Construction method of gate pole line falling identification model, identification method, equipment and system

Abstract

The invention discloses a construction method, an identification method, equipment and a system of a gate pole line falling identification model, and the construction method comprises the steps: obtaining an industrial anomaly detection image set and a historical image set corresponding to gate pole lines of a plurality of converter valve thyristors; wherein the number of images included in the industrial anomaly detection image set is far greater than the number of images included in the historical image set, and the historical image set at least comprises an image subset with a fallen gate line and an image subset with a non-fallen gate line; and carrying out iterative training on a U-Net model by adopting a meta-learning algorithm based on the industrial anomaly detection image set and the historical image set to obtain a gate line falling identification model of the converter valve thyristor. According to the invention, the accuracy and efficiency of thyristor gate line falling identification can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power electronics and high-voltage direct current transmission, and in particular to a construction method, an identification method, a device and a system for a gate line dropout identification model. Background Art

[0002] The converter valve is the core equipment of UHV DC transmission, and its healthy operation determines the safety of the entire DC project. The thyristor is a key power electronic device in the converter valve power transformation process. The gate line shedding is the main cause of thyristor failure or defect, which can easily lead to control information communication failure, making the thyristor uncontrollable and causing a reduction in redundancy.

[0003] In the related art, the gate line drop detection of thyristors usually adopts manual inspection, or uses track monitoring system and valve control equipment monitoring system; in this way, there are problems such as limited and single monitoring means, which will cause the detected thyristor state to be discontinuous, and there will be phenomena such as missed detection and false detection, which will lead to the problem of low recognition accuracy, low efficiency and susceptibility to human factors. Therefore, a high-efficiency and high-accuracy method is urgently needed to solve the problem of thyristor gate line drop defect identification. Summary of the invention

[0004] In order to solve the problems of the prior art, the present invention proposes a method for constructing a gate line dropout identification model, an identification method, a device and a system, aiming to improve the accuracy and efficiency of thyristor gate line dropout identification.

[0005] The purpose of the present invention is achieved by adopting the following technical solutions:

[0006] In one aspect, the present invention provides a method for constructing a gate line dropout identification model for a thyristor of a converter valve, the method comprising:

[0007] Acquire an industrial anomaly detection image set and a historical image set corresponding to gate lines of multiple converter valve thyristors; wherein the number of images included in the industrial anomaly detection image set is much greater than the number of images included in the historical image set, and the historical image set at least includes: an image subset of gate line detachment and an image subset of gate line non-detachment;

[0008] Based on the industrial anomaly detection image set and the historical image set, a meta-learning algorithm is used to iteratively train the U-Net model to obtain a gate line dropout recognition model for a thyristor of a converter valve.

[0009] Optionally, the industrial anomaly detection image set includes image groups of multiple different categories, and each category of image groups includes: a defect-free training image group, a defective image group, and a defect-free image; based on the industrial anomaly detection image set and the historical image set, a meta-learning algorithm is used to iteratively train the U-Net model to obtain a gate line detachment recognition model for a thyristor of a converter valve, including:

[0010] Based on the idea of ​​the meta-learning algorithm, the images in the industrial anomaly detection image set are segmented to obtain a support image set and a query image set, and the images in the historical image set are segmented to obtain a training image set including a first image and a second image and a test image set including a third image and a fourth image; wherein the image subset with gate line detachment includes: the first image and the third image, and the image subset with gate line non-detachment includes: the second image and the fourth image;

[0011] Using the support image set and the query image set, the U-Net model is subjected to multiple rounds of iterative training and performance evaluation until an intermediate U-Net model is obtained whose performance evaluation index meets preset conditions;

[0012] The training image set and the test image set are used to perform multiple rounds of iterative training and performance evaluation on the intermediate U-Net model until the gate line dropout recognition model is obtained.

[0013] Optionally, the using the support image set and the query image set to perform multiple rounds of iterative training and performance evaluation on the U-Net model until an intermediate U-Net model whose performance evaluation index meets preset conditions is obtained includes:

[0014] Determine M meta-tasks from the support image set and the query image set; wherein each meta-task includes: N*K training images and I*J query images; wherein the N*K training images refer to N types of training images in the support image set and each type of training images includes K training images; the I*J query images refer to I types of query images in the query image set and each type of query images includes J query images; N, K, I, J and M are all greater than 0, N is greater than or equal to K; I is greater than or equal to J;

[0015] The M meta-tasks are used to perform multiple rounds of iterative training and performance evaluation on the U-Net model until the intermediate U-Net model whose performance evaluation index meets the preset conditions is obtained.

[0016] Optionally, the acquiring of a historical image set corresponding to gate lines of a plurality of converter valve thyristors includes:

[0017] Acquiring an initial image set corresponding to gate lines of the plurality of thyristor converter valves;

[0018] Rotating and / or shearing each initial image in the initial image set to obtain a plurality of intermediate images corresponding to each initial image;

[0019] The initial image set and a plurality of intermediate images corresponding to each initial image are fused to obtain the historical image set.

[0020] Optionally, fusing the initial image set and a plurality of intermediate images corresponding to each initial image to obtain the historical image set includes:

[0021] The initial image set and a plurality of intermediate images corresponding to each initial image are fused to obtain an intermediate image set;

[0022] Image preprocessing is performed on each intermediate image in the intermediate image set to obtain the historical image set.

[0023] Correspondingly, the present invention also provides a system for constructing a gate line dropout identification model of a converter valve thyristor, the system comprising:

[0024] A first acquisition module is used to acquire an industrial anomaly detection image set and a historical image set corresponding to gate lines of a plurality of thyristor converter valves; wherein the number of images included in the industrial anomaly detection image set is much greater than the number of images included in the historical image set, and the historical image set at least includes: an image subset of gate line detachment and an image subset of gate line non-detachment;

[0025] The iterative training module is used to iteratively train the U-Net model based on the industrial anomaly detection image set and the historical image set using a meta-learning algorithm to obtain a gate line dropout recognition model for the thyristor of the converter valve.

[0026] Optionally, the industrial anomaly detection image set includes multiple image groups of different categories, and each category of image groups includes: a defect-free training image group, a defective image group, and a defect-free image; the iterative training module includes:

[0027] A data segmentation submodule is used to segment the images in the industrial anomaly detection image set based on the idea of ​​the meta-learning algorithm to obtain a support image set and a query image set, and to segment the images in the historical image set to obtain a training image set including a first image and a second image and a test image set including a third image and a fourth image; wherein the image subset with gate line detachment includes: the first image and the third image, and the image subset with gate line non-detachment includes: the second image and the fourth image;

[0028] A training submodule, used to perform multiple rounds of iterative training and performance evaluation on the U-Net model using the support image set and the query image set, until an intermediate U-Net model whose performance evaluation index meets preset conditions is obtained;

[0029] The testing submodule is used to use the training image set and the test image set to perform multiple rounds of iterative training and performance evaluation on the intermediate U-Net model until the gate line dropout recognition model is obtained.

