Power transformer winding turn-to-turn short circuit fault detection method

Abstract

The invention provides a power transformer winding turn-to-turn short circuit fault detection method, and belongs to the technical field of power equipment, and the method comprises the steps: obtaining a power transformer winding turn-to-turn historical short circuit fault record, recognizing a winding turn-to-turn short circuit formation reason from the historical short circuit fault record, and setting a short circuit fault detection point according to the formation reason; determining a detection sensor based on the short-circuit fault detection point, further creating a short-circuit fault detection graph, and determining a reason graph of each formation reason according to the short-circuit fault detection graph; short-circuit fault features are determined according to the short-circuit fault detection graph and the reason graph, a fault detection model is established based on the short-circuit fault features and applied to real-time data, whether a short-circuit fault exists or not is judged according to model output, a detection report is synchronously output, and accurate fault positioning and timely early warning are achieved. And the specific reason of the winding turn-to-turn short circuit fault can be accurately identified and positioned, and the timeliness and effectiveness of fault early warning are enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power equipment, and in particular to a method for detecting short-circuit faults between turns of a power transformer winding. Background Art

[0002] Power transformers are important equipment in power systems. Their stability and safety directly affect the operation of the power grid. Winding turn-to-turn short-circuit faults are one of the common fault types of transformers, which may lead to serious consequences such as equipment damage and power outages. There are currently some sensor-based fault detection technologies, such as temperature monitoring, vibration analysis, and current monitoring. Although these technologies can identify transformer faults to a certain extent, they often lack specificity and cannot accurately locate the cause of the fault, resulting in insufficient timeliness and accuracy in fault warnings.

[0003] Therefore, the present invention provides a method for detecting an inter-turn short-circuit fault of a power transformer winding. Summary of the invention

[0004] The present invention provides a method for detecting short-circuit faults between turns of a power transformer winding. The method analyzes historical short-circuit fault records, identifies the causes of short-circuit faults between turns of the winding, sets corresponding detection points, and selects appropriate sensors to create a fault detection map. Based on the map and the cause map, the short-circuit fault characteristics are extracted, and a fault detection model is established. The model monitors data in real time, determines whether a short-circuit fault exists, and generates a detection report, thereby achieving accurate fault location and timely warning, improving the pertinence and accuracy of power transformer fault detection, being able to accurately identify and locate the specific causes of short-circuit faults between turns of the winding, enhancing the timeliness and effectiveness of fault warnings, reducing the risks of equipment damage and power outages, and thus improving the stability and safety of the power system.

[0005] The present invention provides a method for detecting a short-circuit fault between turns of a power transformer winding, comprising: Step 1: Obtain historical short-circuit fault records between turns of a power transformer winding, identify the causes of short circuits between turns of the winding from the historical short-circuit fault records, and set short-circuit fault detection points of the power transformer according to the causes; Step 2: determining a corresponding detection sensor based on the short-circuit fault detection point, and then creating a short-circuit fault detection map, and determining a cause map for each formation cause according to the short-circuit fault detection map; Step 3: Determine the short-circuit fault characteristics according to the short-circuit fault detection diagram and the cause diagram, and establish a fault detection model based on the short-circuit fault characteristics; Step 4: Apply the fault detection model to real-time data, determine whether a short circuit fault exists based on the model output, and output a detection report simultaneously.

[0006] The present invention provides a method for detecting short-circuit faults between turns of a power transformer winding, obtaining short-circuit fault records between turns of a winding, identifying the causes of the short-circuit between turns of the winding from the historical short-circuit fault records, and setting short-circuit fault detection points of the power transformer according to the causes, including: Obtain historical fault records of the power transformer, classify the historical faults according to the fault results in the historical fault records, and then obtain historical short-circuit fault records between winding turns; Performing time series analysis on the data before and after the occurrence of the historical short-circuit fault, identifying the existing abnormal patterns and abnormal trends, and determining the cause of each historical short-circuit fault based on the abnormal patterns and abnormal trends; Prioritize the causes of the short-circuit faults according to the occurrence frequency and impact of each short-circuit fault in the historical short-circuit fault records; According to the key indicator corresponding to each formation cause in the cause-indicator table, a short-circuit fault detection point between turns of the power transformer winding is set based on the key indicator and the priority sorting result.

[0007] The present invention provides a method for detecting short-circuit faults between turns of a power transformer winding, determining a corresponding detection sensor based on the short-circuit fault detection point, and then creating a short-circuit fault detection map, and determining a cause map of each formation cause according to the short-circuit fault detection map, including: Performing characteristic analysis on the short-circuit fault detection point, selecting a corresponding detection sensor according to the characteristic analysis result, drawing a physical layer detection diagram according to the actual structure of the power transformer at the short-circuit fault detection point and the corresponding detection sensor, and drawing a data layer detection diagram according to the detection data of the detection sensor; Comprehensively analyzing the physical layer detection map and the data layer detection map to identify key detection points and data flows, and creating a short circuit fault detection map based on the comprehensive analysis results; The various cause nodes in the cause graph are determined according to the formation causes, and the cause graph for each formation cause is determined by combining the short-circuit fault detection graph and the various cause nodes.

