Transformer partial discharge signal type identification method and device, terminal equipment and storage medium

Abstract

The invention discloses a transformer partial discharge signal type identification method and apparatus, a terminal device and a storage medium. The method comprises the steps of measuring actual partial discharge signal data of a target transformer; inputting the actual partial discharge signal data into a feature extraction model for feature enhancement, so that the feature extraction model generates the actual partial discharge signal data subjected to feature enhancement according to the input data; and inputting the actual partial discharge signal data subjected to feature enhancement into a pre-constructed discharge signal type classifier, so that the discharge signal type classifier identifies and obtains a partial discharge signal type label of the actual partial discharge signal data. According to the method and the device, the partial discharge signal type label of the actual partial discharge signal data can be accurately identified, operation and maintenance personnel can determine the specific type of internal discharge of the transformer, and safe and stable operation of the transformer is guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of transformer fault detection, and in particular to a transformer partial discharge signal type recognition method, device, terminal equipment and storage medium. Background Art

[0002] With the rapid development of my country's power industry and the continuous increase in voltage levels, the stable operation of power transformers has become particularly important. Insulation failure is the main reason affecting the stable operation of power transformers, and partial discharge is one of the important reasons leading to transformer insulation failure.

[0003] In the existing technology, the local discharge detection technology mainly includes pulse current detection method, ultrasonic detection method and ultra-high frequency detection method. Compared with other local discharge detection technologies, the local discharge signal measured by the ultra-high frequency detection method has very rich time and frequency information due to its wide frequency band. Its monitoring frequency band is often higher than the frequency of the spatial interference electromagnetic wave of the power equipment, so it has good anti-interference ability and measurement sensitivity. Therefore, technicians usually use ultra-high frequency detection method to realize online monitoring of partial discharge of power transformers. However, affected by factors such as the propagation path of electromagnetic wave signals and the difference in defect discharge types, although the detection signal amplitude and the degree of defect degradation are basically consistent in trend, it is still impossible to accurately quantify the degree of defect degradation and it is difficult to quantitatively describe the type of defect discharge signal. In engineering, it is necessary to use other methods in conjunction. Summary of the invention

[0004] The present invention provides a method, a device, a terminal device and a storage medium for identifying the type of a transformer partial discharge signal. The method can accurately and effectively identify the type of a transformer partial discharge signal.

[0005] An embodiment of the present invention provides a method for identifying a transformer partial discharge signal type, comprising:

[0006] Measure actual partial discharge signal data of the target transformer; wherein the actual partial discharge signal data includes amplitude, frequency, duration, pulse interval, rise time, waveform characteristics, signal-to-noise ratio and phase data;

[0007] Inputting the actual partial discharge signal data into a feature extraction model for feature enhancement, so that the feature extraction model generates actual partial discharge signal data with feature enhancement completed according to the input data;

[0008] The feature-enhanced actual partial discharge signal data is input into a pre-built discharge signal type classifier, so that the discharge signal type classifier identifies and obtains a partial discharge signal type label of the actual partial discharge signal data according to the feature-enhanced actual partial discharge signal data.

[0009] Furthermore, the step of inputting the actual partial discharge signal data into a feature extraction model for feature enhancement so that the feature extraction model generates actual partial discharge signal data with feature enhancement completed according to the input data includes:

[0010] Inputting the actual partial discharge signal data into a first feature extraction model, so that the first feature extraction model generates actual partial discharge signal data with preliminary feature enhancement according to the input data;

[0011] The actual partial discharge signal data after the initial feature enhancement is input into a second feature extraction model, so that the second feature extraction model outputs the actual partial discharge signal data after the initial feature enhancement based on the actual partial discharge signal data after the initial feature enhancement.

[0012] Furthermore, the generation process of the first feature extraction model includes:

[0013] Acquire a historical partial discharge signal data set; wherein the historical partial discharge signal data set includes a number of measured historical partial discharge signal data;

[0014] Constructing a stacked autoencoder; wherein the stacked autoencoder is composed of a plurality of autoencoders connected in sequence;

[0015] Taking the historical partial discharge signal dataset as input and the historical partial discharge signal dataset after the extracted features are initially enhanced as output, iteratively training the autoencoders in the stacked autoencoder in sequence;

[0016] The trained stacked autoencoder is used as the first feature extraction model;

[0017] Among them, when the autoencoders in the stacked autoencoders are iteratively trained in sequence, the feature data output by the previous trained autoencoder is used as the input of the current autoencoder to be trained; when the current autoencoder to be trained is the first autoencoder of the stacked autoencoder, the historical partial discharge signal data set is used as the input of the first autoencoder.

