XLPE cable insulation defect type identification and severity assessment method

Abstract

The invention discloses an XLPE cable insulation defect type identification and severity evaluation method. The method comprises the following steps: S1, sample preparation, S2, data acquisition, S3, feature parameter extraction, S4, feature fusion and preprocessing, S5, hierarchical model construction, S6, model training and verification, and S7, model practical application. According to the method, through multi-dimensional feature fusion, dielectric loss (aging overall characterization) and a nonlinear coefficient (local defect sensitivity) are combined, and the recognition accuracy is improved; meanwhile, through a hierarchical model, the defect type is determined firstly, and then the severity is evaluated in a targeted manner, so that feature confusion of different defects can be avoided, and accurate classification and severity quantitative evaluation of the XPLE cable insulation defects are realized; the problem that no diagnosis technology for other defects except for the partial discharge defect exists in XLPE cable insulation at present is solved.

Description

Technical Field

[0001] This invention belongs to the field of cable insulation defect diagnosis technology, specifically relating to a method for identifying the types and severity of insulation defects in XLPE cables. Background Technology

[0002] XLPE cables are mainly composed of a conductor, a cross-linked polyethylene insulation layer, a shielding layer, and a sheath layer. They offer advantages such as low maintenance workload, convenient installation, superior electrical performance, and good heat resistance, making them widely used in medium and low voltage power transmission and distribution lines. The safe operation of XLPE cables is crucial to the stability of the entire power system; a fault in XLPE cables can cause significant power outages in the affected area. Reports indicate that the vast majority of operational faults in XLPE power cables are insulation faults, caused by surface discharge of the solid composite dielectric. Therefore, the safe operation of XLPE power cable lines, especially the safety and reliability of their cross-linked polyethylene insulation layer, should be given sufficient attention.

[0003] Currently, diagnostic technologies for XLPE cable insulation defects are mostly focused on partial discharge defects. For example, Chinese invention patent application CN115201644A, entitled "A Method, Recording Medium and System for Diagnosing Insulation Defects and Moisture Condition of XLPE Cables," uses a bidirectional neural network algorithm to model the partial discharge state of the cable. In the modeling process, the partial discharge state of the cable is quantified and formed into an array, which is then added as a variable parameter to the model. This solves the technical problem that traditional cable defect detection methods have large errors compared to reality and are difficult to correct. At present, there are no diagnostic technologies for other insulation defects of XLPE cables. To solve the above problems, it is necessary to develop a method for identifying the types of insulation defects and assessing their severity in XLPE cables. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for identifying and assessing the types and severity of insulation defects in XLPE cables with accurate defect classification and quantifiable severity assessment. This method can assist in online monitoring and preventive maintenance of XPLE power cables, effectively ensuring the safe and stable operation of the power system.

[0005] The objective of this invention is achieved as follows: a method for identifying the types and severity of insulation defects in XLPE cables, comprising the following steps:

[0006] S1, Sample preparation: Prepare n XLPE cable samples with different insulation defect types in each group, n≥30, and the n samples in each group contain different severity levels;

[0007] S2, Data Acquisition: UI feature tests are performed on each group of samples sequentially under ultra-low frequency voltage excitation;

[0008] S21, apply an ultra-low frequency sinusoidal voltage to the test sample, with a voltage frequency f = 0.1 Hz and a voltage level U = [0.5U0, U0, 1.5U0, 2U0], where U0 is the rated voltage;

[0009] S22, synchronously collect voltage U(t) and current I(t), collect m cycles of data at each voltage level, m≥5, and take the average value to reduce noise;

[0010] S23, calculate the tangent of the ultra-low frequency dielectric loss angle. The calculation formula is: In the formula, I loss This refers to the active component of the current, i.e., the active loss current; I cap Y represents the reactive component of the current, i.e., the reactive capacitor current; Y = I / U is the admittance. The fundamental component is extracted by Fourier transform, and the real part Re(Y) and imaginary part Im(Y) of the admittance are calculated.

