Cable line partial discharge waveform classification method and device

Abstract

The invention relates to a cable line partial discharge waveform classification method and device, and the method comprises the following steps: collecting the partial discharge waveform data of a cable line, extracting the fat degree of the waveform, the maximum slope of the waveform leading edge, and the maximum signal-to-noise ratio, and inputting the data into a pre-trained partial discharge waveform classification model, classifying the partial discharge waveform and the noise waveform to obtain a partial discharge waveform classification result; the training process of the partial discharge waveform classification model comprises the steps of obtaining training data, inputting the training data into an SVM classification model for training, performing parameter adjustment on the SVM classification model through a global optimal parameter algorithm model to obtain optimal parameters for training of the SVM classification model, and finally obtaining the trained SVM classification model as the partial discharge waveform classification model. Compared with the prior art, the method has the advantages that the training time of the whole partial discharge waveform classification model is greatly shortened, and meanwhile, the partial discharge waveform classification model is better in performance and higher in accuracy.

Description

Technical Field

[0001] The invention relates to the technical field of cable line partial discharge detection, and in particular to a cable line partial discharge waveform classification method and device. Background Art

[0002] During the operation of high-voltage equipment, insulation defects may cause partial discharge (PD) phenomena. PD will cause changes in the structure and properties of insulating materials, thereby reducing insulation performance and affecting the normal operation of equipment. The causes of PD signals are relatively complex and varied. Common types of PD include internal air gap discharge, tip discharge, surface discharge, suspended particle discharge, failure to install stress tubes in cable terminals, and incorrect use of insulating tape on joints. As an important characteristic signal of the discharge process, the PD waveform is closely related to the type, location, and discharge intensity of the discharge source. Different types of PD signal waveforms have different characteristics in terms of amplitude, duration, pulse width, and reflected waves. Therefore, accurate identification of PD signal waveform types is of great significance for status assessment and fault diagnosis of power equipment.

[0003] The traditional classification methods of partial discharge waveforms are as follows:

[0004] 1) The method based on time domain features is simple to calculate, but the classification accuracy is not high and the anti-interference ability is weak.

[0005] 2) The method based on frequency domain features has relatively high classification accuracy, but it has high requirements on the measurement environment and cannot reflect the dynamic characteristics of the PD signal.

[0006] 3) The method based on time-frequency features fully combines the time domain and frequency domain features and has a high classification accuracy. However, the algorithm is complex, requires high signal processing capabilities, and requires manual operation, which is also very complex.

[0007] The traditional partial discharge waveform recognition method is complex to operate and requires professionals to classify the waveform. It is a manual experience analysis method with a low degree of intelligence. Summary of the invention

[0008] The purpose of the present invention is to provide a cable line partial discharge waveform classification method and device in order to overcome the defects of the traditional partial discharge waveform recognition method in the above-mentioned prior art, which is complicated to operate, requires professionals to classify the waveform, belongs to the manual experience analysis method, and has a low degree of intelligence.

[0009] The purpose of the present invention can be achieved by the following technical solutions:

[0010] A method for classifying partial discharge waveforms of cable lines comprises the following steps:

[0011] Collect partial discharge waveform data of the cable line, extract the fatness and thinness of the waveform, the maximum slope of the waveform front and the maximum signal-to-noise ratio, and input them into the pre-trained partial discharge waveform classification model to classify the partial discharge waveform and the noise waveform to obtain the partial discharge waveform classification result;

[0012] The training process of the partial discharge waveform classification model includes: obtaining training data, inputting the training data into the SVM classification model for training, and adjusting the parameters of the SVM classification model through a global optimal parameter algorithm model to obtain optimal parameters for training the SVM classification model, and finally obtaining the trained SVM classification model as the partial discharge waveform classification model.