[0018] Furthermore, the generation process of the second feature extraction model includes:

[0019] Acquire a historical partial discharge signal data set after the feature is initially enhanced;

[0020] Building a generative adversarial network; wherein the generative adversarial network includes a generator and a discriminator;

[0021] Generate some random noise;

[0022] Inputting the historical partial discharge signal data set after the initial feature enhancement and the random noise into the generative adversarial network, so that the generative adversarial network performs iterative training according to the input data until the loss function of the generative adversarial network reaches a preset convergence condition;

[0023] The trained generative adversarial network is used as the second feature extraction model.

[0024] Furthermore, the historical partial discharge signal data set after the initial feature enhancement and the random noise are input into the generative adversarial network, so that the generative adversarial network is iteratively trained according to the input data until the loss function of the generative adversarial network reaches a preset convergence condition, including:

[0025] Inputting the historical partial discharge signal data set after the initial feature enhancement and the random noise into the generative adversarial network, so that the generative adversarial network alternately trains the generator and the discriminator according to the input data and a preset training round until the loss function of the generative adversarial network reaches a preset convergence condition;

[0026] Among them, in each iterative training process, the generator generates a quasi-real partial discharge signal with a similar distribution to the historical partial discharge signal data set after the feature is initially enhanced based on the historical partial discharge signal data set after the feature is initially enhanced and the random noise, and inputs the quasi-real partial discharge signal to the discriminator; the discriminator performs discriminant analysis based on the quasi-real partial discharge signal and the historical partial discharge signal data set after the feature is initially enhanced, and returns the result of the discriminant analysis to the generator.

[0027] Furthermore, the construction process of the discharge signal type classifier includes:

[0028] Acquire a known type label corresponding to the historical partial discharge signal data measured in the historical partial discharge signal data set;

[0029] Integrate the historical partial discharge signal data set and the known type labels into a partial discharge signal label sample set;

[0030] Constructing a classifier, taking the partial discharge signal label sample set as input and the predicted signal type label as output, and iteratively training the classifier;

[0031] Using the trained classifier as the discharge signal type classifier;

[0032] Among them, in each iterative training process, the signal type label predicted by the current classifier is compared with the known type label in the partial discharge signal label sample set, and the loss function value is calculated according to the comparison result; it is judged whether the loss function value converges, if so, the training is stopped to obtain the classifier after the training is completed, if not, the model parameters of the classifier are adjusted to obtain the classifier for the next iterative training.

[0033] Furthermore, the types of the partial discharge signal type labels include: corona discharge, internal discharge, oil wedge discharge, suspension discharge, tip discharge and surface discharge.

[0034] An embodiment of the present invention further provides a transformer partial discharge signal type identification device, comprising: a data acquisition module, a feature enhancement module and a type label identification module;

[0035] The data acquisition module is used to measure the actual partial discharge signal data of the target transformer; wherein the actual partial discharge signal data includes amplitude, frequency, duration, pulse interval, rise time, waveform characteristics, signal-to-noise ratio and phase data;

[0036] The feature enhancement module is used to input the actual partial discharge signal data into a feature extraction model for feature enhancement, so that the feature extraction model generates the actual partial discharge signal data with feature enhancement completed according to the input data;

[0037] The type label identification module is used to input the feature-enhanced actual partial discharge signal data into a pre-built discharge signal type classifier, so that the discharge signal type classifier identifies and obtains the partial discharge signal type label of the actual partial discharge signal data based on the feature-enhanced actual partial discharge signal data.

[0038] The present application also provides a terminal device, including:

[0039] one or more processors;

[0040] A memory, coupled to the processor, for storing one or more programs;

[0041] When the one or more programs are executed by the one or more processors, the one or more processors implement the transformer partial discharge signal type identification method as described in the above-mentioned embodiment of the invention.

[0042] The present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for identifying the type of transformer partial discharge signal as described in the above-mentioned embodiment of the invention is implemented.

