Power transformer sensing network monitoring and fault diagnosis method and device

Abstract

The invention relates to a power transformer sensor network monitoring and fault diagnosis method and device, computer equipment and a computer readable storage medium. The method comprises the steps of obtaining multiple pieces of state quantity data of a to-be-diagnosed transformer through a pre-deployed fusion monitoring system, and screening the multiple pieces of state quantity data to obtain target state quantity data; inputting the target state quantity data into a pre-trained fault diagnosis model; obtaining a fault loop identification result through a fault loop classifier in the fault diagnosis model; inputting the fault loop identification result and the target state quantity data into a fault part classifier in the fault diagnosis model to obtain a fault part identification result; inputting the fault part identification result and the target state quantity data into a fault reason classifier in the fault diagnosis model to obtain a fault reason identification result; and obtaining a fault diagnosis result of the to-be-diagnosed transformer according to each identification result. By adopting the method, the fault diagnosis accuracy of the transformer can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of power transformer status diagnosis, and in particular to a power transformer sensor network monitoring and fault diagnosis method, apparatus, computer equipment, and computer-readable storage medium. Background Art

[0002] As power systems continue to expand, transformers, as core components of the grid, shoulder crucial responsibilities. Their operational status directly impacts the safety and stability of the entire power system. Transformer failures often cause widespread grid outages. Therefore, timely and accurate transformer fault diagnosis and early detection of potential fault hazards are crucial to ensuring the safe operation of power systems.

[0003] Traditionally, power transformer fault diagnosis relies primarily on dissolved gas analysis (DGA) technology. DGA analyzes the composition and concentration of gases dissolved in transformer oil to identify the type, severity, and development trend of potential internal faults.

[0004] However, the current dissolved gas analysis technology in oil has great limitations in fault identification. For example, it can mainly identify faults such as overheating and discharge, but cannot identify other types of faults, and has the problem of low fault identification accuracy. Summary of the Invention

[0005] Based on this, it is necessary to provide a power transformer sensor network monitoring and fault diagnosis method, device, computer equipment and computer-readable storage medium that can improve fault identification accuracy in response to the above technical problems.

[0006] In a first aspect, the present application provides a method for monitoring and fault diagnosis of a power transformer sensor network, comprising:

[0007] Acquire multiple state quantity data of the transformer to be diagnosed through a pre-deployed fusion monitoring system, filter the multiple state quantity data, and obtain target state quantity data; the fusion monitoring system includes multiple types of sensors, and the multiple types of sensors are respectively used to monitor multiple different physical quantities of the transformer to be diagnosed;

[0008] Inputting the target state quantity data into a pre-trained fault diagnosis model; the fault diagnosis model includes a fault circuit classifier, a fault location classifier, and a fault cause classifier;

[0009] Obtaining a fault circuit identification result through a fault circuit classifier in the fault diagnosis model;

[0010] Inputting the fault loop identification result and the target state quantity data into the fault location classifier in the fault diagnosis model to obtain a fault location identification result;

[0011] Inputting the fault location identification result and the target state quantity data into the fault cause classifier in the fault diagnosis model to obtain a fault cause identification result;

[0012] A fault diagnosis result of the transformer to be diagnosed is obtained according to the fault circuit identification result, the fault location identification result and the fault cause identification result.

[0013] In one embodiment, the filtering of the plurality of state quantity data to obtain target state quantity data includes:

[0014] Obtaining an association between each state quantity data and each preset fault cause in a preset fault cause set;

[0015] quantifying each of the association relationships through kernel principal component analysis to obtain a contribution rate of each of the state quantity data; the contribution rate is used to characterize the importance of the fault information contained in the state quantity data;

[0016] Obtaining a descending sort order of each of the state quantity data according to the contribution rate;

[0017] Calculate the cumulative contribution rate of each state quantity data in turn according to the descending sort order;

[0018] When the cumulative contribution rate is greater than a preset cumulative contribution rate threshold and the number of state quantity data is the smallest, the plurality of state quantity data corresponding to the cumulative contribution rate are determined as target state quantity data.

[0019] In one embodiment, the fusion monitoring system includes a bushing insulation monitoring device, a neutral point DC current monitoring device, an oil dissolved gas detection device, a partial discharge detection device, a winding optical fiber temperature measurement device, and an iron core grounding current detection device;

[0020] The method of obtaining multiple state quantity data of the transformer to be diagnosed through the pre-deployed fusion monitoring system includes:

[0021] The bushing insulation monitoring state quantity data is obtained through the bushing insulation monitoring device; the neutral point DC current monitoring state quantity data is obtained through the neutral point DC current monitoring device; the dissolved gas in oil detection state quantity data is obtained through the dissolved gas in oil detection device; the partial discharge detection state quantity data is obtained through the partial discharge detection device; the winding optical fiber temperature measurement state quantity data is obtained through the winding optical fiber temperature measurement device; and the core grounding current detection state quantity data is obtained through the core grounding current detection device.

[0022] In one embodiment, after screening the plurality of state quantity data to obtain target state quantity data, the method further includes:

[0023] The target state quantity data is discretized according to a preset discretization rule, and the discretized target state quantity data is used as new target state quantity data.

[0024] In one embodiment, the step of training the fault diagnosis model includes:

[0025] Acquiring historical fault diagnosis data, the historical fault diagnosis data including a plurality of historical state quantity data acquired by the fusion monitoring system, as well as a fault circuit determination result, a fault location determination result, and a fault cause determination result;

[0026] Filtering the plurality of historical state quantity data to obtain target historical state quantity data;

[0027] Inputting the target historical state quantity data into the fault diagnosis model to be trained, and obtaining a historical fault loop identification result through the fault loop classifier in the fault diagnosis model;

[0028] Inputting the historical fault loop identification result and the target historical state quantity data into the fault location classifier in the fault diagnosis model to obtain the historical fault location identification result;

[0029] Inputting the historical fault location identification result and the target historical state quantity data into the fault cause classifier in the fault diagnosis model to obtain a historical fault cause identification result;

[0030] According to the difference between the historical fault loop identification result and the fault loop determination result, the difference between the historical fault location identification result and the fault location determination result, and the difference between the historical fault cause identification result and the fault cause determination result, the fault diagnosis model to be trained is trained to obtain a trained fault diagnosis model.

[0031] In one embodiment, the filtering of the plurality of historical state quantity data to obtain target historical state quantity data includes:

[0032] Obtaining a historical correlation between each of the historical state quantity data and the fault cause in the fault cause determination result;

[0033] Quantifying each of the historical association relationships through kernel principal component analysis to obtain a contribution rate of each of the historical state quantity data; the contribution rate of the historical state quantity data is used to characterize the importance of the fault information contained in the historical state quantity data;

[0034] Obtaining a descending sort order of each of the historical state quantity data according to the contribution rate of the historical state quantity data;

[0035] Calculate the historical cumulative contribution rate of each of the historical state quantity data in descending order;

[0036] When the historical cumulative contribution rate is greater than the cumulative contribution rate threshold and the number of historical state quantity data is the smallest, the plurality of historical state quantity data corresponding to the historical cumulative contribution rate are determined as target historical state quantity data.

[0037] In one embodiment, the training of the to-be-trained fault diagnosis model based on the difference between the historical fault loop identification result and the fault loop determination result, the difference between the historical fault location identification result and the fault location determination result, and the difference between the historical fault cause identification result and the fault cause determination result to obtain a trained fault diagnosis model includes:

[0038] respectively obtaining the calculation time of the fault circuit classifier, the fault location classifier, and the fault cause classifier outputting identification results;

[0039] The diagnostic accuracy of the fault circuit classifier, the fault part classifier, and the fault cause classifier are obtained respectively based on the difference between the historical fault circuit identification result and the fault circuit determination result, the difference between the historical fault part identification result and the fault part determination result, and the difference between the historical fault cause identification result and the fault cause determination result;

[0040] By improving the Grey Wolf optimization algorithm, with the goal of maximizing the diagnostic accuracy and minimizing the calculation time, the fault circuit classifier, the fault location classifier, and the fault cause classifier are trained to obtain a trained fault diagnosis model.