[0013] Furthermore, the calculation process of the maximum slope of the waveform front includes:

[0014] In the entire partial discharge waveform data, the maximum slope of the waveform front between two points is found; the calculation expression of the maximum slope of the waveform front is:

[0015]

[0016] In the formula, K max is the maximum slope of the waveform front edge, V a is the amplitude at point a, V b is the amplitude at point b, T a is the time at point a, T b is the time of point b, and MAX is the maximum value of the slope of the leading edge of the waveform between the two points.

[0017] Furthermore, the maximum signal-to-noise ratio is the maximum signal difference between the signal data and the noise data on the spectrum graph, and the corresponding calculation expression is:

[0018] D=max(S a -S c )

[0019] Where D is the maximum signal-to-noise ratio, S a is the high frequency noise amplitude, S c is the signal amplitude after removing high-frequency noise, and max is the maximum value.

[0020] Furthermore, the fatness or thinness of the waveform is extracted through the kurtosis function of the starts function.

[0021] Furthermore, during the training process, the global optimal parameter algorithm model adjusts the overfitting penalty weight parameters and feature conversion kernel parameters of the SVM classification model in a grid search manner to obtain the optimal parameters with the lowest loss.

[0022] Furthermore, the method also uses in series a partial discharge discrimination method based on fqnt logic threshold and a neural network partial discharge discrimination method based on PRPD spectrum to classify and judge partial discharge waveforms;

[0023] The partial discharge discrimination method based on the fqnt logic threshold comprises: extracting the frequency f, the discharge amount q, the number of pulse discharges n and the discharge time t according to the partial discharge waveform data, and comparing them with the corresponding noise thresholds respectively, and if they are lower than the corresponding noise thresholds, judging them as noise waveforms;

[0024] The neural network partial discharge identification method based on the PRPD spectrum includes: extracting the PRPD spectrum according to the partial discharge waveform data, so as to determine whether it is a partial discharge waveform or a noise waveform.

[0025] Furthermore, the process of using the partial discharge discrimination method based on the fqnt logic threshold and the partial discharge discrimination method based on the neural network of the PRPD spectrum in series to classify and judge the partial discharge waveform is specifically as follows:

[0026] S101: Acquire partial discharge waveform data of the cable line;

[0027] S102: judging the partial discharge waveform data obtained in step S101 by a partial discharge discrimination method based on the fqnt logic threshold, and if it is judged to be a suspected partial discharge signal, executing step S103; if it is judged to be a noise waveform, returning to step S101;

[0028] S103: judging the partial discharge waveform data obtained in step S101 by using a neural network partial discharge discrimination method based on PRPD spectrum. If it is judged to be a suspected partial discharge signal, executing step S104; if it is judged to be a noise waveform, returning to step S101;

[0029] S104: extract the fatness or thinness of the waveform, the maximum slope of the waveform front and the maximum signal-to-noise ratio of the partial discharge waveform data obtained in step S101, and input them into the partial discharge waveform classification model for classification judgment. If it is judged to be a suspected partial discharge signal, the partial discharge waveform data is judged to be a partial discharge signal; if it is judged to be a noise waveform, return to step S101.

[0030] Furthermore, the method also uses in parallel a partial discharge discrimination method based on fqnt logic threshold and a neural network partial discharge discrimination method based on PRPD spectrum to classify and judge the partial discharge waveform.

[0031] Furthermore, the process of using the partial discharge identification method based on the fqnt logic threshold and the partial discharge identification method based on the neural network of the PRPD spectrum in parallel to classify and determine the partial discharge waveform is specifically as follows:

[0032] S201: Acquire partial discharge waveform data of the cable line;

[0033] S202: using the first scheme, the second scheme and the third scheme to judge the partial discharge waveform data obtained in step S201 respectively;

[0034] The first solution extracts the fatness and thinness of the waveform, the maximum slope of the waveform front and the maximum signal-to-noise ratio of the partial discharge waveform data obtained in step S101, and inputs them into the partial discharge waveform classification model for classification judgment;

[0035] The second scheme judges the partial discharge waveform data obtained in step S101 by a partial discharge discrimination method based on the fqnt logic threshold;

[0036] The third solution judges the partial discharge waveform data obtained in step S101 by using a neural network partial discharge discrimination method based on a PRPD spectrum;

[0037] S203: If at least two of the results in the first scheme, the second scheme, and the third scheme are determined to be suspected partial discharge signals, the partial discharge waveform data is determined to be a partial discharge signal; otherwise, the partial discharge waveform data is determined to be a noise signal.

