Data Acquisition and Analysis Method for Perceiving the Opening and Closing States of GCB Based on Acoustic and Vibration Signals

Abstract

A method for collecting and analyzing GCB opening and closing state perception data based on acoustic and vibration signals. It uses a high-frequency and high-sensitivity vibration sensor to collect signals, calculates Mel cepstral coefficients and short-time energy features, and based on segmented data frames, uses empirical mode decomposition and sample entropy methods to obtain feature vectors. Further, the OneClassSVM method is applied to domesticate and obtain a standard reference sample model. Taking this model as a reference, it judges whether each opening and closing operation is normal. The present invention realizes fast judgment under a small model; the adopted transient standard reference model for opening and closing is domesticated from N groups of signal feature vectors by using the OneClassSVM method, and has a powerful machine self-learning ability. It can continuously optimize the standard reference model based on the accumulation of later operation signal samples and continuously improve the accuracy of signal discrimination; the data discrimination result of the present invention has two output methods, local and remote, ensuring timely early warning of abnormal signals.

Description

Technical Field

[0001] The invention belongs to the technical field of on-line monitoring of high-voltage circuit breakers at the generator outlet, and particularly relates to a method for collecting and analyzing GCB switching state perception data based on acoustic and vibration signals. Background Art

[0002] GCB is a high-voltage circuit breaker installed between a generator and a main transformer, which can realize fast short-circuit protection of the generator-transformer unit. At the same time, it can increase the flexibility of power plant operation and dispatching, ensure the continuity, reliability and flexibility of auxiliary power supply, and can also complete the synchronization operation on the generator side. In recent years, abnormal switching of GCB has occurred successively in many power stations in China. Through analysis and research, the above failures are all caused by mechanical failures in the breaker operating link circuit. When and before the above mechanical failures occur, there are obvious abnormal signals in the transmission link, power supply current, etc.

[0003] At present, the relatively mature on-line monitoring methods for GCB are mainly the monitoring of the arc chamber temperature and pressure. The above parameters respond slowly to equipment failures and cannot give early warnings in time. Another monitoring method is to use limit sensors to monitor the switching state, that is, the states at the start and end of the operation. In addition, a method for monitoring the closing and opening states of a generator outlet circuit breaker (CN111157882A) obtains vibration displacement, velocity, acceleration, vibration frequency, damping, and attenuation coefficient through vibration sensors, and then obtains the switching closing or opening state of the output layer through multiple hidden layers of deep neural networks.

[0004] The transient process of switching reflects the response characteristics of different components, and the above three methods cannot identify the details thereof. The processing of the switching transient process has high requirements for real-time performance, so the data needs to be processed using a smaller model and local computing hardware. Therefore, how to realize the early perception and warning analysis of the GCB switching transient process under a smaller data model is an urgent problem to be solved. Summary of the Invention

[0005] In view of the technical problems existing in the background art, the method for collecting and analyzing GCB switching state perception data based on acoustic and vibration signals provided by the invention uses the model downloaded by the DSP local processor to directly judge each operation signal, greatly reducing the data processing workload and realizing fast judgment under a small model; the adopted switching transient standard reference model is obtained by domesticating N groups of signal feature vectors using the OneClassSVM method, and has powerful machine self-learning ability. On the basis of the accumulation of later operation signal samples, the standard reference model can be continuously optimized to continuously improve the accuracy of signal discrimination; the data discrimination result of the invention has two output modes: local and remote, ensuring the timely warning of abnormal signals.