[0041] In a second aspect, the present application further provides a power transformer sensor network monitoring and fault diagnosis device, comprising:

[0042] A data acquisition module is used to obtain multiple state quantity data of the transformer to be diagnosed through a pre-deployed fusion monitoring system, and filter the multiple state quantity data to obtain target state quantity data; the fusion monitoring system includes multiple types of sensors, and the multiple types of sensors are respectively used to monitor multiple different physical quantities of the transformer to be diagnosed;

[0043] a fault identification module configured to input the target state quantity data into a pre-trained fault diagnosis model; the fault diagnosis model comprising a fault loop classifier, a fault location classifier, and a fault cause classifier; obtain a fault loop identification result through the fault loop classifier in the fault diagnosis model; input the fault loop identification result and the target state quantity data into the fault location classifier in the fault diagnosis model to obtain a fault location identification result; and input the fault location identification result and the target state quantity data into the fault cause classifier in the fault diagnosis model to obtain a fault cause identification result;

[0044] The diagnosis result acquisition module is used to obtain the fault diagnosis result of the transformer to be diagnosed according to the fault circuit identification result, the fault location identification result and the fault cause identification result.

[0045] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method described in the first aspect when executing the computer program.

[0046] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the method described in the first aspect when the computer program is executed by a processor.

[0047] The above-described power transformer sensor network monitoring and fault diagnosis method, apparatus, computer equipment, and computer-readable storage medium utilize a fusion monitoring system to acquire and filter data from multiple physical quantities, enabling comprehensive and accurate reflection of the transformer's operating status. Furthermore, a multi-layered classifier-based fault diagnosis model employs a progressively refined and in-depth analysis of the fault circuit, fault location, and fault cause, enabling more precise fault location and improving the accuracy of fault diagnosis. Furthermore, by filtering state quantity data and removing redundant data, the complexity of data processing is reduced. Furthermore, the fault diagnosis model employs a hierarchical structure, with each classifier focusing on a specific task, enabling rapid fault identification and diagnosis, thereby improving the efficiency of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments of the present application or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying any creative work.

[0049] Figure 1Schematic diagram of an application environment of a power transformer sensor network monitoring and fault diagnosis method in one embodiment;

[0050] Figure 2 1 is a flow chart of a method for monitoring and diagnosing faults in a power transformer sensor network according to an embodiment;

[0051] Figure 3 1. A flowchart of steps for screening a plurality of the above-mentioned state quantity data to obtain target state quantity data in one embodiment;

[0052] Figure 4 A schematic diagram of the deployment location of a fusion monitoring system in one embodiment;

[0053] Figure 5 Schematic diagram of the contribution rate of each principal component in one embodiment;

[0054] Figure 6 Schematic diagram of the structure of a diagnostic model composed of a multi-level LSSVM classifier in one embodiment;

[0055] Figure 7 1 is a schematic diagram of an overall diagnostic architecture of a transformer fault diagnosis method based on power sensor network technology in one embodiment;

[0056] Figure 8 is a structural block diagram of a power transformer sensor network monitoring and fault diagnosis device in one embodiment;

[0057] Figure 9 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0059] The power transformer sensor network monitoring and fault diagnosis method provided in the embodiment of the present application can be applied to Figure 1In the application environment shown, power sensor network technology can refer to a comprehensive technology that integrates sensor technology, network communication technology, and information processing technology. In the transformer fault diagnosis scenario, by deploying a large number of sensors on the transformer and its related equipment, these sensors can sense various physical quantities of the transformer in real time and transmit the data via the network to a data processing center for analysis, thereby achieving real-time monitoring of the transformer's operating status and fault diagnosis. The fused monitoring system 102 communicates with the server 104 via the network. The power system can be deployed on the server 104. The fused monitoring system 102 is pre-deployed near the transformer to be diagnosed. The fused monitoring system 102 can include multiple types of sensors, forming multiple different types of monitoring devices. Different monitoring devices use sensors to monitor corresponding physical quantities of the transformer to be diagnosed. Different monitoring devices can be installed or attached to the transformer to be diagnosed. The data storage system can store data that the server 104 needs to process. The data storage system can be integrated with the server 104 or located in the cloud or other network servers. The server 104 obtains multiple state quantity data of the transformer to be diagnosed through the pre-deployed fusion monitoring system 102, filters the multiple state quantity data, and obtains target state quantity data; the above-mentioned fusion monitoring system includes multiple types of sensors, and the multiple types of sensors are respectively used to monitor multiple different physical quantities of the above-mentioned transformer to be diagnosed; the server 104 inputs the above-mentioned target state quantity data into a pre-trained fault diagnosis model; the above-mentioned fault diagnosis model includes a fault loop classifier, a fault location classifier and a fault cause classifier; the server 104 obtains a fault loop identification result through the fault loop classifier in the above-mentioned fault diagnosis model; the above-mentioned fault loop identification result and the above-mentioned target state quantity data are input into the fault location classifier in the above-mentioned fault diagnosis model to obtain a fault location identification result; the above-mentioned fault location identification result and the above-mentioned target state quantity data are input into the fault cause classifier in the above-mentioned fault diagnosis model to obtain a fault cause identification result; the server 104 obtains the fault diagnosis result of the transformer to be diagnosed based on the above-mentioned fault loop identification result, the above-mentioned fault location identification result and the above-mentioned fault cause identification result. The server 104 may be an independent physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services.

[0060] In an exemplary embodiment, Figure 2 As shown, a method for monitoring and fault diagnosis of a power transformer sensor network is provided. Figure 1 The server 104 in the example is used to illustrate the process, which includes the following steps S202 to S212.

[0061] Step S202 : obtaining a plurality of state quantity data of the transformer to be diagnosed through a pre-deployed fusion monitoring system, screening the plurality of state quantity data, and obtaining target state quantity data.

[0062] The fusion monitoring system includes multiple types of sensors, each used to monitor multiple physical quantities of the transformer under diagnosis. This integrated system provides comprehensive, real-time monitoring of the operating status of the transformer under diagnosis and can be pre-deployed on the transformer under diagnosis. Different types of sensors can monitor different physical quantities of the transformer, such as temperature sensors monitoring transformer temperature, current sensors monitoring electrical parameters, and gas sensors performing dissolved gas analysis. By integrating data from these different sensor types, the transformer's operating status can be more accurately reflected.

[0063] Among them, state quantity data can refer to various data that can reflect the operating status of the transformer, including but not limited to temperature, electrical parameters (such as current, voltage, resistance, etc.), dissolved gas analysis results (such as the content of gases such as hydrogen, methane, and ethylene), infrared images, etc.

[0064] The target state data may refer to data closely related to fault diagnosis that is screened out from multiple state data sets. The purpose of screening is to remove redundant data, reduce the complexity of data processing, and improve the accuracy and efficiency of fault diagnosis.

[0065] For example, server 104 uses multiple sensors pre-deployed on the transformer within the fusion monitoring system to collect real-time data on multiple physical quantities of the transformer to be diagnosed. For example, a temperature sensor records the transformer's temperature at regular intervals, while a current sensor monitors the current in real time. These sensors transmit the collected data to the fusion monitoring system's data processing center, where it is further processed and uploaded to server 104. Server 104 then filters the acquired state quantity data. Data mining and feature selection methods can be used to select data closely related to fault diagnosis from a large amount of state quantity data based on factors such as its relevance to fault diagnosis and its reliability, thereby obtaining target state quantity data. For example, by analyzing historical fault data, if it is found that certain electrical parameters change significantly when a fault occurs, while others change only slightly, the parameters with significant changes can be selected as target state quantity data.

