Current fault type judgment method and device, storage medium and electronic equipment

Abstract

The invention discloses a current fault type judgment method and device, a storage medium and electronic equipment. Relates to the technical field of current detection. The method comprises the following steps: receiving a three-phase current signal after a fault occurs; performing image conversion on the three-phase current signal to obtain an image to be subjected to fault judgment; performing semantic segmentation processing on the image to be subjected to fault judgment by adopting a pre-trained semantic segmentation model to obtain a target image comprising a target fault waveform; and judging the fault type of the target fault waveform to obtain a current fault type judgment result. According to the method, the current fault type judgment accuracy can be improved, and the labor cost is saved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of current detection, in particular to a current fault type judgment method and device, a storage medium and an electronic device. BACKGROUND

[0002] When a single-phase ground fault occurs in a small current grounding system, the zero sequence current of each line contains serious noise components and presents complex nonlinearity and non-stationarity, and the signal characteristics of the fault point are also very weak, which brings difficulties to fault line selection. Actual data is difficult to obtain, physical modeling is difficult, measurement devices have detection errors, sampling times are not synchronized, and transient signals under different grounding fault conditions also present large differences in energy distribution and spectral characteristics, and the fault mechanism is complex, and the high-resistance grounding fault signal is even weaker. These factors result in low accuracy and poor reliability of the existing fault line selection method of the distribution network. SUMMARY

[0003] Therefore, the present application provides a current fault type judgment method and device, a storage medium and an electronic device, which mainly aims to solve the problem of low accuracy and poor reliability of current fault type detection.

[0004] To solve the above problems, the present application provides a current fault type judgment method, comprising: receiving a three-phase current signal after a fault occurs; performing image conversion on the three-phase current signal to obtain a to-be-fault-judgment image; performing semantic segmentation processing on the to-be-fault-judgment image by using a pre-trained semantic segmentation model to obtain a target image including a target fault waveform; judging the fault type of the target fault waveform to obtain a current fault type judgment result.

[0005] Optionally, before performing semantic segmentation processing on the to-be-fault-judgment image by using a pre-trained semantic segmentation model, the method further comprises: constructing the semantic segmentation model; The construction of the semantic segmentation model specifically comprises: obtaining a plurality of historical fault waveform images of different fault types; labeling each of the historical fault waveform images to obtain a label image set, and the labeled label data includes background pixel identification, non-fault waveform pixel identification, and fault waveform pixel identification; performing model training on a preset neural network model with an attention mechanism by using the historical fault waveform images and the label image set to obtain the semantic segmentation model; The preset neural network model comprises an input layer, a backbone network layer, an attention module, a decoder, and an output layer.

[0006] Optionally, the pre-trained semantic segmentation model is used to perform semantic segmentation processing on the image to be fault judged to obtain a target image comprising a target fault waveform, and the processing specifically comprises: The input layer of the semantic segmentation model is used to perform standardization processing on the image to be fault judged to obtain a first feature image; The backbone network layer of the semantic segmentation model is used to perform feature extraction on the first feature image to obtain a second feature image, wherein the second feature image comprises a first up-sampling image and a second up-sampling image for up-sampling feature extraction by a decoder of the semantic segmentation model, and a high-level feature image for feature extraction by an attention module of the semantic segmentation model; The attention module of the semantic segmentation model is used to perform channel weighting processing on the high-level feature image to obtain a third feature image; The decoder of the semantic segmentation model is used to perform feature fusion processing on the first up-sampling image, the second up-sampling image, and the third feature image to obtain a predicted segmentation image comprising a target fault waveform; A preset activation function is used to activate the predicted segmentation image in the channel dimension to obtain a target image comprising a target fault waveform.

[0007] Optionally, the backbone network layer comprises a plurality of convolutional pooling layers, and the backbone network layer of the semantic segmentation model is used to perform feature extraction on the first feature image to obtain a second feature image, and the processing specifically comprises: The plurality of convolutional pooling layers are used to perform feature extraction on the first feature image one by one, and the output of a previous convolutional pooling layer is used as the input of a subsequent convolutional pooling layer to obtain the second feature image of a bottom convolutional pooling output.

[0008] Optionally, the attention module of the semantic segmentation model is used to perform channel weighting processing on the high-level feature image to obtain a third feature image, and the processing specifically comprises: A global average pooling operation is performed on the high-level feature image to obtain a global importance index of each channel; The global importance index is calculated to obtain a scaling factor of each channel; The scaling factor and the high-level feature image are used to perform channel weighting processing to obtain the third feature image.

