomaly, forming a complete dataset. This dataset not only provides multi-dimensional information but also supports complex analysis tasks, thereby improving the accuracy and reliability of fault identification.

[0088] In some embodiments, performing anomaly detection on the surface image data based on a target detection algorithm includes:

[0089] Preprocessing the surface image data to obtain preprocessed image data;

[0090] Use convolutional neural network to extract features from preprocessed image data to obtain key features;

[0091] The key features are input into the target detection model to perform anomaly detection, and the surface anomaly category and the corresponding abnormal area are obtained.

[0092] In this example embodiment, the target detection model can be YOLO, Faster R-CNN or SSD, etc. The preprocessing of surface image data is the basis of appearance anomaly detection. Since the images collected by the camera may contain noise, illumination changes and angle deviation and other problems, preprocessing is required. Preprocessing specifically includes image denoising, histogram equalization, color normalization and geometric transformation correction to ensure the consistency and comparability of the input images. For example, Gaussian filtering or median filtering is used to remove noise, Gamma correction is used to adjust the influence of illumination, and perspective transformation is used to correct the shooting angle deviation. In the feature extraction stage, a convolutional neural network (CNN) can be used to perform deep feature learning on key areas in the image; the multi-layer convolution operation of CNN can automatically extract the edge features, texture patterns and color distribution of the transformer surface, and enhance the recognition ability of abnormal areas. In the feature extraction process, mainstream deep learning models such as ResNet, VGG or MobileNet can be used to optimize the network parameters through transfer learning or fine-tuning to adapt to the task of transformer appearance anomaly detection. During the object detection phase, an object detection model (such as YOLO, Faster R-CNN, or SSD) locates abnormal areas on the transformer's exterior and outputs an anomaly category. The object detection model is trained using a dataset of labeled samples to identify common types of appearance anomalies, including but not limited to casing damage, discharge marks, contamination, and tilt. For example, for casing damage, the model can identify cracks or damaged areas based on shape discontinuities and color abrupt changes. For discharge marks, the model can detect corona discharge phenomena through highlight features and edge radioactivity distribution. For contamination, texture features can be used to analyze the coverage of surface deposits. For tilt, the angular deviation between the transformer and the ground can be calculated by combining Hough transforms or geometric transformations. To improve detection accuracy and robustness, a multi-scale feature extraction strategy can be employed, enabling the model to identify anomaly areas of varying sizes. Non-maximum suppression algorithms can also be used to remove duplicate detection boxes, ensuring that each anomaly area is labeled only once. For anomaly detection results, the system outputs the specific anomaly type, location coordinates, and confidence score, and stores the results in a database for subsequent analysis.

[0093] This example uses an object detection algorithm to analyze acquired images and video data of transformer appearances to accurately identify external anomalies. A deep learning model is used to extract image features, combined with object detection technology to automatically identify and classify abnormal areas, generating appearance anomaly detection results. This implementation enables automated and intelligent identification of external anomalies in pole-mounted transformers, providing critical input for subsequent fault diagnosis and improving both efficiency and accuracy.

[0094] In an example implementation, before using the transformer structural fault knowledge graph to associate fault structures with abnormal areas corresponding to various surface abnormality categories to obtain associated fault structures and corresponding association situations, the method further includes:

[0095] Construct entity nodes based on the equipment information and component structure of the transformer;

[0096] Based on the physical connection relationship and functional association relationship between each component structure, relationship edges are constructed between each entity node;

[0097] Based on the historical fault data of the transformer, a fault description is added to each entity node and the fault association relationship between the various component structures of the transformer is determined. Based on the fault association relationship between the various component structures of the transformer, a fault attribute is added to each relationship edge;

[0098] Based on each entity node and its fault description, each relationship edge and its fault attribute, a transformer structural fault knowledge graph is constructed.

