Lightning arrester fault positioning method

Abstract

The invention discloses a lightning arrester fault positioning method, which comprises the following steps of S1, installing an intelligent monitoring terminal externally connected with multiple sensors near a lightning arrester, monitoring the lightning arrester by adopting the multiple sensors, and preprocessing monitored data; s2, transmitting the monitoring data to a power operation and maintenance center by using the Internet of Things (IoT) to realize remote real-time monitoring; and S3, comparing long-term monitoring data with a normal lightning arrester database, and analyzing whether an abnormal trend exists or not. According to the invention, the fault condition of the lightning arrester is analyzed by combining the data monitored by multi-sensor fusion and adopting an intelligent analysis method, the problem that a slight fault and aging are difficult to distinguish is effectively solved, the fault position is corrected through the AI model, the fault positioning precision is improved, and the fault positioning accuracy is improved. In addition, the service life of the lightning arrester can be effectively prolonged in cooperation with the mode of intelligently adjusting the maintenance period through the AI prediction model, and the use requirement is met.

Description

Technical Field

[0001] The present invention relates to the technical field of arrester fault location, and particularly to a method for arrester fault location. Background Art

[0002] Arrester fault location refers to detecting and determining the location where an arrester fails through technical means for timely maintenance, replacement, or repair. In the prior art, the arrester fault location method is usually single fault detection. Since some arresters will have parameter drift after long-term operation, the existing detection methods are difficult to distinguish equipment aging from potential faults, and there is a risk of misjudgment or missed judgment. In addition, in the prior art, the positioning accuracy is low through the measurement and positioning method. Considering the above situations, the present application proposes a method for arrester fault location. Summary of the Invention

[0003] Based on the technical problems existing in the background art, the present invention proposes a method for arrester fault location.

[0004] A method for arrester fault location proposed by the present invention includes the following steps:

[0005] S1: Install an intelligent monitoring terminal with external multi-sensors near the arrester, monitor the arrester using the multi-sensors, and preprocess the monitored data;

[0006] S2: Transmit the monitored data to the power operation and maintenance center using the Internet of Things (IoT) to achieve remote real-time monitoring;

[0007] S3: Analyze whether there is an abnormal trend by comparing the long-term monitored data with the normal arrester database;

[0008] S4: Automatically classify the health status of the arrester through a support vector machine model;

[0009] S5: The intelligent monitoring terminal automatically notifies the operation and maintenance personnel to handle the abnormal situation of the arrester;

[0010] S6: Calculate the approximate location of the arrester fault using the traveling wave method and correct the error using an AI model to improve the positioning accuracy;

[0011] S7: Combine the operating environment of the arrester to intelligently adjust the maintenance cycle and improve the service life of the arrester.

[0012] Preferably, in the step S1, the multi-sensors include a leakage current sensor, an ultrasonic partial discharge sensor, an infrared thermal imager, and a vibration sensor. The leakage current sensor is used to monitor the change trend of the leakage current of the lightning arrester in the long term. The ultrasonic partial discharge sensor is used to detect whether there is a tiny discharge phenomenon inside the lightning arrester. The infrared thermal imager is used to monitor the surface temperature distribution of the lightning arrester and detect abnormal hot spots. The vibration sensor is used to monitor whether the lightning arrester generates abnormal vibration due to internal aging or damage.

[0013] When preprocessing the data, a filtering algorithm is used to remove power grid noise, improve the signal quality, and perform data normalization and feature extraction on the data. The extracted features include time-domain features, frequency-domain features, time-frequency features, and environmental features. Among them, the time-domain features include leakage current, leakage current volatility, maximum leakage current, partial discharge amplitude, partial discharge root mean square value, temperature change rate, and vibration amplitude. The frequency-domain features include leakage current harmonic content, partial discharge main frequency, high-frequency component energy, and low-frequency power spectral density. The time-frequency features include partial discharge pulse time-frequency diagram, instantaneous energy spectrum, and feature entropy value. The environmental features include relative humidity, environmental temperature, and lightning activity.

