Method and device for on-line identification of winding state of transformer, equipment and medium

Abstract

The embodiment of the application discloses a kind of online identification method and device, equipment and medium of the winding state of transformer, method includes: obtaining the current frequency response curve of transformer;According to current frequency response curve and preset normalized frequency response hybrid variance algorithm, determine current normalized frequency response hybrid variance;With current normalized frequency response hybrid variance and preset winding state identification model, obtain the current winding state of transformer, winding state identification model is based on the sample normalized frequency response hybrid variance under different winding states to the convolutional neural network model is obtained by pre-training.The above-mentioned mode can avoid the misjudgment and potential safety risk caused by monitoring sensor error, prolong the service life of transformer and other advantages, and has important significance for ensuring the safe, reliable and efficient operation of power system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power equipment fault diagnosis, and in particular to an online identification method and device for a winding state of a transformer, equipment and a medium. BACKGROUND

[0002] As a core device in the power system, the normal operation of the transformer is crucial to the safety of the power system. The transformer may be subjected to collision, impact during transportation or installation, and sudden short circuit during operation, which may cause winding deformation, thereby seriously threatening the safe operation of the power system. Winding deformation is one of the common faults of the transformer, which may cause permanent deformation phenomena such as local twisting, bulging or displacement, and these deformations will seriously affect the safe operation of the transformer and even cause serious power accidents.

[0003] Through online monitoring of the working state (winding and core state) of the transformer by the monitoring sensor, equipment faults can be found in time, which is of great significance to prolong the service life, operation reliability and economy of the transformer, and has important significance to ensure the safe, reliable and efficient operation of the power system. SUMMARY

[0004] The main purpose of the present application is to provide an online identification method and device for a winding state of a transformer, equipment and a medium, which can solve the problem of monitoring the working state of the transformer in the prior art.

[0005] To achieve the above-mentioned purpose, the present application provides an online identification method for a winding state of a transformer, which comprises:

[0006] obtaining a current frequency response curve of the transformer;

[0007] determining a current normalized frequency response hybrid variance according to the current frequency response curve and a preset normalized frequency response hybrid variance algorithm;

[0008] obtaining a current winding state of the transformer by using the current normalized frequency response hybrid variance and a preset winding state identification model, wherein the winding state identification model is obtained by pre-training a convolutional neural network model based on sample normalized frequency response hybrid variances under different winding states.

[0009] In a feasible implementation manner, the determination of the current normalized frequency response hybrid variance according to the current frequency response curve and the preset normalized frequency response hybrid variance algorithm comprises:

[0010] obtaining current frequency response hybrid variances of different frequency bands by using a reference frequency response curve of the transformer, two sets of current frequency response curves of the transformer and a preset frequency response hybrid variance algorithm;

[0011] The current normalized frequency response mixing variance of different frequency bands is obtained by using the current frequency response mixing variance and the normalized frequency response mixing variance algorithm.

[0012] In a possible implementation, the mathematical expression of the frequency response mixing variance algorithm is as follows:

[0013]

[0014] In the formula, S f is the frequency response mixing variance of the frequency band f, i is the frequency point number of the frequency band f, X is the total number of the selected frequency points, N1(i) is the amplitude of the frequency point i corresponding to the first group of current frequency response curves, N2(i) is the amplitude of the frequency point i corresponding to the second group of current frequency response curves, and M(i) is the amplitude of the frequency point i corresponding to the reference frequency response curve.

[0015] In a possible implementation, the mathematical expression of the normalized frequency response mixing variance algorithm is as follows:

[0016]

[0017] In the formula, S is the normalized frequency response mixing variance of the frequency band f, S f is the frequency response mixing variance of the frequency band f, i is the frequency point number of the frequency band f, N1(i) is the amplitude of the frequency point i corresponding to the first group of current frequency response curves, N2(i) is the amplitude of the frequency point i corresponding to the second group of current frequency response curves, M(i) is the amplitude of the frequency point i corresponding to the reference frequency response curve, min is the minimum value, and max is the maximum value.