[0099] In this example embodiment, device information may include device type and model, etc., and the component structure of the transformer may include the core, winding, oil tank, oil pillow, radiator, insulating bushing (high-voltage bushing, low-voltage bushing), leads, etc. For example, a transformer structural fault knowledge graph can be constructed based on historical fault data and the transformer's three-dimensional structural model / drawings. The construction of the knowledge graph is based on the transformer's device model, structural composition, and component feature information (such as component failure conditions). The device model, main components (such as insulators, radiators, insulating bushings, etc.), and component-level components (leads, relays, etc.) can be used as entity nodes. Each entity node can have material attributes and fault attributes. The fault attribute is used to describe the fault condition of the entity node, and the material attribute is used to describe the material of the entity node. Based on the hierarchical structure and functional connections between the components, relationship edges are constructed between the entity nodes. The relationship edges can include functional relationships and structural connection relationships. For example, a "contains" attribute relationship exists between the transformer model and its components, while a "may be affected" attribute relationship can be established between components and components, or between component / component quality inspections through fault attributes. Finally, a transformer structural fault knowledge graph is constructed through each entity node and its fault description, each relationship edge and its fault attribute.

[0100] In some embodiments, the step S120 of using the transformer structural fault knowledge graph to perform fault structure association on the abnormal areas corresponding to each surface abnormality category to obtain associated fault structures and corresponding association situations includes:

[0101] Based on the pre-trained word vector model, semantic matching is performed between each surface anomaly category and the fault description of each entity node in the transformer structure fault knowledge graph to determine candidate matching nodes;

[0102] Determining associated nodes associated with the candidate matching node based on each relationship edge in the transformer structure fault knowledge graph;

[0103] Performing position matching on the abnormal area corresponding to each surface abnormality category with the candidate matching node and the associated node, and taking the node with successful position matching as the associated fault structure;

[0104] The node fault description and fault association relationship corresponding to the associated fault structure are searched in the transformer structure fault knowledge graph as the associated situation corresponding to the associated fault structure.

[0105] In this example embodiment, during the knowledge graph query process, the corresponding database of the transformer equipment model can be retrieved according to the surface anomaly category to obtain the structural information of the transformer. If damage to the shell of a certain transformer is detected, the system calls the structural drawing library of the model, extracts the structural drawings of the relevant components, and performs semantic matching. The semantic matching process calculates the similarity between the detected anomaly category (such as "crack" or "breakage") and the fault description of each entity node in the knowledge graph based on the pre-trained word vector model, and selects the nodes that may be affected as matching nodes. For example, when shell damage is detected, the heat sink or shell shield of the model may be retrieved, and the matching nodes are determined based on the similarity of the fault description of the heat sink or shell shield. At the same time, in order to improve the matching accuracy, the feature comparison method is used to verify whether the abnormal area corresponds to a specific component. Feature comparison includes position analysis. For example, the three-dimensional structure data of the transformer is called in the three-dimensional model library, and the computer vision algorithm is used to compare the detected abnormal area with the component position in the three-dimensional model to determine whether it matches the expected component, and the component with successful position matching is used as the associated fault structure. In addition, the ownership of the abnormal area can be further verified by comparing material properties (such as whether the shell is made of metal and whether the insulator is made of ceramic). After determining the associated fault structure, the node fault description and fault association relationship corresponding to the associated fault structure can be searched in the transformer structure fault knowledge graph as the associated situation corresponding to the associated fault structure. The final output query result can include the name of the affected component, structural information and its specific location in the transformer, and this information will be used as the basis for subsequent fault diagnosis to provide accurate equipment information support for further analysis. This example performs comparative analysis by calling the knowledge graph. The knowledge graph library contains fault modes based on historical cases, equipment model characteristics, and association information between common faults and affected components. A structured comparison is performed based on the known fault modes of the equipment model, so that the detected appearance anomalies can be accurately mapped to the physical structure of the transformer, thereby improving the accuracy and reliability of fault location.

[0106] In an example implementation, before performing transformer fault diagnosis based on the operating status data and the associated fault structure and corresponding associated conditions to obtain a fault diagnosis result, the method further includes:

[0107] If the internal temperature of the transformer is detected to be abnormal, the associated fault structure and the corresponding associated conditions are corrected based on the temperature abnormality area.