[0008] The present invention provides a method for detecting short-circuit faults between turns of a power transformer winding, determining short-circuit fault characteristics according to a short-circuit fault detection diagram and a cause diagram, and establishing a fault detection model based on the short-circuit fault characteristics, including: Determine the key parameters of each detection node according to the role of each detection point in the power transformer, extract the historical data of each detection point based on the key parameters, and form a detection feature set; Determine the cause characteristics related to the short circuit fault according to each cause node in the cause graph; Associating the cause feature with the detection feature set to form a comprehensive feature set, determining a feature relationship between each cause feature and each feature in the detection feature set according to the feature set, combining the feature relationship with the comprehensive feature set to determine different short-circuit fault modes, and extracting short-circuit fault features of each short-circuit fault mode; A fault detection model is created according to the short circuit fault characteristics.

[0009] The present invention provides a method for detecting short-circuit faults between turns of a power transformer winding, and a fault detection model is created according to the short-circuit fault characteristics, comprising: Arranging the short-circuit fault features into graph structure data, wherein the short-circuit fault features are graph nodes, and the feature relationships are graph connection lines; A feature matrix is ​​obtained according to the graph nodes, and an adjacency matrix of the feature matrix is ​​obtained according to the graph connection lines; A graph convolutional network architecture is determined based on the feature matrix and the adjacency matrix, and a fault detection model is created according to the graph convolutional network architecture.

[0010] The present invention provides a method for detecting short-circuit faults between turns of a power transformer winding, determining a graph convolutional network architecture based on the feature matrix and the adjacency matrix, and creating a fault detection model according to the graph convolutional network architecture, including: The feature matrix and the normalized adjacency matrix are used as the input layer of the graph convolutional network architecture; Perform diversity judgment on the feature matrix and the adjacency matrix, and set the number of layers of the graph convolutional network architecture according to the diversity judgment results. Each layer uses the graph convolution operation to aggregate the information of each graph node and the corresponding adjacent nodes to obtain the aggregated information, and then derive the graph convolution layer of the graph convolutional network architecture. Determine an activation function according to the aggregate information, introduce nonlinear information, and obtain an output layer of the graph convolutional network architecture by combining the aggregate information and the nonlinear information; The input layer, the graph convolution layer and the output layer determine the graph convolution network architecture, and then construct a fault detection model.

[0011] The present invention provides a method for detecting short-circuit faults between turns of power transformer windings, wherein the fault detection model is applied to real-time data, and whether a short-circuit fault exists is determined according to the model output, and a detection report is output synchronously, including: Determine the sub-features corresponding to each short-circuit fault feature according to the cause diagram, and derive a dynamic threshold based on the sub-features and the short-circuit fault feature; The real-time data is input into the fault detection model, and a short circuit fault is determined to exist if the data is not less than the dynamic threshold value, and a corresponding detection report is output according to the cause diagram and the short circuit fault detection diagram.

[0012] The present invention provides a method for detecting short-circuit faults between turns of a power transformer winding, which determines sub-features corresponding to each short-circuit fault feature according to the cause diagram, and derives a dynamic threshold based on the sub-features and the short-circuit fault feature, including: ,in, represents the dynamic threshold at time t; represents the standard threshold value of the power transformer at t=0; represents the linear attenuation factor; represents the quadratic attenuation factor; represents the weight of the i-th short-circuit fault feature; n represents the number of short-circuit fault features; m represents the number of sub-features related to the i-th short-circuit fault feature; represents the weight of the jth sub-feature associated with the i-th short-circuit fault feature; represents the detection value of the jth sub-feature at time t=0; represents the mean of the jth sub-feature; represents the standard deviation of the j-th sub-feature; represents the additional attenuation factor related to the i-th short-circuit fault feature; t represents time.

[0013] Compared with the prior art, the beneficial effects of the present application are as follows: by analyzing historical short-circuit fault records, identifying the causes of short circuits between winding turns, setting corresponding detection points, and selecting appropriate sensors to create a fault detection map, based on this map and the cause map, extracting short-circuit fault characteristics, and establishing a fault detection model. The model monitors data in real time, determines whether there is a short-circuit fault, and generates a detection report, thereby achieving accurate fault location and timely warning, improving the pertinence and accuracy of power transformer fault detection, and being able to accurately identify and locate the specific causes of short-circuit faults between winding turns, enhancing the timeliness and effectiveness of fault warnings, reducing the risk of equipment damage and power outages, and thus improving the stability and safety of the power system.

