Electric tower insulator multi-element damage detection method and equipment based on multi-layer sensor

Abstract

The invention discloses an electric tower insulator multi-element damage detection method and device based on a multi-layer sensor, and the method comprises the following steps: S1, collecting the transverse vibration frequencies of various types of insulators in a normal state and / or various damage abnormal states, and constructing a transverse vibration data set needed by the multi-layer sensor; s2, performing classification training on the multi-layer perceptron through the transverse vibration data set until the accuracy rate, the precision rate, the recall rate and the F1 value of the multi-layer perceptron reach expected targets, and completing classification training of the multi-layer perceptron; and S3, identifying and judging the transverse vibration data of the insulator to be detected through the trained multi-layer sensor, and outputting a corresponding insulator type and a current state. According to the method, transverse vibration signals of various types of insulators in different states are collected, a transverse vibration data set related to multivariate damage is established, then classification prediction training is performed on the multilayer sensor by using the transverse vibration data set, instance detection is performed on the trained multilayer sensor field insulator, and the detection accuracy of the multilayer sensor field insulator is improved. The insulator type and the damage state can be rapidly detected.

Description

Technical Field

[0001] The present invention relates to the technical field of power insulator detection, and in particular to a method and device for detecting multi-element damage of an electric tower insulator based on a multi-layer sensor. Background Art

[0002] As an important insulating control component to protect overhead power transmission lines, insulators can ensure the safe connection between wires / conductors and grounding components at different points, improve electrical transportation safety and the service life of towers. However, since insulators are usually made of brittle materials such as porcelain or glass, and are exposed to high altitudes without shelter for a long time, they are easily affected by the harsh climate of the natural environment and long-term electric charge, resulting in functional failure and damaging the service life of the line. Therefore, regular electrical safety transmission inspections are particularly important.

[0003] Conventional insulator inspection methods currently available generally include manual climbing inspection, climbing robot inspection, and contactless inspection. The timeliness of manual climbing inspection is low, and the inspection results are affected by the subjective factors of the inspectors. At the same time, the safety of the inspectors is also worth considering. The detection accuracy of climbing robot inspection is affected by factors such as the inspection operator's proficiency and the accuracy of the equipment equipped by the inspection robot, and there may be problems with uneven detection accuracy.

[0004] The non-contact inspection mainly relies on computer vision technology, which is easily affected by factors such as image acquisition perspective, light, occlusion, weather, etc. There is a big discrepancy between the training data set and the actual inspection image, and computer vision technology is difficult to detect the type of damage inside the insulator. Although the use of laser, ultrasonic, vibration wave and other detection methods can make up for the above shortcomings, the current research focus is mainly on defect imaging of insulators. The defects after imaging still need to be manually judged, which is difficult to use as an all-weather, all-type, and automated detection method. Summary of the invention

[0005] The object of the present invention is to provide a method and device for detecting multi-element damage of an insulator of an electric tower based on a multi-layer sensor, so as to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The multi-layer perceptron-based method for detecting multivariate damage of an insulator of an electric tower comprises the following steps:

[0008] S1. Collect the lateral vibration frequencies of various types of insulators in normal state and / or various abnormal damage states, and construct the lateral vibration data set required by the multi-layer perceptron;

[0009] S2. Classification training of the multilayer perceptron is performed using the above lateral vibration data set until the accuracy, precision, recall rate and F1 value of the multilayer perceptron reach the expected target, thus completing the classification training of the multilayer perceptron;

[0010] S3. The lateral vibration data of the insulator to be tested is identified and judged through the trained multi-layer perceptron, and the corresponding insulator type and current status are output.

[0011] As a further solution of the present invention, the method for constructing the lateral vibration data set in step S1 is:

[0012] Classify the collected lateral vibration frequencies according to insulator type and condition;

[0013] Obtain the autopower spectrum corresponding to each classified lateral vibration frequency, and calculate the kurtosis index K and mean square frequency index MSF of the autopower spectrum;

[0014] The autopower spectrum, kurtosis index K and mean square frequency index MSF corresponding to various types of insulators in various states are classified and stored to form a lateral vibration data set.

