An experimental monitoring method and system for temperature distribution evolution of transformer windings

Abstract

The present invention discloses an experimental monitoring method and system for the evolution of transformer winding temperature distribution, including: collecting a light signal with a target transformer winding temperature on a reflection grating for decoding to generate a hotspot temperature sequence, matching historical operating status data with the historical hotspot temperature sequence; constructing a digital twin model of the target transformer, and splitting it into several sub-models, dividing the working conditions of the target transformer to analyze the hotspot temperature distribution changes under each working condition; obtaining the temperature change correlation of each sub-model under different working conditions, and graphically representing different sub-models based on the temperature change correlation and physical coupling, constructing a hotspot prediction model to identify the hotspot position of the target transformer, and performing real-time temperature monitoring and early warning. The present invention analyzes the thermal characteristics of different structures according to the evolution of the winding temperature field distribution during the operation of the transformer, monitors the hotspot position and temperature early warning, provides a data basis for determining the fault operation state of the transformer, and ensures the normal operation of the transformer.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer testing, and more specifically, to an experimental monitoring method and system for transformer winding temperature distribution evolution. Background Art

[0002] In today's era of power system prevalence, the power system almost supports the operation of the entire social system, so the safe operation of the power system is crucial to economic development and social stability. Transformer failures are diverse, but failures in key parts of transformer windings are the most common. When serious overload, short circuit, insulation damage and other faults occur inside the transformer, the insulating oil is subjected to high temperature or electric arc, and thermal decomposition produces a large amount of hydrocarbon mixed gas, causing the pressure inside the transformer to rise sharply, thus causing serious explosion accidents. Therefore, real-time monitoring of the temperature of the transformer winding and early warning are of great practical significance.

[0003] Since the transformer is in a strong electromagnetic field, traditional electronic temperature sensors are susceptible to electromagnetic interference and cannot achieve precise measurement. Infrared radiation temperature measurement has great limitations due to the lack of scanning function and the limitation of the temperature measurement distance of the instrument itself, and cannot solve the problem of real-time monitoring of the internal temperature of the transformer. Compared with traditional sensors, distributed fiber optic sensing has great advantages such as strong anti-electromagnetic interference ability and distributed measurement. It constructs a multi-point temperature monitoring field for the transformer, and perceives the abnormal temperature changes of the transformer operation on the basis of the multi-point temperature monitoring field to issue early warnings, solving the problem of real-time monitoring of the temperature of multiple hot spots in the transformer winding. In the process of real-time monitoring, how to construct a hot spot prediction model based on the distribution of the winding temperature field under different load rates to accurately capture the changes in the hot spot position during the actual operation of the transformer is an urgent problem that needs to be solved. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes an experimental monitoring method and system for the evolution of transformer winding temperature distribution.

[0005] A first aspect of the present invention provides an experimental monitoring method for the evolution of transformer winding temperature distribution, comprising:

[0006] Collect the optical signal of the target transformer winding temperature from the reflection grating, decode the optical signal, obtain the hot spot temperature sequence of multiple fixed positions of the transformer winding, obtain the operating status data of the target transformer, and match the historical operating status data of the preset time step with the historical hot spot temperature sequence;

[0007] Construct a digital twin model of the target transformer, split the digital twin model into several sub-models, divide the working conditions of the target transformer, aggregate the historical hot spot temperature sequences of different working conditions, and analyze the temperature distribution changes under each working condition;

[0008] Obtain the temperature distribution change correlation of each sub-model under different working conditions, perform graphical representation according to the temperature distribution change correlation and physical coupling of different sub-models, and learn the graphical representation to build a hotspot prediction model;

[0009] The hotspot position of the target transformer is identified according to the hotspot prediction model, real-time monitoring and early warning are performed based on the temperature of the hotspot position, and the temperature prediction sequence of the hotspot position is fed back to the target transformer entity to make an operation decision for the target transformer.

[0010] In this scheme, the historical operating status data of the preset time step is matched with the historical hot spot temperature sequence, specifically:

[0011] Acquire a three-dimensional physical model of the target transformer, divide the winding structure area based on the three-dimensional physical model, extract index parameters of the winding structure sub-area according to preset structural parameter indicators, and construct a response relationship between the winding temperature and the structural parameters based on the index parameters;

[0012] Setting the point weights of different winding structure sub-areas according to the response relationship, setting the number of sensor points according to the point weights, and setting FBG fiber grating sensors based on the number of points to collect optical signals with winding temperature information;

[0013] The optical signal is photoelectrically converted to obtain temperature data, the temperature data of multiple fixed positions are combined with timestamp analysis to generate a hotspot temperature sequence, and the hotspot temperature sequence is matched with the operating status data using the timestamp;

[0014] The historical operation status data and the historical hot spot temperature sequence of the preset time step are obtained according to the data retrieval, the historical operation status data are preprocessed, and the preprocessed historical operation status data are paired with the historical hot spot temperature sequence.