[0006] To solve the above technical problems, the present invention adopts the following technical solutions to achieve:

[0007] A method for collecting and analyzing GCB opening and closing state perception data based on acoustic vibration signals, the steps are as follows:

[0008] Step 1, install a collection circuit for obtaining acoustic vibration signals: The acoustic vibration signals are collected by vibration sensors, and the vibration sensors are installed at the end of the toggle arm of the GCB breaker; the vibration sensors are electrically connected to the processor, the processor is communicatively connected to the remote server, and the processor is electrically connected to the alarm signal indicator light;

[0009] Step 2, establish a reference model:

[0010] Step 2.1: Establish a closing transient reference model;

[0011] Step 2.2: Establish an opening transient reference model;

[0012] Step 3, perform action signal analysis: After the opening transient reference model and the closing transient reference model are established, analyze and determine each opening and closing acoustic vibration signal;

[0013] Step 4, perform signal discrimination output.

[0014] Preferably, the vibration sensors include A-phase vibration sensors, B-phase vibration sensors and C-phase vibration sensors for collection. The A-phase vibration sensor is used to be installed at the end of the toggle arm of the A-phase GCB breaker, the B-phase vibration sensor is used to be installed at the end of the toggle arm of the B-phase GCB breaker, and the C-phase vibration sensor is used to be installed at the end of the toggle arm of the C-phase GCB breaker.

[0015] Preferably, the establishment method of step 2.1 is: record the average maximum value S of vibration during N closing processes max , select 85% of this value as the reference value, and record the moment T base_start , as the start time of the model data frame;

[0016] Perform the following operations on each signal:

[0017] Set the moving detection frame Signal move , Signal move with a length of 0.1 ms, Signal move average value less than S min , the S min represents the average minimum value of a vibration signal; record the moment T move when Signal base_end first appears, as the cut-off time of the model data frame;

[0018] Get a duration of T base_length, and a signal spectrum model data frame Signal with a certain amplitude window ;

[0019] Calculate the maximum value, minimum value, mean value and variance of the model data frame, and the feature numbers are: 1-4;

[0020] Calculate the 0-12 order Mel cepstrum parameters of the model data frame to obtain 13 groups of signals; then calculate the maximum value, minimum value, mean value and variance of these 13 groups of signals respectively to obtain 52 characteristic signals, and these characteristic signals are numbered in sequence as: 5-56;

[0021] Divide the model data frame Signal window into 10 segmentation data frames; perform variational mode decomposition on each segmentation data frame to obtain K components, and the expression is:

[0022]

[0023] u k represents K modal signals, ε represents the residual, and the permutation entropy V(u k ) of each mode is obtained, and 10K signals are obtained. The 10K signals are continuously numbered, and the numbers are: 57-(56+10K);

[0024] After the above steps, finally obtain N (56+10K)-dimensional V train model feature vectors, and apply OneClassSVM to train the outlier points of these N model feature vectors to obtain the closing transient standard reference model Model close .

[0025] Preferably, in step 2.2, collect N opening operation signals, and obtain the opening transient standard reference model Model open according to the closing signal processing method.

[0026] Preferably, the analysis process of step 3 is as follows: operate the circuit breaker to obtain 1 signal to be analyzed; if the signal to be analyzed reaches 85% S max , select this moment as the start time T test_start of the data frame to be measured;

[0027] Set the moving frame Frame move , with a length of 0.1 ms, and the average value of the moving frame is less than S min， ; take the moment when this frame first appears as the cut-off time T test_end of the data frame to be measured;

[0028] Calculate the length T test_length of the data frame to be measured to obtain the data frame to be measured TestSignal window ;

[0029] Calculate TestSignal window The maximum, minimum, mean and variance of , feature number: T1-T4;

[0030] Calculate TestSignal window Maximum, minimum, mean and variance of Mel-frequency cepstrum parameters of order 0 to 12, feature numbers: T5-T56;

[0031] TestSignal window Divide into ten segmentation data frames, perform variational mode decomposition, obtain K components and obtain the permutation entropy V(u k+1 ), feature number: T57-T(56+10K);

[0032] Finally, we get a (56+10K)-dimensional feature vector V to be tested. test , input the closing feature vector into the model close If it is a gate feature, the vector is input to the model open , make a prediction and get a score S.