[0043] The following beneficial effects are achieved by implementing the present invention:

[0044] The present invention provides a transformer partial discharge signal type identification method, device, terminal equipment and storage medium. In view of the fact that the acquired actual partial discharge signal data has unclear features and is easily covered by interference factors such as noise, the method first inputs the acquired actual partial discharge signal data into a feature extraction model for feature enhancement before type identification, and generates actual partial discharge signal data with feature enhancement completed; thereby, a subsequent discharge signal type classifier is more likely to focus on key feature performance, thereby improving the accuracy and reliability of classification;

[0045] Then, the actual partial discharge signal data after feature enhancement is input into the discharge signal type classifier: the partial discharge signal type label of the actual partial discharge signal data is accurately identified; thereby, the problem of quantitative description of defect discharge signal types in the prior art is effectively solved, which helps operation and maintenance personnel to quickly identify the specific type of internal discharge of the transformer, and then take corresponding maintenance and repair measures in a targeted manner to prevent the fault from further deteriorating and ensure the safe and stable operation of the transformer. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solution of the present application, the drawings required for use in the implementation manner will be briefly introduced below. Obviously, the drawings described below are only some implementation manners of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0047] Figure 1 It is a flowchart of a transformer partial discharge signal type identification method provided by a certain embodiment of the present application;

[0048] Figure 2 It is a structural schematic diagram of a transformer partial discharge signal type identification device provided in a certain embodiment of the present application;

[0049] Figure 3 It is a structural diagram of a terminal device provided in a certain embodiment of the present application. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solutions and advantages of this application clearer, the technical solutions in this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0051] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by technicians in the technical field to which this application belongs; the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" in the specification and claims of this application and the above-mentioned figure descriptions and any variations thereof are intended to cover non-exclusive inclusions.

[0052] In the description of the embodiments of the present application, the technical terms "first", "second", etc. are only used to distinguish different objects, and cannot be understood as indicating or implying relative importance or implicitly indicating the number, specific order or primary and secondary relationship of the indicated technical features. In the description of the embodiments of the present application, the meaning of "multiple" is more than two, unless otherwise clearly and specifically defined.

[0053] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0054] In the description of the embodiments of the present application, the term "and / or" is only a description of the association relationship of the associated objects, indicating that there may be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0055] In the description of the embodiments of the present application, the term "multiple" refers to more than two (including two). Similarly, "multiple groups" refers to more than two groups (including two groups), and "multiple pieces" refers to more than two pieces (including two pieces).

[0056] In the description of the embodiments of the present application, unless otherwise clearly specified and limited, technical terms such as "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, and it can be the internal connection of two elements or the interaction relationship between two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the embodiments of the present application can be understood according to the specific circumstances.

[0057] See also Figure 1 , is a flow chart of a transformer partial discharge signal type identification method provided by an embodiment of the present invention, comprising:

[0058] S1. Measure actual partial discharge signal data of a target transformer; wherein the actual partial discharge signal data includes amplitude, frequency, duration, pulse interval, rise time, waveform characteristics, signal-to-noise ratio and phase data;

[0059] Specifically, first, a transformer to be detected is determined, and the transformer to be detected is used as a target transformer; then, multiple ultra-high frequency sensors are used to measure the key parts of the target transformer to obtain actual partial discharge signal data of the target transformer;

[0060] Specifically, the actual partial discharge signal data includes amplitude, frequency, duration, pulse interval, rise time, waveform characteristics, signal-to-noise ratio and phase data.

[0061] S2, inputting the actual partial discharge signal data into a feature extraction model for feature enhancement, so that the feature extraction model generates actual partial discharge signal data with feature enhancement completed according to the input data;

[0062] In a preferred embodiment, the step of inputting the actual partial discharge signal data into a feature extraction model for feature enhancement so that the feature extraction model generates actual partial discharge signal data with feature enhancement according to the input data, comprises:

[0063] Inputting the actual partial discharge signal data into a first feature extraction model, so that the first feature extraction model generates actual partial discharge signal data with preliminary feature enhancement according to the input data;

[0064] Inputting the actual partial discharge signal data after the initial feature enhancement into a second feature extraction model, so that the second feature extraction model outputs the actual partial discharge signal data after the initial feature enhancement based on the actual partial discharge signal data;

