A feature-assisted partial discharge point location method

Abstract

The application discloses a feature-assisted partial discharge point positioning method, and specifically comprises the following steps: (1) collecting and preliminarily processing partial discharge signals, collecting the partial discharge signals of power equipment through a partial discharge detection device, and obtaining the position measurement information of the partial discharge point, partial discharge pulse signals and spectrum information and the like after processing; (2) modeling according to the time domain and frequency domain features, using the amplitude as the time domain feature and using the frequency spectrum as the frequency domain feature; (3) fusing the time domain and frequency domain features with the Gaussian mixture probability hypothesis density filtering, and correcting the weight of the Gaussian component of the partial discharge point with the features. The application is based on the Gaussian mixture probability hypothesis density filtering, fuses the time-frequency domain features, thereby can suppress the clutter interference, and realizes the positioning of the partial discharge point in the strong clutter environment. The application has the advantages of fast response speed, strong anti-interference capability, high positioning accuracy and the like.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of partial discharge positioning of electrical equipment, and particularly relates to a feature-assisted partial discharge positioning method. BACKGROUND

[0002] The detection of partial discharge signals can be an effective means for evaluating the insulation state of equipment, and is also an important prerequisite for realizing partial discharge point positioning and power equipment fault positioning. In the working environment of power equipment, various noises exist, and it is crucial to extract partial discharge signals and distinguish partial discharge signals from noise signals in the complex noise strong clutter environment.

[0003] The traditional partial discharge positioning problem of power equipment is mainly based on the difference in distance between each sensor and the partial discharge point, and the time difference in receiving signals is not large, and the time difference equation set is established and solved to position according to the time difference of the signals received by the sensors from the partial discharge point and the sensor coordinates. However, due to the large noise in the environment and the unknown and variable generation of the partial discharge point, the existing partial discharge positioning method has some limitations.

[0004] In view of the problem that there is a large noise in the running environment of power equipment, which affects the positioning of partial discharge, the traditional method of removing noise from partial discharge signals mainly uses wavelet analysis and empirical mode decomposition, but the loss of original partial discharge signal characteristics will occur while removing noise, and the limitations of the algorithm itself plus the complex characteristics of partial discharge signals and noise, the overall noise reduction effect is general. The present application considers solving the problem of strong clutter interference on partial discharge signals at the positioning level. SUMMARY

[0005] The present application is a feature-assisted partial discharge positioning method designed to solve the problem that strong clutter noise in the partial discharge point detection and positioning environment makes it impossible to obtain accurate positioning, introduces time domain and frequency domain features, uses state parameters and feature parameters to estimate the position of the partial discharge point, realizes clutter suppression in the partial discharge positioning process, and improves the anti-interference performance and positioning accuracy of the partial discharge point positioning method. The present application has the advantages of fast response speed, high positioning accuracy, high environmental adaptability, etc.

[0006] The principle of the present application is as follows: the time domain and frequency domain features are introduced, and they are combined with the Gaussian Mixture Probability Hypothesis Density (GM-PHD) algorithm, the weight of the Gaussian component is corrected by the features, the weight of the Gaussian component at the position of the partial discharge point is increased, and the weight of the Gaussian component at the position of the noise is reduced, the clutter suppression is completed, and the partial discharge point positioning in the strong noise clutter environment is realized.

[0007] A feature-assisted partial discharge positioning method, the specific steps are as follows:

[0008] S1, collecting and preliminarily processing partial discharge signals, collecting partial discharge signals of power equipment through a partial discharge detection device, and obtaining position measurement information of partial discharge points, partial discharge pulse signals and spectrum information and the like after processing.

[0009] S2, time domain analysis is performed on the collected partial discharge signals: taking the partial discharge signals as time domain signals, using amplitude as time domain feature modeling, considering the characteristics of the partial discharge points, the higher the amplitude, the higher the possibility of existence of the partial discharge points in the unit. The specific modeling is as follows:

[0010]

[0011]

[0012] g=(s(k,i)-s c(k) ) / s c(k) (3)

[0013] α=(1+exp(-2g)) -1 (4)

[0014] Wherein, N represents the total pulse of the i-th unit, s(k,i) represents the average amplitude of the pulse in the i-th unit at k time, M is the total number of units, s c(k) is the average amplitude of all measurement units at k time, g is the pulse amplitude normalization of the i-th unit at k time. Equation 4 obtains the amplitude feature statistics by amplitude feature normalization mapping. When alpha tends to 1, it means that the measured unit has a higher possibility of existence of partial discharge points.

[0015] S3, frequency domain analysis is performed on the collected partial discharge signals: Fourier transform is performed on the collected partial discharge signals to extract frequency spectrum information, frequency spectrum is used as frequency domain feature, and the difference between the frequency spectrum of the partial discharge signal and the noise spectrum is converted into the normalized characteristic statistical quantity value difference, so that the real partial discharge signal and the noise are distinguished in value. The greater the difference, the higher the possibility of existence of the partial discharge point. The specific modeling is as follows:

[0016]

[0017]

[0018]

[0019] Wherein, fft(s(k,i)) represents the frequency spectrum of the i-th measurement unit at k time, f c(k)The average spectrum of M measurement units at time k, taking the measurement unit containing only clutter as the reference unit, is f c(k) The average spectrum of clutter. f is the spectrum normalization of the i-th unit at time k, and β is the normalized frequency domain feature. A certain threshold is set, and when max(f) is lower than the threshold, β = 0.01.