[0033] Preferably, the judgment rule for signal output is:

[0034] TΔ sound =|T base_length -T test_length |,

[0035] The TΔ sound Indicates whether the signal is abnormal or not, T base_length Represents the standard duration set by the algorithm, T test_length Indicates the actual duration of a test signal; if the deviation exceeds the set threshold TΔ th , then output the local IO alarm signal and output the alarm signal to the server;

[0036] Set the model score threshold Threshhold. If S is greater than Threshhold, output a local IO alarm signal and output an alarm signal to the remote server; if S is less than Threshhold, output a local IO normal signal and output a normal signal to the remote server.

[0037] This patent can achieve the following beneficial effects:

[0038] 1. Compared with the prior art, the present invention adopts an integrated acoustic and vibration detection sensor to achieve the synchronous collection of sound signals and three-dimensional vibration signals, with richer data volume;

[0039] 2. Compared with the prior art, the integrated acoustic and vibration detection sensor adopted by the present invention is directly installed using the existing screw holes on the breaker toggle arm, which is simpler and more convenient for installation and removal. Meanwhile, it does not affect the safe and stable operation of the device to be detected itself;

[0040] 3. The model downloaded by the DSP on-site processor adopted by the present invention directly discriminates each operation signal, greatly reducing the data processing workload and achieving fast judgment under a small model;

[0041] 4. The opening and closing transient standard reference model adopted by the present invention is obtained by domesticating N groups of signal feature vectors using the OneClassSVM method, and has a powerful machine self-learning ability. Based on the accumulation of operation signal samples in the later stage, the standard reference model can be continuously optimized to continuously improve the accuracy of signal discrimination;

[0042] 5. The data discrimination result of the present invention has two output methods: on-site and remote, ensuring the timely early warning of abnormal signals;

[0043] 6. After the present invention is used, it can collect the acoustic and vibration signals of the mechanical transmission components of the GCB enclosed core electrical equipment, monitor early abnormal signals, optimize the equipment status evaluation algorithm, support the equipment status maintenance strategy, and check and handle equipment hidden dangers early. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The present invention will be further described below in conjunction with the drawings and embodiments:

[0045] Figure 1 is the signal acquisition and communication system diagram of the present invention;

[0046] Figure 2 is the A-phase acoustic vibration waveform diagram during the breaker closing process of the present invention;

[0047] Figure 3 is the B-phase acoustic vibration waveform diagram during the breaker closing process of the present invention;

[0048] Figure 4 is the C-phase acoustic vibration waveform diagram during the breaker closing process of the present invention;

[0049] Figure 5 is the A-phase acoustic vibration waveform diagram during the breaker opening process of the present invention;

[0050] Figure 6 is the B-phase acoustic vibration waveform diagram during the breaker opening process of the present invention;

[0051] Figure 7 is the C-phase acoustic vibration waveform diagram during the breaker opening process of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0052] Embodiment 1:

[0053] The preferred solution is as follows Figures 1 to 7 shown, a method for collecting and analyzing data on the opening and closing states of GCB based on acoustic vibration signals, the steps are as follows:

[0054] Step 1, install a collection circuit for obtaining acoustic vibration signals: The acoustic vibration signals are collected by vibration sensors, and the vibration sensors are installed at the end of the toggle arm of the GCB circuit breaker; the vibration sensors are electrically connected to the processor, the processor is communicatively connected to the remote server, and the processor is electrically connected to the alarm signal indicator light;

[0055] Specifically, the vibration sensors include A-phase vibration sensors, B-phase vibration sensors, and C-phase vibration sensors for collection. The A-phase vibration sensor is used to be installed at the end of the toggle arm of the A-phase GCB circuit breaker, the B-phase vibration sensor is used to be installed at the end of the toggle arm of the B-phase GCB circuit breaker, and the C-phase vibration sensor is used to be installed at the end of the toggle arm of the C-phase GCB circuit breaker. The installation methods of the vibration sensors include screw fastening, magnet adsorption, or hard glue connection. The processor is a 32-bit DSP processor.