[0018] In a possible implementation, the method further includes:

[0019] obtaining a sample data set, the sample data set including a plurality of sample normalized frequency response mixing variances labeled with real labels of winding states;

[0020] dividing the sample data set into a training set and a verification set by using cross validation;

[0021] performing model training of winding state prediction according to the training set and a preset convolutional neural network model, to obtain a trained convolutional neural network model;

[0022] performing model optimization of winding state prediction according to the verification set and the trained convolutional neural network model, to obtain a winding state recognition model.

[0023] In a possible implementation, the model training of winding state prediction according to the training set and the preset convolutional neural network model, to obtain the trained convolutional neural network model, includes:

[0024] training the convolutional neural network model by using the normalized frequency response hybrid variance of the samples in the training set, to obtain a predicted label of the winding state, wherein the convolutional neural network model comprises convolutional layers and fully connected layers connected in sequence, the number of the convolutional layers is set to three layers, the size of the convolution kernel is selected to be 1*1, the step is 1, and the activation function is selected to be ReLU, and batch normalization is applied after the convolutional layers or the fully connected layers;

[0025] determining whether the convolutional neural network model converges by using the predicted label, the real label of the normalized frequency response hybrid variance of the samples in the training set, and a preset mean square error loss function;

[0026] if the convolutional neural network model converges, obtaining a trained convolutional neural network model.

[0027] In a feasible implementation manner, the model optimization for predicting the fault type according to the verification set and the trained convolutional neural network model obtains a winding state recognition model, comprising:

[0028] adjusting the model learning rate by using a linear learning rate decay strategy, introducing an early stopping method, setting a patience value and a minimum change amount of the verification loss of the verification set, calculating the loss of the trained convolutional neural network model on the verification set after each round of training ends, saving the model parameters when the verification loss is the lowest, and obtaining the winding state recognition model.

[0029] To achieve the above object, the second aspect of the present application provides an online recognition device for the winding state of a transformer, comprising:

[0030] a data acquisition module configured to acquire a current frequency response curve of the transformer;

[0031] a feature determination module configured to determine a current normalized frequency response hybrid variance according to the current frequency response curve and a preset normalized frequency response hybrid variance algorithm;

[0032] a state recognition module configured to obtain a current winding state of the transformer by using the current normalized frequency response hybrid variance and a preset winding state recognition model, wherein the winding state recognition model is obtained by pre-training a convolutional neural network model based on the normalized frequency response hybrid variance of samples under different winding states.

[0033] To achieve the above object, the third aspect of the present application provides a computer readable storage medium storing a computer program, wherein the computer program is executed by a processor to make the processor execute the steps of the first aspect and any feasible implementation manner.

[0034] To achieve the above object, the fourth aspect of the present application provides a computer device, comprising a memory and a processor, the memory stores a computer program, and the computer program is executed by the processor to make the processor execute the steps of the first aspect and any feasible implementation manner.

[0035] By adopting the embodiment of the present application, the following beneficial effects are achieved:

[0036] The present application provides an online identification method for the winding state of a transformer, comprising: obtaining a current frequency response curve of the transformer; determining a current normalized frequency response hybrid variance according to the current frequency response curve and a preset normalized frequency response hybrid variance algorithm; and obtaining a current winding state of the transformer by using the current normalized frequency response hybrid variance and a preset winding state identification model, wherein the winding state identification model is obtained by pre-training a convolutional neural network model based on sample normalized frequency response hybrid variances under different winding states. By the above method, false judgments and potential safety risks caused by monitoring sensor errors can be avoided, the service life of the transformer is prolonged, and the like, which has important significance for ensuring the safe, reliable and efficient operation of the power system. BRIEF DESCRIPTION OF DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings.

