Multi-loop cable state type diagnosis method, device, equipment and medium

Abstract

The invention discloses a multi-loop cable state type diagnosis method, device and equipment and a medium, and the method comprises the steps: obtaining multi-loop cable three-phase grounding circulation data corresponding to a to-be-diagnosed multi-loop cable; and inputting the multi-loop cable three-phase grounding circulation data into a target cable state type diagnosis model, and outputting a state type corresponding to the to-be-diagnosed cable, the target cable state type diagnosis model being generated by training a three-branch parallel convolutional neural network model based on the multi-loop cable three-phase grounding circulation data. Therefore, the state type of the to-be-diagnosed multi-loop cable can be obtained by inputting the three-phase grounding circulation data of the multi-loop cable into the target cable state type diagnosis model, so that the state type diagnosis efficiency is improved; and the target cable state type diagnosis model is generated by training the three-branch parallel convolutional neural network model based on the multi-loop cable three-phase grounding circulation data, so that the limitation in a single-loop feature extraction mode is avoided, and the state type diagnosis accuracy is improved.

Description

Technical Field

[0001] This application relates to the technical field of cable condition diagnosis, and particularly to a method, device, equipment and medium for diagnosing the condition types of multiple-circuit cables. Background Art

[0002] With the rapid development of the power industry, cable lines, as key infrastructure for power transmission, the importance of their safe and stable operation has become increasingly prominent. The multiple-circuit cable system has become a widely applied solution in the power transmission field due to its significant advantages in improving power supply reliability and constructing system redundancy. However, this development trend has correspondingly raised the requirements and challenges for cable defect diagnosis technology.

[0003] Traditional cable defect diagnosis methods mainly rely on manual inspections and simple electrical signal analysis. The manual inspection method is not only inefficient but also prone to missing potential defect hazards, making it difficult to meet the requirements of modern power systems for the efficiency and accuracy of monitoring and diagnosis. The existing electrical signal analysis methods have poor adaptability to complex cable systems and cannot effectively determine the type of defects. That is, the accuracy and efficiency of cable condition diagnosis in the existing technology are relatively low. Therefore, there is an urgent need for a diagnosis method to improve the efficiency and accuracy of cable condition diagnosis. Summary of the Invention

[0004] This application provides a method, device, equipment and medium for diagnosing the condition types of multiple-circuit cables, which can quickly and accurately diagnose the condition types of cables.

[0005] In a first aspect, this application provides a method for diagnosing the condition types of multiple-circuit cables, the method comprising:

[0006] Obtaining multi-phase grounding loop current data of multiple-circuit cables corresponding to the multiple-circuit cables to be diagnosed;

[0007] Inputting the multi-phase grounding loop current data of the multiple-circuit cables into a target cable condition type diagnosis model, and outputting the condition type corresponding to the cable to be diagnosed; the target cable condition type diagnosis model is generated by training a three-branch parallel convolutional neural network model based on the multi-phase grounding loop current data of multiple-circuit cables.

[0008] Optionally, the method further comprises:

[0009] Dividing a target sample set to obtain a target training sample set and a target test sample set; the target sample set includes multi-phase loop current data of multiple-circuit cables and the true condition type labels corresponding to the multi-phase loop current data of the multiple-circuit cables;

[0010] Training the three-branch parallel convolutional neural network model based on the target training sample set to generate an initial cable condition type diagnosis model;

[0011] Test the initial cable state type diagnosis model according to the target test sample set to obtain the test accuracy rate.

[0012] After the test accuracy rate reaches the preset accuracy threshold, obtain the target cable state type diagnosis model as the initial cable state type diagnosis model.

[0013] Optionally, training the three-branch parallel convolutional neural network model based on the target training sample set to generate an initial cable state type diagnosis model includes:

[0014] Obtain the grounding loop current data under defective working conditions and the grounding loop current data under normal working conditions in the target training sample set.

[0015] Input the grounding loop current data under defective working conditions and the grounding loop current data under normal working conditions into the three-branch parallel convolutional neural network model through a multi-head attention mechanism, and output the first feature, the second feature, and the third feature.

[0016] Perform branch merging on the first feature, the second feature, and the third feature, and output the predicted state type label corresponding to the target training sample set.

[0017] Calculate the error result between the true state type label and the predicted state type label according to the loss function of the three-branch parallel convolutional neural network model.

[0018] Based on the error result, calculate the gradient corresponding to the model parameters of the three-branch parallel convolutional neural network model.

[0019] Update and iterate the model parameters of the three-branch parallel convolutional neural network model according to the gradient to complete the training of the three-branch parallel convolutional neural network model and generate the initial cable state type diagnosis model.

[0020] Optionally, the three-branch parallel convolutional neural network model includes a first branch, a second branch, and a third branch. The step of inputting the grounding loop current data under defective working conditions and the grounding loop current data under normal working conditions into the three-branch parallel convolutional neural network model through a multi-head attention mechanism and outputting the first feature, the second feature, and the third feature includes:

[0021] Input the grounding loop current data under defective working conditions into the first branch through the multi-head attention mechanism, and output the first feature; the first branch is used for feature extraction of the grounding loop current data under defective working conditions and outputting the first feature.

[0022] Through the multi-head attention mechanism, input the normal operating condition ground loop current data into the second branch and output the second feature; the second branch is used to extract features from the normal operating condition ground loop current data and output the second feature.

[0023] Through the multi-head attention mechanism, input the defective operating condition ground loop current data and the normal operating condition ground loop current data into the third branch and output the third feature; the third branch is used to extract features from both the defective operating condition ground loop current data and the normal operating condition ground loop current data simultaneously and output the third feature.

[0024] Optionally, the branch merging of the first feature, the second feature, and the third feature to output the predicted status type label corresponding to the target training sample set includes:

[0025] Perform local feature extraction on the first feature, the second feature, and the third feature through a convolutional layer to output local features.