[0065] Indicatively, due to the complex environment of the target transformer and the relatively large external interference when the target transformer is partially discharged, useless interference signals are mixed in the partial discharge signal of the target transformer; if such interference signals are not removed, such interference signals will have a great impact on the subsequent identification. Therefore, it is necessary to pre-process the actual partial discharge signal data obtained by using a nonlinear filtering method to eliminate the interfering random signals;

[0066] In order to improve the accuracy of subsequent identification, the present application introduces a feature enhancement process in step S2 to effectively enhance the features in the actual partial discharge signal data;

[0067] Specifically, step S2 introduces a first feature extraction model and a second feature extraction model; the actual local discharge signal data first enters the first feature extraction model for a feature enhancement operation to generate actual local discharge signal data after preliminary feature enhancement, and then the actual local discharge signal data after preliminary feature enhancement is input into the second feature extraction model to generate actual local discharge signal data after feature enhancement.

[0068] In a preferred embodiment, the generation process of the first feature extraction model includes:

[0069] Acquire a historical partial discharge signal data set; wherein the historical partial discharge signal data set includes a number of measured historical partial discharge signal data;

[0070] Constructing a stacked autoencoder; wherein the stacked autoencoder is composed of a plurality of autoencoders connected in sequence;

[0071] Taking the historical partial discharge signal dataset as input and the historical partial discharge signal dataset after the extracted features are initially enhanced as output, iteratively training the autoencoders in the stacked autoencoder in sequence;

[0072] The trained stacked autoencoder is used as the first feature extraction model;

[0073] Wherein, when the autoencoders in the stacked autoencoders are iteratively trained in sequence, the feature data output by the previously trained autoencoder is used as the input of the current autoencoder to be trained; when the current autoencoder to be trained is the first autoencoder of the stacked autoencoder, the historical partial discharge signal data set is used as the input of the first autoencoder;

[0074] Schematically, the first feature extraction model is trained based on a stacked autoencoder (SAE);

[0075] Specifically, firstly, a stacked autoencoder is constructed, wherein the stacked autoencoder is composed of a plurality of autoencoders connected in sequence;

[0076] When the autoencoders in the stacked autoencoder are iteratively trained in sequence,

[0077] If the current autoencoder to be trained is the first autoencoder of the stacked autoencoder, the acquired historical partial discharge signal data set is used as input data x, mapped to the hidden layer, and the hidden layer feature y is obtained; then the hidden layer feature y is mapped to the output layer to obtain feature data z;

[0078] If the current autoencoder to be trained is not the first autoencoder of the stacked autoencoder, the feature data z output by the previous trained autoencoder is used as the input data x of the current autoencoder to be trained;

[0079] If the current autoencoder to be trained is the last autoencoder of the stacked autoencoders, the output feature data is a historical partial discharge signal data set after preliminary feature enhancement;

[0080] Specifically, the training process expression is as follows:

[0081]

[0082] Where W1 is the weight matrix between the input layer and the hidden layer, W2 is the weight matrix between the hidden layer and the output layer, b is the bias vector of the hidden layer, and d is the bias vector of the output layer;

[0083] wherein the feature data is used as input to a next autoencoder to be trained connected to the first autoencoder;

[0084] Sf and Sg are both sigmoid functions, and the expressions are:

[0085]

[0086] Take W1 as the transpose W2' of W2, that is: W1 = W2' = W, thereby reducing the parameters;

[0087] c(x,z) is the training target of the stacked autoencoder, which is to minimize the difference between the historical partial discharge signal dataset z after the initial feature enhancement and the historical partial discharge signal dataset x, that is:

[0088]

[0089] The overall training objectives are:

[0090] Specifically, in the process of iterative training of each autoencoder, the network weights need to be continuously adjusted until convergence, the training process of the currently trained autoencoder is completed, and the next autoencoder to be trained is trained; each autoencoder is trained in turn, so that each autoencoder learns a higher level of feature representation;

[0091] Specifically, the stacked autoencoder after training is used as the first feature extraction model. By constructing the first feature extraction model, the first feature extraction model can capture subtle changes in the acquired signal and improve recognition accuracy.