[0014] Preferably, the specific logical steps of the step S3 are as follows:

[0015] S301: Retrieve the long-term monitoring data of the lightning arrester with normal operation and the lightning arrester performance parameter standard.

[0016] S302: Calculate the health index, set the normal range, calculate the health status of the current lightning arrester, and compare the health status with the parameters of the lightning arrester with normal operation. If the health index is lower than the typical value in the normal database, there may be an abnormality. The calculation formula is:

[0017] HI = w1×Ileak + w2×PD + w3×T,

[0018] where HI is the health index, Ileak is the leakage current, PD is the partial discharge, T is the operating temperature, and w1, w2, and w3 are influence weight coefficients; if HI ≤ 0.3, the health status is good and the lightning arrester can operate normally; if 0.3 < HI ≤ 0.6, it is slightly aged and the trend change needs to be concerned; if 0.6 < HI ≤ 0.8, it is severely aged and maintenance is recommended; if HI > 0.8, it is a high risk and replacing the lightning arrester is recommended.

[0019] Preferably, the specific logical steps of the step S4 are as follows:

[0020] S401: Label the preprocessed data in S1, and label the normal state (the device is within the normal operating range and there are no obvious signs of failure), minor faults (slight increase in partial discharge, slight temperature fluctuations, but not seriously affecting operation), and severe faults (sharp rise in leakage current and partial discharge, abnormal device operation, possible insulation damage or internal breakdown) respectively;

[0021] S402: Divide the labeled dataset in S401 into a training set and a test set;

[0022] S403: Select a linear SVM model as the training model, use the training set data to train the SVM model, and evaluate the performance of the SVM model through the test set;

[0023] S404: Deploy the trained SVM model in S403 to the intelligent monitoring terminal for real-time monitoring of the health status classification of data;

[0024] S405: Input the real-time monitored arrester data into the SVM model, and the SVM model outputs the health status category of the arrester. The program used by the SVM model for health status classification is:

[0025] from sklearn.svm import SVC

[0026] model = SVC(kernel='rbf')

[0027] model.fit(training data, labels)

[0028] Prediction result = model.predict(test data).

[0029] Preferably, in S6, when calculating the approximate location of the arrester fault using the traveling wave method, the traveling wave signal propagating along the power line starts from the fault point, propagates to the measurement point, and then is received by the measurement system. By measuring the time difference of the signal arriving at each measurement point, the specific location of the fault can be calculated;

[0030] The formula it uses is as follows:

[0031] Where L f is the distance from the fault point to the measurement point, v is the traveling wave propagation speed, and the traveling wave propagation speed is related to the electrical characteristics of the line. Generally, it can be obtained through experiments or theoretical calculations. Δt is the time difference from the fault point to different measurement points.

[0032] Preferably, in S6, the specific steps to correct the error using the AI model are as follows:

[0033] S601: Use the historical traveling wave positioning error data to train the error compensation model, and use cross - validation to optimize the hyperparameters of the error compensation model;

[0034] S602: Use the error compensation model to predict the error of the traveling wave method, and the formula used is: ε = AI(L f , v, Δt, line characteristics), where ε is the error, L f is the initial fault location calculated by the traveling wave method, v is the traveling wave propagation speed, and Δt is the time difference of the traveling wave signals received at the measurement point;

[0035] S603: Correct the fault location according to the error predicted in S602, and the formula used is: L 修正 = L f + ε.

[0036] Preferably, the specific logical steps of S7 are as follows:

[0037] S701: Select the long - short - term memory network model as the AI prediction model;

[0038] S702: Classify the pre - processed monitoring data in S1 into a training set, a validation set, and a test set;

[0039] S703: Use the divided data to train and evaluate the AI prediction model;

[0040] S704: Use the trained AI prediction model to predict the future health index of the arrester, and the formula used is:

[0041] HI future

[0042] = f(Ileakage, Vcapacitance, Rinsulation, Hhumidity, Llightning, Pload), where HI future is the predicted health index and f(·) is the AI prediction model;

[0043] S705: According to the predicted future health index, adjust the maintenance cycle. When HI > 80%, it is healthy and normal inspection is maintained. When 50% ≤ HI ≤ 80%, it is mildly aged and the inspection cycle is shortened. When 30% ≤ HI < 50%, it is moderately aged and maintenance is advanced. When HI < 30%, it is severely aged and immediate repair or replacement is required.