[0108] In this example implementation, the temperature of various locations inside the transformer can be obtained through a temperature sensor, thereby obtaining the temperature distribution inside the transformer. The inside of the transformer can be divided into temperature zones according to the structural distribution and function of the components inside the transformer, and temperature thresholds can be set for different temperature zones according to the component structures of different temperature zones. When the real-time monitored temperature of a certain temperature zone exceeds the corresponding temperature threshold, the temperature zone is determined to be a temperature abnormality zone. Based on the location of the temperature abnormality zone and the component structure of the zone, the associated fault structure and the structure and associated relationship with a low probability of failure in the corresponding associated situation are eliminated. This example uses temperature to correct the associated structure, which can realize the correction of associated faults in the temperature dimension and reduce the false alarm rate.

[0109] Exemplarily, the correcting the associated fault structure and the corresponding associated situation based on the temperature abnormality area includes:

[0110] Screening out a target structure related to temperature from the associated fault structures based on the fault description of the associated fault structure;

[0111] Correcting the target structure and the fault impact range of the target structure based on the temperature anomaly area to obtain the temperature anomaly structure and the corresponding impact range;

[0112] The associated fault structure and the corresponding associated situation are updated based on the temperature anomaly structure and the corresponding impact range.

[0113] In this example implementation, many transformer fault types can cause temperature anomalies. For example, long-term winding overload or debris falling into the winding can lead to excessive winding temperatures and insulation aging. Moisture in the winding can cause insulation expansion and blockage of the oil passages, causing localized overheating. Heat sink failure or circuit damage can also cause localized overheating, among other things. Therefore, the associated fault structure and its associated conditions can be corrected based on the temperature monitoring results within the transformer. For example, the structural fault can be determined to be temperature-related based on the fault description corresponding to the associated fault structure in the knowledge graph. For temperature-related target structures, further corrections can be made based on the temperature monitoring results, namely, the temperature anomaly areas. For example, structures whose target structures overlap with the temperature anomaly area can be identified as temperature anomaly structures. Target structures and their corresponding associations in other normal temperature areas can be eliminated, and the target structures and their corresponding impact ranges in the associated fault structure and corresponding associated conditions can be updated to the temperature anomaly structure and their corresponding impact ranges. This example uses real-time temperature monitoring results to correct the association results of the knowledge graph, achieving precise location of associated faults. For example, if it is detected that the casing of a certain model of transformer is damaged and the internal temperature is abnormal, similar cases of this model in historical data can be retrieved, and the fault structure and related situations (component short circuit or other faults) of past cases can be analyzed.

[0114] In an example implementation, the step S130 of performing transformer fault diagnosis based on the operating status data and the associated fault structure and corresponding associated conditions to obtain a fault diagnosis result includes:

[0115] Searching a transformer historical fault case library based on the associated fault structure and corresponding associated conditions to determine a target fault case that meets the associated fault structure and corresponding associated conditions;

[0116] Performing operation matching between the current situation and each fault case based on a difference between the operation status data and an operation parameter corresponding to each fault case in the target fault case;

[0117] The fault case that is successfully matched is used as the fault diagnosis result.

[0118] In this example embodiment, the historical fault case library can record information such as the historical fault type, fault component, fault impact range, and operating parameters at the time of the fault. Fault cases with a high degree of match between the associated fault structure and the corresponding associated situation can be searched within the historical fault case library as target fault cases. For example, fault cases with a high degree of overlap (greater than 90%) or a high coverage rate between the structure and the associated fault structure and the corresponding associated situation can be selected as target fault cases. After determining the target fault case, a determination is made as to whether the operational match between the operating status data and the operating parameters corresponding to each fault case in the target fault case is successful based on the difference between the two. For example, a difference threshold can be set for each operating parameter. When the differences between multiple operating parameters are all less than the corresponding difference threshold, the current situation is determined to be a successful operational match between the fault case and the current situation. Alternatively, a comprehensive difference threshold can be set. Multiple operating parameters in the current operating status data are weighted and then compared with the comprehensive difference threshold to determine whether the two are a successful match. During the weighted fusion, more important operating parameters can be assigned a higher weight, such as input voltage / current. Fault cases with successful operational matches are selected as fault diagnosis results. This example further verifies the correlation results by running parameters to improve the accuracy of fault diagnosis.