[0026] Based on a flattening layer, convert the local features into a one-dimensional vector.

[0027] Perform integration processing on the one-dimensional vector through the fully connected layer to output the predicted status type label.

[0028] Optionally, the branch merging of the first feature, the second feature, and the third feature to output the predicted status type label corresponding to the target training sample set includes:

[0029] Perform local feature extraction on the first feature, the second feature, and the third feature through a convolutional layer to output local features.

[0030] Based on a flattening layer, convert the local features into a one-dimensional vector.

[0031] Perform integration processing on the one-dimensional vector through the fully connected layer to output the predicted status type label.

[0032] Optionally, the status types include: normal type and status type, and the defect types include coaxial cable fracture, metal sheath grounding, cross-bonding box water ingress, or phase conversion failure.

[0033] In a second aspect, the present application also provides a multi-circuit cable status type diagnosis device, and the device includes:

[0034] An acquisition unit for acquiring multi-circuit cable three-phase ground loop current data corresponding to the multi-circuit cable to be diagnosed.

[0035] A diagnostic unit for inputting the three-phase grounding circulating current data of the multi-circuit cable into a target cable status type diagnostic model and outputting the status type corresponding to the cable to be diagnosed; the target cable status type diagnostic model is generated by training a three-branch parallel convolutional neural network model based on the three-phase grounding circulating current data of the multi-circuit cable.

[0036] Optionally, the device further includes a model training unit;

[0037] The model training unit includes:

[0038] A sample set obtaining module for splitting a target sample set to obtain a target training sample set and a target test sample set; the target sample set includes three-phase circulating current data of multi-circuit cables and the true status type labels corresponding to the three-phase circulating current data of the multi-circuit cables;

[0039] A model training module for training the three-branch parallel convolutional neural network model based on the target training sample set to generate an initial cable status type diagnostic model;

[0040] A test model for testing the initial cable status type diagnostic model according to the target test sample set to obtain a test accuracy rate;

[0041] A model obtaining module for obtaining the target cable status type diagnostic model as the initial cable status type diagnostic model after the test accuracy rate reaches a preset accuracy rate threshold.

[0042] Optionally, the model training module is specifically used for:

[0043] Splitting a target sample set to obtain a target training sample set and a target test sample set; the target sample set includes three-phase circulating current data of multi-circuit cables and the true status type labels corresponding to the three-phase circulating current data of the multi-circuit cables;

[0044] Training the three-branch parallel convolutional neural network model based on the target training sample set to generate an initial cable status type diagnostic model;

[0045] Testing the initial cable status type diagnostic model according to the target test sample set to obtain a test accuracy rate;

[0046] Obtaining the target cable status type diagnostic model as the initial cable status type diagnostic model after the test accuracy rate reaches a preset accuracy rate threshold.

[0047] Optionally, the three-branch parallel convolutional neural network model includes a first branch, a second branch, and a third branch. By means of the multi-head attention mechanism, the grounding loop current data under defective conditions and the grounding loop current data under normal conditions are input into the three-branch parallel convolutional neural network model, and the first feature, the second feature, and the third feature are output, including:

[0048] Through the multi-head attention mechanism, the grounding loop current data under defective conditions is input into the first branch, and the first feature is output; the first branch is used to extract features from the grounding loop current data under defective conditions and output the first feature;

[0049] Through the multi-head attention mechanism, the grounding loop current data under normal conditions is input into the second branch, and the second feature is output; the second branch is used to extract features from the grounding loop current data under normal conditions and output the second feature;

[0050] Through the multi-head attention mechanism, the grounding loop current data under defective conditions and the grounding loop current data under normal conditions are input into the third branch, and the third feature is output; the third branch is used to extract features from the grounding loop current data under defective conditions and the grounding loop current data under normal conditions simultaneously and output the third feature.

[0051] Optionally, the branch merging of the first feature, the second feature, and the third feature outputs the predicted state type label corresponding to the target training sample set, including:

[0052] Local features are extracted from the first feature, the second feature, and the third feature through a convolutional layer, and local features are output;

[0053] Based on a flattening layer, the local features are converted into one-dimensional vectors;

[0054] The one-dimensional vectors are integrally processed through the fully connected layer, and the predicted state type label is output.

[0055] Optionally, the sample set obtaining module is further used for:

[0056] Obtain multiple sets of three-phase grounding loop current data of multiple circuits of cables, where the three-phase grounding loop current data of multiple circuits of cables includes simulated calculation three-phase grounding loop current data of multiple circuits of cables and on-site test three-phase grounding loop current data of multiple circuits of cables;

[0057] Classify multiple sets of the three-phase grounding loop current data of multiple circuits of cables according to the state types corresponding to the three-phase grounding loop current data of multiple circuits of cables, and obtain an initial sample data set;

[0058] According to the state type, assign the corresponding true state type label to the three-phase grounding loop current data of the multi-circuit cable;

[0059] Perform data cleaning on the three-phase grounding loop current data of the multi-circuit cable to obtain the three-phase grounding loop current data of the multi-circuit cable after data cleaning; the data cleaning includes null value processing and invalid value processing;

[0060] Obtain an intermediate sample set, where the intermediate sample set includes the three-phase grounding loop current data of the multi-circuit cable after data cleaning and the corresponding true state type label;

[0061] Based on a sliding window of a preset length, convert the three-phase grounding loop current data of the multi-circuit cable after data cleaning in the intermediate sample set to obtain a target sample set, where the target sample set includes time series data of a preset length and the corresponding true state type label.

[0062] Optionally, the state type includes: normal type and defect type, and the defect type includes coaxial cable breakage, metal sheath grounding, cross-bonding box water ingress, or transposition failure.

