阅读说明：本技术 一种真空灭弧室内真空度监测方法及系统 (Vacuum degree monitoring method and system in vacuum arc extinguishing chamber ) 是由 马强平 陈立 韦云清 李兴文 于 2021-08-27 设计创作，主要内容包括：本发明公开了一种真空灭弧室内真空度监测方法及系统,通过从外部获取真空灭弧室在工作状态下的电磁波信号,对获取的电磁波信号进行小波分解,将小波分解后的电磁波信号进行系数阈值处理后再通过小波包重构得到去噪信号,然后将去噪信号进行S变换得到时频谱,再对时间轴积分得到信号边际谱,基于边际谱提取信号的频域特征量,提取脉冲信号幅值均值和脉冲次数作为时域特征量,采用预训练模型进行真空度预测,提高了对原始信号信息的利用效率,并且基于频域特征量和时域特征量能够提高预测灭弧室内真空度的精度,误差较小,不需要对真空灭弧室进行插接,可实现在线工作监测,避免了与灭弧室的直接接触,也不用内置传感器,不用改变灭弧室结构,简单快捷,提高了监测安全性。(The invention discloses a vacuum degree monitoring method and a system in a vacuum arc extinguish chamber, which can improve the utilization efficiency of original signal information by acquiring electromagnetic wave signals of the vacuum arc extinguish chamber from the outside in a working state, performing wavelet decomposition on the acquired electromagnetic wave signals, performing coefficient threshold processing on the electromagnetic wave signals after the wavelet decomposition, then reconstructing a wavelet packet to obtain a de-noised signal, performing S transformation on the de-noised signal to obtain a time-frequency spectrum, integrating a time axis to obtain a signal marginal spectrum, extracting frequency domain characteristic quantity of the signal based on the marginal spectrum, extracting the amplitude mean value and pulse times of the pulse signal as time domain characteristic quantity, adopting a pre-training model to predict the vacuum degree, improving the precision of predicting the vacuum degree in the arc extinguish chamber based on the frequency domain characteristic quantity and the time domain characteristic quantity, having smaller error, being free from plugging the vacuum arc extinguish chamber and realizing on-line work monitoring, the direct contact with the arc extinguish chamber is avoided, a built-in sensor is not needed, the structure of the arc extinguish chamber is not needed to be changed, the method is simple and rapid, and the monitoring safety is improved.)

1. A vacuum degree monitoring method in a vacuum arc extinguish chamber is characterized by comprising the following steps:

s1, acquiring electromagnetic wave signals of the vacuum arc extinguish chamber in a working state, then performing wavelet decomposition on the acquired electromagnetic wave signals, performing coefficient threshold processing on the electromagnetic wave signals after the wavelet decomposition, and then performing wavelet packet reconstruction to obtain de-noising signals;

s2, performing S transformation on the de-noised signal to obtain a time frequency spectrum, integrating the time frequency spectrum along a time axis to obtain a marginal spectrum, extracting frequency domain characteristic quantity and time domain characteristic quantity of the signal through the marginal spectrum, and then performing characteristic dimension reduction and characteristic screening on the frequency domain characteristic quantity and the time domain characteristic quantity to obtain final characteristic quantity;

and S3, predicting the final characteristic quantity by using a pre-training model to obtain a corresponding vacuum degree value, wherein the pre-training model is obtained by training the characteristic quantity extracted from the electromagnetic wave signal of the vacuum arc-extinguishing chamber with known vacuum degree in the working state.

2. The method for monitoring the vacuum degree in the vacuum arc extinguish chamber according to claim 1, wherein the annular antennas are arranged at intervals on the periphery of the vacuum arc extinguish chamber, and the annular plane is opposite to the vacuum arc extinguish chamber to obtain the electromagnetic wave signals of the vacuum arc extinguish chamber in the working state.

