Transformer turn-to-turn short circuit fault identification method, model training method, medium and equipment
By using multi-layer autoencoder and backpropagation optimization methods, combined with simulation and actual data, the problem of insufficient short-circuit fault samples between turns of the transformer is solved, and efficient and accurate fault identification is achieved.
Patent Information
- Application Number
- CN202411074619.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-07
- Publication Date
- 2024-12-03
AI Technical Summary
The limited number of samples for transformer interturn short circuit faults is caused by insufficient samples when training artificial intelligence models, affecting the accuracy of detection and diagnosis.
Using a method, by obtaining unlabeled simulation data and actual sampled data of labels, the initial model is constructed for unsupervised training, using a multi-layer autoencoder to learn feature representation, and fine-tuning the model parameters through backpropagation, and finally, in The output layer adds a classifier to identify inter-turn short circuit faults.
This method reduces dependence on a large number of tag data, maintains good performance in the identification of inter-turn short-circuit faults, and can efficiently and accurately identify different inter-turn short-circuit faults of the transformer.
Smart Images

Figure CN119066511A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of transformer faults, and in particular to a method for identifying a transformer turn-to-turn short-circuit fault, a model training method, a medium and a device. Background Art
[0002] Among all kinds of transformer faults, the occurrence probability of inter-turn short-circuit fault is the highest, accounting for 60% to 70% of the total faults. Therefore, it is very important to detect and prevent inter-turn short-circuit faults. Systematic research on transformer inter-turn short-circuit faults based on artificial intelligence technology can not only improve the accuracy and efficiency of detection, but also provide new methods and tools for fault prediction and prevention, which has important practical significance.
[0003] However, in actual work, researchers often face the problem of insufficient samples. Specifically, the number of transformer samples with turn-to-turn short-circuit faults is limited, which makes it difficult to train and verify artificial intelligence models. Insufficient samples will cause the model to be unable to fully learn the fault characteristics, which in turn affects the accuracy of detection and diagnosis. Summary of the invention
[0004] Based on this, it is necessary to provide a transformer inter-turn short-circuit fault identification method, a model training method, a medium and equipment to solve the problem of insufficient training samples.
[0005] A training method for a transformer turn-to-turn short-circuit fault identification model, the method comprising:
[0006] Obtain first sample data and second sample data; wherein the first sample data is simulated data of the transformer under different inter-turn short-circuit faults, and the first sample data is not labeled, and the second sample data is actual sampling data of the transformer under different inter-turn short-circuit faults, and the second sample data is labeled;
[0007] Acquire the constructed initial model; wherein the initial model includes an input layer, a multi-layer autoencoder connected in sequence, and an output layer;
[0008] Inputting the first sample data into the input layer of the initial model, and performing unsupervised training on the multi-layer autoencoder layer by layer based on the constructed first comprehensive loss function to obtain a first training model; wherein the first comprehensive loss function is used to indicate the difference between the first sample data and the reconstructed output;
[0009] Inputting the second sample data into the first training model, and performing back propagation optimization on the first training model based on the constructed second comprehensive error function to obtain a second training model; wherein the second comprehensive error function is used to indicate the error between the predicted output and the label;
[0010] A classifier is added to the output layer of the second training model to obtain a turn-to-turn short circuit fault identification model.
[0011] In one embodiment, the method further comprises:
[0012] Constructing a first mean square error term, a weight penalty term and a sparsity penalty term; wherein the mean square error term is used to indicate the difference between the first sample data and the reconstructed data, the weight penalty term is used to control the weights of different neurons, and the sparsity penalty term is used to reduce the proportion of information copying from the input layer to the hidden layer;
[0013] The mean square error term, the weight penalty term and the sparsity penalty term are respectively assigned preset proportional coefficients and summed up to obtain the first comprehensive loss function.
[0014] In one embodiment, the first comprehensive loss function is expressed as:
[0015] J DSAENN =J MSE +λ1J w +λ2J s
[0016]
[0017] In the above formula, J MSE is the mean square error term, J w is the weight penalty term, J S is the sparsity penalty term; λ1 is the proportional coefficient of the weight penalty term, λ2 is the proportional coefficient of the sparsity penalty term; m is the number of samples of the first sample data; X is the first sample data, is the reconstructed output; n l is the number of nodes in layer l, is the weight between the i-th neuron in the l-th layer and the j-th neuron in the l-1-th layer; KL(ρ||ρ j ) is ρ and ρ j The distribution difference between them; J is the total number of hidden layer neurons; ρ is the expected sparsity parameter; ρ j is the sparse parameter of the jth neuron.