[0066] Step S204 , inputting the target state quantity data into a pre-trained fault diagnosis model; the fault diagnosis model includes a fault circuit classifier, a fault location classifier, and a fault cause classifier.

[0067] The fault diagnosis model can be a model for diagnosing transformer faults, consisting of a fault circuit classifier, a fault location classifier, and a fault cause classifier. By learning and training from historical fault data, the model can identify and determine the transformer's fault circuit, fault location, and fault cause based on the input target state data.

[0068] For example, the server 104 inputs the filtered target state quantity data into a pre-trained fault diagnosis model. Before inputting the data, it may be necessary to pre-process the data, such as normalization, standardization, etc., to ensure that the format and range of the data meet the input requirements of the fault diagnosis model.

[0069] Step S206 , obtaining a fault loop identification result through the fault loop classifier in the above fault diagnosis model.

[0070] Among them, the fault circuit classifier can refer to a component in the fault diagnosis model. Its main function is to identify the fault circuit of the transformer based on the input target state quantity data, for example, to determine whether the fault occurs in the electrical circuit, magnetic circuit or oil circuit.

[0071] For example, after receiving target state data, the fault circuit classifier analyzes and processes the data using its internal algorithms and models (such as support vector machines and neural networks). Based on pre-learned fault characteristics and patterns, the classifier determines the transformer's fault circuit. For example, if abnormal changes in certain electrical parameters indicate a problem with current transmission, the fault may be in the electrical circuit; if the oil temperature is too high and the gas composition in the oil is abnormal, the fault may be in the oil circuit.

[0072] Step S208 : inputting the fault loop identification result and the target state quantity data into the fault location classifier in the fault diagnosis model to obtain a fault location identification result.

[0073] Among them, the fault location classifier can refer to a component in the fault diagnosis model. Its main function is to receive the fault circuit identification results and target state quantity data, and further determine the specific location of the transformer fault, such as the winding, tap changer, iron core, etc.

[0074] For example, the fault circuit identification results and target state data are simultaneously input into a fault location classifier. The fault location classifier combines the fault circuit information and target state data to further narrow the fault scope and determine the specific location of the fault. For example, if the fault circuit identification result indicates a circuit fault, the fault location classifier will determine whether the fault occurs in the winding, tap changer, or other circuit components based on the specific changes in electrical parameters and the structural characteristics of the transformer.

[0075] Step S210 , inputting the fault location identification result and the target state quantity data into the fault cause classifier in the fault diagnosis model to obtain a fault cause identification result.

[0076] Among them, the fault cause classifier can refer to the last classifier in the fault diagnosis model. Its main function is to analyze the specific causes of transformer failure, such as inter-turn short circuit, winding deformation, poor contact of tap changer, etc., based on the fault location identification results and target state quantity data.

[0077] For example, the fault location identification results and target state data are input into a fault cause classifier. Based on the fault location information and target state data, the fault cause classifier conducts an in-depth analysis of the specific cause of the fault. For example, if the fault location identification result is a winding fault, the fault cause classifier determines whether the fault is caused by a turn-to-turn short circuit, winding overheating, or another cause based on winding temperature, resistance, partial discharge, and other data.

[0078] Step S212: Obtain the fault diagnosis result of the transformer to be diagnosed based on the fault circuit identification result, the fault location identification result, and the fault cause identification result.

[0079] Specifically, server 104 integrates the fault circuit identification results, fault location identification results, and fault cause identification results to obtain a final fault diagnosis result for the transformer to be diagnosed. This result clearly informs the operation and maintenance personnel of the transformer fault situation, including the fault circuit, location, and cause, providing a basis for subsequent maintenance and treatment.

[0080] The aforementioned power transformer sensor network monitoring and fault diagnosis method utilizes a fusion monitoring system to acquire and filter data from multiple physical quantities, enabling comprehensive and accurate reflection of the transformer's operating status. Furthermore, a multi-layered fault diagnosis model employs a progressively more detailed analysis of the fault circuit, fault location, and fault cause, enabling more precise fault location and improving the accuracy of fault diagnosis. Furthermore, by filtering state quantity data and removing redundant data, the complexity of data processing is reduced. Furthermore, the fault diagnosis model employs a hierarchical structure, with each classifier focusing on a specific task. This allows for rapid fault identification and diagnosis, improving the efficiency of fault diagnosis.

[0081] In an exemplary embodiment, Figure 3 As shown, the above-mentioned state quantity data are screened to obtain target state quantity data, including:

[0082] Step S302: obtaining the association between each of the above-mentioned state quantity data and each preset fault cause in the preset fault cause set.

[0083] The preset fault cause set can be a pre-defined set that includes various possible causes of transformer failure. For example, it could include a winding short circuit, poor core grounding, or a tap changer failure. This set can be determined based on historical transformer failure data, industry experience, and relevant technical standards.

[0084] Among them, the association relationship can refer to the connection between each state quantity data and each preset fault cause in the preset fault cause set. For example, the high oil temperature of the transformer may be associated with preset fault causes such as winding short circuit and cooling system failure, and the correlation is the association relationship.

[0085] For example, a large amount of historical transformer fault data can be collected. This data includes state data (such as temperature, current, voltage, etc.) and the corresponding fault causes. Then, using data mining and statistical analysis methods, the correlation between each state data item and each preset fault cause in the preset fault cause set can be analyzed. For example, correlation coefficients can be calculated, and causal analysis and kernel principal component analysis can be performed to determine the degree of correlation between each state data item and each preset fault cause.

[0086] Step S304 , quantifying each of the above-mentioned association relationships through kernel principal component analysis to obtain the contribution rate of each of the above-mentioned state quantity data.

[0087] The contribution rate is used to represent the importance of the fault information contained in the state data. A higher contribution rate indicates that the state data is more useful for fault diagnosis, contains more fault-related information, or is more effective in representing the fault information.

[0088] Kernel Principal Component Analysis (KPCA) is a nonlinear data dimensionality reduction technique that uses a kernel function to map raw data into a high-dimensional space. Principal component analysis is then performed within that space to identify key features and structures within the data. In this embodiment, KPCA is used to quantify the correlation between state data and pre-defined fault causes.

[0089] For example, a Gaussian kernel function can be used, and its hyperparameter γ determines the distribution of data in high-dimensional space. The larger the γ value, the more concentrated the data points. In this embodiment, the hyperparameter γ can be set to 2, and the value range is 10 -3 to 10 3 As described above, through kernel principal component analysis, the correspondence between each state quantity data and the preset fault cause is quantified, and the contribution rate of each principal component (i.e., the proportion of its explained variance to the total variance) is calculated and displayed.

[0090] Step S306: Obtain the descending sort order of each of the above-mentioned state quantity data according to the contribution rate.

[0091] Step S308 : accumulating and calculating the cumulative contribution rate of each of the above-mentioned state quantity data in sequence according to the above-mentioned descending sorting order.

[0092] Step S310 : when the cumulative contribution rate is greater than a preset cumulative contribution rate threshold and the number of state quantity data is the smallest, determining the multiple state quantity data corresponding to the cumulative contribution rate as target state quantity data.

[0093] Among them, the cumulative contribution rate can refer to the value obtained by sorting the state quantity data in descending order according to the contribution rate, and then accumulating the contribution rate of each state quantity data in turn; it reflects the proportion of the sum of the fault information contained in the previous several state quantity data to the fault information of all state quantity data.

[0094] Among them, the cumulative contribution rate threshold can refer to a pre-set value, which is used to determine how many state quantity data are selected as target state quantity data; when the cumulative contribution rate is greater than this threshold, it means that enough state quantity data have been selected, and these data can contain most of the information related to fault diagnosis.

