A method, device, storage medium and electronic device for identifying discharge type

Abstract

The present application discloses a method, device, storage medium and electronic device for identifying discharge types, the method comprising: determining a PRPD spectrum without phase correction based on a collected ultrasonic discharge signal, and inputting the PRPD spectrum into a pre-trained neural network model for discharge type identification; in the neural network model, extracting features from the PRPD spectrum, and performing phase prediction on the obtained feature extraction results to obtain phase estimation features; performing classification prediction based on the phase estimation features and the feature extraction results to determine the probability information of each classification; and determining the discharge type corresponding to the PRPD spectrum based on the probability information of each classification. The application of the present application can effectively improve the accuracy of discharge type identification.

Description

Technical Field

[0001] The present application relates to ultrasonic signal processing technology, and in particular to a discharge type recognition method, device, storage medium and electronic device. Background Art

[0002] With the development of ultrasonic signal processing technology, non-contact acoustic imagers are currently mostly used for discharge detection. They process, detect and identify the ultrasonic waves generated by the discharge in space to determine whether discharge occurs and the specific type of discharge.

[0003] In the current discharge detection method, the discharge data is generally counted, and the discharge type is identified using simple statistical results or cluster center solutions. In this type of discharge detection method, for scenes with low discharge volume, long-distance scenes, strong noise scenes, and non-discharge scenes, the discharge points have more interference and many noise points. Therefore, the statistical results obtained based on the collected discharge data in these scenes are greatly affected by interference, making the detection accuracy of the discharge type low. Summary of the invention

[0004] The present application provides a method, device, storage medium and electronic device for identifying a discharge type, which can effectively improve the accuracy of identifying the discharge type.

[0005] To achieve the above objectives, this application adopts the following technical solutions:

[0006] A method for identifying a discharge type, comprising:

[0007] Determine a phase-resolved partial discharge PRPD spectrum without phase correction based on the collected ultrasonic discharge signal, and input the PRPD spectrum into a pre-trained neural network model for discharge type recognition;

[0008] In the neural network model, feature extraction is performed on the PRPD spectrum, and phase prediction is performed on the obtained feature extraction results to obtain phase estimation features; classification prediction is performed based on the phase estimation features and the feature extraction results to determine probability information of each classification;

[0009] Based on the probability information of each classification, the discharge type corresponding to the PRPD spectrum is determined.

[0010] Preferably, the training method of the neural network model includes:

[0011] Inputting the training PRPD spectrum with correct phase into the neural network model as training samples;

[0012] In the neural network model, the following processing is performed:

[0013] Performing phase perturbation on the training samples and marking actual phase shift labels for the training samples;

[0014] Performing feature extraction on the spectrum after the phase disturbance, and performing phase prediction on the obtained feature extraction result to obtain an estimated phase and a phase estimation feature;

[0015] Comparing the estimated phase with the actual phase shift label to determine a first loss function;

[0016] Performing classification prediction based on the phase estimation feature and the feature extraction result to obtain a predicted discharge type, comparing the predicted discharge type with an actual type label, and determining a second loss function;

[0017] The parameters of the neural network model are updated based on the first loss function and the second loss function, and the next round of training is performed until the training end condition is met.

[0018] Preferably, after determining the discharge type of the PRPD spectrum, the method further comprises:

[0019] Based on the discharge type of the PRPD spectrum, the PRPD spectrum is phase corrected.

[0020] Preferably, the phase correction of the PRPD spectrum comprises:

[0021] For a single peak tip discharge type, the discharge peak is determined in the PRPD spectrum, and the phase corresponding to the discharge peak is corrected to 90 degrees;

[0022] and / or,

[0023] For a double-peak tip discharge type or a double-peak surface discharge type, a discharge peak is determined in the PRPD spectrum, and a maximum discharge amount is determined among discharge amounts corresponding to other phases that are at least 90 degrees apart from the phase corresponding to the discharge peak, and phase values ​​corresponding to the discharge peak and the maximum discharge amount are determined, and the intermediate phase between the two phase values ​​is corrected to 180 degrees;

[0024] and / or,

[0025] For the suspended discharge type, the discharge peak is determined in the PRPD spectrum, and the second discharge amount whose first discharge amount exceeds delta times of the discharge peak is searched rightward from 45 degrees to the left of the phase corresponding to the discharge peak until the phase corresponding to the discharge peak is reached; if the second discharge amount is found, the phase corresponding to the second discharge amount is corrected to 90 degrees; if the second discharge amount is not found, the phase corresponding to the discharge peak is corrected to 90 degrees;

[0026] and / or,

[0027] For non-discharge types, no correction is performed;

[0028] and / or,

[0029] For other discharge types, the discrimination degree of confidence of various discharge types is less than the discrimination threshold, or the confidence of various discharge types is less than the confidence threshold, the discharge peak is determined in the PRPD spectrum, and the phase corresponding to the discharge peak is corrected to 90 degrees, or no phase correction is performed.

[0030] Preferably, the method further comprises: dividing the entire phase range into N phase intervals, counting the discharge amplitudes of the discharge points in the same phase interval to obtain discharge statistical information of the corresponding phase interval, and performing phase correction based on the discharge statistical information.

[0031] A discharge type identification device comprises: a spectrum generating unit and a type identification unit;

[0032] The spectrum generating unit is used to determine a PRPD spectrum without phase correction based on the collected ultrasonic discharge signal, and input the PRPD spectrum into a pre-trained neural network model for discharge type recognition;

[0033] The type identification unit is used to perform feature extraction on the PRPD spectrum in the neural network model, and perform phase prediction on the feature extraction results to obtain phase estimation features; perform classification prediction based on the phase estimation features and the feature extraction results to determine the discharge type corresponding to the PRPD spectrum.