[0015] As a further solution of the present invention, in step S1, the lateral vibration frequency is obtained by directly irradiating the target insulator with a scanning laser Doppler vibrometer, and the irradiation acquisition time is greater than or equal to 8 seconds.

[0016] As a further solution of the present invention, the type of insulator in step S1 includes but is not limited to: at least one of a porcelain / glass disk type suspension insulator, a composite insulator, a porcelain long rod insulator, and a porcelain / glass composite disk type suspension insulator.

[0017] As a further solution of the present invention, the state of the composite insulator is divided into: normal state and broken string state (damage abnormal state), wherein the causes of the broken string state of the composite insulator include but are not limited to: external force tearing, brittle fracture (stress corrosion) and / or rotten fracture (core rod carbonization);

[0018] The states of porcelain disc type suspension insulators, porcelain long rod insulators and porcelain composite disc type suspension insulators are divided into: normal state and broken string / zero value state (damage abnormal state). The causes of broken string / zero value state include but are not limited to: cracking due to poor glue assembly, damage to the insulator head due to lightning strike and / or cracking of the steel foot due to external force;

[0019] The states of glass disc type suspension insulators and glass composite disc type suspension insulators are divided into: normal state and broken string state (damage abnormal state), among which the broken string state includes but is not limited to: radial self-explosion caused by quality problems, fish scale-like self-explosion caused by external force and / or anode corrosion of insulator steel cap caused by electrolysis of rain, snow and water.

[0020] As a further solution of the present invention, the autopower spectrum is calculated using discrete signals, specifically:

[0021] The lateral vibration frequency data x(n) with a length of N, n = 0, 1, ..., N-1 is divided into L segments, each segment has M data, where the expression of the k-th segment data is:

[0022] x k (n)=x(n+kM-M),0≤n≤M,1≤k≤L;

[0023] Then the auto-power spectrum S xx (e jω ) is:

[0024]

[0025] Wherein, w(n) is the window function, j is an imaginary number, and ω=2π / L.

[0026] The calculation expression of the kurtosis index K is:

[0027]

[0028] The calculation expression of the mean square frequency index MSF is:

[0029]

[0030] Among them, x i is the discrete value of the collected lateral vibration frequency, is the mean value of the collected lateral vibration frequency, N is the length of the collected lateral vibration frequency data, fi is the frequency position corresponding to the auto-power spectrum, S(f i ) is the frequency position f i The corresponding autopower spectrum value.

[0031] As a further solution of the present invention, the lateral vibration data set has a hierarchical structure, the first layer is the insulator type, and the second layer is the kurtosis index K and the mean square frequency index MSF calculated for each type of insulator under the corresponding state.

[0032] As a further solution of the present invention, the desired goal in step S2 is specifically: the accuracy, precision, recall rate and F1 value of the multilayer perceptron meet the performance levels of 0.95, 0.95, 0.95 and 0.95 respectively, thereby ensuring that the error detection rate is within 5%.

[0033] As a further solution of the present invention, in step S3, if the insulator type predicted by the multilayer perceptron is wrong, re-detection is required; if the multilayer perceptron still cannot predict correctly after multiple detections, it is necessary to include the error case records in the lateral vibration data set as supplementary training data to compensate for the poor generalization of the model caused by incomplete early data collection.

[0034] The multi-layered perceptron-based multi-dimensional damage detection device for electric tower insulators stores computer instructions, and when the computer instructions are executed by a processor, the multi-dimensional damage detection method is executed.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] 1. The present invention first collects lateral vibration signals of various types of insulators under different states, establishes a lateral vibration data set related to multivariate damage, and then uses the lateral vibration data set to perform classification prediction training on a multilayer perceptron. The trained multilayer perceptron is then used to perform instance detection on insulators on site, so as to quickly detect the type and damage state of the insulator.

