Harmonic component-based on-line detection and evaluation method for insulation deterioration degree of distribution network cable

Abstract

The application discloses a kind of based on harmonic component's distribution network cable insulation deterioration degree live detection evaluation method, first by current transformer live measurement harmonic current data of the cable to be detected, send into power quality analysis module, analysis extracts each harmonic current of the cable to be detected;Then select appropriate harmonic number initialization weight coefficient, construct Lasso regression analysis model;Then carry out Lasso regression analysis model training, optimize the objective function of Lasso regression model and carry out model verification, evaluate the optimized weight coefficient, so that residual vector is small enough or feature vector has been selected;Finally, the optimized Lasso regression analysis model is obtained, and the deterioration degree of cable is evaluated by using the optimized Lasso regression analysis model.The application can accurately detect and troubleshoot the insulation defect of the entire distribution network cable in a single detection, and has good reliability.

Description

Technical Field

[0001] This invention relates to the field of cable testing technology, and specifically to a live-line testing and evaluation method for the degree of insulation degradation of distribution network cables based on harmonic components. Background Technology

[0002] Cross-linked polyethylene (XLPE) cables are increasingly widely used in urban power transmission and distribution systems due to their superior performance and ease of installation. However, the increasing complexity of urban underground pipe networks, the intricate structure of distribution network cable lines, diverse operating environments, growing equipment density, and the rising proportion of aging cables have led to persistently high failure rates in distribution network cable lines. This poses a growing risk of major fires or widespread power outages, potentially severely impacting the operational safety of urban power grids.

[0003] Since cables are mainly laid underground, and the laying channels often experience overcapacity, accumulation, water accumulation, and siltation, these conditions can easily lead to a decline in cable insulation performance and subsequent failures. Therefore, detecting and assessing the insulation aging and moisture condition of cables has always been a key focus for researchers and maintenance personnel. Currently, the main methods for assessing the condition of distribution network cables are routine manual inspections and offline testing (withstand voltage, oscillating wave partial discharge, ultra-low frequency dielectric loss, etc.). However, due to numerous limitations such as strict control over cable network construction, laying channels, and planned power outage maintenance times (especially in large cities with extremely high power supply reliability requirements), as well as the risk of cumulative damage or accelerated failure of aging cable insulation by offline high-voltage testing, offline high-voltage testing technology is becoming increasingly difficult to apply widely.

[0004] In recent years, research on live-line testing technology for cables has become increasingly popular due to the growing difficulty in applying offline testing techniques. While traditional high-voltage cables have relatively mature live-line testing technologies such as partial discharge and grounding current detection, distribution network cables differ significantly from high-voltage cables in terms of structural characteristics and grounding systems. High-frequency partial discharge and grounding current detection technologies for high-voltage cables are difficult to apply to the condition monitoring of distribution network cables. Currently, common live-line testing methods for distribution network cables include high-frequency pulse partial discharge, ultrasonic partial discharge, and infrared thermal imaging. However, in practical applications, these methods suffer from limitations in detection coverage, difficulty in implementation, limited identifiable defects, and inherent bottlenecks in sensitivity and interference immunity. More importantly, they cannot accurately identify insulation defects along the entire distribution network cable line in a single test, requiring multiple measurements, resulting in high labor and time costs and significant application difficulties. Therefore, it is necessary to conduct research on convenient live-line testing technologies suitable for high-reliability insulation condition sensing of distribution network cables. Summary of the Invention

[0005] This invention addresses the challenge of live-line detection and assessment of insulation degradation in distribution network cables by providing a live-line detection and assessment method for the degree of insulation degradation in distribution network cables based on harmonic components.

[0006] The technical solution adopted in this invention is: a live-line detection and evaluation method for the insulation degradation degree of distribution network cables based on harmonic components. First, harmonic current data of the cable under test is measured using a current transformer under energization and sent to a power quality analysis module to analyze and extract the harmonic currents of the cable under test. Next, appropriate harmonic orders are selected to initialize weight coefficients, and a Lasso regression analysis model is constructed. Then, the Lasso regression analysis model is trained, the objective function of the Lasso regression model is optimized, and the model is validated. The optimized weight coefficients are evaluated to ensure that the residual vector is sufficiently small or that all eigenvectors have been selected. Finally, the optimized Lasso regression analysis model is obtained, and the cable degradation degree is evaluated using the optimized Lasso regression analysis model.