[0030] Optionally, the training submodule includes:

[0031] A determination unit, configured to determine M meta-tasks from the support image set and the query image set; wherein each meta-task includes: N*K training images and I*J query images; wherein the N*K training images refer to N types of training images in the support image set and each type of training images includes K training images; the I*J query images refer to I type of query images in the query image set and each type of query images includes J query images; N, K, I, J and M are all greater than 0, N is greater than or equal to K; I is greater than or equal to J;

[0032] A training unit is used to perform multiple rounds of iterative training and performance evaluation on the U-Net model using the M meta-tasks until the intermediate U-Net model whose performance evaluation index meets the preset conditions is obtained.

[0033] Optionally, the first acquisition module includes:

[0034] A collection submodule, used for collecting an initial image set corresponding to the gate lines of the plurality of converter valve thyristors;

[0035] A processing submodule, configured to rotate and / or shear each initial image in the initial image set to obtain a plurality of intermediate images corresponding to each initial image;

[0036] The fusion submodule is used to fuse the initial image set and the multiple intermediate images corresponding to each initial image to obtain the historical image set.

[0037] Optionally, the fusion submodule includes:

[0038] a fusion unit, configured to fuse the initial image set and a plurality of intermediate images corresponding to each initial image to obtain an intermediate image set;

[0039] The preprocessing unit is used to perform image preprocessing on each intermediate image in the intermediate image set to obtain the historical image set.

[0040] On the other hand, the present invention also provides a method for identifying gate line dropout of a thyristor of a converter valve, the method comprising:

[0041] Acquire an image to be identified corresponding to a gate line of a thyristor to be identified that is in a working state;

[0042] The gate line dropout recognition model constructed by the method described above is used to recognize the image to be recognized, and a gate line dropout recognition result of the thyristor to be recognized is obtained.

[0043] Correspondingly, the present invention further provides a gate line dropout identification system for a thyristor of a converter valve, the system comprising:

[0044] A second acquisition module is used to acquire an image to be identified corresponding to a gate line of a thyristor to be identified in a working state;

[0045] The recognition module is used to recognize the image to be recognized by adopting the gate line drop recognition model constructed by the system as described above, and obtain the gate line drop recognition result of the thyristor to be recognized.

[0046] In another aspect, the present invention further provides a computing device, comprising: one or more processors;

[0047] a processor for executing one or more programs;

[0048] When the one or more programs are executed by the one or more processors, the method for constructing a gate line dropout identification model for a thyristor of a converter valve as described above and the method for identifying a gate line dropout for a thyristor of a converter valve as described above are implemented.

[0049] On the other hand, the present invention also provides a readable storage medium having an execution program stored thereon, which, when executed, implements a method for constructing a gate line dropout identification model for a thyristor of a converter valve as described in any one of the above, and a method for identifying a gate line dropout for a thyristor of a converter valve as described in any one of the above.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] The present invention provides a method and system for constructing a gate line dropout recognition model for a thyristor of a commutator valve. By means of a historical image set corresponding to the gate lines of a plurality of thyristors of commutator valves (including a gate line dropout image subset and a gate line non-dropout image subset) and an industrial anomaly detection image set, a U-Net model is trained using a meta-learning algorithm to obtain a gate line dropout recognition model for a thyristor of a commutator valve. On the one hand, by using small sample data (a historical image set corresponding to the gate lines of a plurality of thyristors of commutator valves) and large sample data (an industrial anomaly detection image set), it is possible to enhance the training of the U-Net model under small sample conditions, thereby avoiding overfitting, and further improving the recognition accuracy of the obtained gate line dropout recognition model for a thyristor of a commutator valve. On the one hand, the scheme involves a classic U-Net model, which can realize a unique downsampling and upsampling process based on the U-Net model, fully guaranteeing the detailed information of the feature map extracted by the U-Net model, so that the constructed gate line dropout recognition model for a thyristor of a commutator valve is more accurate in image edge recognition.

[0052] The present invention provides a method and system for identifying gate line dropout of a thyristor of a converter valve, which detects the image to be identified corresponding to the gate line of the thyristor of the converter valve in real time according to the gate line dropout identification model of the thyristor of the converter valve obtained by improving the U-Net model, and can identify and locate the gate line dropout defect of the thyristor more accurately and efficiently. Compared with the prior art, the gate line dropout identification model of the thyristor of the converter valve is directly used to detect the image to be identified corresponding to the gate line collected in real time in real time, which has the advantages of high automation, intelligence, strong adaptability, etc., and can be widely used in the field of power system equipment detection and fault identification.

[0053] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and are not intended to limit the technical solutions provided by the embodiments of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work, among which:

[0055] Figure 1 A schematic flow chart of a method for constructing a gate line dropout identification model for a thyristor of a converter valve provided by an embodiment of the present invention;

[0056] Figure 2 This is a schematic diagram of the network structure of the U-Net model in practical applications;

[0057] Figure 3 A schematic diagram of a framework involved in the meta-learning algorithm of the present invention;

[0058] Figure 4 A schematic diagram of a flow chart of a method for identifying gate line loss of a thyristor of a converter valve provided by an embodiment of the present invention;

[0059] Figure 5 A schematic diagram of the test data and accuracy rate corresponding to the gate line drop recognition model constructed using the meta-learning algorithm and not using the meta-learning algorithm in practical applications;

[0060] Figure 6 The experimental data-loss diagram corresponding to the gate line dropout recognition model constructed using the meta-learning algorithm and not using the meta-learning algorithm in practical applications;

[0061] Figure 7 A schematic diagram of the composition of a system for building a gate line dropout identification model for a thyristor of a converter valve provided by an embodiment of the present invention;

[0062] Figure 8 A schematic diagram of the composition of a gate line dropout identification system for a thyristor of a converter valve provided by an embodiment of the present invention;

[0063] Fig. 9 A schematic diagram of the composition of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0064] The following will describe the embodiments of the present invention with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention, not for limiting the scope of protection of the present invention.

[0065] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0066] In the following description, the terms "first\second\third" involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present invention described herein can be implemented in an order other than that illustrated or described herein.

[0067] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the technical field of the embodiments of the present invention. The terms used herein are only for the purpose of describing the embodiments of the present invention and are not intended to limit the embodiments of the present invention.

[0068] As the core equipment of UHV DC transmission, the healthy operation of the converter valve determines the safety of the entire DC project. As the key power electronic device in the converter valve power conversion process, the gate line shedding of the thyristor is the main cause of thyristor failure or defect, which can easily lead to control information communication failure, making the thyristor uncontrollable and reducing the redundancy. Therefore, it is necessary to effectively identify the thyristor gate line shedding defect to evaluate the severity of the defect, so as to avoid the sudden occurrence of converter valve failure.