3. The method for monitoring the vacuum degree in the vacuum arc extinguish chamber according to claim 2, wherein the distance between the loop antenna and the vacuum arc extinguish chamber is 0.5-1.5 m, the loop antenna is connected with a data acquisition system, and the sampling rate of the data acquisition system is 500 kHz.

4. The vacuum degree monitoring method in the vacuum arc extinguish chamber according to claim 1, wherein db4 wavelet basis functions are adopted to carry out 3-layer wavelet packet decomposition on the acquired electromagnetic wave signals; the wavelet packet noise reduction selects a fixed threshold, and the expression of the threshold is as follows:

where σ is the noise mean square error and N is the size or length of the signal.

5. The method for monitoring the vacuum degree in the vacuum arc extinguish chamber according to claim 1, wherein a time frequency spectrum is obtained by performing S transformation on the de-noised signal, and the formula of the S transformation is as follows:

wherein h (t) is an input signal; omega (tau-t, f) is a Gaussian window function, tau is a displacement factor, the position of the Gaussian window on the time axis is controlled, and the expression of the Gaussian window function is as follows:

6. the method of claim 1, wherein the frequency domain characteristic parameters include signal spectral mean, variance and bandwidth.

7. The method of claim 1, wherein the time domain signature includes a mean of pulse number and pulse amplitude for positive and negative half cycles of 10 cycles.

8. The method of claim 1, wherein the characteristic quantity of the electromagnetic wave signal is H_k＝(h^1k,h^2k,…,h^nk)^TThen H is_kHas a covariance matrix of

Where k is the number of samples, n is the feature vector dimension,is the mean of the feature vectors of each sample.

9. The method of claim 8, wherein the covariance matrix is solved for all eigenvalues λ_iAnd a feature vector v_iThen, the eigenvalues are ranked from large to small: lambda [ alpha ]₁＞λ₂＞…＞λ_mWhen the value is more than …, the characteristic value is selected to be more than lambda_mThe feature vector of (a) constitutes the principal vector, sample H_iProjection onto a feature vector v_iObtaining principal component in that directionThe cumulative variance contribution of the first m principal components isThe cumulative variance contribution H (m) > 95% is taken.

10. A vacuum degree monitoring system in a vacuum arc extinguish chamber based on the monitoring method of claim 1, which is characterized by comprising a prediction module and a data processing module;

the prediction module is used for storing a prediction model obtained by training according to characteristic quantities extracted from electromagnetic wave signals of a vacuum arc-extinguishing chamber with known vacuum degree in a working state, the data processing module is used for performing wavelet decomposition on the electromagnetic wave signals of the vacuum arc-extinguishing chamber in the working state, performing coefficient threshold processing on the electromagnetic wave signals after the wavelet decomposition, and reconstructing through a wavelet packet to obtain a de-noising signal; and performing S transformation on the de-noised signal to obtain a time-frequency spectrum, integrating the time-frequency spectrum along a time axis to obtain a marginal spectrum, extracting frequency domain characteristic quantity and time domain characteristic quantity of the signal through the marginal spectrum, then performing characteristic dimension reduction and characteristic screening on the frequency domain characteristic quantity and the time domain characteristic quantity to obtain final characteristic quantity, inputting the final characteristic quantity to a prediction module, and predicting to obtain a vacuum degree value corresponding to the characteristic quantity.

Technical Field

The invention belongs to the technical field of electrical fault detection of high-voltage vacuum circuit breakers, and particularly relates to a vacuum degree monitoring method and system in a vacuum arc extinguish chamber.

Background

The vacuum circuit breaker has the advantages of small volume, good breaking performance, 30 years of service life, no explosion and fire hazard, environmental friendliness and small maintenance workload, so that the vacuum circuit breaker is developed rapidly and the production capacity is increased continuously. At present, China is a large producing country. As the technology is more advanced and matured, the vacuum circuit breaker in 12kV and 40.5kV voltage levels gradually replaces SF₆Trend of circuit breakers. The future development trend is high capacity, high voltage class vacuum circuit breakers. In recent years, social development has led to further improvement in economic level, and higher demands have been made on reliability and stability of power systems. Although the vacuum circuit breaker has a small failure rate, it causes a serious economic loss once a failure occurs. Therefore, the condition monitoring of the vacuum circuit breaker is carried out, and defects and faults can be found in time, which is a hotspot and a difficulty of research of people.

