Method and system for determining transformer fault data set

Abstract

The invention discloses a method and a system for determining a transformer fault data set. The method comprises the steps of obtaining sample transformer fault data, modeling the sample transformer fault data into a dynamic pulse width signal model based on an annihilation filtering algorithm, and determining sample pulse data based on model parameters of the dynamic pulse width signal model, the sample pulse data comprises symmetric amplitude data, pulse width data, asymmetric amplitude data and time delay data; generating target pulse data according to the sample pulse data and a pre-trained pulse parameter generation model, and reconstructing target transformer fault data based on the target pulse data and the dynamic pulse width signal model; and determining a transformer fault data set based on the sample transformer fault data and the target transformer fault data. The problem of insufficient transformer fault data is solved, and the determination efficiency of the transformer fault data set is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data expansion, and in particular to a method and system for determining a transformer fault data set. Background Art

[0002] The research on transformer fault data expansion technology is of great significance in the intelligent operation and maintenance of modern power systems. It mainly serves to improve the efficiency of transformer condition monitoring and fault diagnosis systems.

[0003] With the development of digital power grids and the advancement of the Internet of Things and big data technologies, the amount of real-time monitoring data of transformers has increased significantly. However, due to the sporadic nature and diversity of fault sample data, it is often difficult to accumulate enough cases to train high-precision fault diagnosis models. Summary of the invention

[0004] The present invention provides a method and system for determining a transformer fault data set to solve the problem of insufficient transformer fault data.

[0005] According to one aspect of the present invention, a method for determining a transformer fault data set is provided, the method comprising:

[0006] Acquire sample transformer fault data, model the sample transformer fault data as a dynamic pulse width signal model based on an annihilation filtering algorithm, and determine sample pulse data based on model parameters of the dynamic pulse width signal model, wherein the sample pulse data includes symmetrical amplitude data, pulse width data, asymmetrical amplitude data, and time delay data;

[0007] Generate target pulse data according to the sample pulse data and a pre-trained pulse parameter generation model, and reconstruct target transformer fault data based on the target pulse data and the dynamic pulse width signal model;

[0008] A transformer fault data set is determined based on the sample transformer fault data and the target transformer fault data.

[0009] According to another aspect of the present invention, a system for determining a transformer fault data set is provided, the system comprising:

[0010] A sample pulse data determination module is used to obtain sample transformer fault data, model the sample transformer fault data as a dynamic pulse width signal model based on an annihilation filtering algorithm, and determine the sample pulse data based on model parameters of the dynamic pulse width signal model, wherein the sample pulse data includes symmetrical amplitude data, pulse width data, asymmetrical amplitude data and delay data;

[0011] A data reconstruction module, used to generate target pulse data according to the sample pulse data and a pre-trained pulse parameter generation model, and reconstruct target transformer fault data based on the target pulse data;

[0012] The data set determination module is used to determine a transformer fault data set based on the sample transformer fault data and the target transformer fault data.

[0013] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0014] at least one processor; and

[0015] a memory communicatively connected to the at least one processor; wherein,

[0016] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can perform the method for determining a transformer fault data set according to any embodiment of the present invention.

[0017] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the method for determining a transformer fault data set according to any embodiment of the present invention when executed.

[0018] The technical solution of the embodiment of the present invention obtains sample transformer fault data, models the sample transformer fault data as a dynamic pulse width signal model based on the annihilation filtering algorithm, and determines the sample pulse data based on the model parameters of the dynamic pulse width signal model, wherein the sample pulse data includes symmetrical amplitude data, pulse width data, asymmetrical amplitude data and delay data; the sample transformer fault data is modeled as a dynamic pulse width signal model of a spatial trend curve for signal processing, feature extraction and reconstruction. The target pulse data is generated according to the sample pulse data and the pre-trained pulse parameter generation model, and the target transformer fault data is reconstructed based on the target pulse data and the dynamic pulse width signal model; new target transformer fault data is quickly generated through the sample pulse data and the pulse parameter generation model. The transformer fault data set is determined based on the sample transformer fault data and the target transformer fault data. The transformer fault set is quickly generated based on the sample transformer fault data and the target transformer fault data, which solves the problem of insufficient transformer fault data and achieves the beneficial effects of improving the determination efficiency of the transformer fault data set and improving the expansion efficiency of the transformer fault data.