[0015] In this scheme, the working conditions of the target transformer are divided, the historical hot spot temperature sequences of different working conditions are aggregated, and the temperature distribution changes under each working condition are analyzed. Specifically:

[0016] A digital twin model is constructed according to the three-dimensional physical model and the logical model of the target transformer, and the operation status data is used to perform data identification of the digital twin model, and the digital twin model is split into a plurality of sub-models;

[0017] Generate a paired sample set based on the paired historical operating status data and the historical hot spot temperature sequence, use clustering to divide the working conditions in the paired sample set, optimize the K-means algorithm according to the legacy algorithm, and select the best initial cluster center;

[0018] The Mahalanobis distance is selected as the metric function, the membership of each paired sample in the paired sample set to the initial cluster center is calculated, and a hard clustering threshold and a soft clustering threshold are preset. When the membership is greater than or equal to the hard clustering threshold, the initial cluster center with the highest membership is assigned to generate a cluster;

[0019] When the membership degree is greater than or equal to the soft clustering threshold and less than the hard clustering threshold, the paired samples are respectively assigned to all initial clustering centers that meet the standards. After the paired sample allocation is completed, the clustering center is updated using the clustering clusters, and the working condition labels corresponding to the paired samples are output through iterative updating;

[0020] The paired samples are divided into sub-models based on the sampling position, and the historical hot spot temperature sequences of different working conditions are aggregated in each sub-model. The historical hot spot temperature sequences of different working conditions are encoded, and the temperature distribution changes of each working condition in time series are obtained using a multi-layer perceptron.

[0021] In this solution, the temperature distribution change correlation of each sub-model under different working conditions is obtained, and a graph is represented according to the temperature distribution change correlation and physical coupling of different sub-models, specifically:

[0022] Obtain the temperature distribution changes of each sub-model under different working conditions, use convolutional neural network to read the spatial characteristics of the temperature distribution of the sub-model, and calculate the Pearson correlation coefficient between the sub-models based on the spatial characteristics;

[0023] The Pearson correlation coefficient is used to perform spatial weighting on the temperature distribution changes of the sub-models under different working conditions, and the average similarity of the temperature distribution changes corresponding to different working conditions between the sub-models is calculated. If the average similarity is greater than a preset similarity threshold, it is proved that there is a correlation between the temperature distribution changes of the two sub-models;

[0024] Read the coupling relationship between different sub-models according to the digital twin model of the target transformer, and use the coupling relationship to determine whether there is a physical coupling association between the sub-models;

[0025] The connection relationship between the sub-models is determined according to the temperature distribution change association and the physical coupling association, each sub-model is represented by a graph, the sub-model is used as a node in the graph, and the edge structure in the graph is generated according to the connection relationship between the sub-models.

[0026] In this solution, the graph representation is learned to construct a hotspot prediction model, specifically:

[0027] The digital twin model of the target transformer is used to generate twin data based on the paired historical operating status data and the historical hot spot temperature series, and data enhancement is performed through the twin data. The operating status data and hot spot temperatures under different working conditions are randomly combined to construct a training set.

[0028] The graph representation is learned through a graph attention network to build a hotspot prediction model, the model is trained using a training set, a multi-head attention mechanism is introduced into the hotspot prediction model, and a node feature representation is output in the attention layer according to the weights of the submodule node and its adjacent submodule nodes;

[0029] The node feature representations output by multiple attention heads are concatenated and integrated through an average operation in the last layer of the multi-layer structure to achieve node neighbor aggregation and obtain the final node feature representation corresponding to the sub-model node;

[0030] Importing the final node feature representation into a gated recurrent unit to obtain feature time dependency, outputting hotspot prediction data through a fully connected layer, obtaining hotspot prediction errors for back propagation, and updating and optimizing parameters of the hotspot prediction model;

[0031] The current operating status data and current winding temperature distribution of the target transformer are imported into the trained hotspot prediction model, and the winding temperature hotspot position and hotspot temperature prediction value of the target transformer are read according to the output hotspot prediction data.

[0032] In this solution, real-time monitoring and early warning are performed based on the temperature of the hot spot position, and the temperature prediction sequence of the hot spot position is fed back to the target transformer entity to make an operation decision for the target transformer, specifically:

[0033] Obtain the hot spot location of the winding temperature and the predicted value of the hot spot temperature output by the hot spot prediction model and import them into the three-dimensional physical model of the target transformer for visualization;

[0034] Retrieve and extract transformer thermal fault data, cluster the thermal fault data using the component information involved, match different thermal fault data subsets with corresponding sub-models, and adaptively generate a winding hot spot temperature threshold according to the thermal fault data subsets;

[0035] Extracting a corresponding winding hotspot temperature threshold based on the winding temperature hotspot position, and generating a warning mark in the three-dimensional physical model when the hotspot temperature prediction value is greater than the winding hotspot temperature threshold;

[0036] The temperature prediction sequence of the winding temperature hotspot position at the preset time step in the future is obtained and fed back to the target transformer entity. The temperature rise characteristics of the winding hotspot are extracted using the temperature prediction sequence. A preset number of high-probability thermal fault types are obtained based on the temperature rise characteristics of the real-time operating status data set.