[0014] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0015] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1It is a flow chart of a method for detecting short-circuit faults between turns of a power transformer winding provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0018] Embodiment 1: The embodiment of the present invention provides a method for detecting a short-circuit fault in a power transformer winding. Figure 1 As shown, including: Step 1: Obtain historical short-circuit fault records between turns of a power transformer winding, identify the causes of short circuits between turns of the winding from the historical short-circuit fault records, and set short-circuit fault detection points of the power transformer according to the causes; Step 2: determining a corresponding detection sensor based on the short-circuit fault detection point, and then creating a short-circuit fault detection map, and determining a cause map for each formation cause according to the short-circuit fault detection map; Step 3: Determine the short-circuit fault characteristics according to the short-circuit fault detection diagram and the cause diagram, and establish a fault detection model based on the short-circuit fault characteristics; Step 4: Apply the fault detection model to real-time data, determine whether a short circuit fault exists based on the model output, and output a detection report simultaneously.

[0019] In this embodiment, the historical fault record records all fault events that occur during the operation of the power transformer, including information such as fault time, type, and impact. For example, all fault events that occurred in a transformer in the past five years include winding turn short circuit, overload, insulation failure, etc.

[0020] In this embodiment, the formation cause is a specific cause that leads to the short circuit fault, such as insulation aging, overload, manufacturing defects, etc. For example, the formation cause of a short circuit fault is due to aging of the insulation material.

[0021] In this embodiment, the short-circuit fault detection point is a monitoring point set according to key indicators and priority rankings, and is used to monitor the risk of short-circuit faults in real time. For example, a temperature sensor and an insulation resistance monitor are set at the winding of the transformer as a short-circuit fault detection point.

[0022] In this embodiment, the detection sensor is a device used to monitor and collect specific parameters, such as a temperature sensor: monitoring winding temperature, an insulation resistance tester: monitoring insulation status, and a current sensor: monitoring load current.

[0023] In this embodiment, the short-circuit fault detection diagram is a diagram that integrates physical layer and data layer information to display the fault detection process and monitoring results. For example, the diagram shows how the monitoring data of different sensors are collected into the fault analysis module and ultimately generates early warning information.

[0024] In this embodiment, the cause diagram is a diagram showing the causes of short-circuit faults and their interrelationships. For example, the cause diagram indicates insulation aging, overload, and environmental factors as nodes that cause short-circuit faults, and the connection method indicates the causal relationship between them.

[0025] In this embodiment, the short-circuit fault characteristic is a specific indicator or state that describes the short-circuit fault mode. For example, the short-circuit fault characteristic may be a sudden high current value, a sharp rise in temperature, a rapid decrease in insulation resistance, and the like.

[0026] In this embodiment, by determining the key parameters of each detection point of the power transformer, extracting historical data to form a detection feature set, and identifying the cause features related to the short-circuit fault based on the cause diagram, the cause features are associated with the detection feature set, the feature relationship is analyzed, the short-circuit fault mode is identified and the fault features are extracted.

[0027] In this embodiment, the input of the fault detection model is the variables used to train and test the fault detection model, such as the data in the detection feature set (such as temperature, insulation resistance, load current, etc.). The output of the fault detection model is the result generated by the model based on the input data, which is usually a fault state or warning information. For example, the output may be "normal operation", "potential fault" or "fault warning". The number of training times is the number of times the model is iterated during the training process, which is used to optimize the model performance. For example, it is usually set to 1,000 times or more, and the specific number depends on the complexity of the model and the amount of data.

[0028] In this embodiment, by analyzing the short-circuit fault characteristics and their corresponding sub-characteristics, a dynamic threshold is established, and real-time data is input into the fault detection model. If the data exceeds the dynamic threshold, it is determined that a short-circuit fault exists. Combined with the cause diagram and the fault detection diagram, a detailed detection report is generated.

[0029] In this embodiment, the detection report content includes fault type, fault characteristics, sub-characteristic analysis, dynamic threshold, fault location and corresponding recommended measures.

[0030] The working principle and beneficial effects of the above technical solution are: by analyzing historical short-circuit fault records, identifying the causes of short circuits between winding turns, setting corresponding detection points, and selecting appropriate sensors to create a fault detection map, based on this map and the cause map, extracting short-circuit fault characteristics, and establishing a fault detection model. The model monitors data in real time, determines whether there is a short-circuit fault, and generates a detection report, thereby achieving accurate fault location and timely warning, improving the pertinence and accuracy of power transformer fault detection, and being able to accurately identify and locate the specific causes of short-circuit faults between winding turns, enhancing the timeliness and effectiveness of fault warnings, reducing the risk of equipment damage and power outages, and thus improving the stability and safety of the power system.

[0031] Embodiment 2: The embodiment of the present invention provides a method for detecting a short-circuit fault between turns of a power transformer winding, obtaining a short-circuit fault record between turns of the winding, identifying a cause of the short-circuit between turns of the winding from the historical short-circuit fault record, and setting a short-circuit fault detection point of the power transformer according to the cause, including: Obtain historical fault records of the power transformer, classify the historical faults according to the fault results in the historical fault records, and then obtain historical short-circuit fault records between winding turns; Performing time series analysis on the data before and after the occurrence of the historical short-circuit fault, identifying the existing abnormal patterns and abnormal trends, and determining the cause of each historical short-circuit fault based on the abnormal patterns and abnormal trends; Prioritize the causes of the short-circuit faults according to the occurrence frequency and impact of each short-circuit fault in the historical short-circuit fault records; According to the key indicator corresponding to each formation cause in the cause-indicator table, a short-circuit fault detection point between turns of the power transformer winding is set based on the key indicator and the priority sorting result.