[0092] In a preferred embodiment, the generation process of the second feature extraction model includes:

[0093] Acquire a historical partial discharge signal data set after the feature is initially enhanced;

[0094] Building a generative adversarial network; wherein the generative adversarial network includes a generator and a discriminator;

[0095] Generate some random noise;

[0096] Inputting the historical partial discharge signal data set after the initial feature enhancement and the random noise into the generative adversarial network, so that the generative adversarial network performs iterative training according to the input data until the loss function of the generative adversarial network reaches a preset convergence condition;

[0097] Using the trained generative adversarial network as the second feature extraction model;

[0098] Schematically, a generative adversarial network is built, wherein the generative adversarial network consists of two neural networks: a generator and a discriminator;

[0099] Specifically, the network architecture of the generator is a 3-layer fully connected layer, and a 3-layer fully connected neural network is built, including an input layer, an intermediate hidden layer and an output layer; wherein the input layer neuron nodes of the generator are 100, the first layer of the intermediate hidden layer has 256 neurons, the second layer has 128 neurons, the output layer neuron nodes are 6, the training cycle is 1000, the learning rate is 0.0002, and the activation function is the Relu function, and the expression is:

[0100] f(x)=max(0,x);

[0101] Specifically, the network architecture of the discriminator is a 3-layer fully connected layer, and a 3-layer fully connected neural network is built, including an input layer, an intermediate hidden layer and an output layer; wherein, the input layer neuron nodes of the discriminator are 6, the first layer of the intermediate hidden layer has 128 neurons, the second layer has 64 neurons, the output layer neuron node is 1, the training cycle is 1000, the learning rate is 0.0002, the output layer activation function is a sigmoid function, ensuring that the output range is between 0 and 1, and each hidden layer uses a Relu function as an activation function;

[0102] The parameters of the generation model and the discriminant model are randomly initialized to complete the construction of the generator and the discriminator.

[0103] Specifically, the loss function of the generative adversarial network is constructed. Since the original GAN ​​model has the problems of gradient vanishing and mode collapse, the present application introduces Wasserstein distance as the loss function of the generative adversarial network. The objective function expression of the Wasserstein distance is as follows:

[0104]

[0105] Where: E is the expectation, Pdata is the probability distribution of the real sample x; Pz is the prior distribution of the random vector z;

[0106] It should be noted that, in the present application, the real sample is a historical partial discharge signal data set after the feature is initially enhanced.

[0107] In a preferred embodiment, the historical partial discharge signal data set after the initial feature enhancement and the random noise are input into the generative adversarial network, so that the generative adversarial network is iteratively trained according to the input data until the loss function of the generative adversarial network reaches a preset convergence condition, including:

[0108] Inputting the historical partial discharge signal data set after the initial feature enhancement and the random noise into the generative adversarial network, so that the generative adversarial network alternately trains the generator and the discriminator according to the input data and a preset training round until the loss function of the generative adversarial network reaches a preset convergence condition;

[0109] Wherein, in each iterative training process, the generator generates a quasi-real partial discharge signal having a similar distribution to the historical partial discharge signal data set after the initial feature enhancement according to the historical partial discharge signal data set after the initial feature enhancement and the random noise, and inputs the quasi-real partial discharge signal to the discriminator; the discriminator performs discriminant analysis according to the quasi-real partial discharge signal and the historical partial discharge signal data set after the initial feature enhancement, and transmits the result of the discriminant analysis back to the generator;

[0110] Specifically, the RMSProp optimizer is used to optimize the losses of the generator and the discriminator, the learning rate is set to 0.0001, and the minimum optimization method is adopted;

[0111] Specifically, the preset training rounds are: in each iterative training process, the generator is trained once, and the discriminator is trained three times;

[0112] Specifically, in each iterative training process, the generator generates a quasi-real partial discharge signal with a similar distribution to the historical partial discharge signal data set after the initial enhancement of the feature and the random noise, and inputs the quasi-real partial discharge signal to the discriminator, which performs discriminant analysis based on the quasi-real partial discharge signal and the historical partial discharge signal data set after the initial enhancement of the feature, and transmits the result of the discriminant analysis back to the generator; thus, in the training process, the generator attempts to generate a more realistic discharge signal, and the discriminator strives to distinguish between the real signal and the generated signal. Through this adversarial process, the generator gradually learns the characteristics of the partial discharge signal.

