GIS partial discharge fault discrimination method based on feature similarity

Abstract

A GIS partial discharge fault discrimination method based on feature similarity, using VMD algorithm for signal preprocessing, has stronger adaptive ability than wavelet analysis, without the need for artificial determination of wavelet basis function, reducing the influence on signal characteristics. At the same time, as an improved EMD algorithm, VMD algorithm has stronger anti-aliasing ability, fully considers the narrowband nature, and the signal-to-noise ratio of the decomposed components is higher, and each analysis signal has physical meaning. At the same time, the energy information after VMD decomposition and the original signal power spectrum density information are used to judge whether there is a fault, and after the fault is determined, diagnosis is carried out. Compared with directly identifying faults, the calculation can be simplified and the system efficiency can be improved. The traditional discharge frequency, mean, amplitude, phase and other signal characteristics are combined with power spectrum density and analysis signal energy to construct feature quantities, which can more accurately identify the discharge fault type, and according to the system identification process, the partial discharge location information can be estimated.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of online monitoring fault identification of electrical equipment, and particularly relates to a GIS partial discharge fault identification method based on feature similarity. BACKGROUND

[0002] GIS has been widely used in domestic and foreign power systems due to its compact structure, superior performance, strong anti-interference and other advantages. Although the stability of GIS is very high, the failure rate is much lower than that of traditional electrical equipment, but with the increase of application quantity and running time, various fault risks are also faced, and the main factor affecting the normal operation of GIS is insulation defect. When there is an insulation defect inside the GIS, partial discharge may be caused, and continuous partial discharge will further reduce the insulation performance of the GIS, eventually leading to insulation breakdown, and serious accidents will be caused, so monitoring the partial discharge of the GIS can ensure its safe and stable operation.

[0003] The traditional GIS partial discharge fault identification first needs to perform noise reduction processing on the original signal, and then performs fault identification on the original signal collected by the sensor according to the feature quantity through a deep learning algorithm. However, since the partial discharge signal is a non-stationary signal, it is usually necessary to add a window for short-time Fourier transform for noise reduction in the frequency domain, such as wavelet denoising method, but due to the unknown nature of the discharge frequency and discharge amount, the frequency characteristics of abnormal signals and noise signals overlap, and the effective signal features cannot be completely preserved. At the same time, using real-time fault identification algorithm for the original signal will inevitably lead to a large amount of calculation, and the system requirement is high, which cannot be embedded and integrated, so it is of great significance to pre-process the original signal to simplify the calculation while ensuring the accuracy of fault diagnosis and identification. SUMMARY

[0004] In order to overcome the above technical problems, the present application provides a GIS partial discharge fault identification method which can ensure the accuracy of fault diagnosis and identification.

[0005] The technical scheme adopted by the present application to overcome the technical problems is:

[0006] A GIS partial discharge fault identification method based on feature similarity, comprising the following steps:

[0007] a) arranging N partial discharge ultra-high frequency sensors in a GIS partial discharge online monitoring system;

[0008] b) obtaining initial data of the i-th partial discharge ultra-high frequency sensor when the GIS partial discharge online monitoring system is operated for the first time i∈{1,2,...,N}, preprocessing the initial data to obtain several decomposed intrinsic mode functions

[0009] c) obtaining the real-time collected raw data A of the i-th partial discharge ultrahigh frequency sensor i (t), preprocessing the raw data A i (t) to obtain a plurality of decomposed intrinsic mode functions IMF a i,n (t);

[0010] d) calculating the energy E of the intrinsic mode function IMF a ; calculating the energy E of the intrinsic mode function IMF a i,n (t); i,n

[0011] e) calculating the power spectral density P of the initial data A ; calculating the power spectral density P of the raw data A i (t); i,f

[0012] f) calculating the energy feature similarity p and the power spectral density similarity p i,E ; i,P

[0013] g) when p i,E ≥ 0 and p i,P ≥ 0, the current collected data of the i-th partial discharge ultrahigh frequency sensor is normal data, when p i,E < 0 or p i,P < 0, the current collected data of the i-th partial discharge ultrahigh frequency sensor is abnormal data;

[0014] h) calculating the average partial discharge signal pulse intensity of the partial discharge ultrahigh frequency sensor corresponding to the abnormal data arranging the average partial discharge signal pulse intensity of each partial discharge ultrahigh frequency sensor from large to small, and the interval where the partial discharge ultrahigh frequency sensor is located is the partial discharge occurrence position.