[0038] The present invention also provides a cable line partial discharge waveform classification device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor calls the computer program to execute the steps of the above method.

[0039] Compared with the prior art, the present invention has the following advantages:

[0040] (1) The present invention extracts the fatness and thinness of the waveform data, the maximum slope of the waveform front and the maximum signal-to-noise ratio from the partial discharge waveform data of the cable line as the input of the global optimal parameter algorithm model and the SVM algorithm model, so that the reciprocating training of the model is more efficient and the influence of the features with lower weights on the model is greatly reduced;

[0041] During the training process of the SVM algorithm model, the optimal parameters obtained through the global optimal parameter algorithm model training are combined with the method and value of distinguishing local discharge and noise obtained through the SVM algorithm model training to obtain the optimal local discharge waveform classification model parameters. Based on the dual algorithm technology of global optimal parameter algorithm and support vector machine SVM, the training time of the entire local discharge waveform classification model is greatly reduced, while also ensuring the robustness and performance of the model, making the local discharge waveform classification model have better performance and higher accuracy.

[0042] (2) Finding the optimal parameters: The optimal parameters are found by adding the global optimal parameter algorithm model and providing them to the SVM algorithm model to obtain the partial discharge waveform classification model parameters, further improving the classification accuracy of partial discharge and noise waveforms.

[0043] Automatic parameter adjustment: The global optimal parameter algorithm model automatically searches for the optimal parameters, reducing the time required for manual identification and manual parameter adjustment, making the parameter optimization process more automated and simple.

[0044] Data standardization preprocessing: Standardized feature extraction preprocessing of waveform data enhances the performance and stability of the model.

[0045] Improved manual review efficiency: The global optimal parameter algorithm model and SVM algorithm model are combined with automatic classification of partial discharge and noise waveforms to improve the accuracy of waveform data classification and solve the problem of long time and inaccuracy in manual review of waveform data.

[0046] (3) The scheme of the present invention can also be used in series or in parallel with a discrimination program based on fqnt logic threshold and a neural network discrimination program based on PRPD spectrum, which can further improve the classification accuracy of partial discharge waveforms. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A schematic flow chart of a cable line partial discharge waveform classification method provided in an embodiment of the present invention;

[0048] Figure 2 A distribution diagram of fatness and thinness of a waveform provided in an embodiment of the present invention;

[0049] Figure 3 A distribution diagram of the maximum slope of a waveform front provided in an embodiment of the present invention;

[0050] Figure 4 A schematic diagram of a process of finding optimal parameters for a global optimal parameter algorithm model provided in an embodiment of the present invention;

[0051] Figure 5 It is a schematic diagram of partial discharge and noise waveforms automatically classified by a partial discharge waveform classification model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0053] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0055] Example 1

[0056] This embodiment provides a cable line partial discharge waveform classification method, proposes feature extraction preprocessing of waveform data and a global optimal parameter algorithm model, and uses the global optimal parameter algorithm model to train feature data to automatically find optimal parameters. Based on the global optimal parameter algorithm model and support vector machine (SVM) dual algorithm model technology, the shortcomings of traditional and existing partial discharge waveform classification methods are solved, and the specific steps include:

[0057] Collect partial discharge waveform data of the cable line, extract the fatness and thinness of the waveform, the maximum slope of the waveform front and the maximum signal-to-noise ratio, and input them into the pre-trained partial discharge waveform classification model to classify the partial discharge waveform and the noise waveform to obtain the partial discharge waveform classification result;

[0058] The training process of the partial discharge waveform classification model includes: obtaining training data, inputting the training data into the SVM classification model for training, and adjusting the parameters of the SVM classification model through the global optimal parameter algorithm model to obtain the optimal parameters for training the SVM classification model, and finally obtaining the trained SVM classification model as the partial discharge waveform classification model.