[0113] S3, inputting the feature-enhanced actual partial discharge signal data into a pre-built discharge signal type classifier, so that the discharge signal type classifier identifies and obtains a partial discharge signal type label of the actual partial discharge signal data according to the feature-enhanced actual partial discharge signal data;

[0114] In a preferred embodiment, the process of constructing the discharge signal type classifier includes:

[0115] Acquire a known type label corresponding to the historical partial discharge signal data measured in the historical partial discharge signal data set;

[0116] Integrate the historical partial discharge signal data set and the known type labels into a partial discharge signal label sample set;

[0117] Constructing a classifier, taking the partial discharge signal label sample set as input and the predicted signal type label as output, and iteratively training the classifier;

[0118] Using the trained classifier as the discharge signal type classifier;

[0119] In each iterative training process, the signal type label predicted by the current classifier is compared with the known type label in the partial discharge signal label sample set, and the loss function value is calculated according to the comparison result; it is determined whether the loss function value converges, if so, the training is stopped to obtain the classifier after the training is completed, if not, the model parameters of the classifier are adjusted to obtain the classifier for the next iterative training;

[0120] In an illustrative manner, the measured historical partial discharge signal data in the historical partial discharge signal data set are labeled and classified;

[0121] For data with clear discharge types, they are labeled with known types, expressed as:

[0122] {(x(1),X(1)),…,(x(m),X(k))};

[0123] For data with unclear discharge type, no type label is added and it is expressed as:

[0124] {(x(1),…,(x(m)};

[0125] Where x(i) represents the i-th data, X(j) represents the j-th discharge type, X(j)∈{1,2,...,k}, k is the number of categories;

[0126] Integrate the historical partial discharge signal data set and the known type labels into a partial discharge signal label sample set; wherein the known type labels include corona discharge, internal discharge, oil wedge discharge, suspension discharge, tip discharge, and surface discharge;

[0127] Specifically, a classifier is constructed, wherein the classifier uses Softmax to be applicable to multi-classification problems and can give each classification result in a probability form;

[0128] Use labeled samples {(x(1),X(1)),…,(x(m),X(k))} to build the classifier;

[0129] Since each historical partial discharge signal data measured in the historical partial discharge signal data set includes amplitude, frequency, duration, pulse interval, rise time, waveform characteristics, signal-to-noise ratio and phase, the classifier input layer has 8 nodes, and the number of nodes in the output layer is equal to the number of categories of known type labels, then the output layer has 6 nodes;

[0130] Therefore, the regression hypothesis of the constructed classifier is:

[0131]

[0132] Where: p(X (i) =j|x (i) ; θ) represents the probability that the i-th historical partial discharge signal data belongs to the j-th category; θ 1 ,θ 2 ,…,θ k All of them are parameter vectors of the classifier;

[0133] The cost function of the classifier is defined as:

[0134]

[0135] Where: m is the number of data; 1{·} represents the indicator function, when the expression in the brackets is true, the function value is 1, otherwise it is 0; λ is the weight attenuation coefficient;

[0136] Specifically, the classifier is iteratively trained. In each iterative training process, the signal type label predicted by the current classifier is compared with the known type labels in the partial discharge signal label sample set, and the output prediction value is calculated. The expression is:

[0137]

[0138] The loss function value is calculated based on the comparison results, where the chain rule is used to calculate the gradient of the loss to the model parameters, and its expression is:

[0139]

[0140] Determine whether the loss function value converges, if so, stop training and obtain the classifier after training.

[0141] If not, the model parameters of the classifier are adjusted to obtain the classifier for the next iterative training, wherein the expression for adjusting the model parameters is:

[0142]

[0143] Where η is the learning rate.

[0144] Thus, the feature-enhanced actual partial discharge signal data is input into a pre-built discharge signal type classifier, so that the discharge signal type classifier identifies and obtains a partial discharge signal type label of the actual partial discharge signal data according to the feature-enhanced actual partial discharge signal data;

[0145] In a preferred embodiment, the types of the partial discharge signal type labels include: corona discharge, internal discharge, oil wedge discharge, suspension discharge, tip discharge and surface discharge.

