CN115236385A - An automatic identification method of high frequency pulse current waveform polarity - Google Patents

Abstract

The invention discloses an automatic identification method for the polarity of a high-frequency pulse current waveform belonging to the technical field of insulation fault detection of electric power equipment. Then, the first wave propagation characteristics of different typical discharge positions and types are obtained by injecting steep pulses, and the response signal waveforms are measured through each outgoing line coupling terminal, and a typical response waveform sample library for each injection mode and position is established; and then the waveform sequence is used as the input. Vector, the method of training the first wave waveform and polarity of the input waveform sequence through the artificial neural network, and using the artificial neural network to identify the waveform details, so as to realize the automatic, efficient and accurate identification of the polarity of the high-frequency current waveform of the transformer The identification provides key diagnostic information for the diagnosis of key states such as the discharge type and location of partial discharge, and improves the robustness and automation level of the diagnosis.

Description

An automatic identification method of high frequency pulse current waveform polarity

technical field

The invention belongs to the technical field of insulation fault detection of power equipment, and particularly relates to an automatic identification method for the polarity of a high-frequency pulse current waveform

Background technique

Partial discharge is an important symptom of internal insulation defects of power equipment, and it is also an important basis for equipment insulation fault diagnosis and fault location; high-frequency partial discharge detection has a wide bandwidth and rich information, and is easy to install and implement through equipment grounding. The polarity information of the high-frequency pulse current signal is an important basis for the identification of the interference pulse, the discharge position and the discharge type.

Under the environmental conditions of live operation of field equipment, the polarity of the measured pulse current waveform is affected by background noise and interference signals, and the first wave of the discharge waveform will be "polluted", making the first wave difficult to distinguish; in addition, due to the high frequency signal After the propagation of windings, long wires, etc., its high-frequency components will be attenuated and distorted, which also makes it difficult to identify the first wave. At present, in the on-site partial discharge live detection, the tester manually confirms the waveform polarity based on experience, and estimates the time delay accordingly, so as to analyze and judge the discharge type and discharge location. Due to the unavoidable personal safety risks of on-site personnel during the testing of defective equipment, it is necessary to carry out machine identification through online monitoring or intensive care systems. There are several methods for judging the polarity of pulse waveform, mainly as follows:

(1) First, the interference signal is filtered out by various filtering means, the background noise level is reduced, the signal-to-noise ratio of detection is improved, and the recognizability of the first wave signal is improved; the threshold method is to set a threshold according to the background noise level, The polarity of the pulse can be determined according to the level value when the first wave of the pulse waveform exceeds the threshold, as shown in Figure 1. The two parallel horizontal lines in the middle of Figure 1 are the threshold, and the first wave is positive when it passes the threshold. This method is simple and intuitive, but when the background noise level is high or there is an interfering signal, it is very difficult to judge the polarity of the first wave due to the decrease of the signal-to-noise ratio of the detection signal.

(2) Directly through waveform analysis, the polarity of the first passing threshold is used as the polarity of the first wave to be judged. This judgment method is very basic, but it is affected by the threshold setting, background noise and interference signals (as shown in Figure 2) or The correlation analysis method is adopted, and a representative waveform X is selected as a reference. The polarity of the waveform is known. The waveform u ₂ (j) of the polarity to be determined and the template file Y are calculated using the formula (1) to calculate the similarity coefficient,

The similarity coefficient ρ≥k, k is the judgment threshold, and ρ is between 0 and 1. The closer it is to 1, the higher the similarity between the two waveforms, and the higher the consistency between the polarity of the waveform to be tested and the reference waveform. This method is also the basic algorithm for waveform polarity judgment and time delay estimation. However, this method is also greatly affected by the background noise level, and at the same time, due to the attenuation and distortion of the signal propagation process, especially the superposition effect of the catadioptric signal, the effectiveness of the related method will be seriously reduced.