[0059] The overall real-time process of the above solution is as follows Figure 1 As shown, it consists of collecting partial discharge and noise waveform data with known labels 1, extracting features from the waveform data 2, training in a global optimal parameter algorithm model 3, obtaining optimal parameters 4, training in an SVM algorithm model 5, finding a method and value for distinguishing partial discharge and noise 6, obtaining an optimal SVM model 7, collecting unknown waveform data 8, partial discharge waveform types 9, 10, 11, 12, 13, 14 and noise waveform types 15, 16, 17, 18, 19, 20.

[0060] The specific process includes: firstly, collecting partial discharge and noise waveform data with known markers, then extracting features from the collected waveform data, and then sending the extracted data to the global optimal parameter algorithm model and the SVM algorithm model for fusion training. In the global optimal parameter algorithm model, through in-depth training of various characteristic waveform data and the combination of algorithm logic to continuously and automatically adjust parameters, a set of parameters that optimize the model performance is automatically adjusted. In the SVM algorithm model path, the training data automatically finds the method and value of distinguishing partial discharge and noise. The optimal parameters obtained by training the global optimal parameter algorithm model are combined with the method and value of distinguishing partial discharge and noise obtained by training the SVM algorithm model to obtain the optimal partial discharge waveform classification model parameters. For new unknown waveform data, it is input into the optimal partial discharge waveform classification model parameters, and a variety of partial discharge waveforms and noise waveforms are automatically classified by the partial discharge waveform classification model. Based on the dual algorithm technology of global optimal parameter algorithm and support vector machine SVM, the performance of the partial discharge waveform classification model is better and the accuracy is higher.

[0061] The above feature extraction specifically extracts the fatness L of the waveform data and the maximum slope K of the waveform front. max , the maximum signal-to-noise ratio D is used as the input of the global optimal parameter algorithm model and the SVM algorithm model, which makes the reciprocating training of the model more efficient and greatly reduces the impact of features with lower weights on the model.

[0062] 1) The fatness or thinness of the waveform L: It describes the steepness of the waveform. The fatness or thinness of the waveform is calculated according to formula (1), such as Figure 2 As shown, when the fatness or thinness of the waveform is greater than 25, the possibility of partial discharge is greater, otherwise, the possibility of noise is greater.

[0063] L = starts.kurtosis(data) (1)

[0064] In formula (1), starts is the call to the starts function, kurtosis is the function in the starts function, and data is the amplitude data of the waveform.

[0065] 2) Maximum slope K of the waveform front max :Automatically find the maximum slope K of the waveform front between two points in the entire waveform max , calculate the maximum slope K of the frontier according to formula (2) max ,like Figure 3 As shown, when the maximum slope of the waveform is greater than 180, the possibility of partial discharge is greater, otherwise, the possibility of noise is greater.

[0066]

[0067] In formula (2), Kmax is the maximum slope, V a is the amplitude at point a, V b is the amplitude at point b, T a is the time at point a, T b It is the time of point b.

[0068] 3) The maximum signal-to-noise difference D of the waveform is the maximum signal difference between the signal data and the noise data on the spectrum diagram. The maximum signal-to-noise difference D of the waveform is calculated according to formula (3).

[0069] D=max(S a -S c ) (3)

[0070] In formula (3), D is the maximum signal-to-noise ratio, S a is the high frequency noise amplitude, S c is the signal amplitude after removing high-frequency noise.

[0071] The above-mentioned global optimal parameter algorithm model is embedded in the SVM classification model as an optimization model. Specifically, when the global optimal parameter algorithm model and the SVM algorithm model are trained, the global optimal parameter algorithm model is continuously and automatically trained to automatically adjust the parameters that are most suitable for the SVM model, so that the accuracy of the SVM model is maximized. The global optimal parameter algorithm model greatly reduces the training time of the entire partial discharge waveform classification model, while also ensuring the robustness and performance of the model, ensuring the classification effect, and improving the classification accuracy of partial discharge and noise waveforms.