[0015] Further, in step a), each partial discharge ultrahigh frequency sensor is connected to a signal processing unit in a GIS partial discharge online monitoring system through a coaxial cable.

[0016] Further, in step b), the preprocessing method of the raw data A is as follows: the raw data A of the i-th partial discharge ultrahigh frequency sensor is decomposed by using a variational mode decomposition algorithm VMD, wherein K is the number of decomposition layers of the variational mode decomposition algorithm VMD.

[0017] ​​​​​Further, the step c) is to pre-process the original data A i (t) by using a VMD algorithm to decompose the original data A i (t) of the i-th partial discharge ultrahigh frequency sensor. Wherein, N E is the number of VMD decomposition layers.

[0018] Further, the step d) includes the following steps:

[0019] d-1) calculate the energy E by the formula , wherein T * is the data length of the obtained initial data A , to obtain the energy information E of each layer. d-2) calculate the energy E i,n by the formula i , wherein T is the data length of the obtained original data A (t), to obtain the energy information E of each layer.

[0020] Further, the step e) includes the following steps: e-1) perform Fourier transform on the initial data A to obtain the frequency domain information F , and calculate the power spectral density P 2 by the formula , wherein the unit of P i is rad i / Hz.

[0021] e-2) perform Fourier transform on the original data A i,f (t) to obtain the frequency domain information F i,f (ω), and calculate the power spectral density P 2 by the formula

[0022] f-1) calculate the energy feature similarity ρ

[0023]

[0024] by the formula i,E ;

[0025] f-2) calculate the power spectral density similarity ρ i,P by the formula

[0026]

[0027] .

[0028] Preferably, O in step g) is 0.7.

[0029] Further, in step h), the pulse intensity average of the partial discharge signal of the ith partial discharge ultrahigh frequency sensor is calculated by the formula Ai N is the number of pulses in the data collected by the ith partial discharge ultrahigh frequency sensor, and A i,k is the amplitude of the kth pulse.

[0030] The beneficial effects of the present application are: without artificially determining the wavelet base function, the influence on the signal characteristics is reduced, at the same time, the VMD algorithm is an improved EMD algorithm, and the anti-aliasing ability is stronger, the narrowband nature is fully considered, the component signal-to-noise ratio obtained by decomposition is higher, the fault is determined, and then diagnosis is performed, compared with directly performing fault recognition, the calculation can be simplified, and the system efficiency can be improved. The traditional discharge frequency, average, amplitude, phase and other signal characteristics are combined with the power spectrum density and the analytical signal energy to construct the characteristic quantity, the discharge fault type can be more accurately recognized, and at the same time, according to the system recognition process, the local discharge position information can be preliminarily estimated. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1 is a flow chart of the fault diagnosis and recognition method of the present application;

[0032] Figure 2 is a flow chart of the signal preprocessing process of the present application;

[0033] Figure 3 is a flow chart of the abnormal information discrimination and fault diagnosis and recognition process of the present application. DETAILED DESCRIPTION

[0034] The following will be combined with the accompanying Figure 1 to the accompanying Figure 3 Further illustrate the present application.

[0035] A GIS partial discharge fault discrimination method based on feature similarity, comprising the following steps:

[0036] a) arranging N partial discharge ultrahigh frequency sensors in a GIS partial discharge online monitoring system.

[0037] b) when the GIS partial discharge online monitoring system is operated for the first time, obtaining initial data of the ith partial discharge ultrahigh frequency sensor i∈{1,2,...,N}, the initial data is pretreated to obtain several intrinsic mode functions after decomposition.

[0038] ​c) obtaining the real-time collected raw data A of the i-th partial discharge ultrahigh frequency sensor i (t) to obtain several decomposed intrinsic mode functions IMFa i (t). i,n (t).

[0039] d) calculating the energy of the intrinsic mode function IMFa (t). (t). i,n i,n .

[0040] e) calculating the power spectrum density of the initial data (t). (t). i i,f .

[0041] f) calculating the energy feature similarity ρ i,E and the power spectrum density similarity ρ i,P .

