Power cable line deterioration diagnosis system and method using database samples

Abstract

The application discloses a power cable line degradation diagnosis system and method using a database sample, and the method comprises the following steps: 1) extracting the harmonic current characteristics of the sample cable, marking the harmonic current characteristics with expert evaluation data of the sample cable as labels, and dividing the set of the marked harmonic current characteristics into a training set and a test set; 2) constructing a convolutional neural network corresponding to the expert evaluation data, wherein the convolutional neural network corresponding to the expert evaluation data comprises an input layer, a plurality of convolutional layers, a maximum value pooling layer and an output layer; 3) inputting the training set into the convolutional neural network corresponding to the expert evaluation data to obtain a target model; and 4) collecting the harmonic current characteristics of a cable to be diagnosed and inputting the harmonic current characteristics into the target model to obtain the degradation degree of the cable to be diagnosed. According to the embodiment of the application, the cable circuit insulation degradation analysis can be performed without power interruption.

Description

Technical Field

[0001] This invention relates to the field of power facility operation and maintenance technology, and more specifically to a power cable line deterioration diagnosis system and method utilizing database samples. Background Technology

[0002] Currently, cross-linked polyethylene is used for the external insulation layer of cables. However, due to the age of the cables, some cable lines experience aging, which may lead to cable body deterioration due to insulation degradation, thus restricting the power supply reliability of the cable lines. Therefore, the insulation degradation status and lifespan of cables in operation have become issues that urgently need to be assessed.

[0003] For example, patent application number 202111047797.7 discloses a method for assessing the degradation state of cable insulation materials based on dielectric sound velocity, relating to the field of cable degradation state assessment technology. The method includes: selecting a cable of the same model and known service life as the cable to be assessed, processing it to obtain cable slice samples; heating the samples to a preset temperature; applying pulses to the samples using a pulse power supply, and acquiring real-time data of the electroacoustic pulse waveform of the samples; calculating the dielectric sound velocity of the samples at multiple preset temperatures based on the acquired data; plotting the relationship curve between the service life of the cable with known service life, preset temperature, and dielectric sound velocity; measuring the dielectric sound velocity of the cable to be assessed at its operating temperature, and obtaining the service life of the cable to be assessed based on the relationship curve. This invention uses a pulse power supply to test the electroacoustic pulse waveform of the cable, resulting in a simple test circuit, convenient testing, and low equipment cost. It utilizes the data from the waveform to calculate the dielectric sound velocity, achieving a non-destructive assessment of the cable's condition.

[0004] Existing conventional testing methods are insufficient for analyzing insulation degradation and predicting the lifespan of cables in energized operation; most such analyses and predictions are performed in laboratories. Furthermore, current cable lifespan prediction models are relatively simplistic, lacking generalizability and sustainability, and failing to accurately simulate the actual condition of cables under real-world conditions. Therefore, determining how to perform cable circuit insulation degradation analysis without power interruption is a pressing technical problem that needs to be solved. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a power cable line deterioration diagnosis system and method using database samples to perform cable circuit insulation deterioration analysis without power interruption.

[0006] The present invention solves the above-mentioned technical problems through the following technical solutions:

[0007] This invention provides a method for diagnosing the deterioration of power cable lines using database samples, the method comprising:

[0008] 1) Extract the harmonic current characteristics of the sample cable, use the expert evaluation data of the sample cable as labels to label the harmonic current characteristics, and divide the set of labeled harmonic current characteristics into training set and test set.

[0009] 2) Construct a convolutional neural network corresponding to the expert evaluation data, wherein the convolutional neural network corresponding to the expert evaluation data includes an input layer, several convolutional layers, a max pooling layer, and an output layer;

[0010] 3) Input the training set into the convolutional neural network corresponding to the expert evaluation data to obtain a pre-trained model; use the test set to test the pre-trained model, adjust the training parameters according to the test results and retrain, and return to execute the step of inputting the training set into the convolutional neural network corresponding to the expert evaluation data until the convolutional neural network corresponding to the expert evaluation data converges to obtain the target model;

[0011] 4) Collect the harmonic current characteristics of the cable to be diagnosed and input them into the target model to obtain the degree of degradation of the cable to be diagnosed.

[0012] Optionally, the extraction of harmonic current characteristics of the sample cable includes:

[0013] Obtain the load time distribution of the sample cable, and determine the peak load time of the sample cable based on the load time distribution;

[0014] Harmonic current characteristics at the peak load moment.