[0037] 2. The present invention provides a multi-layer perceptron-based method for detecting multivariate damage of electric tower insulators, which can be applied to the detection of different types of insulators in different damage states; it can continuously learn from error records during the detection process and continuously supplement the data set, and has better generalization;

[0038] 3. In specific applications, the present invention only requires one supporting device of a laser Doppler vibrometer to collect the lateral vibration data of the insulator, and the laser scanning distance is relatively long, which is much higher than conventional non-contact detection methods based on photographic images;

[0039] 4. The present invention is relatively simple in processing collected data, does not require complex calculation processes, and has fewer input and output layers for the multi-layer perceptron, which is more conducive to lightweight deployment of on-site detection processes. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic diagram of the process of the multi-element damage detection method of insulators of the present invention;

[0041] Figure 2 It is a specific logic operation flow chart of the insulator multi-element damage detection method of the present invention;

[0042] Figure 3 A schematic diagram of the structure of detecting an insulator using the method of the present invention;

[0043] Figure 4 A specific example of the transverse vibration frequency of an insulator collected using the method of the present invention;

[0044] Figure 5 The back propagation neural network model adopted by the method of the present invention;

[0045] Figure 6 It is a multi-population genetic algorithm structure used to optimize the back-propagation neural network in the method of the present invention.

[0046] In the figure: 1. Scanning laser Doppler vibrometer; 2. Target insulator; 3. Laser Doppler controller; 4. Computer analysis terminal; 5. Laser signal; 6. Data acquisition transmission line; 7. Data input channel; 8. Data output channel; 9. Data output transmission line; 10. Time domain data; 11. Frequency domain data. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] See also Figures 1-2 The multi-layer perceptron-based multi-element damage detection method for electric tower insulators comprises the following steps:

[0049] S1. Collect the lateral vibration frequencies of various types of insulators in normal state and / or various abnormal damage states, and construct the lateral vibration data set required by the multi-layer perceptron;

[0050] S2. Classification training of the multilayer perceptron is performed using the above lateral vibration data set until the accuracy, precision, recall rate and F1 value of the multilayer perceptron reach the expected target, thus completing the classification training of the multilayer perceptron;

[0051] S3. The lateral vibration data of the insulator to be tested is identified and judged through the trained multi-layer perceptron, and the corresponding insulator type and current status are output.

[0052] In the present invention, step S1 and step S2 are training stages, and step S3 is a detection stage.

[0053] The method for constructing the lateral vibration data set in step S1 is:

[0054] Classify the collected lateral vibration frequencies according to insulator type and condition;

[0055] Obtain the autopower spectrum corresponding to each classified lateral vibration frequency, and calculate the kurtosis index K and mean square frequency index MSF of the autopower spectrum;

[0056] The autopower spectrum, kurtosis index K and mean square frequency index MSF corresponding to various types of insulators in various states are classified and stored to form a lateral vibration data set.

[0057] Specifically, Figure 3 As shown, in step S1, the lateral vibration frequency is obtained by directly irradiating the target insulator 2 with the scanning laser Doppler vibrometer 1, and the irradiation and acquisition time of each group of target insulators 2 is greater than or equal to 8 seconds. Figure 4 The time domain data 10 and frequency domain data 11 of the lateral vibration collected in a certain embodiment are shown.

[0058] As a specific solution, the type of insulator in step S1 includes but is not limited to: at least one of a porcelain / glass disk type suspension insulator, a composite insulator, a porcelain long rod insulator, and a porcelain / glass composite disk type suspension insulator.

[0059] The states of the composite insulator are divided into: normal state and broken string state (damage abnormal state), wherein the causes of the broken string state of the composite insulator include but are not limited to: external force tearing, brittle fracture (stress corrosion) and / or rotten fracture (core rod carbonization);

[0060] The states of porcelain disc type suspension insulators, porcelain long rod insulators and porcelain composite disc type suspension insulators are divided into: normal state and broken string / zero value state (damage abnormal state). The causes of broken string / zero value state include but are not limited to: cracking due to poor glue assembly, damage to the insulator head due to lightning strike and / or cracking of the steel foot due to external force;

[0061] The states of glass disc type suspension insulators and glass composite disc type suspension insulators are divided into: normal state and broken string state (damage abnormal state), among which the broken string state includes but is not limited to: radial self-explosion caused by quality problems, fish scale-like self-explosion caused by external force and / or anode corrosion of insulator steel cap caused by electrolysis of rain, snow and water.