[0019] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0021] Figure 1 is a flow chart of a method for determining a transformer fault data set provided according to Embodiment 1 of the present invention;

[0022] Figure 2a is a flow chart of a method for determining a transformer fault data set provided according to Embodiment 2 of the present invention;

[0023] Figure 2b is a sample schematic diagram of a sample transformer fault data curve according to an optional example of a method for determining a transformer fault data set provided in Embodiment 2 of the present invention;

[0024] Figure 3 is a structural schematic diagram of a system for determining a transformer fault data set provided according to Embodiment 3 of the present invention;

[0025] Figure 4 It is a structural schematic diagram of an electronic device for implementing the method for determining a transformer fault data set according to an embodiment of the present invention. DETAILED DESCRIPTION

[0026] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0027] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0028] Embodiment 1

[0029] Figure 1 A flowchart of a method for determining a transformer fault data set is provided for the first embodiment of the present invention. This embodiment is applicable to data expansion. The method can be executed by a system for determining a transformer fault data set. The system for determining a transformer fault data set can be implemented in the form of hardware and / or software. The system for determining a transformer fault data set can be configured in an electronic device. Figure 1 As shown, the method includes:

[0030] S110. Acquire sample transformer fault data, model the sample transformer fault data as a dynamic pulse width signal model based on an annihilation filtering algorithm, and determine sample pulse data based on model parameters of the dynamic pulse width signal model, wherein the sample pulse data includes symmetrical amplitude data, pulse width data, asymmetrical amplitude data, and delay data.

[0031] The sample transformer fault data may be understood as sample fault parameters of the transformer.

[0032] Specifically, the types of sample fault parameters are divided into parameter vector space according to a preset order, the parameter amplitude characteristics are normalized according to the type and divided into trend vector space, and the fault data curve is characterized according to the parameters and trend vector space. Exemplarily, the sample fault parameters include at least one of winding temperature, dissolved gas in oil, oil temperature, oil pressure, ultrasonic partial discharge, high-frequency partial discharge, core grounding current, clamp grounding current, transformer operation soundprint, transformer operation current and voltage, oil level, and insulation resistance impedance.

[0033] Optionally, the pulse width signal model is expressed by the following formula:

[0034]

[0035] Where N is the length of the parameter vector space, and the basis function f k(n) is the double pulse waveform, σ(n) is the model mismatch error, and K is the number of pulses in the double pulse waveform.

[0036] It is worth noting that the number of pulses of the double pulse waveform is usually pre-selected based on experience or determined by an annihilation filtering algorithm.

[0037] Optionally, the double pulse waveform is expressed by the following formula:

[0038]

[0039] in, Represents the symmetrical curve in the double pulse waveform, Represents the asymmetric curve in the double pulse waveform, c k Represents symmetrical amplitude data, r k Indicates pulse width data, d k Represents asymmetric amplitude data, n k Indicates delay data.

[0040] S120, generating target pulse data according to the sample pulse data and a pre-trained pulse parameter generation model, and reconstructing target transformer fault data based on the target pulse data and the dynamic pulse width signal model.

[0041] The target pulse data can be understood as new pulse data, and the target transformer fault data can be understood as new transformer fault data.

[0042] Specifically, a pre-trained pulse parameter generation model is used to generate new pulse data that conforms to a specific distribution based on the input sample pulse data. The model parameters of the pulse parameter generation model can be adjusted to generate target pulse data with different characteristics to simulate different transformer fault conditions. The target pulse data is converted into the form of a transformer fault signal through a dynamic pulse width signal model. The generated target pulse data is mapped to the input end of the dynamic pulse width signal model, and the reconstructed target transformer fault data is output through the conversion of the model. The reconstructed target transformer fault data is verified to ensure that it has similar characteristics and distribution to the sample transformer fault data.

[0043] In the embodiment of the present invention, by generating different target pulse data, various transformer fault conditions can be simulated to provide rich test data for the fault diagnosis and early warning system. The data set is effectively enhanced to improve the generalization ability of the fault diagnosis model. By combining the pre-trained pulse parameter generation model and the dynamic pulse width signal model, transformer fault data can be effectively generated and reconstructed.

[0044] Optionally, the pulse parameter generation model includes an inference network, a transient space, a generation network and a thermal encoder, the transient space is respectively connected to the inference network and the generation network, the thermal encoder is respectively connected to the inference network and the generation network, and the input of the inference network is the target pulse data.

[0045] Optionally, the inference network includes a first input layer, a first feedforward neuron layer connected to the first input layer, a first intermediate neuron layer connected to the first feedforward neuron layer, a first post neuron layer connected to the first neuron layer, and a first output layer connected to the first post neuron layer; the generation network includes a second input layer, a second feedforward neuron layer connected to the second input layer, a second intermediate neuron layer connected to the second feedforward neuron layer, a second post neuron layer connected to the second neuron layer, and a second output layer connected to the second post neuron layer; the transient space is connected to the first output layer and the second output layer, respectively, and the output layer of the thermal encoder is connected to the first intermediate neuron layer, the first post neuron layer, the second intermediate neuron layer, and the second post neuron layer, respectively.

