A method and system for identifying partial discharge type of transformer
By adopting the attention mechanism of the CBAM structure in the local discharge type recognition of transformer, combined with the multi-dimensional feature extraction of PRPD and PRPS maps, the problem of low accuracy of local discharge type recognition in the prior art is solved, and higher recognition accuracy is achieved Rate.
Patent Information
- Application Number
- CN202111369869.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-11-18
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2041-11-18
AI Technical Summary
The prior art uses less deep learning in the identification of local discharge type of transformer, mainly because sensor technology is not widely used and local discharge data is limited and data differences are large, resulting in poor feature extraction effect.
The local discharge type recognition model with CBAM structure is adopted, and the attention mechanism is applied at the channel level and spatial level at the same time. Features are extracted from different dimensions through PRPD and PRPS maps, and phase, amplitude and periodic information are fused to improve the recognition accuracy.
Through the feature extraction of comprehensive channel and spatial dimensions, the weight of invalid features is reduced, the accuracy of the local discharge type recognition of transformer is improved, and the limitations of the difficulty of extracting effective features is overcome by a single map.
Smart Images

Figure CN113884844B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of electrical fault detection, and in particular relates to a transformer partial discharge type identification method and system. Background Art
[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.
[0003] Effectively identifying the type and severity of partial discharge inside the transformer can timely discover some latent faults inside it, provide a basis for diagnosis and maintenance, and ensure the safe and stable operation of the system. The transformer partial discharge diagnosis method mainly converts the partial discharge signal into statistical graphs such as partial discharge phase distribution graph (Phase Resolved Partial Discharge, PRPD) and pulse sequence phase distribution graph (Phase Resolved Pulse Sequence, PRPS) by using various sensor technologies. By analyzing the graphs, the type and severity of discharge inside the transformer can be determined.
[0004] Deep learning has great advantages in extracting key information from data and gaining insights into tiny features. Using deep learning methods to identify and classify statistical graphs and waveform signals of partial discharge has opened up new ideas for transformer partial discharge pattern recognition.
[0005] At present, deep learning is rarely used in the field of transformer partial discharge pattern recognition. There are two main reasons for this. First, sensor technology is not widely used in transformers, and partial discharge is a low-probability event, so the partial discharge data that can be obtained is very limited; second, partial discharge data will vary greatly due to factors such as different transformer types and environments. Most of the data used in most current research is simulation data obtained from laboratories, so the extracted features are mostly single features for a certain transformer, but it is difficult to extract effective features from a single feature. Summary of the invention
[0006] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides a transformer partial discharge type identification method and system, which realizes the complementarity of limitations between different spectra and improves the accuracy of discharge type identification.
[0007] In order to achieve the above object, the present invention adopts the following technical solution:
[0008] A first aspect of the present invention provides a method for identifying a transformer partial discharge type, comprising:
[0009] Obtain at least two types of partial discharge spectra of the transformer;
[0010] Preprocessing the acquired spectrum;
[0011] The preprocessed spectrum is input into the partial discharge type recognition model to obtain the partial discharge type of the transformer;
[0012] Among them, the feature extraction network of the partial discharge type recognition model adopts the CBAM structure to apply the attention mechanism simultaneously in two dimensions: the channel level and the spatial level. The channel level is used to capture the characteristics of different types of partial discharge spectra, and the spatial level is used to capture the aggregation relationship of the phase, amplitude and period of the partial discharge spectrum.
[0013] Furthermore, the number of the feature extraction networks is the same as the number of types of partial discharge maps obtained, and each feature extraction network is used to extract features of one type of partial discharge map.
[0014] Furthermore, after passing through the sparse layer and the activation function, the output layer of each feature extraction network obtains a transformer partial discharge type recognition result of the corresponding type of partial discharge spectrum, and the loss of the recognition result serves as an auxiliary loss of the feature extraction network.