[0018] In one embodiment, the method further comprises:
[0019] Construct a second mean square error term, a similar error term, and a non-similar error term; wherein the second mean square error term is used to indicate the error between the predicted output and the label, the similar error term is used to indicate the cumulative deviation of the predicted values between the same type of data samples, and the non-similar error term is used to indicate the cumulative deviation of the predicted values between the different types of data samples;
[0020] The second mean square error term, the same error term and the non-same error term are respectively assigned preset proportional coefficients, and the same error term is made positively correlated with the second mean square error term, and the non-same error term is made negatively correlated with the second mean square error term, so as to obtain the second comprehensive loss function.
[0021] In one embodiment, the second comprehensive loss function is expressed as:
[0022] J x (ω,b)=J(ω,b)+μ1J1(ω,b)-μ2J2(ω,b)
[0023]
[0024] In the above formula, J(ω,b) is the second mean square error term; J1(ω,b) is the same error term; J2(ω,b) is the different error term; μ1 is the proportional coefficient of the same error term, μ2 is the proportional coefficient of the different error term; n is the sample type size; m is the number of samples of the second sample data; h ω,b (x (i) ) is the second sample data x (i) The predicted output, y (i) The second sample data x (i) Label; m j is the sample size of the jth type; h ω,b (x (i,j) ) is the sample prediction value of the jth class; M (j) is the sample average of the jth type; M (i) is the sample mean of the i-th type.
[0025] In one embodiment, the formula for back propagation optimization is expressed as:
[0026]
[0027] In the above formula, in the above formula, represents the updated weight between the i-th neuron in the l-th layer and the j-th neuron in the l-1-th layer, represents the weight between the i-th neuron in the l-th layer and the j-th neuron in the l-1-th layer before updating, η represents the learning rate, yes The gradient vector of represents the updated bias of the jth neuron in the lth layer, represents the bias of the jth neuron in the lth layer before updating, yes The gradient vector of .
[0028] In one embodiment, the added classifier is a Softmax classifier, and the output of the Softmax classifier is expressed as:
[0029]
[0030] In the above formula, p(y i =1)|x i;θ To judge the input sample x i is the probability of classification i; θ1, θ2, …, θ k ∈R n+1 are model parameters; is the normalization function.
[0031] A method for identifying a turn-to-turn short-circuit fault of a transformer, the method comprising:
[0032] Acquire monitoring data of the transformer to be identified, and input the monitoring data into a turn-to-turn short-circuit fault identification model to obtain an output turn-to-turn short-circuit fault identification result; wherein the turn-to-turn short-circuit fault identification model is trained by the training method of the turn-to-turn short-circuit fault identification model of the above-mentioned transformer.
[0033] A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the processor executes the steps of a transformer inter-turn short-circuit fault identification method and a model training method.
[0034] A terminal device includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above-mentioned transformer inter-turn short-circuit fault identification method and model training method.
[0035] The present invention provides a method for identifying a short-circuit fault between turns of a transformer, a model training method, a medium and a device. Two types of sample data are first obtained, the first type is unlabeled simulation data of different short-circuit faults between turns of the transformer, and the second type is labeled actual sampling data. Then, an initial model is constructed, and the unlabeled simulation data is input into the initial model. The multi-layer autoencoder is trained layer by layer using the first comprehensive loss function in an unsupervised training manner so that the model can effectively learn feature representation. Then, the actual sampled data with labels is input into the trained first training model, and the back propagation optimization is performed using the second comprehensive error function. The goal of this stage is to fine-tune the parameters of the model to improve the accuracy of the error between the predicted output and the label. Finally, a classifier is added to the output layer of the second training model to convert the model output into the probability distribution of the short-circuit fault between turns, thereby obtaining the final short-circuit fault identification model between turns. In this embodiment, since the effective feature representation has been learned in the early unsupervised pre-training stage, the subsequent reverse fine-tuning stage only needs to fine-tune the model parameters on a small amount of labeled actual sampling data. This process greatly reduces the dependence on large amounts of labeled data while maintaining the model's good performance in identifying turn-to-turn short-circuit faults, enabling the model to efficiently and accurately identify different turn-to-turn short-circuit faults of the transformer. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. 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 paying creative work.
