A method for analyzing a corona current pulse waveform, a computer device and a medium

Abstract

The application discloses a corona current pulse waveform analysis method, computer equipment and a medium, and relates to the technical field of deep learning. The method comprises the following steps: acquiring a plurality of original waveforms of corona current; for each original waveform, a sub-waveform is obtained by intercepting; for each sub-waveform, the sub-waveform is input into a trained neural network model to obtain a polarity recognition result of the sub-waveform; the polarity recognition result is a Trichel pulse waveform or a non-Trichel pulse waveform; for the original waveform with the polarity recognition result being the Trichel pulse waveform, a waveform analysis parameter of the original waveform is extracted. The trained neural network model can quickly and accurately identify the sub-waveform with the polarity recognition result being the Trichel pulse waveform, so that the waveform analysis parameter of the original waveform where the sub-waveform is located can be quickly extracted, and support is provided for further analysis of the influence of different working conditions on the corona current pulse waveform.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of deep learning, in particular to a corona current pulse waveform analysis method based on a deep learning algorithm, a computer device and a medium. BACKGROUND

[0002] When the surface field strength of the conductor is higher than the critical field strength at which the conductor generates corona discharge, the conductor will generate corona discharge. Corona discharge of the conductor will cause ionization of the surrounding air molecules, generating a large number of electrons and charged particles. These charged particles move at high speed under the action of electric field force, drift and diffuse from the ionosphere to the previous layer, and induce current in the conductor. Because the current pulse characteristics are different under different working conditions, by analyzing the statistical characteristics, the change trend of the current pulse under different working conditions can be obtained to provide a basis for other research. For example, if the influence of the change of air pressure on the characteristics of corona current pulse is studied, by changing the air pressure parameter, a large amount of corona current pulse data is obtained, and after processing, the change trend of the rising edge and the falling edge can be obtained, and then some rules of the corona current pulse in the process of air pressure change can be analyzed.

[0003] However, at present, the analysis method of the corona current pulse waveform is mostly manual screening, analysis and calculation of a large amount of data obtained by corona discharge experiment. Because the data volume is large, it is difficult for human to process, the accuracy of the analysis result is greatly affected by human factors, and it has great subjectivity. Therefore, it is necessary to propose a corona current pulse waveform analysis method that can quickly, accurately and efficiently analyze the corona current pulse waveform obtained in the corona discharge experiment to complete the data analysis and processing work. SUMMARY

[0004] The purpose of the present application is to provide a corona current pulse waveform analysis method, a computer device and a medium, which can quickly, accurately and efficiently analyze the corona current pulse waveform obtained in the corona discharge experiment.

[0005] To achieve the above purpose, the present application provides the following scheme:

[0006] A corona current pulse waveform analysis method, the method comprising:

[0007] Obtaining a plurality of original waveforms of corona current; the original waveforms are obtained by corona discharge experiment;

[0008] For each original waveform, taking the time point at which the maximum peak value of the original waveform is located as the center, the original waveform with a set time length is intercepted to obtain a sub waveform;

[0009] For each of the sub-waveforms, the sub-waveform is input into a trained neural network model to obtain a polarity recognition result of the sub-waveform; the polarity recognition result is a Trichel pulse waveform or a non-Trichel pulse waveform; the trained neural network model is a model trained by taking a sample sub-waveform as input and taking a sample polarity recognition result of the sample sub-waveform as a label;

[0010] For the original waveform with the polarity recognition result being a Trichel pulse waveform, a waveform analysis parameter of the original waveform is extracted; the waveform analysis parameter includes a peak value I m , a time of first arrival at 0.1I m and 0.9I m , a time of second arrival at 0.1I m and 0.9I m , a time interval of first arrival at 0.1I m and 0.9I m , and a time interval of second arrival at 0.1I m and 0.9I m .

[0011] Optionally, before the step of inputting the sub-waveform into the trained neural network model to obtain the polarity recognition result of the sub-waveform, the method further includes:

[0012] Converting the sub-waveform into an image.