[0015] Furthermore, the partial discharge type recognition model also includes a feature map fusion layer, which is used to fuse the features of the partial discharge maps extracted by each feature extraction network to obtain a fused feature map.
[0016] Furthermore, the fused feature map is outputted using an activation function after passing through a convolution layer, and the output is the partial discharge type of the transformer identified by the partial discharge type recognition model.
[0017] Furthermore, the attention mechanism specifically includes:
[0018] Get input information and give a query vector;
[0019] Calculate the similarity between the query vector and the input information, obtain the weight, and normalize the weight to obtain the attention distribution;
[0020] The attention distribution is attached to the input information for weighting to obtain the output value.
[0021] Further, the preprocessing includes:
[0022] Normalizing the acquired spectrum;
[0023] The normalized spectrum is shifted in the phase dimension to obtain a preprocessed spectrum.
[0024] A second aspect of the present invention provides a transformer partial discharge type identification system, comprising:
[0025] A data acquisition module is configured to: acquire at least two types of partial discharge spectra of the transformer;
[0026] A preprocessing module, which is configured to: preprocess the acquired atlas;
[0027] A partial discharge type identification module is configured to: input the preprocessed spectrum into a partial discharge type identification model to obtain a partial discharge type of the transformer;
[0028] Among them, the feature extraction network of the partial discharge type recognition model adopts the CBAM structure to apply the attention mechanism simultaneously in two dimensions: the channel level and the spatial level. The channel level is used to capture the characteristics of different types of partial discharge spectra, and the spatial level is used to capture the aggregation relationship of the phase, amplitude and period of the partial discharge spectrum.
[0029] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the transformer partial discharge type identification method as described above.
[0030] A fourth aspect of the present invention provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the transformer partial discharge type identification method as described above are implemented.
[0031] Compared with the prior art, the present invention has the following beneficial effects:
[0032] The present invention provides a method for identifying the type of partial discharge of a transformer, which applies the attention mechanism to both the channel level and the spatial level. The channel level reflects different types of features, and the spatial level is reflected in the graph in the clustering relationship between phase, amplitude and period. By integrating the features of the channel level and the spatial dimension, the weight of invalid features can be reduced, and features that are more conducive to improving the recognition accuracy can be extracted.
[0033] The present invention provides a transformer partial discharge type identification method, which extracts features from different dimensions of the PRPD and PRPS spectra collected during the discharge process to achieve complementary limitations between different spectra. The PRPD spectrum focuses more on describing the degree of aggregation of partial discharge in phase and amplitude, while the PRPS spectrum focuses more on representing the changes in the spectrum in different cycles. Based on this, the features of the two spectra are extracted from the two dimensions of phase and cycle, and then the two features are fused using the weights of deep learning self-learning, thereby improving the accuracy of discharge type identification. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.
[0035] Figure 1 is an overall structural diagram of a partial discharge type identification model according to a first embodiment of the present invention;
[0036] Figure 2 is a CBAM network structure diagram of the first embodiment of the present invention;
[0037] Figure 3 is a structural diagram of a Resnet network integrating an attention mechanism according to the first embodiment of the present invention;
[0038] Figure 4 Schematic diagram of the attention mechanism of the first embodiment of the present invention. DETAILED DESCRIPTION
[0039] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.
[0040] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.
[0041] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.
[0042] Embodiment 1
[0043] like Figure 1 As shown, this embodiment provides a method for identifying the type of partial discharge of a transformer, which specifically includes the following steps:
[0044] Step 1: Obtain at least two types of partial discharge spectrum data of a transformer.
[0045] Among them, the types of transformer partial discharge spectra mainly include PRPD spectrum, PRPS spectrum, discharge number spectrum, discharge phase spectrum and time domain waveform.