[0056] The acoustic vibration signal sensor adopted in the present invention is of the X, Y, Z three-axis detection type, with a vibration range: ±200g, a vibration frequency range: 0.1Hz to 1600Hz, and a vibration sensitivity: 4mg. It is installed at the end of the toggle arm of the GCB circuit breaker through a solution of screw fastening, strong magnetism, and hard glue bonding.

[0057] The acoustic vibration signals are transmitted to the remote server through a 32-bit DSP processing unit. After a series of analysis and processing, a standard reference sample model is domesticated and sent back to the DSP processing unit. During the formal operation of the system, the data of each opening and closing acoustic vibration signal is directly processed by the DSP and compared with the standard reference sample model. If there are early warning and alarm signals, they can be displayed locally and sent to the remote server. The signal collection and communication system is shown in Figure 1 .

[0058] Step 2, establish a reference model:

[0059] Step 2.1: Establish a closing transient reference model;

[0060] The establishment method is as follows: Record the average maximum value S of vibration during N closing processes max , select 85% of this value as the reference value, and record the moment T base_start , as the start time of the model data frame;

[0061] Perform the following operations on each signal:

[0062] Set the moving detection frame Signal move , the length of this frame is 0.1ms, and the average value is less than S min , S minRepresents the average minimum value of a vibration signal, which is artificially set; records the moment T when this frame first appears base_end , as the cut-off time of the model data frame;

[0063] The length T of the model data frame base_length (This value is used as the template standard value and allows a certain deviation TΔsound); obtains the model data frame Signal window : Signal window Is a signal spectrum with a duration of T base_length and a certain amplitude.

[0064] The purpose of the above steps is to obtain a signal with a duration of T base_length (T base_end -T base_start ) and an amplitude between 85% S max and S min .

[0065] Calculate the maximum value, minimum value, mean value and variance of the model data frame, and the feature numbers are: 1 - 4;

[0066] Calculate the 0 - 12th order Mel cepstrum parameters of the model data frame to obtain 13 groups of signals. Then, calculate the maximum value, minimum value, mean value and variance of these 13 groups of signals respectively to obtain 52 characteristic signals, and these characteristic signals are numbered in sequence: 5 - 56;

[0067] Divide the model data frame Signal window into 10 segmented data frames; each segmented data frame is subjected to variational mode decomposition to obtain K components, and the expression is:

[0068]

[0069] u k represents K modal signals, ε represents the residual, and the permutation entropy V(u k ) of each mode is obtained, and 10K signals are obtained. These 10K signals continue to be numbered, numbered as: 57 - (56 + 10K);

[0070] After the above steps, finally obtain N (56 + 10K)-dimensional V train model feature vectors, and apply OneClassSVM to train these N model feature vectors for outliers to obtain the closing transient standard reference model Model close .

[0071] Step 2.2: Establish the opening transient reference model;

[0072] In Step 2.2, collect the opening action signals N times, and according to the closing signal processing method, obtain the opening transient standard reference model Model open .

[0073] Step 3: Conduct action signal analysis. After the opening transient reference model and the closing transient reference model are established, analyze and determine each opening and closing sound and vibration signal;

[0074] Operate the circuit breaker to obtain 1 signal to be analyzed; if the action signal to be analyzed reaches 85% S max , select this moment as the start time T of the data frame to be measured test_start ;

[0075] Set the moving frame Frame move , with a length of 0.1 ms, and the average value of the moving frame is less than S min ; Take the moment when this frame first appears as the cut-off time T of the data frame to be measured test_end ;

[0076] Calculate the length T of the data frame to be measured test_length , and obtain the data frame to be measured TestSignal window ;

[0077] Calculate the maximum value, minimum value, mean value and variance of TestSignal window , feature numbers: T1 - T4.

[0078] The final 56 + 10K signals of TestSignalwindow can be given different numbers.

