A method and device for identifying discharge type of insulator

Abstract

The present invention provides a method and device for identifying the discharge type of an insulator, which relates to the field of discharge type identification, and includes: S1: obtaining sample ultraviolet videos of an insulator at different discharge stages, and extracting pixel sequences of the discharge area of ​​each sample ultraviolet video; S2: performing multiple feature extraction on each pixel sequence to obtain a training set; S3: constructing a discharge type identification network, and training the discharge type identification network through the training set to obtain a trained discharge type identification network; S4: inputting the ultraviolet video to be identified into the trained discharge type identification network to obtain the discharge type of the insulator. The present invention forms a training set through pixel sequences, pixel average sequences, and pixel variance sequences at each discharge stage, and trains the discharge type identification network through the training set, so that the discharge type identification network automatically learns the characteristics of sequence data of different discharge types, and realizes automatic discharge identification of the collected ultraviolet video.

Description

Technical Field

[0001] The present invention relates to the field of discharge type identification, and in particular to a method and device for identifying the discharge type of an insulator. Background Art

[0002] The leakage current of contaminated porcelain insulators will increase sharply in weather conditions such as heavy fog, drizzle, light rain, snowmelt, condensation, etc. Due to the evaporation effect of Joule heat, it is easy to form a local dry area, which increases the local field strength. When it exceeds the surface ionization field strength, a surface discharge phenomenon will form. The discharge types can be roughly divided into four types from weak to strong: corona discharge, spark discharge, small arc discharge and large arc discharge. Obviously, the stronger the discharge intensity, the worse the insulation state of the insulator, and the higher the flashover probability and risk. Therefore, identifying the surface discharge type of contaminated insulators is of great practical significance for the prevention of pollution flashover of power system transmission lines.

[0003] The discharge process is accompanied by the radiation of ultraviolet light signals, some of which are located in the day-blind ultraviolet band of 240-280nm. In recent years, day-blind ultraviolet imaging detection technology has begun to be applied to the discharge detection of insulators. It converts the ultraviolet light signals generated by the discharge into image signals that can be displayed on the screen, which is very vivid and intuitive. The ultraviolet imaging method also has excellent discharge positioning capabilities and detection sensitivity far exceeding conventional methods, and has broad application prospects in engineering. However, there is still a lack of relevant methods for identifying the type of discharge through ultraviolet imaging detection videos or images. At present, the project mainly relies on manual evaluation by on-site inspection personnel. With the popularization of drone inspections, the workload of manual evaluation is large, and the accuracy of its judgment is closely related to personal experience, resulting in low consistency and accuracy of the evaluation results. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a method and device for identifying the discharge type of an insulator, so as to solve the technical problem that the existing manual evaluation leads to low consistency and accuracy of the evaluation results.

[0005] The present invention provides a method for identifying the discharge type of an insulator, comprising the steps of:

[0006] S1: Obtain sample UV videos of the insulator at different discharge stages, and extract pixel sequences of the discharge area of ​​each sample UV video;

[0007] S2: extract multiple features from each pixel sequence to obtain a training set;

[0008] S3: construct a discharge type recognition network, train the discharge type recognition network through a training set, and obtain a trained discharge type recognition network;

[0009] S4: Input the ultraviolet video to be identified into the trained discharge type recognition network to obtain the discharge type of the insulator.

[0010] Preferred:

[0011] The discharge stages include: corona discharge stage, spark discharge stage, small arc discharge stage and large arc discharge stage;

[0012] u is used to represent the number of the discharge stage. When u is 1, it represents the corona discharge stage, when u is 2, it represents the spark discharge stage, when u is 3, it represents the small arc discharge stage, and when u is 4, it represents the large arc discharge stage.

[0013] Preferably, step S1 specifically comprises:

[0014] S11: Get the sample UV video w of the discharge stage u u , the sample UV video w u After frame extraction, cropping, grayscale transformation, binarization and mathematical morphological filtering, the sample ultraviolet video w is obtained. u The discharge area image of each frame;

[0015] S12: Extracting the discharge area image L of the vth frame of the discharge stage u uv , in the discharge area image L uv The pixel value of the discharge pixel is 1, and the pixel value of the non-discharge pixel is 0. uv All pixel values ​​in are accumulated to obtain the total pixel value; where v represents the frame number;

[0016] S13: Repeat step S12 to obtain the total pixel value of each frame in the discharge stage u, and arrange the total pixel values ​​in chronological order to obtain the sample ultraviolet video w u The pixel sequence X of the discharge area u ;

[0017] S14: Repeat steps S11 to S13 to obtain a pixel sequence of the discharge area of ​​each sample ultraviolet video.