[0095] For example, all state quantity data can be sorted from large to small according to their contribution rate to form a list in descending order; this way, the importance of each state quantity data to fault diagnosis can be clearly seen; starting from the state quantity data with the largest contribution rate, the contribution rate of each state quantity data is accumulated in turn to obtain the cumulative contribution rate. For example, the cumulative contribution rate of the first state quantity data is its own contribution rate, the cumulative contribution rate of the first two state quantity data is the contribution rate of the first state quantity data plus the contribution rate of the second state quantity data, and so on. Obtain a pre-set cumulative contribution rate threshold, such as 90% or 95%; then, from the list of state quantity data sorted in descending order, accumulate the cumulative contribution rates in turn. When the cumulative contribution rate is greater than the preset cumulative contribution rate threshold for the first time, record the number of state quantity data contained at this time, that is, ensure that under the condition that the cumulative contribution rate is greater than the threshold, select the combination with the smallest number of state quantity data. These state quantity data are the final target state quantity data.

[0096] In this embodiment, by screening out state quantity data with a high correlation degree and a large contribution rate to the preset fault cause as the target state quantity data, the fault condition of the transformer can be reflected more accurately, the interference of redundant data is avoided, and the accuracy of fault diagnosis is improved.

[0097] In an exemplary embodiment, Figure 4As shown, the above-mentioned integrated monitoring system includes a bushing insulation monitoring device, a neutral point DC current monitoring device, an oil-dissolved gas detection device, a partial discharge detection device, a winding optical fiber temperature measurement device, and a core grounding current detection device; the above-mentioned pre-deployed integrated monitoring system obtains multiple state quantity data of the transformer to be diagnosed, including:

[0098] The bushing insulation monitoring state quantity data is obtained through the above-mentioned bushing insulation monitoring device; the neutral point DC current monitoring state quantity data is obtained through the above-mentioned neutral point DC current monitoring device; the dissolved gas in oil detection state quantity data is obtained through the above-mentioned dissolved gas in oil detection device; the partial discharge detection state quantity data is obtained through the above-mentioned partial discharge detection device; the winding optical fiber temperature measurement state quantity data is obtained through the above-mentioned winding optical fiber temperature measurement device; and the core grounding current detection state quantity data is obtained through the above-mentioned core grounding current detection device.

[0099] Among them, bushing insulation monitoring device, neutral point DC current monitoring device, oil dissolved gas detection device, partial discharge detection device, winding optical fiber temperature measurement device and core grounding current detection device can be deployed in such places as Figure 4 In the transformer environment shown.

[0100] In this embodiment, multi-dimensional fault characteristics are obtained through the bushing insulation monitoring device, the neutral point DC current monitoring device, the dissolved gas in oil detection device, the partial discharge detection device, the winding optical fiber temperature measurement device and the core grounding current detection device, so that when diagnosing the fault, multiple state quantity data collected by multiple devices can be cross-verified, thereby improving the accuracy of fault diagnosis.

[0101] In an exemplary embodiment, after screening the plurality of state quantity data to obtain target state quantity data, the method further includes:

[0102] The target state quantity data is discretized according to a preset discretization rule, and the discretized target state quantity data is used as the new target state quantity data.

[0103] Among them, the preset discretization rules can refer to pre-defined standards or methods for guiding the discretization process, which are usually formulated based on domain knowledge, historical data distribution or algorithm requirements. For example: equal-interval binning: dividing the data range into intervals of equal width (such as the temperature of 0~100℃ is divided into intervals of every 20℃) or equal-frequency binning: making each interval contain the same number of data points, or, threshold method: setting the critical value based on expert experience (such as acetylene concentration in oil >5μL / L is considered a fault interval), etc.

[0104] Discretization refers to the process of converting continuous state data into discrete data. For example, continuous temperature values can be divided into intervals such as "low," "medium," and "high," or dissolved gas concentrations in oil can be classified into categories such as "normal," "warning," and "fault." Discretization can reduce data complexity and improve the model's ability to handle nonlinear relationships.

[0105] Exemplarily, a pre-set discretization rule is obtained, and each continuous target state quantity data is mapped to the corresponding discrete category according to the preset rule. The discretized data is presented in the form of category or interval (such as using 0, 1, and 2 to represent different intervals), replacing the original continuous data, as the input of the subsequent fault diagnosis model.

[0106] In this embodiment, the target state quantity data is discretized by using preset discretization rules to eliminate the impact of differences in dimension and evaluation direction and ensure the consistency and comparability of the data.

[0107] In an exemplary embodiment, the training step of the fault diagnosis model may include the following steps:

[0108] Step S11 , obtaining historical fault diagnosis data, the historical fault diagnosis data including a plurality of historical state quantity data obtained by the fusion monitoring system, as well as a fault circuit determination result, a fault location determination result and a fault cause determination result.

[0109] Step S12, screening the plurality of historical state quantity data to obtain target historical state quantity data;

[0110] Step S13: input the above-mentioned target historical state quantity data into the fault diagnosis model to be trained, and obtain the historical fault loop identification result through the fault loop classifier in the above-mentioned fault diagnosis model.

[0111] Step S14 , inputting the above historical fault circuit identification result and the above target historical state quantity data into the fault location classifier in the above fault diagnosis model to obtain the historical fault location identification result.

[0112] Step S15 , inputting the above historical fault location identification result and the above target historical state quantity data into the fault cause classifier in the above fault diagnosis model to obtain the historical fault cause identification result.

[0113] Step S16, based on the difference between the above-mentioned historical fault loop identification results and the above-mentioned fault loop determination results, the difference between the above-mentioned historical fault location identification results and the above-mentioned fault location determination results, and the difference between the above-mentioned historical fault cause identification results and the above-mentioned fault cause determination results, the above-mentioned fault diagnosis model to be trained is trained to obtain a trained fault diagnosis model.

[0114] The historical fault diagnosis data may refer to a dataset accumulated in the past and used to train the fault diagnosis model. After filtering the original historical state data (e.g., through kernel principal component analysis), the key data with high contribution rate and low redundancy is retained for model training.

[0115] In this embodiment, a fault diagnosis model based on historical data is constructed through a closed-loop process of "data collection → screening → hierarchical prediction → error feedback training". Its core advantages lie in the high efficiency of hierarchical diagnosis, the scientific nature of data screening and the automation of model training, which ultimately achieves accurate and rapid diagnosis of transformer faults.

[0116] In an exemplary embodiment, the above-mentioned multiple historical state quantity data are screened to obtain target historical state quantity data, including: obtaining the historical correlation relationship between each of the above-mentioned historical state quantity data and the fault cause in the above-mentioned fault cause determination result; quantifying each of the above-mentioned historical correlation relationships through kernel principal component analysis to obtain the contribution rate of each of the above-mentioned historical state quantity data; the contribution rate of the above-mentioned historical state quantity data is used to characterize the importance of the fault information contained in the above-mentioned historical state quantity data; obtaining the descending sorting order of each of the above-mentioned historical state quantity data according to the size of the contribution rate of the above-mentioned historical state quantity data; according to the descending sorting order of each of the above-mentioned historical state quantity data, sequentially accumulating and calculating the historical cumulative contribution rate of each of the above-mentioned historical state quantity data; when the above-mentioned historical cumulative contribution rate is greater than the above-mentioned cumulative contribution rate threshold and the number of historical state quantity data is the smallest, determining that the multiple historical state quantity data corresponding to the above-mentioned historical cumulative contribution rate are the target historical state quantity data.

[0117] The process of filtering data has been described in detail above and will not be repeated here.

[0118] In this implementation, the historical correlation between historical state data and the fault causes identified in the fault cause determination results is first obtained. Kernel principal component analysis is then used to quantify the nonlinear correlation between the state data and the fault. This method, combined with a cumulative contribution rate threshold and the minimum data size principle, filters key historical data, achieving efficient extraction of high-dimensional, nonlinear fault features. This method reduces data redundancy while retaining core fault information, significantly improving the training efficiency and diagnostic accuracy of the fault diagnosis model and providing a scientific feature engineering solution for the accurate identification of transformer faults.