[0038] Among them:

[0039] Figure 1 The flow chart of the online identification method for the winding state of a transformer in an embodiment of the present application;

[0040] Figure 2 Another flow chart of the online identification method for the winding state of a transformer in an embodiment of the present application;

[0041] Figure 3 The structure block diagram of the online identification device for the winding state of a transformer in an embodiment of the present application;

[0042] Figure 4 The structure block diagram of the computer device in an embodiment of the present application. DETAILED DESCRIPTION

[0043] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments of the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative work fall within the scope of the present application.

[0044] Please refer to Figure 1 , Figure 1 A flowchart of an online identification method for a winding state of a transformer in an embodiment of the present application. The method can be applied to a terminal or a server. The terminal can be a desktop terminal or a mobile terminal, and the mobile terminal can be at least one of a mobile phone, a tablet computer, a notebook computer, etc. The server can be implemented by an independent server or a server cluster composed of multiple servers. This embodiment is exemplified by application to a terminal, and the method is as shown in Figure 1 The method comprises the following steps:

[0045] 101. Obtain a current frequency response curve of the transformer.

[0046] 102. Determine a current normalized frequency response hybrid variance according to the current frequency response curve and a preset normalized frequency response hybrid variance algorithm.

[0047] It should be noted that the online frequency response test is performed on the transformer, and the current frequency response curve is obtained through the online frequency response test. The normalized frequency response hybrid variance is a normalized frequency response hybrid variance, and the frequency response hybrid variance is a hybrid variance between the current frequency response curve and a reference frequency response curve of the transformer. The current normalized frequency response hybrid variance can be obtained by using the normalized frequency response hybrid variance algorithm based on the current frequency response curve.

[0048] The current frequency response curve can be one or more groups, and the current normalized frequency response hybrid variance can be a hybrid variance between two groups of current frequency response curves and the reference frequency response curve, which is taken as an example in the present application.

[0049] And, the current normalized frequency response mixed variance can include current normalized frequency response mixed variances of three frequency bands, i.e., a low frequency band, a middle frequency band and a high frequency band, according to the frequency bands of the current frequency response curve. For example, the current normalized frequency response mixed variance of the low frequency band is calculated by using the curve data of the low frequency band corresponding to the two sets of current frequency response curves and the reference frequency response curve, the current normalized frequency response mixed variance of the middle frequency band is calculated by using the curve data of the middle frequency band corresponding to the two sets of current frequency response curves and the reference frequency response curve, and the current normalized frequency response mixed variance of the high frequency band is calculated by using the curve data of the high frequency band corresponding to the two sets of current frequency response curves and the reference frequency response curve. In this way, the current normalized frequency response mixed variances of the three frequency bands of the current frequency response curve are obtained. The winding state includes but is not limited to a normal state and an abnormal state, and the abnormal state includes but is not limited to various fault types.

[0050] 103. The current winding state of the transformer is obtained by using the current normalized frequency response mixed variance and a preset winding state recognition model, and the winding state recognition model is obtained by pre-training a convolutional neural network model based on sample normalized frequency response mixed variances under different winding states.

[0051] Further, the current normalized frequency response mixed variance is obtained, and the current normalized frequency response mixed variance can be input into the pre-trained convolutional neural network model to obtain the current winding state of the transformer. Since the convolutional neural network model is trained based on sample normalized frequency response mixed variances under different winding states, a large number of corresponding relationships between sample normalized frequency response mixed variances and winding states have been pre-learned, and therefore the corresponding winding state can be output according to the current normalized frequency response mixed variance.

[0052] The current normalized frequency response mixed variance can include current normalized frequency response mixed variances of three frequency bands (high, middle and low), thereby improving the accuracy of online identification of the winding state, and the misjudgment and potential safety risk caused by the error of the monitoring sensor can be avoided.