[0046] It is worth noting that the dimensions of the inference network and the generation network are different.

[0047] Optionally, the target pulse data is generated according to the sample pulse data and a pre-trained pulse parameter generation model, including: inputting the sample pulse data into the inference network to obtain a transient feature vector converted based on the sample pulse data, and sending the transient feature vector to the generation network through the transient space; inputting the sample label corresponding to the sample pulse data into the hot encoder to obtain a hot encoding coefficient corresponding to the sample label encoding, and sending the hot encoding coefficient to the generation network; the generation network generates the target pulse data based on the received transient feature vector and the hot encoding coefficient.

[0048] Specifically, the sample pulse data is input into the inference network, the inference network processes the sample pulse data, extracts its transient features, and outputs a transient feature vector. The transient feature vector is sent to the generation network through the transient space. The sample label corresponding to the sample pulse data is input into the hot encoder. The hot encoder converts the sample label into a hot encoding coefficient and sends the hot encoding coefficient to the generation network. The generation network receives the transient feature vector from the inference network and the hot encoding coefficient from the hot encoder. Based on the transient feature vector and the hot encoding coefficient, the generation network generates the target pulse data.

[0049] In the embodiment of the present invention, the target pulse data with specific labels or features can be quickly and accurately generated through the pulse parameter generation model. By adjusting the hot encoding coefficient, pulse data with different features or labels can be generated, providing a rich data set for transformer fault diagnosis.

[0050] S130. Determine a transformer fault data set based on the sample transformer fault data and the target transformer fault data.

[0051] The transformer fault data set may be understood as a set of a preset number of transformer fault data.

[0052] Specifically, the sample transformer fault data and the newly generated target transformer fault data are integrated, and the integrated data are preprocessed, including data cleaning, data standardization and normalization, and a transformer fault data set is determined based on a preset number of preprocessed transformer fault data.

[0053] Optionally, determining the transformer fault data set based on the sample transformer fault data and the target transformer fault data includes: reducing the sample transformer fault data based on a random downsampling method to obtain processed sample transformer fault data, and constructing a transformer fault data set based on the processed sample transformer fault data and the target transformer fault data.

[0054] Specifically, the original sample transformer fault data is randomly downsampled. A portion of samples are randomly selected from the majority class (for example; non-fault or common fault) and deleted to reduce their number. The number of samples of different categories is more balanced, which helps to improve the performance of the fault diagnosis model. After random downsampling, the processed sample transformer fault data is obtained. The newly generated target transformer fault data is integrated with the processed sample data. The integrated data is preprocessed, and a transformer fault data set is constructed based on the processed sample transformer fault data and the target transformer fault data. Furthermore, each sample in the transformer fault data set can also be labeled, including relevant information such as fault type and fault degree. Further, the data of the transformer fault data set is divided into a training set, a validation set and a test set. The training set is used for model training, the validation set is used for model selection and parameter adjustment, and the test set is used to evaluate the performance of the model. The divided data set is stored in an appropriate storage medium, and a data set management and maintenance mechanism is established. The transformer fault data set is used for training and testing the fault diagnosis model.

[0055] In the embodiment of the present invention, the original sample transformer fault data is pared down to reduce the imbalance of the data and improve the efficiency, thereby enriching the diversity of the data set and including more samples of different fault conditions.

[0056] The technical solution of the embodiment of the present invention obtains sample transformer fault data, models the sample transformer fault data as a dynamic pulse width signal model based on the annihilation filtering algorithm, and determines the sample pulse data based on the model parameters of the dynamic pulse width signal model, wherein the sample pulse data includes symmetrical amplitude data, pulse width data, asymmetrical amplitude data and delay data; the sample transformer fault data is modeled as a dynamic pulse width signal model of a spatial trend curve for signal processing, feature extraction and reconstruction. The target pulse data is generated according to the sample pulse data and the pre-trained pulse parameter generation model, and the target transformer fault data is reconstructed based on the target pulse data and the dynamic pulse width signal model; new target transformer fault data is quickly generated through the sample pulse data and the pulse parameter generation model. The transformer fault data set is determined based on the sample transformer fault data and the target transformer fault data. The transformer fault set is quickly generated based on the sample transformer fault data and the target transformer fault data, which solves the problem of insufficient transformer fault data and achieves the beneficial effects of improving the determination efficiency of the transformer fault data set and improving the expansion efficiency of the transformer fault data.