[0119] In an exemplary embodiment, the above-mentioned fault diagnosis model to be trained is trained based on the difference between the above-mentioned historical fault circuit identification result and the above-mentioned fault circuit determination result, the difference between the above-mentioned historical fault location identification result and the above-mentioned fault location determination result, and the difference between the above-mentioned historical fault cause identification result and the above-mentioned fault cause determination result to obtain a trained fault diagnosis model, including the following steps:

[0120] Step S21 , respectively obtaining the calculation time for the fault circuit classifier, the fault location classifier, and the fault cause classifier to output identification results.

[0121] Step S22, based on the difference between the above-mentioned historical fault circuit identification results and the above-mentioned fault circuit determination results, the difference between the above-mentioned historical fault part identification results and the above-mentioned fault part determination results, and the difference between the above-mentioned historical fault cause identification results and the above-mentioned fault cause determination results, the diagnostic accuracy of the above-mentioned fault circuit classifier, the above-mentioned fault part classifier, and the above-mentioned fault cause classifier are respectively obtained.

[0122] Step S23 , by improving the Grey Wolf optimization algorithm, with the goal of maximizing the diagnostic accuracy and minimizing the calculation time, the fault circuit classifier, the fault location classifier, and the fault cause classifier are trained to obtain a trained fault diagnosis model.

[0123] Among them, the Improved Grey Wolf Optimizer (IGWO) can refer to a heuristic optimization algorithm based on the hunting behavior of wolf packs. It optimizes the objective function by simulating the leadership hierarchy and hunting mechanism of wolf packs. Compared with the standard gray wolf optimization algorithm, it may improve the optimization efficiency and avoid premature convergence by adjusting the convergence factor, introducing mutation operations or adaptive strategies. It is suitable for multi-objective optimization problems (such as optimizing accuracy and computing time at the same time).

[0124] Among them, the diagnostic accuracy can refer to the proportion of samples in which the model correctly identifies the fault type (circuit, location, cause) to the total number of samples, which is used to measure the classification accuracy of the model.

[0125] Among them, computing time can refer to the time required for the model to complete a fault diagnosis (from input data to output results), reflecting the reasoning efficiency and real-time performance of the model.

[0126] For example, during training, historical state data is input to each classifier, and a timer records the time interval (e.g., in milliseconds) from data input to recognition output. The computation time of multiple samples is averaged and used as the computation time indicator for the classifier. The historical fault circuit identification results are compared with the actual fault circuit determination results to calculate the circuit classification accuracy. Similarly, the classification accuracy of the fault location and cause can be calculated. A multi-objective optimization function is then defined, simultaneously maximizing diagnostic accuracy and minimizing computation time. The optimization variables are key classifier parameters (e.g., the learning rate, number of layers, or nodes in a neural network, or kernel function parameters in a support vector machine). The algorithm may include initializing the wolf pack: randomly generating a set of parameter combinations (wolf individuals), with each individual corresponding to a candidate model. Fitness calculation: For each individual, the accuracy and computation time of the corresponding classifier are calculated on the historical data and substituted into the objective function to obtain the fitness value. Wolf pack iteration: The positions (parameters) of individual wolves are updated through operations such as "surrounding prey," "pursuing prey," and "attacking prey," gradually approaching the optimal solution. Termination condition: When the fitness value converges or reaches the maximum number of iterations, the optimization is stopped and the optimal parameter combination is taken as the trained model.

[0127] This embodiment simultaneously optimizes diagnostic accuracy and computation time, avoiding model overfitting and computational inefficiency that can occur with traditional single-metric optimization (e.g., focusing solely on accuracy). This allows the model to achieve both high precision and real-time performance in practical applications. Furthermore, improvements to the Grey Wolf Optimization algorithm (e.g., adaptive adjustment and mutation operations) prevent traditional optimization algorithms from falling into local optima, enabling the model to maintain stable diagnostic performance under varying operating conditions. This makes it particularly suitable for complex diagnostic scenarios involving heterogeneous data from multiple sources in power equipment.

[0128] With the rapid development of intelligent technologies, the use of artificial intelligence algorithms, such as machine learning and deep learning, for power transformer fault diagnosis has become a new research direction. These intelligent algorithms can analyze transformer faults from multiple dimensions using extensive historical and real-time monitoring data, thereby improving the accuracy and real-time nature of diagnosis. However, single physical quantity data (such as temperature, electrical parameters, or gas analysis data) often cannot provide sufficient information to fully reflect the transformer's fault status. Therefore, the use of multi-physical quantity fusion technology, combining multiple data sources such as temperature, electrical parameters, and gas analysis, can provide more comprehensive and accurate information for fault diagnosis.

[0129] While some research has been conducted on multi-physics fusion, most methods still rely on a single classification algorithm, lack multi-level progressive fault analysis, and have limitations, such as low classification accuracy and inability to effectively locate fault locations. Therefore, achieving efficient fusion of multi-physics data in power transformer fault diagnosis and utilizing advanced algorithms (such as the improved Grey Wolf Optimization algorithm and least squares support vector machines) for multi-level, automated fault diagnosis and location remains an urgent challenge.

[0130] In an exemplary embodiment, the present application adopts multi-physical quantity fusion technology to integrate multiple physical quantity data such as temperature, electrical parameters, dissolved gas analysis, infrared images, etc. of the power transformer, extract key information through advanced dimensionality reduction technology (such as kernel principal component analysis, KPCA), and combine the improved grey wolf optimization algorithm (IGWO) with the least squares support vector machine (LSSVM) for fault diagnosis. Kernel principal component analysis is used to reduce the redundancy of data and improve the expressive power of features. IGWO optimizes the hyperparameters of LSSVM and improves classification accuracy. The diagnosis process adopts a multi-layer progressive classification model to ensure the accuracy and real-time performance of fault diagnosis by refining the fault circuit, fault location and fault cause layer by layer. This method can effectively identify and locate the fault type of the power transformer and provide more accurate fault prediction and protection measures. The present application also provides a transformer fault diagnosis method based on power sensor network technology, which can specifically include:

[0131] By deploying multiple types of sensors, the operating status of the power transformer is fully sensed and monitored, such as Figure 4 As shown in the figure, the main data collected includes neutral point DC current, core grounding current, partial discharge signal, dissolved gas content in oil, fiber optic winding temperature, and bushing insulation status. Neutral point DC current monitoring can identify abnormal operating conditions caused by DC system bias; core grounding current monitoring can determine whether there are faults such as multiple grounding points in the core; partial discharge monitoring can proactively detect potential breakdown risks in the transformer insulation system; dissolved gas analysis (DGA) detects the concentration of gases such as hydrogen, acetylene, and ethylene in the oil to identify typical transformer faults such as overheating, partial discharge, and arcing; fiber optic winding temperature measurement accurately collects hotspot temperatures, effectively reflecting the transformer's load status; and bushing insulation monitoring determines whether the high-voltage bushing is at risk of degradation or breakdown. These multi-source data are integrated and feature extracted by the data fusion processing unit to construct a complete state perception model, providing a comprehensive and accurate basis for subsequent fault diagnosis and location.

[0132] The accuracy of fault location and cause diagnosis is further improved through multi-physics fusion technology. To address the high classification accuracy of the least squares support vector machine (LSSVM) in binary classification tasks, a strategy of increasing the model layer and reducing the number of categories per layer is adopted. This reduces the likelihood of close distributions of multiple categories within a large category and avoids misclassification. This strategy effectively improves the accuracy of fault diagnosis and enables progressive hierarchical diagnosis.

[0133] This example uses kernel principal component analysis (KPCA) to reduce data dimensionality, remove redundant information, and select the 10 most representative characteristic state variables. These characteristics can better reflect the operating status of the transformer. Furthermore, the selected features are discretized according to DL / T1685-2017, "Guidelines for Condition Assessment of Oil-Immersed Transformers."