[0044] Preferably, in S5, when the intelligent monitoring terminal notifies the operation and maintenance personnel, the method is one of sending text messages, emails, or work orders, and it notifies in a hierarchical form, which is divided into yellow alarms, orange alarms, and red alarms. The yellow alarm (warning level) is an abnormal trend but does not reach the dangerous value. The orange alarm (fault level) is that the data exceeds the normal range and maintenance is recommended. The red alarm (emergency level) is a serious abnormality in leakage current and partial discharge, and immediate maintenance is required.

[0045] Compared with the existing technology, the beneficial effects of the present invention are:

[0046] The present invention analyzes the fault conditions of the lightning arrester by combining the data of multi-sensor fusion monitoring and using an intelligent analysis method, effectively solves the problem of difficult to distinguish between minor faults and aging, and corrects the fault location through an AI model, improving the fault location accuracy. In addition, by cooperating with the AI prediction model to intelligently adjust the maintenance cycle, the service life of the lightning arrester can be effectively increased to meet the usage requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a flowchart of a method for pinpointing the fault of a lightning arrester proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] The present invention will be further explained below in conjunction with specific embodiments.

[0049] Embodiment

[0050] Refer to Figure 1 , this embodiment proposes a method for pinpointing the fault of a lightning arrester, including the following steps:

[0051] S1: Install an intelligent monitoring terminal with external multi-sensors near the lightning arrester, monitor the lightning arrester using multi-sensors, and preprocess the monitored data;

[0052] Among them, the multi-sensors include a leakage current sensor, an ultrasonic partial discharge sensor, an infrared thermal imager, and a vibration sensor. The leakage current sensor is used to monitor the change trend of the leakage current of the lightning arrester for a long time. The ultrasonic partial discharge sensor is used to detect whether there is a small discharge phenomenon inside the lightning arrester. The infrared thermal imager is used to monitor the surface temperature distribution of the lightning arrester and detect abnormal hot spots. The vibration sensor is used to monitor whether the lightning arrester generates abnormal vibration due to internal aging or damage;

[0053] When preprocessing data, a filtering algorithm is used to remove power grid noise, improve signal quality, and perform data normalization and feature extraction on the data. The extracted features include time-domain features, frequency-domain features, time-frequency features, and environmental features. Among them, the time-domain features include leakage current, leakage current volatility, maximum leakage current, partial discharge amplitude, partial discharge root mean square value, temperature change rate, and vibration amplitude. The frequency-domain features include leakage current harmonic content, partial discharge main frequency, high-frequency component energy, and low-frequency power spectral density. The time-frequency features include partial discharge pulse time-frequency diagram, instantaneous energy spectrum, and characteristic entropy value. The environmental features include relative humidity, ambient temperature, and lightning activity;

[0054] In addition, the intelligent monitoring terminal is built with an AI processing chip to analyze sensor data in real time, reduce data transmission volume, and improve response speed;

[0055] S2: Use the Internet of Things (IoT) to transmit the monitoring data to the power operation and maintenance center to achieve remote real-time monitoring;

[0056] In addition, the intelligent monitoring terminal also supports multiple communication methods, including 5G / 4G LTE, LoRa / NB-IoT, fiber / ethernet, and Wi-Fi / Zigbee. Among them, 5G / 4G LTE is suitable for remote areas or unattended sites, LoRa / NB-IoT is suitable for low-power, long-distance wireless communication requirements, fiber / ethernet is suitable for fixed places such as substations, and Wi-Fi / Zigbee is suitable for short-distance communication between local devices. Operation and maintenance personnel can view real-time data through a PC or mobile phone APP;

[0057] S3: Compare the long-term monitoring data with the normal arrester database to analyze whether there is an abnormal trend. The specific steps are as follows:

[0058] S301: Retrieve the long-term monitoring data and arrester performance parameter standards of normally operating arresters;