[0018] Preferably, step S2 specifically comprises:

[0019] S21: Extract pixel sequence X u ={x1, x2, x3....x n}, where n represents the total number of frames, x n Indicates the total pixel value of the nth frame;

[0020] S22: Set a window with a length of m frames, and the step length of each window sliding is 1 frame;

[0021] S23: Calculate the pixel average value y of the total pixel value in the window i and pixel variance z i The back sliding window, where i represents the number of sliding times;

[0022] S24: Repeat step S23 to calculate and obtain the pixel sequence X u The pixel average sequence Y u ={y1,y2,y3,...y n} and pixel variance sequence Z u ={z1,z2,z3,...z n};

[0023] S25: Repeat steps S21 to S24 to obtain pixel average value sequence and pixel variance sequence of each pixel sequence, and use the pixel sequence, pixel average value sequence and pixel variance sequence of each discharge stage as a training set.

[0024] Preferred:

[0025] Pixel average value y i The calculation formula is:

[0026]

[0027] Pixel variance z i The calculation formula is:

[0028]

[0029] Preferred:

[0030] The discharge type recognition network includes a data input layer, an LSTM network layer, a fully connected layer, a Sofrmax layer and a category output layer which are connected in sequence.

[0031] Preferably, step S3 specifically comprises:

[0032] S31: start the jth training;

[0033] S32: Extract the pixel sequence X of the discharge stage u in the training set u , pixel average sequence Y u and pixel variance sequence Z u ;

[0034] S33: transform the pixel sequence X u , pixel average sequence Y u and pixel variance sequence Z u Input the discharge type recognition network to obtain the predicted recognition result and loss value of the discharge stage u;

[0035] S34: obtaining a true recognition result of the discharge stage u, comparing the true recognition result with the predicted recognition result, and calculating the accuracy of the discharge stage u;

[0036] S35: repeating steps S32 to S35 to calculate and obtain the accuracy and loss value of each discharge stage;

[0037] S36: adjusting the parameter weights of the discharge type recognition network through the optimizer, setting j=j+1;

[0038] S37: Repeat steps S31 to S36 until the accuracy is greater than the first preset value and the loss value is less than the second preset value, thereby obtaining a trained discharge type recognition network.

[0039] An insulator discharge type identification device comprises: a processor and a storage medium; the processor loads and executes instructions and data in the storage medium to implement the insulator discharge type identification method.

[0040] The present invention has the following beneficial effects:

[0041] The present invention proposes a discharge type recognition network based on LSTM (Long Short Term Memory Neural Network), obtains ultraviolet videos generated by insulators in different discharge stages, and forms a training set through pixel sequences, pixel average sequences and pixel variance sequences in each discharge stage. The discharge type recognition network is trained through the training set, so that the discharge type recognition network automatically learns the characteristics of sequence data of different discharge types, realizes automatic discharge recognition of collected ultraviolet videos, and improves the recognition accuracy and robustness of discharge type recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a flow chart of a method according to an embodiment of the present invention;

[0043] Figure 2 Obtain a basic wiring schematic of the experimental system for the discharge phase;

[0044] Figure 3 is the discharge area image;

[0045] Figure 4 is a schematic diagram of a pixel sequence;

[0046] Figure 5 Frame images of sample UV videos at each discharge stage;

[0047] Figure 6 Schematic diagram of pixel sequence of discharge area of ​​each sample UV video;

[0048] Figure 7 It is a schematic diagram of a sliding window;

[0049] Figure 8 It is a schematic diagram of the pixel mean value sequence and pixel variance sequence in the corona discharge stage;

[0050] Fig. 9 It is a schematic diagram of the pixel mean value sequence and pixel variance sequence in the spark discharge stage;

[0051] Fig.10 It is a schematic diagram of the pixel average value sequence and pixel variance sequence in the small arc discharge stage;