[0146] See also Figure 2 , is a transformer partial discharge signal type identification device provided by an embodiment of the present invention, comprising: a data acquisition module, a feature enhancement module and a type label identification module;

[0147] The data acquisition module is used to measure the actual partial discharge signal data of the target transformer; wherein the actual partial discharge signal data includes amplitude, frequency, duration, pulse interval, rise time, waveform characteristics, signal-to-noise ratio and phase data;

[0148] The feature enhancement module is used to input the actual partial discharge signal data into a feature extraction model for feature enhancement, so that the feature extraction model generates the actual partial discharge signal data with feature enhancement completed according to the input data;

[0149] The type label identification module is used to input the feature-enhanced actual partial discharge signal data into a pre-built discharge signal type classifier, so that the discharge signal type classifier identifies and obtains the partial discharge signal type label of the actual partial discharge signal data based on the feature-enhanced actual partial discharge signal data.

[0150] See also Figure 3 , a certain embodiment of the present application further provides a terminal device, including:

[0151] one or more processors;

[0152] A memory, coupled to the processor, for storing one or more programs;

[0153] When the one or more programs are executed by the one or more processors, the one or more processors implement the transformer partial discharge signal type identification method as described above.

[0154] The processor is used to control the overall operation of the terminal device to complete all or part of the steps of the above transformer partial discharge signal type identification method. The memory is used to store various types of data to support the operation of the terminal device, and these data may include instructions for any application or method used to operate on the terminal device, as well as application-related data. The memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory.

[0155] (Static Random Access Memory, SRAM for short), Electrically Erasable Programmable Read-Only Memory (EEPROM for short), Erasable Programmable Read-Only Memory (EPROM for short), Programmable Read-Only Memory (PROM for short), Read-Only Memory (ROM for short), Magnetic Storage, Flash Memory, Disk or Optical Disk.

[0156] In an exemplary embodiment, the terminal device can be implemented by one or more application specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), controllers, microcontrollers, microprocessors or other electronic components to execute the transformer partial discharge signal type identification method as described in any of the above embodiments and achieve the same technical effect as the above method.

[0157] In another exemplary embodiment, a computer-readable storage medium including a computer program is also provided, and when the computer program is executed by a processor, the steps of the transformer partial discharge signal type identification method as described in any of the above embodiments are implemented. For example, the computer-readable storage medium can be the above-mentioned memory including the computer program, and the above-mentioned computer program can be executed by a processor of a terminal device to complete the transformer partial discharge signal type identification method as described in any of the above embodiments, and achieve the technical effect consistent with the above method.

[0158] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for identifying the type of transformer partial discharge signal, characterized in that: include: Measure actual partial discharge signal data of the target transformer; wherein the actual partial discharge signal data includes amplitude, frequency, duration, pulse interval, rise time, waveform characteristics, signal-to-noise ratio and phase data; Inputting the actual partial discharge signal data into a feature extraction model for feature enhancement, so that the feature extraction model generates actual partial discharge signal data with feature enhancement completed according to the input data; The feature-enhanced actual partial discharge signal data is input into a pre-built discharge signal type classifier, so that the discharge signal type classifier identifies and obtains a partial discharge signal type label of the actual partial discharge signal data according to the feature-enhanced actual partial discharge signal data.

2. The transformer partial discharge signal type identification method according to claim 1, characterized in that: The step of inputting the actual partial discharge signal data into a feature extraction model for feature enhancement so that the feature extraction model generates actual partial discharge signal data with feature enhancement completed according to the input data comprises: Inputting the actual partial discharge signal data into a first feature extraction model, so that the first feature extraction model generates actual partial discharge signal data with preliminary feature enhancement according to the input data; The actual partial discharge signal data after the initial feature enhancement is input into a second feature extraction model, so that the second feature extraction model outputs the actual partial discharge signal data after the initial feature enhancement based on the actual partial discharge signal data after the initial feature enhancement.