[0046] As an implementation method, the PRPD spectrum and PRPS spectrum of partial discharge of the transformer are obtained; wherein, the PRPD spectrum represents the relationship between phase-amplitude-number of discharges within a certain discharge time; and the PRPS spectrum represents the relationship between cycle-phase-amplitude. It can be seen that the PRPD spectrum focuses more on describing the degree of aggregation of partial discharge in phase and amplitude, while the PRPS spectrum focuses more on representing the changes in the spectrum in different cycles. Based on this, the features of the two spectra can be extracted from the two dimensions of phase and cycle respectively, and then the two features can be fused using the weights of deep learning self-learning, thereby improving the accuracy of discharge type recognition.
[0047] Step 2: preprocess the acquired atlas data to obtain preprocessed atlas data.
[0048] Step 201: In order to reduce information loss, the data used are the original spectrum data of PRPD spectrum and PRPS. First, the values of the acquired spectrum data are normalized to obtain normalized spectrum data.
[0049] Step 202: In order to reduce the influence of phase offset during data acquisition and enhance the robustness of the model, the normalized atlas data is offset by ±10° in the phase dimension to construct a new data set and obtain the preprocessed atlas data.
[0050] Step 3: Input the preprocessed spectrum data into the partial discharge type recognition model to obtain the partial discharge type of the transformer.
[0051] like Figure 1 As shown, the partial discharge type recognition model includes multiple feature extraction networks (each feature extraction network includes an input layer, a feature extraction layer and an auxiliary output layer), a feature map fusion layer, a convolution layer and a main output layer. The number of feature extraction networks is the same as the number of types of partial discharge map data obtained, and each feature extraction network is used to extract features of one type of partial discharge map data.
[0052] Taking the PRPD spectrum and PRPS spectrum of transformer partial discharge as an example, at this time, the partial discharge type recognition model is a two-input and three-output neural network. The input layer inputs the PRPD spectrum and the PRPS spectrum respectively; the output layers are: the last main output layer of the model is responsible for outputting the final recognition result and the loss of the entire partial discharge type recognition model, the feature output layer of the PRPD spectrum and the feature output layer of the PRPS spectrum are two auxiliary output layers, which are mainly used to insert auxiliary losses for their respective feature extraction layers, so that the two feature extraction modules can be trained smoothly even when the main loss of the model is very high.
[0053] The main body of each feature extraction network is a Resnet network that integrates the attention mechanism. The attention mechanism is a technology that allows the model to focus on important information and fully learn and absorb it. Figure 4 As shown in the figure, the attention mechanism calculates the output value Attention Value by giving a query vector q related to the task, calculating the attention distribution of the query vector q and the feature key and attaching it to the value for weighting. The key is used to extract the weight and is the feature map previously input for training. The value is used to be weighted and is the feature map of the query vector query. Its implementation is mainly divided into three steps:
[0054] (1) Obtain input information, determine the keyword feature Key and the value to be predicted Value from the input information, and calculate the similarity: use X = [X 1 ,X 2 ,……X N ] represents N input information, Value and key are both input information X, let key = value = X, query and key are similarity s(X i ,q) calculation and obtain the weight.
[0055] (2) Calculate the attention distribution α: Use softmax to normalize the weights, and then the attention distribution can be given:
[0056] α i =softmax(s(key i ,q))=softmax(s(X i ,q)) (1)
[0057] Among them, α i It is called attention distribution, s(X i ,q) is the attention scoring mechanism, there are several scoring mechanisms:
[0058] Additive Model:
[0059] Dot product model:
[0060] Scaling the dot product model:
[0061] Bilinear Model:
[0062] (3) Information weighted average: attention distribution α i It can be interpreted as the degree of attention paid to the i-th information when querying the context q. A soft information selection mechanism is used to encode the input information X as follows:
[0063] att(q,X)=∑α i X i (6)
[0064] The feature extraction network used in this invention uses the CBAM (Convolutional Block Attention Module) structure to implement the attention mechanism, and its structure diagram is as follows: Figure 2 As shown in the figure, the conventional attention mechanism only focuses on which layers have stronger feedback capabilities at the channel level. The CBAM structure uses the attention mechanism at both the channel level and the spatial level. The channel level is used to capture the features of different types of partial discharge spectrum data, and the spatial level is used to capture the aggregation relationship of the phase, amplitude and period of the partial discharge spectrum data, thereby improving the feature extraction capability of the feature extraction network without significantly increasing the amount of calculation.