[0013] Optionally, before the step of inputting the sub-waveform into the trained neural network model to obtain the polarity recognition result of the sub-waveform, the method further includes training a neural network model, specifically including:

[0014] Obtaining a data set; the data set includes a plurality of sample sub-waveforms and a sample polarity recognition result of each of the sample sub-waveforms;

[0015] Training the neural network model by using the data set to obtain a trained neural network model.

[0016] Optionally, the step of obtaining a data set specifically includes:

[0017] Obtaining a plurality of sample original waveforms;

[0018] For each of the sample original waveforms, taking a time point at which a peak value maximum of the sample original waveform is located as a center, the sample original waveform with the set time length is intercepted to obtain a sample intercepted waveform, and the sample intercepted waveform is preprocessed to obtain the sample sub-waveform;

[0019] Labeling a sample polarity recognition result of each sample subwaveform to obtain a label corresponding to each sample subwaveform.

[0020] Optionally, the pre-processing of the sample intercepted waveform specifically comprises:

[0021] Converting the sample intercepted waveform into a sample image.

[0022] Processing the sample intercepted waveform in the sample image by using a window function to make current values corresponding to both ends of the sample intercepted waveform as 0, to obtain a sample image processed by the window function.

[0023] Adding noise to the sample image processed by the window function.

[0024] Optionally, before the adding noise to the sample image processed by the window function, the method further comprises:

[0025] Performing maximum value normalization processing on current values corresponding to each sampling moment of the waveform in the sample image processed by the window function.

[0026] Optionally, the neural network model is a ViT neural network model.

[0027] Optionally, the set time length is 100 ns.

[0028] The present application also provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the computer program is executed by the processor to make the processor execute the steps of the corona current pulse waveform analysis method.

[0029] The present application also provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program is suitable for being loaded and executed by a processor to execute the corona current pulse waveform analysis method.

[0030] According to the specific embodiments of the present application, the present application discloses the following technical effects: the present application provides a corona current pulse waveform analysis method, a computer device and a medium, the method comprising: obtaining a plurality of original waveforms of corona current; the original waveforms are obtained through a corona discharge experiment; for each original waveform, taking the time point at which the peak maximum value of the original waveform is located as the center, the original waveform of a set time length is intercepted to obtain a sub waveform; for each sub waveform, the sub waveform is input into a trained neural network model to obtain a polarity recognition result of the sub waveform; the polarity recognition result is a Trichel pulse waveform or a non-Trichel pulse waveform; the trained neural network model is a model trained with sample sub waveforms as input and sample polarity recognition results of the sample sub waveforms as labels; for the original waveform with the polarity recognition result being a Trichel pulse waveform, the waveform analysis parameters of the original waveform are extracted. The trained neural network model can quickly and accurately identify the sub waveform with the polarity recognition result being a Trichel pulse waveform, so that the waveform analysis parameters of the original waveform where the sub waveform is located can be quickly extracted, thereby providing support for further analyzing the influence of different working conditions on the corona current pulse waveform. BRIEF DESCRIPTION OF DRAWINGS

[0031] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0032] Figure 1 A corona current pulse waveform analysis method flowchart is provided for the embodiments of the present application.

[0033] Figure 2 A neural network model training method flowchart is provided for the embodiments of the present application.

[0034] Figure 3 A neural network model architecture based on an attention mechanism is provided for the embodiments of the present application.

[0035] Figure 4 An attention mechanism encoder structure framework is provided for the embodiments of the present application.

[0036] Figure 5 A full connection layer diagram is provided for the embodiments of the present application.

[0037] Figure 6 A sample intercepted waveform diagram is provided for the embodiments of the present application.

[0038] Figure 7 A sample intercepted waveform diagram is provided for the embodiments of the present application. Figure 6Amplified schematic view of the intercepted sample subwaveform;

[0039] Figure 8 Fig. 1 is a structural schematic diagram of a computer device provided by the present application. DETAILED DESCRIPTION

[0040] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0041] The present application aims to provide a corona current pulse waveform analysis method, a computer device and a medium. The trained neural network model can quickly and accurately identify the subwaveform with a polarity recognition result of a Trichel pulse waveform, so that the waveform analysis parameters of the original waveform where the subwaveform is located can be quickly extracted, i.e. the rising edge, falling edge, pulse width, current amplitude and other characteristics of each corona current pulse waveform can be quickly analyzed, which greatly improves the speed and efficiency of data processing and provides support for further analyzing the influence of different working conditions on the corona current pulse waveform. The corona current waveform sequence is used as an input vector to build a deep learning network (neural network model); a typical Trichel pulse waveform sample library is established; the method of training various characteristics of the current pulse waveform by using an artificial neural network to identify the details of the waveform is used to obtain various characteristics of the corona current pulse.