[0009] Optionally, the decoder using the semantic segmentation model performs feature fusion processing on the first up-sampling image, the second up-sampling image and the third feature image to obtain a predicted segmentation image including a target fault waveform, specifically including: performing convolution processing on the third feature image to obtain a first predicted segmentation image with the same number of channels and the same number of categories; performing up-sampling processing on the first predicted segmentation image to obtain a second predicted segmentation image; performing convolution processing on the first up-sampling image to obtain a first target up-sampling image with the same size as the second predicted segmentation image; performing fusion processing on the second predicted segmentation image and the first target up-sampling image to obtain a third predicted segmentation image; performing up-sampling processing on the third predicted segmentation image to obtain a fourth predicted segmentation image; performing convolution processing on the second up-sampling image to obtain a second target up-sampling image with the same size as the fourth predicted segmentation image; performing fusion processing on the fourth predicted segmentation image and the second target up-sampling image to obtain a fifth predicted segmentation image; performing up-sampling processing on the fifth predicted segmentation image using a bilinear interpolation algorithm to obtain the predicted segmentation image.

[0010] Optionally, the judgment on the fault type of the target fault waveform obtains a current fault type judgment result, specifically including: performing similarity calculation processing based on the target fault waveform and a preset fault feature in a preset fault feature library, determining a fault type corresponding to the preset fault feature with the largest similarity as a target current fault type, and determining a target fault feeder according to the target current fault type.

[0011] To solve the above problems, the application provides a current fault type judgment device, which includes: a receiving module configured to receive three-phase current signals after a fault occurs; an image conversion module configured to perform image conversion on the three-phase current signals to obtain a to-be-fault-judgment image; a semantic segmentation module configured to perform semantic segmentation processing on the to-be-fault-judgment image using a pre-trained semantic segmentation model to obtain a target image including a target fault waveform; a judgment module configured to judge the fault type of the target fault waveform to obtain a current fault type judgment result.

[0012] To solve the above problems, the present application provides a storage medium, which stores a computer program, and the computer program is executed by a processor to realize the steps of the current fault type judgment method.

[0013] To solve the above problems, the present application provides an electronic device, which at least includes a memory and a processor, the memory stores a computer program, and the processor realizes the steps of the current fault type judgment method when executing the computer program on the memory.

[0014] The beneficial effects in the present application: the present application realizes the paradigm shift from "numerical calculation" to "visual perception". The present application converts one-dimensional current waveform into two-dimensional image, and converts a complex signal processing problem into an image segmentation problem. This enables the algorithm to intuitively and comprehensively perceive the overall morphology, local mutation and space-time correlation of the waveform like the human eye, so as to capture those weak, nonlinear fault characteristics ignored by traditional methods. Using semantic segmentation model, the fault waveform can be accurately identified to each pixel point, not only to judge "whether there is a fault", but also to accurately locate the specific time and phase of the fault occurrence, eliminating the ambiguity caused by time window and threshold setting in traditional methods. The present application not only realizes performance breakthrough in the specific technical point of small current grounding fault discrimination, but more importantly, provides a systematic method that combines advanced artificial intelligence technology and power system professional knowledge, which can be implemented and evolved, and has important practical value for improving the power supply reliability and intelligent level of distribution network.

[0015] The above description is only a summary of the technical scheme of the present application, in order to more clearly understand the technical means of the present application, the specific embodiments of the present application can be implemented according to the content of the specification, and in order to make the above and other purposes, characteristics and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described. BRIEF DESCRIPTION OF DRAWINGS

[0016] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The accompanying drawings are included to provide a description of the preferred embodiments and are not meant to limit the present application. Moreover, the same reference numerals in the attached drawings indicate the same or similar components. In the drawings: Figure 1 A flowchart of a current fault type judgment method provided by an embodiment of the present application is shown; Figure 2 A flowchart of a current fault type judgment method provided by another embodiment of the present application is shown; Figure 3 A structure block diagram of a current fault type judgment device provided by another embodiment of the present application is shown. Detailed Implementation

[0017] Various embodiments and features of this application are described herein with reference to the accompanying drawings.

[0018] It should be understood that various modifications can be made to the embodiments described herein. Therefore, the above description should not be considered as limiting, but merely as an example of embodiments. Other modifications within the scope and spirit of this application will be apparent to those skilled in the art.

[0019] The accompanying drawings, which are included in and form part of this specification, illustrate embodiments of the present application and, together with the general description of the present application given above and the detailed description of the embodiments given below, serve to explain the principles of the present application.

[0020] These and other features of this application will become apparent from the following description of preferred forms of embodiments given as non-limiting examples, with reference to the accompanying drawings.

[0021] It should also be understood that although this application has been described with reference to some specific examples, those skilled in the art can certainly implement many other equivalent forms of this application.

[0022] The above and other aspects, features and advantages of this application will become more apparent when taken in conjunction with the accompanying drawings and in view of the following detailed description.