[0053] The application provides an online identification method of a winding state of a transformer, which comprises the following steps: obtaining a current frequency response curve of the transformer; determining a current normalized frequency response mixed variance according to the current frequency response curve and a preset normalized frequency response mixed variance algorithm; and obtaining a current winding state of the transformer by using the current normalized frequency response mixed variance and a preset winding state recognition model, wherein the winding state recognition model is obtained by pre-training a convolutional neural network model based on sample normalized frequency response mixed variances under different winding states. By the above method, the misjudgment and potential safety risk caused by the error of the monitoring sensor can be avoided, the service life of the transformer is prolonged, and the like. The method has important significance for ensuring the safe, reliable and efficient operation of the power system.

[0054] Please refer toFigure 2 , Figure 2 Another flow chart of an on-line identification method of a winding state of a transformer in an embodiment of the present application is shown in FIG. 2. Figure 2 The method shown in FIG. 2 comprises the following steps:

[0055] 201. Obtain a current frequency response curve of the transformer;

[0056] It should be noted that step 201 is similar to the content of step 101 shown in FIG. 1, and to avoid repetition, the specific content can be referred to the content of step 101 shown in FIG. 1. Figure 1 Figure 1

[0057] 202. Obtain a current frequency response hybrid variance of different frequency bands by using the reference frequency response curve of the transformer, the two sets of current frequency response curves of the transformer and the preset frequency response hybrid variance algorithm.

[0058] It can be understood that the frequency response hybrid variance in the present application can be the hybrid variance between the reference frequency response curve and the current frequency response curve, and can be the hybrid variance between the reference frequency response curve and the two sets of current frequency response curves. And according to the different frequency bands of the curves, the frequency response hybrid variance of different frequency bands can also be obtained.

[0059] Further, the mathematical expression of the frequency response hybrid variance algorithm is as follows:

[0060]

[0061] In the formula, S f is the frequency response hybrid variance under the frequency band f, i is the selected frequency point number under the frequency band f, X is the total number of selected frequency points i; N1(i) is the amplitude of the frequency point i corresponding to the first set of current frequency response curves, N2(i) is the amplitude of the frequency point i corresponding to the second set of current frequency response curves, and M(i) is the amplitude of the frequency point i corresponding to the reference frequency response curve.

[0062] For example, the reference frequency response curve and the two sets of current frequency response curves of the transformer are selected 100 frequency points in the low frequency band (1 kHz-100 kHz), the medium frequency band (100 kHz-600 kHz) and the high frequency band (600 kHz-1 MHz), that is, X is 100, and the current frequency response hybrid variance of each fault type is calculated. Then, formula (1) is transformed into:

[0063]

[0064] 203. Obtain a current normalized frequency response hybrid variance of different frequency bands by using the current frequency response hybrid variance and the normalized frequency response hybrid variance algorithm.

[0065] ​​Further, the current frequency response mixing variance is normalized to obtain the normalized frequency response mixing variance.

[0066] For example, the mathematical expression of the normalized frequency response mixing variance algorithm is as follows:

[0067]

[0068] In the formula, is the normalized frequency response mixing variance of the frequency band f, S f is the frequency response mixing variance of the frequency band f, i is the frequency point number of the frequency band f; N1(i) is the amplitude of the frequency point i corresponding to the first group of current frequency response curves, N2(i) is the amplitude of the frequency point i corresponding to the second group of current frequency response curves, M(i) is the amplitude of the frequency point i corresponding to the reference frequency response curve, min is the minimum value, and max is the maximum value. It can be understood that different current normalized frequency response mixing variances under different frequency bands can be obtained according to different frequency bands

[0069] 204. Obtain the current winding state of the transformer by using the current normalized frequency response mixing variance and a preset winding state recognition model, wherein the winding state recognition model is obtained by pre-training a convolutional neural network model based on sample normalized frequency response mixing variances under different winding states.