[0042] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0043] As shown in Figure 1 The present application provides a corona current pulse waveform analysis method, which comprises:

[0044] S1: obtaining a plurality of original waveforms of corona current; the original waveforms are obtained through a corona discharge experiment.

[0045] S2: for each original waveform, taking the time point where the peak maximum value of the original waveform is located as the center, intercepting the original waveform with a set time length to obtain a subwaveform.

[0046] S3: inputting each of the sub-waveforms into a trained neural network model to obtain a polarity recognition result of the sub-waveform; the polarity recognition result is a Trichel pulse waveform or a non-Trichel pulse waveform; the trained neural network model is a model trained by taking a sample sub-waveform as input and taking a sample polarity recognition result of the sample sub-waveform as a label.

[0047] S4: for the original waveform with the polarity recognition result of the Trichel pulse waveform, extracting a waveform analysis parameter of the original waveform; the waveform analysis parameter includes a peak value I m , a time of first arrival at 0.1I m and 0.9I m , a time of second arrival at 0.1I m and 0.9I m , a time interval of first arrival at 0.1I m and 0.9I m , and a time interval of second arrival at 0.1I m and 0.9I m .

[0048] Before the step of inputting each of the sub-waveforms into a trained neural network model to obtain a polarity recognition result of the sub-waveform, the method further includes training a neural network model, specifically including:

[0049] obtaining a data set; the data set includes a plurality of sample sub-waveforms and a sample polarity recognition result of each of the sample sub-waveforms.

[0050] training the neural network model by using the data set to obtain a trained neural network model.

[0051] The step of obtaining a data set specifically includes:

[0052] obtaining a plurality of sample original waveforms;

[0053] for each of the sample original waveforms, taking a sample sub-waveform by taking a time point at which a peak value maximum of the sample original waveform as a center and cutting the sample original waveform of the set time length, and pre-processing the sample sub-waveform to obtain the sample sub-waveform;

[0054] annotating a sample polarity recognition result of each of the sample sub-waveforms to obtain a label corresponding to each of the sample sub-waveforms; all the sample sub-waveforms and the label corresponding to each of the sample sub-waveforms constitute the data set.

[0055] As shown in FIG. Figure 2 , the training process of the neural network model specifically includes:

[0056] Step 1: The above sample original waveform is obtained from a corona discharge experiment. First, a wire-plate electrode experimental platform is built, and a negative voltage is applied to the wire by a high-voltage negative DC power supply (model ) to cause corona discharge on the wire.

[0057] Step 2: The corona current signal generated by the corona discharge on the wire in Step 1 is measured by a low-potential measurement method. Specifically, when the wire generates corona discharge, the corona current signal is collected through a coupling capacitor and a sampling resistor. The corona current signal can be obtained by measuring the voltage signal across the sampling resistor.

[0058] Step 3: By the different discharge intensities of the corona current signal, an original waveform sample library is established, which includes a plurality of sample original waveforms. The sample original waveforms in the original waveform sample library are labeled as "1" or "0". "1" represents that the sample original waveform is a target waveform (the target waveform is a Trichel pulse waveform), and "0" represents that the sample original waveform is a non-target waveform (the non-target waveform is a non-Trichel pulse waveform).

[0059] Step 4: A sub-sequence of the waveform in the sample library is obtained. For each sample original waveform, a sample cut-off waveform with a set time length is obtained by cutting around the time point where the maximum value of the sample original waveform is located, and is stored in the sample library. In this embodiment, the set time length is 100 ns. That is, the pulse waveform of 50 ns forward and 50 ns backward from the center point of the maximum value of the sample original waveform is taken as the sample cut-off waveform. The sample cut-off waveform is shown in Figure 6 .