[0057] Embodiment 2

[0058] FIG2 is a flow chart of a method for determining a transformer fault data set provided in Embodiment 2 of the present invention. This embodiment is a further refinement of the method for generating target pulse data according to the sample pulse data and a pre-trained pulse parameter generation model in the above embodiment. Optionally, before generating the target pulse data according to the sample pulse data and a pre-trained pulse parameter generation model, it also includes: training the constructed initial pulse parameter generation model through a pre-defined training objective function, calculating the loss function through a back-propagation algorithm during the training process, and updating the model parameters through an optimizer until a preset number of training times is reached; after the training is completed, evaluating the trained model to obtain a model evaluation index, and determining the target pulse parameter generation model based on at least one of the model evaluation indicators.

[0059] like Figure 2a As shown, the method includes:

[0060] S210, obtaining sample transformer fault data, modeling the sample transformer fault data as a dynamic pulse width signal model based on an annihilation filtering algorithm, and determining sample pulse data based on model parameters of the dynamic pulse width signal model, wherein the sample pulse data includes symmetrical amplitude data, pulse width data, asymmetrical amplitude data and delay data.

[0061] S220, training the constructed initial pulse parameter generation model through a predefined training objective function. During the training process, the loss function is calculated through a back propagation algorithm, and the model parameters are updated through an optimizer until a preset number of training times is reached.

[0062] Specifically, the sample transformer fault data is divided into a training set and a test set, and a training objective function is defined, which is used to measure the difference between the pulse parameters generated by the model and the expected pulse parameters. The training objective function is used to calculate the error between the model prediction value and the true value. Before training, the parameters of the model are initialized. The initial pulse parameter generation model is trained based on the training set. In each training process, the error between the pulse parameters output by the model and the expected pulse parameters is calculated by the training objective function (loss function). The model parameters are updated based on the error value. The back propagation algorithm is used to calculate the gradient of the loss function with respect to the model parameters. After the gradient is calculated, the optimizer is used to update the parameters of the model until the preset number of training times is reached.

[0063] Optionally, the training objective function is expressed as follows:

[0064] J CVAE =E Z～Q [lnP(X|Z,Y)]-D KL [Q(Z|X,Y)||P(Z)];

[0065]

[0066] Among them, E Z～Q [] indicates sampling the transient feature vector Z according to distribution Q and calculating the expected value under the sampling, X indicates the sample pulse data, Z indicates the transient feature vector, Y indicates the hot encoding coefficient of the sample label, P(Z) is the prior probability distribution of the transient feature vector Z, P(X|Z) indicates the posterior probability distribution of the output target pulse data when the transient feature vector Z is input, D KL Represents the KL divergence formula.

[0067] Exemplarily, the dimension of the inference network input layer of the CVAE that has completed training is set to K, the output of the hot encoder is 7, the number of intermediate neuron layer nodes of the generation network and the inference network is set to 156, the number of feedforward (post) neuron layer nodes is set to 24, the number of post (feedforward) neuron nodes is 48, and the dimension of the transient feature vector is 18.

[0068] S230. After the training is completed, the trained model is evaluated to obtain a model evaluation index, and a target pulse parameter generation model is determined based on at least one of the model evaluation indicators.

[0069] Specifically, after the training is completed, the trained model is evaluated based on the test set to obtain a model evaluation index, and the model with the best model evaluation index is determined as the target pulse parameter generation model.

[0070] S240, generating target pulse data according to the sample pulse data and a pre-trained pulse parameter generation model, and reconstructing target transformer fault data based on the target pulse data and the dynamic pulse width signal model.

[0071] S250: Determine a transformer fault data set based on the sample transformer fault data and the target transformer fault data.

[0072] The technical solution of the embodiment of the present invention trains the constructed initial pulse parameter generation model through a predefined training objective function. During the training process, the loss function is calculated through the back propagation algorithm, and the model parameters are updated through the optimizer until the preset number of training times is reached. After the training is completed, the trained model is evaluated to obtain a model evaluation index, and the target pulse parameter generation model is determined based on at least one of the model evaluation indicators. The best performing model is used as the target pulse parameter generation model. The model can be used in subsequent pulse parameter generation or transformer fault diagnosis tasks. Improve the fault diagnosis efficiency of the transformer.

[0073] As an optional example of an embodiment of the present invention, the method for determining a transformer fault data set in this embodiment specifically includes the following steps:

[0074] Step 1: First, use the annihilation filtering algorithm to model the sample transformer fault data into a dynamic pulse width signal model, and determine the sample pulse data based on the model parameters of the dynamic pulse width signal model, wherein the sample pulse data includes symmetrical amplitude data, pulse width data, asymmetrical amplitude data and delay data.