[0032] In this embodiment, the fault result is the impact on the equipment after the fault occurs, such as the degree of damage, downtime, maintenance cost, etc. For example, a short circuit between winding turns causes the transformer to stop operating for 48 hours, and the maintenance cost is 100,000 yuan.

[0033] In this embodiment, the classification process is to systematically classify the fault records according to type, cause, impact, etc., to facilitate subsequent analysis. For example, the fault records are divided into categories such as "winding turn short circuit", "insulation fault", and "overload".

[0034] In this embodiment, the historical short-circuit fault record is a record specifically of the events of the winding turn-to-turn short-circuit fault, including relevant fault data and impacts, for example, the occurrence time, duration, maintenance measures and results of each winding turn-to-turn short-circuit.

[0035] In this embodiment, time series analysis is to analyze the trends and patterns of data changes over time to identify potential anomalies. For example, by analyzing the data of transformer temperature and load current changes over time, the correlation between abnormal temperature rise and short circuit faults is found.

[0036] In this embodiment, the abnormal pattern and abnormal trend are that the abnormal pattern refers to abnormal behavior found in the data, and the abnormal trend refers to the trend of data changes over time. For example, before a short circuit fault occurs, the temperature continues to rise and the load current fluctuations intensify, forming an abnormal pattern.

[0037] In this embodiment, the key indicators are important data indicators related to the cause of the fault, which are used to monitor and judge the fault risk, such as temperature, load current, insulation resistance and other indicators.

[0038] In this embodiment, the cause-indicator table lists the cause of each fault and its corresponding key indicator for easy monitoring and analysis. For example, if the cause is insulation aging, the corresponding key indicator is insulation resistance.

[0039] In this embodiment, the priority sorting result is to sort the causes according to the frequency of occurrence and the degree of impact of the fault to determine the priority of monitoring. For example, the sorting result may be insulation aging, overload, and mechanical damage.

[0040] The working principle and beneficial effects of the above technical solution are: by obtaining the historical fault records of the power transformer, classifying and extracting the short-circuit faults between winding turns, using time series analysis to identify abnormal patterns and trends, determining the cause of each fault, and prioritizing the causes according to the fault frequency and impact, and setting key indicators in combination with the cause-indicator table, so as to determine the short-circuit fault detection point, realize accurate monitoring and fault location, enhance the pertinence and accuracy of power transformer fault detection, and effectively identify and prioritize high-risk fault causes.

[0041] Embodiment 3: The embodiment of the present invention provides a method for detecting short-circuit faults between turns of a power transformer winding, wherein a corresponding detection sensor is determined based on the short-circuit fault detection point, and then a short-circuit fault detection map is created. A cause map of each formation cause is determined according to the short-circuit fault detection map, including: Performing characteristic analysis on the short-circuit fault detection point, selecting a corresponding detection sensor according to the characteristic analysis result, drawing a physical layer detection diagram according to the actual structure of the power transformer at the short-circuit fault detection point and the corresponding detection sensor, and drawing a data layer detection diagram according to the detection data of the detection sensor; Comprehensively analyzing the physical layer detection map and the data layer detection map to identify key detection points and data flows, and creating a short circuit fault detection map based on the comprehensive analysis results; The various cause nodes in the cause graph are determined according to the formation causes, and the cause graph for each formation cause is determined by combining the short-circuit fault detection graph and the various cause nodes.

[0042] In this embodiment, the characteristic analysis process is to evaluate the electrical characteristics, physical characteristics and environmental impacts of the short-circuit fault detection point to determine its monitoring requirements, for example, analyzing the winding temperature, load current, insulation resistance and other characteristics to evaluate its sensitivity to short-circuit faults.

[0043] In this embodiment, the characteristic analysis results are based on the conclusions drawn from the analysis to determine the most suitable monitoring parameters and methods. For example, it is found that temperature increase and insulation resistance decrease are significantly associated with short circuit failures, so temperature and insulation resistance are selected as key monitoring indicators.

[0044] In this embodiment, the actual structure is the physical construction of the power transformer and the layout of its components, such as the specific configuration of the transformer's windings, core, insulation materials and cooling system.

[0045] In this embodiment, the physical layer detection diagram is a diagram showing the position of the detection sensor in the transformer structure and its connection method. For example, the diagram indicates that the temperature sensor is installed near the winding and the current sensor is installed at the input and output lines.

[0046] In this embodiment, the data layer detection diagram is a diagram showing the flow of monitoring data obtained from the sensor and its processing method. For example, the diagram shows that the temperature sensor data flows to the data collection module and then to the analysis module of the monitoring system.

[0047] In this embodiment, the data flow is the transmission path of the monitoring data in the system. For example, the data collected by the temperature sensor is transmitted to the monitoring system through the data collector for analysis and storage.