[0119] In an example implementation, step S130 of performing transformer fault diagnosis based on the operating status data and the associated fault structure and corresponding associated conditions to obtain a fault diagnosis result further includes:

[0120] If there is no fault case that is successfully matched, fault prediction is performed on the transformer based on the historical operating status data and environmental meteorological data of the target period to obtain the corresponding fault prediction result;

[0121] Based on the fault prediction result and the target fault case, a potential fault condition of the transformer is determined as the fault diagnosis result.

[0122] In this example embodiment, if the operation matching fails, that is, if there is no fault case in the target fault case that successfully matches the current situation, it means that there is no fault causing the abnormal operation, but there may be potential risks. Therefore, potential risk prediction can be performed by training a convolutional neural network or a machine learning model (such as a support vector machine, random forest model, etc.) as a fault diagnosis model. Specifically, based on the historical operating status data and environmental meteorological data of the target period, the fault diagnosis model is used to predict faults for the predicted period. For example, if severe weather is detected accompanied by abnormal operating temperature, the model can combine historical data to infer the possible internal short circuit risk. The target period can be a pre-defined historical period, such as the past month or the past week. The environmental meteorological data can include environmental meteorological data for the target period and / or environmental meteorological data for the predicted period. The acquisition of environmental meteorological data relies on meteorological monitoring devices near the transformer or remotely accessed through a meteorological data interface. Environmental meteorological data can include factors such as temperature, humidity, wind speed, and rainfall. These parameters have a significant impact on the operating status of the transformer. For example, high temperature may cause the internal oil temperature of the transformer to rise, humidity changes may affect insulation performance, and strong winds or heavy rain may increase the risk of equipment damage. Therefore, during the data collection process, a time synchronization mechanism is used to ensure the temporal consistency of environmental meteorological data and historical operating data, preventing data misalignment from affecting the prediction process. Finally, based on the similarity between the fault prediction results and the target fault case, the potential fault condition of the transformer is determined as the fault diagnosis result.

[0123] For example, the convolutional neural network model prediction process is as follows: First, the construction of the data input layer requires the integration of multi-source data, including operating data (voltage, current, temperature, etc.) and environmental data. To ensure the uniformity of the input data, the operating data is represented as a time series matrix, and the environmental data is time-synchronized with it as an additional feature. Second, the model design must be tailored to the physical characteristics of the transformer. For time series data (such as the changing trends of current, voltage, and temperature), one-dimensional convolution is used for feature extraction to capture the changing patterns of the operating state. In the feature fusion layer, a fully connected network is used to fuse the different data. Through weighted feature concatenation, abnormal operating parameter patterns and environmental influencing factors are comprehensively analyzed to calculate the probability of fault occurrence and the scope of impact. For specific types of faults, such as casing damage accompanied by temperature anomalies, the model calculates the internal short circuit risk that may be caused by this fault by matching historical data patterns. Finally, the fault prediction model outputs the fault category and its affected area, and the results are passed to the subsequent fault verification stage to further improve the accuracy and reliability of the diagnosis. The construction of this model not only improves the intelligent analysis capability of pole-mounted transformer faults, but also ensures the engineering applicability and credibility of the diagnostic results by combining physical information with operating laws.

[0124] Exemplarily, determining a potential fault condition of the transformer as the fault diagnosis result based on the fault prediction result and the target fault case includes:

[0125] For each prediction result in the fault prediction results, determining a feature similarity of the prediction result based on a Euclidean distance between the prediction result and each fault case in the target fault case;

[0126] Determining a matching degree of the operating state of the prediction result based on a degree of difference between the operating state data corresponding to the prediction result and the operating parameters corresponding to each fault case in the target fault case;

[0127] Performing weighted fusion processing on the feature similarity and the running state matching degree of the prediction result to obtain a matching score of the prediction result;

[0128] Based on the matching score of each prediction result, a potential fault condition of the transformer is determined as the fault diagnosis result.

