Composite insulator thermal defect classification method and system based on random forest algorithm
By using a composite insulator thermal defect classification method based on the random forest algorithm, temperature features are extracted and a random forest model is constructed. This solves the problem of low efficiency in manual qualitative analysis, achieves efficient and accurate identification of thermal defect types, and improves the operational stability and efficiency of the power grid.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ELECTRIC POWER RES INST OF STATE GRID ZHEJIANG ELECTRIC POWER COMAPNY
- Filing Date
- 2022-06-08
- Publication Date
- 2026-05-12
AI Technical Summary
In existing technologies, the qualitative analysis of infrared thermograms of composite insulators by humans is inefficient and inaccurate, making it difficult to accurately identify different types of thermal defects, resulting in indiscriminate replacement and a waste of human and material resources.
A classification method for thermal defects in composite insulators based on the random forest algorithm is adopted. By extracting temperature features, data augmentation is performed using the SMOTE algorithm to construct a random forest classification model. Ten-fold cross-validation is used to select the optimal model, and the classification is performed in combination with the majority rule principle.
This improves the accuracy of thermal defect classification and inspection efficiency of composite insulators, helps maintenance personnel develop targeted solutions, and enhances the power supply capacity and safe operation level of the power grid.
Smart Images
Figure CN115374835B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of defect detection in high-voltage composite insulators, and more particularly to a method and system for classifying thermal defects in composite insulators based on the random forest algorithm. Background Technology
[0002] Composite insulators offer excellent anti-pollution flashover performance and are lightweight, which helps control the construction cost of transmission lines, leading to their widespread use in domestic and international transmission lines. However, during operation, composite insulators may experience string breaks or internal breakdown faults, threatening the safe operation of the line. Composite insulators with aged and damp sheaths show almost no heat generation under low humidity conditions, and only a small heat generation amplitude under high humidity conditions; continued temperature monitoring is sufficient. Composite insulators with surface contamination can be cleaned or replaced depending on the degree of heat generation. Decayed core rods may cause composite insulator breakage accidents; once discovered, they must be replaced immediately. Indiscriminately replacing overheating composite insulators would result in a significant waste of manpower and resources; therefore, identifying the temperature characteristics of composite insulators with different types of thermal defects is particularly important. Currently, qualitative analysis of infrared thermograms is an important means of detecting early internal defects in composite insulators, but it is greatly affected by subjective human factors, resulting in low efficiency and accuracy. Summary of the Invention
[0003] In view of this, the present invention provides a method and system for classifying thermal defects in composite insulators based on the random forest algorithm, which can efficiently and accurately classify composite insulators with different thermal defect types using temperature features.
[0004] This invention provides a method for classifying thermal defects in composite insulators based on a random forest algorithm, comprising the following steps:
[0005] S1. Extract the temperature data of the central axis of the surface of composite insulators with different thermal defect types to obtain temperature curves;
[0006] S2. Perform mathematical analysis on the temperature curves of the central axis of the surface of composite insulators with different thermal defect types, calculate the temperature characteristic quantities of composite insulators with different thermal defect types, and form a temperature characteristic quantity sample set.
[0007] S3. The SMOTE algorithm is used to expand the temperature characteristic sample of composite insulators with rotten core rods to include the number of composite insulators with surface contamination and sheath aging and moisture.
[0008] S4. Use RandomForestClassifier from the sklearn library in Python to construct a random forest algorithm and establish a classification model for thermal defects in composite insulators;
[0009] S5. The composite insulator thermal defect classification model is validated by ten-fold cross-validation, and the classification effect evaluation value of each thermal defect type in each model is calculated. The optimal composite insulator thermal defect classification model is selected based on the effect evaluation value.
[0010] S6. Using the obtained optimal composite insulator thermal defect classification model, classify the thermal defect types of the composite insulators to be classified.
[0011] Furthermore, step S1 specifically includes:
[0012] The types of thermal defects in composite insulators include three categories: core rod rot, surface contamination, and sheath aging and moisture absorption.
[0013] Temperature data along the central axis of the composite insulator surface are extracted, with the extraction direction from the high-voltage end to the low-voltage end. The resulting temperature curve is a series of discrete points along the central axis of the composite insulator surface.
[0014] Furthermore, step S2 specifically includes:
[0015] Seven temperature features were extracted: temperature difference, temperature standard deviation, number of peaks, proportion of heating length, standard deviation of peak value, relative position of maximum temperature, and standard deviation of peak position.