[0075] Step 2: Train the pulse parameter generation model.

[0076] Step three: Generate target pulse data based on the sample pulse data and a pre-trained pulse parameter generation model.

[0077] Step 4: Reconstruct target transformer fault data based on the target pulse data and the dynamic pulse width signal model.

[0078] Step 5: The sample transformer fault data is reduced by a random downsampling method to obtain processed sample transformer fault data, and a transformer fault data set is constructed based on the processed sample transformer fault data and the target transformer fault data.

[0079] The specific step one is:

[0080] Figure 2b is a sample schematic diagram of a sample transformer fault data curve according to an optional example of a method for determining a transformer fault data set provided in Embodiment 2 of the present invention, such as Figure 2b As shown, a sample transformer fault data curve is constructed, and its main contents include: dividing the fault parameter types into parameter vector space in order, dividing the parameter amplitude characteristics into trend vector space after normalizing them according to the type, and characterizing the fault data curve according to the parameter and trend vector space.

[0081] The dynamic pulse width signal model is used to model the fault data curve of the sample transformer fault data. The main content is: the sample transformer fault data curve is modeled as the sum of multiple dynamic pulse width signals (dynamic-pulse-width) and model mismatch errors, that is,

[0082]

[0083] Where N is the length of the parameter vector space, and the basis function f k (n) is the double pulse waveform, σ(n) is the model mismatch error, K is the number of pulses in the double pulse waveform, which is usually selected based on experience or determined by the annihilation filter algorithm, and the basis function f k (n) is a double pulse waveform, which can be divided into:

[0084]

[0085] in, Represents the symmetrical curve in the double pulse waveform, Represents the asymmetric curve in the double pulse waveform, c k Represents symmetrical amplitude data, r k Indicates pulse width data, d k Represents asymmetric amplitude data, n k Indicates delay data.

[0086] Due to the expansibility of the period of the double pulse waveform, ignoring the matching error, the Fourier coefficients of the pulse width signal model of the fault data can be expressed as:

[0087]

[0088] Among them, v k is the amplitude sequence coefficient required for subsequent reconstruction, u k For the subsequent reconstruction of the filter roots required, the parameters of the model can also be expressed as:

[0089] Among them, the main contents of the annihilation filtering algorithm are:

[0090] After obtaining discrete samples of the fault data curve, fast Fourier transform is calculated on the samples to obtain the spectrum;

[0091] Constructing such an annihilation filter containing signal parameter information is as follows:

[0092]

[0093] Among them, A[k], k∈[0,K] are the coefficients of the filter, A[0]=1, and R+1 is the number of frequency domain poles.

[0094] Construct the Toeplitz matrix of the estimated annihilation filter coefficient and combine the K sample spectral coefficients to obtain the pulse width r in the dynamic perception coefficient by calculating the root of the filter. k and delay n k .

[0095] Construct the Vandermonde matrix and combine the sample M spectral coefficients to obtain the symmetric amplitude c in the dynamic perception coefficient by calculating the amplitude sequence k , asymmetric amplitude d k .

[0096] The Toeplitz matrix is:

[0097]

[0098] Where G[M] is the Fourier coefficient of the discrete sample of the fault data curve, M=2K.

[0099] The pulse width r in the dynamic perception coefficient is obtained by calculating the root of the filter: k and delay n k .

[0100]

[0101] A'=(T T' T') -1 T T G;

[0102]

[0103] Among them, u k is the root of filter A, which can be calculated by the least squares method.

[0104] (4.3) The Vandermonde matrix is ​​constructed and combined with the sample M spectral coefficients to obtain the symmetric amplitude c in the dynamic perception coefficient by calculating the amplitude sequence k , asymmetric amplitude d k ,include:

[0105] Construct the Vandermonde matrix U:

[0106]

[0107] The symmetrical amplitude c in the target pulse data is obtained by calculating the amplitude sequence through the following formula k , asymmetric amplitude d k .

[0108] V=(U T U) -1 U T G;

[0109]

[0110] The specific step two is:

[0111] Construct a pulse parameter generation model: inference network, transient space, generation network, hot encoder.

[0112] The inference network includes a first input layer, a first feedforward neuron layer connected to the first input layer, a first intermediate neuron layer connected to the first feedforward neuron layer, a first post neuron layer connected to the first neuron layer, and a first output layer connected to the first post neuron layer;

[0113] The generating network includes a second input layer, a second feedforward neuron layer connected to the second input layer, a second intermediate neuron layer connected to the second feedforward neuron layer, a second post neuron layer connected to the second neuron layer, and a second output layer connected to the second post neuron layer;

[0114] The transient space is connected to the first output layer and the second output layer respectively, and the output layer of the thermal encoder is connected to the first intermediate neuron layer, the first post neuron layer, the second intermediate neuron layer and the second post neuron layer respectively.