[0020] S4, fusion of time domain and frequency domain features with GM-PHD: the features in time domain and frequency domain are used to update the weight of the GM-PHD filter in the form of product, and the features are used to correct the weight of the local discharge point Gaussian component, so as to realize the positioning of the local discharge point under the influence of strong clutter. The specific process is as follows:

[0021] Let the posterior intensity at time k-1 be:

[0022]

[0023] In the formula, N(x; m, P) represents a Gaussian distribution with mean m and covariance P, represents the weight of the Gaussian component at time k-1, and J k-1 is the number of Gaussian components at time k-1.

[0024] The prediction equation at time k can be represented as:

[0025] D k|k-1 (x) = D S,k|k-1 (x) + Y k (x) (9)

[0026] In the formula, D S,k|k-1 (x) is the posterior intensity of the surviving local discharge point, Y k (x) is the posterior intensity of the newly generated local discharge point.

[0027] For the local discharge point existing at time k, it may be detected with a probability of P D,k and generate a measurement according to the likelihood function, or it may be missed with a probability of 1-P D,k . Assuming that the prediction at time k is D k|k-1 (x), and new observation z k is obtained at time k, then the probability hypothesis density after data update at time k is:

[0028]

[0029] In the formula, (1-P D,k )D k|k-1 (x) represents the reservation of the predicted probability hypothesis density for the local discharge point that is not detected, P D,k is the detection probability at time k; D D,k(x;z) is the update of the partial probability hypothesis density detected at k moment, which can be expressed as:

[0030]

[0031] A certain threshold is set to reduce the missed detection as much as possible, when alpha and beta are greater than the set threshold at the same time, the time domain and frequency domain features are fused with GM-PHD, and the Gaussian weight can be updated as:

[0032]

[0033] In the formula, K k (z) is the probability hypothesis density of false alarm or noise in measurement, which can be expressed by the product of clutter number and clutter distribution. H k (z) represents the observation matrix at k moment, R k (z) is the covariance of observation noise at k moment, (z) represents the predicted weight, (z) represents the covariance of process noise.

[0034] The Gaussian components with weights lower than the preset pruning threshold are discarded in the Gaussian components obtained at k moment, the Gaussian components with larger weights are retained, new Gaussian components are obtained, some Gaussian components satisfying the merging threshold are merged into one Gaussian component, the Gaussian mean values with weights greater than a set value are selected to extract multiple partial discharge point states, and the positioning result is obtained.

[0035] The beneficial effects of the present application are:

[0036] The partial discharge positioning method with feature assistance of the present application fuses the time domain and frequency domain features of the partial discharge signal by using the Gaussian mixture probability hypothesis density filter, so that the positioning of the partial discharge point with double randomness of number and state in a complex noise environment can be more accurately realized, the strong clutter can be suppressed, and the influence of the clutter on the partial discharge positioning can be reduced. The present application can filter out a large amount of clutter while completing the positioning of the partial discharge point, suppress the interference of strong clutter, and has the advantages of fast positioning response speed, high positioning accuracy, strong anti-interference performance and the like. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 is a flowchart of the method of the present application;

[0038] Figure 2 is the OSPA error of GM-PHD partial discharge positioning;

[0039] Figure 3 is a comparison diagram of the number of partial discharge true value and estimated value of GM-PHD partial discharge positioning;

[0040] Figure 4is the OSPA error of the application in simulation analysis;

[0041] Figure 5 is a comparison chart of the actual value and the estimated value of the number of partial discharge points in the simulation process of the application. DETAILED DESCRIPTION

[0042] The application will be further described below with reference to the accompanying drawings.

[0043] As Figure 1 shown, a feature-assisted partial discharge positioning method comprises the following steps:

[0044] S1, collecting and preliminarily processing the partial discharge signal, collecting the partial discharge signal of the power equipment through the partial discharge detection device, and obtaining the position measurement information of the partial discharge point, the partial discharge pulse signal and the spectrum diagram and other information after preliminary processing.

[0045] S2, time domain analysis of the collected partial discharge signal: taking the partial discharge signal as the time domain signal, using the amplitude as the time domain feature modeling, assuming that the time interval of state updating is 1s, and N=5000 pulses can be used for feature extraction per second.

[0046] S3, frequency domain analysis of the collected partial discharge signal: Fourier transform is performed on the collected partial discharge signal to extract the frequency spectrum information, the frequency spectrum is used as the frequency domain feature, the normalized threshold is set to 3000, the difference between the partial discharge signal frequency spectrum and the clutter frequency spectrum is converted into the normalized feature statistical quantity difference, and the real partial discharge signal is distinguished from the clutter in the numerical value.