(3) The method of reading the first wave of the pulse signal by the energy accumulation method and the cross-correlation method. Compared with the direct determination by the threshold method, this method will improve the consistency and stability of the identification, but its effect is still greatly affected. Constrained by background noise and interfering signals (as shown in Figure 2), and significant degradation occurs as the level of interfering signals increases; the energy accumulation method is essentially a second-order statistic method. After squaring the signal, A method of observing the starting position of the first wave of a signal. Since the signal energy is proportional to the square of the voltage, the voltage waveform of the pulse signal can be converted into an energy-related value accumulation curve. When the PD signal is much larger than the background noise, an obvious inflection point will occur on the curve, and the inflection point can be regarded as a partial discharge. The starting moment of occurrence. Un is the voltage value of the nth point on the signal waveform, h(h<N) is the number of points accumulated and calculated by the signal, then the accumulated energy is

In the formula: ti is the start time of signal acquisition;

u(t') is the UHF signal amplitude at time t';

R is the input impedance of the acquisition system.

This forms an energy accumulation curve, the inflection point of which is considered to be the start of the signal. The starting time of finding the original signal is converted into finding the inflection point of the energy accumulation curve, as shown in Figure 3. It can be seen from Figure 3 that the transition region of the energy accumulation curve is very gentle, and the arrival position of the first wave is not easy to distinguish, and the square transformation makes the polarity information of the waveform disappear. Obviously, the background noise and interfering signals will seriously degrade the turning information of this method.

(4) Fractional Low-Order Statistics Theory (FLOS) is developed from the second-order statistics theory. The introduction of FLOS into the high-resolution multipath delay estimation algorithm can improve the algorithm’s ability to resist impulse noise and solve classical problems in distributed noise environments. The performance of the algorithm is degraded, but it has a high dependence on the prior knowledge of the characteristics of the noise signal.

To sum up, the existing first wave and waveform polarity identification methods have good effects in the case of high signal-to-noise ratio. However, with the increase of the background noise level and the strength of the interfering signal, and the decrease of the signal-to-noise ratio, the identification accuracy of the first wave and the polarity of the waveform will be seriously reduced, which cannot meet the requirements for the diagnosis and evaluation of the discharge type, location and state in engineering. requirements. The lack of exploration and utilization of waveform features is an important reason for the poor performance of these methods when the signal-to-noise ratio is low. Therefore, based on the observation and analysis of a large number of measured data, an automatic identification method of the first wave polarity of pulse signal based on neural network is proposed. High and adaptable, suitable for fault analysis and diagnosis without manual intervention during online monitoring and intensive care.

SUMMARY OF THE INVENTION

The purpose of the present invention is to propose an automatic identification method for the polarity of a high-frequency pulse current waveform, which is characterized in that the method is to deeply utilize the first wave and its polarity identification method of the waveform characteristics of the pulse signal; The first wave propagation characteristics of different typical discharge positions and types are obtained by the steep pulse method, and the response signal waveform is measured through the coupling end of each outlet line, and the typical response waveform sample library of each injection mode and position is established; and then the waveform sequence is used as the input vector. The neural network trains the waveform and polarity of the first wave on the input waveform sequence, and uses the artificial neural network to identify the details of the waveform. The weight of the attenuated oscillating signal will be automatically reduced, thereby improving the accuracy of the first wave and its polarity identification; the specific steps include:

(1) Build the transformer entity model platform, and simulate the internal discharge of the equipment by injecting signals into the entity transformer model through the steep pulse generator. discharge;

(2) Install high-frequency CT sensors at the end screen of transformer bushing, neutral point, iron core, clamp grounding wire, fuel tank grounding wire, etc., and sample high-frequency impulse response waveforms synchronously through the acquisition device; at this time, due to the injection signal condition The lower device is not powered, so basically no external interference signal enters the test loop, and the background noise level is very low;

(3) Establish a waveform sample library through injection in different locations and methods, and mark the polarity of its first wave;

