GIS latent fault diagnosis method based on vibration signals

Abstract

The invention discloses a GIS latent fault diagnosis method based on vibration signals, and relates to the technical field of GIS equipment fault diagnosis. Compared with a previous fault diagnosis method, the problems that signals are easily interfered by subjective factors, fault features are difficult to extract and distinguish, and misjudgment is easy to generate are solved, corresponding digital twin models are built for different GIS devices, prediction and analysis of the GIS devices in a virtual environment are achieved, a maintenance plan can be optimized, the utilization rate and reliability of the devices can be improved, and the method is suitable for popularization and application. The maintenance cost is reduced, the operation risk caused by direct contact of personnel with equipment is avoided, and the safety is improved; various data information is integrated through a multi-dimensional feature extraction method, comprehensive understanding and accurate judgment of the equipment state are improved, state data of real equipment are synchronized in real time according to the established optimized diagnosis and prediction model, real-time diagnosis and prediction of the equipment state are achieved, efficiency is improved, and the method is suitable for popularization and application. Potential faults can be found in time, and measures are taken to avoid loss.

Description

Technical Field

[0001] The invention relates to the technical field of GIS equipment fault diagnosis, and in particular to a GIS latent fault diagnosis method based on vibration signals. Background Art

[0002] With the construction of ultra-high voltage and extra-high voltage power grids, the demand for power system capacity and energy continues to increase, and higher and higher requirements are also put forward for the reliability and economy of power systems. Gas-insulated metal-enclosed switchgear (GIS) has been widely used in the construction of ultra-high voltage and extra-high voltage power grids due to its advantages such as small footprint, reliable operation and convenient maintenance. As an important component of the power system, GIS plays an important role in improving the reliability and economy of the power system. However, due to its all-metal enclosed structure, the internal mechanical structure of GIS equipment is relatively complex and large. Once a fault occurs, the impact range is large and it is difficult to accurately locate and quickly repair it, resulting in serious economic losses. Therefore, timely and effective discovery of latent defects in GIS equipment and accurate judgment of the types of defects in GIS equipment are of great significance to ensure the safe and stable operation of GIS equipment.

[0003] In the existing process of diagnosing mechanical faults of GIS equipment based on vibration characteristics, the vibration signals of GIS equipment collected by the acquisition equipment are easily affected by multiple factors such as load changes, environmental interference, and equipment defects, resulting in signal interference. It is impossible to take into account the different types of faults of GIS equipment and they may appear similar in the vibration signals, which makes it difficult to extract and distinguish fault features and easily leads to misjudgment, resulting in low final prediction efficiency.

[0004] In order to solve the above problems, the present invention proposes a GIS latent fault diagnosis method based on vibration signals. Summary of the invention

[0005] The purpose of the present invention is to propose a GIS latent fault diagnosis method based on vibration signals to solve the problems raised in the background technology:

[0006] In the existing process of diagnosing mechanical faults of GIS equipment based on vibration characteristics, the vibration signals of GIS equipment collected by the acquisition equipment are easily affected by various factors, which makes the signals susceptible to interference from subjective factors. Different types of faults may appear similar in vibration signals, making it difficult to extract and distinguish fault features, which is prone to misjudgment and leads to low final prediction efficiency.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] The GIS latent fault diagnosis method based on vibration signals is as follows:

[0009] S1: Build a digital twin model of GIS equipment;

[0010] S2: Vibration sensors are used to monitor and collect vibration signals generated by GIS equipment during operation in real time, and denoise the signals based on wavelet packet transform optimized by CEEMDAN fusion. The denoised vibration signals are then input into the digital twin model.

[0011] S3: Compare and analyze the collected vibration signal data with the simulation data in the digital twin model to verify and optimize the digital twin model;

[0012] S4: Fuse the simulation data in the digital twin model with the measured data, extract the features of the vibration signal based on the optimized wavelet transform, and diagnose and predict the latent faults of GIS equipment based on the optimized CNN-SVM;

[0013] S5: Provide maintenance decision support for GIS equipment based on diagnostic results and prediction information.

