A Fault Detection Method and System for a Pumping Station Circuit Breaker

Abstract

The present invention relates to the technical field of power equipment fault diagnosis, and specifically relates to a method and system for detecting faults of a pump station circuit breaker, including the following steps: S1: Collect vibration signals below 50 Hz during the operation of the circuit breaker; S2: Preprocess the vibration data collected in S1; S3: Analyze the vibration data preprocessed in S2, extract characteristic parameters related to the low-frequency vibration of the circuit breaker; and form the extracted characteristic parameters into a characteristic vector; S4: Train the characteristic pattern training set to generate a circuit breaker fault pattern recognition model; S5: Convert the current vibration data into a real-time characteristic vector, and input it into the fault pattern recognition model generated in S4 to determine whether the circuit breaker is abnormal and generate a fault diagnosis report. With the present invention, through the accurate extraction of low-frequency vibration signals and the construction of characteristic vectors, combined with an optimized support vector machine fault pattern recognition model, high-precision real-time detection and reliable diagnosis of circuit breaker faults are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment fault diagnosis, and particularly to a fault detection method and system for a pump station circuit breaker. Background Art

[0002] The pump station circuit breaker is an important device for ensuring the safe operation of electricity in the pump station system. Its main function is to disconnect and protect the circuit to avoid damage to equipment caused by short circuits, overloads or other abnormal conditions. However, the mechanical components of the circuit breaker will experience problems such as wear, aging or loosening during long-term operation, which may cause mechanical failures or electrical abnormalities. Traditional circuit breaker maintenance methods mostly rely on manual regular inspections or simple electrical parameter monitoring, and it is difficult to detect early hidden dangers in a timely manner. Especially in critical continuous operation scenarios such as pump stations, traditional methods often have problems of low efficiency and untimely detection, which are prone to the accumulation of hidden failures and ultimately pose a serious threat to the safe operation of equipment.

[0003] Although some technical means based on vibration monitoring or data analysis have been introduced in the field of circuit breaker fault detection in the prior art, the following technical problems still need to be solved urgently: First, the monitored signal frequency bands are usually concentrated in the high-frequency range, and the low-frequency vibration signals, which are important features reflecting the mechanical component failures of the circuit breaker, are often not fully utilized; Second, there is a lack of organic combination among data preprocessing, feature extraction and fault mode recognition, resulting in insufficient reliability of the detection results; Third, the classification accuracy and robustness of the existing fault detection models still need to be improved when dealing with the mapping relationship between complex vibration features and fault states. Therefore, it is of great technical significance and application value to develop a detection method and system that can comprehensively analyze low-frequency vibration signals and achieve high-precision fault identification. Summary of the Invention

[0004] Based on the above purpose, the present invention provides a fault detection method and system for a pump station circuit breaker.

[0005] A fault detection method for a pump station circuit breaker includes the following steps:

[0006] S1: Install a low-frequency vibration sensor at a specified position of the pump station circuit breaker, and collect vibration signals below 50 Hz during the operation of the circuit breaker to form initial vibration data;

[0007] S2: Perform preprocessing on the initial vibration data collected in S1, including removing noise, filtering and removing abnormal data to obtain preprocessed vibration data;

[0008] S3: Analyze the vibration data after preprocessing in S2, extract the characteristic parameters related to the low-frequency vibration of the circuit breaker. The characteristic parameters include vibration amplitude fluctuation, low-frequency harmonic intensity, resonance frequency deviation, and nonlinear characteristics; and form a feature vector with the extracted characteristic parameters.

[0009] S4: Establish a characteristic pattern training set using historical circuit breaker operation data and historical fault data, and combine with the feature vector in S3. Use the support vector machine algorithm to train the characteristic pattern training set to generate a circuit breaker fault pattern recognition model.

[0010] S5: Convert the current vibration data into a real-time feature vector and input it into the fault pattern recognition model generated in S4 to determine whether there is an abnormality in the circuit breaker; when an abnormal pattern is detected, output the fault recognition result, record the corresponding characteristic parameters, and generate a fault diagnosis report.

[0011] Optionally, the specific steps of S1 include:

[0012] S11: Select the vibration signal sensitive area as the installation position of the sensor. The sensitive area is the support frame of the circuit breaker or the fixed connection point of the switch operating mechanism.

[0013] S12: At the installation position determined in S11, use threaded fasteners to fix the low-frequency vibration sensor to the surface of the circuit breaker, and use silicone to fill the gaps around the sensor.

[0014] S13: Simulate the vibration signal from 0 Hz to 50 Hz by connecting to a standard signal generator, and adjust the sensitivity and output amplitude of the low-frequency vibration sensor to ensure that the error of the vibration signal output by the low-frequency vibration sensor within the target frequency range is controlled within ±1%.

[0015] S14: Directly connect the vibration signal output by the low-frequency vibration sensor to a high-speed data recording device, and set the sampling frequency to 100 Hz; the collected vibration signals are continuously recorded in the form of amplitude and time series.