[0034] Preferably, the device further comprises a model training unit for training the neural network model;

[0035] Wherein, the training method of the neural network model includes:

[0036] Inputting the training PRPD spectrum with correct phase into the neural network model as training samples;

[0037] In the neural network model, the following processing is performed:

[0038] Performing phase perturbation on the training samples and marking actual phase shift labels for the training samples;

[0039] Performing feature extraction on the spectrum after the phase disturbance, and performing phase prediction on the obtained feature extraction result to obtain an estimated phase and a phase estimation feature;

[0040] Comparing the estimated phase with the actual phase label to determine a first loss function;

[0041] Performing classification prediction based on the phase estimation feature and the feature extraction result to obtain a predicted discharge type, comparing the predicted discharge type with an actual type label, and determining a second loss function;

[0042] The parameters of the neural network model are updated based on the first loss function and the second loss function, and the next round of training is performed until the training end condition is met.

[0043] Preferably, the device further comprises a phase correction unit for performing phase correction on the PRPD spectrum based on the discharge type of the PRPD spectrum.

[0044] Preferably, in the phase correction unit, the phase correction of the PRPD spectrum comprises:

[0045] For a single peak tip discharge type, the discharge peak is determined in the PRPD spectrum, and the phase corresponding to the discharge peak is corrected to 90 degrees;

[0046] and / or,

[0047] For a double-peak tip discharge type or a double-peak surface discharge type, a discharge peak is determined in the PRPD spectrum, and a maximum discharge amount is determined among discharge amounts corresponding to other phases that are at least 90 degrees apart from the phase corresponding to the discharge peak, and phase values ​​corresponding to the discharge peak and the maximum discharge amount are determined, and the intermediate phase between the two phase values ​​is corrected to 180 degrees;

[0048] and / or,

[0049] For the suspended discharge type, the discharge peak is determined in the PRPD spectrum, and the second discharge amount whose first discharge amount exceeds delta times of the discharge peak is searched rightward from 45 degrees to the left of the phase corresponding to the discharge peak until the phase corresponding to the discharge peak is reached; if the second discharge amount is found, the phase corresponding to the second discharge amount is corrected to 90 degrees; if the second discharge amount is not found, the phase corresponding to the discharge peak is corrected to 90 degrees;

[0050] and / or,

[0051] For non-discharge types, no phase correction is performed;

[0052] and / or,

[0053] For other discharge types, the discrimination degree of confidence of various discharge types is less than the discrimination threshold, or the confidence of various discharge types is less than the confidence threshold, the discharge peak is determined in the PRPD spectrum, and the phase corresponding to the discharge peak is corrected to 90 degrees, or no phase correction is performed.

[0054] Preferably, the phase correction unit is further used to divide the entire phase range into N phase intervals, count the discharge amplitudes of the discharge points in the same phase interval, obtain discharge statistical information of the corresponding phase interval, and perform phase correction based on the discharge statistical information.

[0055] A computer-readable storage medium stores computer instructions, which, when executed by a processor, can implement any of the above-mentioned discharge type detection methods.

[0056] An electronic device comprising at least a computer-readable storage medium and a processor;

[0057] The processor is used to read executable instructions from the computer-readable storage medium and execute the instructions to implement any of the above-mentioned discharge type detection methods.

[0058] It can be seen from the above technical solution that in this application, a PRPD spectrum without phase correction is determined based on the collected ultrasonic discharge signal, and the PRPD spectrum is input into a pre-trained neural network model for discharge type identification; in the neural network model, feature extraction is performed on the PRPD spectrum, and phase prediction is performed on the feature extraction results to obtain phase estimation features; thus, a phase prediction function is introduced in the neural network model to obtain phase estimation information. Next, classification prediction is performed based on the phase estimation features and feature extraction results to determine the discharge type corresponding to the PRPD spectrum. Since in the classification prediction of the neural network model, not only the image features of the PRPD spectrum are used, but also the phase information is further introduced, more accurate classification can be performed, effectively improving the accuracy of discharge type identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 A schematic diagram of the basic flow of the discharge type identification method in this application;

[0060] Figure 2 This is a basic flow chart of the discharge type identification method in a specific embodiment of the present application;

[0061] Figure 3 It is a time domain diagram of the discharge signal collected in the specific embodiment of the present application;

[0062] Figure 4It is a frequency domain diagram of the discharge signal collected in the specific embodiment of the present application;

[0063] Figure 5 This is a schematic diagram of signal processing of an original ultrasonic discharge signal in a specific embodiment of the present application;

[0064] Figure 6a~Figure 6c These are the example diagrams of PRPD spectra of single peak tip discharge, suspension discharge and surface discharge;

[0065] Figure 7 This is a schematic diagram of a training model of a discharge type recognition model in a specific embodiment of the present application;

[0066] Figure 8 A schematic diagram of a discharge type prediction in a specific embodiment of the present application;

[0067] Fig. 9 A histogram of the discharge amount in each phase interval in a specific embodiment of the present application;

[0068] Fig.10a It is the discharge amount histogram corresponding to the single peak tip discharge type before and after phase correction;

[0069] Fig.10b The histogram of discharge volume corresponding to the double peak tip discharge type before and after phase correction;

[0070] Fig.10c The histogram of discharge amount corresponding to the double-peak surface discharge type before and after phase correction;

[0071] Fig.10d The discharge amount histogram of the suspension discharge type before and after phase correction;

[0072] Fig.11 A schematic diagram of the basic structure of the electronic device provided in this application. DETAILED DESCRIPTION

[0073] In order to make the objectives, technical means and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings.