[0134] In order to improve the accuracy of fault diagnosis, this chapter adopts the LSSVM model (i.e., fault diagnosis model) optimized based on the improved grey wolf optimization algorithm (IGWO), and divides the fault diagnosis into three layers, with a total of 7 classifiers for automatic diagnosis: the first layer (fault circuit classifier) is responsible for identifying the four types of transformer faults - normal state, circuit fault, magnetic circuit fault and oil circuit fault; the second layer (fault location classifier) is used to determine the specific location of the fault, including winding, tap changer, iron core, metal accessories, screen, oil channel, insulating oil paper and other 7 parts; the third layer (fault cause classifier) is further refined to 11 possible fault causes, including winding strands, layers, inter-turn short circuit, winding deformation, poor contact of tap changer, etc.

[0135] In this model, by using an if loop, samples can be automatically passed to the next layer of classifiers for step-by-step diagnosis, and the prediction results for each sample are verified to be consistent with the actual results. Finally, the diagnostic results of this model are compared with other methods and verified with actual fault cases, demonstrating the effectiveness of the proposed method and highlighting its advantages in overall fault diagnosis.

[0136] When relying solely on DGA data for power transformer status diagnosis, it is usually only possible to distinguish overheating or discharge faults, and it is difficult to determine whether the fault involves solid insulation. Such diagnostic results have certain limitations and one-sidedness. By integrating multiple state quantities for diagnosis, the fault circuit and fault location can be more deeply identified, and the specific cause of the fault can be analyzed. A hierarchical and progressive fault diagnosis model proposed in this embodiment is designed to automate diagnosis and significantly improve the precision and accuracy of fault diagnosis. The state quantities and their numbers obtained by power transformers through various tests are shown in the table below.

[0137]

[0138] In various tests for power transformer fault diagnosis, various state variables interact with each other, and the same fault may manifest itself in different test data. Relying on all state variables to determine the transformer's operating status increases the workload for testers. Furthermore, redundant information can affect fault location and condition assessment, increasing algorithm complexity and computation time. This can also lead to inaccurate or contradictory diagnostic results, reducing diagnostic efficiency and accuracy.

[0139] In order to improve this problem, this embodiment uses kernel principal component analysis (KPCA) to perform dimensionality reduction processing on 24 power transformer state quantities based on the collected common transformer fault data. The principle of KPCA is to screen out linearly independent features from a set of possibly related variables through orthogonal transformation, remove redundant data, and enhance the independence of information. The kernel function maps the original data to a high-dimensional space to ensure the separability of the data. Common kernel functions include linear kernels, polynomial kernels, and Gaussian kernels. This embodiment uses the Gaussian kernel function, whose hyperparameter γ determines the distribution of data in high-dimensional space. The larger the γ value, the more concentrated the data points. This embodiment sets γ to 2, and the value range is 10-3 to 103. When analyzing the output results of the third layer (i.e., 11 fault causes), linearly independent state quantities are screened out and dimensionality reduction processing is performed. Through KPCA analysis, the correspondence between each state quantity and the fault cause is quantified, and the contribution rate of each principal component (state quantity) (i.e., the proportion of its explained variance to the total variance) is calculated and displayed, as shown in the figure. Figure 5 The contribution rate is used to determine the dimension after dimensionality reduction. The results show that the cumulative contribution rate of the first 10 principal components has reached more than 94%. As shown in the table below, the state quantity after dimensionality reduction can effectively represent fault information.

[0140]

[0141] In the above table, due to the different voltage levels of power transformers, the precautionary values of water content in transformer oil are different. The * of water content in oil indicates the per-unit value, that is, test value / cautionary value.

[0142] Further analysis revealed that while the first-level fault circuit classification might yield ambiguous diagnostic results due to the wide range of faults, the state variables filtered out for the third-level fault causes were sufficient to complete fault classification at that level. The second-level fault location was closely related to the third-level fault cause, and the state variables filtered out for the third-level fault cause also met the second-level classification requirements, thereby improving the accuracy and efficiency of the overall diagnostic process.

[0143] Because the magnitudes of various state quantities in power transformers vary and their corresponding evaluation criteria are not uniform, the absorption ratio, for example, is a highly preferred indicator, with "larger values are preferred," while most other characteristic state quantities are less preferred, with "smaller values are preferred." Therefore, in subsequent modeling and analysis, the ten selected characteristic state quantities must be uniformly discretized to eliminate the impact of differences in dimensionality and evaluation direction and ensure data consistency and comparability. The specific discretization rules refer to the division criteria given in the table below for conversion.

[0144]

[0145] In the table above, due to different voltage levels of power transformers, the cautionary values for water content in transformer oil vary. The * for water content in oil indicates the per-unit value, i.e., test value divided by cautionary value. Discrete coding can be considered a score, with a code of 1 being the best and a code of 3 being the worst.

[0146] There are various ways to categorize power transformer faults. Based on the device structure, they can be divided into internal and external faults; based on the functional circuit, they include circuit faults, magnetic circuit faults, and oil circuit faults; based on the specific location, they can be further subdivided into winding faults, core faults, and tap changer faults; and based on the nature of the fault manifestation, they can generally be classified as overheating faults and discharge faults. As the number of classifications continues to increase, the accuracy of model recognition is declining, and the boundaries between different categories are becoming increasingly blurred.

[0147] In order to improve the classification effect, this embodiment constructs a step-by-step diagnosis model based on multi-level LSSVM. By increasing the number of model layers and reducing the number of classification categories in each layer, the diagnostic tasks are gradually refined, effectively improving the distinguishability of each category. This hierarchical diagnosis strategy not only conforms to the physical characteristics of the evolution of power transformer faults, but also fits the maintenance logic in actual operation and maintenance. Specifically, this embodiment designs a diagnostic model consisting of 7 LSSVM classifiers optimized by the improved grey wolf optimization algorithm (IGWO). During the operation of the model, all samples are imported at one time, and the samples are passed to the next level classifier layer by layer for judgment according to the diagnosis results based on the if judgment logic. Each layer of the model can independently output the corresponding diagnostic accuracy and calculation time, which is convenient for performance evaluation. The entire diagnostic system consists of three layers:

[0148] The first layer (fault circuit classifier): fault circuit identification, performed by IGWO-LSSVM1, classifies input samples into four categories: normal state, circuit fault, magnetic circuit fault, and oil circuit fault;

[0149] The second layer (fault location classifier): Fault location location. IGWO-LSSVM2 is used to subdivide circuit faults into winding faults and tap changer faults. IGWO-LSSVM3 further divides magnetic circuit faults into core faults and metal accessory faults. IGWO-LSSVM4 subdivides oil circuit faults into enclosure, oil channel, and insulation oil paper faults.

[0150] The third layer (fault cause classifier) analyzes the fault cause and conducts in-depth diagnosis based on the specific location. IGWO-LSSVM5 classifies winding faults into strand short circuits, interlayer short circuits, interturn short circuits, and winding deformation. IGWO-LSSVM6 identifies suspended discharge, magnetic flux leakage / magnetic shielding heating, and bare metal heating in metal accessory faults. IGWO-LSSVM7 further identifies insulation oil-paper faults as aging or moisture.

[0151] The model relies on comprehensive analysis of DGA data, electrical testing and oil-chemical experiments to achieve automatic fault diagnosis from shallow to deep and layer by layer. Figure 6 In the multi-level structure shown in the figure, during the model training process, the initial population of the improved gray wolf optimization algorithm is set to 20, and the maximum number of evolutionary iterations is 100. In the parameter optimization stage, the search range of the penalty coefficient C is set to [0.01, 500], and the search range of the kernel function parameter g is set to [0.01, 100] to ensure that the model achieves the best balance between diagnostic accuracy and computational efficiency.