[0069] In the related art, the gate line drop detection of thyristors is usually carried out by manual inspection, or by means of a track-type monitoring system and a valve-controlled equipment monitoring system; in this way, there are problems such as limited and relatively single monitoring means, which will result in discontinuous thyristor states being detected, and there will be phenomena such as missed detection and false detection, which will lead to low recognition accuracy, low efficiency and susceptibility to human factors in the gate line drop of thyristors. In addition, when a thyristor fails, it is necessary to stop running and use manual methods to find the cause of the failure, which wastes a lot of manpower, material resources and financial resources, and is inefficient. Therefore, there is an urgent need for a high-efficiency and high-accuracy method to solve the problem of identifying thyristor gate line drop defects.

[0070] As image recognition technology has been widely used in the fields of detection and monitoring, if image recognition technology is applied to the detection of thyristor gate line detachment defects, a large number of samples are required for training. However, the data collection and storage sample data in the project is limited, and there is a serious shortage of useful samples, which can easily lead to problems such as low defect recognition accuracy and overfitting in training. In view of the above shortcomings, the mainstream method of industrial sample image recognition mainly adopts transfer learning, which pre-trains the model on other rich data sets and then uses a small number of samples for transfer. However, this method still cannot meet the needs of the project in terms of recognition accuracy and target detection efficiency.

[0071] Based on the above problems, an embodiment of the present invention provides a method for constructing a gate line dropout identification model of a converter valve thyristor and a method for identifying a gate line dropout of a converter valve thyristor.

[0072] Embodiment 1:

[0073] like Figure 1 FIG. 1 is a flow chart of a method for constructing a gate line dropout identification model for a thyristor of a converter valve provided by an embodiment of the present invention. Figure 1 The steps shown in the following instructions are:

[0074] Step 101: Acquire an industrial anomaly detection image set and a historical image set corresponding to gate lines of multiple converter valve thyristors.

[0075] The number of images included in the industrial anomaly detection image set is much larger than the number of images included in the historical image set, and the historical image set at least includes: a subset of images in which the gate line has fallen off and a subset of images in which the gate line has not fallen off.

[0076] In some embodiments of the present invention, the industrial anomaly detection image set may refer to the publicly available MVTec AD industrial image set, which is generally used for anomaly detection in industrial detection, and the industrial anomaly detection image set contains more than 5,000 high-resolution images and 15 different object and texture categories. Each category includes a set of defect-free training images, a set of images with various defects, and defect-free images. Here, the MVTec AD dataset is used to train the U-Net model, so that the trained U-Net model has the ability to perform preliminary defect recognition tasks, thereby further realizing the applicability of learning on the thyristor gate line defect dataset.

[0077] In some embodiments of the present invention, multiple converter valve thyristors may be in the same converter valve or in different converter valves, and the present invention does not impose any limitation on this. At the same time, the number of images included in the historical image set corresponding to the gate lines of multiple converter valve thyristors may be determined according to actual needs, such as 100 or 200 images. Among them, the image data corresponding to the image subset corresponding to the gate line falling off and the image subset corresponding to the gate line not falling off are also determined according to actual needs. Here, multiple converter valve thyristors can be in working state and / or non-working state at the same time.

[0078] It should be noted that the thyristor of the converter valve can be any one of a common thyristor, a bidirectional thyristor, a reverse conducting thyristor, etc., and the present invention does not impose any limitation on this.

[0079] In some embodiments of the present invention, the number of images included in the industrial anomaly detection image set is much larger than the number of images included in the historical image set, for example, the number of images included in the industrial anomaly detection image set is 5,000, and the number of images included in the historical image set is 50.

[0080] It should be noted that the degree of thyristor shedding corresponding to the gate line shedding image subset may be determined according to actual needs, that is, the gate line shedding image subset may include but is not limited to: images with more serious gate line shedding, images with slight gate line shedding, etc.

[0081] In some embodiments of the present invention, the step 101 of “obtaining a historical image set corresponding to gate lines of a plurality of thyristor converter valves” may be implemented by the following steps A1 to A3:

[0082] Step A1: collecting an initial image set corresponding to the gate lines of the plurality of thyristor converter valves.

[0083] In some embodiments of the present invention, an image acquisition device deployed in a converter valve may be used to acquire an initial image set corresponding to gate lines of multiple converter valve thyristors, that is, to acquire an initial image set at the engineering site. Here, the initial image set may also include: an initial image subset corresponding to gate line detachment and an initial image subset corresponding to gate line non-detachment.

[0084] Step A2: rotating and / or shearing each initial image in the initial image set to obtain a plurality of intermediate images corresponding to each initial image.

[0085] In some embodiments of the present invention, each initial image in the initial image set may be subjected to enhancement processing such as image rotation and image shearing. Here, image rotation is the process of rotating a certain angle with a certain point on the original initial image as a fixed point to generate a new image. Image shearing is to cut an image of a certain size from a certain point on the original initial image. In the present invention, two hybrid methods (image rotation and / or image shearing) can be used to enhance the image of the initial image set; wherein, each initial image in the initial image set can be rotated by 15°, 45°, 75°, 90°, 120°, 135°, and 180°, respectively, so that each initial image is expanded from 1 to 8, thereby obtaining multiple intermediate images corresponding to each initial image, thereby achieving the enhancement of the number of images in the initial image set.

[0086] Step A3: Fusing the initial image set and a plurality of intermediate images corresponding to each initial image to obtain the historical image set.

[0087] In some embodiments of the present invention, the acquired initial image set and a plurality of intermediate images corresponding to each initial image are directly fused to obtain a historical image set, so as to enrich the number of acquired images.

[0088] In this way, the initial images in the acquired initial image set are rotated and / or sheared to enhance (enrich) the number of images in the initial image set, thereby providing certain sample data for subsequent related model training and evaluation.

[0089] In some embodiments of the present invention, the above step A3 may also be implemented by the following process:

[0090] Firstly, the initial image set and a plurality of intermediate images corresponding to each initial image are fused to obtain an intermediate image set;

[0091] Then, image preprocessing is performed on each intermediate image in the intermediate image set to obtain the historical image set.

[0092] In some embodiments of the invention, image preprocessing includes but is not limited to: denoising (high-frequency noise and interference signals in image data can be removed by a bandpass filter), normalization (using the minimum-maximum normalization method to convert data to a uniform scale to eliminate dimensional effects and ensure the stability of model training), and other operations.

[0093] In this way, the image quality in the historical image set can be further improved with the help of image preprocessing, thereby improving the accuracy of feature extraction by subsequent related models.

[0094] Step 102: Based on the industrial anomaly detection image set and the historical image set, a meta-learning algorithm is used to iteratively train the U-Net model to obtain a gate line dropout recognition model for the thyristor of the converter valve.