[0037] in:
[0038] Figure 1 A flowchart of a training method for a transformer inter-turn short-circuit fault identification model;
[0039] Figure 2 is a schematic diagram of the transformer primary and secondary coil circuit model;
[0040] Figure 3a It is a schematic diagram of the transformer in normal operation;
[0041] Figure 3b This is a schematic diagram of a ground fault occurring in a single winding of a transformer;
[0042] Figure 3c This is a schematic diagram when two ground faults occur between different windings of the transformer;
[0043] Figure 3d This is a schematic diagram of a transformer with two ground faults in a single winding;
[0044] Figure 4 Schematic diagram of the model training process;
[0045] Figure 5 This is a structural block diagram of the terminal equipment. DETAILED DESCRIPTION
[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions 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 are within the scope of protection of the present invention.
[0047] The terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. 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 limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices.
[0048] 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.
[0049] like Figure 1 As shown, Figure 1 The flowchart of the training method of the turn-to-turn short-circuit fault identification model of the transformer in one embodiment is shown in FIG. The training method of the turn-to-turn short-circuit fault identification model of the transformer in this embodiment includes the following steps:
[0050] S101, obtaining first sample data and second sample data.
[0051] Among them, the first sample data is the simulation data of the transformer under different inter-turn short-circuit faults, and the first sample data is not labeled. The second sample data is the actual sampling data of the transformer under different inter-turn short-circuit faults, and the second sample data is labeled.
[0052] Specifically, we can first build a system based on MATLAB / Simu link. Figure 2The transformer primary and secondary coil circuit models shown are used to simulate different inter-turn short-circuit faults of the transformer.
[0053] like Figure 3a As shown, the primary and secondary winding modules form a simple transformer, where A, B, C represent the high voltage side of the corresponding phase; a, b, c represent the low voltage side of the corresponding phase. The transformer is currently in normal operation.
[0054] like Figure 3b As shown in the figure, when a single winding has a ground fault, the short-circuit point divides the winding into two parts, and the simulation data under the Ag fault can be simulated. Similarly, the simulation data under the Bg and Cg faults can also be simulated. At the same time, because the insulation strength between the low-voltage winding and the ground is not enough to break down the system, this embodiment only simulates the above three high-voltage winding to ground faults.
[0055] When there are two short-circuit points in the fault, such as Figure 3c The figure shows one of the scenarios of double winding faults. In this scenario, the simulation data of six faults, Aa, Bb, Cc, AB, BC and CA, can be simulated. Figure 3d The figure shows one of the scenarios of single winding fault. In this scenario, simulation data of six other faults including aa, bb, cc, AA, BB and CC can be simulated.
[0056] At the same time, considering that the low-voltage winding is generally covered by the high-voltage winding, the interphase faults between different sides of the transformer, such as Ab, are not considered in this embodiment. In this embodiment, we can set the above 15 types of faults for simulation. By adjusting different parameters of the system, we can record the voltage and current data under these fault conditions as the first sample data.
[0057] Correspondingly, the second sample data is the actual sampled data of the transformer under different inter-turn short-circuit faults. These data are collected during the actual operation of the transformer, and each data sample is marked with a corresponding fault label.
[0058] S102, obtaining the constructed initial model.
[0059] Among them, Figure 4 As shown, the initial model includes an input layer, a multi-layer autoencoder (corresponding to a hidden layer) connected in sequence, and an output layer.
[0060] An autoencoder is a neural network structure whose goal is to compress the first sample data into a low-dimensional representation (encode) and then reconstruct it back to the original data. The specific process is as follows: Encoder part: The encoder part of each autoencoder converts the first sample data into a low-dimensional representation. Decoder part: The decoder part of each autoencoder reconstructs the low-dimensional representation back to the original data.