[0070] It should be noted that step 204 is similar to the content of step 103 shown in FIG. 1, and details can be referred to the content of step 103 shown in FIG. 1. Figure 1 Figure 1

[0071] In a feasible implementation manner, the convolutional neural network model needs to be pre-trained, so that the convolutional neural network model can determine the winding state based on the normalized frequency response mixing variance. Therefore, the above method further includes steps A01 to A04:

[0072] A01. Obtain a sample data set, wherein the sample data set includes a plurality of sample normalized frequency response mixing variances labeled with real labels of winding states;

[0073] It can be understood that a plurality of normalized frequency response mixing variances of different winding states are collected as training samples to obtain a sample data set required for model training, wherein the sample data set includes a plurality of sample normalized frequency response mixing variances labeled with real labels of winding states.

[0074] ​​Exemplarily, the reference frequency response curve of transformer operation and two sets of fault frequency response curves of a certain degree and a certain type are selected at 100 frequency points in a low frequency band (1 kHz-100 kHz), a medium frequency band (100 kHz-600 kHz), and a high frequency band (600 kHz-1 MHz), and the normalized frequency response mixed variance of each fault type is calculated. The data in the three frequency bands is a sample, and 100 samples of each fault type are used as a data set. Then, the sample data set can be used to train the model to obtain a target model that can predict the winding state.

[0075] A02, dividing the sample data set into a training set and a validation set by cross-validation;

[0076] Further, in order to improve the accuracy of the model, the application adopts the cross-validation method to train the model, that is, the data of the training set is used for preliminary training, and then the validation set is used to verify the preliminary training model. Specifically, the sample data set is divided into a training set and a validation set by cross-validation, and the division ratio is not limited here.

[0077] A03, training the winding state prediction model according to the training set and a preset convolutional neural network model to obtain a trained convolutional neural network model;

[0078] After the training set is divided, the convolutional neural network model is first trained using the training set. The winding state prediction model is trained according to the training set and a preset convolutional neural network model to obtain a trained convolutional neural network model. Specifically, step A03 includes:

[0079] B01, training the convolutional neural network model using the sample normalized frequency response mixed variance in the training set to obtain a prediction label of the winding state, wherein the convolutional neural network model includes convolutional layers and fully connected layers connected in sequence, the number of convolutional layers is set to three, the convolution kernel size is selected to be 1*1, the step is 1, the activation function is selected to be ReLU, and batch normalization is applied after the convolutional layer or the fully connected layer;

[0080] B02, determining whether the convolutional neural network model converges using the prediction label, the true label of the sample normalized frequency response mixed variance in the training set, and a preset mean square error loss function;

[0081] B03, if the convolutional neural network model converges, a trained convolutional neural network model is obtained.

[0082] Wherein, the predicted label is used to indicate the predicted winding state, and the loss value of the convolutional neural network model at this time is obtained by the predicted label and the true label, so as to determine whether the convolutional neural network model converges. If it converges, it means that the difference between the predicted label and the true label is small, and the prediction of the true label can be almost completed. Then the training can be stopped, and step B03 is executed. If it does not converge, it means that the difference between the predicted label and the true label is large, the prediction accuracy is low, or the prediction requirement has not been reached. Then the loss value of the predicted label and the true label is used to adjust the model parameters of the current convolutional neural network model, and step B01 is returned until the convolutional neural network model converges, and the trained convolutional neural network model is obtained. Wherein, the calculation of the loss value includes the mean square error loss function.

[0083] A04, according to the verification set and the trained convolutional neural network model, the model optimization of winding state prediction is carried out, and the winding state recognition model is obtained.

[0084] Further, the trained convolutional neural network model is verified by using the verification set, and the prediction ability of the trained convolutional neural network model for winding state is evaluated by samples not participating in the preliminary training, that is, according to the verification set and the trained convolutional neural network model, the model optimization of winding state prediction is carried out, and the winding state recognition model is obtained.

[0085] In a feasible implementation, step A04 includes: using a linear learning rate decay strategy to adjust the model learning rate, introducing early stopping method, setting patience value and minimum change amount of verification loss of the verification set, calculating the loss of the trained convolutional neural network model on the verification set after each training round, saving the model parameters when the verification loss is lowest, and obtaining the winding state recognition model.