[0052] Fig.11 Schematic diagram of pixel mean value sequence and pixel variance sequence in the large arc discharge stage;

[0053] Fig.12 The curves of accuracy and loss changing with the iteration rounds;

[0054] Fig.13 Confusion matrix results for the training set of the discharge type recognition network;

[0055] Fig.14 Confusion matrix results for the test set of the discharge type recognition network;

[0056] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0057] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0058] Reference Figure 1 The present invention provides a method for identifying the discharge type of an insulator, comprising the steps of:

[0059] S1: Obtain sample UV videos of the insulator at different discharge stages, and extract pixel sequences of the discharge area of ​​each sample UV video;

[0060] As an example:

[0061] The discharge stages include: corona discharge stage, spark discharge stage, small arc discharge stage and large arc discharge stage;

[0062] u is used to represent the number of the discharge stage. When u is 1, it represents the corona discharge stage, when u is 2, it represents the spark discharge stage, when u is 3, it represents the small arc discharge stage, and when u is 4, it represents the large arc discharge stage.

[0063] Specifically, since the LSTM deep learning model requires sample data for training, in order to obtain sample data at different discharge stages, the present invention carried out experimental research in the laboratory using an artificial climate chamber. Since the environmental humidity has a significant impact on the dirty discharge, the entire experiment was carried out in a fog chamber. The basic wiring schematic diagram of the experimental system is shown in FIG. Figure 2 shown.

[0064] The porcelain insulator is XWP2-70 commonly used in power systems, and the equivalent salt density is 0.03 mg / cm 2 , 0.05mg / cm 2 , 0.1mg / cm 2 , 0.2mg / cm 2 To indicate the first, second, third and fourth level of pollution, the ash density is selected as 2mg / cm according to the regulations of the State Grid. 2 As the discharge develops from weak to strong, the present invention divides the discharge into four stages: corona discharge, spark discharge, small arc discharge and large arc discharge.

[0065] As an example:

[0066] Step S1 is specifically as follows:

[0067] S11: Get the sample UV video w of the discharge stage u u , the sample UV video w u After frame extraction, cropping, grayscale transformation, binarization and mathematical morphological filtering, the sample ultraviolet video w is obtained. u The discharge area image of each frame;

[0068] Specifically, the process of acquiring the discharge area image is as follows:

[0069] (1) Frame extraction: The video file captured by the UV imager is imported into the computer, and the image frames are continuously extracted from the video using the video analysis software. The time interval between adjacent image frames is 40 ms, and the size of each original image frame is 784*576 pixels.

[0070] (2) Image cropping: The background image outside the discharge area in the ultraviolet image is often complex, and the discharge area often only occupies a part of the image. In order to improve the calculation speed of subsequent grayscale transformation, binarization transformation and mathematical morphological filtering, the present invention crops the area outside the discharge area. Since the discharge fluctuates over a period of time, causing the size of the image to change, the present invention uses the circumscribed rectangle of the maximum discharge image over a period of time as the cropping area.

[0071] (3) Grayscale transformation and binarization

[0072] The cropped image is an RGB color image with three color components. Since the image of the discharge area is close to white, its color information is not helpful for subsequent recognition and also increases the amount of calculation. Based on this, the present invention performs grayscale transformation on the cropped image, and the transformation formula is as follows:

[0073] Y=0.299R+0.587G+0.114B(1)

[0074] In formula (1), Y is the grayscale value, which ranges from 0 to 255; R, G, and B are the three color component values ​​of red, green, and blue of the original color image. The above algorithm is implemented using the rgb2gray function.

[0075] The grayscale value of each pixel in the grayscale image ranges from (0 to 255). The present invention also performs binarization on the grayscale image, converting each pixel into white and black. The algorithm is as follows:

[0076]

[0077] x, y represent the horizontal and vertical coordinates of the pixel in the image respectively; (x, y) represents the value of the pixel at the corresponding position. The relevant algorithm is implemented using the imbinarize function in digital image processing, and its threshold is set to 0.85.

[0078] (4) Mathematical morphology filtering

[0079] The binarized image retains the time, date, photon counting frame and other noise points in the original image, which will affect the statistics of subsequent discharge pixel values ​​if not filtered out. The invention uses the opening and closing operations of mathematical morphology to filter out noise.