[0048] In this embodiment, the comprehensive analysis result is obtained by analyzing the physical layer detection diagram and the data layer detection diagram to identify the relationship between key parameters and data flows, including key detection points and data flows. For example, it is determined that temperature and insulation resistance are key monitoring points for short circuit faults, and their data flows to the monitoring system.

[0049] In this embodiment, the relationship between the key detection point and the comprehensive analysis result is that the key detection point is the monitoring focus determined by comprehensive analysis, which directly affects the effectiveness of fault detection. For example, if the comprehensive analysis result shows that temperature rise is a precursor to a short circuit fault, the temperature sensor becomes a key detection point.

[0050] The working principle and beneficial effects of the above technical solution are: by analyzing the characteristics of the short-circuit fault detection points, selecting suitable detection sensors, and drawing physical layer and data layer detection maps according to the transformer structure, combining the detection data, identifying key detection points and data flows, creating a short-circuit fault detection map, and constructing a cause map based on the cause of the fault. Comprehensive analysis is performed to improve the accuracy and pertinence of fault detection, accurately select and configure detection sensors, and improve the pertinence and accuracy of fault detection.

[0051] Embodiment 4: The embodiment of the present invention provides a method for detecting short-circuit faults between turns of a power transformer winding, determining short-circuit fault characteristics according to a short-circuit fault detection diagram and a cause diagram, and establishing a fault detection model based on the short-circuit fault characteristics, including: Determine the key parameters of each detection node according to the role of each detection point in the power transformer, extract the historical data of each detection point based on the key parameters, and form a detection feature set; Determine the cause characteristics related to the short circuit fault according to each cause node in the cause graph; Associating the cause feature with the detection feature set to form a comprehensive feature set, determining a feature relationship between each cause feature and each feature in the detection feature set according to the feature set, combining the feature relationship with the comprehensive feature set to determine different short-circuit fault modes, and extracting short-circuit fault features of each short-circuit fault mode; A fault detection model is created according to the short circuit fault characteristics.

[0052] In this embodiment, the key parameter is a key indicator used by each detection point in the power transformer to monitor a specific state or performance, such as a temperature sensor: the key parameter is the winding temperature (for example, exceeding 85°C may indicate a fault).

[0053] In this embodiment, the insulation resistance tester: the key parameter is the insulation resistance value (for example, less than 1 MΩ may indicate a fault).

[0054] In this embodiment, historical data is past monitoring records related to key parameters, which are used for analysis and modeling, such as monitoring data of temperature and insulation resistance in the past year, including temperature changes in daily operation and periodic test results of insulation resistance.

[0055] In this embodiment, the detection feature set is a feature extracted from historical data and is used to describe the state of the detection point. For example, the feature set may include average temperature, temperature fluctuation range, insulation resistance change trend, etc.

[0056] In this embodiment, the cause node represents a factor or condition that may cause a short circuit fault in the cause graph, such as insulation aging, overload, humid environment, etc.

[0057] In this embodiment, the cause characteristics are characteristics related to the cause node and are used to describe the status of these factors. For example, the characteristic of insulation aging may be a gradual decrease in insulation resistance, and the characteristic of overload may be a continuous increase in load current.

[0058] In this embodiment, the association process is to match the cause feature with the detection feature set to identify potential failure modes. For example, analysis finds that when the insulation resistance is below a certain threshold, the abnormal increase in the temperature feature is highly correlated with a short circuit failure.

[0059] In this embodiment, the characteristic relationship describes the degree and manner of mutual influence or association between different characteristics. For example, the established model shows that there is a positive correlation between temperature increase and insulation resistance decrease, and this relationship is more obvious before a short circuit fault occurs.

[0060] In this embodiment, the short-circuit fault mode is different short-circuit fault types identified based on feature analysis, for example, mode 1: short circuit caused by insulation aging, characterized by a significant increase in temperature and a rapid decrease in insulation resistance, mode 2: short circuit caused by overload, characterized by a load current that continuously exceeds the rated value.

[0061] The working principle and beneficial effects of the above technical solution are: by determining the key parameters of each detection point of the power transformer, extracting historical data to form a detection feature set, and identifying the cause characteristics related to the short-circuit fault based on the cause diagram, associating the cause characteristics with the detection feature set, analyzing the feature relationship, identifying the short-circuit fault mode and extracting the fault characteristics, and finally building a fault detection model to improve the accuracy and timeliness of fault detection and better adapt to the complex power transformer operating environment.

[0062] Embodiment 5: The embodiment of the present invention provides a method for detecting a short-circuit fault between turns of a power transformer winding, and creates a fault detection model according to the short-circuit fault characteristics, comprising: Arranging the short-circuit fault features into graph structure data, wherein the short-circuit fault features are graph nodes, and the feature relationships are graph connection lines; A feature matrix is ​​obtained according to the graph nodes, and an adjacency matrix of the feature matrix is ​​obtained according to the graph connection lines; A graph convolutional network architecture is determined based on the feature matrix and the adjacency matrix, and a fault detection model is created according to the graph convolutional network architecture.