[0060] Step 5: For the sample cut-off waveform obtained in Step 4, the sample cut-off waveform is preprocessed in this embodiment, specifically including:

[0061] Converting the sample cut-off waveform into a sample image.

[0062] Processing the sample cut-off waveform in the sample image using a window function to make the current values corresponding to the two ends of the sample cut-off waveform be 0, to obtain a window function processed sample image.

[0063] Adding noise to the window function processed sample image.

[0064] Before adding noise to the window function processed sample image, it further includes:

[0065] Performing maximum value normalization processing on the current values corresponding to each sampling time of the waveform in the window function processed sample image.

[0066] Specifically: the waveform intercepted sub-sequence in the sample library (the waveform intercepted sub-sequence includes the above-mentioned obtained sample intercepted waveform) is windowed to ensure that the sub-sequence sample library waveform is obtained at both ends, including: locally enlarging the above-mentioned intercepted sub-waveform to convert the sub-waveform into a picture, that is, multiplying each point with the corresponding term of the window function to make the current value at both ends of the waveform 0 to obtain a two-dimensional waveform image (i.e., a sample image); then sampling the current value in the two-dimensional waveform image, taking each pixel point corresponding time as a sampling time, and performing maximum value normalization processing on the current value corresponding to each sampling time to obtain a sub-waveform sequence, and adding the sub-waveform sequence to the sample library.

[0067] After maximum value normalization processing, the embodiment artificially adds a noise signal to each normalized sub-waveform of the sub-waveform sequence in the sample library, adjusts the amplitude level of the noise signal to control the signal-to-noise ratio, and is above a specified signal-to-noise ratio SNRth, wherein SNRth is a set signal-to-noise ratio threshold, and the signal-to-noise ratio threshold is not less than 10 dB.

[0068] The waveform image after adding the noise is a sample sub-waveform, and the enlarged image of the sample sub-waveform is as shown in Figure 7 , and the label of the sample original waveform to which the sample sub-waveform belongs is taken as the label of the sample sub-waveform.

[0069] Step 6: build a deep learning network (neural network model), the structure of which is as shown in Figure 3 . Specifically, the deep learning network is a standard transformer that receives a one-dimensional token embedding sequence as input. It can autonomously explore and extract color, texture, shape and topological structure information contained therein, and has certain robustness and higher computational efficiency. In the embodiment, the deep learning network adopts a ViT (Vision Transformer) neural network model.

[0070] The ViT neural network model is a multi-layer independent network in series, which is composed of an image matrix input layer, an image block embedding module and a position embedding module, a learnable embedding module, a transformer layer and a softmax output layer. Each layer of the transformer layer is composed of multiple parallel attention mechanism modules, which are responsible for extracting features from the features calculated by the previous layer of the network, and extracting features contained in the data in this process. After concatenating these pre-layers and fully connected layers, the feature quantity calculated and output by them is sent to the softmax layer to give the recognition result.

[0071] (1) Image block embedding module. In order to process the two-dimensional waveform image, the two-dimensional waveform image x ∈ R H×W×C is reshaped into a flattened two-dimensional image fragment sequence where (H, W) is the resolution of the two-dimensional waveform image, C is the number of channels, (P, P) is the resolution of each image patch, and N = HW / P 2 is the number of patches generated, which is also the effective input sequence length of the transformer. The transformer uses a constant latent vector size D in all its layers, so this embodiment flattens the two-dimensional waveform image patches and maps them to D dimensions through a trainable linear projection (equation (1)). The output of this projection is called the image patch embedding.

[0072]

[0073] where z0is the embedding patch sequence; x class is the two-dimensional waveform image sequence; is the nth flattened two-dimensional image patch sequence; E is a learnable parameter matrix; E pos is a learnable absolute position encoding.

[0074] Position embedding module. Position information is preserved by adding position embeddings to the patch embeddings. A standard learnable one-dimensional position embedding is used, and the resulting embedding vector sequence can be used as input to the transformer encoder.