[0014] Preferably, in S1, a virtual model is built based on a finite element model combined with physical characteristics of GIS equipment and real-time data.

[0015] Preferably, in S3, the data set is further enhanced and expanded based on data enhancement technology.

[0016] Preferably, in S3, vibration signals of different modes are obtained based on CEEMDAN, and then the processed wavelet coefficients are decomposed and denoised by threshold through optimized wavelet packet transform, and then restored to the denoised signal through inverse wavelet transform; the optimized wavelet packet transform performs threshold optimization of the soft threshold function in the wavelet coefficients based on the optimized WOA algorithm,

[0017] The soft threshold function is as follows:

[0018]

[0019] Where, d is the wavelet transform coefficient; sgn(g) is the sign function; is the threshold value;

[0020] The optimized WOA algorithm is specifically as follows:

[0021] According to the common behavior of humpback whales surrounding prey, a mathematical model of target surrounding is established:

[0022] D=|CgX * (n)-X(n)|

[0023] X(n+1)=X * (n)-AD

[0024] Where D is the distance between the best individual and the average individual in the whale group; n is the number of iterations; X(n) represents the distance between the best individual and the average individual in the whale group at the nth iteration; X(n+1) represents the distance between the best individual and the average individual in the whale group at the n+1th iteration; X * (n) represents the position vector of the best individual of the whale group in the nth iteration; A and C are coefficient vectors used to control the swimming mode of the whale, and the expression is as follows:

[0025] A=2agr1-a

[0026] C=2ar2

[0027]

[0028] Among them, r1 and r2 are random numbers between (0,1); a is the adjustment factor; N is the maximum number of iterations;

[0029] The distance between individual whales and the position of prey is calculated based on the spiral swimming behavior of humpback whales, and an update model between the position of individual whales and prey is established based on this:

[0030]

[0031] B=|X * (n)-X(n)|

[0032] Where x(n+1) represents the distance between the individual whales and their prey at the n+1th iteration; b is the spiral coefficient; l r is a random number between (-1,1); B is the distance between the best individual position of the whale group and the prey;

[0033] In the local development stage, the whale assumes the current prey position as the target position, and continuously approaches the prey with a probability of 50% between the shrinking and encircling mechanism and the spiral update mechanism to obtain the optimal solution:

[0034]

[0035] Where x(n) represents the distance between the individual whale group and the target position at the nth iteration; b is the spiral coefficient; s is a random number on (0,1);

[0036] Build an updated model for prey hunting:

[0037] X'(n+1)=X r -AD0

[0038] D0=|CX r -X(n)|

[0039] Where X'(n+1) represents the distance between the random position vectors of the whale group and the individual whale at the n+1th iteration; D0 represents the distance between the random position vectors of the whale group and the individual whale; X r Represents the random position vector of an individual whale; when A is greater than 1, this vector is used to update the position information of other humpback whales;

[0040] The threshold of wavelet packet transform is used as the whale's position vector, the minimum value of vibration signal error is used as the fitness function, the whale's position vector is initialized within the value range, and the optimal position vector of the whale is obtained by optimization.

[0041] Preferably, in S4, multi-dimensional features of the vibration signal are extracted, and the fusion of the features of each dimension of the vibration signal is performed based on a normalized fusion screening method:

[0042]

[0043] Among them, a ij represents the j-th dimensionless parameter of the i-th node component; r ij is the j-th dimension parameter value of the i-th node component; r imax is the parameter with the largest absolute value of the parameter of the i-th node component;

[0044] g i =a ij

[0045] Among them, g i Represents the screening index of the i-th node; the dimensionless parameter information of each node component is fused using the screening index, and the screening index is sorted, and the first p node components are selected to extract the vibration signal features based on the wavelet transform improved by optimizing WOA.

[0046] Preferably, the multidimensional features include resonance frequency, harmonic frequency, characteristic frequency, skewness value, kurtosis value, permutation entropy, root mean square, crest factor, signal order, peak-to-peak value, nonlinear index, mean value and standard deviation.