[0011] S3, Feature parameter extraction:

[0012] S31, Nonlinear UI Feature Extraction: First, perform a 3rd-order polynomial fitting on the UI curve: I = aU + bU 2 +cU 3 The coefficients [a, b, c] are extracted as nonlinear features; then the nonlinear coefficient k is defined as the contribution of higher-order terms. It reflects the degree of deviation from a linear relationship; k=0 indicates perfect linearity, while k→1 indicates significant nonlinearity.

[0013] S32, Ultra-low frequency dielectric loss feature extraction: Extracting tanδ at different voltage levels, denoted as...

[0014] S33, Statistical Feature Extraction: First, extract the peak current I. peak =max|I(t)|, which is the maximum absolute value of the current time-domain signal; then extract the rate of change of the slope of the UI curve. In the formula Let be the slope of the UI curve at voltage 2U0. The slope of the UI curve is given by a voltage of 0.5U0.

[0015] S4, Feature Fusion and Preprocessing:

[0016] S41, Construct the feature vector for each sample

[0017] A total of 10 features;

[0018] S42 uses min-max normalization preprocessing for each feature. Ensure that the values ​​of each feature are within the range of [0,1].

[0019] S5, Hierarchical Model Construction: A hierarchical model is constructed using a hierarchical neural network. First, a defect type identification model is built, and then a severity assessment model is constructed by combining defect type information.

[0020] S51, Defect Type Classification Model:

[0021] Input layer: 10-dimensional feature vector;

[0022] Hidden layer: 1 layer, 20 neurons, ReLU activation function;

[0023] Output layer: 5 neurons, normalized exponential function Softmax;

[0024] Loss function: Cross-entropy loss, optimizer Adam;

[0025] S52, Severity Assessment Model:

[0026] Input layer: 10-dimensional feature vector + 5-dimensional defect type one-hot encoding, for a total of 15-dimensional feature vector;

[0027] Hidden layers: 2 layers, 30 → 15 neurons, ReLU activation function;

[0028] Output layer: 3 neurons, normalized exponential function Softmax, corresponding to mild, moderate and severe;

[0029] Loss function; cross-entropy loss;

[0030] S6, Model Training and Validation:

[0031] S61, Dataset partitioning: The training set and the test set are divided in a 7:3 ratio, and 5-fold cross-validation is used;

[0032] S62, Training strategy: First, train the defect type classification model independently, then fix the parameters of the defect type classification model, and then use transfer learning to train the severity assessment model.

[0033] S63, Evaluation metrics: accuracy, precision, recall, and F1-score;

[0034] S7, Practical Application of the Model:

[0035] S71, Real-time Testing: Apply ultra-low frequency voltage to the cable under test, collect UI data, and calculate the normalized feature vector x;

[0036] S72, Defect Type Identification: Input x into the defect type classification model and output the defect type with the highest probability;

[0037] S73, Severity Assessment: Input the x+5 dimension defect type one-hot encoding into the severity assessment model and output the severity value;

[0038] S74, Result Mapping: Based on a preset threshold table solidified by expert experience, the model output is used to determine the type, severity level, and corresponding treatment measures for defects in the cable under test.

[0039] Preferably, in step S1, the insulation defect types of the XLPE cable include thermal aging, water treeing, electrical treeing, mechanical damage, and chemical corrosion, wherein the different severity levels of thermal aging, water treeing, electrical treeing, mechanical damage, and chemical corrosion are different aging times, different lengths, different densities, different scratch depths, and different acid and alkali immersion times, respectively.

[0040] Preferably, the unique thermal codes for thermal aging, water treeing, electrical treeing, mechanical damage and chemical corrosion are [1,0,0,0,0], [0,1,0,0,0], [0,0,1,0,0], [0,0,0,1,0] and [0,0,0,0,1], respectively.