[0059] S302: Calculate the health index, set the normal range, calculate the health status of the current arrester, and compare the health status with the parameters of normally operating arresters. If the health index is lower than the typical value in the normal database, there may be an abnormality. The calculation formula is:

[0060] HI = w1×Ileak + w2×PD + w3×T,

[0061] Among them, HI is the health index, Ileak is the leakage current, PD is the partial discharge, T is the operating temperature, and w1, w2, and w3 are the influence weight coefficients; if HI ≤ 0.3, the health state is good and the arrester can operate normally; if 0.3 < HI ≤ 0.6, it is slightly aged and the trend change needs to be concerned; if 0.6 < HI ≤ 0.8, it is severely aged and maintenance is recommended; if HI > 0.8, it is a high risk and replacing the arrester is recommended;

[0062] S4: Automatically classify the health state of the arrester through the support vector machine model;

[0063] The specific logical steps are as follows:

[0064] S401: Label the data preprocessed in S1, and label the normal state (the equipment is within the normal operating range and there are no obvious fault signs), minor faults (the partial discharge increases slightly and the temperature fluctuates slightly, but it has not seriously affected the operation), and serious faults (the leakage current and partial discharge rise sharply, the equipment operates abnormally, and there may be insulation damage or internal breakdown);

[0065] S402: Divide the data set labeled in S401 into a training set and a test set;

[0066] S403: Select the linear SVM model as the training model, use the training set data to train the SVM model, and evaluate the performance of the SVM model through the test set;

[0067] S404: Deploy the trained SVM model in S403 to the intelligent monitoring terminal for real-time monitoring of the health state classification of data;

[0068] S405: Input the real-time monitored arrester data into the SVM model, and the SVM model outputs the health state category of the arrester. The program used by the SVM model for health state classification is:

[0069] from sklearn.svm import SVC

[0070] model = SVC(kernel = 'rbf')

[0071] model.fit(training data, labels)

[0072] prediction result = model.predict(test data);

[0073] S5: The intelligent monitoring terminal automatically notifies the operation and maintenance personnel of the abnormal conditions of the lightning arrester. When the intelligent monitoring terminal notifies the operation and maintenance personnel, the method is one of sending text messages, emails or work orders. It notifies in a hierarchical form, which is divided into yellow alarms, orange alarms and red alarms. The yellow alarm (warning level) is for abnormal trends but not reaching the dangerous value. The orange alarm (fault level) is for data exceeding the normal range, and maintenance is recommended. The red alarm (emergency level) is for serious abnormalities in leakage current and partial discharge, and immediate maintenance is required;

[0074] S6: Use the traveling wave method to calculate the approximate location of the lightning arrester fault, and use the AI model to correct the error and improve the positioning accuracy;

[0075] When using the traveling wave method to calculate the approximate location of the lightning arrester fault, the traveling wave signal propagating along the power line starts from the fault point, propagates to the measurement point, and then is received by the measurement system. By measuring the time difference of the signal arriving at each measurement point, the specific location of the fault can be calculated;

[0076] The formula it uses is as follows:

[0077] Among them, L f is the distance from the fault point to the measurement point, v is the traveling wave propagation speed, and the traveling wave propagation speed is related to the electrical characteristics of the line. Generally, it can be obtained through experiments or theoretical calculations. Δt is the time difference from the fault point to different measurement points;

[0078] The specific steps to correct the error using the AI model are as follows:

[0079] S601: Use historical traveling wave positioning error data to train the error compensation model, and use cross-validation to optimize the hyperparameters of the error compensation model;

[0080] S602: Use the error compensation model to predict the error of the traveling wave method. The formula it uses is: ε = AI(L f , v, Δt, line characteristics), where ε is the error, L f is the initial fault location calculated by the traveling wave method, v is the traveling wave propagation speed, and Δt is the time difference for the measurement point to receive the traveling wave signal;

[0081] S603: Correct the fault location according to the error predicted in S602. The formula it uses is: L 修正 = L f + ε;

[0082] S7: Combine the operating environment of the lightning arrester to intelligently adjust the maintenance cycle and improve the service life of the lightning arrester;