[0080] The opening and closing operations are composed of a cascade of corrosion and expansion operations, and their algorithm definitions are shown in equations (3) and (4):

[0081]

[0082] In formulas (3) and (4), ο and · respectively represent opening and closing operations. Opening operations can be used to eliminate images smaller than the structural element; closing operations can be used to fill small holes in the image that are smaller than the structural element and connect adjacent images. The relevant algorithm is implemented using the imopen and imclose functions, and the resulting discharge area image is as follows: Figure 3 shown.

[0083] S12: Extracting the discharge area image L of the vth frame of the discharge stage u uv , in the discharge area image L uv The pixel value of the discharge pixel is 1, and the pixel value of the non-discharge pixel is 0. uvAll pixel values ​​in are accumulated to obtain the total pixel value; where v represents the frame number;

[0084] Specifically, after the image is filtered as described above, the remaining image is almost only the discharge area image, in which the white area has a value of "1" in the image matrix and the black area has a value of "0", so by adding up the number of "1"s in each image frame matrix, the total number of pixel values ​​of the image frame can be obtained.

[0085] S13: Repeat step S12 to obtain the total pixel value of each frame in the discharge stage u, and arrange the total pixel values ​​in chronological order to obtain the sample ultraviolet video w u The pixel sequence X of the discharge area u ;

[0086] Specifically, the pixel values ​​of the discharge area of ​​each frame in the ultraviolet video are calculated and extracted in sequence, and the discharge pixel sequence is obtained by arranging them in chronological order. Taking a video collected during the arc stage of a contaminated insulator as an example, the pixel sequence of the extracted 500-frame ultraviolet image is as follows: Figure 4 shown.

[0087] S14: Repeat steps S11 to S13 to obtain a pixel sequence of the discharge area of ​​each sample ultraviolet video.

[0088] Specifically, the frame images of the sample UV video at each discharge stage are as follows: Figure 5 As shown, the pixel sequence of the discharge area of ​​each sample UV video is as follows Figure 6 shown.

[0089] (1) Corona discharge stage: Experimental studies have shown that when the humidity reaches about 65%, corona discharge first appears in the small air gap around the insulator cap. At this time, the size of the discharge spot area in the UV image video is relatively stable, and the spot area value generally does not exceed 4000 pixels. The spot area sequence fluctuates slightly during the whole process, and no obvious large area appears. The typical UV image of this stage is as follows: Figure 5 As shown in (a), the pixel sequence of the light spot is as follows: Figure 6 (a) shown.

[0090] (2) Spark discharge stage: When the humidity reaches about 80%, brush discharge begins around the iron cap of the insulator. Its color is blue-purple. Compared with the corona discharge stage, the spot area shows an obvious increase trend, but the volatility also increases. Its typical UV image and spot area are as follows: Figure 5 As shown in (b), the pixel sequence of the light spot is as follows: Figure 6 (b) as shown.

[0091] (3) Small arc discharge: When the humidity reaches about 85%, small orange arc discharge begins to appear on the insulator. The arc is mainly concentrated around the iron cap, but generally the arc length does not exceed 2 cm. The discharge area has covered the entire insulator, and the fluctuation of the discharge spot is more intense. Its typical UV image is as follows: Figure 5 (c), the pixel sequence of the light spot is as follows Figure 6 (c) as shown.

[0092] (4) Large arc discharge stage: When the humidity reaches 90%, very bright red arcs begin to appear on the upper and lower surfaces of the insulator at the same time. When the arc is burning, the spot area in the UV image increases sharply, and after the arc is extinguished, the spot area decreases sharply. Typical UV images are as follows: Figure 5 As shown in (d), the pixel sequence of the light spot is Figure 6 (d) as shown.

[0093] S2: extract multiple features from each pixel sequence to obtain a training set;

[0094] As an embodiment, the above-mentioned only obtains the original pixel sequence of the discharge, in order to better characterize the characteristics of its discharge, further enrich the characteristics of different discharge types, and help improve the recognition accuracy of the LSTM network. To this end, the present invention proposes a method for extracting a feature sequence by sliding a window on the original area sequence, performs a correlation operation on the original sequence, and establishes a pixel average sequence and a pixel variance sequence. The basic idea of ​​the sequence extraction is shown in step S2.