[0072] The global optimal parameter algorithm model is mainly used for data optimization of the model. Its parameters such as overfitting penalty weight parameters and feature conversion kernel parameters are automatically adjusted to make the target model optimal. During the training process, the global optimal parameter algorithm model will automatically grid search for the optimal parameters. After countless attempts and optimizations, the loss is minimized. The model will use the historical search records as experience to help the next step of parameter adjustment. The obtained parameters are passed to the SVM model for model training. The optimal value is calculated according to formula (4), and the optimal partial discharge waveform classification model is finally output. Figure 4 It means that the global optimal parameter algorithm model loops through a given range until the parameters that make the SVM model achieve the highest accuracy are found.

[0073]

[0074] In formula (4), P(A) is the probability of A occurring, and P(B) is the probability of B occurring. Assuming that event A is bound to occur, the probability of event B is inferred, and the probability of event B is increased by continuous adjustment. A means that the accuracy of the SVM model is assumed to reach 95%, and B is adjusted continuously to make A occur.

[0075] The SVM algorithm model is mainly used to classify waveform data. Its principle is to automatically find the suspected classification decision area and the absolute classification judgment area in the data, and finally obtain the method and specific value of distinguishing partial discharge and noise. The absolute classification judgment area should distinguish different types of data as much as possible to achieve the effect of data classification. The classification calculation formula (5) calculates the optimal value.

[0076]

[0077] In formula (5), an absolute classification judgment area is found in the plane, Lmax is the maximum value of the absolute classification judgment area of ​​noise and partial discharge distance, w is equivalent to the parameter a in the univariate linear equation and can be used as an adjustable parameter, i is the value of the above three characteristics, and b is the error value.

[0078] For new unknown waveform data, it can be input into the optimal partial discharge waveform classification model. The partial discharge waveform classification model can automatically classify a variety of partial discharge waveforms and noise waveforms. The results of partial discharge and noise waveform classification are as follows: Figure 5 shown.

[0079] When in use, the unknown waveform data collected from the cable line partial discharge online monitoring or live detection is input into the partial discharge waveform classification program to perform partial discharge waveform classification and directly obtain the waveform judgment result.

[0080] Example 2

[0081] Based on Example 1, this embodiment applies the method of Example 1 in series with the discrimination program based on the fqnt logic threshold and the neural network discrimination program based on the PRPD spectrum. The data collected by the cable line partial discharge online monitoring or live detection system includes characteristic value data such as frequency f, discharge amount q, pulse discharge number n and discharge time t, PRPD spectrum, and waveform data. Specifically, the following steps are included:

[0082] S101: Acquire partial discharge waveform data of the cable line;

[0083] S102: judging the partial discharge waveform data obtained in step S101 by a partial discharge discrimination method based on the fqnt logic threshold, and if it is judged to be a suspected partial discharge signal, executing step S103; if it is judged to be a noise waveform, returning to step S101;

[0084] S103: The partial discharge waveform data obtained in step S101 is judged by a neural network partial discharge discrimination method based on a PRPD spectrum. If it is judged to be a suspected partial discharge signal, step S104 is executed; if it is judged to be a noise waveform, the process returns to step S101 to continue reading new data for discrimination;

[0085] In this embodiment, if the percentage value of the collected signal output is greater than or equal to 90%, it is judged as suspected partial discharge, and the last step of partial discharge waveform classification is continued; otherwise, if the percentage value is less than 90%, it is judged as noise, and the process returns to continue reading new data for judgment;

[0086] S104: extract the fatness or thinness of the waveform, the maximum slope of the waveform front and the maximum signal-to-noise ratio of the partial discharge waveform data obtained in step S101, and input them into the partial discharge waveform classification model for classification judgment. If it is judged to be a suspected partial discharge signal, the partial discharge waveform data is judged to be a partial discharge signal; if it is judged to be a noise waveform, return to step S101.