[0129] In this example implementation, the matching process includes two parts: one is to calculate the similarity by matching the fault type with the historical case, and the other is to calculate the degree of deviation by matching the fault type with the operating data of the historical case. The fault prediction result may include structured results of multiple fault types, affected components, possible causes and impact ranges. The output result is usually stored in a standardized data format and associated with the fault time and transformer operating status data. The fault type, affected components, possible causes and impact range of each prediction result are encoded with the corresponding features of each fault case to form a vector and then a similarity calculation is performed to obtain the corresponding feature similarity; the operating status matching degree is determined based on the degree of difference between the operating status data corresponding to each prediction result and the operating parameters of each fault case. The greater the difference, the lower the operating status matching degree. By weighted fusion of feature similarity and operating status matching degree, the matching score of the corresponding prediction result and the fault case is determined, and the prediction result with a matching score exceeding the matching threshold is regarded as a potential fault condition. Exemplarily, the matching score calculation is as follows:

[0130]

[0131] Where: S f is the matching score, D feature Indicates feature similarity, which is used to measure the similarity of fault features; D operation is the operating state matching degree, which is used to characterize the degree of deviation between the current operating state parameters (current, voltage, load condition) of the transformer and the historical case; α and β are weight coefficients to ensure the balance of contributions of each part; λ is the parameter for adjusting the degree of exponential smoothing.f When it is higher than the set matching threshold T, the model prediction result is considered to be highly matched with the fault case, and the prediction result is considered accurate, and the potential fault location is completed for reference by subsequent operation and maintenance personnel. f If the value falls below T or there is a significant inconsistency, the system re-executes the fault diagnosis process. Compared to traditional hard-rule matching methods, this formula can more precisely quantify the reliability of diagnostic results, improving the model's generalization across different transformer models and environmental conditions, thereby enhancing the accuracy and robustness of the overall diagnostic system. This step ensures the accuracy and stability of fault diagnosis and, through the association mechanism of the knowledge graph, reduces the misdiagnosis rate, making transformer fault detection more reliable.

[0132] In power systems, transformers have long been crucial for power supply and energy conversion. However, their operation and maintenance often face numerous interference factors and harsh environmental conditions. Due to the significant randomness of load fluctuations and external meteorological influences, any single data dimension cannot fully capture the transformer's operating status. This results in traditional technologies often failing to identify anomalies in a timely or accurate manner. Existing solutions often rely on partial or single data sources to assess the health of pole-mounted transformers. A common approach is to set thresholds based on operating parameters, triggering alarms when values ​​exceed these limits. However, this threshold-based monitoring approach fails to fully reflect more complex fault symptoms. Especially in environments with fluctuating external conditions and aging equipment, fixed criteria can lead to high rates of missed and false alarms. Many fault predictions rely on manual inspections or post-event checks, often revealing serious consequences such as overheating, electrical discharge, or component damage. Furthermore, existing approaches based solely on analyzing a small number of operating parameters fail to promptly identify potential risks in the external environment and the environment. Failure to intervene early in the early stages of a fault can quickly escalate, compromising power supply security. The above-mentioned deficiencies not only increase the difficulty of inspection and maintenance, but also easily lead to high-cost accidents such as large-scale power outages and equipment damage, posing hidden dangers to the reliability of the power system.

[0133] By utilizing data of different dimensions and artificial intelligence models, the present invention designs a method of fault location by gradually associating from the outside to the inside, cleverly applies multi-dimensional data and realizes refined judgment of complex conditions of transformers. Through the present invention, intelligent and efficient diagnosis of transformer faults is realized, the accuracy, intelligence and diagnostic efficiency of transformer fault detection are improved, the misjudgment rate is reduced, and the operation and maintenance management level is improved.