[0007] Further optimization involves analyzing and extracting the harmonic currents of the cable under test, including: Let n be the phasor of the nth current harmonic, where n = 1, 2, ..., 7. ;I n Let θ be the amplitude of the nth current harmonic. n The phase angle of the nth current harmonic; when n is 1, it represents the fundamental component of the current. The occupancy rates of each harmonic were calculated as follows: Among them, h n The occupancy rate of the nth harmonic.

[0008] Further preferred, the step of selecting appropriate harmonic orders to initialize weight coefficients and constructing a Lasso regression analysis model includes: The objective function z(b) for constructing the Lasso regression analysis model is: ; in, λ is the predicted value, y is the actual value, λ is the penalty coefficient, and b is the weighting coefficient. The penalty coefficient λ was determined using cross-validation. Then, the variables selected by the Lasso regression analysis model are determined by the penalty coefficient λ and the regression coefficient plot, where variables with a weight coefficient of zero are removed by the Lasso regression analysis model. After determining the penalty coefficient λ, the weight coefficient b is selected by optimizing the least angle regression method.

[0009] Further optimization, using the least angle regression method to select the weight coefficient b, involves the following steps: (1) Initialize the weight coefficients b, for example, initialize them as a zero vector; (2) Initialize the residual vector as the target label vector. ; (3) Select an eigenvector h that is most correlated with the residual vector, find a set of weight coefficients b along the direction of the eigenvector, and another eigenvector h2 that is most correlated with the residual vector appears, so that the new residual vector is equal to the correlation of the two eigenvectors, and recalculate the residual vector. (4) Repeat step (3) to find a set of weight coefficients such that the third eigenvector with the greatest correlation to the residual vector makes the new residual equal to the correlation of these three eigenvectors, and so on. (5) When the residual vector is small enough or all eigenvectors have been selected, the iteration ends.

[0010] Further optimization yielded the following formula for predicting the degradation level of power distribution cables using the Lasso regression analysis model: ; In the formula, b0 is a constant, b2 is the weighting coefficient of the second harmonic, and h2 is the occupancy rate of the second harmonic; b n The nth harmonic weighting coefficient is used; the Lasso regression analysis model continuously and dynamically adjusts the regression coefficients to achieve quantitative prediction and assessment of the degree of degradation of distribution network cables.

[0011] This invention uses an optimized Lasso regression analysis model to assess the degree of cable degradation. It can accurately identify insulation defects along the entire distribution network cable in a single test, demonstrating good reliability. Attached Figure Description

[0012] Figure 1 This is a flowchart of the present invention.

[0013] Figure 2 This is a wiring diagram for live testing.

[0014] Figure 3 This is a cross-validation plot for Lasso regression.

[0015] Figure 4 This is a graph showing the variation of the weighting coefficient with the penalty coefficient λ.

[0016] Figure 5 The graph shows the changes in the occupancy rates of the 3rd, 5th, and 7th harmonics in the example.

[0017] Figure 6 This is a comparison chart of the predicted values ​​and the original values. Detailed Implementation

[0018] The present invention will be further explained in detail below with reference to the accompanying drawings and embodiments.

[0019] Reference Figure 1 A live-line detection and assessment method for the insulation degradation of distribution network cables based on harmonic components is proposed. Based on the analysis of harmonic component data of distribution network cables under water-thermal aging, the method first measures the harmonic current data of the cable under test using a high-precision current transformer, and then inputs this data into a power quality analysis module to analyze and extract the harmonic currents of the cable under test. Next, appropriate harmonic orders are selected to initialize weight coefficients, and a Lasso regression analysis model is constructed. Then, the Lasso regression analysis model is trained, the objective function of the Lasso regression model is optimized, and the model is validated. The optimized weight coefficients are evaluated to ensure that the residual vector is sufficiently small or that all eigenvectors have been selected. Finally, the optimized Lasso regression analysis model is obtained, and the cable degradation degree is assessed using the optimized Lasso regression analysis model.

[0020] like Figure 2 As shown, harmonic current data of the cable under test is collected by a high-precision current transformer under energization and sent to the power quality analysis module to analyze and extract the harmonic currents of the cable under test. In this embodiment, the extracted current harmonic order is 2 to 7, i.e., n=1,2,…,7;

[0021] in, (n=1,2,…,7) represents the nth current harmonic phasor. I n Let θ be the amplitude of the nth current harmonic. n This represents the phase angle of the nth current harmonic. When n is 1, it represents the fundamental component of the current.