[0152] Compared with the existing technology, this application has significant advantages in fault diagnosis accuracy, automation level and adaptability. The overall schematic diagram of its diagnostic architecture is as follows: Figure 7As shown in the figure. First, by integrating multi-physical quantity information and optimization algorithms, the accuracy and reliability of fault identification are significantly improved, enabling comprehensive and accurate diagnosis of transformer internal faults from type to location and cause. Second, the diagnostic process implements fully automated progressive analysis, greatly improving the efficiency of fault detection. The improved Grey Wolf algorithm accelerates model convergence and supports online real-time diagnosis, facilitating early warning of faults. Third, the method of this application adapts to complex operating conditions. Even in the case of complex faults caused by multiple stresses, the cross-correlation of multi-source data ensures the reliability of diagnostic results. This technology can effectively compensate for the inability of traditional methods to locate faults, assisting operation and maintenance personnel in taking timely and targeted maintenance measures, preventing fault escalation, and improving the safety and stability of power transformer operation. This application has high engineering application value and can be embedded in power transformer online monitoring and condition assessment systems to achieve unmanned intelligent diagnosis. Through its promotion and application in power grid substations, it can significantly improve the automation and standardization of transformer fault diagnosis, reduce power outage maintenance time and economic losses, and ensure the safe and reliable operation of the power system. In summary, this application has important engineering value and broad promotion prospects in the safe operation of power systems, and will provide strong support for the intelligent operation and maintenance of power equipment.

[0153] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0154] Based on the same inventive concept, embodiments of the present application also provide a power transformer sensor network monitoring and fault diagnosis device for implementing the aforementioned power transformer sensor network monitoring and fault diagnosis method. The solution provided by this device is similar to the solution described in the aforementioned method. Therefore, the specific limitations of one or more embodiments of the power transformer sensor network monitoring and fault diagnosis device provided below can be found in the above-described limitations of the power transformer sensor network monitoring and fault diagnosis method and will not be further elaborated here.

[0155] In an exemplary embodiment, Figure 8As shown, a power transformer sensor network monitoring and fault diagnosis device 900 is provided, comprising: a data acquisition module 901, a fault identification module 902 and a diagnosis result acquisition module 903, wherein:

[0156] The data acquisition module 901 is used to obtain multiple state quantity data of the transformer to be diagnosed through a pre-deployed fusion monitoring system, and filter the multiple state quantity data to obtain target state quantity data; the fusion monitoring system includes multiple types of sensors, and the multiple types of sensors are respectively used to monitor multiple different physical quantities of the transformer to be diagnosed;

[0157] Fault identification module 902 is configured to input the target state quantity data into a pre-trained fault diagnosis model; the fault diagnosis model includes a fault loop classifier, a fault location classifier, and a fault cause classifier; obtain a fault loop identification result through the fault loop classifier in the fault diagnosis model; input the fault loop identification result and the target state quantity data into the fault location classifier in the fault diagnosis model to obtain a fault location identification result; and input the fault location identification result and the target state quantity data into the fault cause classifier in the fault diagnosis model to obtain a fault cause identification result.

[0158] The diagnosis result acquisition module 903 is used to obtain the fault diagnosis result of the transformer to be diagnosed according to the fault circuit identification result, the fault location identification result and the fault cause identification result.

[0159] In an exemplary embodiment, the above-mentioned data acquisition module 901 is also used to obtain the correlation relationship between each of the above-mentioned state quantity data and each preset fault cause in the preset fault cause set; quantify each of the above-mentioned correlation relationships through kernel principal component analysis to obtain the contribution rate of each of the above-mentioned state quantity data; the above-mentioned contribution rate is used to characterize the importance of the fault information contained in the above-mentioned state quantity data; obtain the descending sorting order of each of the above-mentioned state quantity data according to the size of the contribution rate; according to the above-mentioned descending sorting order, accumulate and calculate the cumulative contribution rate of each of the above-mentioned state quantity data in sequence; when the above-mentioned cumulative contribution rate is greater than the preset cumulative contribution rate threshold and the number of state quantity data is the smallest, determine that the multiple state quantity data corresponding to the above-mentioned cumulative contribution rate are the target state quantity data.

[0160] In an exemplary embodiment, the above-mentioned fusion monitoring system includes a bushing insulation monitoring device, a neutral point DC current monitoring device, a dissolved gas in oil detection device, a partial discharge detection device, a winding optical fiber temperature measurement device, and a core grounding current detection device;

[0161] In an exemplary embodiment, the above-mentioned data acquisition module 901 is also used to obtain bushing insulation monitoring state quantity data through the above-mentioned bushing insulation monitoring device; obtain neutral point DC current monitoring state quantity data through the above-mentioned neutral point DC current monitoring device; obtain dissolved gas in oil detection state quantity data through the above-mentioned dissolved gas in oil detection device; obtain partial discharge detection state quantity data through the above-mentioned partial discharge detection device; obtain winding optical fiber temperature measurement state quantity data through the above-mentioned winding optical fiber temperature measuring device; and obtain core grounding current detection state quantity data through the above-mentioned core grounding current detection device.

[0162] In an exemplary embodiment, the data acquisition module 901 is further configured to discretize the target state quantity data according to a preset discretization rule, and use the discretized target state quantity data as new target state quantity data.

[0163] In an exemplary embodiment, the power transformer sensor network monitoring and fault diagnosis device 900 further includes a model training module for acquiring historical fault diagnosis data, the historical fault diagnosis data including multiple historical state quantity data acquired by the fusion monitoring system, as well as fault circuit determination results, fault location determination results, and fault cause determination results; filtering the multiple historical state quantity data to obtain target historical state quantity data; inputting the target historical state quantity data into the fault diagnosis model to be trained, and obtaining a historical fault circuit identification result through a fault circuit classifier in the fault diagnosis model; inputting the historical fault circuit identification result and the target historical state quantity data into the fault location classifier in the fault diagnosis model to obtain a historical fault location identification result; inputting the historical fault location identification result and the target historical state quantity data into the fault cause classifier in the fault diagnosis model to obtain a historical fault cause identification result; and training the fault diagnosis model to be trained based on the differences between the historical fault circuit identification result and the fault circuit determination result, the differences between the historical fault location identification result and the fault location determination result, and the differences between the historical fault cause identification result and the fault cause determination result to obtain a trained fault diagnosis model.

[0164] In an exemplary embodiment, the above-mentioned model training module is also used to obtain the historical correlation relationship between each of the above-mentioned historical state quantity data and the fault cause in the above-mentioned fault cause determination result; quantify each of the above-mentioned historical correlation relationships through kernel principal component analysis to obtain the contribution rate of each of the above-mentioned historical state quantity data; the contribution rate of the above-mentioned historical state quantity data is used to characterize the importance of the fault information contained in the above-mentioned historical state quantity data; obtain the descending sorting order of each of the above-mentioned historical state quantity data according to the size of the contribution rate of the above-mentioned historical state quantity data; according to the descending sorting order of each of the above-mentioned historical state quantity data, accumulate and calculate the historical cumulative contribution rate of each of the above-mentioned historical state quantity data in sequence; when the above-mentioned historical cumulative contribution rate is greater than the above-mentioned cumulative contribution rate threshold and the number of historical state quantity data is the smallest, determine that the multiple historical state quantity data corresponding to the above-mentioned historical cumulative contribution rate are the target historical state quantity data.

[0165] In an exemplary embodiment, the model training module is further configured to respectively obtain the calculation time of the fault loop classifier, the fault part classifier, and the fault cause classifier outputting identification results; the diagnostic accuracy of the fault loop classifier, the fault part classifier, and the fault cause classifier is respectively obtained based on the difference between the historical fault loop identification result and the fault loop determination result, the difference between the historical fault part identification result and the fault part determination result, and the difference between the historical fault cause identification result and the fault cause determination result; and the fault loop classifier, the fault part classifier, and the fault cause classifier are trained by improving the gray wolf optimization algorithm with the goal of achieving the highest diagnostic accuracy and the minimum calculation time to obtain a trained fault diagnosis model.