[0087] Example 3

[0088] Based on Example 1, this embodiment further uses in parallel a partial discharge discrimination method based on fqnt logic threshold and a neural network partial discharge discrimination method based on PRPD spectrum to perform classification and judgment of partial discharge waveforms.

[0089] The corresponding process is as follows:

[0090] S201: Acquire partial discharge waveform data of the cable line;

[0091] S202: using the first scheme, the second scheme and the third scheme to judge the partial discharge waveform data obtained in step S201 respectively;

[0092] The first solution extracts the fatness and thinness of the waveform, the maximum slope of the waveform front and the maximum signal-to-noise ratio of the partial discharge waveform data obtained in step S101, and inputs them into the partial discharge waveform classification model for classification judgment;

[0093] The second solution judges the partial discharge waveform data obtained in step S101 by using a partial discharge discrimination method based on the fqnt logic threshold;

[0094] The third solution judges the partial discharge waveform data obtained in step S101 by using a neural network partial discharge discrimination method based on a PRPD spectrum;

[0095] S203: If at least two of the results in the first scheme, the second scheme, and the third scheme are determined to be suspected partial discharge signals, the partial discharge waveform data is determined to be a partial discharge signal; otherwise, the partial discharge waveform data is determined to be a noise signal.

[0096] That is, there are three results in the first step of the judgment: Result 1: If the logic threshold fqnt of the collected signal exceeds the set value, it is judged as suspected partial discharge, otherwise it is judged as noise; Result 2: If the percentage value of the PRPD spectrum neural network output of the collected signal is greater than or equal to 90%, it is judged as suspected partial discharge, otherwise, if the percentage value is less than 90%, it is judged as noise; Result 3: The cable line partial discharge waveform classification method classifies the waveform of the collected signal as partial discharge or noise. Finally, the above judgment procedures are combined to judge the three judgment results of the collected signal, and the final judgment is made whether the detection signal is partial discharge or noise. If 2 or more results are judged as partial discharge, the signal is finally judged as partial discharge, otherwise the signal is judged as noise.

[0097] Example 4

[0098] This embodiment provides a cable line partial discharge waveform classification device, including a memory and a processor, the memory stores a computer program, and the processor calls the computer program to execute the steps of the method described in Example 1, Example 2 or Example 3.

[0099] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for classifying partial discharge waveforms of cable lines, characterized in that: The following steps are involved: Collect partial discharge waveform data of the cable line, extract the fatness and thinness of the waveform, the maximum slope of the waveform front and the maximum signal-to-noise ratio, and input them into the pre-trained partial discharge waveform classification model to classify the partial discharge waveform and the noise waveform to obtain the partial discharge waveform classification result; The training process of the partial discharge waveform classification model includes: obtaining training data, inputting the training data into the SVM classification model for training, and adjusting the parameters of the SVM classification model through a global optimal parameter algorithm model to obtain optimal parameters for training the SVM classification model, and finally obtaining the trained SVM classification model as the partial discharge waveform classification model.

2. A cable line partial discharge waveform classification method according to claim 1, characterized in that: The calculation process of the maximum slope of the waveform front includes: In the entire partial discharge waveform data, the maximum slope of the waveform front between two points is found; the calculation expression of the maximum slope of the waveform front is: In the formula, is the maximum slope of the waveform front edge, is the amplitude at point a, is the amplitude at point b, is the time of point a, is the time of point b, and MAX is the maximum value of the slope of the leading edge of the waveform between the two points.

3. A cable line partial discharge waveform classification method according to claim 1, characterized in that: The maximum signal-to-noise ratio is the maximum signal difference between the signal data and the noise data on the spectrum graph, and the corresponding calculation expression is: In the formula, is the maximum signal-to-noise ratio, is the high frequency noise amplitude, is the signal amplitude after removing high-frequency noise, and max is the maximum value.