[0023] Specific embodiments of this application are described thereafter with reference to the accompanying drawings; however, it should be understood that the claimed embodiments are merely examples of this application, which can be implemented in various ways. Well-known and / or repeated functions and structures are not described in detail to avoid unnecessary or redundant details that could obscure the application. Therefore, the specific structural and functional details claimed herein are not intended to be limiting, but merely serve as the basis and representative basis for the claims to teach those skilled in the art to use this application in a variety of substantially any suitable detailed structures.

[0024] This specification may use the phrases “in one embodiment,” “in another embodiment,” “in yet another embodiment,” or “in other embodiments,” all of which may refer to one or more of the same or different embodiments according to this application.

[0025] This application provides a method for determining the type of current fault, such as... Figure 1 As shown, it includes: Step S101: Receive the three-phase current signal after the fault occurs; In the implementation process, the source of the three-phase current signal includes a three-phase current signal collected by a fault recording device, a three-phase current signal obtained by a real model test using an analog power grid platform, or a three-phase current signal generated by a digital simulation platform; the above signals are input to the fault type identification module after analog-to-digital conversion, so as to ensure that the data time synchronization and the sampling accuracy are consistent.

[0026] Step S102: image conversion is performed on the three-phase current signal to obtain a to-be-fault-judged image. In the implementation process, the one-dimensional current signal is converted into a two-dimensional image. The image contains information such as amplitude, phase, harmonic, and transient. GPU or special hardware is used to accelerate image generation to meet real-time requirements.

[0027] Step S103: a pre-trained semantic segmentation model is used to perform semantic segmentation processing on the to-be-fault-judged image to obtain a target image including a target fault waveform. In the implementation process, the input layer of the semantic segmentation model is used to perform standardization processing on the to-be-fault-judged image to obtain a first feature image; the backbone network layer of the semantic segmentation model is used to perform feature extraction on the first feature image to obtain a second feature image, the second feature image including a first up-sampling image and a second up-sampling image for decoder feature extraction of the semantic segmentation model and a high-level feature image for feature extraction of an attention module of the semantic segmentation model; the attention module of the semantic segmentation model is used to perform channel weighting processing on the high-level feature image to obtain a third feature image; the decoder of the semantic segmentation model is used to perform feature fusion processing on the first up-sampling image, the second up-sampling image, and the third feature image to obtain a predicted segmentation image including a target fault waveform; a preset activation function is used to activate the predicted segmentation image in the channel dimension to obtain a target image including a target fault waveform. The semantic segmentation method is used to classify each pixel point in the image: to distinguish background pixels, non-fault waveform pixels, and fault waveform pixels, to realize pixel-level fault positioning, and to improve recognition accuracy and interpretability.

[0028] Step S104: the fault type of the target fault waveform is judged to obtain a current fault type judgment result.

[0029] In the specific implementation process of this step, similarity calculation processing is performed based on the target fault waveform and the preset fault characteristics in the preset fault characteristic library, the fault type corresponding to the preset fault characteristic with the maximum similarity is determined as the target current fault type, and the target fault feeder is determined according to the target current fault type. The segmented fault waveform is matched with the preset fault characteristic. The fault type and the fault feeder are determined according to the highest similarity. The fault type is an A-phase ground fault, a B-phase ground fault, or the like.

[0030] The application realizes the paradigm shift from “numerical calculation” to “visual perception”. The application converts one-dimensional current waveform into two-dimensional image, and converts a complex signal processing problem into an image segmentation problem. This enables the algorithm to intuitively and comprehensively perceive the overall morphology, local mutation and space-time correlation of the waveform like the human eye, so as to capture those weak, nonlinear fault characteristics ignored by traditional methods. Using a semantic segmentation model, the fault waveform can be accurately identified to each pixel point, not only to determine “whether there is a fault”, but also to accurately locate the specific time and phase of the fault occurrence, eliminating the ambiguity caused by the time window and threshold setting in traditional methods. The application not only realizes performance breakthrough in the specific technical point of small current ground fault discrimination, but more importantly, provides a systematic method that integrates advanced artificial intelligence technology and power system professional knowledge, which can be implemented and evolved, and has important practical value for improving the power supply reliability and intelligent level of distribution network.

[0031] Another embodiment of the application provides another current fault type judgment method, as shown in Figure 2 , comprising: Step S201: receiving three-phase current signals after a fault occurs; In the specific implementation process of this step, the source of the three-phase current signal includes a three-phase current signal collected by a fault recording device, a three-phase current signal obtained by a real test using a simulation power grid platform, or a three-phase current signal generated by a digital simulation platform; the above signals are input to a fault type discrimination module after analog-to-digital conversion, ensuring that the data time synchronization and sampling accuracy are consistent.