[0015] Optionally, the extraction of harmonic current characteristics of the sample cable includes:

[0016] Obtain the load time distribution of the sample cable, and determine the average load of the sample cable based on the load time distribution;

[0017] Harmonic current characteristics were collected during periods when the load was greater than the average load value.

[0018] Optionally, the step of labeling harmonic current characteristics using expert evaluation data for the sample cables includes:

[0019] The temperature distribution characteristics of the sample cable are acquired synchronously, and the expert evaluation data of the sample cable is used as a label to mark the temperature distribution characteristics.

[0020] Optionally, when obtaining the temperature distribution characteristics of the sample cable, the method further includes:

[0021] The magnetic field distribution characteristics of the sample cable are acquired simultaneously, and the expert evaluation data of the sample cable are used as tags to label the magnetic field distribution characteristics.

[0022] Optionally, the set of labeled harmonic current features is divided into a training set and a test set, including:

[0023] The labeled harmonic current features are expanded using color histogram equalization, horizontal mirroring, Gaussian blurring, and the addition of random noise. The expanded sample set is then divided into a training set and a test set.

[0024] Optionally, the training process of the convolutional neural network includes:

[0025] A convolutional neural network was trained using labeled harmonic current features as training samples, and the formula was used...

[0026] Calculate the harmonic error loss of the convolutional neural network at the current iteration, where,

[0027] loss1 represents the harmonic error loss of the detection result; S1 represents the number of harmonic current features; x i The expert evaluation degradation index corresponds to the i-th harmonic current characteristic; is the predicted degradation index corresponding to the i-th harmonic current characteristic output by the model; ∑ is the summation function;

[0028] The first pre-trained model is obtained when the harmonic error loss is below a set threshold.

[0029] The first pre-trained model is trained using the labeled temperature distribution features as training samples, and the formula is used... Calculate the temperature error loss of the convolutional neural network in the current iteration, where,

[0030] loss2 represents the temperature error loss of the detection result; S2 represents the number of temperature distribution features; y j The expert evaluation degradation index corresponds to the j-th temperature distribution feature; The predicted degradation index corresponds to the j-th temperature distribution feature output by the model.

[0031] When the temperature error loss is below a set threshold, a second pre-trained model is obtained;

[0032] The second pre-trained model is trained using the labeled magnetic field distribution features as training samples, and the formula is used... Calculate the magnetic field error loss of the convolutional neural network at the current iteration, where,

[0033] loss3 represents the magnetic field error loss of the detection result; S3 represents the number of harmonic current characteristics; z k The expert evaluation degradation index corresponds to the kth magnetic field distribution feature; This is the predicted degradation index corresponding to the kth magnetic field distribution feature output by the model.

[0034] The third pre-trained model is obtained when the magnetic field error loss is below a set threshold.

[0035] Using the formula Loss = λ1*loss1 + λ2*loss2 + loss3, calculate the comprehensive weighted loss of the third pre-trained model in the current iteration, until the comprehensive weighted loss converges to obtain the target model.

[0036] Loss is the comprehensive weighted loss of the convolutional neural network in the current iteration; λ1 is the weight corresponding to the harmonic current feature; λ2 is the weight corresponding to the temperature distribution feature;

[0037] This invention provides a pedestrian re-identification system based on feature-enhanced guided convolutional neural networks, the system comprising:

[0038] The sample generation module is used to extract the harmonic current characteristics of the sample cable, use expert evaluation data for the sample cable as labels to label the harmonic current characteristics, and divide the set of labeled harmonic current characteristics into training set and test set.

[0039] The training module is used to construct a convolutional neural network corresponding to the expert evaluation data. The convolutional neural network corresponding to the expert evaluation data includes an input layer, several convolutional layers, a max pooling layer, and an output layer.

[0040] The training set is input into the convolutional neural network corresponding to the expert evaluation data to obtain a pre-trained model; the pre-trained model is tested using the test set, the training parameters are adjusted and retrained according to the test results, and the process of inputting the training set into the convolutional neural network corresponding to the expert evaluation data is repeated until the convolutional neural network corresponding to the expert evaluation data converges to obtain the target model.

[0041] The output module is used to collect the harmonic current characteristics of the cable under diagnosis and input them into the target model to obtain the degree of degradation of the cable under diagnosis.