[0095] Step S2 is specifically as follows:

[0096] S21: Extract pixel sequence X u ={x1, x2, x3....x n}, where n represents the total number of frames, x n Indicates the total pixel value of the nth frame;

[0097] S22: Set a window with a length of m frames, and the step length of each window sliding is 1 frame;

[0098] Specifically, a window of length m is used to slide from the beginning to the end of the original sequence step by step, with a step length of 1 each time, and then the average and variance of the m pixel data within the sliding window are calculated. To ensure that the length of the intercepted new sequence is equal to the original sequence, a sequence of length m is added to the end of the original sequence before the window is sliding. The schematic diagram of using a sliding window to intercept a sequence of length m is shown in the figure below. Figure 7 shown.

[0099] S23: Calculate the pixel average value y of the total pixel value in the window i and pixel variance z iThe back sliding window, where i represents the number of sliding times;

[0100] Specific:

[0101] Pixel average value y i The calculation formula is:

[0102]

[0103] Pixel variance z i The calculation formula is:

[0104]

[0105] S24: Repeat step S23 to calculate and obtain the pixel sequence X u The pixel average sequence Y u ={y1,y2,y3,...y n} and pixel variance sequence Z u ={z1,z2,z3,...z n};

[0106] Specifically, here Figure 6 As an example, the pixel average sequence of the corona discharge stage obtained by the above calculation is as follows Figure 8 As shown in (a), the pixel variance sequence is Figure 8 (b) shows; the pixel average sequence in the spark discharge stage is as follows Fig. 9 As shown in (a), the pixel variance sequence is Fig. 9 (b) shows; the pixel average sequence in the small arc discharge stage is as follows Fig.10 As shown in (a), the pixel variance sequence is Fig.10 (b) shows; the pixel average sequence in the large arc discharge stage is as follows Fig.11 As shown in (a), the pixel variance sequence is Fig.11 (b) as shown.

[0107] S25: Repeat steps S21 to S24 to obtain pixel average value sequence and pixel variance sequence of each pixel sequence, and use the pixel sequence, pixel average value sequence and pixel variance sequence of each discharge stage as a training set.

[0108] Specifically, the deep learning network needs to establish a sample database to train the model. Based on the above experiments, 12 groups of original discharge pixel sequences of four categories, namely corona discharge, spark discharge, small arc discharge and large arc discharge, were obtained. The length of each group was 1000 frames. Then, the mean value and variance sequence were obtained using formulas (5) and (6).

[0109] The duration of discharge of the 1000-frame data sequence is 40s. Based on the discharge mechanism, it is known that the correlation between the data points with longer intervals in the sequence is not strong over a long period of time. If the 1000-frame sequence is directly input into the LSTM network training, the recognition effect is not good. In the present invention, the 1000-frame sequence is divided into 20 groups of short sequences, each of which is 200 frames long and has a corresponding time length of 8s. In this way, a total of 240 groups of sample data can be obtained, of which 200 groups are used as training data and 40 groups are used as test data.

[0110] In order to facilitate the input of samples into the LSTM network for training, the "original discharge pixel sequence", "pixel average sequence after sliding window processing", and "pixel variance sequence after sliding window processing" of each segment sequence are stored in a matrix, each matrix size is 200*3, and the matrix is ​​stored in a cell array, and the size of the cell array is set to 200*1. In the cell array, each cell corresponds to an output category, which is marked as "corona discharge", "spark discharge", "small arc discharge" and "large arc discharge" in the present invention, and is also stored in the label cell array, whose size is also 200*1. The data establishment method of the test set is similar to that of the training set.

[0111] S3: construct a discharge type recognition network, train the discharge type recognition network through a training set, and obtain a trained discharge type recognition network;

[0112] As an example:

[0113] The discharge type recognition network includes a data input layer, an LSTM network layer, a fully connected layer, a Sofrmax layer and a category output layer which are connected in sequence.

[0114] Specific:

[0115] (1) Data input layer: The function of the data output layer is to read data from the sample database and input it into the network for training. The sample database contains 240 sets of data, 200 sets of sequence data are randomly selected as training sets, and 40 sets of data are used as test sets.