[0063] In a third aspect, the present application also provides an electronic device, where the electronic device includes a processor and a memory:

[0064] The memory is used to store a computer program;

[0065] The processor is used to execute the multi-circuit cable state type diagnosis method provided in the first aspect according to the computer program.

[0066] In a fourth aspect, the present application also provides a computer-readable storage medium, where the computer-readable storage medium is used to store a computer program, and the computer program is used to execute the multi-circuit cable state type diagnosis method provided in the first aspect.

[0067] Thus, the present application has the following beneficial effects:

[0068] The present application provides a method, apparatus, device and medium for diagnosing the state types of multi-circuit cables. In this method, multi-circuit three-phase grounding loop current data corresponding to the multi-circuit cables to be diagnosed is obtained; the multi-circuit three-phase grounding loop current data is input into a target cable state type diagnosis model to output the state types corresponding to the cables to be diagnosed. The target cable state type diagnosis model is generated by training a three-branch parallel convolutional neural network model based on the multi-circuit three-phase grounding loop current data. In this way, by inputting the multi-circuit three-phase grounding loop current data into the target cable state type diagnosis model, the state types of the multi-circuit cables to be diagnosed can be obtained without further analysis, improving the efficiency of state type diagnosis. And because the target cable state type diagnosis model is generated by training a three-branch parallel convolutional neural network model based on the multi-circuit three-phase grounding loop current data, the three-branch structure of the convolutional neural network after structural adjustment is used to extract features and train the multi-circuit three-phase grounding loop current data. Compared with the prior art, the present application performs more comprehensive feature extraction, avoids the limitations of the single-loop feature extraction mode, reduces the probability of misjudgment or missed judgment, and improves the accuracy of state type diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present application. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0070] Figure 1 It is a schematic flowchart of a method for diagnosing the state types of multi-circuit cables in an embodiment of the present application;

[0071] Figure 2 It is a schematic flowchart of a method for obtaining a target cable state type diagnosis model in an embodiment of the present application;

[0072] Figure 3 It is a schematic structural diagram of a three-branch parallel convolutional neural network model provided in an embodiment of the present application;

[0073] Figure 4 It is a schematic structural diagram of a multi-circuit cable state type diagnosis device 400 provided in an embodiment of the present application;

[0074] Figure 5 It is a schematic structural diagram of an electronic device 500 provided in an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0075] The "multiple" involved in the embodiments of the present application means greater than or equal to two. It should be noted that in the description of the embodiments of the present application, terms such as "first" and "second" are only used for the purpose of distinguishing descriptions, and cannot be understood as indicating or implying relative importance, nor can they be understood as indicating or implying an order.

[0076] To make the above objects, features, and advantages of the present application more obvious and understandable, the embodiments of the present application will be further described in detail below with reference to the accompanying drawings and specific implementation manners. It can be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. In addition, it should be noted that for the convenience of description, only parts related to the present application are shown in the accompanying drawings, not all structures.

[0077] Traditional cable defect diagnosis methods mainly rely on manual inspections and simple electrical signal analysis. The method of manual inspections is not only inefficient but also prone to missing potential defect hazards, making it difficult to meet the requirements of modern power systems for the efficiency and accuracy of monitoring and diagnosis.

[0078] Regarding the method of electrical signal analysis in the prior art, the applicant has found through research that the prior art has poor adaptability to complex cable systems because the feature extraction objects for cables in the prior art are all electrical or non-electrical parameters of a single-loop cable, and it does not consider that in a multi-loop line, when a defect occurs in one loop, not only will the ground loop current of the defective loop change significantly, but the ground loop currents of other non-defective loops will also change due to electromagnetic induction. Under such relatively single input conditions, since the defect current waveforms in some cable defect cases are relatively similar, there may be a high probability of misjudgment, and the type of defect cannot be effectively determined, resulting in a low accuracy rate of defect diagnosis. In summary, the accuracy and efficiency of cable defect diagnosis in the prior art are low.

[0079] Based on this, the embodiments of the present application provide a method, device, equipment, and medium for diagnosing the state types of multi-loop cables. In this method, multi-loop cable three-phase ground loop current data corresponding to the multi-loop cables to be diagnosed is obtained; the multi-loop cable three-phase ground loop current data is input into a target cable state type diagnosis model, and the state type corresponding to the cable to be diagnosed is output. The target cable state type diagnosis model is generated by training a convolutional neural network model with three branches in parallel based on the multi-loop cable three-phase ground loop current data.

[0080] In this way, by inputting the three-phase grounding loop current data of multiple cables into the target cable state type diagnosis model, the state type of the multiple cables to be diagnosed can be obtained without further analysis, improving the efficiency of state type diagnosis. Since the target cable state type diagnosis model is generated by training a three-branch parallel convolutional neural network model based on the three-phase grounding loop current data of multiple cables, the three-branch structure of the convolutional neural network after structural adjustment is used to extract features and train the three-phase grounding loop current data of multiple cables. Compared with the prior art, the present application performs more comprehensive feature extraction, avoids the limitations of the single-loop feature extraction mode, reduces the probability of misjudgment or missed judgment, and improves the accuracy of state type diagnosis.

[0081] To facilitate the understanding of the specific implementation of the method for diagnosing the state type of multiple cables provided in the embodiments of the present application, the following will be described with reference to the accompanying drawings.

[0082] It should be noted that the main body for implementing the method for diagnosing the state type of multiple cables may be the device for diagnosing the state type of multiple cables provided in the embodiments of the present application, or may be carried in an electronic device or a functional module of an electronic device. The electronic device in the embodiments of the present application may be any device capable of implementing the method for diagnosing the state type of multiple cables in the embodiments of the present application, for example, it may be an Internet of Things (IoT) device.