[0095] In some embodiments of the invention, the meta-learning algorithm is a task-based training method that enables the model to "learn to learn" by learning image generalization features through multi-task. This method can effectively solve the problems of traditional neural networks, such as overfitting caused by weak learning generalization ability of U-Net models and insufficient data samples, and can also enable the U-Net model to quickly learn other new knowledge based on the existing knowledge with the characteristic of "applying knowledge to other similar cases".

[0096] It should be noted that the U-Net model is a fully convolutional neural network with an encoder-decoder structure. It has a simple network structure and was first used to solve the problem of biomedical image segmentation when sample collection is difficult. It can maximize the extraction of more detailed features of the feature map through the context sampling process. Therefore, the present invention hopes to use its advantages to solve the problem of thyristor gate line detachment defect recognition under small sample data conditions. Here, the structural schematic diagram corresponding to the U-Net model is as follows: Figure 2 As shown, Figure 2 Take the U-Net model as an example to build the gate line drop recognition model of the thyristor of the converter valve, and then apply it to the thyristor gate line drop defect recognition. Figure 2 As shown, the network structure of the U-Net model is a U-shaped structure. Here, the corresponding input is the collected original image 201 (the image corresponding to the gate line of the thyristor to be identified), and the output is the encoded and decoded image 202 as an example.

[0097] Continue to refer Figure 2As shown, the skip connection operation ( Figure 2 The concatenation operation shown in the figure is used for feature fusion in the U-Net model; the activation function (a 3*3 convolution kernel can be used) is used for feature extraction in the U-Net model; the downsampling operation (the pooling operation in the figure can use a 2*2 convolution kernel, i.e., maximum pooling) is used to reduce the feature dimension in the U-Net model; the upsampling operation is used to restore the feature dimension in the U-Net model. Figure 2 The feature map corresponding to each operation and the number of channels of the feature map are shown accordingly.

[0098] Here, one of the most prominent features of the U-Net model is its skip connection operation, which can effectively solve the loss of details (such as boundary information) lost due to downsampling operations, thereby helping the model to better complete accurate positioning. And compared with the conventional addition operation, the skip connection operation increases the amount of calculation, but can retain more spatial information.

[0099] It should be noted that the encoding module of the U-Net model, that is, the feature extraction module, mainly includes the convolution module and the downsampling module; the decoding module mainly includes the convolution module and the upsampling module. Here, the upsampling module uses a deconvolution operation, and the deconvolution of a pixel point uses a 2*2 convolution kernel and a pixel point replication method. At the same time, the feature fusion of the U-Net model mainly splices the features together in the channel dimension to form a thicker feature, thereby realizing the cross-layer connection of the features to obtain the effect of optimizing the details, and then input the output results of the pooling layer of the encoding at different stages into the decoding stage.

[0100] In some embodiments of the present invention, the convolution module of the U-Net model uses a 3*3 convolution kernel, a padding of 0, and a stride of 1. When constructing a gate line dropout recognition model for a thyristor of a converter valve based on the U-Net model to extract features from image data, the convolution module can be used to complete feature extraction. Among them, the convolution operation corresponding to the convolution module can complete multi-dimensional data fusion and feature extraction, as shown in the following formulas (1) and (2):

[0101]

[0102]

[0103] in, represents the eigenvalue after the jth convolution of the L+1th layer, m represents the number of one-dimensional convolution kernels, and p is the size of the convolution kernel. It represents the i-th convolution eigenvalue in the c-th convolution kernel from layer L to layer L+1 when solving the j-th eigenvalue. It represents the i-th convolution weight value in the c-th convolution kernel from layer L to layer L+1 when solving the j-th eigenvalue. Indicates the bias value from layer L to layer L+1, n out is the output feature parameter size, n in is the input feature parameter size, q is the boundary size, k is the convolution kernel size, and s is the step size. σ(*) represents the activation function, and the Relu nonlinear activation function is used, as shown in formula (3):

[0104] Relu(x i )=max(0,x i ) formula (3);

[0105] Among them, x i is the output of any network layer in the U-Net model.

[0106] Pooling operation (downsampling operation or downsampling operation) is used to reduce the probability of overfitting. The pooling operation of the U-Net model is shown in formula (4):

[0107]

[0108] in, represents the j-th pooled eigenvalue of the L+1th layer, It represents the feature value after the i-th convolution of the L-th layer, r is the size of the pooling module, and MaxPool(*) represents the maximum pooling function.

[0109] In some embodiments of the present invention, a 10-layer core module (i.e. Figure 2 The model involved: upsampling five layers, downsampling five layers), and the adaptive learning rate gradient optimization algorithm: Adagrad is used as the optimization algorithm for network parameters, the corresponding learning rate is 0.001, the initial gradient accumulation is 0.1, the batch number is 16, and the number of iterations is 100, as the model training parameters, to train the U-Net model; the corresponding training parameters of the U-Net model can be referred to in Table 1 below.

[0110] Table 1 Some training parameters corresponding to the U-Net model

[0111] Training parameters value Activation Function Relu Pooling function MaxPooling Optimizer Adagrad Loss Function Softmax

[0112] Correspondingly, the gate line dropout recognition model of the converter valve thyristor obtained above can be applied to actual scenarios, that is, to realize real-time detection of the gate line image of the converter valve thyristor, so as to automatically identify and locate the gate line dropout defect of the thyristor, and output the corresponding recognition result.

[0113] In practical applications, accurate identification of thyristor gate line detachment images requires a large number of samples for training, while in actual engineering, there is a serious shortage of samples, resulting in insufficient corresponding defect recognition capabilities. Based on this, in the present invention, with the help of historical image sets corresponding to gate lines of multiple converter valve thyristors (including image subsets of gate line detachment and image subsets of gate line non-detachment) and industrial anomaly detection image sets, a meta-learning algorithm is used to train a U-Net model to obtain a gate line detachment recognition model for converter valve thyristors. On the one hand, through small sample data (historical image sets corresponding to gate lines of multiple converter valve thyristors) and large sample data (industrial anomaly detection image set), it is possible to enhance the training of the U-Net model under small sample conditions, thereby avoiding overfitting, and further improving the recognition accuracy of the obtained gate line detachment recognition model for converter valve thyristors. On the one hand, this solution involves the classic U-Net model, which can implement the unique downsampling and upsampling process inside it, fully guaranteeing the detailed information of the feature map extracted by the U-Net model, thereby making the image edge recognition more accurate, thereby improving the accuracy and efficiency of the subsequent recognition of the thyristor gate line detachment defect of the converter valve.