The vacuum arc-extinguishing chamber is in a sealed state once being manufactured, the vacuum degree inside the vacuum arc-extinguishing chamber is difficult to monitor, and two major types of off-line detection methods and on-line detection methods are mainly adopted at present. The method is characterized in that the vacuum circuit breaker in operation is regularly inspected or the circuit breaker stops operating for detection, and the method is called off-line detection and mainly comprises a power frequency withstand voltage method, a pulse magnetic control discharge method, a pulse current detection method and the like. The off-line detection method is mature, has high detection precision and is available in the market, but generally requires a circuit where the circuit breaker is located to stop running and even disassemble an arc extinguish chamber, which is contradictory to the increasingly improved reliability and economy of an electric power system, and the required detection device has large volume and complex operation. Therefore, online detection methods are a future development trend. The online detection means that the detection of the vacuum degree is realized without influencing the normal operation of the vacuum circuit breaker, and mainly comprises an electromagnetic wave method, a coupling capacitance method, an electro-optical conversion method and the like. However, the existing online monitoring technology is not mature enough and has poor applicability. When the vacuum degree is reduced, partial discharge can occur between the contact and the shielding cover in the arc extinguish chamber, but in the analysis method for detecting the vacuum degree of the arc extinguish chamber by utilizing the principle, the relationship between the discharge amount and the phase and the relationship between the discharge times and the phase are more utilized, so that a large amount of statistics needs to be carried out, the time consumption is long, the analysis time is long, the inconvenience is very high, the qualitative analysis can be realized, and the corresponding relationship with the vacuum degree quantification is not established. Therefore, people need to be able to more effectively utilize the related signals of partial discharge and quickly and accurately judge the vacuum degree in the arc extinguish chamber.

Disclosure of Invention

The invention aims to provide a method and a system for monitoring the vacuum degree in a vacuum arc extinguish chamber, which are used for overcoming the defects of the prior art.

In order to achieve the purpose, the invention adopts the following technical scheme:

a vacuum degree monitoring method in a vacuum arc extinguishing chamber comprises the following steps:

s1, acquiring electromagnetic wave signals of the vacuum arc extinguish chamber in a working state, then performing wavelet decomposition on the acquired electromagnetic wave signals, performing coefficient threshold processing on the electromagnetic wave signals after the wavelet decomposition, and then performing wavelet packet reconstruction to obtain de-noising signals;

s2, performing S transformation on the de-noised signal to obtain a time frequency spectrum, integrating the time frequency spectrum along a time axis to obtain a marginal spectrum, extracting frequency domain characteristic quantity and time domain characteristic quantity of the signal through the marginal spectrum, and then performing characteristic dimension reduction and characteristic screening on the frequency domain characteristic quantity and the time domain characteristic quantity to obtain final characteristic quantity;

and S3, predicting the final characteristic quantity by using a pre-training model to obtain a corresponding vacuum degree value, wherein the pre-training model is obtained by training the characteristic quantity extracted from the electromagnetic wave signal of the vacuum arc-extinguishing chamber with known vacuum degree in the working state.

Furthermore, the annular antennas are arranged on the periphery of the vacuum arc-extinguishing chamber at intervals, and the annular plane is opposite to the vacuum arc-extinguishing chamber, so that electromagnetic wave signals of the vacuum arc-extinguishing chamber in a working state are obtained.

Furthermore, the distance between the loop antenna and the vacuum arc extinguish chamber is 0.5-1.5 m, the loop antenna is connected with a data acquisition system, and the sampling rate of the data acquisition system is 500 kHz.