[0047] Preferably, in S4, fault diagnosis and prediction are performed based on a support vector machine, and parameters in the CNN-SVM are tuned based on an optimized WOA algorithm.

[0048] Preferably, in said S4, the fault diagnosis and prediction process of the GIS equipment is also parallelized based on a distributed computing framework.

[0049] Compared with the prior art, the present invention provides a GIS latent fault diagnosis method based on vibration signals, which has the following beneficial effects:

[0050] The present invention builds corresponding digital twin models for different GIS equipment, which can realize the prediction and analysis of GIS equipment in a virtual environment, take into account the external interference factors of the vibration signal of GIS equipment, optimize the maintenance plan, improve equipment utilization and reliability, reduce maintenance costs, and avoid operational risks brought by direct contact with equipment, thereby improving work safety; and through multi-dimensional feature extraction methods, different data information can be integrated to improve the comprehensive understanding and accurate judgment of equipment status. The digital twin model also synchronizes the status data of real equipment in real time according to the established optimized diagnosis and prediction model, realizes real-time diagnosis and prediction of equipment status and improves efficiency, timely discovers potential faults and takes measures to avoid losses. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a flow chart of the method mentioned in Example 1 of the present invention;

[0052] Figure 2 It is a schematic diagram of the iteration of the optimized WOA algorithm and the traditional WOA algorithm mentioned in Embodiment 2 of the present invention;

[0053] Figure 3 This is a schematic diagram comparing the noise reduction results of different algorithms mentioned in Example 2 of the present invention. DETAILED DESCRIPTION

[0054] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0055] The present invention builds corresponding digital twin models for different GIS equipment, which can realize the prediction and analysis of GIS equipment in a virtual environment, take into account the external interference factors of GIS equipment vibration signals, optimize maintenance plans, improve equipment utilization and reliability, reduce maintenance costs, and avoid operational risks caused by direct contact with equipment, thereby improving work safety; and through multi-dimensional feature extraction methods, it can integrate different data information to improve the comprehensive understanding and accurate judgment of equipment status. The digital twin model also synchronizes the status data of real equipment in real time based on the established optimized diagnosis and prediction model, realizes real-time diagnosis and prediction of equipment status and improves efficiency, timely discovers potential faults and takes measures to avoid losses. Specifically including the following contents.

[0056] Embodiment 1:

[0057] See also Figure 1 The GIS latent fault diagnosis method based on vibration signals of the present invention comprises:

[0058] S1: Build a digital twin model of GIS equipment; the details are as follows:

[0059] First, real-time data of GIS equipment is collected, and relevant physical characteristic parameters of GIS equipment are collected. A digital twin model of GIS equipment is built based on the finite element model, which can reflect the actual situation of the equipment during operation. By integrating the digital twin model and vibration signal analysis technology, more accurate and timely diagnosis of latent faults of GIS equipment can be achieved, thereby improving the reliability and safety of equipment, while reducing maintenance costs and downtime.

[0060] S2: Vibration sensors are used to monitor and collect vibration signals generated by GIS equipment during operation in real time, and denoise the signals based on wavelet packet transform optimized by CEEMDAN fusion. The denoised vibration signals are then input into the digital twin model. The details are as follows:

[0061] First, vibration sensors are used to monitor and collect vibration signal data generated by GIS equipment during operation in real time to ensure the accuracy and completeness of the data.

[0062] Then, based on the CEEMDAN fusion optimized wavelet packet transform method, the vibration signal is denoised. The collected vibration signal is subjected to CEEMDAN denoising, which can effectively decompose the signal and remove noise, extract useful information of the signal, and then the optimized wavelet packet transform is applied for further signal denoising. The threshold of the wavelet packet is obtained based on the optimized WOA algorithm. The optimized WOA algorithm is based on the meta-heuristic optimization calculation that imitates the hunting action of humpback whales. It uses random or optimal search agents to imitate the actions of prey, and uses spirals to imitate the attack strategy of humpback whales, which can better capture prey. The optimized WOA has a simple and easy mechanism, can effectively reduce the amount of parameters, and has powerful optimization capabilities.