[0016] Furthermore, step S3 specifically includes:
[0017] S31. For each sample x in the core rod decayed composite insulator, calculate the k nearest neighbors using Euclidean distance as the standard;
[0018] S32. Randomly select a sample x from the k nearest neighbors. n And the new sample x is obtained by combining it with the original sample x using the following formula. new ;
[0019] x new =x + rand(0,1)·(x n -x)
[0020] S33. Determine the sampling ratio based on the degree of imbalance in the temperature characteristic quantity sample set;
[0021] S34. Add the augmented data to the temperature feature quantity sample set, and label the data in the sample set.
[0022] Furthermore, the construction of the random forest algorithm in step S4 specifically includes:
[0023] S41. Using the Bootstrap method, 80% of the temperature features are randomly selected with replacement from the temperature feature sample set of composite insulators with different thermal defect types to construct m training sets.
[0024] S42. Randomly select 80% of the temperature features in each training set, and use the CART algorithm to model a decision tree for each training set to obtain m decision trees;
[0025] S43. Test each decision tree with the test set to obtain the classification results of m thermal defect types of composite insulators;
[0026] S44. For the m thermal defect classification results obtained, the majority rule is adopted to vote and determine the thermal defect type of the composite insulator.
[0027] Furthermore, step S5, calculating the performance evaluation value of the composite insulator thermal defect classification model, specifically includes:
[0028] The classification results are presented using a confusion matrix. Based on the results, the precision, recall, and F1 score of each classification model for composite insulators with different thermal defect types are calculated. The classification model with the highest average F1 score is selected as the optimal classification model for thermal defects in composite insulators.
[0029] Precision rate represents the proportion of samples that were predicted to be positive and actually were positive, out of the total number of samples that were predicted to be positive.
[0030]
[0031] Recall rate represents the ratio of the number of samples that were predicted to be positive and were actually positive to the number of samples that were actually positive.
[0032]
[0033] The F1 measure is a weighted harmonic average of precision and recall:
[0034]
[0035] Furthermore, step S6 specifically includes:
[0036] S61. Extract the temperature along the central axis of the surface of the composite insulator to be classified to obtain the temperature curve;
[0037] S62. Analyze the temperature curves and calculate the temperature characteristic quantities of the composite insulators to be classified;
[0038] S63. Input the temperature characteristics of the composite insulator to be classified into the optimal composite insulator thermal defect classification model to obtain multiple classification results;
[0039] S64. The majority rule is adopted for the multiple classification results obtained, and the thermal defect type of the composite insulator to be classified is determined by voting.
[0040] This invention also provides a composite insulator thermal defect classification system based on the random forest algorithm, characterized in that: the system consists of a temperature acquisition unit, a temperature feature acquisition unit, a random forest classification model unit, and a result output unit, wherein the temperature acquisition unit, temperature feature acquisition unit, random forest classification model unit, and result output unit are connected sequentially, wherein:
[0041] Temperature acquisition unit: used to acquire the temperature curve of the central axis of the surface of the composite insulator to be classified, and its output end is connected to the input end of the temperature characteristic quantity acquisition unit;
[0042] Temperature feature acquisition unit: used to acquire the temperature feature of the composite insulator to be classified, and its output is connected to the input of the random forest classification model unit;
[0043] Random forest classification model unit: used to test temperature features and obtain multiple test results; its output is connected to the input of the result display unit.
[0044] The result output unit is used to process multiple classification results obtained from the classification model, and to vote on the thermal defect type of the composite insulator to be classified, based on the principle of majority rule.
[0045] Compared with the prior art, the present invention has the following advantages:
[0046] 1. This invention extracts the overall heating characteristics of composite insulators and introduces seven temperature characteristic quantities to describe composite insulators with thermal defects, which is more accurate;
[0047] 2. This invention improves the inspection efficiency of operation and maintenance personnel and helps them develop different solutions for composite insulators with different types of thermal defects, thereby enhancing the power grid's continuous power supply capacity, stability, and safe operation level. Attached Figure Description
[0048] The present invention will be further described below with reference to the accompanying drawings and embodiments:
[0049] Figure 1 Here is a flowchart of a method for classifying thermal defects in composite insulators based on the random forest algorithm.