[0074] The basic idea of ​​the present application is to use a neural network model to identify the discharge type and introduce phase information for classification of the neural network model.

[0075] Figure 1 FIG. 1 is a schematic diagram of the basic flow of the discharge type identification method in this application. Figure 1 As shown, the method includes:

[0076] Step 101, determining a PRPD spectrum without phase correction based on the collected ultrasonic discharge signal, and inputting the PRPD spectrum into a pre-trained neural network model for discharge type recognition.

[0077] This step is used to collect ultrasonic discharge signals and determine a PRPD (Phase Resolved Partial Discharge) spectrum based on the signal. The PRPD spectrum obtained here is a phase-uncorrected PRPD spectrum. The method of the present application can use this inaccurate PRPD spectrum to accurately classify the discharge type. Optionally, the PRPD spectrum can be further corrected according to different classifications of the discharge type.

[0078] In this application, a neural network model is used to identify the discharge type. Therefore, after obtaining the PRPD spectrum, the spectrum is input into a pre-trained neural network model for discharge type identification. Hereinafter, the neural network model used for discharge type identification in this application may also be referred to as a discharge type identification model.

[0079] The discharge type recognition model is pre-trained. When the discharge type recognition is actually performed, the PRPD spectrum to be recognized is directly input into the discharge type recognition model, and the recognized discharge type is output after being processed by the model. The specific discharge type may include single peak tip discharge, double peak tip discharge, etc. Optionally, the recognized type may also include a non-discharge type to realize the detection function of determining whether there is a discharge.

[0080] Step 102, in the neural network model, feature extraction is performed on the PRPD spectrum, and phase prediction is performed on the feature extraction results to obtain phase estimation features.

[0081] In the processing of the discharge type recognition model, the PRPD spectrum is firstly subjected to image feature extraction to obtain feature extraction results; next, in order to ensure the classification accuracy, phase prediction is performed based on the feature extraction results to obtain phase estimation features, thereby further obtaining phase information through image features, thereby providing richer information reference for subsequent classification detection.

[0082] Step 103, performing classification prediction based on the phase estimation feature and the feature extraction result to determine the probability information of each classification; based on the probability information of each classification, determining the discharge type corresponding to the PRPD spectrum.

[0083] In the discharge type recognition model, the discharge type is classified and predicted based on the image feature extraction result and phase estimation feature determined in step 103, and the probability information of each different discharge type can be obtained. Finally, the discharge type corresponding to the PRPD spectrum is determined based on the corresponding probability information.

[0084] At this point, the basic process of the discharge type identification method in this application is completed. In the above process, a neural network model is trained for discharge type identification, and a phase prediction function is introduced into the neural network model. The additional phase information is used as a reference for discharge classification, which can directly identify the discharge type of the PRPD spectrum without phase correction and effectively improve the accuracy of discharge type identification.

[0085] In the above Figure 1 On the basis of the method flow shown, in order to achieve phase correction of the PRPD spectrum, optionally, the present application can further perform phase correction on the PRPD spectrum based on the determined discharge type of the PRPD spectrum after step 103. The phase correction here is carried out in a targeted manner according to the characteristics of different discharge types, and differentiated phase correction strategies can be adopted for different discharge types so that each discharge type can be correctly corrected in phase.

[0086] Optionally, for a single peak tip discharge type, the discharge peak value may be determined in the PRPD spectrum, and the phase corresponding to the discharge peak value may be corrected to 90 degrees;

[0087] For the double-peak tip discharge type or double-peak surface discharge type, the discharge peak value can be determined in the PRPD spectrum, and the maximum discharge amount can be determined among the discharge amounts corresponding to other phases that are at least 90 degrees apart from the phase corresponding to the discharge peak value, and the phase values ​​corresponding to the discharge peak value and the maximum discharge amount can be determined, and the intermediate phase between the two phase values ​​can be corrected to 180 degrees;

[0088] For the suspended discharge type, the discharge peak can be determined in the PRPD spectrum, and the second discharge amount whose first discharge amount exceeds delta times the discharge peak amount can be searched rightward from the phase corresponding to the discharge peak amount 45 degrees to the left, until the phase corresponding to the discharge peak amount is reached; if the second discharge amount is found, the phase corresponding to the second discharge amount is corrected to 90 degrees; if the second discharge amount is not found, the phase corresponding to the discharge peak amount is corrected to 90 degrees;

[0089] For non-discharge types, phase correction may not be performed;

[0090] For other discharge types, the discrimination degree of confidence of various discharge types is less than the discrimination threshold, or the confidence of various discharge types is less than the confidence threshold, the discharge peak value can be determined in the PRPD spectrum, and the phase corresponding to the discharge peak value can be corrected to 90 degrees, or no phase correction can be performed.

[0091] The specific implementation of this application is described below through specific embodiments. Figure 2 FIG. 1 is a schematic diagram of a specific flow chart of the discharge type identification method in this application. Figure 2 As shown, the method includes:

[0092] Step 201, collecting ultrasonic discharge signals.

[0093] Usually, an acoustic imaging device can be used as a collection device to collect ultrasonic discharge signals. Taking the classic device as an example, N microphones can be distributed in a spiral plane, with a collection frequency band of 0 Hz~F0 Hz (corresponding to a sampling rate of 2×F0 Hz), and the multi-channel audio signals collected by the N microphones can be stored, that is, the original data of the ultrasonic discharge signal.

[0094] The data collected by the above equipment can be analyzed in time domain and frequency domain to obtain the power frequency periodicity and significance of its discharge. The processed discharge characteristics are displayed in a fixed periodic manner, which is the PRPD spectrum. Figure 3 and Figure 4 They are the time domain and frequency domain diagrams of the discharge signal respectively.