[0115] In the embodiment of the present invention, the prior probability distribution of the transient feature vector Z is set to follow the Gaussian distribution, that is, the inference network learns the distribution parameters of the transient feature vector Z: mean μ and standard deviation σ. The inference network can be expressed as Q(Z|X), that is, the posterior distribution of the transient feature vector Z represented by the known input sample X.

[0116] μ=E μ (X);

[0117] σ=E σ (X);

[0118] Z = [μ, σ];

[0119] When the generative network obtains the prior probability distribution P(Z) of the transient feature vector Z, it restores the reconstructed sample of the input sample X. Then the process of generating the network can be expressed as:

[0120]

[0121] is the prior probability distribution of the output, P(X|Z) is the posterior probability distribution of the output target pulse data represented by the known input transient eigenvector Z, and P(Z) is the prior probability distribution of the transient eigenvector Z.

[0122] The training objective function of the initial pulse parameter generation model is obtained as follows:

[0123] J CVAE =E Z～Q [ln P(X|Z,Y)]-D KL [Q(Z|X,Y)||P(Z)];

[0124]

[0125] Among them, E Z～Q [] indicates sampling the transient feature vector Z according to distribution Q and calculating the expected value under the sampling, X indicates the sample pulse data, Z indicates the transient feature vector, Y indicates the hot encoding coefficient of the sample label, P(Z) is the prior probability distribution of the transient feature vector Z, P(X|Z) indicates the posterior probability distribution of the output target pulse data when the transient feature vector Z is input, D KL Represents the KL divergence formula.

[0126] The specific step five is: once all target pulse data are estimated, they can be substituted into the model The expression is used to reconstruct the target transformer fault data. A random undersampling algorithm is used to reduce the data in the normal samples, and finally a new data set is obtained by merging the samples that generate faults and the corrected normal samples. The random undersampling method arranges the normal samples according to the overall trend according to the parameter type, randomly reduces the values ​​of the corresponding samples, and finally recombines them to generate new normal samples.

[0127] The technical solution of the embodiment of the present invention models the sample transformer fault data as a spatial trend curve, defines it as a signal processing feature extraction and reconstruction problem to quickly generate target transformer fault data, achieves target transformer fault data generation results with higher computational efficiency by finding the optimal and simplest feature representation, and balances the overall transformer fault data samples to improve the disadvantage of uneven data set.

[0128] Embodiment 3

[0129] Figure 3 This is a schematic diagram of the structure of a system for determining a transformer fault data set provided by Embodiment 3 of the present invention. Figure 3As shown, the system includes: a sample pulse data determination module 310 , a data reconstruction module 320 and a data set determination module 330 .

[0130] Among them, the sample pulse data determination module 310 is used to obtain sample transformer fault data, model the sample transformer fault data as a dynamic pulse width signal model based on the annihilation filtering algorithm, and determine the sample pulse data based on the model parameters of the dynamic pulse width signal model, wherein the sample pulse data includes symmetrical amplitude data, pulse width data, asymmetrical amplitude data and delay data; the data reconstruction module 320 is used to generate target pulse data according to the sample pulse data and a pre-trained pulse parameter generation model, and reconstruct target transformer fault data based on the target pulse data; the data set determination module 330 is used to determine the transformer fault data set based on the sample transformer fault data and the target transformer fault data.

[0131] The technical solution of the embodiment of the present invention is to obtain sample transformer fault data through a sample pulse data determination module, model the sample transformer fault data as a dynamic pulse width signal model based on the annihilation filtering algorithm, and determine the sample pulse data based on the model parameters of the dynamic pulse width signal model, wherein the sample pulse data includes symmetrical amplitude data, pulse width data, asymmetrical amplitude data and delay data; model the sample transformer fault data as a dynamic pulse width signal model of a spatial trend curve, so as to perform signal processing, feature extraction and reconstruction. Through a data reconstruction module, target pulse data is generated according to the sample pulse data and a pre-trained pulse parameter generation model, and target transformer fault data is reconstructed based on the target pulse data and the dynamic pulse width signal model; new target transformer fault data is quickly generated through sample pulse data and the pulse parameter generation model. Through a data set determination module, a transformer fault data set is determined based on the sample transformer fault data and the target transformer fault data. A transformer fault set is quickly generated based on the sample transformer fault data and the target transformer fault data, which solves the problem of insufficient transformer fault data and achieves the beneficial effects of improving the determination efficiency of the transformer fault data set and improving the expansion efficiency of the transformer fault data.