[0016] S15: Save the vibration signal collected in S14 in digital format as the initial vibration data.

[0017] Optionally, the specific steps of S2 include:

[0018] S21: Use the noise removal method based on wavelet transform to decompose the initial vibration data into multiple layers of wavelet coefficients, perform threshold filtering on the high-frequency coefficient part to remove the background noise caused by environmental interference, and retain the low-frequency signal components.

[0019] S22: Filter the vibration data processed by S21 using an FIR bandpass filter with a frequency range of 0.5 Hz to 50 Hz to remove non-target frequency components with frequencies lower than 0.5 Hz and higher than 50 Hz;

[0020] S23: Apply a statistical method to perform abnormality detection on the vibration signal processed by S22, calculate the mean and standard deviation of the signal amplitude, determine the data points whose amplitude exceeds the range of two times the standard deviation of the mean as abnormal data, and eliminate the abnormal data;

[0021] S24: The vibration signal obtained after removing the abnormal data in S23 is smoothed by a sliding average algorithm to eliminate small peaks caused by signal fluctuations.

[0022] Optionally, the S3 specifically includes:

[0023] S31: Divide the vibration data preprocessed by S2 into segments according to time windows, calculate the amplitude range of the vibration signal in each segment, and calculate the mean and standard deviation of the amplitude range of all time windows as characteristic parameters of vibration amplitude fluctuation;

[0024] S32: Perform fast Fourier transform on the vibration signal preprocessed by S2, extract the frequency components in the spectrum, and calculate the square sum of the amplitudes of each frequency component below 10 Hz as the characteristic parameter of the low-frequency harmonic intensity;

[0025] S33: Based on the spectrum analysis of S32, determine the frequency point with the largest amplitude in the vibration signal as the resonance frequency; compare the resonance frequency with the reference resonance frequency when the circuit breaker is working normally, and calculate the frequency offset as the characteristic parameter of the resonance frequency offset;

[0026] S34: Use Hilbert-Huang transform to perform nonlinear analysis on the vibration signal preprocessed by S2, decompose it to obtain several intrinsic mode functions IMF, and calculate the IMF energy proportion and kurtosis as characteristic parameters of the nonlinear characteristics of the vibration signal;

[0027] S35: Arrange the characteristic parameters extracted from S31 to S34 in a fixed order, including vibration amplitude fluctuation, low-frequency harmonic intensity, resonance frequency offset and nonlinear characteristics, to form a characteristic vector of four dimensions, where each dimension corresponds to a characteristic parameter.

[0028] Optionally, the S35 specifically includes:

[0029] S351: normalizing all feature parameters extracted in S31 to S34, converting the original value of each feature parameter into a relative value so that its range is unified into a fixed interval;

[0030] S352: Assign the normalized characteristic parameters to the four dimensions of the feature vector in a fixed order; the first dimension represents the comprehensive situation of the vibration amplitude fluctuation, which is calculated by combining the mean and standard deviation of the vibration amplitude fluctuation; the second dimension represents the low-frequency harmonic intensity, which is directly assigned by the normalized low-frequency harmonic intensity characteristic parameter; the third dimension represents the resonance frequency offset, which is assigned by the normalized resonance frequency offset characteristic; the fourth dimension is used to express the non-linear characteristics of the vibration signal, which is determined by weighting the normalized energy ratio and kurtosis parameter.

[0031] S353: Arrange the characteristic parameters of the above four dimensions in a predetermined order to form a four-dimensional feature vector, and each dimension of the feature vector represents the vibration amplitude fluctuation, low-frequency harmonic intensity, resonance frequency offset and non-linear characteristics respectively.

[0032] Optionally, the S4 specifically includes:

[0033] S41: Collect the historical operation data and historical fault data of the circuit breaker in the normal operation state and known fault states. The collected data includes the original records of the vibration signals and the corresponding operation condition labels, and the operation condition labels clearly mark the normal state and different fault types.

[0034] S42: Preprocess the historical data collected in S41, and use the same denoising, filtering and outlier removal methods as in S2 to ensure that the quality of the historical data is consistent with the real-time data.

[0035] S43: Process the historical data processed in S42 according to the method of extracting characteristic parameters in S3, extract parameters such as vibration amplitude fluctuation, low-frequency harmonic intensity, resonance frequency offset and non-linear characteristics, generate a feature vector consistent with S35, and correspond the feature vector with the operation condition label one by one to form a feature pattern training set.

[0036] S44: Divide the feature pattern training set generated in S43 into a training set and a validation set according to the ratio of 8:2, where 80% of the data is used to train the support vector machine model, and 20% of the data is used to test the model performance.

[0037] S45: Input the feature vector extracted in S43 into the support vector machine algorithm, and use the data in the training set and its corresponding operation condition label to train the support vector machine model so that it can recognize the mapping relationship between the specified feature pattern and the fault state of the circuit breaker.