[0032] Step S202: image conversion is performed on the three-phase current signals to obtain a fault judgment image; In the specific implementation process of this step, one-dimensional current signals are converted into two-dimensional images. The image contains amplitude, phase, harmonic, transient and other information. GPU or special hardware is used to accelerate image generation to meet real-time requirements.

[0033] Step S203: constructing a semantic segmentation model; The step acquires a plurality of historical fault waveform images of different fault types in the implementation process. The fault types include A-phase ground fault, B-phase ground fault and the like. Each historical fault waveform image is labeled to obtain a label image set. The label data of the label image set includes background pixel identification, non-fault waveform pixel identification and fault waveform pixel identification. For example, pixel identification can be represented by color. For example, black represents background pixels, blue represents non-fault waveforms, and red represents fault waveforms. Different fault waveform categories can be identified by digital identification. For example, the number 1 represents an A-phase ground fault, the number 2 represents a B-phase ground fault, and the like. The historical fault waveform images and the label image set are used to train a preset neural network model with an attention mechanism to obtain the semantic segmentation model. The preset neural network model includes an input layer, a backbone network layer, an attention module, a decoder and an output layer. Specifically, each historical fault waveform image represents the change of three-phase current in a time window. The label image corresponds to the historical fault waveform image one by one. The label image is a color-coded image, in which the color of each pixel represents its category. For example, black represents background pixels, blue represents non-fault waveforms, and red represents fault waveforms. The process of generating label data can be completed by a domain expert. The expert reviews historical fault data or simulation data, and manually and accurately labels the starting point, ending point and phase of the fault occurrence on the waveform image. The backbone network layer uses a pre-trained convolutional neural network (such as VGG or ResNet) as a feature extractor. The fully connected layer at the end of the backbone network is removed and replaced with a convolutional layer. This allows the network to accept input images of any size and output a feature map of the same spatial size (but with a different number of channels). The decoder gradually recovers the size of the original input image through upsampling operations (such as transpose convolution) on the deep and abstract features extracted by the backbone network. The features from different levels of the encoder are fused to combine high-level semantic information and low-level detail information, thereby achieving accurate pixel positioning. An attention module is introduced at the bottleneck of the network or the jump connection. The module automatically learns a "weight map" with the same size as the feature map. The weight value at each position is between 0 and 1. When performing feature fusion or forward propagation, the weight map is multiplied point by point with the feature map. Areas with high weights (close to 1) indicate that the model should "focus" on these areas, which usually contain significant fault features (such as current mutation points). Areas with low weights (close to 0) indicate that these are irrelevant information or noise and should be "suppressed".The specific training process is as follows: step one, building a preset neural network model with an attention module; step two, randomly initializing or loading pre-training weights; step three, using the preset neural network model to predict a plurality of historical fault waveform images by using a forward propagation method to obtain a predicted segmentation image; step four, comparing the predicted segmentation image with a label image, and using a cross-entropy function to calculate a loss value; step five, calculating the gradient of the model parameters based on the loss value; step six, updating the model weights according to the gradient using an optimizer; step seven, determining whether a preset training termination condition is met, and when the preset training termination condition is met, executing step eight; when the preset training termination condition is not met, repeating steps one to six until the preset training termination condition is met, thereby obtaining the semantic segmentation model; step eight, when the preset training termination condition is met, determining the preset neural network model as the semantic segmentation model. The preset training termination condition can be that the current iteration period is greater than or equal to the maximum iteration period; the preset training termination condition can also be that the loss value of the current iteration period is less than or equal to a preset threshold.

[0034] Step S204: normalizing the to-be-fault-judgment image using an input layer of the semantic segmentation model to obtain a first feature image; In the specific implementation process of this step, the size of the to-be-fault-judgment image is Height x Width x 3, and the three channels correspond to the amplitudes of three-phase currents, etc. 、 、 The to-be-fault-judgment image is normalized to obtain a first feature image, and the size of the first feature image is (H, W, 3), where H represents the image height, W represents the image width, and 3 represents the number of image channels.