[0050] Figure 2 This is a structural diagram of a composite insulator thermal defect classification system based on the random forest algorithm.
[0051] Figure 3 Schematic diagram for constructing the random forest algorithm
[0052] Figure 4 A graph showing the trend of the F1 measure mean as the number of decision trees increases.
[0053] Figure 5Ranking the importance of temperature characteristics of thermal defect composite insulators Detailed Implementation
[0054] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0055] This invention provides a method for classifying thermal defects in composite insulators based on a random forest algorithm, the specific steps of which include:
[0056] S1. Extract the temperature data of the central axis of the surface of composite insulators with different thermal defect types to obtain temperature curves;
[0057] S2. Perform mathematical analysis on the temperature curves of the central axis of the surface of composite insulators with different thermal defect types, calculate the temperature characteristic quantities of composite insulators with different thermal defect types, and form a temperature characteristic quantity sample set.
[0058] S3. The SMOTE algorithm is used to expand the temperature characteristic sample of composite insulators with rotten core rods to include the number of composite insulators with surface contamination and sheath aging and moisture.
[0059] S4. Use RandomForestClassifier from the sklearn library in Python to construct a random forest algorithm and establish a classification model for thermal defects in composite insulators;
[0060] S5. The composite insulator thermal defect classification model is validated by ten-fold cross-validation, and the classification effect evaluation value of each thermal defect type in each model is calculated. The optimal composite insulator thermal defect classification model is selected based on the effect evaluation value.
[0061] S6. Using the obtained optimal composite insulator thermal defect classification model, classify the thermal defect types of the composite insulators to be classified.
[0062] In this embodiment, step S1 specifically includes:
[0063] The types of thermal defects in composite insulators include three categories: core rod rot, surface contamination, and sheath aging and moisture absorption.
[0064] Temperature data along the central axis of the composite insulator surface were extracted using FTIR Tloos software, which is used with an infrared thermal imager. The extraction direction was from the high-voltage end to the low-voltage end, and the exported temperature curve was a series of discrete points along the central axis of the composite insulator surface.
[0065] In this embodiment, step S2 specifically includes:
[0066] Seven temperature features were extracted: temperature difference, temperature standard deviation, number of peaks, proportion of heating length, standard deviation of peak value, relative position of maximum temperature, and standard deviation of peak position.
[0067] In this embodiment, step S3, since the number of composite insulator samples with rotten core rods is relatively small, uses the SMOTE algorithm to augment the data of these samples, thereby increasing the proportion of minority class samples. This can alleviate the overfitting problem without losing valuable information. Specifically, this includes:
[0068] S31. For each sample x in the core rod decayed composite insulator, calculate the k nearest neighbors using Euclidean distance as the standard;
[0069] S32. Randomly select a sample x from the k nearest neighbors. n And the new sample x is obtained by combining it with the original sample x using the following formula. new ;
[0070] x new =x + rand(0,1)·(x n -x)
[0071] S33. Determine the sampling ratio based on the degree of imbalance in the temperature characteristic sample set.
[0072] S34. Add the augmented data to the temperature feature quantity sample set and label the data in the sample set.
[0073] In this embodiment, the construction of the random forest algorithm in step S4 specifically includes:
[0074] S41. Using the Bootstrap method, 80% of the temperature features are randomly selected with replacement from the temperature feature sample set of composite insulators with different thermal defect types to construct m training sets.
[0075] S42. Randomly select 80% of the temperature features in each training set, and use the CART algorithm to model a decision tree for each training set to obtain m decision trees;
[0076] S43. Test each decision tree with the test set to obtain the classification results of m thermal defect types of composite insulators;
[0077] S44. For the obtained m defect type composite insulator classification results, the majority rule is adopted to vote and determine the thermal defect type of the composite insulator.
[0078] In this embodiment, step S5, calculating the effectiveness evaluation value of the composite insulator thermal defect classification model, specifically includes:
[0079] The classification results are displayed using a confusion matrix, which visually shows the prediction performance for each category. The columns of the matrix represent instances of the predicted class, and the rows represent instances of the actual class. Based on the results, the precision, recall, and F1 score of each classification model for composite insulators with different thermal defect types are calculated. The classification model with the highest average F1 score is selected as the optimal composite insulator thermal defect classification model.
[0080] Precision rate represents the proportion of samples that were predicted to be positive and actually were positive, out of the total number of samples that were predicted to be positive.