[0063] In this embodiment, graph structured data is a form of data representation, in which data is organized in the form of a graph, nodes represent features, and edges represent the relationship between features. For example, short-circuit fault features (such as temperature, insulation resistance, and load current) are used as nodes, and the relationship between features (such as the correlation between temperature and insulation resistance) are used as edges to form a graph structure.

[0064] In this embodiment, the graph node is a basic unit in the graph, representing a specific feature or state, for example, node 1: winding temperature, node 2: insulation resistance, node 3: load current.

[0065] In this embodiment, the graph connection line is an edge connecting the graph nodes, which is used to represent the relationship or mutual influence between the nodes. For example, edge 1: connects node 1 (winding temperature) and node 2 (insulation resistance), which represents the relationship between temperature increase and insulation resistance decrease; edge 2: connects node 2 (insulation resistance) and node 3 (load current), which represents the influence of load current change on insulation resistance.

[0066] In this embodiment, the feature matrix is ​​a matrix in which each row represents the feature value of a graph node. For example, the feature matrix , The adjacency matrix is ​​a matrix that represents the connection relationship between nodes in a graph, usually 0 (no connection) and 1 (connection). For example, the adjacency matrix .

[0067] In this embodiment, the graph convolutional network architecture is a deep learning model based on graph structured data, which can perform information propagation and feature learning through the features and connection relationships of nodes. For example, the graph convolutional network can be designed as a multi-layer structure, each layer aggregates the features of neighbor nodes through the adjacency matrix, updates the representation of each node, and thus identifies short-circuit fault modes.

[0068] The working principle and beneficial effect of the above technical solution are: organize the short-circuit fault features into graph structure data, use the features as graph nodes, use the feature relationships as graph connection lines, build a feature matrix through graph nodes, and use the connection lines to generate an adjacency matrix. Design a graph convolutional network architecture based on the feature matrix and adjacency matrix, create a fault detection model, thereby achieving efficient fault pattern recognition and location, and improving the accuracy and pertinence of fault detection.

[0069] Embodiment 6: An embodiment of the present invention provides a method for detecting short-circuit faults between turns of a power transformer winding, determining a graph convolutional network architecture based on the feature matrix and the adjacency matrix, and creating a fault detection model according to the graph convolutional network architecture, including: The feature matrix and the normalized adjacency matrix are used as the input layer of the graph convolutional network architecture; Perform diversity judgment on the feature matrix and the adjacency matrix, and set the number of layers of the graph convolutional network architecture according to the diversity judgment results. Each layer uses the graph convolution operation to aggregate the information of each graph node and the corresponding adjacent nodes to obtain the aggregated information, and then derive the graph convolution layer of the graph convolutional network architecture. Determine an activation function according to the aggregate information, introduce nonlinear information, and obtain an output layer of the graph convolutional network architecture by combining the aggregate information and the nonlinear information; The input layer, the graph convolution layer and the output layer determine the graph convolution network architecture, and then construct a fault detection model.

[0070] In this embodiment, the normalization process is to normalize the adjacency matrix to ensure that the influence of each node is balanced and to avoid excessive influence of some nodes on the aggregation results. For example, for the adjacency matrix A, the normalization process can be implemented by the following formula: A~=D−1 / 2AD−1 / 2A~=D−1 / 2AD−1 / 2, where D is the node degree matrix.

[0071] In this embodiment, the process of diversity judgment is to analyze the diversity of the feature matrix and the adjacency matrix to determine the number of network layers. For example, the diversity can be judged by calculating the variance or correlation between the features. If the correlation between the features is low, it means that the feature diversity is high, and more layers may be needed to capture complex relationships; conversely, if the correlation is high, the number of layers can be appropriately reduced.

[0072] In this embodiment, the process of setting the number of layers is based on the result of diversity judgment to set the number of layers of the graph convolutional network. For example, if the correlation between temperature, insulation resistance and load current in the feature matrix is ​​low (high diversity), it may be set to 3 layers; if the correlation is high, it can be set to only 2 layers.

[0073] In this embodiment, the adjacent nodes are nodes directly connected to the current node, representing its neighbors in the graph. For example, if node 1 represents "winding temperature", its adjacent nodes may be "insulation resistance" and "load current", representing the direct relationship between these features.

[0074] In this embodiment, the aggregated information is information aggregated from adjacent nodes through graph convolution operations to update the feature representation of the current node. For example, in a layer of graph convolution, the aggregated information of the node "winding temperature" may be the weighted average of its own temperature value and the adjacent nodes "insulation resistance" and "load current".

[0075] In this embodiment, nonlinear information is a nonlinear feature introduced by an activation function, which enables the model to learn more complex patterns. For example, using a ReLU (Rectified Linear Unit) activation function, after the aggregated information is processed by ReLU, negative values ​​can be turned into zero, thereby enhancing the model's response to positive features.

[0076] In this embodiment, the input layer is the first layer of the graph convolutional network architecture, which receives a feature matrix and a normalized adjacency matrix as input. For example, the feature matrix received by the input layer may contain numerical values ​​of different temperatures, insulation resistances, and load currents, and the adjacency matrix represents the relationship between these features.