[0042] g) when ρ i,E ≥ 0 and ρ i,P ≥ 0, the current collected data of the i-th partial discharge ultrahigh frequency sensor is normal data, when ρ i,E < 0 or ρ i,P < 0, the current collected data of the i-th partial discharge ultrahigh frequency sensor is abnormal data.

[0043] h) calculating the average partial discharge signal pulse intensity of the partial discharge ultrahigh frequency sensor corresponding to the abnormal data arranging the average partial discharge signal pulse intensity of each partial discharge ultrahigh frequency sensor from large to small, and the interval of the partial discharge ultrahigh frequency sensor is the partial discharge occurrence position.

[0044] In an embodiment of the present application, each partial discharge ultrahigh frequency sensor in step a) is connected to a signal processing unit in the GIS partial discharge online monitoring system through a coaxial cable.

[0045] ​​​The VMD algorithm is used for signal pretreatment, has stronger self-adaptive ability than wavelet analysis, does not need to determine the wavelet base function artificially, reduces the influence on the signal characteristics, meanwhile, as an improved EMD algorithm, the VMD algorithm has stronger anti-aliasing ability, fully considers the narrowband property, and the signal-to-noise ratio of the decomposed components is higher, and meanwhile, each analysis signal has physical meaning. Meanwhile, the energy information after VMD decomposition and the original signal power spectrum density information are used to judge whether a fault exists, and after the fault is determined, diagnosis is carried out, so that the calculation can be simplified, and the system efficiency can be improved, compared with directly carrying out fault identification. The traditional discharge frequency, mean value, amplitude, phase and other signal characteristics are combined with the power spectrum density and the analysis signal energy to construct the characteristic quantity, so that the discharge fault type can be more accurately identified, and meanwhile, according to the identification process of the system, the local discharge position information can be preliminarily estimated.

[0046] In an embodiment of the present application, the method for pretreating the original data A in step b) is that the original data A of the i th local discharge ultrahigh frequency sensor is decomposed by using a VMD (Variational Mode Decomposition) algorithm, wherein N is the number of decomposition layers of the VMD (Variational Mode Decomposition) algorithm.

[0047] In an embodiment of the present application, the method for pretreating the original data A i (t) in step c) is that the original data A i (t) of the i th local discharge ultrahigh frequency sensor is decomposed by using a VMD (Variational Mode Decomposition) algorithm, wherein N E is the number of decomposition layers of the VMD (Variational Mode Decomposition) algorithm.

[0048] In an embodiment of the present application, step d) comprises the following steps:

[0049] d-1) the energy E is calculated by using the formula wherein T * is the data length of the obtained initial data A , and the energy information E of each layer is obtained.

[0050] d-2) the energy E is calculated by using the formula i,n wherein T is the data length of the obtained original data A i (t), and the energy information E of each layer is obtained. In an embodiment of the present application, step e) comprises the following steps:

[0051] e-1) the initial data A Fourier transform is performed to obtain frequency domain information The power spectral density P is calculated by the formula The unit of P 2 is rad / Hz.

[0052] e-2) Fourier transform is performed on the original data A i (t) to obtain frequency domain information F i (ω), and the power spectral density P i,f is calculated by the formula The unit of P i,f is rad 2 / Hz. In an embodiment of the present application, the Pearson algorithm is used to calculate the similarity of the signal collected by the i-th sensor and the data characteristic of the normal signal without partial discharge. The normal signal without partial discharge is based on the signal collected by the initial sensor when the GIS is started. The specific steps f) include the following steps:

[0053] f-1) The similarity of the energy characteristics ρ i,E is calculated by the formula

[0054]

[0055]

[0056] f-2) The similarity of the power spectral density ρ i,P is calculated by the formula

[0057]

[0058]

[0059] In an embodiment of the present application, preferably, the value of O in step g) is 0.7.

[0060] In an embodiment of the present application, the pulse intensity average of the partial discharge signal of the i-th partial discharge ultrahigh frequency sensor is calculated by the formula in step h), where N Ai is the number of pulses in the data collected by the i-th partial discharge ultrahigh frequency sensor, and A i,k is the amplitude of the k-th pulse.