Further, performing 3-layer wavelet packet decomposition on the acquired electromagnetic wave signals by adopting a db4 wavelet basis function; the wavelet packet noise reduction selects a fixed threshold, and the expression of the threshold is as follows:

where σ is the noise mean square error and N is the size or length of the signal.

Further, performing S transform on the denoised signal to obtain a time-frequency spectrum, where the formula of the S transform is:

wherein h (t) is an input signal; omega (tau-t, f) is a Gaussian window function, tau is a displacement factor, the position of the Gaussian window on the time axis is controlled, and the expression of the Gaussian window function is as follows:

further, the frequency domain characteristic quantity includes a signal spectrum mean, a variance and a bandwidth.

Further, the time domain characteristic quantity comprises the pulse times and the average value of the pulse amplitude in the positive half period and the negative half period of 10 periods.

Further, let the characteristic quantity of the obtained electromagnetic wave signal be H_k＝(h^1k,h^2k,…,h^nk)^TThen H is_kHas a covariance matrix of

Where k is the number of samples, n is the feature vector dimension,is the mean of the feature vectors of each sample.

Further, solving all eigenvalues λ of covariance matrix_iAnd a feature vector v_iThen the characteristic value is selected fromSize to size alignment: lambda [ alpha ]₁＞λ₂＞…＞λ_mWhen the value is more than …, the characteristic value is selected to be more than lambda_mThe feature vector of (a) constitutes the principal vector, sample H_iProjection onto a feature vector v_iObtaining principal component in that directionThe cumulative variance contribution of the first m principal components isThe cumulative variance contribution H (m) > 95% is taken.

A vacuum degree monitoring system in a vacuum arc extinguish chamber comprises a prediction module and a data processing module;

the prediction module is used for storing a prediction model obtained by training according to characteristic quantities extracted from electromagnetic wave signals of a vacuum arc-extinguishing chamber with known vacuum degree in a working state, the data processing module is used for performing wavelet decomposition on the electromagnetic wave signals of the vacuum arc-extinguishing chamber in the working state, performing coefficient threshold processing on the electromagnetic wave signals after the wavelet decomposition, and reconstructing through a wavelet packet to obtain a de-noising signal; and performing S transformation on the de-noised signal to obtain a time-frequency spectrum, integrating the time-frequency spectrum along a time axis to obtain a marginal spectrum, extracting frequency domain characteristic quantity and time domain characteristic quantity of the signal through the marginal spectrum, then performing characteristic dimension reduction and characteristic screening on the frequency domain characteristic quantity and the time domain characteristic quantity to obtain final characteristic quantity, inputting the final characteristic quantity to a prediction module, and predicting to obtain a vacuum degree value corresponding to the characteristic quantity.

Compared with the prior art, the invention has the following beneficial technical effects:

the invention relates to a vacuum degree monitoring method in a vacuum arc extinguish chamber, which obtains electromagnetic wave signals of the vacuum arc extinguish chamber in a working state from the outside, then carries out wavelet decomposition on the obtained electromagnetic wave signals, carries out coefficient threshold processing on the electromagnetic wave signals after the wavelet decomposition, then carries out wavelet packet reconstruction to obtain de-noised signals, then carries out time axis integration to obtain signal marginal spectrums, extracts the signal frequency spectrum mean value, variance and bandwidth as frequency domain characteristic quantities based on the marginal spectrums, extracts the pulse signal amplitude mean value and the pulse times as time domain characteristic quantities, adopts a pre-training model to carry out vacuum degree prediction, improves the utilization efficiency of original signal information, can improve the precision of predicting the vacuum degree in the arc extinguish chamber based on the frequency domain characteristic quantities and the time domain characteristic quantities, has small error, does not need to carry out splicing on the vacuum arc extinguish chamber, can realize on-line work monitoring, and avoids direct contact with the arc extinguish chamber, and a built-in sensor is not needed, the structure of the arc extinguish chamber is not needed to be changed, the method is simple and rapid, and the monitoring safety is improved.