[0095] Step 202: performing signal processing on the collected ultrasonic discharge signal.

[0096] Usually, the original ultrasonic discharge signal can be processed by frequency band fusion and bandpass filtering to achieve the effect of signal noise reduction and enhance the signal-to-noise ratio of the discharge signal in the high frequency band.

[0097] Specifically, you can Figure 5 As shown, the multi-channel data is merged to obtain single-channel audio data; the audio data is then band-pass filtered to obtain a high-frequency narrow-band audio signal; the discharge point of the high-frequency narrow-band audio signal is selected to obtain the discharge point of each power frequency cycle. The original phase P0 and amplitude S0 of each discharge point are recorded.

[0098] Step 203, selecting discharge points based on the processed data, and generating a PRPD spectrum without phase correction.

[0099] The processed data is used to select discharge points, and the selected discharge points usually have characteristics such as "extreme points" and "local maximum points". Each discharge point has a pair of numerical coordinates: [P, S], where P represents the phase value of the point (converted to the 0~360 degree coordinate system), and S represents the energy value of the point, that is, the signal amplitude value. All points are plotted on a graph according to the [P, S] coordinates to obtain the phase amplitude scatter plot of the discharge point in this time period, that is, the PRPD spectrum.

[0100] Specifically, firstly, the discharge points of all power frequency cycles within 1s are obtained (taking 1s as an example, the time period can also be adjusted), and their "discharge coordinates" are obtained, namely: [P0_1, S0_1], [P0_2, S0_2], …, [P0_M, S0_M]; among them, P0_i is a number normalized to the numerical range of 0~360, representing the phase value of the i-th discharge point, and S0_i is the amplitude value, representing the discharge amplitude of the i-th discharge point; next, according to S0_max, each amplitude value S0_i is normalized to the interval [0,1] to obtain S_i; finally, all coordinate points [P0_i, S_i] are plotted on a graph to obtain the PRPD spectrum, which is a version obtained directly from the ultrasonic discharge signal without phase correction.

[0101] Here are some typical PRPD graph examples: Figure 6a~Figure 6c They are example diagrams of single-peak tip discharge, suspension discharge and surface discharge respectively.

[0102] Through the processing of the above steps 201 to 203, a PRPD spectrum without phase correction can be determined based on the collected ultrasonic discharge signal. Specifically, the processing of the above steps 201 to 203 is only an example, and various existing methods can be used in actual applications, not limited to the specific processing method given above.

[0103] Step 204: input the PRPD spectrum into a pre-trained discharge type recognition model, and determine the discharge type of the PRPD spectrum after processing using the model.

[0104] The discharge type recognition model is a pre-trained neural network model used for discharge type classification. The following is an example of the training process of the neural network model. Figure 7 As shown, the training model provided in the present application includes a category classifier and a phase predictor, which are used to classify the discharge type and predict the phase respectively, and the predicted phase information is used to classify the discharge type. Specifically, in the training stage, based on a large amount of discharge data, a training PRPD spectrum is obtained, and the training PRPD spectrum is a spectrum with correct phase. The spectrum is used as training data for model training. The specific training process may include:

[0105] Step 1: Perform phase perturbation on the input training PRPD spectrum and mark the actual phase shift label;

[0106] The phase perturbation can be to perform data enhancement on the phase of the training PRPD spectrum to obtain a perturbed PRPD spectrum, simulating an actual PRPD spectrum without phase correction. In order to train the parameters of the phase prediction part in the neural network model, after the training PRPD spectrum is phase perturbed in this step, the actual phase shift label corresponding to the perturbation is marked.

[0107] Step 2: Perform image feature extraction on the training graph after adding disturbance to obtain feature extraction results;

[0108] Among them, the feature extractor is based on the structure of the CNN network. Of course, in practical applications, the temporal neural network, convolutional neural network, Transformer neural network, and derivative versions of the above basic networks can also be used according to actual needs. The specific processing of image feature extraction using these feature extractors can be implemented in existing ways, which will not be repeated here.

[0109] Step 3: Based on the feature extraction result, a phase predictor is used to perform phase prediction to obtain a phase estimation feature;

[0110] Among them, the phase extractor can extract the phase estimation feature vector based on the image feature extraction result, which can be implemented based on the structure of the CNN network. Of course, in practical applications, temporal neural networks, convolutional neural networks, Transformer neural networks, and derivative versions of the above basic networks can also be used according to actual needs.

[0111] Step 4: Compare the estimated phase obtained by the phase predictor with the actual phase label to determine a first loss function;

[0112] For the training PRPD spectrum, the actual phase shift label is marked after the perturbation processing. In this step, the estimated phase is compared with the actual phase label to determine the first loss function, which is used to measure the accuracy of the current model parameters related to phase prediction. The first loss function can be an error loss function such as MSE.

[0113] Step 5: Based on the feature extraction result of step 2 and the phase estimation feature of step 3, a category classifier is used to perform classification prediction;

[0114] In the training model, classification prediction is performed based on the image feature extraction results and phase estimation features, so that more phase information can be introduced into the classification process and the accuracy of the training model can be improved.

[0115] Step 6, comparing the predicted classification obtained by the classification prediction with the actual type label of the training PRPD atlas to determine the second loss function;

[0116] Through classification prediction, the probability information of each pre-set classification can be obtained. Usually, the classification with the highest probability is selected as the predicted classification. The predicted classification is compared with the actual type label marked in advance for the training sample, and the second loss function can be determined to measure the accuracy of the current model parameters related to the classification process. Among them, the second loss function can be various loss functions commonly used in classification processing, such as CE.