[0075] (2) Learnable embedding module. A learnable embedding is pre-embedded on the embedding patch sequence (z00= Xclass), which is used as input to the transformer encoder Output state of the end as the image representation y (the calculation formula of y is formula (4)). During pre-training and fine-tuning, there is a classifier connected to The classifier is implemented by an MLP with hidden layers during pre-training and by a single linear layer during fine-tuning.

[0076] (3) The transformer encoder (transformer layer) is composed of multiple alternating layers of multi-head self-attention and MLP blocks, and the calculation formula is shown in formulas (1), (2) and (3). Layernorm (LN) is applied before each block, and residual connection is applied after each block. The MLP contains two layers with GELU nonlinearity.

[0077] z′ l = MSA (LN (z l-1 ))+ z l-1 , l = 1... L. (2)

[0078] z l = MLP (LN (z′ l ))+ z′ l , l = 1... L. (3)

[0079]

[0080] wherein, l is the number of layers; LN() is the residual link; is the class label; L is the total number of transformer layers.

[0081] (4) Softmax layer

[0082] The full connection layer structure is shown in Figure 5 , wherein the circle represents a neuron. The number of neurons contained in the input layer is the dimension of the column vector obtained by one-dimensional stretching of the feature map obtained by convolution calculation in the preceding layer; followed by a Softmax layer, the Softmax layer contains a specified number of neurons for classification, and the Softmax layer is used to give the classification result. Figure 5 After output by the last full connection layer, the feature map is transformed into a one-dimensional feature number column. The neurons in the Softmax layer calculate the one-dimensional number column according to formula (2) to obtain the probability that the sample to be recognized belongs to each type (types include target waveforms and non-target waveforms), and take the classification with the largest probability as the recognition result.

[0083]

[0084]

[0085] wherein, Softmax(z j ) is the probability of the jth neuron in the full connection layer, z j is the output result of the jth neuron in the full connection layer (the output result is the classification type obtained by the neuron), and M is the total number of neurons contained in the full connection layer.

[0086] (5) After the input embedding vector is obtained by integrating the class vector, the image block embedding and the position coding, it can be fed into the Transformer Encoder (transformer layer). ViT is similar to CNN, which continuously passes through the Transformer Encoder composed of serial stacking of Transformer Encoder Blocks, and finally extracts the learnable class embedding vector corresponding to the feature of the class token for image classification.

[0087] (6) Fine-tuning and higher resolution

[0088] ​The neural network model is pre-trained on the large dataset ImageNet and then fine-tuned for the waveform recognition task. For this, the pre-trained prediction head is removed and a zero-initialized D x K feed-forward layer is appended, where K is the number of waveform prediction classes. Fine-tuning at a higher resolution than pre-training is usually beneficial. When inputting higher resolution images, the size of the patches is kept constant, which results in a larger effective sequence length. The ViT neural network model can handle arbitrary sequence lengths (up to memory limitations), however, the pre-trained position embeddings can no longer be meaningful. Therefore, the pre-trained position embeddings are two-dimensionally interpolated according to their position in the original image.

[0089] The sample sub-waveform is used as the input of the neural network model, and the binary classification (Trichel pulse waveform or non-Trichel pulse waveform) is used as the output. The network parameter matrix of the neural network model is trained until the set recognition accuracy is reached, and the initial stage is set to 100%. If the set iteration limit is reached and it still does not converge, the accuracy limit can be reduced by 0.1% and re-iterated training is performed.

[0090] The image data of the sample sub-waveform sub-sequence is used to train the top layer, and the specific implementation steps are as follows:

[0091] 1) Normalize the pixels of the waveform image input into the network.

[0092] 2) Remove the pre-trained VIT network top layer structure, build a new transformer layer for feature extraction and dimension reduction, and the softmax layer outputs the recognition result of the waveform polarity (the recognition result of the waveform polarity is Trichel pulse waveform or non-Trichel pulse waveform).

[0093] 3) Import the model parameters of each network trained on the ImageNet image set, and prohibit the update of the network parameters before the newly built fully connected layer during the parameter update process.