[0083] Please refer to Figure 1 , Figure 1 which is a schematic flow chart of a method for diagnosing the state type of multiple cables provided in the embodiments of the present application. This method can be applied to a device for diagnosing the state type of multiple cables. The device for diagnosing the state type of multiple cables may be, for example, Figure 4 the device 400 for diagnosing the state type of multiple cables shown in Figure 5 , or the device for diagnosing the state type of multiple cables may also be a functional module integrated in the

[0084] In the embodiments of the present application, for example, the following steps may be included:

[0085] S1: Obtain the three-phase grounding loop current data of multiple cables corresponding to the multiple cables to be diagnosed.

[0086] It should be noted that "multiple" refers to multiple loops. The three-phase grounding loop current data of multiple cables may include, for example, the three-phase grounding loop current data of the first-loop cable and the three-phase grounding loop current data of the second-loop cable. "Three-phase" refers to the three phase lines contained in the cable, which respectively represent the three phases A, B, and C. The three-phase grounding loop current data of multiple cables may also be expressed as the A, B, and C three-phase grounding loop current data of multiple cables.

[0087] S2: Input the three-phase grounding loop current data of multiple cables into the target cable status type diagnosis model, and output the status type corresponding to the cable to be diagnosed.

[0088] It should be noted that the target cable status type diagnosis model is generated by training a three-branch parallel Convolutional Neural Networks (CNN) model based on the three-phase grounding loop current data of multiple cables.

[0089] It should be noted that the status types include normal types and defect types, and the defect types can include coaxial cable breakage, metal sheath grounding, cross-bonding box water ingress, or phase conversion failure.

[0090] In this way, compared with analyzing the data of a single-circuit cable in the prior art, in the embodiment of the present application, the target cable status type diagnosis model is generated by training a three-branch parallel convolutional neural network model based on the three-phase grounding loop current data of multiple cables. The target cable status type diagnosis model obtained by training with the three-branch parallel convolutional neural network model analyzes the three-phase grounding loop current data of multiple cables A, B, and C. The present application performs more comprehensive feature extraction, avoids the limitations in the single-loop feature extraction mode, reduces the probability of misjudgment or missed judgment, realizes real-time monitoring of the operating status of multiple cables and defect identification, and improves the accuracy and efficiency of the diagnosis of the status type of multiple cables.

[0091] In a possible implementation manner, the process of obtaining the target cable status type diagnosis model in the embodiment of the present application may include the following steps, for example:

[0092] S01: Divide the target sample set to obtain a target training sample set and a target test sample set.

[0093] It should be noted that the target sample set includes multiple groups of three-phase loop current data of multiple cables and the corresponding true status type labels for each group of three-phase loop current data of multiple cables.

[0094] In the specific implementation process, the data set can be divided into an 80% target training sample set and a 20% target test sample set. The target training sample set is used for model training, and the target test sample set is used to evaluate the performance of the trained model. The embodiment of the present application does not limit the division ratio of the target training sample set and the target test sample set.

[0095] S02: Train a three-branch parallel convolutional neural network model based on the target training sample set to generate an initial cable status type diagnosis model.

[0096] Please refer to Figure 3 , which is a schematic structural diagram of a three-branch parallel convolutional neural network model provided by the embodiment of the present application.

[0097] It should be noted that the convolutional neural network model with branch parallelism in the embodiments of the present application includes a first branch, a second branch, and a third branch.

[0098] In a possible implementation manner, S02 provided in the embodiments of the present application may include:

[0099] S021, obtaining the ground loop current data under defective conditions and the ground loop current data under normal conditions in the target training sample set.

[0100] S022, inputting the ground loop current data under defective conditions and the ground loop current data under normal conditions into the convolutional neural network model with three branches in parallel through the multi-head attention mechanism, and outputting a first feature, a second feature, and a third feature.

[0101] It should be noted that the multi-head attention mechanism enhances the expressive ability and learning ability of the trained model by parallelly learning and processing different parts of the input data.

[0102] S023, performing branch merging on the first feature, the second feature, and the third feature, and outputting the predicted defect type label corresponding to the target training sample set.

[0103] It should be noted that the first branch is used to extract features from the ground loop current data under defective conditions and output the first feature, the second branch is used to extract features from the ground loop current data under normal conditions and output the second feature, and the third branch is used to extract features from both the ground loop current data under defective conditions and the ground loop current data under normal conditions and output the third feature.

[0104] In the embodiments of the present application, it is set that the first branch includes two convolutional layers and two pooling layers, the second branch includes one convolutional layer and one pooling layer, and the third branch includes one convolutional layer. Since the first branch extracts features from the ground loop current data under defective conditions, and compared with the ground loop current data under normal conditions, the features possessed by the ground loop current data under defective conditions are richer, so it is set that the first branch includes two convolutional layers and two pooling layers, and the second branch includes one convolutional layer and one pooling layer. And the third branch processes the ground loop current data under all conditions, and only one convolutional layer can achieve the feature extraction effect.

[0105] In the specific implementation process, the convolutional layer and the pooling layer are connected through the ReLU function. The ReLU function introduces a non-linear transformation, which can improve the sparsity of the model, alleviate the problem of gradient disappearance, and improve the calculation efficiency.

[0106] Of course, the above is only an example, and the specific number of layers set for each branch is not set in the embodiments of the present application. It should be noted that when setting the number of convolutional layers and pooling layers, it is necessary to comprehensively consider feature extraction and model training speed. If the number of layers is too small, accurate feature extraction cannot be performed. If the number of layers is too large, the model training speed will be reduced.

[0107] In a possible implementation manner, S022 in the embodiments of the present application may include:

[0108] S0221, input the grounding loop current data of the defective working condition into the first branch through the multi-head attention mechanism, and output the first feature.

[0109] S0222, input the grounding loop current data of the normal working condition into the second branch through the multi-head attention mechanism, and output the second feature.