[0041] Preferably, in step S2, an ultra-low frequency signal generator is used to apply an ultra-low frequency sinusoidal voltage, a high-precision digital multimeter is used to synchronously collect voltage and current, and a shielded test box is used to house the ultra-low frequency signal generator and the high-precision digital multimeter to reduce the interference of environmental noise on the data acquisition.

[0042] Preferably, in step S63, accuracy refers to the proportion of correctly classified samples to the total number of samples, and the calculation formula is:

[0043] Preferably, in step S63, the precision rate refers to the proportion of samples predicted as positive that are actually positive, and the calculation formula is:

[0044] Preferably, in step S63, recall rate refers to the proportion of samples that were actually positive but were predicted to be positive, and the calculation formula is:

[0045] Preferably, in step S63, the F1-score refers to the harmonic mean of precision and recall, and is calculated using the following formula:

[0046] Preferably, in step S7, the preset threshold table is as follows:

[0047]

[0048] A warning will be issued when the condition is determined to be mild; a planned maintenance will be arranged when the condition is moderate; and an immediate maintenance will be arranged when the condition is severe.

[0049] Due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0050] This invention employs a methodological approach encompassing sample preparation, data acquisition, feature parameter extraction, feature fusion and preprocessing, hierarchical model construction, model training and validation, and practical application of the model. From model construction and training to model usage, through multi-dimensional feature fusion and hierarchical modeling, it achieves accurate classification and quantitative assessment of the severity of insulation defects in XPLE cables. This addresses the current lack of diagnostic technologies for defects in XLPE cable insulation other than partial discharge defects, thereby facilitating online monitoring and preventative maintenance of XPLE power cables and effectively ensuring the safe and stable operation of power systems. Attached Figure Description

[0051] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0052] The technical solution of the present invention will be further described in detail below through embodiments and in conjunction with the accompanying drawings.

[0053] like Figure 1 As shown, this invention provides a method for identifying the types and severity of insulation defects in XLPE cables, comprising the following steps:

[0054] S1, Sample preparation: Prepare n XLPE cable samples with different insulation defect types in each group, n≥30. For example, if there are 40 samples in 5 groups, a total of 200 samples need to be prepared. Each group of 40 samples should include samples with different severity levels.

[0055] Among them, the insulation defects of XLPE cables include thermal aging, water treeing, electrical treeing, mechanical damage, and chemical corrosion. Different degrees of severity correspond to different aging times (e.g., 500h, 1000h, 1500h…), different lengths (e.g., 1mm, 2mm, 3mm…), and different densities (e.g., 5 strands / cm²). 3 10 roots / cm 3 15 roots / cm 3 …), different scratch depths (e.g., 0.1mm, 0.3mm, 0.5mm…), different acid and alkali soaking times (e.g., 24h, 48h, 72h…), etc.

[0056] S2, Data Acquisition: Under ultra-low frequency voltage excitation, UI feature tests are performed on each sample group in sequence.

[0057] S21. Apply an ultra-low frequency sinusoidal voltage to the test sample. The voltage frequency is f = 0.1 Hz, and the voltage level is U = [0.5U0, U0, 1.5U0, 2U0], where U0 is the rated voltage.

[0058] S22, synchronously collect voltage U(t) and current I(t), collect m cycles of data at each voltage level, m≥5. If 10 cycles of data are collected, take the average of the 10 cycles of data to reduce noise.

[0059] S23, calculate the tangent of the ultra-low frequency dielectric loss angle. The calculation formula is: In the formula, I loss This refers to the active component of the current, i.e., the active loss current; I cap Y represents the reactive component of the current, i.e., the reactive capacitor current; Y = I / U represents the admittance. The fundamental component is extracted by Fourier transform, and the real part Re(Y) and imaginary part Im(Y) of the admittance are calculated.