[0037] The thermal fault type with the highest risk level is obtained from the high-probability thermal fault types and marked, and the target transformer operation status data is adjusted according to the indicated thermal fault type.

[0038] The second aspect of the present invention further provides an experimental monitoring system for the evolution of transformer winding temperature distribution, the system comprising: a memory, a processor, the memory comprising an experimental monitoring method program for the evolution of transformer winding temperature distribution, the experimental monitoring method program for the evolution of transformer winding temperature distribution being executed by the processor to implement the following steps:

[0039] Collect the optical signal of the target transformer winding temperature from the reflection grating, decode the optical signal, obtain the hot spot temperature sequence of multiple fixed positions of the transformer winding, obtain the operating status data of the target transformer, and match the historical operating status data of the preset time step with the historical hot spot temperature sequence;

[0040] Construct a digital twin model of the target transformer, split the digital twin model into several sub-models, divide the working conditions of the target transformer, aggregate the historical hot spot temperature sequences of different working conditions, and analyze the temperature distribution changes under each working condition;

[0041] Obtain the temperature distribution change correlation of each sub-model under different working conditions, perform graphical representation according to the temperature distribution change correlation and physical coupling of different sub-models, and learn the graphical representation to build a hotspot prediction model;

[0042] The hotspot position of the target transformer is identified according to the hotspot prediction model, real-time monitoring and early warning are performed based on the temperature of the hotspot position, and the temperature prediction sequence of the hotspot position is fed back to the target transformer entity to make an operation decision for the target transformer.

[0043] The present invention discloses an experimental monitoring method and system for the evolution of transformer winding temperature distribution, including: collecting a light signal with a target transformer winding temperature on a reflection grating for decoding to generate a hotspot temperature sequence, matching the historical operating status data of a preset time step with the historical hotspot temperature sequence; constructing a digital twin model of the target transformer, and splitting it into several sub-models, dividing the working conditions of the target transformer to analyze the hotspot temperature distribution changes under each working condition; obtaining the temperature change correlation of each sub-model under different working conditions, and performing a graphical representation based on the temperature change correlation and physical coupling of different sub-models, constructing a hotspot prediction model to identify the hotspot position of the target transformer, and performing real-time monitoring and early warning based on the temperature of the hotspot position. The present invention analyzes the thermal characteristics of different structures according to the evolution of the winding temperature field distribution during the operation of the transformer, monitors the hotspot position and temperature early warning, provides a data basis for determining the faulty operating state of the transformer, ensures the normal operation of the transformer and maintains the stability of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A flow chart of an experimental monitoring method for the evolution of transformer winding temperature distribution according to the present invention is shown;

[0045] Figure 2 A flowchart of the present invention for graphically representing different sub-models is shown;

[0046] Figure 3 A flow chart of constructing a hotspot prediction model according to the present invention is shown;

[0047] Figure 4 A block diagram of an experimental monitoring system for temperature distribution evolution of transformer windings according to the present invention is shown. DETAILED DESCRIPTION

[0048] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0049] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0050] Figure 1 A flow chart of an experimental monitoring method for the evolution of transformer winding temperature distribution according to the present invention is shown.

[0051] like Figure 1 As shown, the first aspect of the present invention provides an experimental monitoring method for the evolution of transformer winding temperature distribution, comprising:

[0052] S102, collecting the optical signal of the target transformer winding temperature from the reflection grating, decoding the optical signal, obtaining the hot spot temperature sequence of multiple fixed positions of the transformer winding, obtaining the operating status data of the target transformer, and matching the historical operating status data of the preset time step with the historical hot spot temperature sequence;

[0053] S104, constructing a digital twin model of a target transformer, splitting the digital twin model into a plurality of sub-models, dividing the working conditions of the target transformer, aggregating the historical hot spot temperature sequences of different working conditions, and analyzing the temperature distribution changes under each working condition;

[0054] S106, obtaining the temperature distribution change correlation of each sub-model under different working conditions, performing graphical representation according to the temperature distribution change correlation and physical coupling of different sub-models, and learning the graphical representation to construct a hotspot prediction model;

[0055] S108, identifying the hotspot position of the target transformer according to the hotspot prediction model, performing real-time monitoring and early warning based on the temperature of the hotspot position, feeding back the temperature prediction sequence of the hotspot position to the target transformer entity, and making an operation decision for the target transformer.