[0094] 4) Use the training set waveform graph (the sample sub-waveform image obtained by adding noise in step 5) to train the top layer weight parameters of the network, calculate the output result of the training set, calculate the error between the output class and the corresponding real class label, use the back propagation algorithm and perform network parameter update for a specified number of iterations.

[0095] The application applies VIT transfer learning based on the ImageNet data set in deep learning to the recognition of high-frequency discharge waveform polarity. Network parameters of a feature extraction part trained by the ImageNet image set are modified to adapt to weight parameter calculation of a small data set, the number of neurons of a top layer full connection layer is modified to adapt to weight parameter calculation of a small data set, and the structure of a softmax layer is modified for Trichel pulse waveform recognition. The application uses a trained neural network model to recognize waveforms, effectively accumulates experience accumulated in the early stage, and continuously expands samples through adversarial learning. The application is efficient, simple and suitable for real-time algorithm application in online monitoring.

[0096] After the above process, a trained neural network model is obtained. After a sub waveform is obtained, the sub waveform is converted into an image, and then the image of the sub waveform is input into the trained neural network model, so that a polarity recognition result of the sub waveform is obtained.

[0097] For sub waveform data with a polarity recognition result of, the embodiment extracts waveform analysis parameters of an original waveform with the polarity recognition result of Trichel pulse waveform. m , the time of the first arrival of 0.1I m and 0.9I m , the time of the second arrival of 0.1I m and 0.9I m , the time interval of the first arrival of 0.1I m and 0.9I m , and the time interval of the second arrival of 0.1I m and 0.9I m , etc.

[0098] The sampling frequency of the time domain waveform of the corona discharge current is 50 MS / s, and the specific steps of waveform analysis are as follows:

[0099] B1, determining the peak value I m of the Trichel pulse in the analyzed waveform, and calculating the current values corresponding to 0.1I m and 0.9I m .

[0100] B2, determining the time of the first arrival of 0.1I m and 0.9I m , and recording them as t1 and t2 respectively.

[0101] B3, determining the time of the second arrival of 0.9I m and 0.1I m in the same pulse waveform, and recording them as t3 and t4 respectively.

[0102] B4, calculating the rising edge tr the time interval, i.e. t2-t1, the pulse rises from the first time 0.1I m to 0.9I m the time interval, i.e. t4-t3, the pulse falls from the second time 0.9I m to 0.1I m the time interval, i.e. t4-t1, the pulse width is the time interval of twice at 0.1I m value.

[0103] The duration of the corona current pulse is 150-450 ns.

[0104] It should be noted that the above steps B1-B4, the calculated and recorded parameters are all waveform analysis parameters.

[0105] The present application applies the transfer learning in deep learning based on the ImageNet data set to the recognition of high-frequency discharge waveform polarity. The network structure used is a ViT neural network. The ViT neural network model is a multi-layer independent network serial connection splicing structure, which is composed of an image matrix input layer, an image block embedding module and a position embedding module, a learnable embedding module, a transformer layer and a softmax output layer. The network parameters of the feature extraction part of each model after ImageNet image set training are retained, the number of neurons of the top full connection layer of the network is modified to adapt to the weight parameter calculation of the small data set, and the Softmax layer structure is modified for Trichel pulse waveform recognition. The present application has the following characteristics:

[0106] 1. By artificially adding noise signals to the waveform sequence in the sample library, the signal-to-noise ratio is controlled by adjusting the amplitude level of the noise signal, and above the specified signal-to-noise ratio SNRth, the neural network can be further trained to improve the adaptability and achieve the set recognition accuracy.

[0107] 2. Import the saved model parameters into the network, input the test set data after pixel matrix normalization into the network, and obtain the recognition result of the waveform polarity.

[0108] The present application also provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the computer program is executed by the processor to make the processor execute the steps of the corona current pulse waveform analysis method.