[0063] The vibration signal data after CEEMDAN denoising and wavelet packet transform is then used as the input of the digital twin model. Combined with the digital twin model, the vibration signal data can be effectively processed to provide accurate diagnosis and prediction of the operating status of GIS equipment, thereby improving the reliability and safety of the equipment.

[0064] S3: Compare and analyze the collected vibration signal data with the simulation data in the digital twin model to verify and optimize the digital twin model; the details are as follows:

[0065] By comparing and analyzing the actual collected vibration signal data with the simulation data in the digital twin model, the similarities or differences of the data can be compared. Based on the results of the data comparison, the accuracy and reliability of the digital twin model can be evaluated to verify whether the model can accurately simulate the actual situation. If a large deviation is found between the actual data and the model simulation data, the digital twin model can be optimized and adjusted based on the results of the comparative analysis to improve the model's predictive ability and applicability. By comparing and analyzing the collected vibration signal data with the simulation data in the digital twin model, the effectiveness of the digital twin model can be verified and guidance can be provided for further optimization of the model. This comprehensive experimental and simulation method helps to improve the application effect of the digital twin model in fault diagnosis and prediction.

[0066] S4: Fuse the simulation data in the digital twin model with the measured data, extract the features of the vibration signal based on the optimized wavelet transform, and diagnose and predict the latent faults of the GIS equipment based on the optimized support vector machine; the details are as follows:

[0067] First, the simulation data in the digital twin model is fused with the measured data to ensure the consistency and integrity of the data. Then, the fused vibration signal data is feature extracted based on the wavelet transform optimized by WOA, and the appropriate wavelet basis function and scale are selected to extract the useful information of the vibration signal. Then, the optimized support vector machine algorithm is used to build a fault diagnosis and prediction model, and the data after feature extraction and preprocessing is used as input to train the model to identify potential faults of GIS equipment. The support vector machine model is also evaluated, including the evaluation of indicators such as accuracy and recall rate. According to the evaluation results, the model parameters are optimized to improve the accuracy of diagnosis and prediction. The vibration signal data of GIS equipment is analyzed using the optimized support vector machine model to realize the diagnosis and prediction of potential faults, and to discover and solve possible problems in advance.

[0068] By integrating simulation data and measured data, applying optimized wavelet transform for feature extraction, and combining optimized support vector machine for fault diagnosis and prediction, the monitoring efficiency and accuracy of GIS equipment operation status can be improved, helping to improve the reliability and safety of equipment. This method of comprehensive use of data fusion and advanced algorithms helps to improve the level and application effect of fault diagnosis technology.

[0069] S5: Provide maintenance decision support for GIS equipment based on diagnostic results and prediction information.

[0070] The details are as follows:

[0071] Based on the results of fault diagnosis and prediction information, evaluate each component or system of GIS equipment to determine which components have potential failure risks, so as to set maintenance priorities. Combined with the prediction information and equipment maintenance records, formulate a maintenance plan for GIS equipment, including regular inspections, preventive maintenance and necessary repairs to ensure the normal operation of the equipment. Based on the diagnosis results, take necessary measures in a timely manner to deal with the faults that have occurred to avoid the expansion of the faults and more serious consequences. Based on the prediction information and maintenance plan, rationally manage the spare parts and consumables required for GIS equipment to ensure the timely supply and use of spare parts. Based on the diagnosis results and fault predictions, provide necessary training and guidance to maintenance personnel to improve their ability to identify and handle equipment failures. Based on the prediction information and equipment maintenance history, evaluate the update and upgrade needs of GIS equipment, and formulate plans and programs for equipment updates.

[0072] Through the above methods, combined with diagnostic results and predictive information, it can help equipment managers formulate scientific and reasonable GIS equipment maintenance strategies, extend the service life of equipment, improve equipment reliability and safety, reduce maintenance costs, and ensure the normal operation of equipment.