Furthermore, a signal marginal spectrum is obtained after S conversion, time domain characteristic vectors of discharge characteristics can be directly represented based on combination of signal frequency domain characteristics extracted from the marginal spectrum, statistics of relations between discharge quantity and phase and between discharge times and phase is not needed, original signals are well represented, meanwhile, the characteristic quantity extraction process is greatly simplified, and effectiveness of the characteristic quantity is remarkably improved.

Furthermore, the frequency domain characteristic quantity and the time domain characteristic quantity are subjected to characteristic dimension reduction and characteristic screening to obtain the final characteristic quantity, the problem that SVM identification accuracy is reduced due to the fact that noise and redundancy exist in sample characteristic information is solved through principal component analysis, formed characteristic vectors are independent of one another to form an orthogonal relation, the identification accuracy of a test sample is effectively improved, time is shortened, and the final PCA-SVM regression prediction model can accurately judge different vacuum degrees.

Drawings

FIG. 1 is a flow chart of a method for monitoring vacuum in a vacuum interrupter chamber according to an embodiment of the present invention;

FIG. 2 is a diagram of a vacuum interrupter according to an embodiment of the present invention;

FIG. 3 is a schematic diagram of an antenna receiving an electromagnetic wave signal according to an embodiment of the present invention;

FIG. 4a is an electromagnetic wave signal output by an antenna when vacuum level is reduced for vacuum chamber vacuum level monitoring using the present invention;

FIG. 4b is a time-frequency spectrum of an electromagnetic wave signal when S transform is applied;

FIG. 4c is a graph of a marginal spectrum of an electromagnetic wave signal obtained based on a time-frequency spectrum;

FIG. 5 shows the result of the PCA-SVM model prediction for vacuum degree monitoring in the vacuum interrupter chamber, according to the present invention.

In the figure: 1. a stationary end cover plate; 2. a movable end cover plate; 3. an insulating housing; 4. a static conductive rod; 5. a movable conductive rod; 6. a contact; 7. a bellows; 8. a shield case; 9. an antenna; 10. electromagnetic waves; 11. a discharge signal.

Detailed Description

The invention is described in further detail below with reference to the accompanying drawings:

a vacuum degree monitoring method in a vacuum arc extinguishing chamber comprises the following steps:

s1, acquiring electromagnetic wave signals of the vacuum arc extinguish chamber in a working state, then performing wavelet decomposition on the acquired electromagnetic wave signals, performing coefficient threshold processing on the electromagnetic wave signals after the wavelet decomposition, and then performing wavelet packet reconstruction to obtain de-noising signals;

s2, performing S transformation on the de-noised signal to obtain a time frequency spectrum, integrating the time frequency spectrum along a time axis to obtain a marginal spectrum, and extracting frequency domain characteristic quantity and time domain characteristic quantity of the signal through the marginal spectrum;

and S3, predicting the frequency domain characteristic quantity and the time domain characteristic quantity by using a pre-training model to obtain corresponding vacuum degree values, wherein the pre-training model is obtained by training the characteristic quantity extracted from the electromagnetic wave signals of the vacuum arc-extinguishing chamber with known vacuum degree in the working state.

Specifically, an annular antenna is adopted to obtain an electromagnetic wave signal of the vacuum arc-extinguishing chamber in a working state; the annular antennas are arranged on the periphery of the vacuum arc extinguish chamber at intervals, and the annular plane is opposite to the vacuum arc extinguish chamber; specifically, the distance between the loop antenna and the vacuum arc-extinguishing chamber is 0.5-1.5 m.

The loop antenna is connected with a data acquisition system, and the sampling rate of the data acquisition system is 500 kHz.

According to the method, db4 wavelet basis functions are adopted to carry out 3-layer wavelet packet decomposition on the obtained electromagnetic wave signals; the wavelet packet noise reduction selects a fixed threshold, and the expression of the threshold is as follows:

where σ is the noise mean square error and N is the size or length of the signal.