[0083] The specific logical steps are as follows:

[0084] S701: Select the long short-term memory network model as the AI prediction model;

[0085] S702: Classify the preprocessed monitoring data in S1 into a training set, a validation set, and a test set;

[0086] S703: Use the divided data to train and evaluate the AI prediction model;

[0087] S704: Use the trained AI prediction model to predict the future health index of the lightning arrester. The formula used is:

[0088] HI future

[0089] = f(Ileakage, Vcapacitance, Rinsulation, Hhumidity, Llightning, Pload), where HI future is the predicted health index and f(·) is the AI prediction model;

[0090] S705: According to the predicted future health index, adjust the maintenance cycle. When HI > 80%, it is healthy and normal inspection is maintained. When 50% ≤ HI ≤ 80%, it is mildly aged and the inspection cycle is shortened. When 30% ≤ HI < 50%, it is moderately aged and maintenance is advanced. When HI < 30%, it is severely aged and immediate repair or replacement is required;

[0091] This embodiment analyzes the fault situation of the lightning arrester by combining the data monitored by multi-sensor fusion and using an intelligent analysis method, effectively solving the problem of difficult distinction between minor faults and aging. Moreover, the fault location is corrected by the AI model, improving the fault location accuracy. In addition, by cooperating with the AI prediction model to intelligently adjust the maintenance cycle, the service life of the lightning arrester can be effectively increased to meet the usage requirements.

[0092] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A method for locating the fault point of a lightning arrester, characterized in that, It includes the following steps: S1: Install an intelligent monitoring terminal with external multi-sensors near the arrester, use the multi-sensors to monitor the arrester, and preprocess the monitored data; S2: Transmit the monitored data to the power operation and maintenance center through the Internet of Things to achieve remote real-time monitoring; S3: Analyze whether there is an abnormal trend by comparing the long-term monitoring data with the normal arrester database; S4: Automatically classify the health status of the arrester through a support vector machine model; S5: The intelligent monitoring terminal automatically notifies the operation and maintenance personnel to handle the abnormal situation of the arrester; S6: Use the traveling wave method to calculate the approximate location of the arrester fault, and use the AI model to correct the error to improve the positioning accuracy; S7: Combine the operating environment of the arrester, intelligently adjust the maintenance cycle, and improve the service life of the arrester.

2. The method for locating the fault of a lightning arrester according to claim 1, wherein In the above S1, the multi-sensors include a leakage current sensor, an ultrasonic partial discharge sensor, an infrared thermal imager, and a vibration sensor. The leakage current sensor is used to monitor the change trend of the leakage current of the arrester for a long time. The ultrasonic partial discharge sensor is used to detect whether there is a tiny discharge phenomenon inside the arrester. The infrared thermal imager is used to monitor the surface temperature distribution of the arrester and detect abnormal hot spots. The vibration sensor is used to monitor whether the arrester generates abnormal vibration due to internal aging or damage; When preprocessing the data, a filtering algorithm is used to remove the power grid noise to improve the signal quality, and the data is normalized and feature extracted. The extracted features include time domain features, frequency domain features, time-frequency features, and environmental features. Among them, the time domain features include leakage current, leakage current volatility, maximum leakage current, partial discharge amplitude, partial discharge root mean square value, temperature change rate, and vibration amplitude. The frequency domain features include leakage current harmonic content, partial discharge main frequency, high-frequency component energy, and low-frequency power spectral density. The time-frequency features include partial discharge pulse time-frequency diagram, instantaneous energy spectrum, and feature entropy value. The environmental features include relative humidity, environmental temperature, and lightning activity.