[0061] The multi-layer autoencoder stacks multiple autoencoder layers in sequence to form a deep network structure, and the output of each autoencoder layer is used as the input of the next autoencoder layer. These autoencoder layers learn the feature representation of the first sample data through layer-by-layer training, so that the network can extract and capture important information in the data.
[0062] S103, inputting the first sample data into the input layer of the initial model, and performing unsupervised training on the multi-layer autoencoder layer by layer based on the constructed first comprehensive loss function to obtain a first training model.
[0063] The first comprehensive loss function is used to indicate the difference between the first sample data and the reconstructed output. The goal of the first comprehensive loss function is to minimize the error between the first sample data and the reconstructed output, so that the autoencoder can effectively learn and capture the features of the first sample data.
[0064] In a specific embodiment, the first comprehensive loss function is constructed as follows:
[0065] (1) Construct the first mean square error term, weight penalty term and sparsity penalty term.
[0066] Among them, the mean square error term is used to indicate the difference between the first sample data and the reconstructed data, and the weight penalty term is used to control the weights of different neurons, that is, to add a limit to the weights of the model to prevent them from becoming too large, thereby controlling the complexity of the model. The sparsity penalty term is used to reduce the proportion of information copying from the input layer to the hidden layer, thereby forcing the model to learn more features instead of simply copying the first sample data.
[0067] (2) Assign preset proportional coefficients to the mean square error term, the weight penalty term, and the sparsity penalty term respectively and sum them up to obtain the first comprehensive loss function.
[0068] Optionally, the first comprehensive loss function is expressed as:
[0069] J DSAENN =J MSE +λ1J w +λ2J s
[0070]
[0071]
[0072] In the above formula, J MSE is the mean square error term, J w is the weight penalty term, J Sis the sparsity penalty term; λ1 is the proportional coefficient of the weight penalty term, λ2 is the proportional coefficient of the sparsity penalty term; m is the number of samples of the first sample data; X is the first sample data, is the reconstructed output; n1 is the number of nodes in layer l, is the weight between the i-th neuron in the l-th layer and the j-th neuron in the l-1-th layer; KL(ρ||ρ j ) is ρ and ρ j The distribution difference between them; J is the total number of hidden layer neurons; ρ is the expected sparsity parameter; ρ j is the sparse parameter of the jth neuron.
[0073] The layer-by-layer training method in this step means that when training multiple layers of autoencoders, only one autoencoder layer is trained at a time. When training each layer of autoencoder, the output of the current layer is used as the input of the next layer, and training is performed in sequence until all layers are trained.
[0074] The goal of training in this step is to make the reconstructed output as close as possible to the first sample data, thereby achieving effective extraction of the features of the first sample data and ultimately forming a first training model.
[0075] S104, inputting the second sample data into the first training model, and performing back propagation optimization on the first training model based on the constructed second comprehensive error function to obtain a second training model.
[0076] The second comprehensive error function is used to indicate the error between the predicted output and the label. The goal of the second comprehensive error function is to minimize the error between the predicted output and the actual label, thereby improving the classification accuracy of the model.
[0077] In a specific embodiment, the second comprehensive loss function is constructed as follows:
[0078] (1) Construct the second mean square error term, similar error term and non-similar error term.
[0079] Among them, the second mean square error term is used to indicate the error between the predicted output and the label, and the homogeneous error term is used to indicate the cumulative deviation of the predicted values between the same type of data samples, that is, to measure the consistency of the predicted values of the data samples of the same category. Ideally, the predicted values of the homogeneous data samples should be as close as possible, and the deviation should be as small as possible. The non-homogeneous error term is used to indicate the cumulative deviation of the predicted values between the different types of data samples. That is, to measure the separation of the predicted values of the data samples of different categories. Ideally, the predicted values of the data samples of different categories should be significantly different, and the deviation should be as large as possible.
[0080] (2) Assigning preset proportional coefficients to the second mean square error term, the same error term and the different error term respectively, and making the same error term positively correlated with the second mean square error term, and making the different error term negatively correlated with the second mean square error term, so as to obtain a second comprehensive loss function.