[0086] It should be noted that after the preliminary training, model optimization is also carried out, the linear learning rate decay strategy is used to adjust the model learning rate, the early stopping method is introduced, the patience value and the minimum change amount of the verification loss are set, the loss of the model on the verification set is calculated after each training round, and the model parameters when the verification loss is lowest are saved.

[0087] For example, the linear learning rate decay strategy is used to adjust the model learning rate, and the early stopping method is introduced, the patience value and the minimum change amount of the verification loss are set, the loss of the model on the verification set is calculated after each training round, and the model parameters when the verification loss is lowest are saved. The following links can be followed:

[0088] 1) The first link, initialize the parameters:

[0089] Set the initial learning rate `initial_lr`.

[0090] Set the learning rate decay rate `lr_decay`.

[0091] Set the patience value `patience` for early stopping, which is the number of consecutive epochs allowed when the loss on the validation set does not improve.

[0092] Set the minimum change in validation loss `min_delta` that is considered a significant improvement in model performance.

[0093] 2) Second step, set up the training environment:

[0094] Choose an optimizer such as SGD, Adam, etc., and set its initial learning rate.

[0095] Prepare the training data and validation data.

[0096] 3) Third step, train the model:

[0097] For each epoch:

[0098] Calculate the learning rate for the current epoch, usually using a linear decay formula: `current_lr = initial_lr * (1 - epoch / total_epochs)`.

[0099] Update the learning rate of the optimizer.

[0100] Train the model for one epoch.

[0101] Evaluate the model performance on the validation set and calculate the validation loss.

[0102] 4) Fourth step, apply early stopping:

[0103] Maintain a variable to record the best validation loss of the model `best_val_loss`, and initialize it to a large number.

[0104] Maintain a counter `counter` to record the number of consecutive epochs when the validation loss does not improve.

[0105] After each epoch ends:

[0106] If the current validation loss is lower than `best_val_loss`, and the improvement is greater than `min_delta`, update `best_val_loss`, save the current model parameters, and reset `counter`.

[0107] If the current validation loss does not improve, or the improvement is less than `min_delta`, increment `counter`.

[0108] If `counter` reaches `patience`, stop training.

[0109] 5) The fifth link, save the best model:

[0110] During training, save model parameters whenever there is a significant improvement in validation loss.

[0111] After training, use the saved best model parameters to restore the model state to obtain the winding state recognition model.

[0112] Further, after obtaining the winding state recognition model, the winding state recognition model can be used to monitor the winding state online, that is, the processes of steps 201 to 204 are executed. Specifically, the trained winding state recognition model is integrated into a hardware device or deployed on a cloud server, real-time receives and processes data from sensors, extracts key features from the collected data, calculates the normalized frequency response mixed variance of different frequency bands of the current state using formula (1) and formula (3);

[0113] Using the extracted features as input, the trained winding state recognition model is used for inference to obtain the state evaluation result of the transformer winding, and the running state of the transformer is adjusted in real time according to the output result of the model.

[0114] In addition, new monitoring data is regularly used to retrain and optimize the model through online learning or incremental learning, so that the model can adapt to changes in the state of the transformer.

[0115] The present application provides an online identification method for the winding state of a transformer, which comprises the following steps: obtaining the current frequency response curve of the transformer; using the reference frequency response curve of the transformer, the two sets of current frequency response curves of the transformer and the preset frequency response mixed variance algorithm to obtain the current frequency response mixed variance of different frequency bands; using the current frequency response mixed variance and the normalization frequency response mixed variance algorithm to obtain the current normalization frequency response mixed variance of different frequency bands; using the current normalization frequency response mixed variance and the preset winding state recognition model to obtain the current winding state of the transformer, wherein the winding state recognition model is obtained by pre-training a convolutional neural network model based on sample normalized frequency response mixed variances under different winding states. Through the above method, the normalized frequency response mixed variance is calculated based on the frequency response curve, the model is trained using a convolutional neural network, the model is optimized, and finally the winding state is monitored online using the model, and the model is periodically retrained and optimized. The present application has the advantages of avoiding misjudgment and potential safety risks caused by sensor errors, prolonging the service life of the transformer, and the like, and has important significance for ensuring the safe, reliable and efficient operation of the power system.