[0110] S0223, input the grounding loop current data of the defective working condition and the grounding loop current data of the normal working condition into the third branch through the multi-head attention mechanism, and output the third feature.

[0111] In the embodiments of the present application, there is no sequential execution order for steps S0221, S0222, and S0223, that is, steps S0221, S0222, and S0223 can be executed simultaneously.

[0112] In a possible implementation manner, S023 in the embodiments of the present application may include:

[0113] S0231, perform local feature extraction on the first feature, the second feature, and the third feature through a convolutional layer, and output local features.

[0114] S0232, based on a flattening layer, convert the local features into one-dimensional vectors.

[0115] S0233, perform integration processing on the one-dimensional vectors through a fully connected layer, and output a predicted defect type label.

[0116] In the embodiments of the present application, S0231, S0232, and S0233 are the forward propagation processes of a three-branch parallel convolutional neural network model. During the entire forward propagation process, global feature extraction is performed to predict whether a defect occurs, local feature extraction is performed to determine the defect type, and finally a predicted defect type label is output through a fully connected layer.

[0117] In the embodiments of the present application, considering the characteristics of the grounding loop current data of multiple cable lines comprehensively, a non-single grounding loop current data input mode is used, and the features extracted by different convolutional layers are integrated, enhancing the richness and robustness of the overall feature representation, thereby improving the accuracy of state type diagnosis.

[0118] S024. Calculate the error result between the true state type label and the predicted state type label according to the loss function of the three-branch parallel convolutional neural network model.

[0119] In the embodiments of the present application, for example, CrossEntropyLoss can be used as the loss function to calculate the error result between the predicted state type label and the true state type label, that is, the difference between the predicted result and the true result is measured by the loss function CrossEntropyLoss.

[0120] S025. Based on the error result, calculate the gradient corresponding to the model parameters of the three-branch parallel convolutional neural network model.

[0121] S026. Update and iterate the model parameters of the three-branch parallel convolutional neural network model according to the gradient to complete the training of the three-branch parallel convolutional neural network model and generate an initial cable state type diagnosis model.

[0122] It should be noted that batch gradient descent or the Adam optimizer can be used to update and iterate the model parameters. Among them, both batch gradient descent and the Adam optimizer can dynamically adjust the learning rate according to the information of historical gradients, improving the training efficiency and performance of the training model.

[0123] In the embodiments of the present application, S024, S025, and S026 are the backpropagation process of the three-branch parallel convolutional neural network model. By continuously updating and iterating the model parameters through the backpropagation process, the generalization ability and diagnostic performance of the obtained initial cable state type diagnosis model can be improved.

[0124] S03: Test the initial cable state type diagnosis model according to the target test sample set to obtain the test accuracy rate.

[0125] S04: After the test accuracy rate reaches the preset accuracy rate threshold, obtain the target cable state type diagnosis model as the initial cable state type diagnosis model.

[0126] In the embodiments of the present application, when the test accuracy rate of the initial cable state type diagnosis model reaches the preset accuracy rate threshold, it proves that the initial cable state type diagnosis model can be put into actual use, that is, the initial cable state type diagnosis model can be used as the target cable state type diagnosis model.

[0127] In a possible implementation, it is also possible to determine that the initial cable state type diagnosis model can be put into actual use when the loss function converges or the confusion matrix shows good classification performance. Among them, it is judged whether the loss function converges by plotting the training loss and test loss curves, and the confusion matrix is used to evaluate the classification performance of the model on different defect categories, which is converted into a percentage form and plotted as an image to intuitively display the classification results.

[0128] In the embodiments of the present application, deep learning models are used to automatically extract features and classify, without relying on manual experience and complex signal processing steps, which simplifies the operation process and system complexity. The grounding loop current characteristics of multiple cable lines are introduced, the grounding loop current of the defective loop and the grounding loop current of the non-defective loop are introduced together, and the convolutional neural network structure is adjusted. A convolutional neural network model structure with three-branch parallelism is proposed. Three-layer branches are used to extract features and train the grounding loop current data. Multi-head feature extraction is performed on the defective loop, non-defective loop, and total loop data for model training, avoiding the limitations of the single-loop feature extraction mode, while extracting the common features of the defective loop and the non-defective loop, reducing the probability of misjudgment or missed judgment, and thus improving the diagnostic accuracy of the target cable state type diagnosis model.

[0129] In a possible implementation, before step S01, the embodiments of the present application may further include:

[0130] S001, obtaining multiple sets of three-phase grounding loop current data of multiple cable lines.

[0131] It should be noted that the three-phase grounding loop current data of multiple cable lines includes simulated calculation of three-phase grounding loop current data of multiple cable lines and on-site test of three-phase grounding loop current data of multiple cable lines. The simulated calculation of three-phase grounding loop current data of multiple cable lines is the three-phase grounding loop current data obtained by simulation calculation, and the on-site test of three-phase grounding loop current data of multiple cable lines is the three-phase grounding loop current data obtained by on-site test.

[0132] It should be noted that using only the simulated calculation or only the on-site test of the three-phase grounding loop current data of multiple cable lines for model training will reduce the accuracy in actual use. Moreover, the simulated calculation is less affected by the real environment, and the extracted features are more targeted at single defects, but there is still a certain gap between the current waveforms obtained by simulation and the actual current waveforms; the current waveforms obtained by on-site test are closer to the actual current waveforms, but affected by various real factors, it is difficult to obtain defective data for on-site test. If defective data is obtained by artificially creating defects, it will affect the safe operation of the power grid. Therefore, the simulated calculation of three-phase grounding loop current data of multiple cable lines is also required for compensation.

[0133] In summary, the multi - circuit cable three - phase grounding circulating current data obtained in the embodiments of the present application includes two types of multi - circuit cable three - phase grounding circulating current data, namely simulation calculation and on - site test data, which can improve the accuracy of the target cable state type diagnosis model generated by subsequent training for state type diagnosis.