[0062] Specifically, in a laboratory environment, different preparation methods can be used to produce insulator samples with different damage types; the preparation methods used must be related to the environmental factors to which the insulators are subjected during operation, and the preparation methods include but are not limited to: using freeze-thaw cycle tests to simulate the damage to insulators caused by harsh rain and snow environments, using a lateral vibration table to simulate the damage to insulators caused by strong wind environments, and using tension tests to simulate external tearing damage to insulators.

[0063] The auto-power spectrum of each type of insulator in different states is calculated. In a preferred embodiment, the auto-power spectrum is calculated using discrete signals, specifically:

[0064] The lateral vibration frequency data x(n) with a length of N, n = 0, 1, ..., N-1 is divided into L segments, each segment has M data, where the expression of the k-th segment data is:

[0065] x k (n)=x(n+kM-M),0≤n≤M,1≤k≤L;

[0066] Then the auto-power spectrum S xx (e jω ) is:

[0067]

[0068] Wherein, w(n) is the window function, j is an imaginary number, and ω=2π / L.

[0069] The kurtosis index K and mean square frequency index MSF of each type of insulator in different states are calculated. In a preferred embodiment, the calculation expression of the kurtosis index K is:

[0070]

[0071] The calculation expression of the mean square frequency index MSF is:

[0072]

[0073] Among them, x i is the discrete value of the collected lateral vibration frequency, is the mean value of the collected lateral vibration frequency, N is the length of the collected lateral vibration frequency data, fi is the frequency position corresponding to the auto-power spectrum, S(f i ) is the frequency position f i The corresponding autopower spectrum value.

[0074] The autopower spectra, kurtosis index K and mean square frequency index MSF of various types of insulators in normal state and / or different damage abnormal state are classified and stored to prepare the lateral vibration data set required by the multi-layer perceptron, wherein the lateral vibration data set has a hierarchical structure, and the first layer is the insulator type, specifically including but not limited to: porcelain / glass disk type suspension insulator, composite insulator, porcelain long rod insulator, porcelain / glass composite disk type suspension insulator, etc.;

[0075] The second layer is the kurtosis index K and mean square frequency index MSF calculated for each type of insulator under the corresponding state. The state types include but are not limited to: normal state, tearing due to external force, brittle fracture (stress corrosion), decay fracture (core rod carbonization), cracking due to poor glue assembly, damage to the insulator head due to lightning strike, cracking of the steel foot due to external force, radial self-explosion caused by quality problems, fish scale self-explosion caused by external force, and anode corrosion of the insulator steel cap caused by electrolysis of rain and snow water.

[0076] As a specific solution, the desired goal in step S2 is that the accuracy, precision, recall and F1 value of the multilayer perceptron meet the performance levels of 0.95, 0.95, 0.95 and 0.95 respectively, thereby ensuring that the error detection rate is within 5%.

[0077] In a preferred embodiment, the multilayer perceptron can specifically use a multi-population genetic algorithm driven back propagation neural network model (MPGA-BPNN) to learn the classified lateral vibration data set, wherein, for example Figure 5 As shown in Figure 2, the hidden layer weight matrix and bias vector of the back propagation neural network (BPNN) model are obtained by global optimization of the multi-population genetic algorithm (MPGA). The structure of the multi-population genetic algorithm (MPGA) is as follows: Figure 6 As shown;

[0078] Furthermore, the input of the back propagation neural network model (BPNN) is the autopower spectrum, kurtosis index K and mean square frequency index MSF of each type of insulator in different states, that is, the input is a three-dimensional array, the number of hidden layers of BPNN is 3 layers, the first layer has 12 neurons, the second layer has 9 neurons, and the third layer has 6 neurons, and the output is the type and state of the insulator;

[0079] Furthermore, the output of the BPNN is divided into two stages. The first stage is the insulator type. When the accuracy, precision, recall and F1 value of the first stage can meet 0.95, 0.95, 0.95 and 0.95 respectively, the state of the insulator in the second stage is output, which is the normal state and / or other damage abnormal state.

[0080] In specific applications, the trained multi-layer perceptron is integrated into the computer analysis terminal 4, which can be Figure 3 This concludes the training phase in a device such as a portable laptop computer with a multi-layer perceptron deployed.