[0060] The process involves applying an ultra-low frequency sinusoidal voltage using an ultra-low frequency signal generator, synchronously acquiring voltage and current using a high-precision digital multimeter, and employing a shielded test box to reduce environmental noise interference with the acquisition.

[0061] S3, Feature parameter extraction:

[0062] S31, Nonlinear UI Feature Extraction: First, perform a 3rd-order polynomial fitting on the UI curve: I = aU + bU 2 +cU 3 The coefficients [a, b, c] are extracted as nonlinear features; generally, the larger the coefficient of the nonlinear term, the stronger the nonlinearity. Then, the nonlinear coefficient k is defined as the contribution of higher-order terms. It is used to reflect the degree of deviation from linearity. k=0 means perfectly linear, and k→1 means significant nonlinearity, that is, the more serious the defect.

[0063] S32, Ultra-low frequency dielectric loss feature extraction: Extracting tanδ at different voltage levels, denoted as... It is used to reflect insulation degradation and generally increases with increasing voltage and the severity of defects.

[0064] S33, Statistical Feature Extraction: First, extract the peak current I. peak =max|I(t)|, which is the maximum absolute value of the current time-domain signal; then extract the rate of change of the slope of the UI curve. In the formula Let be the slope of the UI curve at voltage 2U0. The slope of the UI curve is given by a voltage of 0.5U0.

[0065] S4, Feature Fusion and Preprocessing:

[0066] S41, Construct the feature vector for each sample A total of 10 features;

[0067] S42 uses min-max normalization preprocessing for each feature. Ensure that the value of each feature is within the range of [0,1].

[0068] S5, Hierarchical Model Construction: A hierarchical model is constructed using a hierarchical neural network. First, a defect type identification model is built, and then a severity assessment model is constructed by combining defect type information.

[0069] S51, Construct a defect type classification model, as follows:

[0070] Input layer: 10-dimensional feature vector;

[0071] Hidden layer: 1 layer, 20 neurons, ReLU activation function;

[0072] Output layer: 5 neurons, corresponding to 5 types of defects, normalized exponential function Softmax;

[0073] Loss function: Cross-entropy loss, optimizer Adam.

[0074] S52, Construct a severity assessment model, as follows:

[0075] Input layer: 10-dimensional feature vector + 5-dimensional defect type one-hot encoding, for a total of 15-dimensional feature vector;

[0076] Hidden layers: 2 layers, 30 → 15 neurons, ReLU activation function;

[0077] Output layer: 3 neurons, normalized exponential function Softmax, corresponding to mild, moderate and severe;

[0078] Loss function: Cross-entropy loss.

[0079] Among them, the unique thermal codes for thermal aging, water treeing, electrical treeing, mechanical damage and chemical corrosion are [1,0,0,0,0], [0,1,0,0,0], [0,0,1,0,0], [0,0,0,1,0], [0,0,0,1,0] and [0,0,0,0,1], respectively.

[0080] Specifically, the hierarchical model of this invention first determines the defect type and then assesses the severity in a targeted manner, which can avoid confusion of the characteristics of different defects. For example, thermal aging and chemical corrosion have similar dielectric loss characteristics, but their nonlinear coefficients are very different. At the same time, multi-dimensional feature fusion is used to combine dielectric loss (overall aging characterization) and nonlinear coefficient (sensitivity to local defects) to improve the identification accuracy.

[0081] S6, Model Training and Validation:

[0082] S61, Dataset partitioning: The training set and the test set are divided in a 7:3 ratio, and 5-fold cross-validation is used to avoid overfitting;

[0083] S62, Training strategy: First, train the defect type classification model independently, then fix the parameters of the defect type classification model, and then use transfer learning, that is, use the prior knowledge of defect types to train the severity assessment model.

[0084] S63, Evaluation metrics: accuracy, precision, recall, and F1-score.