[0116] (2) Construction of LSTM network layer: Based on RNN, LSTM adds a path that runs through each operation unit, called a memory cell. After the memory cell retains or forgets the result of each operation unit, it updates the new cell state, which includes the forget gate, input gate, cell state, and output gate. Therefore, it is necessary to set the parameters one by one.

[0117] The optimizer of the LSTM model in the present invention adopts "Adam", and the execution environment is set to "gpu", that is, the GPU of the computer is used for training. Compared with CPU calculation, in the present invention, its training speed is increased by 23 times, the gradient threshold is specified as 1, the maximum number of epochs is specified as 100, the output mode is set to "last", the state activation function is set to "tanh", the gate activation function is set to "sigmoid", the number of hidden layers is set to 500, and the batch size is set to 128. The training rounds of the model are 600. The input parameters of the model are 3, that is, each sample corresponds to 3 feature sequences, namely "original discharge pixel sequence", "pixel average sequence after sliding window processing", "pixel variance sequence after sliding window processing", and the number of classifications is set to 4, namely "corona discharge", "spark discharge", "small arc discharge" and "large arc discharge".

[0118] As an example:

[0119] Step S3 is specifically as follows:

[0120] S31: start the jth training;

[0121] S32: Extract the pixel sequence X of the discharge stage u in the training set u , pixel average sequence Y u and pixel variance sequence Z u ;

[0122] S33: transform the pixel sequence X u , pixel average sequence Y u and pixel variance sequence Z u Input the discharge type recognition network to obtain the predicted recognition result and loss value of the discharge stage u;

[0123] S34: obtaining a true recognition result of the discharge stage u, comparing the true recognition result with the predicted recognition result, and calculating the accuracy of the discharge stage u;

[0124] S35: repeating steps S32 to S35 to calculate and obtain the accuracy and loss value of each discharge stage;

[0125] S36: adjusting the parameter weights of the discharge type recognition network through the optimizer, setting j=j+1;

[0126] S37: Repeat steps S31 to S36 until the accuracy is greater than the first preset value and the loss value is less than the second preset value, thereby obtaining a trained discharge type recognition network.

[0127] Specifically, after the input data is operated by LSTM, a lot of local features are generated. The fully connected layer integrates these local features by connecting to all nodes of the previous neural network to form a global feature that can classify the input data.

[0128] The Softmax layer is mainly used in deep learning and machine learning to convert the K-dimensional vector output by the previous layer (usually the output of the fully connected layer) into a probability distribution, that is, to output a K-dimensional probability vector, in which each element represents the probability that the sample belongs to the corresponding category.

[0129] After completing the above parameter settings, the accuracy and loss curves during training change with the iteration rounds as shown below: Fig.12 As shown in the figure, the analysis shows that with the increase of the number of iterative calculation steps, the accuracy shows an overall increasing trend. When the iterative calculation reaches 320 steps, the accuracy reaches 100%.

[0130] The corresponding confusion matrix results are as follows Fig.13 As shown in the figure, the analysis shows that the model accurately achieves classification and recognition of all samples in the training set, with good results.

[0131] S4: Input the ultraviolet video to be identified into the trained discharge type recognition network to obtain the discharge type of the insulator.

[0132] Specifically, 40 groups of test samples were input into the system for testing, and the results are as follows: Fig.14 As shown, it is found that its accuracy still reaches 100%, indicating that the model has good recognition accuracy.

[0133] The invention provides an insulator discharge type identification device, a processor and a storage medium.

[0134] An insulator discharge type identification device: the insulator discharge type identification device implements the insulator discharge type identification method.

[0135] Processor: The processor loads and executes the instructions and data in the storage medium to implement the insulator discharge type identification method.

[0136] Storage medium: The storage medium stores instructions and data; the storage medium is used to implement the insulator discharge type identification method.

[0137] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0138] The serial numbers of the embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. In a unit claim that lists several means, several of these means may be embodied by the same hardware item. The use of the words first, second, and third, etc. does not indicate any order and these words may be interpreted as identifiers.