[0166] Each module in the aforementioned power transformer sensor network monitoring and fault diagnosis device can be implemented in whole or in part through software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in hardware form, or can be stored in a computer device memory in software form, so that the processor can call and execute the corresponding operations of each module.

[0167] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Figure 9As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected via a system bus, and the communication interface is connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method for monitoring and fault diagnosis of a power transformer sensor network is implemented.

[0168] Those skilled in the art will understand that Figure 9 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0169] In one embodiment, a computer device is further provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0170] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0171] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0172] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the various embodiments provided herein may be, but are not limited to, general-purpose processors, central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), programmable logic devices (PLDs), quantum computing-based data processing logic devices, artificial intelligence (AI) processors, and the like.

[0173] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0174] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A method for monitoring and fault diagnosis of a power transformer sensor network, characterized in that: The method comprises: Acquire multiple state quantity data of the transformer to be diagnosed through a pre-deployed fusion monitoring system, filter the multiple state quantity data, and obtain target state quantity data; the fusion monitoring system includes multiple types of sensors, and the multiple types of sensors are respectively used to monitor multiple different physical quantities of the transformer to be diagnosed; Inputting the target state quantity data into a pre-trained fault diagnosis model; the fault diagnosis model includes a fault circuit classifier, a fault location classifier, and a fault cause classifier; Obtaining a fault circuit identification result through a fault circuit classifier in the fault diagnosis model; Inputting the fault loop identification result and the target state quantity data into the fault location classifier in the fault diagnosis model to obtain a fault location identification result; Inputting the fault location identification result and the target state quantity data into the fault cause classifier in the fault diagnosis model to obtain a fault cause identification result; A fault diagnosis result of the transformer to be diagnosed is obtained according to the fault circuit identification result, the fault location identification result and the fault cause identification result.

2. The method according to claim 1, characterized in that The step of screening the plurality of state quantity data to obtain target state quantity data includes: Obtaining an association between each state quantity data and each preset fault cause in a preset fault cause set; quantifying each of the association relationships through kernel principal component analysis to obtain a contribution rate of each of the state quantity data; the contribution rate is used to characterize the importance of the fault information contained in the state quantity data; Obtaining a descending sort order of each of the state quantity data according to the contribution rate; Calculate the cumulative contribution rate of each state quantity data in turn according to the descending sort order; When the cumulative contribution rate is greater than a preset cumulative contribution rate threshold and the number of state quantity data is the smallest, the plurality of state quantity data corresponding to the cumulative contribution rate are determined as target state quantity data.

3. The method according to claim 1, characterized in that The fusion monitoring system includes a bushing insulation monitoring device, a neutral point DC current monitoring device, a dissolved gas in oil detection device, a partial discharge detection device, a winding optical fiber temperature measurement device, and an iron core grounding current detection device; The method of obtaining multiple state quantity data of the transformer to be diagnosed through the pre-deployed fusion monitoring system includes: Acquiring bushing insulation monitoring state data through the bushing insulation monitoring device; The neutral point DC current monitoring state quantity data is obtained through the neutral point DC current monitoring device; the dissolved gas in oil detection state quantity data is obtained through the dissolved gas in oil detection device; the partial discharge detection state quantity data is obtained through the partial discharge detection device; the winding optical fiber temperature measurement state quantity data is obtained through the winding optical fiber temperature measurement device; and the core grounding current detection state quantity data is obtained through the core grounding current detection device.

4. The method according to claim 1, wherein After the plurality of state quantity data are screened to obtain target state quantity data, the method further includes: The target state quantity data is discretized according to a preset discretization rule, and the discretized target state quantity data is used as new target state quantity data.

5. The method according to any one of claims 1 to 4, characterized in that The training steps of the fault diagnosis model include: Acquiring historical fault diagnosis data, the historical fault diagnosis data including a plurality of historical state quantity data acquired by the fusion monitoring system, as well as a fault circuit determination result, a fault location determination result, and a fault cause determination result; Filtering the plurality of historical state quantity data to obtain target historical state quantity data; Inputting the target historical state quantity data into the fault diagnosis model to be trained, and obtaining a historical fault loop identification result through the fault loop classifier in the fault diagnosis model; Inputting the historical fault loop identification result and the target historical state quantity data into the fault location classifier in the fault diagnosis model to obtain the historical fault location identification result; Inputting the historical fault location identification result and the target historical state quantity data into the fault cause classifier in the fault diagnosis model to obtain a historical fault cause identification result; According to the difference between the historical fault loop identification result and the fault loop determination result, the difference between the historical fault location identification result and the fault location determination result, and the difference between the historical fault cause identification result and the fault cause determination result, the fault diagnosis model to be trained is trained to obtain a trained fault diagnosis model.

6. The method according to claim 5, characterized in that The filtering of the plurality of historical state quantity data to obtain target historical state quantity data includes: Obtaining a historical correlation between each of the historical state quantity data and the fault cause in the fault cause determination result; Quantifying each of the historical association relationships through kernel principal component analysis to obtain a contribution rate of each of the historical state quantity data; the contribution rate of the historical state quantity data is used to characterize the importance of the fault information contained in the historical state quantity data; Obtaining a descending sort order of each of the historical state quantity data according to the contribution rate of the historical state quantity data; Calculate the historical cumulative contribution rate of each of the historical state quantity data in descending order; When the historical cumulative contribution rate is greater than the cumulative contribution rate threshold and the number of historical state quantity data is the smallest, the plurality of historical state quantity data corresponding to the historical cumulative contribution rate are determined as target historical state quantity data.

7. The method according to claim 5, characterized in that The method further comprises: training the fault diagnosis model to be trained based on the difference between the historical fault circuit identification result and the fault circuit determination result, the difference between the historical fault location identification result and the fault location determination result, and the difference between the historical fault cause identification result and the fault cause determination result to obtain a trained fault diagnosis model, including: respectively obtaining the calculation time of the fault circuit classifier, the fault location classifier, and the fault cause classifier outputting identification results; The diagnostic accuracy of the fault circuit classifier, the fault part classifier, and the fault cause classifier are obtained respectively based on the difference between the historical fault circuit identification result and the fault circuit determination result, the difference between the historical fault part identification result and the fault part determination result, and the difference between the historical fault cause identification result and the fault cause determination result; By improving the Grey Wolf optimization algorithm, with the goal of maximizing the diagnostic accuracy and minimizing the calculation time, the fault circuit classifier, the fault location classifier, and the fault cause classifier are trained to obtain a trained fault diagnosis model.

8. A power transformer sensor network monitoring and fault diagnosis device, characterized in that: The device comprises: A data acquisition module is used to obtain multiple state quantity data of the transformer to be diagnosed through a pre-deployed fusion monitoring system, and filter the multiple state quantity data to obtain target state quantity data; the fusion monitoring system includes multiple types of sensors, and the multiple types of sensors are respectively used to monitor multiple different physical quantities of the transformer to be diagnosed; a fault identification module configured to input the target state quantity data into a pre-trained fault diagnosis model; the fault diagnosis model comprising a fault loop classifier, a fault location classifier, and a fault cause classifier; obtain a fault loop identification result through the fault loop classifier in the fault diagnosis model; input the fault loop identification result and the target state quantity data into the fault location classifier in the fault diagnosis model to obtain a fault location identification result; and input the fault location identification result and the target state quantity data into the fault cause classifier in the fault diagnosis model to obtain a fault cause identification result; The diagnosis result acquisition module is used to obtain the fault diagnosis result of the transformer to be diagnosed according to the fault circuit identification result, the fault location identification result and the fault cause identification result.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images