[0132] Optionally, the sample pulse data determination module is specifically used to:

[0133] The pulse width signal model is expressed by the following formula:

[0134]

[0135] Where N is the length of the parameter vector space, and the basis function f k (n) is the double pulse waveform, σ(n) is the model mismatch error, and K is the number of pulses in the double pulse waveform.

[0136] Optionally, the sample pulse data determination module is specifically used to:

[0137] The double pulse waveform is expressed by the following formula:

[0138]

[0139] in, Represents the symmetrical curve in the double pulse waveform, Represents the asymmetric curve in the double pulse waveform, c k Represents symmetrical amplitude data, r k Indicates pulse width data, d k Represents asymmetric amplitude data, n k Indicates delay data.

[0140] Optionally, the pulse parameter generation model includes an inference network, a transient space, a generation network and a thermal encoder, the transient space is respectively connected to the inference network and the generation network, the thermal encoder is respectively connected to the inference network and the generation network, and the input of the inference network is the target pulse data.

[0141] Optionally, the inference network includes a first input layer, a first feedforward neuron layer connected to the first input layer, a first intermediate neuron layer connected to the first feedforward neuron layer, a first post neuron layer connected to the first neuron layer, and a first output layer connected to the first post neuron layer;

[0142] The generating network includes a second input layer, a second feedforward neuron layer connected to the second input layer, a second intermediate neuron layer connected to the second feedforward neuron layer, a second post neuron layer connected to the second neuron layer, and a second output layer connected to the second post neuron layer;

[0143] The transient space is connected to the first output layer and the second output layer respectively, and the output layer of the thermal encoder is connected to the first intermediate neuron layer, the first post neuron layer, the second intermediate neuron layer and the second post neuron layer respectively.

[0144] Optionally, the data reconstruction module includes:

[0145] A feature vector acquisition unit, used for inputting the sample pulse data into the inference network to obtain a transient feature vector converted based on the sample pulse data, and sending the transient feature vector to the generation network through a transient space;

[0146] An encoding unit, configured to input a sample label corresponding to the sample pulse data into the hot encoder to obtain a hot encoding coefficient corresponding to the sample label encoding, and send the hot encoding coefficient to the generating network;

[0147] A data generating unit is used for the generating network to generate target pulse data based on the received transient feature vector and the hot encoding coefficient.

[0148] Optionally, the system also includes a training module and a model determination module.

[0149] The training module is used to train the constructed initial pulse parameter generation model by a predefined training objective function before generating the target pulse data according to the sample pulse data and the pre-trained pulse parameter generation model. During the training process, the loss function is calculated by a back propagation algorithm, and the model parameters are updated by an optimizer until a preset number of training times is reached;

[0150] The model determination module is used to evaluate the trained model after training to obtain a model evaluation index, and determine a target pulse parameter generation model based on at least one of the model evaluation indicators.

[0151] Optionally, the training objective function is expressed as follows:

[0152] J CVAE =E Z～Q [ln P(X|Z,Y)]-D KL [Q(Z|X,Y)||P(Z)];

[0153]

[0154] Among them, E Z～Q [] indicates sampling the transient feature vector Z according to distribution Q and calculating the expected value under the sampling, X indicates the sample pulse data, Z indicates the transient feature vector, Y indicates the hot encoding coefficient of the sample label, P(Z) is the prior probability distribution of the transient feature vector Z, P(X|Z) indicates the posterior probability distribution of the output target pulse data when the transient feature vector Z is input, D KL Represents the KL divergence formula.

[0155] Optionally, the data set determination module is specifically used to:

[0156] The sample transformer fault data is reduced based on a random downsampling method to obtain processed sample transformer fault data, and a transformer fault data set is constructed based on the processed sample transformer fault data and the target transformer fault data.

[0157] The system for determining a transformer fault data set provided in the embodiment of the present invention can execute the method for determining a transformer fault data set provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0158] Embodiment 4

[0159] Figure 4 A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0160] like Figure 4 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0161] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0162] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The processor 11 executes the various methods and processes described above, such as the method for determining a transformer fault data set.

[0163] In some embodiments, the method transformer fault data set determination may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the method transformer fault data set determination described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to perform the method transformer fault data set determination in any other suitable manner (e.g., by means of firmware).

[0164] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0165] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0166] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0167] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0168] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0169] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0170] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0171] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for determining a transformer fault data set, characterized in that: include: Acquire sample transformer fault data, model the sample transformer fault data as a dynamic pulse width signal model based on an annihilation filtering algorithm, and determine sample pulse data based on model parameters of the dynamic pulse width signal model, wherein the sample pulse data includes symmetrical amplitude data, pulse width data, asymmetrical amplitude data, and time delay data; Generate target pulse data according to the sample pulse data and a pre-trained pulse parameter generation model, and reconstruct target transformer fault data based on the target pulse data and the dynamic pulse width signal model; A transformer fault data set is determined based on the sample transformer fault data and the target transformer fault data.