[0035] Step S205: extracting features from the first feature image using a backbone network layer of the semantic segmentation model to obtain a second feature image, wherein the second feature image includes a first up-sampling image and a second up-sampling image for decoding feature extraction of the semantic segmentation model, and a high-level feature image for feature extraction of an attention module of the semantic segmentation model; In the implementation process of this step, the first feature image is sequentially subjected to feature extraction by multiple convolution pooling layers, the output of a previous convolution pooling layer is used as the input of a subsequent convolution pooling layer, and the second feature image of the bottom convolution pooling output is obtained. Specifically, the backbone network layer includes a first convolution layer, a first pooling layer, a second convolution layer, a third convolution layer, a fourth convolution layer, and a fifth convolution layer; the first convolution layer is used to perform convolution processing on the first feature image, and the first pooling layer is used to perform maximum pooling processing on the feature image after convolution processing, thereby obtaining a first sub-feature image with a halved spatial size and an increased number of channels; for example, the size of the first sub-feature image is (H / 2, W / 2, 64). The second convolution layer is used to perform convolution processing on the first sub-feature image, and the first sub-feature image is processed by multiple residual blocks to obtain a second sub-feature image; for example, the size of the second sub-feature image is (H / 4, W / 4, 256); the third convolution layer is used to perform convolution processing on the second sub-feature image, and the second sub-feature image is processed by more and deeper residual blocks to obtain a first up-sampling image; for example, the size of the first up-sampling image is (H / 8, W / 8, 512) The first up-sampling image is used for up-sampling by the decoder of the semantic segmentation model, and subsequent feature fusion processing is performed to obtain a predicted segmentation image including a target fault waveform. The fourth convolution layer is used to perform convolution processing on the first up-sampling image to obtain a second up-sampling image; for example, the size of the second up-sampling image is (H / 16, W / 16, 1024) The second up-sampling image is used for up-sampling by the decoder of the semantic segmentation model, and subsequent feature fusion processing is performed to obtain a predicted segmentation image including a target fault waveform. The fifth convolution layer is used to perform convolution processing on the second up-sampling image to obtain a high-level feature image used for feature extraction by the attention module of the semantic segmentation model. For example, the size of the high-level feature image is (H / 32, W / 32, 2048).

[0036] Step S206: performing channel weighting processing on the high-level feature image by using the attention module of the semantic segmentation model to obtain a third feature image. In the implementation process of this step, a global average pooling operation is performed on the high-level feature image to obtain a global importance indicator of each channel; a scaling factor of each channel is obtained by calculating and processing the global importance indicator; and the third feature image is obtained by performing channel weighting processing on the high-level feature image according to the scaling factor. After channel weighting, the size of the feature image remains unchanged, and is still (H / 32, W / 32, 2048), but the information content is more concentrated, and more attention is paid to fault features.

[0037] Step S207: performing feature fusion processing on the first up-sampling image, the second up-sampling image and the third feature image by using a decoder of the semantic segmentation model to obtain a predicted segmentation image including a target fault waveform; In this step, the third feature image is subjected to convolution processing to obtain a first predicted segmentation image with the same number of channels and the same number of categories. The number of channels is reduced to the number of categories K by using 1x1 convolution, and K=3 in this application: background category, non-fault category and fault category. The size of the first predicted segmentation image is (H / 32, W / 32, K). The first predicted segmentation image is subjected to up-sampling processing to obtain a second predicted segmentation image. Specifically, the first predicted segmentation image is subjected to 2 times up-sampling to obtain a second predicted segmentation image with a size of (H / 16, W / 16, K). The first up-sampling image is subjected to convolution processing to obtain a first target up-sampling image with the same size as the second predicted segmentation image. The first up-sampling image connected by the skip connection is also adjusted to K channels by 1x1 convolution, and the size is (H / 16, W / 16, K). The second predicted segmentation image and the first target up-sampling image are subjected to fusion processing to obtain a third predicted segmentation image. The second predicted segmentation image and the first target up-sampling image are added element by element to obtain a third predicted segmentation image that fuses high-level semantics and low-level details. The third predicted segmentation image is subjected to up-sampling processing to obtain a fourth predicted segmentation image. The third predicted segmentation image is subjected to 2 times up-sampling to obtain a fourth predicted segmentation image with an image size of (H / 8, W / 8, K). The second up-sampling image is subjected to convolution processing to obtain a second target up-sampling image with the same size as the fourth predicted segmentation image. The fourth predicted segmentation image and the second target up-sampling image are subjected to fusion processing to obtain a fifth predicted segmentation image. The fifth predicted segmentation image has an image size of (H / 8, W / 8, K). The fifth predicted segmentation image is subjected to up-sampling processing by using a bilinear interpolation algorithm to obtain the predicted segmentation image. The size of the final predicted segmentation image is (H, W, K).

[0038] Step S208: activating the predicted segmentation image by using a preset activation function in the channel dimension to obtain a target image including a target fault waveform. In the specific implementation process, a Softmax function is applied in the channel dimension. For each pixel position (i, j) in the image, the values of the K channels are converted into a probability distribution representing the probability of the pixel belonging to each class. The sum of the K probabilities is 1. The probability map has a size of (H, W, K). During inference, the class with the maximum probability at each pixel position is taken, and the final pixel-level classification result is obtained, that is, the target image including the target fault waveform is obtained.

[0039] Step S209: judging the fault type of the target fault waveform to obtain a current fault type judgment result.

[0040] In the specific implementation process, similarity calculation processing is performed based on the target fault waveform and the preset fault features in the preset fault feature library. The fault type corresponding to the preset fault feature with the maximum similarity is determined as the target current fault type, and a target fault feeder is determined according to the target current fault type.