[0134] Example 2

[0135] Based on the same inventive concept, the present invention also provides an artificial intelligence-based transformer fault diagnosis system, comprising:

[0136] An anomaly detection module is used to obtain operating status data and surface image data of the transformer in operation; perform anomaly detection on the surface image data based on the target detection algorithm to obtain the surface anomaly category of the transformer and the corresponding abnormal area;

[0137] An association module is used to use the transformer structural fault knowledge graph to perform fault structure association on the abnormal area corresponding to each surface abnormality category, and obtain the associated fault structure and corresponding association situation;

[0138] A fault diagnosis module, configured to perform transformer fault diagnosis based on the operating status data, the associated fault structure, and the corresponding associated conditions, and obtain a fault diagnosis result;

[0139] Among them, the transformer structural fault knowledge graph is constructed based on the connection relationship between the component structures in the transformer and the fault association relationship between the component structures.

[0140] In a possible implementation, the surface abnormality category includes at least one of the following: chassis damage, discharge marks, surface contamination, and device tilt.

[0141] In one possible implementation, the system further includes a knowledge graph construction module, which includes:

[0142] The node construction submodule is used to construct entity nodes based on the device information and component structure of the transformer;

[0143] The variable construction submodule is used to build relationship edges between various entity nodes based on the physical connection relationship and functional association relationship between various component structures;

[0144] The attribute adding submodule is used to add fault descriptions to each entity node based on the historical fault data of the transformer and determine the fault association relationship between the various component structures of the transformer, and add fault attributes to each relationship edge based on the fault association relationship between the various component structures of the transformer;

[0145] The graph construction submodule is used to construct a transformer structural fault knowledge graph based on each entity node and its fault description, each relationship edge and its fault attribute.

[0146] In a possible implementation, the association module includes:

[0147] A node matching submodule is used to perform semantic matching between each surface anomaly category and the fault description of each entity node in the transformer structure fault knowledge graph based on a pre-trained word vector model to determine candidate matching nodes; and to determine associated nodes associated with the candidate matching nodes based on each relationship edge in the transformer structure fault knowledge graph;

[0148] A position matching submodule is used to perform position matching on the abnormal area corresponding to each surface abnormality category with the candidate matching node and the associated node, and use the node with successful position matching as the associated fault structure;

[0149] The search submodule is used to search the transformer structure fault knowledge graph for the node fault description and fault association relationship corresponding to the associated fault structure as the associated situation corresponding to the associated fault structure.

[0150] In one possible implementation, the system further includes:

[0151] The relationship correction module is used to correct the associated fault structure and corresponding associated conditions based on the temperature abnormality area if the internal temperature of the transformer is monitored to be abnormal.

[0152] In a possible implementation, the relationship correction module includes:

[0153] a structure screening submodule, configured to screen out temperature-related target structures from the associated fault structures based on the fault descriptions of the associated fault structures;

[0154] a correction submodule, configured to correct the target structure and the fault impact range of the target structure based on the temperature anomaly area, to obtain the temperature anomaly structure and the corresponding impact range;

[0155] An updating submodule is configured to update the associated fault structure and the corresponding associated situation based on the temperature anomaly structure and the corresponding impact range.

[0156] In a possible implementation, the fault diagnosis module includes:

[0157] A case search submodule is used to search the transformer historical fault case library based on the associated fault structure and the corresponding associated conditions, and determine a target fault case that meets the associated fault structure and the corresponding associated conditions;

[0158] The operation matching submodule is used to perform operation matching between the current situation and each fault case based on the difference between the operation status data and the operation parameters corresponding to each fault case in the target fault case; and the fault case with successful operation matching is used as the fault diagnosis result.

[0159] In a possible implementation, the fault diagnosis module further includes:

[0160] A fault prediction submodule is used to predict the fault of the transformer based on the historical operating status data and environmental meteorological data of the target period if there is no fault case that is successfully matched;

[0161] The diagnosis submodule is configured to determine a potential fault condition of the transformer as the fault diagnosis result based on the fault prediction result and the target fault case.