[0042] The present invention has the following advantages over the prior art:

[0043] This invention uses harmonic current characteristic data as raw data and artificial degradation evaluation data as sample labels to obtain training samples between harmonic current characteristics and cable degradation. Then, a neural network model is used to learn the hidden relationship between harmonic current characteristics and cable degradation in the training samples. After the neural network model is trained, the target model can be used to assess the degree of degradation of the cable to be diagnosed, thus achieving the effect of analyzing cable circuit insulation degradation without power interruption. Attached Figure Description

[0044] Figure 1 A schematic flowchart of a power cable line degradation diagnosis method using database samples provided in an embodiment of the present invention;

[0045] Figure 2 This is a schematic diagram of the structure of a power cable line deterioration diagnosis system utilizing database samples, provided in an embodiment of the present invention. Detailed Implementation

[0046] The embodiments of the present invention are described in detail below. These embodiments are implemented based on the technical solution of the present invention, and provide detailed implementation methods and specific operation processes. However, the scope of protection of the present invention is not limited to the following embodiments.

[0047] Example 1

[0048] Harmonics typically originate from three sources: poor power generation quality, harmonics generated by the power transmission and distribution system, and harmonics generated by electrical equipment. The third point is the primary cause of harmonic generation, and examples include nonlinear loads such as rectifiers, switching power supplies, UPS, frequency converters, and inverters. Figure 1 As shown, the device generates harmonics.

[0049] First, harmonics cause cable dielectric loss, increase transmission loss, raise leakage current, and increase partial discharge in dry cables, which may lead to single-phase grounding faults.

[0050] Furthermore, because the distributed capacitance of power cables amplifies harmonic currents, when the system load is low and the system voltage rises, the harmonic voltage also increases accordingly. The higher the rated voltage level of the cable, the greater the risk of instability in the cable dielectric caused by harmonics, and the more prone it is to failure.

[0051] Furthermore, increased cable line losses will cause the line temperature to rise. Since the conductors are covered with insulation and protective layers, temperature analysis is complex. The conductor temperature can be determined by analyzing the bare conductor.

[0052] Finally, referring to the chemical reaction kinetics rate equation, it can be seen that polymer materials have a thermal aging effect, which is mainly manifested in the fact that the life of cables decreases sharply as the temperature increases. When the average operating temperature of power cables increases by 10°C, the lifespan is only 1 / 4 of that under normal operating temperature.

[0053] Therefore, the overload and temperature rise caused by harmonics can affect the performance of cable insulation. Thus, harmonic analysis can, to some extent, diagnose the deterioration state of cable insulation.

[0054] Figure 1 This is a flowchart illustrating the power cable line degradation diagnosis method using database samples provided in an embodiment of the present invention, as shown below. Figure 1 As shown, the method includes:

[0055] S101: Extract the harmonic current characteristics of the sample cable, use the expert evaluation data of the sample cable as a label to label the harmonic current characteristics, and divide the set of labeled harmonic current characteristics into training set and test set.

[0056] For example, since the load peak places high demands on the cable insulation, the load time distribution of the sample cable can be obtained, and the load peak time of the sample cable can be determined based on the load time distribution; the harmonic current characteristics at the load peak time can be collected, and the harmonic current characteristics corresponding to the peak time can be used as sample data.

[0057] Furthermore, considering that cables operate under heavy loads for extended periods, leading to more severe insulation degradation due to cumulative effects, the inventors also obtained the load time distribution of the sample cable and determined its average load based on this distribution. Harmonic current characteristics were then collected during periods when the load exceeded the average load. For example, if the sample cable outputs a total of 12000 kWh of electricity at a voltage of 110 kV over 24 hours, the corresponding average load (average power) is 500 kW. If the average load during the period from 9:00:00 to 3:00:00 exceeds 500 kW, then the harmonic current characteristics during this period can be sampled.

[0058] Similarly, insulation degradation can lead to severe localized heating and partial discharge in the cable; localized heating alters the temperature distribution of the sample cable, while partial discharge alters its magnetic field distribution. Therefore, the temperature distribution characteristics of the sample cable can be acquired simultaneously, and expert evaluation data for the sample cable can be used as labels to mark these temperature distribution characteristics. The magnetic field distribution characteristics of the sample cable can also be acquired simultaneously, and expert evaluation data for the sample cable can be used as labels to mark these magnetic field distribution characteristics.