[0056] It should be noted that the three-dimensional geometric models of the oil tank, core and winding of the transformer are combined and spliced ​​to obtain the three-dimensional physical model of the target transformer, the winding structure area is divided based on the three-dimensional physical model, and the index parameters such as the distance between the low-voltage winding and the core, the conductor cross-sectional width, the conductor cross-sectional length, the inner oil channel width, and the outer oil channel width of the winding structure sub-area are extracted according to the preset structural parameter index, and the response relationship between the winding temperature and the structural parameters is constructed based on the index parameters; the distribution weights of different winding structure sub-areas are set according to the response relationship, when the structural parameters are more complex, the corresponding temperature rise of the winding temperature is more significant, so the number of sensors set in the corresponding area is more, the number of sensor distribution points is set according to the distribution weight, and the FBG fiber grating sensor is set based on the number of distribution points to collect the optical signal with the winding temperature information; the optical signal is converted into the corresponding current signal by photoelectric conversion, the current signal is converted into the voltage by amplification and amplified, the analog value of the voltage is converted into the digital value, the digital value of the voltage is analyzed, and the corresponding temperature is generated, and the temperature of multiple fixed positions is measured in real time through multi-channel grating scanning. The temperature data of multiple fixed-point positions are combined with timestamp analysis to compare the temperature data of multiple fixed points to obtain the winding temperature hotspots, and generate a hotspot temperature sequence. The hotspot temperature sequence is matched with operating status data such as voltage signals, current signals, impedance voltage, vibration signals, oil flow velocity signals and load rates using timestamps; historical operating status data and historical hotspot temperature sequences of preset time steps are obtained according to data retrieval, and the historical operating status data are preprocessed by removing outliers, normalizing, etc., and the preprocessed historical operating status data are paired with the historical hotspot temperature sequence.

[0057] It should be noted that a digital twin model is constructed based on the three-dimensional physical model and logical model of the target transformer, a three-dimensional physical model is constructed based on the three-dimensional geometric model of the component, a controllable closed-loop logical model is established for the logical mechanism relationship between the physical entities of each component, and the physical model is mapped to the logical model; the transformer operating status data is used to identify the data of the digital twin model, the historical operating status data is imported into the digital twin, the parameters of the digital twin model are tuned by the error between the twin data output by the digital twin and the true value, the digital twin model of the target transformer after identification and correction is output, and the digital twin model is split into several sub-models; a paired sample set is generated based on the paired historical operating status data and the historical hot spot temperature sequence, the working conditions are divided by clustering in the paired sample set, and the K-means algorithm is used to perform the multi-function matching. The algorithm is optimized to select the best initial clustering center; the Mahalanobis distance is selected as the metric function, the membership of each paired sample in the paired sample set to the initial clustering center is calculated, and the hard clustering threshold and the soft clustering threshold are preset. When the membership is greater than or equal to the hard clustering threshold, the initial clustering center with the highest membership is assigned to generate a cluster; when the membership is greater than or equal to the soft clustering threshold and less than the hard clustering threshold, the paired samples are respectively assigned to all the initial clustering centers that meet the standards. After the paired sample allocation is completed, the clustering center is updated using the clustering clusters, and the working condition labels corresponding to the paired samples are output through iterative updating; the paired samples are divided according to the sub-models based on the sampling position, and the historical hot spot temperature sequences of different working conditions are aggregated in each sub-model, the historical hot spot temperature sequences of different working conditions are encoded, and the temperature distribution changes of each working condition under time series are obtained using a multi-layer perceptron.

[0058] Figure 2 A flow chart of the present invention for graphically representing different sub-models is shown.

[0059] According to an embodiment of the present invention, the temperature distribution change correlation of each sub-model under different working conditions is obtained, and a graph is represented according to the temperature distribution change correlation and physical coupling of different sub-models, specifically:

[0060] S202, obtaining the temperature distribution changes of each sub-model under different working conditions, using a convolutional neural network to read the spatial characteristics of the temperature distribution of the sub-model, and calculating the Pearson correlation coefficient between the sub-models based on the spatial characteristics;

[0061] S204, using the Pearson correlation coefficient to perform spatial weighting on the temperature distribution changes of the sub-models under different working conditions, and calculating the average similarity of the temperature distribution changes corresponding to different working conditions between the sub-models. If the average similarity is greater than a preset similarity threshold, it is proved that there is a correlation between the temperature distribution changes of the two sub-models.

[0062] S206, reading the coupling relationship between different sub-models according to the digital twin model of the target transformer, and using the coupling relationship to determine whether there is a physical coupling association between the sub-models;

[0063] S208, determining the connection relationship between the sub-models according to the temperature distribution change correlation and the physical coupling correlation, representing each sub-model in a graph, using the sub-model as a node in the graph, and generating an edge structure in the graph according to the connection relationship between the sub-models.

[0064] It should be noted that the digital twin model of the target transformer is split to obtain n sub-models, and the sub-models are represented by graphs. , is the vertex set of the submodel of the graph, is the edge structure set of the graph, The type corresponding to the edge structure is divided into temperature distribution change association and physical coupling association.

[0065] Figure 3 A flow chart of constructing a hotspot prediction model according to the present invention is shown.