[0162] In a possible implementation, the diagnosis submodule is specifically configured to:

[0163] For each prediction result in the fault prediction results, determining a feature similarity of the prediction result based on a Euclidean distance between the prediction result and each fault case in the target fault case;

[0164] Determining a matching degree of the operating state of the prediction result based on a degree of difference between the operating state data corresponding to the prediction result and the operating parameters corresponding to each fault case in the target fault case;

[0165] Performing weighted fusion processing on the feature similarity and the running state matching degree of the prediction result to obtain a matching score of the prediction result;

[0166] Based on the matching score of each prediction result, a potential fault condition of the transformer is determined as the fault diagnosis result.

[0167] Example 3

[0168] like Figure 2 As shown, the present invention also provides an electronic device, which may be a computer, a single-chip microcomputer, a smart mobile device, or the like. The electronic device in this embodiment may include a processor, a memory, a transceiver component, and the like. The memory, processor, and transceiver component are connected via a bus; the memory may be used to store an execution program, which may include instructions; and the processor may be used to execute the instructions stored in the memory. The memory may also be used to store data, which may be accessed and / or modified during the execution of the instructions.

[0169] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in a storage medium to implement the corresponding method flow or corresponding function, so as to implement the steps of an artificial intelligence-based transformer fault diagnosis method in the above embodiment.

[0170] Example 4

[0171] Based on the same inventive concept, the present invention also provides a readable storage medium, specifically an electronic device readable storage medium (Memory), which is a memory device in an electronic device for storing programs and data. It is understandable that the storage medium here can include both built-in storage media in the electronic device and, of course, extended storage media supported by the electronic device. The storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space. These instructions can be one or more execution programs (including program codes). It should be noted that the storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. The processor loads and executes one or more instructions stored in the storage medium, which can implement the steps of an artificial intelligence-based transformer fault diagnosis method in the above embodiment.

[0172] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0173] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0174] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0175] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0176] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit its scope of protection. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that after reading the present invention, those skilled in the art may still make various changes, modifications or equivalent substitutions to the specific implementation methods of the application, but these changes, modifications or equivalent substitutions are all within the scope of protection of the pending claims.

Claims

1. A transformer fault diagnosis method based on artificial intelligence, characterized in that: include: Acquiring operating state data and surface image data of the transformer in an operating state; Performing anomaly detection on the surface image data based on a target detection algorithm to obtain a surface anomaly category and a corresponding abnormal area of ​​the transformer; The transformer structural fault knowledge graph is used to associate the fault structures of the abnormal areas corresponding to each surface abnormality category, and the associated fault structures and corresponding association situations are obtained; Perform transformer fault diagnosis based on the operating status data, the associated fault structure, and the corresponding associated conditions to obtain a fault diagnosis result; Among them, the transformer structural fault knowledge graph is constructed based on the connection relationship between the component structures in the transformer and the fault association relationship between the component structures.

2. The method according to claim 1, characterized in that The surface abnormality category includes at least one of the following: chassis damage, discharge marks, surface contamination, and device tilt.

3. The method according to claim 1, characterized in that Before using the transformer structural fault knowledge graph to associate the fault structures of the abnormal areas corresponding to each surface abnormality category and obtain the associated fault structures and corresponding associated situations, the following steps are also included: Construct entity nodes based on the equipment information and component structure of the transformer; Based on the physical connection relationship and functional association relationship between each component structure, relationship edges are constructed between each entity node; Based on the historical fault data of the transformer, a fault description is added to each entity node and the fault association relationship between the various component structures of the transformer is determined. Based on the fault association relationship between the various component structures of the transformer, a fault attribute is added to each relationship edge; Based on each entity node and its fault description, each relationship edge and its fault attribute, a transformer structural fault knowledge graph is constructed.