[0061] Finally, it should be noted that the above description is only for the preferred embodiments of the present application and is not intended to limit the present application. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent replacements to some technical features. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.​

Claims

1. A method for identifying partial discharge faults in GIS based on feature similarity, characterized in that, Includes the following steps: a) Deploy N ultra-high frequency partial discharge sensors in the GIS partial discharge online monitoring system; b) During the first run of the GIS partial discharge online monitoring system, the initial data of the i-th partial discharge UHF sensor is acquired. For initial data Preprocessing is performed to obtain several decomposed intrinsic mode functions. c) Obtain the raw data A collected in real time by the i-th partial discharge UHF sensor. i (t), for the original data A i (t) is preprocessed to obtain several decomposed intrinsic mode functions IMFa. i,n (t); d) Calculate the intrinsic mode functions energy Calculate the intrinsic mode function (IMF) a i,n The energy E of (t) i,n ; e) Calculate initial data power spectral density Calculate the original data A i The power spectral density P of (t) i,f ; f) Calculate the energy feature similarity ρ i,E and power spectral density similarity ρ i,P ; g) When ρ i,E ≥O and ρ i,P When ≥ 0, the data currently collected by the i-th partial discharge UHF sensor is normal data. i,E <O or ρ i,P When < 0, the data currently collected by the i-th partial discharge UHF sensor is abnormal data; h) Calculate the average pulse intensity of the partial discharge signal from the UHF sensor corresponding to the abnormal data. The average pulse intensity of the partial discharge signal from each UHF partial discharge sensor is calculated. Arranged from largest to smallest, the interval where the partial discharge UHF sensor is located indicates the location where the partial discharge occurs.

2. The GIS partial discharge fault discrimination method based on feature similarity according to claim 1, characterized in that: In step a), each of the UHF partial discharge sensors is connected to the signal processing unit of the GIS partial discharge online monitoring system via a coaxial cable.

3. The GIS partial discharge fault discrimination method based on feature similarity according to claim 1, characterized in that, In step b), the raw data The preprocessing method is as follows: the variational mode decomposition algorithm (VMD) is used to process the raw data of the i-th partial discharge UHF sensor. Decompose it. in, This represents the number of decomposition layers in the Variational Mode Decomposition (VMD) algorithm.

4. The GIS partial discharge fault discrimination method based on feature similarity according to claim 3, characterized in that, In step c), the original data A i (t) The preprocessing method is as follows: the variational mode decomposition algorithm (VMD) is used to process the raw data A of the i-th partial discharge UHF sensor. i (t) is decomposed. Where, N E This represents the number of decomposition layers in the Variational Mode Decomposition (VMD) algorithm.

5. The GIS partial discharge fault discrimination method based on feature similarity according to claim 4, characterized in that, Step d) includes the following steps: d-1) Through the formula Calculated energy In the formula T * Initial data to be acquired The data length is used to obtain energy information for each layer. d-2) By formula The energy E was calculated. i,n In the formula, T represents the original data A obtained. i The data length of (t) is used to obtain the energy information of each layer.

6. The GIS partial discharge fault discrimination method based on feature similarity according to claim 5, characterized in that, Step e) includes the following steps: e-1) Initial data The frequency domain information F is obtained by performing a Fourier transform. i * (ω), through the formula The power spectral density was calculated. The unit is rad. 2 / Hz; e-2) Transfer the original data A i (t) is subjected to Fourier transform to obtain frequency domain information F. i (ω), through the formula The power spectral density P was calculated. i,f P i,f The unit is rad. 2 / Hz.

7. The GIS partial discharge fault discrimination method based on feature similarity according to claim 6, characterized in that, Step f) includes the following steps: f-1) through formula The energy feature similarity ρ was calculated. i,E ; f-2) via formula The power spectral density similarity ρ was calculated. i,P .

8. The GIS partial discharge fault discrimination method based on feature similarity according to claim 1, characterized in that: In step g), O takes the value of 0.

7.

9. The GIS partial discharge fault discrimination method based on feature similarity according to claim 1, characterized in that: In step h), the formula is used. The average pulse intensity of the partial discharge signal from the i-th ultra-high frequency partial discharge sensor is calculated, N. Ai Let A be the number of pulses in the data collected by the i-th partial discharge ultra-high frequency sensor. i,k Let be the amplitude of the k-th pulse.

Images