[0066] According to an embodiment of the present invention, the graph representation is learned to construct a hotspot prediction model, specifically:

[0067] S302, using the digital twin model of the target transformer to generate twin data based on the paired historical operating status data and the historical hot spot temperature sequence, performing data enhancement through the twin data, and randomly combining the operating status data and hot spot temperatures under different working conditions to construct a training set;

[0068] S304, learning the graph representation through a graph attention network to build a hotspot prediction model, using a training set to train the model, introducing a multi-head attention mechanism into the hotspot prediction model, and outputting node feature representations in the attention layer according to the weights of the submodule nodes and their adjacent submodule nodes;

[0069] S306, concatenating the node feature representations output by the multiple attention heads, and concatenating and integrating them through an average operation in the last layer of the multi-layer structure to achieve node neighbor aggregation and obtain the final node feature representation corresponding to the sub-model node;

[0070] S308, importing the final node feature representation into a gated recurrent unit to obtain feature time dependency, outputting hotspot prediction data through a fully connected layer, obtaining hotspot prediction errors for back propagation, and updating and optimizing parameters of the hotspot prediction model;

[0071] S310, importing the current operating status data and current winding temperature distribution of the target transformer into the trained hotspot prediction model, and reading the winding temperature hotspot position and hotspot temperature prediction value of the target transformer according to the output hotspot prediction data.

[0072] It should be noted that the use of graph attention network to build a hotspot prediction model gets rid of the constraints of the Laplace matrix and reduces the amount of calculation. In the hotspot prediction model, a multi-head attention mechanism is introduced to achieve parallel calculation through multiple attention heads to accelerate the calculation speed of the model. In the multi-layer structure, the concatenation and integration of the last layer uses an average operation to obtain the final node feature representation corresponding to the sub-model node. , specifically:

[0073] ;

[0074] in is the activation function, is the total number of attention heads, is the number of attention heads, is the parameter vector between nodes in the attention head, is the weight matrix between nodes in the attention head, It is the feature representation of the adjacent sub-model nodes.

[0075] After the attention layer, the information extracted by the hotspot prediction model is not only the information of each sub-model node itself, but also the features corresponding to each sub-model node are scaled according to the attention score, and the relevant information of adjacent sub-model nodes is aggregated. A gated recurrent unit is added after the graph attention network. The gated recurrent unit overcomes the problems of gradient vanishing and gradient exploding in complex networks. The final node feature representation of the sub-model node is updated using the hidden state to capture the time dependency implied by the hotspot temperature distribution and operating state parameters of the transformer winding. In addition, the graph attention network adopts semi-supervised learning. When the transformer operating state data fluctuates greatly, the graph attention layer adaptively adjusts the weight according to the sub-model node characteristics, and performs network self-correction according to the real-time operating state data to ensure the robustness of the hotspot prediction model.

[0076] It should be noted that the hotspot temperature position and predicted value of the hotspot temperature output by the hotspot prediction model are imported into the three-dimensional physical model of the target transformer for visual display; the transformer thermal fault data is retrieved and extracted, the thermal fault data is clustered using the component information involved, different thermal fault data subsets are matched with the corresponding sub-models, and the winding hotspot temperature threshold is adaptively generated according to the thermal fault data subset; the corresponding winding hotspot temperature threshold is extracted based on the winding temperature hotspot position, and when the hotspot temperature predicted value is greater than the winding hotspot temperature threshold, a warning mark is generated in the three-dimensional physical model; the temperature prediction sequence of the winding temperature hotspot position at a preset time step in the future is obtained and fed back to the target transformer entity, the temperature prediction sequence is used to extract the temperature rise characteristics of the winding hotspot, and a preset number of high-probability thermal fault types are obtained according to the temperature rise characteristics of the real-time operating status data set; the fault loss and other indicators are extracted according to the instance data of different thermal fault types to evaluate the risk level; the thermal fault type with the highest risk level is obtained among the high-probability thermal fault types and marked, and the operating status data of the target transformer is adjusted according to the indicated thermal fault type.

[0077] An experimental monitoring database is constructed to obtain the winding hotspot temperature distribution corresponding to transformers with different aging degrees, and the spatiotemporal characteristics are obtained according to the winding hotspot temperature distribution. The spatiotemporal characteristics are used to construct a mapping relationship with the thermal life based on the aging degree label, and the mapping relationship is stored in the experimental monitoring database. The target transformer is sampled according to a preset sampling time interval to generate a change characteristic of the winding hotspot temperature distribution, and is located in the experimental monitoring database according to the change characteristic. The corresponding thermal life is read using the mapping relationship, and the environmental coefficient is set according to the application environment of the target transformer to weight the thermal life threshold. When the thermal life is less than the weighted thermal life threshold, a transformer remaining life warning is generated.

[0078] Figure 4 A block diagram of an experimental monitoring system for temperature distribution evolution of transformer windings according to the present invention is shown.