[0081] The detection stage of the present invention may specifically include the following process:

[0082] The scanning laser Doppler vibrometer 2 is used to collect the laser signal 5 of the lateral vibration of the insulator to be tested, and the collection time is not less than 8s. The collected data is connected to the data input channel 7 of the laser Doppler controller 3 via the data collection transmission line 6, and then transmitted to the laser Doppler controller 3. The laser Doppler controller 3 converts the collected signal into a digital signal, and transmits it from the data output channel 8 of the laser Doppler controller 3 to the computer analysis terminal 4 via the data output transmission line 9.

[0083] The computer analysis terminal 4 calculates the corresponding autopower spectrum, kurtosis index K and mean square frequency index MSF according to the collected lateral vibration frequency.

[0084] The obtained autopower spectrum, kurtosis index K and mean square frequency index MSF are input into the integrated multilayer perceptron for inference prediction. First, the type of the insulator is output. If the type prediction is correct, the state of the insulator is output.

[0085] If the type prediction is wrong, the lateral vibration frequency of the insulator to be tested can be collected again, and prediction can be made based on the secondary collected data. If the predicted insulator type is correct, the state of the insulator will continue to be output;

[0086] If the type of secondary prediction is wrong, the error case record is included in the lateral vibration dataset as a supplementary dataset to improve the performance of the subsequent multi-layer perceptron, thereby compensating for the poor generalization of the model caused by incomplete data collection in the early stage.

[0087] The multi-layered perceptron-based multi-dimensional damage detection device for electric tower insulators stores computer instructions, and when the computer instructions are executed by a processor, the multi-dimensional damage detection method is executed.

[0088] Example 1

[0089] The computer analysis terminal 4 in this embodiment has integrated the trained MPGA-BPNN model. It is necessary to test the insulators of the GJ31 (GN83+1) tower of the Chengde East (Kuancheng) ~ Luanxian 500 kV line project. The number of insulator samples tested is 10 groups, and the insulator types are all porcelain disc type suspension insulators. The specific operation steps are as follows:

[0090] A PSV500 scanning laser Doppler vibrometer, a laser Doppler controller 3 and a computer analysis terminal 4 (notebook) were set up 10 meters away from the GJ31 (GN83+1) power tower. The above equipment are all supporting equipment for the PSV500 scanning laser Doppler vibrometer. The computer analysis terminal 4 is additionally equipped with a trained MPGA-BPNN model and a python test environment.

[0091] Align the transmitter of the PSV500 scanning laser Doppler vibrometer to the tower body for coordinate and laser calibration;

[0092] The calibrated PSV500 scanning laser Doppler vibrometer then raises the transmitter elevation angle, aligns it with the insulator to be tested, and opens the data acquisition channel to collect the lateral vibration frequency of the insulator to be tested. The acquisition time is 15 seconds per set.

[0093] After collecting data from 10 groups of insulators to be tested, the collected data are transmitted to the computer analysis terminal via the laser Doppler controller;

[0094] The computer analysis terminal converts the collected data into the corresponding autopower spectrum, kurtosis index K and mean square frequency index MSF through Python;

[0095] Immediately start the deployed python test environment, input the above 10 groups of autopower spectra, kurtosis index K and mean square frequency index MSF into the MPGA-BPNN model, and predict the insulator type. The prediction results are all correct.

[0096] Finally, the MPGA-BPNN model further outputs the predicted insulator state. The specific results are shown in the following table.

[0097] Table 1: Insulator status detection results of Example 1.

[0098]

[0099] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A multi-layer perceptron-based multi-element damage detection method for electric tower insulators, characterized in that: The steps include: S1. Collect the lateral vibration frequencies of various types of insulators in normal state and / or various abnormal damage states, and construct the lateral vibration data set required by the multi-layer perceptron; S2. Classification training of the multilayer perceptron is performed using the above lateral vibration data set until the accuracy, precision, recall rate and F1 value of the multilayer perceptron reach the expected target, thus completing the classification training of the multilayer perceptron; S3. The lateral vibration data of the insulator to be tested is identified and judged through the trained multi-layer perceptron, and the corresponding insulator type and current status are output.