[0081]
[0082] Recall rate represents the ratio of the number of samples that were predicted to be positive and were actually positive to the number of samples that were actually positive.
[0083]
[0084] The F1 measure is a weighted harmonic average of precision and recall:
[0085]
[0086] In this embodiment, a composite insulator thermal defect classification model constructed using ten-fold cross-validation is shown. The change in the average F1 score of the cross-validation with the increase of the number of decision trees is as follows: Figure 4 As shown, when the number of decision trees n_estimators = 12, the average F1 score is 0.94, reaching its highest value. The F1 scores for composite insulators with surface contamination, sheath aging and dampness, and core rod decay are shown in the table below, which are 0.94, 0.95, and 0.94, respectively. Therefore, the classification model with the number of decision trees n_estimators = 12 is the optimal classification model for thermal defects in composite insulators.
[0087]
[0088] In this embodiment, the feature_importances parameter in the random forest algorithm model was also used to rank the importance of temperature features, and the results are as follows: Figure 5 As shown, in this composite insulator thermal defect classification model based on the random forest algorithm, the most important feature among the seven temperature features is the location of the highest temperature.
[0089] In this embodiment, step S6 specifically includes:
[0090] S61. Use the FTIR Tloos software to extract the temperature data of the central axis of the composite insulator surface. The extraction direction is from the high voltage end to the low voltage end. The exported temperature curve is a series of discrete points along the central axis of the composite insulator surface.
[0091] S62. Analyze the temperature curves and calculate the seven temperature characteristic quantities of the composite insulators to be classified, namely: temperature difference, temperature standard deviation, number of peaks, proportion of heating length, standard deviation of peak value, relative position of maximum temperature and standard deviation of peak position;
[0092] S63. Input the temperature characteristics of the composite insulator to be classified into the optimal composite insulator thermal defect classification model to obtain multiple classification results;
[0093] S64. The majority rule is adopted for the multiple classification results obtained, and the thermal defect type of the composite insulator to be classified is determined by voting.
[0094] This invention also provides a composite insulator thermal defect classification system based on the random forest algorithm, characterized in that: the system consists of a temperature acquisition unit, a temperature feature acquisition unit, a random forest classification model unit, and a result output unit, wherein the temperature acquisition unit, temperature feature acquisition unit, random forest classification model unit, and result output unit are connected sequentially, wherein:
[0095] Temperature acquisition unit: used to acquire the temperature curve of the central axis of the surface of the composite insulator to be classified, and its output end is connected to the input end of the temperature characteristic quantity acquisition unit;
[0096] Temperature feature acquisition unit: Used to acquire the temperature features of the composite insulator to be classified; its output is connected to the input of the random forest classification model unit.
[0097] Random forest classification model unit: used to test temperature features and obtain multiple test results; its output is connected to the input of the result display unit.
[0098] The result output unit is used to process multiple classification results obtained from the classification model, and to vote on the thermal defect type of the composite insulator to be classified, based on the principle of majority rule.
[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A method for classifying thermal defects in composite insulators based on the random forest algorithm, characterized in that: Including the following steps: S1. Extract the temperature data of the central axis of the composite insulator surface for different thermal defect types to obtain temperature curves; the thermal defect types of the composite insulator include three categories: core rod rot, surface contamination, and sheath aging and moisture; use infrared thermal imager software to extract the temperature data of the central axis of the composite insulator surface, with the extraction direction from the high voltage end to the low voltage end, and the exported temperature curve is a series of discrete points along the central axis of the composite insulator surface. S2. Perform mathematical analysis on the temperature curves of the central axis of the surface of composite insulators with different thermal defect types, calculate the temperature characteristic quantities of composite insulators with different thermal defect types, and form a temperature characteristic quantity sample set. S3. The SMOTE algorithm is used to expand the temperature characteristic sample of composite insulators with rotten core rods to include the number of composite insulators with surface contamination and sheath aging and moisture. S4. Use the RandomForestClassifier library in the Python software sklearn to construct a random forest algorithm and establish a classification model for thermal defects in composite insulators; S5. The composite insulator thermal defect classification model is validated by ten-fold cross-validation, and the classification effect evaluation value of each thermal defect type in each model is calculated. The optimal composite insulator thermal defect classification model is selected based on the effect evaluation value. S6. Using the obtained optimal composite insulator thermal defect classification model, classify the thermal defect types of the composite insulators to be classified.