[0116] Please refer to Figure 3 ,Figure 3 A structural block diagram of an online identification device for a winding state of a transformer in an embodiment of the present application is shown in Figure 3 The device shown comprises:

[0117] A data acquisition module 301 is configured to acquire a current frequency response curve of the transformer.

[0118] A feature determination module 302 is configured to determine a current normalized frequency response hybrid variance according to the current frequency response curve and a preset normalized frequency response hybrid variance algorithm.

[0119] A state identification module 303 is configured to obtain a current winding state of the transformer by using the current normalized frequency response hybrid variance and a preset winding state identification model, which is obtained by pre-training a convolutional neural network model based on sample normalized frequency response hybrid variances under different winding states.

[0120] It should be noted that, Figure 3 the roles of the various modules in the device shown are similar to the contents of the various steps in the method shown in Figure 1 to avoid repetition, details can be referred to the contents of the various steps in the method shown in Figure 1 .

[0121] The present application provides an online identification device for a winding state of a transformer, which comprises: a data acquisition module configured to acquire a current frequency response curve of the transformer; a feature determination module configured to determine a current normalized frequency response hybrid variance according to the current frequency response curve and a preset normalized frequency response hybrid variance algorithm; and a state identification module configured to obtain a current winding state of the transformer by using the current normalized frequency response hybrid variance and a preset winding state identification model, which is obtained by pre-training a convolutional neural network model based on sample normalized frequency response hybrid variances under different winding states. In this way, false judgments and potential safety risks caused by monitoring sensor errors can be avoided, the service life of the transformer is prolonged, and the like. This has important significance for ensuring the safe, reliable and efficient operation of the power system.

[0122] Figure 4 An internal structure diagram of a computer device in an embodiment is shown. The computer device can be a terminal or a server. As shown in Figure 4As shown, the computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system, and can also store a computer program, which, when executed by the processor, can enable the processor to implement the above method. The internal memory can also store a computer program, which, when executed by the processor, can enable the processor to execute the above method. Those skilled in the art can understand that Figure 4 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0123] In one embodiment, a computer device is provided, comprising a memory and a processor, the memory stores a computer program, the computer program is executed by the processor, so that the processor executes the method as Figure 1 or Figure 2 The steps shown.

[0124] In one embodiment, a computer readable storage medium is provided, storing a computer program, the computer program is executed by the processor, so that the processor executes the method as Figure 1 or Figure 2 The steps shown.

[0125] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The program can be stored in a non-volatile computer readable storage medium, and when the program is executed, the processes of the above-mentioned embodiment methods can be included. Any reference to memory, storage, database, or other medium used in the embodiments provided in the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0126] Any combination of the technical features of the above embodiments can be made. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combination of the technical features does not exist, it should be considered as the scope of the present application.