[0079] The second aspect of the present invention further provides an experimental monitoring system 4 for the evolution of transformer winding temperature distribution, the system comprising: a memory 41, a processor 42, the memory comprising an experimental monitoring method program for the evolution of transformer winding temperature distribution, the experimental monitoring method program for the evolution of transformer winding temperature distribution being executed by the processor to implement the following steps:

[0080] Collect the optical signal of the target transformer winding temperature from the reflection grating, decode the optical signal, obtain the hot spot temperature sequence of multiple fixed positions of the transformer winding, obtain the operating status data of the target transformer, and match the historical operating status data of the preset time step with the historical hot spot temperature sequence;

[0081] Construct a digital twin model of the target transformer, split the digital twin model into several sub-models, divide the working conditions of the target transformer, aggregate the historical hot spot temperature sequences of different working conditions, and analyze the temperature distribution changes under each working condition;

[0082] Obtain the temperature distribution change correlation of each sub-model under different working conditions, perform graphical representation according to the temperature distribution change correlation and physical coupling of different sub-models, and learn the graphical representation to build a hotspot prediction model;

[0083] The hotspot position of the target transformer is identified according to the hotspot prediction model, real-time monitoring and early warning are performed based on the temperature of the hotspot position, and the temperature prediction sequence of the hotspot position is fed back to the target transformer entity to make an operation decision for the target transformer.

[0084] The third aspect of the present invention also provides a computer-readable storage medium, which includes an experimental monitoring method program for the evolution of transformer winding temperature distribution. When the experimental monitoring method program for the evolution of transformer winding temperature distribution is executed by a processor, the steps of the experimental monitoring method for the evolution of transformer winding temperature distribution as described in any one of the above items are implemented.

[0085] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0086] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0087] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0088] A person of ordinary skill in the art can understand that: all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: a mobile storage device, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0089] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0090] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. An experimental monitoring method for the evolution of transformer winding temperature distribution, characterized in that: The following steps are involved: Collect the optical signal of the target transformer winding temperature from the reflection grating, decode the optical signal, obtain the hot spot temperature sequence of multiple fixed positions of the transformer winding, obtain the operating status data of the target transformer, and match the historical operating status data of the preset time step with the historical hot spot temperature sequence; Construct a digital twin model of the target transformer, split the digital twin model into several sub-models, divide the working conditions of the target transformer, aggregate the historical hot spot temperature sequences of different working conditions, and analyze the temperature distribution changes under each working condition; Obtain the temperature distribution change correlation of each sub-model under different working conditions, perform graphical representation according to the temperature distribution change correlation and physical coupling of different sub-models, and learn the graphical representation to build a hotspot prediction model; Identify the hotspot position of the target transformer according to the hotspot prediction model, perform real-time monitoring and early warning based on the temperature of the hotspot position, feed back the temperature prediction sequence of the hotspot position to the target transformer entity, and make an operation decision for the target transformer; Obtain the temperature distribution change correlation of each sub-model under different working conditions, and make a graphical representation based on the temperature distribution change correlation and physical coupling of different sub-models, specifically: Obtain the temperature distribution changes of each sub-model under different working conditions, use convolutional neural network to read the spatial characteristics of the temperature distribution of the sub-model, and calculate the Pearson correlation coefficient between the sub-models based on the spatial characteristics; The Pearson correlation coefficient is used to perform spatial weighting on the temperature distribution changes of the sub-models under different working conditions, and the average similarity of the temperature distribution changes corresponding to different working conditions between the sub-models is calculated. If the average similarity is greater than a preset similarity threshold, it is proved that there is a correlation between the temperature distribution changes of the two sub-models; Read the coupling relationship between different sub-models according to the digital twin model of the target transformer, and use the coupling relationship to determine whether there is a physical coupling association between the sub-models; Determine the connection relationship between the sub-models according to the temperature distribution change association and the physical coupling association, represent each sub-model in a graph, use the sub-model as a node in the graph, and generate an edge structure in the graph according to the connection relationship between the sub-models; The graph representation is learned to construct a hotspot prediction model, specifically: The digital twin model of the target transformer is used to generate twin data based on the paired historical operating status data and the historical hot spot temperature series, and data enhancement is performed through the twin data. The operating status data and hot spot temperatures under different working conditions are randomly combined to construct a training set. The graph representation is learned through a graph attention network to build a hotspot prediction model, the model is trained using a training set, a multi-head attention mechanism is introduced into the hotspot prediction model, and a node feature representation is output in the attention layer according to the weights of the submodule node and its adjacent submodule nodes; The node feature representations output by multiple attention heads are concatenated and integrated through an average operation in the last layer of the multi-layer structure to achieve node neighbor aggregation and obtain the final node feature representation corresponding to the sub-model node; Importing the final node feature representation into a gated recurrent unit to obtain feature time dependency, outputting hotspot prediction data through a fully connected layer, obtaining hotspot prediction errors for back propagation, and updating and optimizing parameters of the hotspot prediction model; The current operating state data and current winding temperature distribution of the target transformer are imported into the trained hotspot prediction model, and the winding temperature hotspot position and hotspot temperature prediction value of the target transformer are read according to the output hotspot prediction data; Based on the temperature of the hot spot position, real-time monitoring and early warning are performed, and the temperature prediction sequence of the hot spot position is fed back to the target transformer entity to make an operation decision for the target transformer, specifically: Obtain the hot spot location of the winding temperature and the predicted value of the hot spot temperature output by the hot spot prediction model and import them into the three-dimensional physical model of the target transformer for visualization; Retrieve and extract transformer thermal fault data, cluster the thermal fault data using the component information involved, match different thermal fault data subsets with corresponding sub-models, and adaptively generate a winding hot spot temperature threshold according to the thermal fault data subsets; Extracting a corresponding winding hotspot temperature threshold based on the winding temperature hotspot position, and generating a warning mark in the three-dimensional physical model when the hotspot temperature prediction value is greater than the winding hotspot temperature threshold; The temperature prediction sequence of the winding temperature hotspot position at the preset time step in the future is obtained and fed back to the target transformer entity. The temperature rise characteristics of the winding hotspot are extracted using the temperature prediction sequence. A preset number of high-probability thermal fault types are obtained based on the temperature rise characteristics of the real-time operating status data set. The thermal fault type with the highest risk level is obtained from the high-probability thermal fault types and marked, and the target transformer operation status data is adjusted according to the indicated thermal fault type.