[0085] Accuracy refers to the proportion of correctly classified samples out of the total number of samples, and the calculation formula is: Precision rate refers to the proportion of samples predicted as positive that are actually positive. The formula is: Recall rate is the proportion of samples that were predicted to be positive out of the samples that were actually positive. The formula is: The F1 score is the harmonic mean of precision and recall, calculated using the following formula:

[0086] S7, Practical Application of the Model:

[0087] S71, Real-time Testing: Apply ultra-low frequency voltage to the cable under test, collect UI data, and calculate the normalized feature vector x;

[0088] S72, Defect Type Identification: Input x into the defect type classification model and output the defect type with the highest probability;

[0089] S73, Severity Assessment: Input the x+5 dimension defect type one-hot encoding into the severity assessment model and output the severity value;

[0090] S74, Result Mapping: Based on a preset threshold table solidified by expert experience, the model output is used to determine the type, severity level, and corresponding treatment measures for defects in the cable under test.

[0091] The preset thresholds for the five common defect types are as follows:

[0092]

[0093] The corresponding handling measures are as follows: when the problem is determined to be mild, the system will issue an early warning; when it is moderate, a planned maintenance will be arranged; and when it is severe, immediate maintenance will be arranged.

[0094] S8, in the later stages, with the accumulation of sample data, can be used to repeatedly train, validate and continuously optimize the constructed hierarchical model using more and more sample datasets, thereby improving the classification and evaluation capabilities of the two models.

[0095] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of the claims of the present invention.

Claims

1. A method for identifying the types and assessing the severity of insulation defects in XLPE cables, characterized in that, Includes the following steps: S1, Sample preparation: Prepare n XLPE cable samples with different insulation defect types in each group, n≥30, and the n samples in each group contain different severity levels; S2, Data Acquisition: UI feature tests are performed on each group of samples sequentially under ultra-low frequency voltage excitation; S21, apply an ultra-low frequency sinusoidal voltage to the test sample, with a voltage frequency f = 0.1 Hz and a voltage level U = [0.5U0, U0, 1.5U0, 2U0], where U0 is the rated voltage; S22, synchronously collect voltage U(t) and current I(t), collect m cycles of data at each voltage level, m≥5, and take the average value to reduce noise; S23, calculate the tangent of the ultra-low frequency dielectric loss angle. The calculation formula is: In the formula, I loss This refers to the active component of the current, i.e., the active loss current; I cap Y represents the reactive component of the current, i.e., the reactive capacitor current; Y = I / U is the admittance. The fundamental component is extracted by Fourier transform, and the real part Re(Y) and imaginary part Im(Y) of the admittance are calculated. S3, Feature parameter extraction: S31, Nonlinear UI Feature Extraction: First, perform a 3rd-order polynomial fitting on the UI curve: I = aU + bU 2 +cU 3 The coefficients [a,b,c] are extracted as nonlinear features; Then, the nonlinear coefficient k is defined as the contribution of higher-order terms. It reflects the degree of deviation from a linear relationship; k=0 indicates perfect linearity, while k→1 indicates significant nonlinearity. S32, Ultra-low frequency dielectric loss feature extraction: Extracting tanδ at different voltage levels, denoted as... S33, Statistical Feature Extraction: First, extract the peak current I. peak =max|I(t)|, which is the maximum absolute value of the current time-domain signal; then extract the rate of change of the slope of the UI curve. In the formula Let be the slope of the UI curve at voltage 2U0. The slope of the UI curve is given by a voltage of 0.5U0. S4, Feature Fusion and Preprocessing: S41, Construct the feature vector for each sample A total of 10 features; S42 uses min-max normalization preprocessing for each feature. Ensure that the values ​​of each feature are within the range of [0,1]. S5, Hierarchical Model Construction: A hierarchical model is constructed using a hierarchical neural network. First, a defect type identification model is built, and then a severity assessment model is constructed by combining defect type information. S51, Defect Type Classification Model: Input layer: 10-dimensional feature vector; Hidden layer: 1 layer, 20 neurons, ReLU activation function; Output layer: 5 neurons, normalized exponential function Softmax; Loss function: Cross-entropy loss, optimizer Adam; S52, Severity Assessment Model: Input layer: 10-dimensional feature vector + 5-dimensional defect type one-hot encoding, for a total of 15-dimensional feature vector; Hidden layers: 2 layers, 30 → 15 neurons, ReLU activation function; Output layer: 3 neurons, normalized exponential function Softmax, corresponding to mild, moderate and severe; Loss function; cross-entropy loss; S6, Model Training and Validation: S61, Dataset partitioning: The training set and the test set are divided in a 7:3 ratio, and 5-fold cross-validation is used; S62, Training strategy: First, train the defect type classification model independently, then fix the parameters of the defect type classification model, and then use transfer learning to train the severity assessment model. S63, Evaluation metrics: accuracy, precision, recall, and F1-score; S7, Practical Application of the Model: S71, Real-time Testing: Apply ultra-low frequency voltage to the cable under test, collect UI data, and calculate the normalized feature vector x; S72, Defect Type Identification: Input x into the defect type classification model and output the defect type with the highest probability; S73, Severity Assessment: Input the x+5 dimension defect type one-hot encoding into the severity assessment model and output the severity value; S74, Result Mapping: Based on a preset threshold table solidified by expert experience, the model output is used to determine the type, severity level, and corresponding treatment measures for defects in the cable under test.