[0109] Please refer to Figure 8 , Figure 8 is a structural schematic diagram of a computer device provided by the present application. As Figure 8As shown, the computer device 1000 can include a processor 1001, a network interface 1004 and a memory 1005, in addition, the computer device 1000 can further include a user interface 1003, and at least one communication bus 1002. Wherein the communication bus 1002 is used to realize the connection communication between the components. Wherein the user interface 1003 can include a display, a keyboard, and the optional user interface 1003 can further include a standard wired interface, a wireless interface. The network interface 1004 can optionally include a standard wired interface, a wireless interface (such as a WI-FI interface). The memory 1005 can be a high-speed RAM memory, or a non-volatile memory, for example, at least one disk storage. The memory 1005 can optionally be at least one storage device located away from the aforementioned processor 1001. For example Figure 8 As shown, the memory 1005 as a computer storage medium can include an operating system, a network communication module, a user interface module and a device control application.

[0110] In Figure 8 As shown in the computer device 1000, the network interface 1004 can provide network communication function; while the user interface 1003 is mainly used to provide an interface for user input; and the processor 1001 can be used to call the device control application stored in the memory 1005 to realize the corona current pulse waveform analysis method described in the above embodiments, which will not be repeated here.

[0111] The application further provides a computer readable storage medium, the computer readable storage medium stores a computer program, the computer program is suitable for being loaded by a processor and executing the corona current pulse waveform analysis method described in the above embodiments, which will not be repeated here.

[0112] The above program can be deployed on a computer device for execution, or deployed on multiple computer devices located in one place for execution, or on multiple computer devices distributed in multiple places and interconnected through a communication network for execution. The multiple computer devices distributed in multiple places and interconnected through a communication network can constitute a blockchain network.

[0113] The computer readable storage medium can be an internal storage unit of the computer device, for example, a hard disk or a memory of the computer device. The computer readable storage medium can also be an external storage device of the computer device, for example, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. Further, the computer readable storage medium can include both the internal storage unit and the external storage device of the computer device. The computer readable storage medium is used to store the computer program and other programs and data required by the computer device. The computer readable storage medium can also be used to temporarily store data that has been output or will be output.

[0114] The embodiments are described in a progressive manner in the specification, and each embodiment focuses on the difference from other embodiments. The same or similar parts between the embodiments can be mutually referred to. For the system disclosed by the embodiments, the description is relatively simple because it corresponds to the method disclosed by the embodiments. The relevant parts can be referred to the description of the method.

[0115] The principles and implementation manners of the present application are described by using specific examples in the specification. The above description of the embodiments is only used to help understand the method of the present application and its core idea. For those skilled in the art, the specific implementation manners and application scopes can be changed according to the idea of the present application. In conclusion, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A method of analyzing a corona current pulse waveform, characterized by, The method comprises: a plurality of original waveforms of corona current are obtained; the original waveforms are obtained through a corona discharge experiment; for each original waveform, a sub-waveform of a set time length is obtained by taking the time point at which the maximum peak value of the original waveform is located as the center; for each sub-waveform, a polarity recognition result of the sub-waveform is obtained by inputting the sub-waveform into a trained neural network model; the polarity recognition result is a Trichel pulse waveform or a non-Trichel pulse waveform; the trained neural network model is a model trained by taking a sample sub-waveform as input and taking a sample polarity recognition result of the sample sub-waveform as a label; the neural network model is a ViT neural network model, which comprises an image matrix input layer, an image block embedding module and a position embedding module, a learnable embedding module, a transformer layer, and a softmax output layer; The image patch embedding module flattens a two-dimensional waveform image x ∈ R H×W×C into a sequence of flattened two-dimensional image patches where (H, W) is the resolution of the two-dimensional waveform image, C is the number of channels, (P, P) is the resolution of each image patch, and N = HW / P 2 is the number of patches produced, flattens the two-dimensional waveform image patches, and maps them to D dimensions by equation (1). The output of the projection is called the image patch embedding: where z0is the embedded patch sequence; x class is a sequence of two-dimensional waveform images; is the nth flattened two-dimensional image patch sequence; E is a learnable parameter matrix; E pos is a learnable absolute position encoding; position embedding is added to patch embedding, a standard learnable one-dimensional position embedding is used, and the generated embedding vector sequence can be used as the input of the transformer encoder; the learnable embedding module pre-sets a learnable embedding on the embedding patch sequence, and the state at the output end of the transformer encoder is used as the image representation y; during pre-training and fine-tuning, a classifier is connected to the transformer encoder, and the classifier is implemented by an MLP with hidden layers during pre-training and by a single linear layer during fine-tuning; the transformer encoder is composed of alternating layers of multi-head self-attention and MLP blocks, and the calculation formulas are shown in formulas (2), (3) and (4); Layernorm is applied before each block, and residual connection is applied after each block; the MLP includes two layers with GELU nonlinearity: z′ l = MSA(LN(z l-1 ))+ z l-1 , l = 1...L (2); z l = MLP(LN(z' + z) + z') (2) l )+ z' l , l = 1...L (3); wherein, l is the number of layers; LN() is the residual link; is the class label; L is the total number of transformer layers; the number of neurons included in the input layer is the dimension of the column vector obtained by one-dimensional stretching of the feature map obtained by convolution calculation in the pre-stage; the input layer is followed by a Softmax layer, which includes a specified number of neurons for classification; the Softmax layer is used to give a classification result; after output by the last fully connected layer, the feature map is transformed into a one-dimensional feature number sequence, the neurons in the Softmax layer calculate the one-dimensional number sequence according to formula (5), obtain the probability that the sample to be recognized belongs to each type, and take the classification output with the maximum probability as the recognition result: wherein Softmax(z j ) is the probability of the jth neuron in the fully connected layer, z j is the output result of the jth neuron in the fully connected layer, and M is the total number of neurons contained in the fully connected layer; For the original waveform with the polarity recognition result being a Trichel pulse waveform, waveform analysis parameters of the original waveform are extracted; the waveform analysis parameters include a peak value I m , a time of first reaching 0.1I m and 0.9I m , a time of second reaching 0.1I m and 0.9I m , a time interval of first reaching 0.1I m and 0.9I m , and a time interval of second reaching 0.1I m and 0.9I m .