[0081] Optionally, the second comprehensive loss function is expressed as:
[0082] J x (ω,b)=J(ω,b)+μ1J1(ω,b)-μ2J2(ω,b)
[0083]
[0084] In the above formula, J(ω,b) is the second mean square error term; J1(ω,b) is the same error term; J2(ω,b) is the different error term; μ1 is the proportional coefficient of the same error term, μ2 is the proportional coefficient of the different error term; n is the sample type size; m is the number of samples of the second sample data; h ω,b (x (i) ) is the second sample data x (i) The predicted output, y (i) The second sample data x (i) Label; m j is the sample size of the jth type; h ω,b (x (i,j) ) is the sample prediction value of the jth class; M (j) is the sample average of the jth type; M (i) is the sample mean of the i-th type.
[0085] like Figure 4 As shown, this embodiment then calculates the gradient of the loss function and adjusts the parameters of the model through back propagation to minimize the loss. During the back propagation process, the error is propagated layer by layer from the output layer to the input layer, and the weights and biases of each layer are updated.
[0086] Alternatively, the back-propagation optimization formula is expressed as:
[0087]
[0088] In the above formula, in the above formula, represents the updated weight between the i-th neuron in the l-th layer and the j-th neuron in the l-1-th layer, represents the weight between the i-th neuron in the l-th layer and the j-th neuron in the l-1-th layer before updating, η represents the learning rate, yes The gradient vector of represents the updated bias of the jth neuron in the lth layer, represents the bias of the jth neuron in the lth layer before updating, yes The gradient vector of .
[0089] S105, adding a classifier to the output layer of the second training model to obtain a turn-to-turn short circuit fault identification model.
[0090] In a specific embodiment, Figure 4 As shown, the added classifier is a Softmax classifier, which can convert the input vector into an output vector representing the probability of each category, and then determine which type of inter-turn short-circuit fault has the highest probability.
[0091] Optionally, the output of the Softmax classifier is expressed as:
[0092]
[0093] In the above formula, p(y i =1)|x i;θ To judge the input sample x i is the probability of classification i; θ1, θ2, …, θ k ∈R n+1 are model parameters; is the normalization function.
[0094] The training method of the above-mentioned transformer inter-turn short-circuit fault identification model first obtains two types of sample data, the first type is unlabeled simulation data of different inter-turn short-circuit faults of the transformer, and the second type is labeled actual sampling data. Then construct the initial model, input the unlabeled simulation data into the initial model, and train the multi-layer autoencoder layer by layer using the first comprehensive loss function in an unsupervised training manner so that the model can effectively learn feature representation. Then input the labeled actual sampling data into the trained first training model, and use the second comprehensive error function for back propagation optimization. The goal of this stage is to fine-tune the parameters of the model to improve the accuracy of the error between the predicted output and the label. Finally, a classifier is added to the output layer of the second training model to convert the model output into the probability distribution of the inter-turn short-circuit fault, thereby obtaining the final inter-turn short-circuit fault identification model. In this embodiment, since the effective feature representation has been learned in the early unsupervised pre-training stage, the subsequent reverse fine-tuning stage only needs to fine-tune the model parameters on a small amount of labeled actual sampling data. This process greatly reduces the dependence on large amounts of labeled data while maintaining the model's good performance in identifying turn-to-turn short-circuit faults, enabling the model to efficiently and accurately identify different turn-to-turn short-circuit faults of the transformer.
[0095] In one embodiment, the present invention further provides a method for identifying a turn-to-turn short circuit fault of a transformer. The method for identifying a turn-to-turn short circuit fault of a transformer in this embodiment includes the following steps:
[0096] Acquire monitoring data of the transformer to be identified, and input the monitoring data into a turn-to-turn short-circuit fault identification model to obtain an output turn-to-turn short-circuit fault identification result; wherein the turn-to-turn short-circuit fault identification model is trained by the above-mentioned model training method, and the turn-to-turn short-circuit fault identification model can effectively infer whether the transformer has a turn-to-turn short-circuit fault from the input monitoring data.