[0117] Step 7: Update the parameters of the neural network model based on the first loss function and the second loss function;

[0118] The first loss function and the second loss function of this round of training are obtained through the above processing, and the model parameters of this round of training are updated based on the above two loss functions. Specifically, when updating the parameters based on the two loss functions, the two loss functions can be fused, and then the parameters are updated based on the fusion result.

[0119] As described above, through the processing of steps 1 to 7, a round of model training using training samples is completed.

[0120] Step 8: Determine whether the training end condition is met. If not, input other training samples into the training model, return to step 1, and perform the next round of training. If it is met, end the training and obtain a trained discharge type recognition model.

[0121] The training end condition may be, for example, reaching a set training round, or reaching a specified loss function value, etc. If the training end condition is not met, the next round of training will be continued; if the training end condition is met, the training will be terminated, and the current discharge type recognition model will be used as the trained discharge type recognition model.

[0122] At this point, the exemplary training process ends. Of course, the above training process is only an example, and the specific implementation in actual applications is not limited to this, and a more appropriate training process can be set according to actual needs.

[0123] The training process of the above-mentioned discharge type recognition model is completed in advance. In step 204, the trained model is directly used to determine the discharge type. The process of using the model to predict the discharge type in step 204 can be as follows: Figure 8As shown, the PRPD spectrum without phase correction (that is, the PRPD spectrum of random phase) is subjected to feature extraction by a feature extractor to obtain a feature extraction result, and the feature extraction result is input into a phase predictor to obtain a phase estimation feature, and then the phase estimation feature and the feature extraction result are input into a category classifier to obtain the prediction probability of various discharge types. Optionally, the discharge type may specifically include: single-peak tip discharge, double-peak tip discharge, double-peak surface discharge, suspended discharge, non-discharge type, etc. Based on the prediction probability of all discharge types, the discharge type corresponding to the PRPD spectrum is determined.

[0124] Since in the aforementioned model training process, the training PRPD spectrum with correct phase is perturbed before being used for model training, and the phase prediction function is introduced to use the phase information as a reference for classification, the trained model can accurately predict the discharge type based on the PRPD spectrum without phase correction.

[0125] After the prediction probabilities of different discharge types are obtained through the model, the discharge type corresponding to the PRPD spectrum is determined by combining the prediction probabilities. Usually, the discharge type with the largest prediction probability is selected as the discharge type corresponding to the PRPD spectrum.

[0126] As described above, the discharge type identification of the PRPD spectrum can be realized. In addition, in the method of using ultrasonic discharge signals to identify the discharge type, since the electrical signal is not directly detected, the true AC phase value cannot be known. Based on this, it is usually necessary to perform phase correction on the PRPD spectrum obtained by this method to obtain accurate phase information. In order to achieve accurate phase correction, the following processing is further performed:

[0127] Step 205: In the PRPD spectrum, count the discharge amount in each phase interval.

[0128] When performing phase correction, the discharge amount of each discharge point is expressed by interval statistics. Specifically, the entire phase interval of 0-360 can be divided into N different phase intervals, and the discharge amplitudes of the discharge points in the same phase interval are counted to obtain the discharge statistical information of the corresponding phase interval, and phase correction is performed based on the discharge statistical information.

[0129] Specifically in this embodiment, according to the discharge points obtained in step 203, the average discharge intensity of each interval [A_0, A_1, A_2, ... A_N] is counted as the discharge statistical information; wherein A_i is the discharge intensity of the ith interval, and A_max is the maximum discharge intensity of each phase interval, that is, the discharge peak value. That is to say, in this embodiment, the average discharge intensity of the discharge points in the interval is used to represent the interval discharge amount. Of course, other statistical results such as the total discharge amount can also be used to represent the discharge amount. Fig. 9 : is a histogram of the discharge amount in each phase interval obtained after statistics in this embodiment.

[0130] Step 206 , for the discharge type determined in step 204 , a corresponding phase correction strategy is used to perform phase correction on the PRPD spectrum.

[0131] Some examples of phase correction strategies corresponding to typical discharge types are given below.

[0132] 1. Single peak tip discharge

[0133] Specifically, the maximum value A_max can be found among all A_i, the phase interval P_max corresponding to A_max is corrected to 90 degrees, and the phases of other phase intervals are adaptively adjusted. Fig.10a Shown is the discharge amount histogram corresponding to the single-peak tip discharge type before and after phase correction.

[0134] The reason why this phase correction strategy is adopted here is that: through a large number of experiments, it is found that for the single-peak tip discharge type, the maximum value of the actual discharge amount usually appears at the phase position of 90 degrees. Therefore, for the single-peak tip discharge, the phase corresponding to the discharge peak is corrected to 90 degrees.

[0135] 2. Double peak tip discharge

[0136] Specifically, the maximum value A1_max and the phase interval P1_max corresponding to A1_max can be found in all A_i; then, the maximum value A2_max of the average discharge intensity and the phase interval P2_max corresponding to A2_max are found in the phase interval outside the phase of 90 degrees from P1_max; finally, the intermediate phase between P1_max and P2_max is corrected to 180 degrees, and the phases of other phase intervals are adaptively adjusted so that P1_max and P2_max are symmetrical along 180 degrees. Fig.10b Shown is the discharge amount histogram corresponding to the double-peak tip discharge type before and after phase correction.

[0137] The reason why this phase correction strategy is adopted here is that: through a large number of experiments, it is found that under the double-peak tip discharge type, the phases of the two discharge peaks are symmetrical along 180 degrees. Therefore, for the double-peak tip discharge, the middle value of the phases corresponding to the two discharge peaks is corrected to 180 degrees, so that the phases corresponding to the two discharge peaks are symmetrical along 180 degrees.