[0114] In some embodiments of the present invention, the industrial anomaly detection image set includes multiple different categories of image groups, and each category of image groups includes: a defect-free training image group, a defective image group, and a defect-free image; correspondingly, the above step 102 can be implemented by the following steps B1 to B3:

[0115] Step B1: Based on the idea of ​​the meta-learning algorithm, the images in the industrial anomaly detection image set are segmented to obtain a support image set and a query image set, and the images in the historical image set are segmented to obtain a training image set including a first image and a second image and a test image set including a third image and a fourth image.

[0116] The image subset with gate line detachment includes: the first image and the third image, and the image subset with gate line non-detachment includes: the second image and the fourth image.

[0117] Here, taking the preset industrial image set including 5000 images and the historical image set including 60 images as an example, the idea of ​​meta-learning algorithm is adopted, and the preset industrial image set is used as the source data set for meta-learning training, and the historical image set is used as the target data set for meta-learning testing. Here, taking the image size of 800*800 as an example, the image data segmentation is shown in Table 2:

[0118] Table 2 Image segmentation corresponding to the meta-learning algorithm

[0119]

[0120] It should be noted that the meta-learning algorithm is a method that enables the model to have meta-learning capabilities. The algorithm includes two stages: a meta-training stage and a meta-testing stage. In the meta-training stage, the model trained using the meta-learning algorithm is exposed to different meta-tasks and learns how to quickly adapt from the support image set to the query image set. In the meta-testing stage, the model trained using the meta-learning algorithm is applied to new tasks to test its performance. Correspondingly, in the present invention, the preset industrial image set can be used as the image data involved in the meta-training stage, and the historical image set can be used as the image data involved in the meta-testing stage.

[0121] Step B2: using the support image set and the query image set, perform multiple rounds of iterative training and performance evaluation on the U-Net model until an intermediate U-Net model is obtained whose performance evaluation index meets preset conditions.

[0122] Step B3: using the training image set and the test image set, perform multiple rounds of iterative training and performance evaluation on the intermediate U-Net model until the gate line dropout recognition model is obtained.

[0123] In some embodiments of the present invention, step B2 may be implemented by following steps B21 to B22:

[0124] Step B21: Determine M meta-tasks from the support image set and the query image set.

[0125] Each meta-task includes: N*K training images and I*J query images; wherein the N*K training images refer to N types of training images in the support image set and each type of training images includes K training images; the I*J query images refer to I type of query images in the query image set and each type of query images includes J query images; N, K, I, J and M are all greater than 0, N is greater than or equal to K; I is greater than or equal to J.

[0126] In some embodiments of the present invention, the meta-learning algorithm uses a small sample task as a basic unit (meta-task); for example, the image data involved in each training unit involved in the meta-training stage comes from the support image set and the query image set, that is, the image data of the meta-task involved in the meta-training stage is provided by the support image set and the query image set.

[0127] In some embodiments of the present invention, the framework diagram of the meta-learning algorithm is as follows: Figure 3 As shown, the meta-learning algorithm is divided into three parts: meta-learner 301, learner of U-Net model 302 and new task 303:

[0128] First, the corresponding training images and query images are extracted with the help of the support image set and the query image set to form a meta-learner 301 (involved in the meta-training stage here), where the meta-tasks are composed of N-way K-shot. Here, each meta-task ( Figure 3 Task 1, Task 2, ..., Task n) in the meta-task selects N types of training images from the support image set, and selects K training images for each of the N types, so that N*K training images are selected as the support data set during the training process in the meta-training phase. At the same time, each meta-task ( Figure 3 For task 1, task 2, …, task n) in the test phase of the meta-training stage, I*J query images can be selected, where I*J query images refer to I types of query images in the query image set and each type of query image includes J query images. Generally, N or I is 5, 10, 15, 20, and K or J is 1 or 5 for small sample training.

[0129] Then, meta-training of the U-Net model is implemented based on the meta-learner 301, thereby obtaining a learner 302 of the U-Net model.

[0130] Finally, the learner 302 based on the learned U-Net model learns the new task 303 to achieve the initial model parameters with strong generalization ability corresponding to the U-Net model obtained through training on the industrial anomaly detection image set, and completes the adaptation training of the new task on the test set (the historical image set corresponding to the gate lines of multiple thyristor converter valves) to achieve the purpose of quickly learning the new task.

[0131] Step B22: Using the M meta-tasks, perform multiple rounds of iterative training and performance evaluation on the U-Net model until the intermediate U-Net model whose performance evaluation index meets the preset conditions is obtained.

[0132] In some embodiments of the present invention, the determined M tasks may be used to perform multiple rounds of iterative training and corresponding performance evaluation on the U-Net model until an intermediate U-Net model whose performance evaluation index satisfies certain performance indicators is obtained. The meta-learning algorithm samples M meta-tasks (meta-training stage) in the process of acquiring meta-knowledge, which can be shown by formula (5):

[0133]

[0134] Where D source is the training sample set of M meta-tasks, are the training images and query images corresponding to the ith meta-task. Correspondingly, the U-Net model is meta-trained on the industrial anomaly detection image set to optimize the model parameters, as shown in formula (6):

[0135] θ *=argmaxlogp(θ|D source ) formula (6);

[0136] Among them, θ * is the optimal parameter set of the U-Net model after training, and θ is the model parameter set of the U-Net model.

[0137] In this way, by performing multiple rounds of iterative training and performance evaluation on the U-Net model based on M meta-tasks determined from the support image set and the query image set, the U-Net model can be exposed to different types of meta-tasks, thereby obtaining an intermediate U-Net model that learns how to quickly adapt from the industrial anomaly detection image set to the historical image set, so that the intermediate U-Net model can quickly learn new tasks in the future.

[0138] Step B3: using the training image set and the test image set, perform multiple rounds of iterative training and performance evaluation on the intermediate U-Net model until the gate line dropout recognition model is obtained.

[0139] In some embodiments of the present invention, in the meta-testing phase corresponding to the meta-learning algorithm, a training image set and a test image set (obtained by segmenting the historical image set) can be used to perform multiple rounds of iterative training and performance evaluation on the intermediate U-Net model until a gate line dropout recognition model is obtained.

[0140] Correspondingly, as described above, in the meta-testing phase, the meta-learning algorithm can determine Q test tasks from the training image set and the test image set, which can be shown in formula (7):

[0141]

[0142] Where D target is the test sample set of Q test tasks, are the training image set and test image set corresponding to the i-th test task.

[0143] Here, the meta-training phase and meta-testing phase involved in iterative training of the U-Net model using the meta-learning algorithm, the loss function involved and the evaluation index corresponding to the performance evaluation, respectively, can be determined according to actual needs. That is, the loss function involved in the meta-training phase and the meta-testing phase can be the same or different, and the evaluation index of the performance evaluation of the corresponding model can be the same or different.