[0134] S002, classify multiple sets of multi - circuit cable three - phase grounding circulating current data according to the state types corresponding to the multi - circuit cable three - phase grounding circulating current data to obtain an initial sample data set.

[0135] It should be noted that the state types include normal types and defect types; the defect types include coaxial cable breakage, metal sheath grounding, cross - bonding box water ingress, or phase conversion failure. Among them, coaxial cable breakage includes coaxial cable breakage with different phase sequence combinations. Of course, the above state types are only for illustrative purposes, and the embodiments of the present application do not limit the specific state types.

[0136] To facilitate the understanding of the coaxial cable breakage with different phase sequence combinations in the embodiments of the present application, an example is given below. For example, the multi - circuit cable A, B, C three - phase grounding circulating current data includes the first - loop cable A, B, C three - phase grounding circulating current data and the second - loop cable A, B, C three - phase grounding circulating current data. Then, the coaxial cable breakage with different phase sequence combinations includes coaxial breakage ABC - ABC, coaxial breakage ACB - ABC, coaxial breakage BAC - ABC, coaxial breakage BCA - ABC, coaxial breakage CAB - ABC, and coaxial breakage CBA - ABC.

[0137] In the specific implementation process, the multi - circuit cable three - phase current data belonging to the same state type can be saved in the same Excel file, and a corresponding state type label can be assigned to each file for subsequent comprehensive diagnosis.

[0138] For example, if the target cable state type diagnosis model is to be able to distinguish normal types, coaxial cable breakage with different phase sequence combinations, metal sheath grounding, cross - bonding box water ingress, or phase conversion failure, then the storage files 1, 2, 3... 10 of the multi - circuit cable three - phase grounding circulating current data storing state types such as normal type, coaxial breakage ABC - ABC, coaxial breakage ACB - ABC, coaxial breakage BAC - ABC, coaxial breakage BCA - ABC, coaxial breakage CAB - ABC, coaxial breakage CBA - ABC, metal sheath grounding, cross - bonding box water ingress, and phase conversion failure can be obtained respectively. It should be noted that the above is only an example, and the embodiments of the present application do not limit the manifestation form of the state type label. In the specific implementation process, the corresponding storage files of the multi - circuit cable three - phase grounding circulating current data required for model training are merged to obtain an initial sample data set, and the initial sample data set includes multiple sets of multi - circuit cable three - phase grounding circulating current data.

[0139] S003. Assign corresponding true status type labels to the three-phase grounding loop current data of multi-circuit cables according to the status type.

[0140] Among them, the true status type labels correspond to the status type. By assigning the corresponding true status type labels, the model can distinguish the true status type corresponding to the three-phase grounding loop current data of multi-circuit cables more quickly.

[0141] For example, if the three-phase grounding loop current data of a multi-circuit cable corresponds to the normal status, assign the corresponding true status type label "Normal Status" "1" to the three-phase grounding loop current data of the multi-circuit cable. In the specific implementation process, for the convenience of model training, LabelEncoder can be used to encode the true status type labels into an integer format, that is, encode the normal status as "1". Again, for example, "1", "2", "3", "4", "5", "6", "7", "8", "9", "10" can be used as the encodings corresponding to the status type labels of normal type, coaxial fracture ABC-ABC, coaxial fracture ACB-ABC, coaxial fracture BAC-ABC, coaxial fracture BCA-ABC, coaxial fracture CAB-ABC, coaxial fracture CBA-ABC, metal sheath grounding, cross-bonding box water ingress, and transposition failure, etc.

[0142] S004. Clean the three-phase grounding loop current data of multi-circuit cables to obtain the three-phase grounding loop current data of multi-circuit cables after data cleaning.

[0143] It should be noted that data cleaning can include null value processing and invalid value processing.

[0144] In the embodiments of the present application, data cleaning is used to process the null values and values that cannot be converted to floating-point numbers in the three-phase grounding loop current data of multi-circuit cables, ensuring the integrity and accuracy of the three-phase grounding loop current data of multi-circuit cables.

[0145] S005. Obtain an intermediate sample set.

[0146] It should be noted that the intermediate sample set includes the three-phase grounding loop current data of multi-circuit cables after data cleaning and the corresponding true status type labels.

[0147] S006. Based on a sliding window of a preset length, convert the three-phase grounding loop current data of multi-circuit cables after data cleaning in the intermediate sample set to obtain a target sample set.

[0148] It should be noted that the target sample set includes time series data of a preset length and the corresponding true status type labels.

[0149] Compared with the traditional signal processing and machine learning methods in the prior art, in the embodiments of the present application, based on a sliding window with a preset length, the multi-circuit cable three-phase grounding circulating current data after data cleaning in the intermediate sample set is converted to obtain a target sample set, realizing efficient processing of time series information.

[0150] In the embodiments of the present application, by introducing the multi-circuit cable three-phase grounding circulating current data and performing a series of processes on the obtained initial sample set to obtain a target sample set, the diagnostic accuracy of the target cable state type diagnostic model generated subsequently through the target sample set is improved.

[0151] See Figure 4 , the embodiments of the present application further provide a multi-circuit cable state type diagnostic device 400, and the device 400 includes:

[0152] An acquisition unit 401, configured to acquire multi-circuit cable three-phase grounding circulating current data corresponding to the multi-circuit cable to be diagnosed;

[0153] A diagnosis unit 402, configured to input the multi-circuit cable three-phase grounding circulating current data into the target cable state type diagnostic model and output the state type corresponding to the cable to be diagnosed; so the target cable state type diagnostic model is generated by training a three-branch parallel convolutional neural network model based on the multi-circuit cable three-phase grounding circulating current data.