[0065] When F is used as the input of the feature extraction network, CBAM mainly performs the following two operations on it:
[0066]
[0067]
[0068] Among them, M c It represents the operation of extracting attention distribution in the channel dimension, M s It represents the operation of extracting attention distribution in the spatial dimension. The operator It represents the element-wise product.
[0069] As a technology to improve the model, the CBAM structure can be embedded in the Resnet network to obtain Figure 3 feature extraction network.
[0070] The output layer of each feature extraction network obtains the transformer partial discharge type recognition result of the corresponding type of partial discharge map data after passing through the sparse layer and the sigmoid activation function. The loss of the recognition result is used as the auxiliary loss of the feature extraction network; the feature map fusion layer fuses the features of the partial discharge map data extracted by each feature extraction network to obtain a fused feature map; the fused feature map is output using an activation function after passing through one convolution layer, and the output is the partial discharge type of the transformer identified by the partial discharge type recognition model.
[0071] For example, the output layers of the two feature extraction networks obtain the prediction results of the PRPD spectrum and the PRPS spectrum respectively after the sparse layer and the sigmoid activation function. The loss of this prediction result can be used as the auxiliary loss of each feature extraction network. At the same time, the features of the two N*1 partial discharge spectrum data are fused into an N*2 feature map, and then after a 1*1 convolution layer, the sigmoid activation function is used to output it. This output is the final output result of the entire model.
[0072] At present, most of the research on partial discharge is still based on high-voltage test measurement. Common partial discharge detection methods can be divided into electrical measurement and non-electrical measurement methods. Electrical measurement methods include conventional pulse current method, broadband pulse current method, and ultra-high frequency method. Non-electrical measurement methods include ultrasonic detection method, chromatographic analysis method, optical measurement method, etc. The above methods can be used to convert discharge signals into statistical graphs or time domain waveforms for analysis.
[0073] The local discharge spectra commonly used in the field of local discharge research mainly include PRPD spectra, PRPS spectra, discharge frequency spectra, discharge phase spectra, time domain waveforms, etc. Most of the research in recent years has been conducted on one of the spectra. Although there have been relatively fruitful interim results in the research on the mechanism of the generation and development of different types of local discharges, due to the influence of factors such as detection methods, transformer types, and surrounding environment, even the discharge spectra of the same discharge type often have huge differences, and it is difficult to extract effective features from a single spectra. The present invention uses the characteristics of different spectra to describe different features, extracts effective features of data from a multi-dimensional perspective, and improves the feature description ability of the model.
[0074] In view of the fact that the features of the discharge spectrum are not obvious, the present invention adopts the attention mechanism to enhance the feature extraction ability of the model, and weights it in the two dimensions of channel and space. Compared with ordinary convolutional neural networks, it can focus on effective information more effectively.
[0075] The present invention also has strong scalability. The method used in this paper is the fusion method of PRPD and PRPS graph features. In the application process, different graphs can be used for fusion according to actual needs.
[0076] Embodiment 2
[0077] This embodiment provides a transformer partial discharge type identification system, which specifically includes the following modules:
[0078] A data acquisition module is configured to: acquire at least two types of partial discharge spectra of the transformer;
[0079] A preprocessing module, which is configured to: preprocess the acquired atlas;
[0080] A partial discharge type identification module is configured to: input the preprocessed spectrum into a partial discharge type identification model to obtain a partial discharge type of the transformer;
[0081] Among them, the feature extraction network of the partial discharge type recognition model adopts the CBAM structure to apply the attention mechanism simultaneously in two dimensions: the channel level and the spatial level. The channel level is used to capture the characteristics of different types of partial discharge spectra, and the spatial level is used to capture the aggregation relationship of the phase, amplitude and period of the partial discharge spectrum.