[0127] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent of the present application. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. An on-line method of identifying the state of a winding of a transformer, characterized in that, The method comprises: acquiring a current frequency response curve of the transformer; determining a current normalized frequency response hybrid variance according to the current frequency response curve and a preset normalized frequency response hybrid variance algorithm; obtaining a current winding state of the transformer by using the current normalized frequency response hybrid variance and a preset winding state recognition model, wherein the winding state recognition model is obtained by pre-training a convolutional neural network model based on sample normalized frequency response hybrid variances under different winding states; wherein the mathematical expression of the frequency response hybrid variance algorithm is as follows: ； In the formula, S f is the frequency band f , and i is the selected frequency point number in the frequency band f , X is the total number of selected frequency points; i N 1( i ) is the amplitude of the frequency point i corresponding to the first group of current frequency response curves, N 2( i ) is the amplitude of the frequency point i corresponding to the second group of current frequency response curves, M ( i ) is the amplitude of the frequency point i corresponding to the reference frequency response curve.​ wherein the mathematical expression of the normalized frequency response hybrid variance algorithm is as follows: ； In the formula, is the normalized frequency response hybrid variance in the frequency band f S f is the frequency response hybrid variance in the frequency band f i is the frequency point number in the frequency band f N 1( i ) is the amplitude of the frequency point i in the first group of current frequency response curves, N 2( i ) is the amplitude of the frequency point i in the second group of current frequency response curves, M ( i ) is the amplitude of the frequency point i in the reference frequency response curve, min is the minimum value, max is the maximum value.​​​ wherein the obtaining of the current winding state of the transformer by using the current normalized frequency response hybrid variance and the preset winding state recognition model comprises: training a convolutional neural network model by using sample normalized frequency response hybrid variances in a training set to obtain a prediction label of the winding state, wherein the convolutional neural network model comprises convolutional layers and fully connected layers connected in sequence, the number of the convolutional layers is set to three, the convolution kernel size is selected to be 1*1, the step is 1, the activation function is selected to be ReLU, and batch normalization is applied after the convolutional layers or the fully connected layers; determining whether the convolutional neural network model converges by using the prediction label, real labels of the sample normalized frequency response hybrid variances in the training set and a preset mean square error loss function; if the convolutional neural network model converges, obtaining a trained convolutional neural network model.

2. The method of claim 1, wherein, The determining of the current normalized frequency response hybrid variance according to the current frequency response curve and the preset normalized frequency response hybrid variance algorithm comprises: obtaining current frequency response hybrid variances of different frequency bands by using a reference frequency response curve of the transformer, two current frequency response curves of the transformer and a preset frequency response hybrid variance algorithm; obtaining current normalized frequency response hybrid variances of different frequency bands by using the current frequency response hybrid variances and the normalized frequency response hybrid variance algorithm.

3. The method of claim 1, wherein, The method further comprises: acquiring a sample data set, wherein the sample data set comprises sample normalized frequency response hybrid variances labeled with real labels of winding states; dividing the sample data set into a training set and a validation set by using cross-validation; training a winding state prediction model according to the training set and a preset convolutional neural network model to obtain a trained convolutional neural network model; optimizing the winding state prediction model according to the validation set and the trained convolutional neural network model to obtain a winding state recognition model.

4. The method of claim 3, wherein, The optimizing of the winding state prediction model according to the validation set and the trained convolutional neural network model comprises: adjusting a model learning rate by using a linear learning rate decay strategy, introducing an early stopping method, setting a patience value and a minimum change amount of a validation loss of the validation set, calculating a loss of the trained convolutional neural network model on the validation set after each round of training, saving model parameters when the validation loss is lowest, and obtaining the winding state recognition model.

5. An apparatus for on-line identification of the state of a winding of a transformer, characterized in that The method according to any one of claims 1 to 4 is applied to the device, and the device comprises: a data acquisition module configured to acquire a current frequency response curve of the transformer; a feature determination module configured to determine a current normalized frequency response hybrid variance according to the current frequency response curve and a preset normalized frequency response hybrid variance algorithm; a state recognition module configured to obtain a current winding state of the transformer by using the current normalized frequency response hybrid variance and a preset winding state recognition model, the winding state recognition model being obtained by pre-training a convolutional neural network model based on sample normalized frequency response hybrid variances under different winding states.

6. A computer readable storage medium storing a computer program, characterized in that, The computer program is executed by the processor, so that the processor executes the steps of the method according to any one of claims 1 to 4. 7.A computer device, comprising a memory and a processor, and characterized in that, The memory stores a computer program, and the computer program is executed by the processor, so that the processor executes the steps of the method according to any one of claims 1 to 4.

Images