[0059] By applying the embodiments of the present invention, the degree of insulation degradation of cables can be evaluated from multiple perspectives, including harmonics, temperature, and magnetic fields, thereby further improving the accuracy of the evaluation.

[0060] In one specific implementation of this step, color histogram equalization, horizontal mirroring, Gaussian blurring, and adding random noise can be used to expand the labeled harmonic current features, and the expanded sample set can be divided into a training set and a test set.

[0061] S102: Construct a convolutional neural network corresponding to the expert evaluation data, wherein the convolutional neural network corresponding to the expert evaluation data includes an input layer, several convolutional layers, a max pooling layer, and an output layer.

[0062] Specifically, a convolutional neural network contains six convolutional units consisting of a 3x3 convolutional layer, a ReLU activation layer, and a BatchNorm normalization layer.

[0063] There are three max pooling layers. In practical applications, a backpropagation sigmoid layer can be inserted after each max pooling layer; the output layer is after the sigmoid layer.

[0064] The weights in a convolutional neural network are not pre-set, but rather generated iteratively during network training. The network autonomously selects its weights and performs dot products with feature maps at different levels to continuously guide the network's feature enhancement learning, particularly in its attention to detail. This allows expert evaluation data to focus more on cable degradation characteristics, avoiding background interference while enhancing adaptability to different scenarios.

[0065] S103: Input the training set into the convolutional neural network corresponding to the expert evaluation data to obtain a pre-trained model; use the test set to test the pre-trained model, adjust the training parameters according to the test results and retrain, and return to execute the step of inputting the training set into the convolutional neural network corresponding to the expert evaluation data until the convolutional neural network corresponding to the expert evaluation data converges to obtain the target model.

[0066] Specifically, the training process of setting the learning strategy and training parameters, and using the training set to train the convolutional neural network can be as follows:

[0067] A convolutional neural network was trained using labeled harmonic current features as training samples, and the formula was used...

[0068] Calculate the harmonic error loss of the convolutional neural network at the current iteration, where,

[0069] loss1 represents the harmonic error loss of the detection result; S1 represents the number of harmonic current features; x i The expert evaluation degradation index corresponds to the i-th harmonic current characteristic; is the predicted degradation index corresponding to the i-th harmonic current characteristic output by the model; ∑ is the summation function;

[0070] The first pre-trained model is obtained when the harmonic error loss is below a set threshold.

[0071] The first pre-trained model is trained using the labeled temperature distribution features as training samples, and the formula is used... Calculate the temperature error loss of the convolutional neural network in the current iteration, where,

[0072] loss2 represents the temperature error loss of the detection result; S2 represents the number of temperature distribution features; y j The expert evaluation degradation index corresponds to the j-th temperature distribution feature; The predicted degradation index corresponds to the j-th temperature distribution feature output by the model.

[0073] When the temperature error loss is below a set threshold, a second pre-trained model is obtained;

[0074] The second pre-trained model is trained using the labeled magnetic field distribution features as training samples, and the formula is used... Calculate the magnetic field error loss of the convolutional neural network at the current iteration, where,

[0075] loss3 represents the magnetic field error loss of the detection result; S3 represents the number of harmonic current characteristics; z k The expert evaluation degradation index corresponds to the kth magnetic field distribution feature; This is the predicted degradation index corresponding to the kth magnetic field distribution feature output by the model.

[0076] The third pre-trained model is obtained when the magnetic field error loss is below a set threshold.

[0077] Using the formula Loss = λ1*loss1 + λ2*loss2 + loss3, calculate the comprehensive weighted loss of the third pre-trained model in the current iteration, until the comprehensive weighted loss converges to obtain the target model.

[0078] Loss is the comprehensive weighted loss of the convolutional neural network in the current iteration; λ1 is the weight corresponding to the harmonic current feature; λ2 is the weight corresponding to the temperature distribution feature;

[0079] Typically, training parameters are also called hyperparameters. These parameters that need adjustment include: the learning rate, batch size, learning strategy, maximum number of iterations, and activation function for training the neural network model. Specifically, the batch size can be adjusted based on the actual situation; the learning rate should be reduced to relearn new data; the learning strategy can remain unchanged or be tried; and the maximum number of iterations should be reduced. Understandably, the purpose of these adjustments is to optimize the performance of the new model during training.

[0080] Then, use the training set step in step S101 to train the pre-trained model with adjusted training parameters. Based on the test results of the pre-trained model on the test set and the changes in neural network loss during training, adjust the training parameters and network parameters to continue training.