2. The experimental monitoring method for the evolution of transformer winding temperature distribution according to claim 1, characterized in that: Match the historical operating status data of the preset time step with the historical hot spot temperature sequence, specifically: Acquire a three-dimensional physical model of the target transformer, divide the winding structure area based on the three-dimensional physical model, extract index parameters of the winding structure sub-area according to preset structural parameter indicators, and construct a response relationship between the winding temperature and the structural parameters based on the index parameters; Setting the point weights of different winding structure sub-areas according to the response relationship, setting the number of sensor points according to the point weights, and setting FBG fiber grating sensors based on the number of points to collect optical signals with winding temperature information; The optical signal is photoelectrically converted to obtain temperature data, the temperature data of multiple fixed positions are combined with timestamp analysis to generate a hotspot temperature sequence, and the hotspot temperature sequence is matched with the operating status data using the timestamp; The historical operation status data and the historical hot spot temperature sequence of the preset time step are obtained according to the data retrieval, the historical operation status data are preprocessed, and the preprocessed historical operation status data are paired with the historical hot spot temperature sequence.

3. The experimental monitoring method for the evolution of transformer winding temperature distribution according to claim 1, characterized in that: Divide the working conditions of the target transformer, aggregate the historical hot spot temperature sequences under different working conditions, and analyze the temperature distribution changes under each working condition, specifically: A digital twin model is constructed according to the three-dimensional physical model and the logical model of the target transformer, and the operation status data is used to perform data identification of the digital twin model, and the digital twin model is split into a plurality of sub-models; Generate a paired sample set based on the paired historical operating status data and the historical hot spot temperature sequence, use clustering to divide the working conditions in the paired sample set, optimize the K-means algorithm according to the legacy algorithm, and select the best initial cluster center; The Mahalanobis distance is selected as the metric function, the membership of each paired sample in the paired sample set to the initial cluster center is calculated, and a hard clustering threshold and a soft clustering threshold are preset. When the membership is greater than or equal to the hard clustering threshold, the initial cluster center with the highest membership is assigned to generate a cluster; When the membership degree is greater than or equal to the soft clustering threshold and less than the hard clustering threshold, the paired samples are respectively assigned to all initial clustering centers that meet the standards. After the paired sample allocation is completed, the clustering center is updated using the clustering clusters, and the working condition labels corresponding to the paired samples are output through iterative updating; The paired samples are divided into sub-models based on the sampling position, and the historical hot spot temperature sequences of different working conditions are aggregated in each sub-model. The historical hot spot temperature sequences of different working conditions are encoded, and the temperature distribution changes of each working condition in time series are obtained using a multi-layer perceptron.