[0138] 3. Double peak surface discharge

[0139] The phase correction strategy for double-peak surface discharge is the same as that for double-peak tip discharge. Specifically, the maximum value A1_max and the phase interval P1_max corresponding to A1_max can be found among all A_i; then, the maximum value A2_max of the average discharge intensity and the phase interval P2_max corresponding to A2_max can be found in the phase interval beyond 90 degrees from the phase of P1_max; finally, the intermediate phase between P1_max and P2_max is corrected to 180 degrees, and the phases of other phase intervals are adaptively adjusted to make P1_max and P2_max symmetrical along 180 degrees. Fig.10c Shown is the discharge amount histogram corresponding to the double-peak surface discharge type before and after phase correction.

[0140] The reason why this phase correction strategy is adopted here is that: through a large number of experiments, it is found that the double-peak surface discharge type is similar to the double-peak tip discharge type, and the phases of the two discharge peaks are symmetrical along 180 degrees. Therefore, for double-peak surface discharge, the middle value of the phases corresponding to the two discharge peaks is corrected to 180 degrees, so that the phases corresponding to the two discharge peaks are symmetrical along 180 degrees.

[0141] 4. Suspension discharge

[0142] Specifically, the maximum value A1_max and the phase interval P1_max corresponding to A1_max can be found among all A_i; then, the phase 45 degrees to the left of phase P1_max is used as the starting point to search rightward until phase P1_max is used as the search end point, and the first phase exceeding A1_max is found. The average discharge intensity A2_max of delta, and the phase interval P2_max corresponding to A2_max; finally, if A2_max is found, P2_max is corrected to 90 degrees, if A2_max is not found, P1_max is corrected to 90 degrees, and the phases of other phase intervals are adaptively adjusted. Fig.10d Shown is the discharge amount histogram corresponding to the suspension discharge type before and after phase correction.

[0143] The reason why this phase correction strategy is adopted here is that: through a large number of experiments, it is found that the "first stripe" under the suspension discharge type usually appears at the position of 90 degrees in phase. Therefore, after the first stripe is determined, it is corrected to 90 degrees.

[0144] 5. Non-discharge type

[0145] For non-discharge types, phase correction is usually not required.

[0146] 6. Other discharge types

[0147] For other discharge types except the typical discharge types mentioned above, the same method as the single peak tip discharge type can be used to correct the phase interval corresponding to the discharge peak value A1_max to 90 degrees.

[0148] In addition, among the predicted probabilities corresponding to various discharge types output by the discharge type recognition model, it may be that the predicted probabilities of various discharge types are not high or the predicted probabilities of different discharge types have a small degree of distinction. In this case, the same method as the single-peak tip discharge type can be used to correct the phase interval corresponding to the discharge peak value A1_max to 90 degrees, or no phase correction may be performed.

[0149] At this point, the method flow in this embodiment ends. Through the above-mentioned discharge type identification method in this application, using the idea of ​​data perturbation, a neural network model with a phase prediction module for discharge type identification is designed, which can identify the discharge type based on the corrected PRPD spectrum and effectively improve the accuracy of discharge type identification; at the same time, this application can make differentiated phase correction strategies for different discharge types, so that each type can be correctly corrected, and the correct phase correction amount can be calculated to cover all discharge scenarios; further, the discharge amount statistical information of the discharge point is characterized by the average discharge amount in the phase interval, which can directly reflect the overall contour trend of the PRPD spectrum. The use of this statistic can quickly find the discharge concentration area (discharge peak, the method is simple and efficient, and avoids complex statistical operations such as "clustering" and "finding eigenvalues".

[0150] The above is a specific implementation of the discharge type identification method in this application. This application also provides a discharge type identification device, which can be used to implement the above identification method. Specifically, the basic structure of the discharge type identification device provided by this application includes: a spectrum generation unit and a type identification unit;

[0151] The spectrum generating unit is used to determine the PRPD spectrum without phase correction based on the collected ultrasonic discharge signal, and input the PRPD spectrum into a pre-trained neural network model for discharge type recognition;

[0152] The type recognition unit is used to extract features from the PRPD spectrum in the neural network model, and perform phase prediction on the feature extraction results to obtain phase estimation features; perform classification prediction based on the phase estimation features and the feature extraction results to determine the discharge type corresponding to the PRPD spectrum.

[0153] Optionally, the device may further include a model training unit for training the neural network model;

[0154] Among them, the training method of the neural network model may include:

[0155] The training PRPD spectrum with correct phase is input into the neural network model as training samples;

[0156] In the neural network model, the following processing is performed:

[0157] Perform phase perturbation on training samples and label the training samples with actual phase shift labels;

[0158] Perform feature extraction on the spectrum after phase disturbance, and perform phase prediction on the feature extraction results to obtain estimated phase and phase estimation features;

[0159] Comparing the estimated phase with the actual phase label to determine a first loss function;

[0160] Performing classification prediction based on the phase estimation feature and the feature extraction result to obtain a predicted discharge type, comparing the predicted discharge type with the actual type label, and determining a second loss function;

[0161] The parameters of the neural network model are updated based on the first loss function and the second loss function, and the next round of training is carried out until the training end conditions are met.

[0162] Optionally, the device may further include a phase correction unit for performing phase correction on the PRPD spectrum based on the discharge type of the PRPD spectrum.