[0097] Figure 5 FIG. 2 shows an internal structure diagram of a terminal device in an embodiment. Figure 5 As shown, the terminal device includes a processor, a memory and a network interface connected via a system bus. Among them, the memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the terminal device stores an operating system and may also store a computer program. When the computer program is executed by the processor, the processor can implement a method for identifying a short-circuit fault between turns of a transformer. The internal memory may also store a computer program. When the computer program is executed by the processor, the processor can implement a method for identifying a short-circuit fault between turns of a transformer. Those skilled in the art can understand that Figure 5 The structure shown in the figure is only a block diagram of a partial structure related to the scheme of the present application, and does not constitute a limitation on the terminal device to which the scheme of the present application is applied. The specific terminal device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0098] A computer-readable storage medium stores a computer program, which implements the following steps when executed by a processor: obtaining first sample data and second sample data; wherein the first sample data is simulated data of a transformer under different turn-to-turn short-circuit faults, the first sample data is not labeled, and the second sample data is actual sampling data of the transformer under different turn-to-turn short-circuit faults, the second sample data is labeled; obtaining a constructed initial model; wherein the initial model includes an input layer, a multi-layer autoencoder connected in sequence, and an output layer; inputting the first sample data into the input layer of the initial model, and performing unsupervised training on the multi-layer autoencoder layer by layer based on a constructed first comprehensive loss function to obtain a first training model; wherein the first comprehensive loss function is used to indicate the difference between the first sample data and the reconstructed output; inputting the second sample data into the first training model, and performing back-propagation optimization on the first training model based on a constructed second comprehensive error function to obtain a second training model; wherein the second comprehensive error function is used to indicate the error between the predicted output and the label; adding a classifier to the output layer of the second training model to obtain a turn-to-turn short-circuit fault recognition model.
[0099] That is, the following steps are implemented: obtaining monitoring data of the transformer to be identified, and inputting the monitoring data into the turn-to-turn short-circuit fault identification model to obtain the output turn-to-turn short-circuit fault identification result.
[0100] A terminal device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program: obtaining first sample data and second sample data; wherein the first sample data is simulated data of a transformer under different inter-turn short-circuit faults, the first sample data is not labeled, and the second sample data is actual sampling data of the transformer under different inter-turn short-circuit faults, the second sample data is labeled; obtaining a constructed initial model; wherein the initial model comprises an input layer, a multi-layer autoencoder connected in sequence and an output layer; inputting the first sample data into the input layer of the initial model, and performing unsupervised training on the multi-layer autoencoder layer by layer based on the constructed first comprehensive loss function to obtain a first training model; wherein the first comprehensive loss function is used to indicate the difference between the first sample data and the reconstructed output; inputting the second sample data into the first training model, and performing back-propagation optimization on the first training model based on the constructed second comprehensive error function to obtain a second training model; wherein the second comprehensive error function is used to indicate the error between the predicted output and the label; adding a classifier to the output layer of the second training model to obtain an inter-turn short-circuit fault recognition model.
[0101] That is, the following steps are implemented: obtaining monitoring data of the transformer to be identified, and inputting the monitoring data into the turn-to-turn short-circuit fault identification model to obtain the output turn-to-turn short-circuit fault identification result.
[0102] It should be noted that the above-mentioned transformer turn-to-turn short-circuit fault identification method, model training method, medium and equipment belong to a general inventive concept, and the contents in the embodiments of the transformer turn-to-turn short-circuit fault identification method, model training method, medium and equipment are applicable to each other.
[0103] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing related hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct RAMbus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.
[0104] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0105] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.
Claims
1. A training method for a transformer turn-to-turn short-circuit fault identification model. It is characterized in that The method comprises: Obtain first sample data and second sample data; wherein the first sample data is simulated data of the transformer under different inter-turn short-circuit faults, and the first sample data is not labeled, and the second sample data is actual sampling data of the transformer under different inter-turn short-circuit faults, and the second sample data is labeled; Acquire the constructed initial model; wherein the initial model includes an input layer, a multi-layer autoencoder connected in sequence, and an output layer; Inputting the first sample data into the input layer of the initial model, and performing unsupervised training on the multi-layer autoencoder layer by layer based on the constructed first comprehensive loss function to obtain a first training model; wherein the first comprehensive loss function is used to indicate the difference between the first sample data and the reconstructed output; Inputting the second sample data into the first training model, and performing back propagation optimization on the first training model based on the constructed second comprehensive error function to obtain a second training model; wherein the second comprehensive error function is used to indicate the error between the predicted output and the label; A classifier is added to the output layer of the second training model to obtain a turn-to-turn short circuit fault identification model.