(4) For the waveform interception subsequence in the sample library, take k times the root mean square value of the sequence as the threshold, and take the pulse subsequence of 1us forward and 2us backward from the first threshold point in the sequence as the threshold The artificial neural network is input, and the first wave polarity is used as the output to train the network parameter matrix of the neural network. It stops until the set recognition accuracy is reached. The initial stage is set to 100%. Convergence, the precision limit value can be reduced by 0.1% and re-iterative training is performed; among them, k takes 1.3 to 1.5;

(5) By artificially adding a noise signal to the waveform sequence in the sample library, and adjusting the amplitude level of the noise signal to control the signal-to-noise ratio, when the specified signal-to-noise ratio SNR _th is above, further strengthen the training of the neural network to improve the adaptability The network parameters can be output after reaching the set recognition accuracy; SNR _th is the set signal-to-noise ratio threshold, and the signal-to-noise ratio threshold is not less than 10dB.

(6) Using the network parameters after training to intercept the collected pulse waveform, and using the intercepted subsequence as the input vector of the neural network, the automatic identification of the first wave polarity of the high-frequency partial discharge pulse can be realized.

The beneficial effect of the present invention is that the method uses artificial neural network to identify the waveform details, can effectively accumulate the experience accumulated in the early stage, and can continuously expand the samples through confrontation learning, realize the fully automatic first wave polarity identification, and is efficient in application. , concise, suitable for the application of real-time algorithm in online monitoring.

Description of drawings

Figure 1 is the threshold method to determine the waveform polarity;

FIG. 2 is a schematic diagram of a mailbox for determining the polarity of waveforms against background noise and interference signals.

Fig. 3 is the energy accumulation curve of pulse waveform;

Figure 4 is the injection signal in the simulation of external interference and different discharge forms, wherein a) external interference signal; b) winding to ground; c) winding turn/cake;

Figure 5 is a schematic diagram of the position of the signal coupling point;

Figure 6 is a typical waveform sample library a) the multi-terminal detection waveform when the first end of the winding is injected, b) the multi-terminal detection waveform injected between the turns of the pie in the middle of the winding;

Figure 7 is a flowchart of automatic identification of the polarity of the high-frequency pulse current waveform.

Detailed ways

The invention proposes an automatic identification method for the polarity of a high-frequency pulse current waveform; the method is to deeply utilize the first wave and its polarity identification method of the waveform characteristics of the pulse signal. The first wave propagation characteristics of the discharge position and type are measured by the response signal waveform of each outlet coupling end, and the typical response waveform sample library of each injection mode and position is established, and then the waveform sequence is used as the input vector. The method for training the first wave waveform and polarity in a sequence will be further described below with reference to the accompanying drawings.

The specific process steps of the first wave of the described depth utilization pulse signal waveform feature and its polarity identification are shown in the flow chart of automatic identification of the polarity of the high-frequency pulse current waveform in Figure 7:

(1) Build the transformer entity model platform, and simulate the internal discharge of the equipment by injecting signals into the entity transformer model through the steep pulse generator. Discharge, etc., as shown in Figure 4, respectively, are the injection signals in the simulation of external interference and different discharge forms, where a) external interference signal; b) winding to ground; c) winding turn/cake;

(2) Install high-frequency CT sensors at the end screen of the transformer bushing, neutral point, iron core, clamp grounding wire, fuel tank grounding wire, etc., and synchronously sample the high-frequency impulse response waveform through the acquisition device, as shown in the signal coupling point in Figure 5 The location diagram is shown. Since the device is not powered when the signal is injected, almost no external interference signal enters the test loop, and the background noise level is very low.