2. The method for classifying thermal defects in composite insulators based on the random forest algorithm according to claim 1, characterized in that: Step S2 specifically includes: Seven temperature features were extracted: temperature difference, temperature standard deviation, number of peaks, proportion of heating length, standard deviation of peak value, relative position of maximum temperature, and standard deviation of peak position.
3. The method for classifying thermal defects in composite insulators based on the random forest algorithm according to claim 1, characterized in that: Step S3 specifically includes: S31. For each sample in the mandrel-rotted composite insulator. The k-nearest neighbor is calculated using Euclidean distance as the standard. S32. Randomly select a sample from the k nearest neighbors. and the original sample The new sample is obtained according to the following formula. ; S33. Determine the sampling ratio based on the degree of imbalance in the temperature characteristic quantity sample set; S34. Add the augmented data to the temperature feature quantity sample set, and label the data in the sample set.
4. The method for classifying thermal defects in composite insulators based on the random forest algorithm according to claim 1, characterized in that: The construction of the random forest algorithm in step S4 specifically includes: S41. Using the Bootstrap method, 80% of the temperature features are randomly selected with replacement from the temperature feature sample set of composite insulators with different thermal defect types to construct m training sets. S42. Randomly select 80% of the temperature features in each training set, and use the CART algorithm to model a decision tree for each training set to obtain m decision trees; S43. Test each decision tree with the test set to obtain the classification results of m thermal defect types of composite insulators; S44. For the m thermal defect classification results obtained, the majority rule is adopted to vote and determine the thermal defect type of the composite insulator.
5. The method for classifying thermal defects in composite insulators based on the random forest algorithm according to claim 1, characterized in that: Step S5, calculating the performance evaluation value of the composite insulator thermal defect classification model, specifically includes: The classification results are presented using a confusion matrix. Based on the results, the precision, recall, and F1 score of each classification model for composite insulators with different thermal defect types are calculated. The classification model with the highest average F1 score is selected as the optimal classification model for thermal defects in composite insulators. Precision rate represents the proportion of samples that were predicted to be positive and actually were positive, out of the total number of samples that were predicted to be positive. Where TP represents the number of samples that were predicted to be positive but were actually positive; FP represents the number of samples that were predicted to be positive but were actually negative; and Precision represents the accuracy. Recall rate represents the proportion of samples that were predicted to be positive and were actually positive, out of the total number of samples that were actually positive. Where FN represents the number of samples that were predicted to be negative but were actually positive; Recall represents the recall rate. The F1 measure is a weighted harmonic average of precision and recall: 。 6. The method for classifying thermal defects in composite insulators based on the random forest algorithm according to claim 1, characterized in that: Step S6 specifically includes: S61. Extract the temperature along the central axis of the surface of the composite insulator to be classified to obtain the temperature curve; S62. Analyze the temperature curves and calculate the temperature characteristic quantities of the composite insulators to be classified; S63. Input the temperature characteristics of the composite insulator to be classified into the optimal composite insulator thermal defect classification model to obtain multiple classification results; S64. The majority rule is adopted for the multiple classification results obtained, and the thermal defect type of the composite insulator to be classified is determined by voting.
7. A composite insulator thermal defect classification system based on random forest algorithm, characterized in that: The system consists of a temperature acquisition unit, a temperature feature acquisition unit, a random forest classification model unit, and a result output unit. These units are connected sequentially. Temperature acquisition unit: used to acquire the temperature curve of the central axis of the surface of the composite insulator to be classified, and its output end is connected to the input end of the temperature characteristic quantity acquisition unit; the thermal defect types of the composite insulator include three categories: core rod decay, surface contamination and sheath aging and moisture; the temperature data of the central axis of the surface of the composite insulator is extracted using the software of the infrared thermal imager, the extraction direction is from the high voltage end to the low voltage end, and the exported temperature curve is a series of discrete points along the central axis of the surface of the composite insulator; Temperature feature acquisition unit: used to acquire the temperature feature of the composite insulator to be classified, and its output is connected to the input of the random forest classification model unit; Random forest classification model unit: used to test temperature features and obtain multiple test results; its output is connected to the input of the result display unit. The result output unit is used to process multiple classification results obtained from the classification model, and to vote on the thermal defect type of the composite insulator to be classified, based on the principle of majority rule.