[0144] Here, the loss function Loss involved in the multiple rounds of iterative training of the U-Net model by the meta-learning algorithm in the meta-training stage is taken as an example, which is the average loss of all M meta-tasks, as shown in formula (8):

[0145]

[0146] Among them, l i is the loss value of the i-th meta-task.

[0147] In some embodiments of the present invention, when the performance evaluation indicators include at least accuracy, precision and recall, the performance evaluation involved satisfies the preset conditions and may include the following:

[0148] The accuracy is greater than or equal to the accuracy threshold;

[0149] The accuracy is greater than or equal to the accuracy threshold;

[0150] The recall is greater than or equal to the recall threshold.

[0151] In order to evaluate the performance of the classification results (both in the meta-training stage and the meta-testing stage), evaluation indicators including Accuracy, Precision and Recall are used. The calculation is as follows:

[0152]

[0153] Among them, TP represents the number of samples correctly evaluated as positive, TN represents the number of samples correctly estimated as negative, FP represents the number of samples incorrectly evaluated as positive, and FN represents the number of samples incorrectly evaluated as positive.

[0154] In this way, the U-Net model is meta-trained with the support image set and the query image set in turn using the meta-learning algorithm to obtain an intermediate U-Net model, and then the intermediate U-Net model is meta-tested with the training image set and the test image set. On the one hand, it can achieve the transfer from the industrial anomaly detection image set to the new task, that is, the learning of the image corresponding to the gate line of the thyristor of the converter valve, so as to obtain the gate line detachment recognition model of the thyristor of the converter valve. On the other hand, it can solve the problems of insufficient generalization ability and overfitting on small sample data sets, so as to improve the recognition accuracy of the gate line detachment recognition model of the thyristor of the converter valve.

[0155] Embodiment 2:

[0156] Correspondingly, an embodiment of the present invention further provides a method for identifying gate line loss of a thyristor of a converter valve, such as Figure 4 FIG. 1 is a flow chart of a method for identifying gate line loss of a thyristor of a converter valve provided by an embodiment of the present invention. Figure 4 The steps shown in the following instructions are:

[0157] Step 401 : Acquire an image to be identified corresponding to a gate line of a thyristor to be identified that is in a working state.

[0158] Step 402: using a gate line dropout recognition model constructed by any of the above methods to recognize the image to be recognized, and obtaining a gate line dropout recognition result of the thyristor to be recognized.

[0159] In some embodiments of the present invention, an image acquisition device may be deployed in a converter valve to acquire real-time images of gate lines of thyristors to be identified that are in working state inside the converter valve.

[0160] It should be noted that the number of thyristors to be identified may be one, or two or more. The present invention is described by taking the case where the image to be identified includes one thyristor to be identified as an example.

[0161] In some embodiments of the present invention, the gate line dropout recognition model of the thyristor of the converter valve constructed in any of the above embodiments can be used to perform abnormal recognition on the image to be recognized, so as to obtain the dropout recognition result of the gate line of the thyristor to be recognized. Here, the dropout recognition result includes but is not limited to: dropout, no dropout, severe dropout, slight dropout, etc. The present invention does not make any limitation to this.

[0162] In this way, according to the gate line drop recognition model of the converter valve thyristor obtained by improving the U-Net model, the image to be identified corresponding to the gate line of the converter valve thyristor to be identified is detected in real time, which can accurately and efficiently identify and locate the gate line drop defect of the thyristor. And compared with the prior art, the gate line drop recognition model of the converter valve thyristor is directly used to detect the image to be identified corresponding to the gate line collected in real time in real time, which has the advantages of high automation, intelligence, strong adaptability, etc., so it can be widely used in the field of power system equipment detection and fault identification.

[0163] Referring to the above description, the present invention also provides relevant actual experimental data, that is, the following are some experimental data corresponding to the gate line dropout identification model 1 of the thyristor assembly of the converter valve constructed by the meta-learning algorithm and the gate line dropout identification model 2 of the thyristor assembly of the converter valve constructed without the meta-learning algorithm, and can also be referred to Figure 5 and Figure 6 As shown in the figure, they are the accuracy and loss of gate line dropout recognition model 1 and gate line dropout recognition model 2 in their respective training epochs; 501 is the recognition accuracy corresponding to gate line dropout recognition model 1, 502 is the recognition accuracy corresponding to gate line dropout recognition model 2; 601 is the loss corresponding to gate line dropout recognition model 1, 602 is the loss corresponding to gate line dropout recognition model 2. Figure 5 and Figure 6It can be concluded that the accuracy and loss corresponding to the gate line dropout recognition model 2 are unstable, while the convergence speed of the model corresponding to the gate line dropout recognition model 1 is faster and more stable.

[0164] At the same time, in the actual experimental stage, 60 images corresponding to the thyristor gate lines are used to verify the gate line off recognition model 1 and the gate line off recognition model 2 respectively. The corresponding data are shown in Table 3. It can be concluded that the gate line off recognition model 1, that is, the use of the meta-learning algorithm to train the U-Net model, can improve the accuracy of model recognition, and the convergence time is shortened.

[0165] Based on this, this paper conducts comparative analysis on the model recognition methods using meta-learning algorithm and those not using meta-learning algorithm. The experimental results show that under the scale of a small sample set, the use of meta-learning algorithm can improve the recognition rate of thyristor gate line detachment defects (recognition of 60 defective images, including 15 defective images), and the accuracy rate as shown in Table 3 below can reach 100%, which has great guiding significance in engineering applications. It can also be further shown that in the construction method and recognition method of the gate line detachment recognition model of the thyristor of the converter valve provided by the present invention, it makes full use of the meta-training method suitable for the small sample training process to construct an improved U-Net model, and with the help of the downsampling and upsampling process inside the U-Net model, the detailed information of the feature map extracted in the U-Net model is fully guaranteed, so that the image edge recognition is more accurate, and the model training is enhanced under the condition of small sample data, avoiding the overfitting problem, so as to improve the generalization ability of the model.

[0166] Table 3 Test results

[0167] Meta-Learning Sample size Correct quantity Evaluation time(s) Accuracy (%) Gate line drop detection model 2 60 54 0.95 90 Gate line drop detection model 1 60 60 0.43 100

[0168] In addition, as shown in Table 4, the traditional method is used to perform training and testing on this data set, and the method provided by the present invention is compared with the parameters such as model performance evaluation, which can further prove that the gate line dropout identification model for the thyristor of the converter valve constructed by the present invention can achieve the best effect.