[0139] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for identifying the discharge type of an insulator, characterized in that: Includes steps: S1: Obtain sample UV videos of the insulator at different discharge stages, and extract pixel sequences of the discharge area of ​​each sample UV video; S2: extract multiple features from each pixel sequence to obtain a training set; S3: construct a discharge type recognition network, train the discharge type recognition network through a training set, and obtain a trained discharge type recognition network; S4: Input the ultraviolet video to be identified into the trained discharge type recognition network to obtain the discharge type of the insulator.

2. The method for identifying the discharge type of an insulator according to claim 1, characterized in that: The discharge stages include: corona discharge stage, spark discharge stage, small arc discharge stage and large arc discharge stage; u is used to represent the number of the discharge stage. When u is 1, it represents the corona discharge stage, when u is 2, it represents the spark discharge stage, when u is 3, it represents the small arc discharge stage, and when u is 4, it represents the large arc discharge stage.

3. The method for identifying the discharge type of an insulator according to claim 2, characterized in that: Step S1 is specifically as follows: S11: Get the sample UV video w of the discharge stage u u , the sample UV video w u After frame extraction, cropping, grayscale transformation, binarization and mathematical morphological filtering, the sample ultraviolet video w is obtained. u The discharge area image of each frame; S12: Extracting the discharge area image L of the vth frame of the discharge stage u uv , in the discharge area image L uv The pixel value of the discharge pixel is 1, and the pixel value of the non-discharge pixel is 0. uv All pixel values ​​in are accumulated to obtain the total pixel value; where v represents the frame number; S13: Repeat step S12 to obtain the total pixel value of each frame in the discharge stage u, and arrange the total pixel values ​​in chronological order to obtain the sample ultraviolet video w u The pixel sequence X of the discharge area u ; S14: Repeat steps S11 to S13 to obtain a pixel sequence of the discharge area of ​​each sample ultraviolet video.

4. The method for identifying the discharge type of an insulator according to claim 2, characterized in that: Step S2 is specifically as follows: S21: Extract pixel sequence X u ={x1, x2, x3....x n }, where n represents the total number of frames, x n Indicates the total pixel value of the nth frame; S22: Set a window with a length of m frames, and the step length of each window sliding is 1 frame; S23: Calculate the pixel average value y of the total pixel value in the window i and pixel variance z i The back sliding window, where i represents the number of sliding times; S24: Repeat step S23 to calculate and obtain the pixel sequence X u The pixel average sequence Y u ={y1,y2,y3,...y n } and pixel variance sequence Z u ={z1,z2,z3,...z n }; S25: Repeat steps S21 to S24 to obtain pixel average value sequence and pixel variance sequence of each pixel sequence, and use the pixel sequence, pixel average value sequence and pixel variance sequence of each discharge stage as a training set.

5. The method for identifying the discharge type of an insulator according to claim 4, characterized in that: Pixel average value y i The calculation formula is: Pixel variance z i The calculation formula is:

6. The method for identifying the discharge type of an insulator according to claim 1, characterized in that: The discharge type recognition network includes a data input layer, an LSTM network layer, a fully connected layer, a Sofrmax layer and a category output layer which are connected in sequence.

7. The method for identifying the discharge type of an insulator according to claim 6, characterized in that: Step S3 is specifically as follows: S31: start the jth training; S32: Extract the pixel sequence X of the discharge stage u in the training set u , pixel average sequence Y u and pixel variance sequence Z u ; S33: transform the pixel sequence X u , pixel average sequence Y u and pixel variance sequence Z u Input the discharge type recognition network to obtain the predicted recognition result and loss value of the discharge stage u; S34: obtaining a true recognition result of the discharge stage u, comparing the true recognition result with the predicted recognition result, and calculating the accuracy of the discharge stage u; S35: repeating steps S32 to S35 to calculate and obtain the accuracy and loss value of each discharge stage; S36: adjusting the parameter weights of the discharge type recognition network through the optimizer, setting j=j+1; S37: Repeat steps S31 to S36 until the accuracy is greater than the first preset value and the loss value is less than the second preset value, thereby obtaining a trained discharge type recognition network.

8. An insulator discharge type identification device, characterized in that: include: A processor and a storage medium; the processor loads and executes instructions and data in the storage medium to implement the insulator discharge type identification method according to any one of claims 1 to 7.

Images