[0041] Another embodiment of the present application provides a current fault type judgment device, as shown in the figure, comprising: Figure 3 A receiving module 1 is configured to receive a three-phase current signal after a fault occurs. An image conversion module 2 is configured to perform image conversion on the three-phase current signal to obtain a to-be-fault-judged image. A semantic segmentation module 3 is configured to perform semantic segmentation processing on the to-be-fault-judged image by using a pre-trained semantic segmentation model to obtain a target image including a target fault waveform. A judgment module 4 is configured to judge the fault type of the target fault waveform to obtain a current fault type judgment result.

[0042] In the specific implementation process, the device further comprises a model construction module, which is specifically configured to obtain a plurality of historical fault waveform images of different fault types; label each historical fault waveform image to obtain a label image set, and the labeled label data includes background pixel identification, non-fault waveform pixel identification, and fault waveform pixel identification; and perform model training on a preset neural network model with an added attention mechanism by using the historical fault waveform images and the label image set to obtain the semantic segmentation model. The preset neural network model includes an input layer, a backbone network layer, an attention module, a decoder, and an output layer.

[0043] ​In the specific implementation process, the semantic segmentation module 3 is specifically configured to: perform standardization processing on the to-be-failure-judgment image by using an input layer of the semantic segmentation model to obtain a first feature image; perform feature extraction on the first feature image by using a backbone network layer of the semantic segmentation model to obtain a second feature image, the second feature image including a first up-sampling image and a second up-sampling image for performing up-sampling feature extraction by a decoder of the semantic segmentation model, and a high-level feature image for performing feature extraction by an attention module of the semantic segmentation model; perform channel weighting processing on the high-level feature image by using the attention module of the semantic segmentation model to obtain a third feature image; perform feature fusion processing on the first up-sampling image, the second up-sampling image, and the third feature image by using the decoder of the semantic segmentation model to obtain a predicted segmentation image including a target failure waveform; and activate the predicted segmentation image by using a preset activation function in the channel dimension to obtain a target image including the target failure waveform.

[0044] In the specific implementation process, the semantic segmentation module 3 is further configured to: perform feature extraction on the first feature image by using a plurality of convolutional pooling layers one by one, with the output of a previous convolutional pooling layer serving as the input of a subsequent convolutional pooling layer, to obtain the second feature image of a bottom-layer convolutional pooling output.

[0045] In the specific implementation process, the semantic segmentation module 3 is further configured to: perform a global average pooling operation on the high-level feature image to obtain a global importance indicator of each channel; perform calculation processing according to the global importance indicator to obtain a scaling factor of each channel; and perform channel weighting processing according to the scaling factor and the high-level feature image to obtain the third feature image.

[0046] In the specific implementation process, the semantic segmentation module 3 is further configured to: perform convolution processing on the third feature image to obtain a first predicted segmentation image having the same number of channels and the same number of categories; perform up-sampling processing on the first predicted segmentation image to obtain a second predicted segmentation image; perform convolution processing on the first up-sampling image to obtain a first target up-sampling image having the same size as the second predicted segmentation image; perform fusion processing on the second predicted segmentation image and the first target up-sampling image to obtain a third predicted segmentation image; perform up-sampling processing on the third predicted segmentation image to obtain a fourth predicted segmentation image; perform convolution processing on the second up-sampling image to obtain a second target up-sampling image having the same size as the fourth predicted segmentation image; perform fusion processing on the fourth predicted segmentation image and the second target up-sampling image to obtain a fifth predicted segmentation image; and perform up-sampling processing on the fifth predicted segmentation image by using a bilinear interpolation algorithm to obtain the predicted segmentation image.

[0047] In a specific implementation process, the determination module 4 is specifically configured to perform similarity calculation processing based on the target fault waveform and preset fault features in a preset fault feature library, determine a fault type corresponding to a preset fault feature with the largest similarity as a target current fault type, and determine a target fault feeder according to the target current fault type.

[0048] The application realizes the paradigm shift from "numerical calculation" to "visual perception". The application converts one-dimensional current waveform into two-dimensional image, and converts a complex signal processing problem into an image segmentation problem. This enables the algorithm to intuitively and comprehensively perceive the overall shape, local mutation and space-time correlation of the waveform like the human eye, so as to capture those weak, nonlinear fault features ignored by traditional methods. Using a semantic segmentation model, the fault waveform can be accurately identified to each pixel point, not only to determine "whether there is a fault", but also to accurately locate the specific time and phase of the fault occurrence, eliminating the ambiguity caused by time window and threshold setting in traditional methods. The application not only realizes performance breakthrough in the specific technical point of small current grounding fault discrimination, but more importantly, provides a systematic method that integrates advanced artificial intelligence technology and power system professional knowledge, which can be implemented and evolved, and has important practical value for improving the power supply reliability and intelligent level of distribution network.