[0022] The occupancy rates of each harmonic were calculated as follows: Among them, h n The occupancy rate of the nth harmonic is given; the changes in the occupancy rates of the 3rd, 5th, and 7th harmonics obtained from the test over the test time are shown as follows: Figure 5 As shown. By Figure 5 It can be seen that for cables with different thermal aging times, the amplitude of the third harmonic gradually increases in an oscillating manner as the aging degree deepens, the amplitude of the fifth harmonic first decreases and then increases with the deterioration state, and the seventh harmonic generally shows a downward trend, but there is an abnormal sudden increase in the middle process.

[0023] Figure 5The harmonic current characteristics of XLPE cables under different degradation states are given, but a quantitative indicator for judging the degree of cable degradation is still lacking. Therefore, Lasso regression analysis is introduced to provide a degradation degree assessment method. Lasso regression adds a regularization term to prevent overfitting, based on linear regression which fits the data by minimizing the sum of squared residuals. Appropriate harmonic orders are selected to initialize the weight coefficients b, and a Lasso regression analysis model is constructed. The objective function z(b) of the Lasso regression analysis model is: (1) in, λ is the predicted value, y is the actual value, λ is the penalty coefficient, and b is the weighting coefficient.

[0024] To minimize the objective function, the penalty coefficient λ is first determined using cross-validation. Then, the value of λ is determined using a graph of the regression coefficients (e.g., ...). Figure 3 As shown, the variables selected by the Lasso regression analysis model are determined, where variables with a weight coefficient of zero can be considered to be removed by the Lasso regression analysis model.

[0025] Figure 3 The vertical axis represents the mean squared error of the Lasso regression model, i.e., the cross-validation error. The upper horizontal axis represents the number of variables corresponding to different values ​​of λ, and the lower horizontal axis represents the base-10 logarithm of the penalty coefficient λ. The optimization objective is to select the Lasso regression model with the smallest possible number of feature variables and the smallest possible error, corresponding to the penalty coefficient λ. The graph contains two dashed lines: the left line represents the lowest error, and the right line represents the number of feature variables.

[0026] In this embodiment, the minimum mean square error of the Lasso regression analysis model is obtained when λ = 0.000045. The Lasso regression analysis model has the fewest characteristic variables when λ = 0.005. Once the value of λ reaches a certain size, further increasing the number of independent variables does not significantly improve the model performance.

[0027] After determining λ, the weight coefficient b is selected by optimizing the least angle regression method. The specific steps are as follows: (1) Initialize the weight coefficients b, for example, initialize them as a zero vector; (2) Initialize the residual vector as the target label vector. ; (3) Select an eigenvector h that is most correlated with the residual vector, find a set of weight coefficients b along the direction of the eigenvector, and another eigenvector h2 that is most correlated with the residual vector appears, so that the new residual vector is equal to the correlation of the two eigenvectors, and recalculate the residual vector. (4) Repeat step (3) to find a set of weight coefficients such that the third eigenvector with the greatest correlation to the residual vector makes the new residual equal to the correlation of these three eigenvectors, and so on. (5) When the residual vector is small enough or all eigenvectors have been selected, the iteration ends.

[0028] For the thermally aged cable in this embodiment, the trajectory diagram of its Lasso regression fitting weight coefficient b is as follows: Figure 4 As shown, Figure 4 Each curve in the graph represents the trajectory of the coefficient of each independent variable. The vertical axis represents the value of the coefficient, and the horizontal axis represents the penalty coefficient λ. As the penalty coefficient λ changes, the variable corresponding to the one whose penalty coefficient is compressed to 0 later is more important. Observation Figure 4 It can be observed that the 4th and 6th harmonics first become zero, and therefore are deleted. As the value of λ changes, the 5th harmonic is finally compressed to 0, therefore the 5th harmonic is the most important to the dependent variable.

[0029] The final weighting coefficient b was determined, and the following formula for predicting the degree of degradation of power distribution cables using the Lasso regression analysis model was obtained: (2) In the formula, b0 is a constant, b2 is the weighting coefficient of the second harmonic, and h2 is the occupancy rate of the second harmonic; b n The nth harmonic weighting coefficient; It should be noted that Equation (2) is only a local linear regression formula under a certain test data. As the live test harmonic data increases and the model is trained and verified, the Lasso regression analysis model of this invention will continuously and dynamically adjust the regression coefficients to achieve quantitative prediction and evaluation of the degree of degradation of distribution network cables.