4. A cable line partial discharge waveform classification method according to claim 1, characterized in that: The fatness or thinness of the waveform is determined by Function Feature extraction.

5. A cable line partial discharge waveform classification method according to claim 1, characterized in that: During the training process, the global optimal parameter algorithm model adjusts the overfitting penalty weight parameters and feature conversion kernel parameters of the SVM classification model in a grid search manner to obtain the optimal parameters with the lowest loss.

6. A cable line partial discharge waveform classification method according to claim 1, characterized in that: The method also uses in series a partial discharge discrimination method based on an fqnt logic threshold and a neural network partial discharge discrimination method based on a PRPD spectrum to classify and judge partial discharge waveforms; The partial discharge discrimination method based on the fqnt logic threshold comprises: extracting the frequency f, the discharge amount q, the number of pulse discharges n and the discharge time t according to the partial discharge waveform data, and comparing them with the corresponding noise thresholds respectively, and if they are lower than the corresponding noise thresholds, judging them as noise waveforms; The neural network partial discharge identification method based on the PRPD spectrum includes: extracting the PRPD spectrum according to the partial discharge waveform data, so as to determine whether it is a partial discharge waveform or a noise waveform.

7. A cable line partial discharge waveform classification method according to claim 6, characterized in that: The process of using the partial discharge discrimination method based on the fqnt logic threshold and the partial discharge discrimination method based on the neural network of the PRPD spectrum in series to classify and judge the partial discharge waveform is specifically as follows: S101: Acquire partial discharge waveform data of the cable line; S102: judging the partial discharge waveform data obtained in step S101 by a partial discharge discrimination method based on the fqnt logic threshold, and if it is judged to be a suspected partial discharge signal, executing step S103; if it is judged to be a noise waveform, returning to step S101; S103: judging the partial discharge waveform data obtained in step S101 by using a neural network partial discharge discrimination method based on PRPD spectrum. If it is judged to be a suspected partial discharge signal, executing step S104; if it is judged to be a noise waveform, returning to step S101; S104: extract the fatness or thinness of the waveform, the maximum slope of the waveform front and the maximum signal-to-noise ratio of the partial discharge waveform data obtained in step S101, and input them into the partial discharge waveform classification model for classification judgment. If it is judged to be a suspected partial discharge signal, the partial discharge waveform data is judged to be a partial discharge signal; if it is judged to be a noise waveform, return to step S101.

8. A cable line partial discharge waveform classification method according to claim 1, characterized in that: The method also uses in parallel a partial discharge discrimination method based on an fqnt logic threshold and a neural network partial discharge discrimination method based on a PRPD spectrum to classify and judge partial discharge waveforms.

9. A cable line partial discharge waveform classification method according to claim 8, characterized in that: The process of using the partial discharge identification method based on the fqnt logic threshold and the partial discharge identification method based on the neural network of the PRPD spectrum in parallel to classify and determine the partial discharge waveform is specifically as follows: S201: Acquire partial discharge waveform data of the cable line; S202: using the first scheme, the second scheme and the third scheme to judge the partial discharge waveform data obtained in step S201 respectively; The first solution extracts the fatness and thinness of the waveform, the maximum slope of the waveform front and the maximum signal-to-noise ratio of the partial discharge waveform data obtained in step S101, and inputs them into the partial discharge waveform classification model for classification judgment; The second scheme judges the partial discharge waveform data obtained in step S101 by a partial discharge discrimination method based on the fqnt logic threshold; The third solution judges the partial discharge waveform data obtained in step S101 by using a neural network partial discharge discrimination method based on a PRPD spectrum; S203: If at least two of the results in the first scheme, the second scheme, and the third scheme are determined to be suspected partial discharge signals, the partial discharge waveform data is determined to be a partial discharge signal; otherwise, the partial discharge waveform data is determined to be a noise signal.

10. A cable line partial discharge waveform classification device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and the processor calls the computer program to execute the steps of any one of the methods according to claims 1 to 9.

Images