[0163] Optionally, in the phase correction unit, the phase correction process for the PRPD spectrum may specifically include:

[0164] For the single peak tip discharge type, the discharge peak is determined in the PRPD spectrum, and the phase corresponding to the discharge peak is corrected to 90 degrees;

[0165] and / or,

[0166] For a double-peak tip discharge type or a double-peak surface discharge type, a discharge peak is determined in the PRPD spectrum, and a maximum discharge amount is determined among the discharge amounts corresponding to other phases that are at least 90 degrees apart from the phase corresponding to the discharge peak, and phase values ​​corresponding to the discharge peak and the maximum discharge amount are determined, and the intermediate phase between the two phase values ​​is corrected to 180 degrees;

[0167] and / or,

[0168] For the suspended discharge type, determine the discharge peak in the PRPD spectrum, and start from the phase corresponding to the discharge peak 45 degrees to the left and search rightward for the second discharge whose first discharge amount exceeds delta times the discharge peak until the phase corresponding to the discharge peak is reached; if the second discharge amount is found, the phase corresponding to the second discharge amount is corrected to 90 degrees; if the second discharge amount is not found, the phase corresponding to the discharge peak is corrected to 90 degrees;

[0169] and / or,

[0170] For non-discharge types, no phase correction is performed;

[0171] and / or,

[0172] For other discharge types, the discrimination degree of the confidence of various discharge types is less than the discrimination threshold, or the confidence of various discharge types is less than the confidence threshold, the discharge peak is determined in the PRPD spectrum, and the phase corresponding to the discharge peak is corrected to 90 degrees, or no phase correction is performed.

[0173] Optionally, the phase correction unit may be further configured to divide the entire phase range into N phase intervals, collect statistics on discharge amplitudes of discharge points in the same phase interval, obtain discharge statistical information of the corresponding phase interval, and perform phase correction based on the discharge statistical information.

[0174] The present application also provides a computer-readable storage medium, which stores instructions, and when the instructions are executed by a processor, the steps in the method for identifying a discharge type as described above can be executed. In practical applications, the computer-readable medium can be included in each device / apparatus / system in the above-mentioned embodiments, or it can exist independently without being assembled into the device / apparatus / system. Instructions are stored in a computer-readable storage medium, and when the instructions stored therein are executed by a processor, the steps in the method for identifying a discharge type as described above can be executed.

[0175] According to the embodiments disclosed in the present application, the computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, it may include but is not limited to: 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), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof, but it is not used to limit the scope of protection of the present application. In the embodiments disclosed in the present application, the computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, apparatus, or device.

[0176] Fig.11This application also provides an electronic device. Fig.11 As shown, it shows a schematic diagram of the structure of the electronic device involved in the embodiment of the present application, specifically:

[0177] The electronic device may include a processor 1101 with one or more processing cores, a memory 1102 with one or more computer-readable storage media, and a computer program stored in the memory and executable on the processor. When the program in the memory 1102 is executed, a discharge type identification method may be implemented.

[0178] Specifically, in practical applications, the electronic device may further include components such as a power supply 1103, an input / output unit 1104, etc. Those skilled in the art may understand that Fig.11 The structure of the electronic device shown in the figure does not constitute a limitation on the electronic device, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0179] Processor 1101 is the control center of the electronic device. It uses various interfaces and lines to connect various parts of the entire electronic device. It executes various functions of the server and processes data by running or executing software programs and / or modules stored in memory 1102 and calling data stored in memory 1102, thereby monitoring the electronic device as a whole.

[0180] The memory 1102 can be used to store software programs and modules, that is, the above-mentioned computer-readable storage medium. The processor 1101 executes various functional applications and data processing by running the software programs and modules stored in the memory 1102. The memory 1102 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function, etc.; the data storage area may store data created according to the use of the server, etc. In addition, the memory 1102 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage devices. Accordingly, the memory 1102 may also include a memory controller to provide the processor 1101 with access to the memory 1102.

[0181] The electronic device also includes a power supply 1103 for supplying power to various components, which can be logically connected to the processor 1101 through a power management system, so as to manage charging, discharging, power consumption and other functions through the power management system. The power supply 1103 can also include any components such as one or more DC or AC power supplies, recharging systems, power failure detection circuits, power converters or inverters, and power status indicators.

[0182] The electronic device may also include an input-output unit 1104, which may be used to receive input digital or character information and generate keyboard, mouse, joystick, and optical signal inputs related to user settings and function control. The input-output unit 1104 may also be used to display information input by the user or information provided to the user and various graphical user interfaces, which may be composed of graphics, text, icons, videos, and any combination thereof.

[0183] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for identifying a discharge type, characterized in that: include: Determine a phase-resolved partial discharge PRPD spectrum without phase correction based on the collected ultrasonic discharge signal, and input the PRPD spectrum into a pre-trained neural network model for discharge type recognition; In the neural network model, feature extraction is performed on the PRPD spectrum, and phase prediction is performed on the obtained feature extraction results to obtain phase estimation features; Performing classification prediction based on the phase estimation feature and the feature extraction result to determine probability information of each classification; Determining the discharge type corresponding to the PRPD spectrum based on the probability information of each classification; Wherein, the training method of the neural network model includes: Inputting the training PRPD spectrum with correct phase into the neural network model as training samples; In the neural network model, the following processing is performed: Performing phase perturbation on the training samples and marking actual phase shift labels for the training samples; Performing feature extraction on the spectrum after the phase disturbance, and performing phase prediction on the obtained feature extraction result to obtain an estimated phase and a phase estimation feature; Comparing the estimated phase with the actual phase shift label to determine a first loss function; Performing classification prediction based on the phase estimation feature and the feature extraction result to obtain a predicted discharge type, comparing the predicted discharge type with an actual type label, and determining a second loss function; The parameters of the neural network model are updated based on the first loss function and the second loss function, and the next round of training is performed until the training end condition is met.