4. An experimental monitoring system for the evolution of transformer winding temperature distribution, characterized in that: The system includes: a memory and a processor, wherein the memory includes an experimental monitoring method program for the evolution of transformer winding temperature distribution, and when the experimental monitoring method program for the evolution of transformer winding temperature distribution is executed by the processor, the following steps are implemented: Collect the optical signal of the target transformer winding temperature from the reflection grating, decode the optical signal, obtain the hot spot temperature sequence of multiple fixed positions of the transformer winding, obtain the operating status data of the target transformer, and match the historical operating status data of the preset time step with the historical hot spot temperature sequence; Construct a digital twin model of the target transformer, split the digital twin model into several sub-models, divide the working conditions of the target transformer, aggregate the historical hot spot temperature sequences of different working conditions, and analyze the temperature distribution changes under each working condition; Obtain the temperature distribution change correlation of each sub-model under different working conditions, perform graphical representation according to the temperature distribution change correlation and physical coupling of different sub-models, and learn the graphical representation to build a hotspot prediction model; Identify the hotspot position of the target transformer according to the hotspot prediction model, perform real-time monitoring and early warning based on the temperature of the hotspot position, feed back the temperature prediction sequence of the hotspot position to the target transformer entity, and make an operation decision for the target transformer; Obtain the temperature distribution change correlation of each sub-model under different working conditions, and make a graphical representation based on the temperature distribution change correlation and physical coupling of different sub-models, specifically: Obtain the temperature distribution changes of each sub-model under different working conditions, use convolutional neural network to read the spatial characteristics of the temperature distribution of the sub-model, and calculate the Pearson correlation coefficient between the sub-models based on the spatial characteristics; The Pearson correlation coefficient is used to perform spatial weighting on the temperature distribution changes of the sub-models under different working conditions, and the average similarity of the temperature distribution changes corresponding to different working conditions between the sub-models is calculated. If the average similarity is greater than a preset similarity threshold, it is proved that there is a correlation between the temperature distribution changes of the two sub-models; Read the coupling relationship between different sub-models according to the digital twin model of the target transformer, and use the coupling relationship to determine whether there is a physical coupling association between the sub-models; Determine the connection relationship between the sub-models according to the temperature distribution change association and the physical coupling association, represent each sub-model in a graph, use the sub-model as a node in the graph, and generate an edge structure in the graph according to the connection relationship between the sub-models; The graph representation is learned to construct a hotspot prediction model, specifically: The digital twin model of the target transformer is used to generate twin data based on the paired historical operating status data and the historical hot spot temperature series, and data enhancement is performed through the twin data. The operating status data and hot spot temperatures under different working conditions are randomly combined to construct a training set. The graph representation is learned through a graph attention network to build a hotspot prediction model, the model is trained using a training set, a multi-head attention mechanism is introduced into the hotspot prediction model, and a node feature representation is output in the attention layer according to the weights of the submodule node and its adjacent submodule nodes; The node feature representations output by multiple attention heads are concatenated and integrated through an average operation in the last layer of the multi-layer structure to achieve node neighbor aggregation and obtain the final node feature representation corresponding to the sub-model node; Importing the final node feature representation into a gated recurrent unit to obtain feature time dependency, outputting hotspot prediction data through a fully connected layer, obtaining hotspot prediction errors for back propagation, and updating and optimizing parameters of the hotspot prediction model; The current operating state data and current winding temperature distribution of the target transformer are imported into the trained hotspot prediction model, and the winding temperature hotspot position and hotspot temperature prediction value of the target transformer are read according to the output hotspot prediction data; Based on the temperature of the hot spot position, real-time monitoring and early warning are performed, and the temperature prediction sequence of the hot spot position is fed back to the target transformer entity to make an operation decision for the target transformer, specifically: Obtain the hot spot location of the winding temperature and the predicted value of the hot spot temperature output by the hot spot prediction model and import them into the three-dimensional physical model of the target transformer for visualization; Retrieve and extract transformer thermal fault data, cluster the thermal fault data using the component information involved, match different thermal fault data subsets with corresponding sub-models, and adaptively generate a winding hot spot temperature threshold according to the thermal fault data subsets; Extracting a corresponding winding hotspot temperature threshold based on the winding temperature hotspot position, and generating a warning mark in the three-dimensional physical model when the hotspot temperature prediction value is greater than the winding hotspot temperature threshold; The temperature prediction sequence of the winding temperature hotspot position at the preset time step in the future is obtained and fed back to the target transformer entity. The temperature rise characteristics of the winding hotspot are extracted using the temperature prediction sequence. A preset number of high-probability thermal fault types are obtained based on the temperature rise characteristics of the real-time operating status data set. The thermal fault type with the highest risk level is obtained from the high-probability thermal fault types and marked, and the target transformer operation status data is adjusted according to the indicated thermal fault type.

5. The experimental monitoring system for transformer winding temperature distribution evolution according to claim 4, characterized in that: Divide the working conditions of the target transformer, aggregate the historical hot spot temperature sequences under different working conditions, and analyze the temperature distribution changes under each working condition, specifically: A digital twin model is constructed according to the three-dimensional physical model and the logical model of the target transformer, and the operation status data is used to perform data identification of the digital twin model, and the digital twin model is split into a plurality of sub-models; Generate a paired sample set based on the paired historical operating status data and the historical hot spot temperature sequence, use clustering to divide the working conditions in the paired sample set, optimize the K-means algorithm according to the legacy algorithm, and select the best initial cluster center; The Mahalanobis distance is selected as the metric function, the membership of each paired sample in the paired sample set to the initial cluster center is calculated, and a hard clustering threshold and a soft clustering threshold are preset. When the membership is greater than or equal to the hard clustering threshold, the initial cluster center with the highest membership is assigned to generate a cluster; When the membership degree is greater than or equal to the soft clustering threshold and less than the hard clustering threshold, the paired samples are respectively assigned to all initial clustering centers that meet the standards. After the paired sample allocation is completed, the clustering center is updated using the clustering clusters, and the working condition labels corresponding to the paired samples are output through iterative updating; The paired samples are divided into sub-models based on the sampling position, and the historical hot spot temperature sequences of different working conditions are aggregated in each sub-model. The historical hot spot temperature sequences of different working conditions are encoded, and the temperature distribution changes of each working condition in time series are obtained using a multi-layer perceptron.

Images