[0154] Optionally, the device 400 further includes a model training unit 403;

[0155] The model training unit 403 includes:

[0156] A sample set obtaining module, configured to divide the target sample set to obtain a target training sample set and a target test sample set; the target sample set includes multi-circuit cable three-phase circulating current data and the true state type labels corresponding to the multi-circuit cable three-phase circulating current data;

[0157] A model training module, configured to train the three-branch parallel convolutional neural network model based on the target training sample set to generate an initial cable state type diagnostic model;

[0158] A test model, configured to test the initial cable state type diagnostic model according to the target test sample set to obtain a test accuracy rate;

[0159] A model obtaining module, configured to obtain the target cable state type diagnostic model as the initial cable state type diagnostic model after the test accuracy rate reaches a preset accuracy rate threshold.

[0160] Optionally, the model training module is specifically configured to:

[0161] The target sample set is segmented to obtain a target training sample set and a target test sample set; the target sample set includes multi-circuit cable three-phase circulating current data and the true state type labels corresponding to the multi-circuit cable three-phase circulating current data;

[0162] Based on the target training sample set, the three-branch parallel convolutional neural network model is trained to generate an initial cable state type diagnosis model;

[0163] According to the target test sample set, the initial cable state type diagnosis model is tested to obtain the test accuracy rate;

[0164] After the test accuracy rate reaches the preset accuracy rate threshold, the target cable state type diagnosis model is obtained as the initial cable state type diagnosis model.

[0165] Optionally, the three-branch parallel convolutional neural network model includes a first branch, a second branch, and a third branch. By means of the multi-head attention mechanism, the grounding circulating current data under the defective working condition and the grounding circulating current data under the normal working condition are input into the three-branch parallel convolutional neural network model, and the first feature, the second feature, and the third feature are output, including:

[0166] Through the multi-head attention mechanism, the grounding circulating current data under the defective working condition is input into the first branch, and the first feature is output; the first branch is used to extract features from the grounding circulating current data under the defective working condition and output the first feature;

[0167] Through the multi-head attention mechanism, the grounding circulating current data under the normal working condition is input into the second branch, and the second feature is output; the second branch is used to extract features from the grounding circulating current data under the normal working condition and output the second feature;

[0168] Through the multi-head attention mechanism, the grounding circulating current data under the defective working condition and the grounding circulating current data under the normal working condition are input into the third branch, and the third feature is output; the third branch is used to extract features from the grounding circulating current data under the defective working condition and the grounding circulating current data under the normal working condition simultaneously and output the third feature.

[0169] Optionally, the branch merging of the first feature, the second feature, and the third feature is performed to output the predicted state type label corresponding to the target training sample set, including:

[0170] Local feature extraction is performed on the first feature, the second feature, and the third feature through a convolutional layer, and local features are output;

[0171] Based on a flattening layer, the local features are converted into one-dimensional vectors;

[0172] Integrate and process the one-dimensional vector through the fully connected layer, and output the predicted state type label.

[0173] Optionally, the sample set obtaining module is further configured to:

[0174] Obtain multiple groups of three-phase grounding loop current data of multi-circuit cables, where the three-phase grounding loop current data of multi-circuit cables includes simulated calculation of three-phase grounding loop current data of multi-circuit cables and on-site test of three-phase grounding loop current data of multi-circuit cables;

[0175] Classify multiple groups of the three-phase grounding loop current data of multi-circuit cables according to the state types corresponding to the three-phase grounding loop current data of multi-circuit cables, and obtain an initial sample data set;

[0176] According to the state type, assign the corresponding true state type label to the three-phase grounding loop current data of multi-circuit cables;

[0177] Perform data cleaning on the three-phase grounding loop current data of multi-circuit cables to obtain the three-phase grounding loop current data of multi-circuit cables after data cleaning; the data cleaning includes null value processing and invalid value processing;

[0178] Obtain an intermediate sample set, where the intermediate sample set includes the three-phase grounding loop current data of multi-circuit cables after data cleaning and the corresponding true state type label;

[0179] Based on a sliding window with a preset length, convert the three-phase grounding loop current data of multi-circuit cables after data cleaning in the intermediate sample set to obtain a target sample set, where the target sample set includes time series data with a preset length and the corresponding true state type label.

[0180] Optionally, the state types include: normal type and defect type, and the defect type includes coaxial cable breakage, metal sheath grounding, cross-bonding box water ingress, or transposition failure.

[0181] It should be noted that for the specific implementation manner and the achieved technical effects of the device 400, reference can be made to the relevant descriptions in the Figure 1 method shown.

[0182] In addition, an embodiment of the present application further provides an electronic device 500, as Figure 5 shown, the electronic device 500 includes a processor 501 and a memory 502:

[0183] The memory 502 is used to store a computer program;

[0184] The processor 501 is configured to execute according to the computer program Figure 1 the provided method for diagnosing the state type of multi-circuit cables.

[0185] In addition, an embodiment of the present application further provides a computer-readable storage medium, which is used to store a computer program, and the computer program is used to execute the multi-cable state type diagnosis method provided by the embodiment of the present application.

[0186] From the description of the above embodiments, those skilled in the art can clearly understand that all or part of the steps in the above embodiment methods can be implemented by means of software plus a general hardware platform. Based on such an understanding, the technical solution of the present application can be embodied in the form of a software product, and the computer software product can be stored in a storage medium, such as a read-only memory (ROM) / RAM, magnetic disk, optical disk, etc., including several instructions for causing a computer device (which can be a personal computer, a server, or a network communication device such as a router) to execute the methods described in each embodiment or some parts of the embodiments of the present application.

[0187] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiment. The device embodiments described above are only illustrative. The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the objectives of the embodiment solution. Those of ordinary skill in the art can understand and implement without creative work.