[0049] Another embodiment of the application provides a storage medium storing a computer program, wherein the computer program is executed by a processor to implement the following method steps: Step one, receiving three-phase current signals after a fault occurs; Step two, image conversion is performed on the three-phase current signals to obtain a to-be-fault-determined image; Step three, a pre-trained semantic segmentation model is used to perform semantic segmentation processing on the to-be-fault-determined image to obtain a target image including a target fault waveform; Step four, the fault type of the target fault waveform is determined to obtain a current fault type determination result.

[0050] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, storage, database or other medium used in the embodiments provided in the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0051] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above-mentioned division of functional units and modules is exemplified, and in actual application, the above-mentioned functions can be completed by different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above.

[0052] The specific implementation process of the above method steps can refer to the embodiments of any of the current fault type judgment methods described above, and this embodiment will not be repeated here.

[0053] The application realizes the paradigm shift from "numerical calculation" to "visual perception". The application converts one-dimensional current waveform into two-dimensional image, and converts a complex signal processing problem into an image segmentation problem. This enables the algorithm to intuitively and comprehensively perceive the overall morphology, local mutation and space-time correlation of the waveform like the human eye, so as to capture the weak, nonlinear fault characteristics ignored by traditional methods. By using a semantic segmentation model, the fault waveform can be accurately identified to each pixel point, not only to judge "whether there is a fault", but also to accurately locate the specific time and phase of the fault occurrence, eliminating the ambiguity caused by the time window and threshold setting of the traditional method. The application not only realizes performance breakthrough in the specific technical point of small current grounding fault discrimination, but more importantly, provides a systematic method that combines advanced artificial intelligence technology and power system professional knowledge, which can be implemented and evolved, and has important practical value for improving the power supply reliability and intelligent level of distribution network.

[0054] Another embodiment of the application provides an electronic device, which can be a server, comprising a processor, a memory, a network interface and a database connected by a system bus. The processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes non-volatile and / or volatile storage media, internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the electronic device is used to communicate with the external client through the network connection. The electronic device program is executed by the processor to realize the function or step of the server side of the current fault type judgment method.

[0055] In one embodiment, an electronic device is provided, which can be a client. The electronic device comprises a processor, a memory, a network interface, a display screen and an input device connected by a system bus. The processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the electronic device is used to communicate with the external server through the network connection. The electronic device program is executed by the processor to realize the function or step of the client side of the current fault type judgment method.

[0056] Another embodiment of the application provides an electronic device, which at least includes a memory and a processor, the memory stores a computer program, and the processor realizes the following method steps when executing the computer program on the memory: Step one, receiving three-phase current signals after fault occurs; Step two, image conversion is performed on the three-phase current signal to obtain a to-be-fault-judgment image; Step three, a pre-trained semantic segmentation model is used to perform semantic segmentation processing on the to-be-fault-judgment image to obtain a target image including a target fault waveform; Step four, the fault type of the target fault waveform is judged to obtain a current fault type judgment result.

[0057] The specific implementation process of the above method steps can be referred to the embodiments of any current fault type judgment method described above, and will not be repeated here.

[0058] The present application realizes the paradigm shift from "numerical calculation" to "visual perception". The present application converts one-dimensional current waveform into two-dimensional image, and converts a complex signal processing problem into an image segmentation problem. This enables the algorithm to intuitively and comprehensively perceive the overall shape, local mutation and space-time correlation of the waveform like the human eye, so as to capture those weak, nonlinear fault characteristics ignored by traditional methods. Using a semantic segmentation model, the fault waveform can be accurately identified to each pixel point, not only to judge "whether there is a fault", but also to accurately locate the specific time and phase of the fault occurrence, eliminating the ambiguity caused by the time window and threshold setting of traditional methods. The present application not only realizes performance breakthrough in the specific technical point of small current grounding fault discrimination, but more importantly, provides a systematic method that combines advanced artificial intelligence technology and power system professional knowledge, which can be implemented and evolved, and has important practical value for improving the power supply reliability and intelligent level of distribution network.

[0059] The above embodiments are only exemplary embodiments of the present application and are not used to limit the present application, and the protection scope of the present application is defined by the claims. Those skilled in the art can make various modifications or equivalent replacements to the present application within the spirit and protection scope of the present application, and such modifications or equivalent replacements should also be considered to fall within the protection scope of the present application.

Claims

1. A method of determining a type of current fault, characterized by, The method comprises the following steps: receiving a three-phase current signal after a fault occurs; image conversion is performed on the three-phase current signal to obtain a to-be-fault-judged image; a pre-trained semantic segmentation model is used to perform semantic segmentation processing on the to-be-fault-judged image to obtain a target image including a target fault waveform; the fault type of the target fault waveform is judged to obtain a current fault type judgment result.