[0169] Table 4 Model comparison

[0170] Model Accuracy Accuracy Recall Support Vector Machine 0.874 0.852 0.862 Bayes' Theorem 0.874 0.883 0.835 Convolutional Neural Networks 0.883 0.871 0.932 The present invention 1 0.991 0.993

[0171] Embodiment 3:

[0172] Based on the same inventive concept, the present invention also provides a system for constructing a gate line dropout identification model for a thyristor of a converter valve, such as Figure 7 FIG. 7 is a schematic diagram of a system for constructing a gate line dropout identification model for a thyristor of a converter valve provided by an embodiment of the present invention. The system 700 includes:

[0173] The first acquisition module 701 is used to acquire an industrial anomaly detection image set and a historical image set corresponding to gate lines of multiple converter valve thyristors; wherein the number of images included in the industrial anomaly detection image set is much larger than the number of images included in the historical image set, and the historical image set at least includes: an image subset of gate line detachment and an image subset of gate line non-detachment;

[0174] The iterative training module 702 is used to iteratively train the U-Net model based on the industrial anomaly detection image set and the historical image set using a meta-learning algorithm to obtain a gate line dropout recognition model for the thyristor of the converter valve.

[0175] In some embodiments of the present invention, the industrial anomaly detection image set includes image groups of multiple different categories, and each category of image groups includes: a defect-free training image group, a defective image group, and a defect-free image; the iterative training module 702 includes: a data segmentation submodule, which is used to segment the images in the industrial anomaly detection image set based on the idea of ​​the meta-learning algorithm to obtain a support image set and a query image set, and to segment the images in the historical image set to obtain a training image set including a first image and a second image and a test image set including a third image and a fourth image; wherein the image subset with gate line detachment includes: the first image and the third image, and the image subset with gate line non-detachment includes: the second image and the fourth image; a training submodule, which is used to use the support image set and the query image set to perform multiple rounds of iterative training and performance evaluation on the U-Net model until an intermediate U-Net model whose performance evaluation index meets preset conditions is obtained; a testing submodule, which is used to use the training image set and the test image set to perform multiple rounds of iterative training and performance evaluation on the intermediate U-Net model until the gate line detachment recognition model is obtained.

[0176] In some embodiments of the present invention, the training submodule includes: a determination unit, used to determine M meta-tasks from the support image set and the query image set; wherein each meta-task includes: N*K training images and I*J query images; wherein the N*K training images refer to N types of training images in the support image set and each type of training images includes K training images; the I*J query images refer to I type of query images in the query image set and each type of query images includes J query images; N, K, I, J and M are all greater than 0, N is greater than or equal to K; I is greater than or equal to J; a training unit, used to use the M meta-tasks to perform multiple rounds of iterative training and performance evaluation on the U-Net model until the intermediate U-Net model whose performance evaluation index meets the preset conditions is obtained.

[0177] In some embodiments of the present invention, the first acquisition module 701 includes: an acquisition submodule, used to acquire an initial image set corresponding to the gate lines of the multiple thyristor converters; a processing submodule, used to rotate and / or shear each initial image in the initial image set to obtain a plurality of intermediate images corresponding to each initial image; and a fusion submodule, used to fuse the initial image set and the plurality of intermediate images corresponding to each initial image to obtain the historical image set.

[0178] In some embodiments of the present invention, the fusion submodule includes: a fusion unit, used to fuse the initial image set and multiple intermediate images corresponding to each initial image to obtain an intermediate image set; a preprocessing unit, used to perform image preprocessing on each intermediate image in the intermediate image set to obtain the historical image set.

[0179] It should be noted that the description of the construction system of the gate line dropout identification model of the converter valve thyristor is similar to the description of the construction method embodiment of the gate line dropout identification model of the converter valve thyristor, and has similar beneficial effects as the method embodiment. For technical details not disclosed in the system embodiment of the present invention, please refer to the description of the method embodiment of the present invention for understanding.

[0180] Embodiment 4:

[0181] Correspondingly, the present invention also provides a gate line dropout identification system for a thyristor of a converter valve, such as Figure 8 FIG. 8 is a schematic diagram of a gate line drop identification system for a thyristor of a converter valve provided by an embodiment of the present invention. The system 800 includes:

[0182] The second acquisition module 801 is used to acquire the image to be identified corresponding to the gate line of the thyristor to be identified in the working state;

[0183] The identification module 802 is used to use the gate line dropout identification model constructed by any of the above-mentioned systems to identify the image to be identified and obtain a gate line dropout identification result of the thyristor to be identified.

[0184] It should be noted that the description of the corresponding embodiment of the gate line dropout identification system of the converter valve thyristor is similar to the description of the gate line dropout identification method embodiment of the converter valve thyristor, and has similar beneficial effects as the method embodiment. For technical details not disclosed in the system embodiment of the present invention, please refer to the description of the method embodiment of the present invention for understanding.

[0185] Embodiment 5:

[0186] Based on the same inventive concept, Fig. 9As shown, the present invention also provides an electronic device, which may be a computer device, a single-chip device, an intelligent mobile device, etc. The electronic device in this embodiment may include a processor 910, a memory 920, a transceiver component 930, etc. The processor 910, the memory 920 and the transceiver component 930 are connected via a bus 940; the memory 920 can be used to store an execution program, and an exemplary execution program may include instructions; the processor 910 is used to execute the instructions stored in the memory. The memory 920 can also be used to store data, which can be called and / or modified when executing instructions.

[0187] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, and is specifically suitable for loading and executing one or more instructions in a storage medium to implement corresponding method flows or corresponding functions, so as to implement the method for constructing a gate line dropout identification model for a thyristor of a commutator valve in the above-mentioned embodiment, or the steps of the gate line dropout identification method for a thyristor of a commutator valve.

[0188] Embodiment 6:

[0189] Based on the same inventive concept, the present invention also provides a readable storage medium, specifically an electronic device readable storage medium (Memory), which is a memory device in the electronic device for storing programs and data. It can be understood that the storage medium here can include both the built-in storage medium in the electronic device and the extended storage medium supported by the electronic device. The storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions can be one or more execution programs (including program codes). It should be noted that the storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. The processor loads and executes one or more instructions stored in the storage medium to implement the method for constructing the gate line dropout identification model of the thyristor of the converter valve in the above embodiment, or the steps of the gate line dropout identification method of the thyristor of the converter valve.

[0190] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0191] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0192] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0193] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0194] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for constructing a gate line dropout identification model for a thyristor of a converter valve, characterized in that: The method comprises: Acquire an industrial anomaly detection image set and a historical image set corresponding to gate lines of multiple converter valve thyristors; wherein the number of images included in the industrial anomaly detection image set is much greater than the number of images included in the historical image set, and the historical image set at least includes: an image subset of gate line detachment and an image subset of gate line non-detachment; Based on the industrial anomaly detection image set and the historical image set, a meta-learning algorithm is used to iteratively train the U-Net model to obtain a gate line dropout recognition model for a thyristor of a converter valve.