[0077] In this embodiment, the graph convolution layer is the middle layer of the graph convolution network architecture, and the node features are updated through the graph convolution operation. For example, the first graph convolution layer may aggregate the information of "winding temperature", "insulation resistance" and "load current" to generate a new node representation for further processing in subsequent layers.

[0078] In this embodiment, the output layer is the last layer of the graph convolutional network architecture, which generates the final prediction result. For example, the output layer may generate a fault detection result, such as "normal" or "short circuit fault", based on the node features after multi-layer aggregation and nonlinear transformation.

[0079] The working principle and beneficial effect of the above technical solution are: the number of layers of the graph convolutional network is set by diversity judgment. In each layer, the graph convolution operation aggregates the information of the node and its adjacent nodes to generate aggregate information, and the activation function is combined to introduce nonlinear information to finally form the output layer. Through this structure, the model can effectively extract and learn the characteristics of short-circuit faults, thereby achieving efficient fault detection.

[0080] Embodiment 7: The embodiment of the present invention provides a method for detecting short-circuit faults between turns of a power transformer winding, wherein the fault detection model is applied to real-time data, and whether a short-circuit fault exists is determined according to the model output, and a detection report is output synchronously, including: Determine the sub-features corresponding to each short-circuit fault feature according to the cause diagram, and derive a dynamic threshold based on the sub-features and the short-circuit fault feature; The real-time data is input into the fault detection model, and a short circuit fault is determined to exist if the data is not less than the dynamic threshold value, and a corresponding detection report is output according to the cause diagram and the short circuit fault detection diagram.

[0081] In this embodiment, the sub-feature is a subordinate feature of the short-circuit fault feature. For example, if the short-circuit fault feature is temperature change, the corresponding sub-features are the winding temperature and the oil temperature.

[0082] The working principle and beneficial effects of the above technical solution are: by analyzing the short-circuit fault characteristics and their corresponding sub-characteristics, a dynamic threshold is established, and real-time data is input into the fault detection model. If the data exceeds the dynamic threshold, it is determined that a short-circuit fault exists. Combining the cause diagram with the fault detection diagram, a detailed detection report is generated to achieve fast and accurate fault location and early warning, thereby improving the safety and stability of the power transformer.

[0083] Embodiment 8: The embodiment of the present invention provides a method for detecting short-circuit faults between turns of a power transformer winding, which determines sub-features corresponding to each short-circuit fault feature according to the cause diagram, and derives a dynamic threshold based on the sub-features and the short-circuit fault feature, including: ,in, represents the dynamic threshold at time t; represents the standard threshold value of the power transformer at t=0; represents the linear attenuation factor; represents the quadratic attenuation factor; represents the weight of the i-th short-circuit fault feature; n represents the number of short-circuit fault features; m represents the number of sub-features related to the i-th short-circuit fault feature; represents the weight of the jth sub-feature associated with the i-th short-circuit fault feature; represents the detection value of the jth sub-feature at time t=0; represents the mean of the jth sub-feature; represents the standard deviation of the j-th sub-feature; represents the additional attenuation factor related to the i-th short-circuit fault feature; t represents time.

[0084] The working principle and beneficial effects of the above technical solution are: monitoring short-circuit fault characteristics through dynamic thresholds, combining standard thresholds, linear and quadratic attenuation factors, calculating the weight of each feature and its sub-features, and comparing real-time data with the threshold. If it exceeds the dynamic threshold, it is determined to be a fault, which improves the sensitivity and accuracy of fault detection, can locate the cause of the fault in time, and enhance the safety and stability of the power transformer.

[0085] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting short-circuit faults between turns of a power transformer winding, characterized in that: include: Step 1: Obtain historical short-circuit fault records between turns of a power transformer winding, identify the causes of short circuits between turns of the winding from the historical short-circuit fault records, and set short-circuit fault detection points of the power transformer according to the causes; Step 2: determining a corresponding detection sensor based on the short-circuit fault detection point, and then creating a short-circuit fault detection map, and determining a cause map for each formation cause according to the short-circuit fault detection map; Step 3: Determine the short-circuit fault characteristics according to the short-circuit fault detection diagram and the cause diagram, and establish a fault detection model based on the short-circuit fault characteristics; Step 4: Apply the fault detection model to real-time data, determine whether a short circuit fault exists based on the model output, and output a detection report simultaneously.

2. A method for detecting short-circuit faults in power transformer windings according to claim 1, characterized in that: Obtaining short-circuit fault records between winding turns, identifying the cause of the short circuit between winding turns from the historical short-circuit fault records, and setting the short-circuit fault detection point of the power transformer according to the cause, including: Obtain historical fault records of the power transformer, classify the historical faults according to the fault results in the historical fault records, and then obtain historical short-circuit fault records between winding turns; Performing time series analysis on the data before and after the occurrence of the historical short-circuit fault, identifying the existing abnormal patterns and abnormal trends, and determining the cause of each historical short-circuit fault based on the abnormal patterns and abnormal trends; Prioritize the causes of the short-circuit faults according to the occurrence frequency and impact of each short-circuit fault in the historical short-circuit fault records; According to the key indicator corresponding to each formation cause in the cause-indicator table, a short-circuit fault detection point between turns of the power transformer winding is set based on the key indicator and the priority sorting result.