[0079] Calculate the maximum value, minimum value, mean value and variance of the 0 - 12th order Mel cepstrum parameters of TestSignal window , feature numbers: T5 - T56;

[0080] Divide TestSignal window into ten segmented data frames, perform variational mode decomposition, obtain K components and obtain the permutation entropy V(u k+1 ) of each component mode, feature numbers: T57 - T(56 + 10K).

[0081] Finally, obtain 1 (56 + 10K)-dimensional feature vector V to be measured test , input the closing feature vector into the model Model close (input the opening feature vector into the model Model open ), perform prediction and obtain the score S.

[0082] In addition, after step 3, there are only the data frame to be measured and the moving data frame. The data frame to be measured and the moving data frame are two different signals. Among them, the moving data frame is a standard detection signal. After the device action acquires the original signal, the moving data frame is needed to detect the original signal, and finally a data frame to be measured is obtained.

[0083] Step 4: Perform signal discrimination and output.

[0084] The judgment rule for signal output is as follows:

[0085] TΔ sound = |T base_length - T test_length |

[0086] TΔ sound is a determination condition for whether the signal is abnormal. T base_length represents the standard duration set by the algorithm, and T test_length represents the actual duration of a test signal. The absolute value of the difference between the two times is TΔ sound . If this deviation exceeds the set threshold TΔ th , then a local IO alarm signal is output and an alarm signal is output to the server;

[0087] Set the model score threshold Threshhold. If S is greater than Threshhold, then a local IO alarm signal is output and an alarm signal is output to the remote server; if S is less than Threshhold, then a local IO normal signal is output and a normal signal is output to the remote server.

[0088] As Figures 2 - 7 shown, by using the data acquisition and analysis method provided by the present invention, a vibration sensor with high frequency and high sensitivity is used to collect signals, and waveform diagrams of the acoustic vibration signals during closing and opening are obtained. The present invention calculates the Mel cepstral coefficients and short-time energy characteristics on a relatively small model (usually a model below 10MB, suitable for embedded systems), and based on the segmented data frames, uses the methods of empirical mode decomposition and sample entropy to obtain feature vectors. The closing and opening states of the GCB are evaluated with these feature parameters. At the same time, an online monitoring and hierarchical alarm strategy is adopted to feedback abnormal information to the monitoring system and the big data system, optimize the equipment status evaluation, improve the rapid response ability of equipment faults, and support the equipment status maintenance strategy.

[0089] The above embodiments are only the preferred technical solutions of the present invention and should not be regarded as limitations to the present invention. The protection scope of the present invention should be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A method for collecting and analyzing data on the opening and closing states of a GCB based on acoustic and vibration signals, characterized in that It includes the following steps: Step 1, install an acquisition loop for obtaining acoustic and vibration signals: The acoustic and vibration signals are collected by vibration sensors, and the vibration sensors are installed at the end of the toggle arm of the GCB circuit breaker; the vibration sensors are electrically connected to the processor, the processor is communicatively connected to the remote server, and the processor is electrically connected to the alarm signal indicator light; Step 2, establish a reference model: Step 2.1: Establish a closing transient reference model; Step 2.2: Establish a tripping transient reference model; Step 3, conduct action signal analysis: After the tripping transient reference model and the closing transient reference model are established, analyze and determine each closing and tripping acoustic and vibration signal; Step 4, conduct signal discrimination and output.

2. The method for collecting and analyzing GCB switching state perception data based on acoustic vibration signals according to claim 1, characterized in that: The vibration sensors include an A-phase vibration sensor, a B-phase vibration sensor, and a C-phase vibration sensor for collection. The A-phase vibration sensor is used to be installed at the end of the toggle arm of the A-phase GCB circuit breaker, the B-phase vibration sensor is used to be installed at the end of the toggle arm of the B-phase GCB circuit breaker, and the C-phase vibration sensor is used to be installed at the end of the toggle arm of the C-phase GCB circuit breaker.