2. The method according to claim 1, characterized in that The industrial anomaly detection image set includes multiple different categories of image groups, and each category of image groups includes: a defect-free training image group, a defective image group, and a defect-free image; The method of iteratively training the U-Net model based on the industrial anomaly detection image set and the historical image set using a meta-learning algorithm to obtain a gate line dropout recognition model for a thyristor of a converter valve includes: Based on the idea of ​​the meta-learning algorithm, the images in the industrial anomaly detection image set are segmented to obtain a support image set and a query image set, and the images in the historical image set are segmented to obtain a training image set including a first image and a second image and a test image set including a third image and a fourth image; wherein the image subset with gate line detachment includes: the first image and the third image, and the image subset with gate line non-detachment includes: the second image and the fourth image; Using the support image set and the query image set, the U-Net model is subjected to multiple rounds of iterative training and performance evaluation until an intermediate U-Net model is obtained whose performance evaluation index meets preset conditions; The training image set and the test image set are used to perform multiple rounds of iterative training and performance evaluation on the intermediate U-Net model until the gate line dropout recognition model is obtained.

3. The method according to claim 2, characterized in that The method of using the support image set and the query image set to perform multiple rounds of iterative training and performance evaluation on the U-Net model until an intermediate U-Net model whose performance evaluation index meets preset conditions is obtained includes: Determine M meta-tasks from the support image set and the query image set; wherein each meta-task includes: N*K training images and I*J query images; wherein the N*K training images refer to N types of training images in the support image set and each type of training images includes K training images; the I*J query images refer to I types of query images in the query image set and each type of query images includes J query images; N, K, I, J and M are all greater than 0, N is greater than or equal to K; I is greater than or equal to J; The M meta-tasks are used to perform multiple rounds of iterative training and performance evaluation on the U-Net model until the intermediate U-Net model whose performance evaluation index meets the preset conditions is obtained.

4. The method according to claim 1, characterized in that: The step of obtaining a historical image set corresponding to gate lines of a plurality of thyristor converter valves comprises: Acquiring an initial image set corresponding to gate lines of the plurality of thyristor converter valves; Rotating and / or shearing each initial image in the initial image set to obtain a plurality of intermediate images corresponding to each initial image; The initial image set and a plurality of intermediate images corresponding to each initial image are fused to obtain the historical image set.

5. The method according to claim 4, characterized in that The step of fusing the initial image set and a plurality of intermediate images corresponding to each initial image to obtain the historical image set includes: The initial image set and a plurality of intermediate images corresponding to each initial image are fused to obtain an intermediate image set; Image preprocessing is performed on each intermediate image in the intermediate image set to obtain the historical image set.

6. A system for constructing a gate line dropout identification model for a thyristor of a converter valve, characterized in that: The system comprises: A first acquisition module is used to acquire an industrial anomaly detection image set and a historical image set corresponding to gate lines of a plurality of thyristor converter valves; wherein the number of images included in the industrial anomaly detection image set is much greater than the number of images included in the historical image set, and the historical image set at least includes: an image subset of gate line detachment and an image subset of gate line non-detachment; The iterative training module is used to iteratively train the U-Net model based on the industrial anomaly detection image set and the historical image set using a meta-learning algorithm to obtain a gate line dropout recognition model for the thyristor of the converter valve.

7. The system according to claim 6, characterized in that The industrial anomaly detection image set includes multiple different categories of image groups, each category of image groups includes: a defect-free training image group, a defective image group and a defect-free image; the iterative training module includes: A data segmentation submodule is used to segment the images in the industrial anomaly detection image set based on the idea of ​​the meta-learning algorithm to obtain a support image set and a query image set, and to segment the images in the historical image set to obtain a training image set including a first image and a second image and a test image set including a third image and a fourth image; wherein the image subset with gate line detachment includes: the first image and the third image, and the image subset with gate line non-detachment includes: the second image and the fourth image; A training submodule, used to perform multiple rounds of iterative training and performance evaluation on the U-Net model using the support image set and the query image set, until an intermediate U-Net model whose performance evaluation index meets preset conditions is obtained; The testing submodule is used to use the training image set and the test image set to perform multiple rounds of iterative training and performance evaluation on the intermediate U-Net model until the gate line dropout recognition model is obtained.

8. The system according to claim 7, characterized in that The training submodule comprises: A determination unit, configured to determine M meta-tasks from the support image set and the query image set; wherein each meta-task includes: N*K training images and I*J query images; wherein the N*K training images refer to N types of training images in the support image set and each type of training images includes K training images; the I*J query images refer to I type of query images in the query image set and each type of query images includes J query images; N, K, I, J and M are all greater than 0, N is greater than or equal to K; I is greater than or equal to J; A training unit is used to perform multiple rounds of iterative training and performance evaluation on the U-Net model using the M meta-tasks until the intermediate U-Net model whose performance evaluation index meets the preset conditions is obtained.

9. The system according to claim 6, characterized in that The first acquisition module includes: A collection submodule, used for collecting an initial image set corresponding to the gate lines of the plurality of converter valve thyristors; A processing submodule, configured to rotate and / or shear each initial image in the initial image set to obtain a plurality of intermediate images corresponding to each initial image; The fusion submodule is used to fuse the initial image set and the multiple intermediate images corresponding to each initial image to obtain the historical image set.

10. The system according to claim 9, characterized in that The fusion submodule comprises: a fusion unit, configured to fuse the initial image set and a plurality of intermediate images corresponding to each initial image to obtain an intermediate image set; The preprocessing unit is used to perform image preprocessing on each intermediate image in the intermediate image set to obtain the historical image set.

11. A method for identifying gate line dropout of a thyristor of a converter valve, characterized in that: The method comprises: Acquire an image to be identified corresponding to a gate line of a thyristor to be identified that is in a working state; The gate line dropout recognition model constructed by the method according to any one of claims 1 to 5 is used to recognize the image to be recognized, so as to obtain a gate line dropout recognition result of the thyristor to be recognized.

12. A gate line drop identification system for a thyristor of a converter valve, characterized in that: The system comprises: A second acquisition module is used to acquire an image to be identified corresponding to a gate line of a thyristor to be identified in a working state; The identification module is used to identify the image to be identified by using the gate line dropout identification model constructed by the system according to any one of claims 6 to 10, and obtain a gate line dropout identification result of the thyristor to be identified.

13. An electronic device, characterized in that: include: at least one processor and memory; The memory and the processor are connected via a bus; The memory is used to store one or more programs; When the one or more programs are executed by the at least one processor, the method for constructing a gate line dropout identification model for a converter valve thyristor as described in any one of claims 1 to 5 and the method for identifying a gate line dropout for a converter valve thyristor as described in claim 6 are implemented.

14. A readable storage medium, characterized in that: An execution program is stored thereon, and when the execution program is executed, the method for constructing a gate line dropout identification model for a thyristor of a converter valve as described in any one of claims 1 to 5 and the method for identifying a gate line dropout for a thyristor of a converter valve as described in claim 6 are implemented.

Images