[0073] Embodiment 2:

[0074] For common GIS faults, such as insulation faults (insulation breakdown, insulation aging, insulation damage), grounding faults, mechanical faults (mechanism faults, loose mechanical connection parts, mechanism jamming), electrical faults (such as poor contact of contacts, circuit breaker failure, damage to electrical components), protection system faults, etc., the above faults are coded. For details, please refer to Table 1:

[0075] Table 1 Partial fault codes

[0076]

[0077] The population size of the optimized WOA algorithm and the traditional WOA algorithm is set to 15, and the maximum number of iterations is 11. The iterative optimization can be referred to Figure 2 ,As can be seen from the figure, the optimized WOA algorithm has fewer ,iterations and faster speed compared with the traditional WOA algorithm.

[0078] The original vibration signal, the signal denoised by wavelet packet transform fused with CEEMDAN optimization, the signal denoised by CEEMDAN, and the signal denoised by wavelet packet transform fused with CEEMDAN are tested and compared. The specific denoising results can be referred to Figure 3As can be seen from the figure, the three denoising methods have good denoising effects on vibration signals, but the CEEMDAN denoising processing and the CEEMDAN fused wavelet packet transform denoising processing still contain a lot of noise at the peaks and troughs of the vibration signal waveform, and the noise interference in the tail waveform is more obvious; but the signal after CEEMDAN fused optimized wavelet packet transform denoising not only solves the problem of a small amount of noise in the peaks and troughs, but also removes a large amount of noise at the tail of the waveform, greatly improving the quality of the vibration signal.

[0079] The digital twin model is also based on CNN-SVM as experimental group 1, WOA optimized CNN-SVM as experimental group 2, optimized CNN-SVM and CNN-SVM as experimental group 3, and 28182 training sets and 9393 test sets containing the above GIS fault types for latent fault diagnosis and prediction. The specific prediction results can be found in Table 2:

[0080] Table 2 Prediction results of different prediction models

[0081] Model Number of features Classification accuracy of training set Cross-validation classification accuracy Test set classification accuracy Average prediction time Experimental Group 1 36 93.68％ 91.10％ 88.26％ 12.82 Experimental Group 2 27 94.32％ 92.08％ 89.91％ 16.72 Experimental Group 3 29 95.29％ 93.15％ 90.96％ 4.85

[0082] As can be seen from the table, the classification prediction accuracy of experimental group three is significantly higher than that of experimental group one and experimental group two. Experimental group three also applied a distributed computing framework for parallel processing, which greatly improved the prediction accuracy while improving the prediction accuracy, and had a better recognition effect.

[0083] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A GIS latent fault diagnosis method based on vibration signals, characterized in that: The steps include: S1: Build a digital twin model of GIS equipment; S2: Vibration sensors are used to monitor and collect vibration signals generated by GIS equipment during operation in real time, and denoise the signals based on wavelet packet transform optimized by CEEMDAN fusion. The denoised vibration signals are then input into the digital twin model. S3: Compare and analyze the collected vibration signal data with the simulation data in the digital twin model to verify and optimize the digital twin model; S4: Fuse the simulation data in the digital twin model with the measured data, extract the features of the vibration signal based on the optimized wavelet transform, and diagnose and predict the latent faults of GIS equipment based on the optimized CNN-SVM; S5: Provide maintenance decision support for GIS equipment based on diagnostic results and prediction information.

2. The GIS latent fault diagnosis method based on vibration signals according to claim 1 is characterized in that: In S1, a virtual model is built based on the finite element model combined with the physical characteristics of the GIS equipment and real-time data.

3. The GIS latent fault diagnosis method based on vibration signals according to claim 1 is characterized in that: In S3, the data set is further enhanced and expanded based on data enhancement technology.