2. The multi-layer perceptron-based multi-element damage detection method for electric tower insulators according to claim 1 is characterized in that: The method for constructing the lateral vibration data set in step S1 is: Classify the collected lateral vibration frequencies according to insulator type and condition; Obtain the autopower spectrum corresponding to each classified lateral vibration frequency, and calculate the kurtosis index K and mean square frequency index MSF of the autopower spectrum; The autopower spectrum, kurtosis index K and mean square frequency index MSF corresponding to various types of insulators in various states are classified and stored to form a lateral vibration data set.

3. The multi-layer perceptron-based multi-element damage detection method for electric tower insulators according to claim 1 is characterized in that: In step S1, the lateral vibration frequency is obtained by directly irradiating the target insulator with a scanning laser Doppler vibrometer, and the irradiation acquisition time is greater than or equal to 8 seconds.

4. The multi-layer perceptron-based multi-element damage detection method for electric tower insulators according to claim 1 is characterized in that: The type of insulator in step S1 includes but is not limited to: at least one of a porcelain / glass disk type suspension insulator, a composite insulator, a porcelain long rod insulator, and a porcelain / glass composite disk type suspension insulator.

5. The multi-layer perceptron-based multi-element damage detection method for electric tower insulators according to claim 1 is characterized in that: The states of the composite insulator are divided into: normal state and broken string state (damage abnormal state), wherein the causes of the broken string state of the composite insulator include but are not limited to: external force tearing, brittle fracture (stress corrosion) and / or rotten fracture (core rod carbonization); The states of porcelain disc type suspension insulators, porcelain long rod insulators and porcelain composite disc type suspension insulators are divided into: normal state and broken string / zero value state (damage abnormal state). The causes of broken string / zero value state include but are not limited to: cracking due to poor glue assembly, damage to the insulator head due to lightning strike and / or cracking of the steel foot due to external force; The states of glass disc type suspension insulators and glass composite disc type suspension insulators are divided into: normal state and broken string state (damage abnormal state), among which the broken string state includes but is not limited to: radial self-explosion caused by quality problems, fish scale-like self-explosion caused by external force and / or anode corrosion of insulator steel cap caused by electrolysis of rain, snow and water.

6. The method for detecting multi-dimensional damage of an electric tower insulator based on a multi-layer perceptron according to claim 2, characterized in that: The autopower spectrum is calculated using discrete signals, specifically: The lateral vibration frequency data x(n) with a length of N, n = 0, 1, ..., N-1 is divided into L segments, each segment has M data, where the expression of the k-th segment data is: x k (n)=x(n+kM-M),0≤n≤M,1≤k≤L; Then the auto-power spectrum S xx (e jω ) is: Wherein, w(n) is the window function, j is an imaginary number, and ω=2π / L. The calculation expression of the kurtosis index K is: The calculation expression of the mean square frequency index MSF is: Among them, x i is the discrete value of the collected lateral vibration frequency, is the mean value of the collected lateral vibration frequency, N is the length of the collected lateral vibration frequency data, fi is the frequency position corresponding to the auto-power spectrum, S(f i ) is the frequency position f i The corresponding autopower spectrum value.

7. The multi-layer perceptron-based multi-element damage detection method for electric tower insulators according to claim 2 is characterized in that: The lateral vibration data set has a hierarchical structure, the first layer is the insulator type, and the second layer is the kurtosis index K and the mean square frequency index MSF calculated for each type of insulator under the corresponding state.

8. The multi-layer perceptron-based multi-element damage detection method for electric tower insulators according to claim 2 is characterized in that: The desired goal in step S2 is specifically that the accuracy, precision, recall and F1 value of the multilayer perceptron meet the performance levels of 0.95, 0.95, 0.95 and 0.95 respectively.

9. The multi-layer perceptron-based multi-element damage detection method for electric tower insulators according to claim 8, characterized in that: In step S3, if the insulator type predicted by the multilayer perceptron is wrong, re-testing is required; if the multilayer perceptron still cannot make a correct prediction after multiple tests, the error case records need to be included in the lateral vibration data set as supplementary training data to compensate for the poor generalization of the model caused by incomplete early data collection.

10. A multi-layer sensor-based multi-element damage detection device for electric tower insulators, characterized in that: Computer instructions are stored, and when the computer instructions are executed by a processor, a multivariate damage detection method as described in any one of claims 1-9 is executed.

Images