4. The method according to claim 3, characterized in that The transformer structural fault knowledge graph is used to associate the fault structures of the abnormal areas corresponding to each surface abnormality category to obtain associated fault structures and corresponding association situations, including: Based on the pre-trained word vector model, semantic matching is performed between each surface anomaly category and the fault description of each entity node in the transformer structure fault knowledge graph to determine candidate matching nodes; Determining associated nodes associated with the candidate matching node based on each relationship edge in the transformer structure fault knowledge graph; Performing position matching on the abnormal area corresponding to each surface abnormality category with the candidate matching node and the associated node, and taking the node with successful position matching as the associated fault structure; The node fault description and fault association relationship corresponding to the associated fault structure are searched in the transformer structure fault knowledge graph as the associated situation corresponding to the associated fault structure.

5. The method according to claim 1, wherein Before performing transformer fault diagnosis based on the operating status data, the associated fault structure, and the corresponding associated conditions to obtain a fault diagnosis result, the method further includes: If the internal temperature of the transformer is detected to be abnormal, the associated fault structure and the corresponding associated conditions are corrected based on the temperature abnormality area.

6. The method according to claim 5, characterized in that The correcting of the associated fault structure and the corresponding associated situation based on the temperature abnormality area includes: Screening out a target structure related to temperature from the associated fault structures based on the fault description of the associated fault structure; Correcting the target structure and the fault impact range of the target structure based on the temperature anomaly area to obtain the temperature anomaly structure and the corresponding impact range; The associated fault structure and the corresponding associated situation are updated based on the temperature anomaly structure and the corresponding impact range.

7. The method according to claim 1 or 5, characterized in that The transformer fault diagnosis is performed based on the operating status data, the associated fault structure, and the corresponding associated situation to obtain a fault diagnosis result, including: Searching a transformer historical fault case library based on the associated fault structure and corresponding associated conditions to determine a target fault case that meets the associated fault structure and corresponding associated conditions; performing operational matching between the current situation and each fault case based on a difference between the operational status data and the operational parameters corresponding to each fault case in the target fault case; The fault case that is successfully matched is used as the fault diagnosis result.

8. The method according to claim 7, characterized in that The performing transformer fault diagnosis based on the operating status data and the associated fault structure and corresponding associated conditions to obtain a fault diagnosis result further includes: If there is no fault case that is successfully matched, fault prediction is performed on the transformer based on the historical operating status data and environmental meteorological data of the target period to obtain the corresponding fault prediction result; Based on the fault prediction result and the target fault case, a potential fault condition of the transformer is determined as the fault diagnosis result.

9. The method according to claim 8, characterized in that The determining, based on the fault prediction result and the target fault case, a potential fault condition of the transformer as the fault diagnosis result, includes: For each prediction result in the fault prediction results, determining a feature similarity of the prediction result based on a Euclidean distance between the prediction result and each fault case in the target fault case; Determining a matching degree of the operating state of the prediction result based on a degree of difference between the operating state data corresponding to the prediction result and the operating parameters corresponding to each fault case in the target fault case; Performing weighted fusion processing on the feature similarity and the running state matching degree of the prediction result to obtain a matching score of the prediction result; Based on the matching score of each prediction result, a potential fault condition of the transformer is determined as the fault diagnosis result.

10. A transformer fault diagnosis system based on artificial intelligence, characterized in that: include: an anomaly detection module, used to obtain operating status data and surface image data of the transformer in operating state; Performing anomaly detection on the surface image data based on a target detection algorithm to obtain a surface anomaly category and a corresponding abnormal area of ​​the transformer; An association module is used to use the transformer structural fault knowledge graph to perform fault structure association on the abnormal area corresponding to each surface abnormality category, and obtain the associated fault structure and corresponding association situation; A fault diagnosis module, configured to perform transformer fault diagnosis based on the operating status data, the associated fault structure, and the corresponding associated conditions, and obtain a fault diagnosis result; Among them, the transformer structural fault knowledge graph is constructed based on the connection relationship between the component structures in the transformer and the fault association relationship between the component structures.

Images