[0030] The Lasso regression results show that, based on the standardized coefficients of the intercept term, the 3rd harmonic, the 4th harmonic, the 5th harmonic, the 6th harmonic, and the 7th harmonic, the intercept term, the 3rd harmonic, and the 5th harmonic are retained, while the 4th harmonic, the 6th harmonic, and the 7th harmonic are deleted. Therefore, the standardized formula for the Lasso regression analysis model used for assessing the degradation degree of power distribution cables in this embodiment is as follows: =10.679+1795.82h3+5348.4h5 It should be noted that this formula is only a local linear regression formula under the test data of this embodiment. As the live test harmonic data increases and the model is trained and validated, the Lasso regression prediction model of this invention will continuously and dynamically adjust the regression coefficients to achieve a quantitative prediction and assessment of the degree of cable degradation in the distribution network. The comparison between the predicted cable degradation value and the actual value obtained from Lasso regression is shown below. Figure 6 As shown.

[0031] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A live-line detection and evaluation method for the insulation degradation degree of distribution network cables based on harmonic components, characterized in that, First, harmonic current data of the cable under test is measured using a current transformer under energized conditions and sent to the power quality analysis module to analyze and extract the harmonic currents of the cable under test. Next, appropriate harmonic orders are selected to initialize the weight coefficients, and a Lasso regression analysis model is constructed. Then, the Lasso regression analysis model is trained, the objective function of the Lasso regression model is optimized, and the model is validated. The optimized weight coefficients are evaluated to ensure that the residual vector is sufficiently small or that all feature vectors have been selected. Finally, the optimized Lasso regression analysis model is obtained, and the optimized Lasso regression analysis model is applied to assess the degree of cable degradation. The analysis and extraction of harmonic currents in the cable under test includes: Let n be the phasor of the nth current harmonic, where n = 1, 2, ..., 7. ;I n Let θ be the amplitude of the nth current harmonic. n The phase angle of the nth current harmonic; when n is 1, it represents the fundamental component of the current. The occupancy rates of each harmonic were calculated as follows: ； Among them, h n The occupancy rate of the nth harmonic.

2. The live-line detection and evaluation method for the insulation degradation degree of distribution network cables based on harmonic components according to claim 1, characterized in that, The process of selecting appropriate harmonic orders to initialize weight coefficients and constructing a Lasso regression analysis model includes: The objective function z(b) for constructing the Lasso regression analysis model is: ； in, λ is the predicted value, y is the actual value, λ is the penalty coefficient, and b is the weighting coefficient; The penalty coefficient λ was determined using cross-validation. Then, the variables selected by the Lasso regression analysis model are determined by the penalty coefficient λ and the regression coefficient plot, where variables with a weight coefficient of zero are removed by the Lasso regression analysis model. After determining the penalty coefficient λ, the weight coefficient b is selected by optimizing the least angle regression method.

3. The live-line detection and evaluation method for the insulation degradation degree of distribution network cables based on harmonic components according to claim 2, characterized in that, The steps for optimizing the selection of weight coefficient b using the least angle regression method are as follows: (1) Initialize the weight coefficient b; (2) Initialize the residual vector as the target label vector. ; (3) Select an eigenvector h that is most correlated with the residual vector, find a set of weight coefficients b along the direction of the eigenvector, and another eigenvector h2 that is most correlated with the residual vector appears, so that the new residual vector is equal to the correlation of the two eigenvectors, and recalculate the residual vector. (4) Repeat step (3) to find a set of weight coefficients such that the third feature vector with the greatest correlation to the residual vector makes the new residual vector have the same correlation with these three feature vectors, and so on; (5) When the residual vector is small enough or all eigenvectors have been selected, the iteration ends.

4. The live-line detection and evaluation method for the insulation degradation degree of distribution network cables based on harmonic components according to claim 3, characterized in that, The Lasso regression analysis model is used to predict the degree of degradation of power distribution cables as follows: ； In the formula, b0 is a constant, b2 is the weighting coefficient of the second harmonic, and h2 is the occupancy rate of the second harmonic; b n The nth harmonic weighting coefficient is used; the Lasso regression analysis model continuously and dynamically adjusts the regression coefficients to achieve quantitative prediction and assessment of the degree of degradation of distribution network cables.

Images