[0081] Repeat the above steps until the best pre-trained model is obtained, that is, until the pre-trained model achieves the required results on the validation set pedestrian data, or until the neural network loss converges within a certain range during training.

[0082] In practical applications, the threshold can be set to a fixed value or a dynamically adjusted value. The size of this value generally depends on the working environment of the cable to be diagnosed. When the working environment requirements are high, a smaller threshold can be used, and when the working environment requirements are low, a larger threshold can be used.

[0083] S104: Collect the harmonic current characteristics of the cable to be diagnosed and input them into the target model to obtain the degree of degradation of the cable to be diagnosed.

[0084] Then, the harmonic current characteristics of the cable to be diagnosed are collected, the target model is used to predict the insulation degradation, and the degree of insulation degradation is output. When the degree of degradation is high, the cable can be replaced or the cable can be insulated.

[0085] Furthermore, the harmonic current characteristics, temperature distribution characteristics, and magnetic field distribution characteristics of the cable to be diagnosed can be collected. The target model can be used to predict insulation degradation and output the degree of insulation degradation. When the degree of degradation is high, the cable can be replaced or the cable can be insulated.

[0086] This invention uses harmonic current characteristic data as raw data and artificial degradation evaluation data as sample labels to obtain training samples between harmonic current characteristics and cable degradation. Then, a neural network model is used to learn the hidden relationship between harmonic current characteristics and cable degradation in the training samples. After the neural network model is trained, the target model can be used to assess the degree of degradation of the cable to be diagnosed, thus achieving the effect of analyzing cable circuit insulation degradation without power interruption.

[0087] Finally, high-voltage cables generally cannot be de-energized. Therefore, the embodiments of the present invention can avoid power outages of high-voltage cables, thus avoiding the impact on users' electricity consumption. At the same time, it can also enable power transmission and distribution companies to supply power without interruption, thereby improving the efficiency of power transmission and distribution companies.

[0088] Example 2

[0089] Figure 2 This is a schematic diagram of the structure of a power cable line degradation diagnosis system utilizing database samples provided in an embodiment of the present invention, as shown below. Figure 2 As shown, corresponding to Embodiment 1 of the present invention, Embodiment 2 of the present invention provides a pedestrian re-identification system based on feature-enhanced guided convolutional neural networks, the system comprising:

[0090] The sample generation module 201 is used to extract the harmonic current characteristics of the sample cable, use the expert evaluation data of the sample cable as a label to label the harmonic current characteristics, and divide the set of labeled harmonic current characteristics into a training set and a test set.

[0091] Training module 202 is used to construct a convolutional neural network corresponding to the expert evaluation data. The convolutional neural network corresponding to the expert evaluation data includes an input layer, several convolutional layers, a max pooling layer, and an output layer.

[0092] The training set is input into the convolutional neural network corresponding to the expert evaluation data to obtain a pre-trained model; the pre-trained model is tested using the test set, the training parameters are adjusted and retrained according to the test results, and the process of inputting the training set into the convolutional neural network corresponding to the expert evaluation data is repeated until the convolutional neural network corresponding to the expert evaluation data converges to obtain the target model.

[0093] Output module 203 is used to collect the harmonic current characteristics of the cable to be diagnosed and input them into the target model to obtain the degree of degradation of the cable to be diagnosed.