3. A method for detecting short-circuit faults in power transformer windings according to claim 1, characterized in that: Determining a corresponding detection sensor based on the short-circuit fault detection point, and then creating a short-circuit fault detection map, and determining a cause map of each formation cause according to the short-circuit fault detection map, including: Performing characteristic analysis on the short-circuit fault detection point, selecting a corresponding detection sensor according to the characteristic analysis result, drawing a physical layer detection diagram according to the actual structure of the power transformer at the short-circuit fault detection point and the corresponding detection sensor, and drawing a data layer detection diagram according to the detection data of the detection sensor; Comprehensively analyzing the physical layer detection map and the data layer detection map to identify key detection points and data flows, and creating a short circuit fault detection map based on the comprehensive analysis results; The various cause nodes in the cause graph are determined according to the formation causes, and the cause graph for each formation cause is determined by combining the short-circuit fault detection graph and the various cause nodes.

4. A method for detecting short-circuit faults in power transformer windings according to claim 1, characterized in that: Determine the short-circuit fault characteristics according to the short-circuit fault detection diagram and the cause diagram, and establish a fault detection model based on the short-circuit fault characteristics, including: Determine the key parameters of each detection node according to the role of each detection point in the power transformer, extract the historical data of each detection point based on the key parameters, and form a detection feature set; Determine the cause characteristics related to the short circuit fault according to each cause node in the cause graph; Associating the cause feature with the detection feature set to form a comprehensive feature set, determining a feature relationship between each cause feature and each feature in the detection feature set according to the feature set, combining the feature relationship with the comprehensive feature set to determine different short-circuit fault modes, and extracting short-circuit fault features of each short-circuit fault mode; A fault detection model is created according to the short circuit fault characteristics.

5. A method for detecting short-circuit faults between turns of a power transformer winding according to claim 4, characterized in that: Creating a fault detection model according to the short circuit fault characteristics includes: Arranging the short-circuit fault features into graph structure data, wherein the short-circuit fault features are graph nodes, and the feature relationships are graph connection lines; A feature matrix is ​​obtained according to the graph nodes, and an adjacency matrix of the feature matrix is ​​obtained according to the graph connection lines; A graph convolutional network architecture is determined based on the feature matrix and the adjacency matrix, and a fault detection model is created according to the graph convolutional network architecture.

6. A method for detecting short-circuit faults between turns of a power transformer winding according to claim 5, characterized in that: Determining a graph convolutional network architecture based on the feature matrix and the adjacency matrix, and creating a fault detection model according to the graph convolutional network architecture, including: The feature matrix and the normalized adjacency matrix are used as the input layer of the graph convolutional network architecture; Perform diversity judgment on the feature matrix and the adjacency matrix, and set the number of layers of the graph convolutional network architecture according to the diversity judgment results. Each layer uses the graph convolution operation to aggregate the information of each graph node and the corresponding adjacent nodes to obtain the aggregated information, and then derive the graph convolution layer of the graph convolutional network architecture. Determine an activation function according to the aggregate information, introduce nonlinear information, and obtain an output layer of the graph convolutional network architecture by combining the aggregate information and the nonlinear information; The input layer, the graph convolution layer and the output layer determine the graph convolution network architecture, and then construct a fault detection model.

7. A method for detecting short-circuit faults in power transformer windings according to claim 1, characterized in that: Apply the fault detection model to real-time data, determine whether there is a short circuit fault based on the model output, and output a detection report simultaneously, including: Determine the sub-features corresponding to each short-circuit fault feature according to the cause diagram, and derive a dynamic threshold based on the sub-features and the short-circuit fault feature; The real-time data is input into the fault detection model, and a short circuit fault is determined to exist if the data is not less than the dynamic threshold value, and a corresponding detection report is output according to the cause diagram and the short circuit fault detection diagram.

8. A method for detecting short-circuit faults between turns of a power transformer winding according to claim 7, characterized in that: Determining a sub-feature corresponding to each short-circuit fault feature according to the cause diagram, and deriving a dynamic threshold based on the sub-feature and the short-circuit fault feature, including: ,in, represents the dynamic threshold at time t; represents the standard threshold of the power transformer at t=0; represents the linear attenuation factor; represents the quadratic attenuation factor; represents the weight of the i-th short-circuit fault feature; n represents the number of short-circuit fault features; m represents the number of sub-features related to the i-th short-circuit fault feature; represents the weight of the jth sub-feature associated with the i-th short-circuit fault feature; represents the detection value of the jth sub-feature at time t=0; represents the mean of the jth sub-feature; represents the standard deviation of the j-th sub-feature; represents the additional attenuation factor related to the i-th short-circuit fault feature; t represents time.

Images