[0188] The above is only the preferred embodiment of the present application, and is not used to limit the protection scope of the present application. It should be noted that for those of ordinary skill in the art in the technical field of the present application, without departing from the premise of the present application, several improvements and retouches can still be made, and these improvements and retouches should also be regarded as the protection scope of the present application.

Claims

1. A method for diagnosing multi-circuit cable status types, characterized in that: The method comprises: Obtaining the three-phase grounding circulation current data of multiple cables corresponding to the multiple cables to be diagnosed; The multi-circuit cable three-phase grounding circulation data is input into the target cable state type diagnosis model, and the state type corresponding to the cable to be diagnosed is output; therefore, the target cable state type diagnosis model is generated by training a three-branch parallel convolutional neural network model based on the multi-circuit cable three-phase grounding circulation data.

2. The method according to claim 1, characterized in that Also includes: The target sample set is segmented to obtain a target training sample set and a target test sample set; the target sample set includes multiple cable three-phase circulating current data and real state type labels corresponding to the multiple cable three-phase circulating current data; Training the three-branch parallel convolutional neural network model based on the target training sample set to generate an initial cable status type diagnosis model; Testing the initial cable status type diagnosis model according to the target test sample set to obtain a test accuracy rate; After the test accuracy reaches a preset accuracy threshold, the target cable state type diagnosis model is obtained as the initial cable state type diagnosis model.

3. The method according to claim 2, characterized in that The training of the three-branch parallel convolutional neural network model based on the target training sample set to generate an initial cable status type diagnosis model includes: Acquire the ground loop current data of defective working condition and the ground loop current data of normal working condition in the target training sample set; The defective working condition ground loop current data and the normal working condition ground loop current data are input into the three-branch parallel convolutional neural network model through a multi-head attention mechanism, and a first feature, a second feature and a third feature are output; Perform branch merging on the first feature, the second feature, and the third feature, and output a predicted state type label corresponding to the target training sample set; Calculate the error result between the real state type label and the predicted state type label according to the loss function of the three-branch parallel convolutional neural network model; Based on the error result, calculating the gradients corresponding to the model parameters of the three-branch parallel convolutional neural network model; The model parameters of the three-branch parallel convolutional neural network model are updated and iterated according to the gradient to complete the training of the three-branch parallel convolutional neural network model and generate the initial cable state type diagnosis model.

4. The method according to claim 3, characterized in that The three-branch parallel convolutional neural network model includes a first branch, a second branch and a third branch. The defective working condition ground loop current data and the normal working condition ground loop current data are input into the three-branch parallel convolutional neural network model through a multi-head attention mechanism, and a first feature, a second feature and a third feature are output, including: Through the multi-head attention mechanism, the defective working condition ground loop current data is input into the first branch, and the first feature is output; the first branch is used to extract features from the defective working condition ground loop current data and output the first feature; Through the multi-head attention mechanism, the normal working condition ground loop current data is input into the second branch, and the second feature is output; the second branch is used to extract features from the normal working condition ground loop current data and output the second feature; Through the multi-head attention mechanism, the defective working condition ground loop current data and the normal working condition ground loop current data are input into the third branch, and the third feature is output; the third branch is used to simultaneously extract features from the defective working condition ground loop current data and the normal working condition ground loop current data and output the third feature.

5. The method according to claim 3, characterized in that: The branching and merging of the first feature, the second feature, and the third feature to output a predicted state type label corresponding to the target training sample set includes: Performing local feature extraction on the first feature, the second feature, and the third feature through a convolution layer, and outputting local features; Based on the flattening layer, the local features are converted into a one-dimensional vector; The one-dimensional vector is integrated through the fully connected layer to output the predicted state type label.

6. The method according to claim 2, characterized in that Before segmenting the target sample set to obtain the target training sample set and the target test sample set, the method further includes: Acquire multiple groups of multi-circuit cable three-phase grounding circulation data, wherein the multi-circuit cable three-phase grounding circulation data includes simulation calculation multi-circuit cable three-phase grounding circulation data and field test multi-circuit cable three-phase grounding circulation data; Classify multiple groups of the three-phase grounding circulating current data of the multiple-circuit cables according to the state types corresponding to the three-phase grounding circulating current data of the multiple-circuit cables to obtain an initial sample data set; According to the state type, assigning corresponding real state type labels to the multi-circuit cable three-phase grounding circulation current data; Performing data cleaning on the three-phase grounding circulation data of the multiple cables to obtain the three-phase grounding circulation data of the multiple cables after data cleaning; the data cleaning includes null value processing and invalid value processing; Obtaining an intermediate sample set, the intermediate sample set including the multiple-circuit cable three-phase grounding circulating current data after data cleaning and the corresponding real state type label; Based on a sliding window of preset length, the three-phase grounding circulating current data of multiple cables after data cleaning in the intermediate sample set are converted to obtain a target sample set, which includes time series data of preset length and the corresponding real state type label.

7. The method according to any one of claims 1 to 6, characterized in that: The state types include: a normal type and a defect type, and the defect type includes a coaxial cable break, a metal sheath grounding, water ingress into a cross-connection box, or a transposition failure.

8. A multi-circuit cable status type diagnostic device, characterized in that: The device comprises: An acquisition unit, used for acquiring the three-phase grounding circulation current data of multiple cables corresponding to the multiple cables to be diagnosed; The diagnostic unit is used to input the three-phase grounding circulation data of the multiple cables into the target cable state type diagnostic model, and output the state type corresponding to the cable to be diagnosed; therefore, the target cable state type diagnostic model is generated by training a three-branch parallel convolutional neural network model based on the three-phase grounding circulation data of the multiple cables.

9. An electronic device, characterized in that: The electronic device comprises a processor and a memory: The memory is used to store computer programs; The processor is configured to execute the method according to any one of claims 1 to 7 according to the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store a computer program, and the computer program is used to execute the method according to any one of claims 1 to 7.

Images