3. The transformer partial discharge signal type identification method according to claim 2, characterized in that: The generation process of the first feature extraction model includes: Acquire a historical partial discharge signal data set; wherein the historical partial discharge signal data set includes a number of measured historical partial discharge signal data; Constructing a stacked autoencoder; wherein the stacked autoencoder is composed of a plurality of autoencoders connected in sequence; Taking the historical partial discharge signal dataset as input and the historical partial discharge signal dataset after the extracted features are initially enhanced as output, iteratively training the autoencoders in the stacked autoencoder in sequence; The trained stacked autoencoder is used as the first feature extraction model; Among them, when the autoencoders in the stacked autoencoders are iteratively trained in sequence, the feature data output by the previous trained autoencoder is used as the input of the current autoencoder to be trained; when the current autoencoder to be trained is the first autoencoder of the stacked autoencoder, the historical partial discharge signal data set is used as the input of the first autoencoder.

4. The transformer partial discharge signal type identification method according to claim 3, characterized in that: The generation process of the second feature extraction model includes: Acquire a historical partial discharge signal data set after the feature is initially enhanced; Building a generative adversarial network; wherein the generative adversarial network includes a generator and a discriminator; Generate some random noise; Inputting the historical partial discharge signal data set after the initial feature enhancement and the random noise into the generative adversarial network, so that the generative adversarial network performs iterative training according to the input data until the loss function of the generative adversarial network reaches a preset convergence condition; The trained generative adversarial network is used as the second feature extraction model.

5. The transformer partial discharge signal type identification method according to claim 4, characterized in that: The step of inputting the historical partial discharge signal data set after the initial feature enhancement and the random noise into the generative adversarial network so that the generative adversarial network performs iterative training according to the input data until the loss function of the generative adversarial network reaches a preset convergence condition comprises: Inputting the historical partial discharge signal data set after the initial feature enhancement and the random noise into the generative adversarial network, so that the generative adversarial network alternately trains the generator and the discriminator according to the input data and a preset training round until the loss function of the generative adversarial network reaches a preset convergence condition; Among them, in each iterative training process, the generator generates a quasi-real partial discharge signal with a similar distribution to the historical partial discharge signal data set after the feature is initially enhanced based on the historical partial discharge signal data set after the feature is initially enhanced and the random noise, and inputs the quasi-real partial discharge signal to the discriminator; the discriminator performs discriminant analysis based on the quasi-real partial discharge signal and the historical partial discharge signal data set after the feature is initially enhanced, and returns the result of the discriminant analysis to the generator.

6. The transformer partial discharge signal type identification method according to claim 3, characterized in that: The construction process of the discharge signal type classifier includes: Acquire a known type label corresponding to the historical partial discharge signal data measured in the historical partial discharge signal data set; Integrate the historical partial discharge signal data set and the known type labels into a partial discharge signal label sample set; Constructing a classifier, taking the partial discharge signal label sample set as input and the predicted signal type label as output, and iteratively training the classifier; Using the trained classifier as the discharge signal type classifier; Among them, in each iterative training process, the signal type label predicted by the current classifier is compared with the known type label in the partial discharge signal label sample set, and the loss function value is calculated according to the comparison result; it is judged whether the loss function value converges, if so, the training is stopped to obtain the classifier after the training is completed, if not, the model parameters of the classifier are adjusted to obtain the classifier for the next iterative training.

7. The transformer partial discharge signal type identification method according to claim 1, characterized in that: The types of partial discharge signal type labels include: corona discharge, internal discharge, oil wedge discharge, suspension discharge, tip discharge and surface discharge.

8. A transformer partial discharge signal type identification device, characterized in that: include: Data acquisition module, feature enhancement module and type label recognition module; The data acquisition module is used to measure the actual partial discharge signal data of the target transformer; wherein the actual partial discharge signal data includes amplitude, frequency, duration, pulse interval, rise time, waveform characteristics, signal-to-noise ratio and phase data; The feature enhancement module is used to input the actual partial discharge signal data into a feature extraction model for feature enhancement, so that the feature extraction model generates the actual partial discharge signal data with feature enhancement completed according to the input data; The type label identification module is used to input the feature-enhanced actual partial discharge signal data into a pre-built discharge signal type classifier, so that the discharge signal type classifier identifies and obtains the partial discharge signal type label of the actual partial discharge signal data based on the feature-enhanced actual partial discharge signal data.

9. A terminal device, characterized in that: include: one or more processors; A memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the transformer partial discharge signal type identification method according to any one of claims 1 to 7.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for identifying the type of transformer partial discharge signal is implemented as described in any one of claims 1 to 7.

Images