[0082] It should be noted here that the various modules in this embodiment correspond one-to-one to the various steps in Example 1, and the specific implementation process is the same, which will not be repeated here.
[0083] Embodiment 3
[0084] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps in the transformer partial discharge type identification method described in the above-mentioned embodiment 1 are implemented.
[0085] Embodiment 4
[0086] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the transformer partial discharge type identification method described in the first embodiment are implemented.
[0087] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program code.
[0088] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0089] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0090] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.
[0091] A person skilled in the art can 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, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.
[0092] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A method for identifying the type of partial discharge of a transformer, characterized in that: include: Obtain at least two types of partial discharge spectra of the transformer; Preprocessing the acquired spectra; The preprocessing includes: normalizing the values of the acquired spectrum data; shifting the normalized spectrum data in the phase dimension, constructing a new data set, and obtaining a preprocessed spectrum; The preprocessed spectrum is input into the partial discharge type recognition model to obtain the partial discharge type of the transformer; The partial discharge type recognition model includes multiple feature extraction networks, feature map fusion layers, convolution layers and main output layers, each feature extraction network includes an input layer, a feature extraction layer and an auxiliary output layer; the number of the feature extraction networks is the same as the number of types of partial discharge maps obtained, and each feature extraction network is used to extract features of one type of partial discharge map; The input layer of the partial discharge type recognition model inputs the acquired graphs respectively; the output layers are: the final main output layer of the model is responsible for outputting the final recognition result and the loss of the entire partial discharge type recognition model, and the feature output layer of the acquired graph is an auxiliary output layer, which inserts auxiliary losses for the respective feature extraction layers; Among them, the feature extraction network of the partial discharge type recognition model adopts the CBAM structure to apply the attention mechanism to both the channel level and the spatial level. The channel level is used to capture the features of different types of partial discharge spectra, and the spatial level is used to capture the aggregation relationship of the phase, amplitude and period of the partial discharge spectrum. When F is used as the input of the feature extraction network, CBAM performs the following two operations on it: Among them, M c It represents the operation of extracting attention distribution in the channel dimension, M s It represents the operation of extracting attention distribution in the spatial dimension. The operator It represents the element-wise multiplication; The output layer of each feature extraction network obtains a transformer partial discharge type recognition result of a corresponding type of partial discharge spectrum after passing through a sparse layer and an activation function, and the loss of the recognition result is used as an auxiliary loss of the feature extraction network; The feature map fusion layer is used to fuse the features of the partial discharge maps extracted by each feature extraction network to obtain a fused feature map; The fused feature map is outputted using an activation function after passing through a convolution layer, and the output is the partial discharge type of the transformer identified by the partial discharge type recognition model.
2. A transformer partial discharge type identification method as claimed in claim 1, characterized in that: The attention mechanism specifically includes: Get input information and give a query vector; Calculate the similarity between the query vector and the input information, obtain the weight, and normalize the weight to obtain the attention distribution; The attention distribution is attached to the input information for weighting to obtain the output value.
3. A transformer partial discharge type identification system based on the method according to any one of claims 1-2, characterized in that: include: A data acquisition module is configured to: acquire at least two types of partial discharge spectra of the transformer; A preprocessing module, which is configured to: preprocess the acquired atlas; A partial discharge type identification module is configured to: input the preprocessed spectrum into a partial discharge type identification model to obtain a partial discharge type of the transformer; Among them, the feature extraction network of the partial discharge type recognition model adopts the CBAM structure to apply the attention mechanism simultaneously in two dimensions: the channel level and the spatial level. The channel level is used to capture the characteristics of different types of partial discharge spectra, and the spatial level is used to capture the aggregation relationship of the phase, amplitude and period of the partial discharge spectrum.
4. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in a transformer partial discharge type identification method as described in any one of claims 1-2 are implemented.
5. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the transformer partial discharge type identification method according to any one of claims 1-2 are implemented.