[0047] S4, fusion of time domain and frequency domain features and GM-PHD: the two-dimensional features of time domain and frequency domain are used in the form of product to update the weight of GM-PHD, and the feature is used to correct the weight of the Gaussian component of the partial discharge point.

[0048] The survival probability P S of the partial discharge point is set to 0.99 and the detection probability P D is set to 0.9. The simulation environment is constructed in a space area with a length, width and height of 1000m, the simulation time is set to 100s, the clutter distribution obtained at each monitoring time is independent and the number thereof obeys the Poisson distribution, dense clutter is used, and the average number of clutters obtained at each sampling time is 100. Figure 2 and Figure 3 are the OSPA error of the GM-PHD algorithm for realizing partial discharge positioning under dense clutter and the comparison chart of the estimated difference of the number of partial discharge points. Figure 4 and Figure 5The OSPA error and the number estimation difference of the partial discharge positioning using the method of the application under dense clutter are respectively shown in Table 1. Table 1 shows more intuitive data error comparison. As can be seen, the method of the application can realize the positioning of the partial discharge point under dense clutter, and the error is smaller than that of the traditional GM-PHD method, indicating that the clutter suppression effect is good.

[0049] Table 1

[0050]

[0051] The above series of detailed descriptions are only specific descriptions of the feasible implementation modes of the application, and are not used to limit the protection scope of the application. Any equivalent mode or change without departing from the technology of the application should be included in the protection scope of the application.

Claims

1. A feature-assisted partial discharge point location method, characterized in that, Comprise the following steps: S1, collect and preliminarily process partial discharge signal, after processing, the position measurement information of partial discharge point, partial discharge pulse signal and spectrum information are obtained; S2, time domain analysis is carried out on the collected partial discharge signal: using amplitude as time domain feature modeling, considering the characteristics of partial discharge point, the higher the amplitude, the higher the possibility of existence of partial discharge point; S3, frequency domain analysis is carried out on the collected partial discharge signal: Fourier transform is carried out on the collected partial discharge signal to extract frequency spectrum information, using frequency spectrum as frequency domain feature, the difference between partial discharge signal frequency spectrum and noise frequency spectrum is converted into normalized characteristic statistical quantity value difference, and the difference between true partial discharge signal and noise is distinguished in value, the greater the difference, the higher the possibility of existence of partial discharge point; S4, time domain and frequency domain features are fused with GM-PHD: the features of two dimensions of time domain and frequency domain are used to update the weight of GM-PHD filter in the form of product, the weight of local discharge point Gaussian component is corrected by the features, and the positioning of local discharge point under the influence of strong noise is realized; The updating method of GM-PHD filter weight of S4 is as follows: The posterior intensity at k-1 moment is: where N(x; m, P) denotes a Gaussian distribution with mean m and covariance P, denotes the weight of the Gaussian component at time k - 1, k-1 is the number of Gaussian components at time k - 1. The prediction equation at k moment is expressed as: D k|k-1 (x) = D S,k|k-1 (x) + Y k (x) (2) where D S,k|k-1 (x) is the posterior intensity of surviving partial discharge points, Y k (x) is the posterior intensity of newly created partial discharge points; For a local discharge point existing at time k, it can be detected with a probability of P D,k and generate a measurement according to the likelihood function, or it can be missed with a probability of 1-P D,k . Assuming that the prediction at time k is D k|k-1 (x), and a new observation z k is obtained at time k, the updated probability hypothesis density at time k is: where (1 - P D,k )D k|k-1 (x) represents the reservation of the predictive probability density of the non-detected partial discharge point, P D,k is the detection probability at time k; D D,k (x; z) is the update of the partial probability density that is detected at time k, and is expressed as: A certain threshold is set to reduce the missed detection part as much as possible, when the amplitude characteristic statistical quantity α and the normalized frequency domain feature β are greater than the set threshold at the same time, the time domain and frequency domain features are fused with GM-PHD, and the Gaussian weight is updated as: where K k (z) is the probability hypothesis density of false alarm or noise in the measurement, expressed as the product of clutter number and clutter distribution; H k represents the observation matrix at time k, R k is the covariance of observation noise at time k, represents the weight of prediction, represents the covariance of process noise; The positioning method of S4 local discharge point: the Gaussian components with weight lower than the preset pruning threshold are discarded in the Gaussian components obtained at k moment, the Gaussian components with larger weight are retained, new Gaussian components are obtained, some Gaussian components meeting the merging threshold are merged into one Gaussian component, the Gaussian mean with weight greater than the set value is selected to extract multiple partial discharge point states, and the positioning result is obtained.

2. The partial discharge point locating method assisted by features according to claim 1, characterized in that, The time domain feature modeling of S2 is as follows: g = (s(k,i) - s c(k) ) / s c(k) (8) a = (1 + exp(-2g)) -1 (9) where N represents the total number of pulses of the i-th cell, s(k,i) represents the average amplitude of the pulses in the i-th cell at the k-th time, M is the total number of cells, and s(k) is the average amplitude of all the measured cells at the k-th time. c(k) is the average amplitude of all the measured cells at the k-th time.

Images