4. The GIS latent fault diagnosis method based on vibration signals according to claim 3 is characterized in that: In S3, vibration signals of different modes are obtained based on CEEMDAN, and then the processed wavelet coefficients are decomposed and denoised by threshold through optimized wavelet packet transform, and then restored to the denoised signal through inverse wavelet transform; the optimized wavelet packet transform performs threshold optimization of the soft threshold function in the wavelet coefficients based on the optimized WOA algorithm. The soft threshold function is as follows: Where, d is the wavelet transform coefficient; sgn(g) is the sign function; is the threshold value; The optimized WOA algorithm is specifically as follows: According to the common behavior of humpback whales surrounding prey, a mathematical model of target surrounding is established: D=|CgX * (n)-X(n)| X(n+1)=X * (n)-AD Where D is the distance between the best individual and the average individual in the whale group; n is the number of iterations; X(n) represents the distance between the best individual and the average individual in the whale group at the nth iteration; X(n+1) represents the distance between the best individual and the average individual in the whale group at the n+1th iteration; X * (n) represents the position vector of the best individual of the whale group in the nth iteration; A and C are coefficient vectors used to control the swimming mode of the whale, and the expression is as follows: A=2agr1-a C=2ar2 Among them, r1 and r2 are random numbers between (0,1); a is the adjustment factor; N is the maximum number of iterations; The distance between individual whales and the position of prey is calculated based on the spiral swimming behavior of humpback whales, and an update model between the position of individual whales and prey is established based on this: B=|X * (n)-X(n) Where x(n+1) represents the distance between the individual whales and their prey at the n+1th iteration; b is the spiral coefficient; l r is a random number between (-1,1); B is the distance between the best individual position of the whale group and the prey; In the local development stage, the whale assumes the current prey position as the target position, and continuously approaches the prey with a probability of 50% between the shrinking and encircling mechanism and the spiral update mechanism to obtain the optimal solution: Where x(n) represents the distance between the individual whale group and the target position at the nth iteration; b is the spiral coefficient; s is a random number on (0,1); Build an updated model for prey hunting: X'(n+1)=X r -AD0 D0=|CX r -X(n)| Where X'(n+1) represents the distance between the random position vectors of the whale group and the individual whale at the n+1th iteration; D0 represents the distance between the random position vectors of the whale group and the individual whale; X r Represents the random position vector of an individual whale; when A is greater than 1, this vector is used to update the position information of other humpback whales; The threshold of wavelet packet transform is used as the whale's position vector, the minimum value of vibration signal error is used as the fitness function, the whale's position vector is initialized within the value range, and the optimal position vector of the whale is obtained by optimization.

5. The GIS latent fault diagnosis method based on vibration signals according to claim 1 is characterized in that: In S4, the multi-dimensional features of the vibration signal are extracted, and the dimensional features of the vibration signal are fused based on the normalized fusion screening method: Among them, a ij represents the j-th dimensionless parameter of the i-th node component; r ij is the j-th dimension parameter value of the i-th node component; r imax is the parameter with the largest absolute value of the parameter of the i-th node component; g i =a ij Among them, g i Represents the screening index of the i-th node; the dimensionless parameter information of each node component is fused using the screening index, and the screening index is sorted, and the first p node components are selected to extract the vibration signal features based on the wavelet transform improved by optimizing WOA.

6. The GIS latent fault diagnosis method based on vibration signals according to claim 5 is characterized in that: The multi-dimensional features include resonance frequency, harmonic frequency, characteristic frequency, skewness value, kurtosis value, permutation entropy, root mean square, crest factor, signal order, peak-to-peak value, nonlinear index, mean value and standard deviation.

7. The GIS latent fault diagnosis method based on vibration signals according to claim 6 is characterized in that: In S4, fault diagnosis and prediction are performed based on a support vector machine, and parameters in the CNN-SVM are tuned based on an optimized WOA algorithm.

8. The GIS latent fault diagnosis method based on vibration signals according to claim 7 is characterized in that: In S4, the fault diagnosis and prediction process of the GIS equipment is also parallelized based on the distributed computing framework.

Images