2. The method for identifying and assessing the severity of insulation defects in XLPE cables according to claim 1, characterized in that: In step S1, the insulation defect types of XLPE cables include thermal aging, water treeing, electrical treeing, mechanical damage, and chemical corrosion. The different severity levels of thermal aging, water treeing, electrical treeing, mechanical damage, and chemical corrosion are different aging times, different lengths, different densities, different scratch depths, and different acid and alkali immersion times, respectively.

3. The method for identifying and assessing the severity of insulation defects in XLPE cables according to claim 2, characterized in that: The unique thermal codes for thermal aging, water treeing, electrical treeing, mechanical damage, and chemical corrosion are [1,0,0,0,0], [0,1,0,0,0], [0,0,1,0,0], [0,0,0,1,0], [0,0,0,1,0], and [0,0,0,0,1], respectively.

4. The method for identifying the types and assessing the severity of insulation defects in XLPE cables according to claim 1, characterized in that: In step S2, an ultra-low frequency sinusoidal voltage is applied using an ultra-low frequency signal generator, and the voltage and current are synchronously collected using a high-precision digital multimeter. A shielded test box is used to house the ultra-low frequency signal generator and the high-precision digital multimeter to reduce the interference of environmental noise on the data acquisition.

5. The method for identifying and assessing the severity of insulation defects in XLPE cables according to claim 1, characterized in that, In step S63, accuracy refers to the proportion of correctly classified samples to the total number of samples, and the calculation formula is:

6. The method for identifying and assessing the severity of insulation defects in XLPE cables according to claim 1, characterized in that, In step S63, the precision rate refers to the proportion of samples predicted as positive that are actually positive, and the calculation formula is:

7. The method for identifying and assessing the severity of insulation defects in XLPE cables according to claim 1, characterized in that, In step S63, recall rate refers to the proportion of samples that were predicted to be positive out of the samples that were actually positive, and the calculation formula is:

8. The method for identifying and assessing the severity of insulation defects in XLPE cables according to claim 1, characterized in that, In step S63, the F1-score refers to the harmonic mean of precision and recall, and is calculated using the following formula:

9. The method for identifying the types and assessing the severity of insulation defects in XLPE cables according to claim 2, characterized in that, In step S7, the preset threshold table is as follows: A warning will be issued when the condition is determined to be mild; a planned maintenance will be arranged when the condition is moderate; and an immediate maintenance will be arranged when the condition is severe.

Images