2. The method according to claim 1, characterized in that The pulse width signal model is expressed by the following formula: Where N is the length of the parameter vector space, and the basis function f k (n) is the double pulse waveform, σ(n) is the model mismatch error, and K is the number of pulses in the double pulse waveform.

3. The method according to claim 2, characterized in that The double pulse waveform is expressed by the following formula: in, Represents the symmetrical curve in the double pulse waveform, Represents the asymmetric curve in the double pulse waveform, c k Represents symmetrical amplitude data, r k Indicates pulse width data, d k Represents asymmetric amplitude data, n k Indicates delay data.

4. The method according to claim 1, characterized in that The pulse parameter generation model includes an inference network, a transient space, a generation network and a thermal encoder, the transient space is respectively connected to the inference network and the generation network, the thermal encoder is respectively connected to the inference network and the generation network, and the input of the inference network is the target pulse data.

5. The method according to claim 4, characterized in that The inference network includes a first input layer, a first feedforward neuron layer connected to the first input layer, a first intermediate neuron layer connected to the first feedforward neuron layer, a first post neuron layer connected to the first neuron layer, and a first output layer connected to the first post neuron layer; The generating network includes a second input layer, a second feedforward neuron layer connected to the second input layer, a second intermediate neuron layer connected to the second feedforward neuron layer, a second post neuron layer connected to the second neuron layer, and a second output layer connected to the second post neuron layer; The transient space is connected to the first output layer and the second output layer respectively, and the output layer of the thermal encoder is connected to the first intermediate neuron layer, the first post neuron layer, the second intermediate neuron layer and the second post neuron layer respectively.

6. The method according to claim 5, characterized in that The generating target pulse data according to the sample pulse data and a pre-trained pulse parameter generation model comprises: Inputting the sample pulse data into the inference network to obtain a transient feature vector converted based on the sample pulse data, and sending the transient feature vector to the generation network through a transient space; Inputting a sample label corresponding to the sample pulse data into the hot encoder to obtain a hot encoding coefficient corresponding to the sample label encoding, and sending the hot encoding coefficient to the generating network; The generation network generates target pulse data based on the received transient feature vector and the hot encoding coefficient.

7. The method according to claim 1, characterized in that Before generating target pulse data according to the sample pulse data and the pre-trained pulse parameter generation model, the method further includes: The constructed initial pulse parameter generation model is trained through a predefined training objective function. During the training process, the loss function is calculated through a back-propagation algorithm, and the model parameters are updated through an optimizer until a preset number of training times is reached; After the training is completed, the trained model is evaluated to obtain a model evaluation index, and a target pulse parameter generation model is determined based on at least one of the model evaluation indicators.

8. The method according to claim 7, characterized in that The training objective function is expressed as follows: J CVAE =E Z～Q [ln P(X|Z,Y)]-D KL [Q(Z|X,Y)||P(Z)]; Among them, E Z～Q [] indicates sampling the transient feature vector Z according to distribution Q and calculating the expected value under the sampling, X indicates the sample pulse data, Z indicates the transient feature vector, Y indicates the hot encoding coefficient of the sample label, P(Z) is the prior probability distribution of the transient feature vector Z, P(X|Z) indicates the posterior probability distribution of the output target pulse data when the transient feature vector Z is input, D KL Represents the KL divergence formula.

9. The method according to claim 1, characterized in that: The determining of the transformer fault data set based on the sample transformer fault data and the target transformer fault data comprises: The sample transformer fault data is reduced based on a random downsampling method to obtain processed sample transformer fault data, and a transformer fault data set is constructed based on the processed sample transformer fault data and the target transformer fault data.

10. A system for determining a transformer fault data set, characterized in that: include: A sample pulse data determination module is used to obtain sample transformer fault data, model the sample transformer fault data as a dynamic pulse width signal model based on an annihilation filtering algorithm, and determine the sample pulse data based on model parameters of the dynamic pulse width signal model, wherein the sample pulse data includes symmetrical amplitude data, pulse width data, asymmetrical amplitude data and delay data; A data reconstruction module is used to generate target pulse data according to the sample pulse data and a pre-trained pulse parameter generation model, and reconstruct target transformer fault data based on the target pulse data; a data set determination module is used to determine a transformer fault data set based on the sample transformer fault data and the target transformer fault data.

Images