2. The corona current pulse waveform analysis method according to claim 1, characterized by, before the step of inputting the sub-waveform into the trained neural network model to obtain the polarity recognition result of the sub-waveform, the method further comprises: converting the sub-waveform into an image.

3. The corona current pulse waveform analysis method according to claim 1, characterized by, before the step of inputting the sub-waveform into the trained neural network model to obtain the polarity recognition result of the sub-waveform, the method further comprises training the neural network model, specifically comprising: obtaining a data set; the data set comprises a plurality of sample sub-waveforms and a sample polarity recognition result of each sample sub-waveform; training the neural network model using the data set to obtain a trained neural network model.

4. The corona current pulse waveform analysis method according to claim 3, characterized by, the step of obtaining a data set specifically comprises: acquiring a plurality of sample original waveforms; for each of the sample original waveforms, taking the sample original waveform of the set time length as the center, obtaining a sample intercepted waveform, and preprocessing the sample intercepted waveform to obtain a sample sub waveform, wherein the sample intercepted waveform is obtained at the time when the peak maximum value of the sample original waveform is located; labeling the sample polarity identification result of each sample sub waveform to obtain a label corresponding to each sample sub waveform; all sample sub waveforms and the label corresponding to each sample sub waveform constitute the data set.

5. The corona current pulse waveform analysis method according to claim 4, characterized by, The preprocessing of the sample intercepted waveform specifically includes: converting the sample intercepted waveform into a sample image; processing the sample intercepted waveform in the sample image by using a window function to make the current values corresponding to both ends of the sample intercepted waveform be 0, to obtain a sample image processed by the window function; adding noise to the sample image processed by the window function.

6. The corona current pulse waveform analysis method according to claim 5, characterized by, Before the step of adding noise to the sample image processed by the window function, the method further includes: performing maximum value normalization processing on the current value corresponding to each sampling time of the waveform in the sample image processed by the window function.

7. The corona current pulse waveform analysis method according to claim 1, characterized by, The set time length is 100 ns.

8. A computer device, comprising: The computer readable storage medium stores a computer program, and the computer program is suitable for being loaded and executed by the processor to execute the method in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is suitable for being loaded and executed by the processor to execute the method in any one of claims 1-7.

Images