3. The method for acquiring and analyzing the GCB opening and closing state perception data based on the acoustic and vibration signals according to claim 1, wherein: The establishment method of Step 2.1 is as follows: record the average maximum value S of vibration during N closing processes max , select 85% of this value as the reference value and record it as time T base_start , which serves as the start time of the model data frame; Perform the following operations on each signal: Set the moving detection frame Signal move , Signal move with a length of 0.1 ms, Signal move whose average value is less than S min , where the S min represents the average minimum value of a vibration signal; record Signal move the moment T of the first occurrence base_end , as the cut-off time of the model data frame; Thus, a signal spectrum model data frame Signal with a duration of T is obtained base_length and a certain amplitude window ; Calculate the maximum value, minimum value, mean value, and variance of the model data frame, and the feature numbers are respectively: 1-4; Calculate the 0-12th order Mel cepstrum parameters of the model data frame to obtain 13 groups of signals; then respectively obtain the maximum value, minimum value, mean value, and variance of these 13 groups of signals to obtain 52 characteristic signals, and these characteristic signals are numbered in sequence as: 5-56; Divide the model data frame Signal window into 10 segmented data frames; each segmented data frame is subjected to variational mode decomposition to obtain K components, and the expression is: u k represents K modal signals, ε represents the residual, and the permutation entropy V(u k ) of each mode is obtained. 10K signals are obtained, and the 10K signals are numbered continuously as: 57 - (56 + 10K); After the above steps, finally, N V's with (56 + 10K) dimensions are obtained. train The model feature vectors are used to train outliers with OneClassSVM for these N model feature vectors to obtain the closing transient standard reference model Model. close .

4. The method for collecting and analyzing the GCB opening and closing state perception data based on the acoustic and vibration signals according to claim 3, wherein: In Step 2.2, collect the opening operation signals N times, and obtain the opening transient standard reference model Model according to the closing signal processing method open .

5. The method for acquiring and analyzing the GCB opening and closing state perception data based on the acoustic and vibration signals according to claim 1, wherein: The analysis process of Step 3 is as follows: Operate the circuit breaker to obtain 1 signal to be analyzed; if the action signal to be analyzed reaches 85% S max , select to analyze the moment when the action signal reaches 85% S max as the start time T of the data frame to be measured test_start ; Set a moving frame Frame move , with a length of 0.1 ms, and the average value of the moving frame is less than S min ; Use the moment when the moving frame first appears as the cut-off time T of the data frame to be measured test_end ; Calculate the length T of the data frame to be measured test_length , and obtain the data frame to be measured TestSignal window ; Calculate TestSignal window for maximum value, minimum value, mean value and variance, feature numbers: T1 - T4; Calculate TestSignal window The maximum, minimum, mean, and variance of the Mel cepstral coefficients of orders 0 to 12, feature numbers: T5 - T56; Divide TestSignal window into ten segmented data frames, perform variational mode decomposition, obtain K components and obtain the permutation entropy V(u k+1 ) of each component mode; Feature numbers: T57 - T(56 + 10K); Finally, a feature vector V to be measured with a dimension of (56 + 10K) is obtained. test , and the closing feature vector is input into the model Model. close , if it is a tripping feature, the vector is input into the model Model. open , and prediction is performed to obtain the score S.

6. The method for collecting and analyzing the GCB switching state perception data based on the acoustic and vibration signals according to claim 1, wherein: The judgment rule for signal output is: TΔ sound = |T base_length - T test_length |, The described TΔ sound represents a determination condition for whether the signal is abnormal, and T base_length represents the standard duration set by the algorithm, and T test_length represents the actual duration of a test signal; T test_length If the deviation exceeds the set threshold value TΔ th , a local IO alarm signal is output and an alarm signal is output to the server; Set the model score threshold Threshhold. If S is greater than Threshhold, output a local IO alarm signal and output an alarm signal to the remote server; if S is less than Threshhold, output a local IO normal signal and output a normal signal to the remote server.

Images