2. The method of claim 1, wherein, Before the pre-trained semantic segmentation model is used to perform semantic segmentation processing on the to-be-fault-judged image, the method further comprises the following steps: constructing the semantic segmentation model; the construction of the semantic segmentation model specifically comprises the following steps: obtaining a plurality of historical fault waveform images of different fault types; labeling each of the historical fault waveform images to obtain a label image set, and the labeled label data comprises a background pixel identifier, a non-fault waveform pixel identifier, and a fault waveform pixel identifier; a pre-set neural network model with an attention mechanism is trained using the historical fault waveform images and the label image set to obtain the semantic segmentation model; 3. The method of claim 1, wherein, wherein the pre-set neural network model comprises an input layer, a backbone network layer, an attention module, a decoder, and an output layer. The method comprises the following steps: the input layer of the semantic segmentation model is used to perform standardization processing on the to-be-fault-judged image to obtain a first feature image; the backbone network layer of the semantic segmentation model is used to perform feature extraction on the first feature image to obtain a second feature image, wherein the second feature image comprises a first up-sampling image and a second up-sampling image for up-sampling feature extraction by the decoder of the semantic segmentation model, and a high-level feature image for feature extraction by the attention module of the semantic segmentation model; the attention module of the semantic segmentation model is used to perform channel weighting processing on the high-level feature image to obtain a third feature image; the decoder of the semantic segmentation model is used to perform feature fusion processing on the first up-sampling image, the second up-sampling image, and the third feature image to obtain a predicted segmentation image including a target fault waveform; 4. The method of claim 3, wherein, a pre-set activation function is used to activate the predicted segmentation image in the channel dimension to obtain a target image including a target fault waveform. The backbone network layer comprises a plurality of convolutional pooling layers, and the backbone network layer of the semantic segmentation model is used to perform feature extraction on the first feature image to obtain a second feature image, specifically comprising the following steps:

5. The method of claim 3, wherein, a plurality of convolutional pooling layers are used to perform feature extraction on the first feature image one by one, and the output of a previous convolutional pooling layer is used as the input of a subsequent convolutional pooling layer to obtain the second feature image of the bottom convolutional pooling output. The attention module of the semantic segmentation model is used to perform channel weighting processing on the high-level feature image to obtain a third feature image, specifically comprising the following steps: global average pooling is performed on the high-level feature image to obtain a global importance indicator of each channel; According to the global importance index, a scaling factor of each channel is calculated; According to the scaling factor and the high-level feature image, a channel weighting process is performed to obtain the third feature image.

6. The method of claim 3, wherein, The decoder of the semantic segmentation model performs feature fusion processing on the first up-sampling image, the second up-sampling image and the third feature image to obtain a predicted segmentation image including a target fault waveform, specifically including: The third feature image is convoluted to obtain a first predicted segmentation image with the same number of channels and categories as the third feature image; The first predicted segmentation image is up-sampled to obtain a second predicted segmentation image; The first up-sampling image is convoluted to obtain a first target up-sampling image with the same size as the second predicted segmentation image; The second predicted segmentation image and the first target up-sampling image are fused to obtain a third predicted segmentation image; The third predicted segmentation image is up-sampled to obtain a fourth predicted segmentation image; The second up-sampling image is convoluted to obtain a second target up-sampling image with the same size as the fourth predicted segmentation image; The fourth predicted segmentation image and the second target up-sampling image are fused to obtain a fifth predicted segmentation image; The fifth predicted segmentation image is up-sampled using a bilinear interpolation algorithm to obtain the predicted segmentation image.

7. The method of claim 1, wherein, The fault type of the target fault waveform is determined to obtain a current fault type determination result, specifically including: Based on similarity calculation processing between the target fault waveform and a preset fault feature in a preset fault feature library, the fault type corresponding to the preset fault feature with the largest similarity is determined as a target current fault type, and a target fault feeder is determined according to the target current fault type.

8. A current fault type determination device, characterized by comprising: Including: A receiving module is configured to receive three-phase current signals after a fault occurs; An image conversion module is configured to perform image conversion on the three-phase current signals to obtain a to-be-fault-judged image; A semantic segmentation module is configured to perform semantic segmentation processing on the to-be-fault-judged image using a pre-trained semantic segmentation model to obtain a target image including a target fault waveform; A determination module is configured to determine a fault type of the target fault waveform to obtain a current fault type determination result.

9. A storage medium, characterized by The storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the current fault type determination method in any one of claims 1-7.

10. An electronic device, comprising: At least including a memory and a processor, the memory stores a computer program, and the processor implements the steps of the current fault type determination method in any one of claims 1-7 when executing the computer program on the memory.

Images