3. A method for pinpointing the location of a lightning arrester fault according to claim 1, characterized in that, The specific logical steps of the above S3 are as follows: S301: Retrieve the long-term monitoring data and arrester performance parameter standards of the normally operating arrester; S302: Calculate the health index, set the normal range, calculate the health status of the current arrester, and compare the health status with the parameters of the normally operating arrester. If the health index is lower than the typical value in the normal database, there may be an abnormality. The calculation formula is: HI = w1×Ileak + w2×PD + w3×T, where HI is the health index, Ileak is the leakage current, PD is the partial discharge, T is the operating temperature, and w1, w2, and w3 are influence weight coefficients; when HI≤0.3, the health status is good and the arrester can operate normally; when 0.3 < HI≤0.6, it is slightly aged and the trend change needs to be concerned; when 0.6 < HI≤0.8, it is severely aged and overhaul is recommended; when HI > 0.8, it is a high risk and replacing the arrester is recommended.

4. A method for pinpointing the location of an arrester fault according to claim 1, characterized in that, The specific logical steps of the above S4 are as follows: S401: Label the preprocessed data in S1, and label the normal state, minor fault, and severe fault respectively; S402: Divide the labeled data set in S401 into a training set and a test set; S403: Select the linear SVM model as the training model, use the training set data to train the SVM model, and evaluate the performance of the SVM model through the test set; S404: Deploy the trained SVM model in S403 to the intelligent monitoring terminal for real-time monitoring of the health status classification of data; S405: Input the real-time monitored arrester data into the SVM model, and the SVM model outputs the health status category of the arrester. The program used by the SVM model for health status classification is as follows: from sklearn.svm import SVC model = SVC(kernel='rbf') model.fit(training data, labels) prediction result = model.predict(test data).

5. A method for pinpointing the location of an arrester fault according to claim 1, characterized in that, In S6, when calculating the approximate location of the arrester fault using the traveling wave method, the traveling wave signal propagating along the power line starts from the fault point, propagates to the measurement point, and then is received by the measurement system. By measuring the time difference of the signal reaching each measurement point, the specific location of the fault can be calculated; The formula used is as follows: where L f is the distance from the fault point to the measurement point, v is the traveling wave propagation speed, and the traveling wave propagation speed is related to the electrical characteristics of the line, which can generally be obtained through experiments or theoretical calculations. Δt is the time difference from the fault point to different measurement points.

6. A method for locating the fault point of a lightning arrester according to claim 1, characterized in that, In S6, the specific steps to correct the error using the AI model are as follows: S601: Use the historical traveling wave positioning error data to train the error compensation model, and use cross-validation to optimize the hyperparameters of the error compensation model; S602: Use the error compensation model to predict the error of the traveling wave method. The formula used is: ε = AI(L f , v, Δt, line characteristics); where ε is the error, L f is the initial fault location calculated by the traveling wave method, v is the traveling wave propagation speed, and Δt is the time difference of the traveling wave signals received at the measurement points; S603: Correct the fault location according to the prediction error in S602, and the formula used is: L 修正 = L f + ε.

7. A method for locating the fault point of a lightning arrester according to claim 1, characterized in that The specific logical steps of S7 are as follows: S701: Select the long short-term memory network model as the AI prediction model; S702: Classify the preprocessed monitoring data in S1 into a training set, a validation set, and a test set; S703: Use the divided data to train and evaluate the AI prediction model; S704: Use the trained AI prediction model to predict the future health index of the arrester. The formula used is: HI future = f(Ileakage, Vcapacitance, Rinsulation, Hhumidity, Llightning, Pload), where HI future is the predicted health index and f(·) is the AI prediction model; S705: According to the predicted future health index, adjust the maintenance cycle. When HI > 80%, it is healthy and normal inspection is maintained. When 50% ≤ HI ≤ 80%, it is slightly aged and the inspection cycle is shortened. When 30% ≤ HI < 50%, it is moderately aged and maintenance is advanced. When HI < 30%, it is severely aged and immediate repair or replacement is required.

8. A method for pinpointing the location of an arrester fault according to claim 1, characterized in that, In S5, the way the intelligent monitoring terminal notifies the operation and maintenance personnel is one of sending text messages, emails, or work orders. It notifies in a hierarchical form, which is divided into yellow warnings, orange warnings, and red warnings. A yellow warning means the trend is abnormal but does not reach the dangerous value. An orange warning means the data exceeds the normal range and maintenance is recommended. A red warning means a serious abnormality in leakage current and partial discharge, and immediate repair is required.

Images