2. The method according to claim 1, It is characterized in that The method further comprises: Constructing a first mean square error term, a weight penalty term and a sparsity penalty term; wherein the mean square error term is used to indicate the difference between the first sample data and the reconstructed data, the weight penalty term is used to control the weights of different neurons, and the sparsity penalty term is used to reduce the proportion of information copying from the input layer to the hidden layer; The mean square error term, the weight penalty term and the sparsity penalty term are respectively assigned preset proportional coefficients and summed up to obtain the first comprehensive loss function.
3. The method according to claim 2, It is characterized in that The first comprehensive loss function is expressed as: I DSAENN =J MSE +λ 1 I w +λ 2 I s In the above formula, J MSE is the mean square error term, J w is the weight penalty term, J S is the sparsity penalty term; 1 is the proportional coefficient of the weight penalty term, λ 2 is the proportional coefficient of the sparsity penalty term; m is the number of samples of the first sample data; X is the first sample data, is the reconstructed output; n 1 is the number of nodes in layer l, is the weight between the i-th neuron in the l-th layer and the j-th neuron in the l-1-th layer; KL(ρ||ρ j ) is ρ and ρ j The distribution difference between them; J is the total number of hidden layer neurons; ρ is the expected sparsity parameter; ρ j is the sparse parameter of the jth neuron.
4. The method according to claim 1, It is characterized in that The method further comprises: Construct a second mean square error term, a similar error term, and a non-similar error term; wherein the second mean square error term is used to indicate the error between the predicted output and the label, the similar error term is used to indicate the cumulative deviation of the predicted values between the same type of data samples, and the non-similar error term is used to indicate the cumulative deviation of the predicted values between the different types of data samples; The second mean square error term, the same error term and the non-same error term are respectively assigned preset proportional coefficients, and the same error term is made positively correlated with the second mean square error term, and the non-same error term is made negatively correlated with the second mean square error term, so as to obtain the second comprehensive loss function.
5. The method according to claim 4, It is characterized in that The second comprehensive loss function is expressed as: J x (ω,b)=J(ω,b)+μ 1 J 1 (ω,b)-μ 2 J 2 (ω,b) In the above formula, J(ω,b) is the second mean square error term; J 1 (ω,b) is the same type of error term; J 2 (ω,b) is the non-similar error term; μ 1 is the proportionality coefficient of similar error terms, μ 2 is the proportional coefficient of non-similar error terms; n is the sample type size; m is the sample size of the second sample data; h ω,b (x (i) ) is the second sample data x (i) The predicted output, y (i) is the second sample data x (i) Label; m j is the sample size of the jth type; h ω,b (x (i,j) ) is the sample prediction value of the jth class; M (j) is the sample average of the jth type; M (i) is the sample mean of the i-th type.
6. The method according to claim 1, It is characterized in that The formula for back propagation optimization is expressed as: In the above formula, in the above formula, represents the updated weight between the i-th neuron in the l-th layer and the j-th neuron in the l-1-th layer, represents the weight between the i-th neuron in the l-th layer and the j-th neuron in the l-1-th layer before updating, η represents the learning rate, yes The gradient vector of represents the updated bias of the jth neuron in the lth layer, represents the bias of the jth neuron in the lth layer before updating, yes The gradient vector of .
7. The method according to claim 1, It is characterized in that The added classifier is a Softmax classifier, and the output of the Softmax classifier is expressed as: In the above formula, p(y i =1)|x i;θ To judge the input sample x i is the probability of classification i; θ 1 ,θ 2 ,…,θ k ∈R n+1 are model parameters; is the normalization function.
8. A method for identifying a transformer turn-to-turn short-circuit fault. It is characterized in that The method comprises: Acquire monitoring data of the transformer to be identified, and input the monitoring data into a turn-to-turn short-circuit fault identification model to obtain an output turn-to-turn short-circuit fault identification result; wherein the turn-to-turn short-circuit fault identification model is trained by the method described in any one of claims 1-7.
9. A computer-readable storage medium, It is characterized in that A computer program is stored, and when the computer program is executed by a processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 8.
10. A terminal device, It is characterized in that The method comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method according to any one of claims 1 to 8.