(3) Establish a waveform sample library through injection in different locations and methods, and mark the polarity of its first wave. Figure 6 shows a typical waveform sample library. Among them, a) the multi-terminal detection waveform when the winding head is injected, b) the multi-terminal detection waveform injected between the turns of the pie in the middle of the winding;

(4) For the waveform truncation subsequence in the sample library, the method is to take k times the root mean square value of the sequence (k is 1.3 to 1.5) as the threshold, and take 1us forward from the first threshold point in the sequence , Take the pulse subsequence of 2us backward as the input of the artificial neural network, and use the polarity of the first wave as the output to train the network parameter matrix of the neural network, until it reaches the set recognition accuracy and stop, the initial stage is set to 100%, if If it still does not converge after reaching the limit of the number of iterations, the limit of accuracy can be reduced by 0.1% and re-iterative training is performed;

(5) By artificially adding a noise signal to the waveform sequence in the sample library, and adjusting the amplitude level of the noise signal to control the signal-to-noise ratio, when the specified signal-to-noise ratio SNR _th is above, further strengthen the training of the neural network to improve the adaptability The network parameters can be output after reaching the set recognition accuracy;

(6) Using the network parameters after training to intercept the collected pulse waveform, and using the intercepted subsequence as the input vector of the neural network, the automatic identification of the first wave polarity of the high-frequency partial discharge pulse can be realized.

In summary, the present invention organically integrates prior knowledge such as high-frequency pulse current propagation characteristics into transformer high-frequency partial discharge detection, and uses artificial neural network to identify waveform details. Subsequent reflected superimposed signals and attenuated oscillating signals in the pulse waveform that have nothing to do with the first wave will automatically reduce the weight, thus realizing the automatic, efficient and accurate identification of the polarity of the high-frequency current waveform of the transformer. The diagnosis of such critical states provides key diagnostic information and improves the accuracy of the first wave and its polarity identification. Improved robustness and automation of diagnostics.

Claims (1)

1. a kind of automatic identification method of high frequency pulse current waveform polarity, it is characterized in that this method is the first wave and its polarity identification method of deeply utilizing pulse signal waveform feature; At first under laboratory conditions, obtain by injecting steep pulse mode The first wave propagation characteristics of different typical discharge positions and types are measured through the coupling end of each outlet line, and the typical response waveform sample library of each injection mode and position is established; then the waveform sequence is used as the input vector, and the artificial neural network The method of training the waveform and polarity of the first wave by using the waveform sequence, and using artificial neural network to identify the details of the waveform. Due to the nonlinear characteristics of the neural network, the subsequent reflection superimposed signals and attenuated oscillation signals in the pulse waveform that are not related to the first wave are all The weight will be automatically reduced to improve the accuracy of the first wave and its polarity identification; the specific steps include: (1) Build the transformer entity model platform, and simulate the internal discharge of the equipment by injecting signals into the entity transformer model through the steep pulse generator. discharge; (2) Install high-frequency CT sensors at the end screen of transformer bushing, neutral point, iron core, clamp grounding wire, fuel tank grounding wire, etc., and sample high-frequency impulse response waveforms synchronously through the acquisition device; at this time, due to the injection signal condition The lower device is not powered, so basically no external interference signal enters the test loop, and the background noise level is very low; (3) Establish a waveform sample library through injection in different locations and in different ways, and mark the polarity of the first wave; (4) For the waveform interception subsequence in the sample library, take k times the root mean square value of the sequence as the threshold, and take the pulse subsequence of 1us forward and 2us backward from the first threshold point in the sequence as the threshold The artificial neural network is input, and the first wave polarity is used as the output to train the network parameter matrix of the neural network. It stops until the set recognition accuracy is reached. The initial stage is set to 100%. Convergence, the precision limit value can be reduced by 0.1% and re-iterative training is performed; among them, k takes 1.3 to 1.5; (5) By artificially adding a noise signal to the waveform sequence in the sample library, and adjusting the amplitude level of the noise signal to control the signal-to-noise ratio, when the specified signal-to-noise ratio SNR _th is above, further strengthen the training of the neural network to improve the adaptability The network parameters can be output after reaching the set recognition accuracy; SNR _th is the set signal-to-noise ratio threshold, and the signal-to-noise ratio threshold is not less than 10dB; (6) Using the network parameters after training to intercept the collected pulse waveform, and using the intercepted subsequence as the input vector of the neural network, the automatic identification of the first wave polarity of the high-frequency partial discharge pulse can be realized.

Images