[0094] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for diagnosing the deterioration of power cable lines using database samples, characterized in that, The method includes: 1) Extract the harmonic current characteristics of the sample cable, use the expert evaluation data of the sample cable as labels to label the harmonic current characteristics, and divide the set of labeled harmonic current characteristics into training set and test set. 2) Construct a convolutional neural network corresponding to the expert evaluation data, wherein the convolutional neural network corresponding to the expert evaluation data includes an input layer, several convolutional layers, a max pooling layer, and an output layer; 3) Input the training set into the convolutional neural network corresponding to the expert evaluation data to obtain a pre-trained model; use the test set to test the pre-trained model, adjust the training parameters according to the test results and retrain, and return to execute the step of inputting the training set into the convolutional neural network corresponding to the expert evaluation data until the convolutional neural network corresponding to the expert evaluation data converges to obtain the target model; 4) Collect the harmonic current characteristics of the cable to be diagnosed and input them into the target model to obtain the degree of degradation of the cable to be diagnosed; The training process of the convolutional neural network includes: A convolutional neural network was trained using labeled harmonic current features as training samples, and the formula was used... Calculate the harmonic error loss of the convolutional neural network at the current iteration, where S1 is the number of harmonic current features; x i The expert evaluation degradation index corresponds to the i-th harmonic current characteristic; ∑ is the predicted degradation index corresponding to the i-th harmonic current feature output by the model; ∑ is the summation function; when the harmonic error loss is lower than the set threshold, the first pre-trained model is obtained; The first pre-trained model is trained using the labeled temperature distribution features as training samples, and the formula is used... Calculate the temperature error loss of the convolutional neural network in the current iteration, where S2 is the number of temperature distribution features; y j The expert evaluation degradation index corresponds to the j-th temperature distribution feature; The predicted degradation index is the j-th temperature distribution feature output by the model; when the temperature error loss is lower than a set threshold, the second pre-trained model is obtained. The second pre-trained model is trained using the labeled magnetic field distribution features as training samples, and the formula is used... Calculate the magnetic field error loss of the convolutional neural network in the current iteration, where S3 is the number of harmonic current features; z k The expert evaluation degradation index corresponds to the kth magnetic field distribution feature; The predicted degradation index is the k-th magnetic field distribution feature output by the model; when the magnetic field error loss is lower than the set threshold, the third pre-trained model is obtained. Using the formula Loss=λ1*loss1+λ2*loss2+loss3, the comprehensive weighted loss of the third pre-trained model in the current iteration is calculated until the comprehensive weighted loss converges to obtain the target model. Here, Loss is the comprehensive weighted loss of the convolutional neural network in the current iteration; λ1 is the weight corresponding to the harmonic current feature; and λ2 is the weight corresponding to the temperature distribution feature.

2. The method for diagnosing power cable line deterioration using database samples according to claim 1, characterized in that, The process of extracting harmonic current characteristics of the sample cable includes: acquiring the load time distribution of the sample cable, determining the load peak value of the sample cable based on the load time distribution, and collecting the harmonic current characteristics at the load peak value time.

3. The method for diagnosing power cable line deterioration using database samples according to claim 1, characterized in that, The process of extracting harmonic current characteristics of the sample cable includes: obtaining the load time distribution of the sample cable, determining the average load of the sample cable based on the load time distribution, and collecting harmonic current characteristics during periods when the load is greater than the average load.

4. The method for diagnosing power cable line deterioration using database samples according to claim 1, characterized in that, The step of using expert evaluation data for the sample cables as labels to tag harmonic current characteristics includes: The temperature distribution characteristics of the sample cable are acquired synchronously, and the expert evaluation data of the sample cable is used as a label to mark the temperature distribution characteristics.

5. The method for diagnosing power cable line deterioration using database samples according to claim 4, characterized in that, When obtaining the temperature distribution characteristics of the sample cable, the method further includes: The magnetic field distribution characteristics of the sample cable are acquired simultaneously, and the expert evaluation data of the sample cable are used as tags to label the magnetic field distribution characteristics.

6. The method for diagnosing power cable line deterioration using database samples according to claim 5, characterized in that, The set of labeled harmonic current characteristics is divided into a training set and a test set, including: The labeled harmonic current features are expanded using color histogram equalization, horizontal mirroring, Gaussian blurring, and the addition of random noise. The expanded sample set is then divided into a training set and a test set.

7. A power cable line deterioration diagnosis system utilizing database samples, used to execute the power cable line deterioration diagnosis method utilizing database samples as described in any one of claims 1-6, characterized in that, The system includes: The sample generation module is used to extract the harmonic current characteristics of the sample cable, use expert evaluation data for the sample cable as labels to label the harmonic current characteristics, and divide the set of labeled harmonic current characteristics into training set and test set. The training module is used to construct a convolutional neural network corresponding to the expert evaluation data. The convolutional neural network corresponding to the expert evaluation data includes an input layer, several convolutional layers, a max pooling layer, and an output layer. The training set is input into the convolutional neural network corresponding to the expert evaluation data to obtain a pre-trained model; the pre-trained model is tested using the test set, the training parameters are adjusted and retrained according to the test results, and the process of inputting the training set into the convolutional neural network corresponding to the expert evaluation data is repeated until the convolutional neural network corresponding to the expert evaluation data converges to obtain the target model. The output module is used to collect the harmonic current characteristics of the cable under diagnosis and input them into the target model to obtain the degree of degradation of the cable under diagnosis.

Images