2. The method according to claim 1, characterized in that After determining the discharge type of the PRPD spectrum, the method further comprises: Based on the discharge type of the PRPD spectrum, the PRPD spectrum is phase corrected.

3. The method according to claim 2, characterized in that The phase correction of the PRPD spectrum comprises: For a single peak tip discharge type, the discharge peak is determined in the PRPD spectrum, and the phase corresponding to the discharge peak is corrected to 90 degrees; and / or, For a double-peak tip discharge type or a double-peak surface discharge type, a discharge peak is determined in the PRPD spectrum, and a maximum discharge amount is determined among discharge amounts corresponding to other phases that are at least 90 degrees apart from the phase corresponding to the discharge peak, and phase values ​​corresponding to the discharge peak and the maximum discharge amount are determined, and the intermediate phase between the two phase values ​​is corrected to 180 degrees; and / or, For the suspended discharge type, the discharge peak is determined in the PRPD spectrum, and the second discharge amount whose first discharge amount exceeds delta times of the discharge peak is searched rightward from 45 degrees to the left of the phase corresponding to the discharge peak until the phase corresponding to the discharge peak is reached; if the second discharge amount is found, the phase corresponding to the second discharge amount is corrected to 90 degrees; if the second discharge amount is not found, the phase corresponding to the discharge peak is corrected to 90 degrees; and / or, For non-discharge types, no correction is performed; and / or, For other discharge types, the discrimination degree of confidence of various discharge types is less than the discrimination threshold, or the confidence of various discharge types is less than the confidence threshold, the discharge peak is determined in the PRPD spectrum, and the phase corresponding to the discharge peak is corrected to 90 degrees, or no phase correction is performed.

4. The method according to claim 2, characterized in that: The method further includes: dividing the entire phase range into N phase intervals, counting the discharge amplitudes of the discharge points in the same phase interval to obtain discharge statistical information of the corresponding phase interval, and performing phase correction based on the discharge statistical information.

5. A discharge type identification device, characterized in that: include: Graph generation unit, type recognition unit and model training unit; The spectrum generating unit is used to determine a PRPD spectrum without phase correction based on the collected ultrasonic discharge signal, and input the PRPD spectrum into a pre-trained neural network model for discharge type recognition; The type identification unit is used to perform feature extraction on the PRPD spectrum in the neural network model, and perform phase prediction on the obtained feature extraction results to obtain phase estimation features; Perform classification prediction based on the phase estimation feature and the feature extraction result to determine the discharge type corresponding to the PRPD spectrum; The model training unit is used to train the neural network model; Wherein, the training method of the neural network model includes: Inputting the training PRPD spectrum with correct phase into the neural network model as training samples; In the neural network model, the following processing is performed: Performing phase perturbation on the training samples and marking actual phase shift labels for the training samples; Performing feature extraction on the spectrum after the phase disturbance, and performing phase prediction on the obtained feature extraction result to obtain an estimated phase and a phase estimation feature; Comparing the estimated phase with the actual phase label to determine a first loss function; Performing classification prediction based on the phase estimation feature and the feature extraction result to obtain a predicted discharge type, comparing the predicted discharge type with an actual type label, and determining a second loss function; The parameters of the neural network model are updated based on the first loss function and the second loss function, and the next round of training is performed until the training end condition is met.

6. The device according to claim 5, characterized in that The device further includes a phase correction unit for performing phase correction on the PRPD spectrum based on the discharge type of the PRPD spectrum.

7. The device according to claim 6, characterized in that In the phase correction unit, the phase correction of the PRPD spectrum includes: For a single peak tip discharge type, the discharge peak is determined in the PRPD spectrum, and the phase corresponding to the discharge peak is corrected to 90 degrees; and / or, For a double-peak tip discharge type or a double-peak surface discharge type, a discharge peak is determined in the PRPD spectrum, and a maximum discharge amount is determined among discharge amounts corresponding to other phases that are at least 90 degrees apart from the phase corresponding to the discharge peak, and phase values ​​corresponding to the discharge peak and the maximum discharge amount are determined, and the intermediate phase between the two phase values ​​is corrected to 180 degrees; and / or, For the suspended discharge type, the discharge peak is determined in the PRPD spectrum, and the second discharge amount whose first discharge amount exceeds delta times of the discharge peak is searched rightward from 45 degrees to the left of the phase corresponding to the discharge peak until the phase corresponding to the discharge peak is reached; if the second discharge amount is found, the phase corresponding to the second discharge amount is corrected to 90 degrees; if the second discharge amount is not found, the phase corresponding to the discharge peak is corrected to 90 degrees; and / or, For non-discharge types, no phase correction is performed; and / or, For other discharge types, the discrimination degree of confidence of various discharge types is less than the discrimination threshold, or the confidence of various discharge types is less than the confidence threshold, the discharge peak is determined in the PRPD spectrum, and the phase corresponding to the discharge peak is corrected to 90 degrees, or no phase correction is performed.

8. The device according to claim 6, characterized in that The phase correction unit is further used to divide the entire phase range into N phase intervals, count the discharge amplitudes of the discharge points in the same phase interval, obtain the discharge statistical information of the corresponding phase interval, and perform phase correction based on the discharge statistical information.

9. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the instruction is executed by the processor, the discharge type detection method described in any one of claims 1 to 4 can be implemented.

10. An electronic device, characterized in that: The electronic device includes at least a computer-readable storage medium and also includes a processor; The processor is used to read executable instructions from the computer-readable storage medium and execute the instructions to implement the discharge type detection method described in any one of claims 1 to 4 above.

Images