And carrying out S transformation on the de-noised signal to obtain a time frequency spectrum, wherein the formula of the S transformation is as follows:

wherein h (t) is an input signal; omega (tau-t, f) is a Gaussian window function, tau is a displacement factor, the position of the Gaussian window on the time axis is controlled, and the expression of the Gaussian window function is as follows:

the obtained frequency domain characteristic quantity comprises a signal spectrum mean value, a variance and a bandwidth; the time domain characteristic quantity comprises the pulse times and the average value of the pulse amplitude in the positive half period and the negative half period of 10 periods.

Let the characteristic quantity of the electromagnetic wave signal be H_k＝(h^1k,h^2k,…,h^nk)^TThen H is_kHas a covariance matrix of

Where k is the number of samples, n is the feature vector dimension,is the mean of the feature vectors of each sample.

Solving all eigenvalues λ of the covariance matrix_iAnd a feature vector v_iThen, the eigenvalues are ranked from large to small: lambda [ alpha ]₁＞λ₂＞…＞λ_mGreater than …, and selecting characteristic value greater than lambda for lowering dimension_mThe feature vector of (a) constitutes the principal vector, sample H_iProjection onto a feature vector v_iObtaining principal component in that directionThe cumulative variance contribution of the first m principal components isThe method selects the original data information with the accumulated variance contribution rate H (m) larger than 95 percent, namely more than 95 percent, to be reserved in the first m principal components, and utilizes the first m principal components to represent the original information, thereby removing the noise and redundancy of the feature vector of the original vibration signal and realizing the dimension reduction.

Example (b):

step 1: the method comprises the following steps of placing an annular antenna at a position 0.5-1.5 m away from a vacuum arc extinguish chamber, enabling an annular plane to face the arc extinguish chamber, collecting electromagnetic wave signals generated by discharge, and collecting N data points by a data collection system (a data collection card) at a sampling rate of 500 kHz;

step 2: adopting db4 wavelet basis functions to carry out 3-layer wavelet packet decomposition on the collected noisy electromagnetic wave signals, then selecting a fixed threshold value to carry out coefficient threshold value processing on the signals, and finally obtaining a de-noising signal through wavelet packet reconstruction;

and step 3: denoising signals are processed according to a formula:

performing S transformation to obtain a time-frequency spectrum, integrating the time-frequency spectrum along a time axis to obtain a marginal spectrum, and extracting a signal spectrum mean value, a signal spectrum variance and a signal spectrum bandwidth as frequency domain characteristic quantities through the marginal spectrum; respectively extracting the pulse times and the pulse amplitude mean values in the positive half period and the negative half period of 10 periods as time domain characteristic quantities;

and 4, step 4: an SVM regression prediction model (pre-training model) is adopted, firstly, the obtained frequency domain characteristic quantity and time domain characteristic quantity are subjected to characteristic dimension reduction and characteristic screening through Principal Component Analysis (PCA), and the obtained final characteristic quantity is used as the input of an SVM to predict the vacuum degree in the arc extinguish chamber.

The pre-training model is based on electromagnetic wave signals of a vacuum arc-extinguishing chamber with known vacuum degree in a working state, the electromagnetic wave signals are subjected to wavelet decomposition, the electromagnetic wave signals subjected to the wavelet decomposition are subjected to coefficient threshold processing and then are subjected to wavelet packet reconstruction to obtain de-noised signals, the de-noised signals are subjected to S transformation to obtain time-frequency spectrums, the time-frequency spectrums are integrated along a time axis to obtain marginal spectrums, frequency domain characteristic quantities and time domain characteristic quantities of the signals are extracted through the marginal spectrums, and model training is carried out by using the extracted frequency domain characteristic quantities and the extracted time domain characteristic quantities.

As shown in fig. 2, the pulse discharge occurs between the contact 6 and the shield cover 8 when the degree of vacuum is reduced, and as shown in fig. 3, the electromagnetic wave is detected by the loop antenna according to the radiation of the electromagnetic wave from the vacuum interrupter. When the vacuum degree in the vacuum arc extinguish chamber of the high-voltage vacuum circuit breaker is reduced, the insulation strength between the contact 6 and the shielding case 8 is reduced according to the Paschen law, and the breakdown voltage V is reduced_BDecreases when the potential difference DeltaU between the contact 6 and the shield 8 reaches the breakdown voltage V_BAt this time, a pulse discharge occurs between the contact and the shield case, resulting in a sudden change in potential and radiating electromagnetic waves outward.

According to the test, even if the power supply voltage U is changed, and the capacity of the high-voltage vacuum circuit breaker is changed, the frequency of the electromagnetic wave is 2-20 kHz.

With reference to fig. 4a to 5, the effectiveness of the present invention in determining the vacuum degree in the vacuum interrupter chamber by using the frequency spectrum mean, the variance, and the bandwidth as the frequency domain characteristic quantities and using the pulse signal amplitude mean and the pulse times as the time domain characteristic quantities is explained. At the sample frequency f_sCollecting electromagnetic wave signals discharged between the contact and the shielding cover under different vacuum degrees at 500 kHz. Fig. 4a shows electromagnetic wave signals when the pressure inside the vacuum arc-extinguishing chamber is respectively 0.26Pa, 110Pa, 150Pa and 180Pa, and it can be seen that a discharge signal exists only in the negative half period of the power frequency voltage at 0.26Pa, and as the pressure increases, the positive half period and the negative half period start to discharge, and at the same time, the number of discharge pulses also increases, and the pulse discharge amplitude increases, which illustrates that the characteristic quantity in the invention can well represent the discharge characteristics between the contact and the shield cover under different vacuum degrees. As shown in FIG. 4b, the S-transform used in the present invention includes variable factors to improve the time-frequency resolution of the electromagnetic wave signals, and FIG. 4c is a diagram of the followingThe electromagnetic wave signal marginal spectrogram obtained by the time-frequency spectrum proves that the frequency of the electromagnetic wave signal generated by pulse discharge is below 50 kHz.

As shown in figure 5, the method adopts a PCA-SVM regression prediction model to judge the vacuum degree in the vacuum arc-extinguishing chamber, is more suitable for the classification regression problem of small samples, has the root mean square error of only about 3Pa, can accurately reflect the vacuum degree in the vacuum arc-extinguishing chamber in real time, avoids the waste of manpower and material resources caused by periodic shutdown maintenance, can remind workers to find and process in time when the vacuum degree is reduced, timely eliminates the fault of the high-voltage vacuum circuit breaker caused by the reduction of the vacuum degree, avoids accidents, and ensures the safe and stable operation of a power system.

The invention adopts the loop antenna to remotely receive the electromagnetic wave signal generated by the discharge between the contact and the shielding case in the arc extinguish chamber, thereby avoiding the direct contact with the arc extinguish chamber, avoiding the need of a built-in sensor and changing the structure of the arc extinguish chamber;

the method improves the effectiveness of the extracted characteristic quantity of the electromagnetic wave signal generated by the discharge between the contact inside the arc extinguish chamber and the shielding case, obtains the signal marginal spectrum after S conversion, combines the signal frequency domain characteristic extracted based on the marginal spectrum and can directly represent the time domain characteristic vector of the discharge characteristic without counting the relationship between the discharge quantity and the phase, and the discharge frequency and the phase, greatly simplifies the characteristic quantity extraction process while well representing the original signal, and obviously improves the effectiveness of the characteristic quantity;

the method has high detection accuracy, the problem of SVM identification accuracy reduction caused by noise and redundancy of sample characteristic information is solved through principal component analysis, the formed characteristic vectors are independent of each other to form an orthogonal relation, the identification accuracy of a test sample is effectively improved, time is shortened, and the final PCA-SVM regression prediction model can accurately judge different vacuum degrees.