[0038] S46: After completing the model training, verify the classification accuracy of the support vector machine model through the validation set. If the accuracy reaches the preset requirement, save the model as the circuit breaker fault pattern recognition model.

[0039] Optionally, S45 specifically includes:

[0040] S451: Define the training parameters of the support vector machine model, including the kernel function, penalty coefficient, and tolerance parameter;

[0041] S452: Extract feature vectors from the training set, and arrange the feature vectors of all samples in the order of samples to form an input data matrix; at the same time, extract the operating condition labels corresponding to each sample to construct an operating state label vector;

[0042] S453: Set the training objective of the support vector machine model, so that the model can find a classification hyperplane in the high-dimensional space, which can maximize the interval of the classification boundary and reduce classification errors at the same time;

[0043] S454: During the training process, use the kernel function to map the input data from the original feature space to the high-dimensional feature space; and in the high-dimensional space, through iterative optimization of the model's weight and bias parameters, construct an optimal classification hyperplane for distinguishing different operating states;

[0044] S455: Save the support vector machine model obtained by iterative optimization for identifying the mapping relationship between the input feature vector and the circuit breaker fault state.

[0045] Optionally, S46 specifically includes:

[0046] S461: Input the feature vectors in the validation set divided by S44 into the trained support vector machine model one by one, classify each input feature vector according to the classification decision function, and output the predicted operating condition label;

[0047] S462: Record the actual operating condition labels and the prediction results of the support vector machine model for each sample in the validation set to form a corresponding relationship table of actual labels and predicted labels;

[0048] S463: Statistically analyze the label correspondence table, calculate the ratio of the number of samples with correct model classification in the validation set to the total number of samples in the validation set as the classification accuracy of the model. The calculation formula is: , where represents the classification accuracy; represents the number of samples with correct classification; represents the total number of samples in the validation set;

[0049] S464: Compare the classification accuracy calculated in S463 with the preset classification accuracy threshold When the classification accuracy is greater than or equal to the classification accuracy threshold If yes, it is determined that the preset requirements are met, and the model that meets the requirements is saved as the circuit breaker fault mode recognition model; otherwise, return to S45 to re-adjust the model parameters and perform retraining.

[0050] Optionally, the S5 specifically includes:

[0051] S51: Collect vibration data in real time and perform preprocessing according to the same preprocessing method as in S2; subsequently, according to the method in S3, extract four characteristic parameters of vibration amplitude fluctuation, low-frequency harmonic intensity, resonance frequency offset, and nonlinear characteristics from the preprocessed real-time vibration data, and construct a real-time feature vector;

[0052] S52: Input the constructed real-time feature vector into the fault mode recognition model generated by S4, and use its classification decision function to determine whether the current vibration data belongs to the defined fault mode; if the output classification result is a fault mode, mark that there is an abnormality in the circuit breaker;

[0053] S53: Record the characteristic parameters extracted in S51 to clarify the specific characteristics of the abnormal vibration signal, including the vibration amplitude fluctuation range, low-frequency harmonic intensity value, resonance frequency offset amount, and nonlinear characteristic index;

[0054] S54: Generate a fault diagnosis report based on the abnormal characteristic parameters recorded in S53 and the model classification result. The report includes the specific values of the abnormal characteristic parameters, the classification result of the fault mode, and the fault type.

[0055] A pump station circuit breaker fault detection system for implementing the above-mentioned pump station circuit breaker fault detection method, including the following modules: Vibration signal acquisition module: A low-frequency vibration sensor installed at a specified part of the pump station circuit breaker, used to collect vibration signals below 50 Hz in real time, and transmit the collected vibration signals to the signal processing module in digital form; Signal processing module: Connected to the vibration signal acquisition module, used to preprocess the collected vibration signals. The preprocessing includes removing noise, filtering, and removing abnormal data to generate processed vibration data;

[0056] Feature extraction module: Connected to the signal processing module, used to analyze the preprocessed vibration data, extract characteristic parameters of the low-frequency vibration of the circuit breaker, including vibration amplitude fluctuation, low-frequency harmonic intensity, resonance frequency offset, and nonlinear characteristics, and construct the extracted characteristic parameters into a feature vector;

[0057] Fault mode recognition module: Connected to the feature extraction module, used to receive the feature vector, and train it in combination with the operation data and historical fault data of the historical circuit breaker, and then generate a circuit breaker fault mode recognition model, used to determine whether the current feature vector belongs to the preset fault mode, and output the fault detection result;

[0058] Data recording and diagnosis module: Connected to the fault mode recognition module, it is used to record the fault detection results output by the fault mode recognition module and generate a fault diagnosis report, which includes the specific values of abnormal characteristic parameters, the classification results of fault modes, and the fault types.

[0059] Advantages of the present invention:

[0060] In the present invention, through the real-time acquisition, preprocessing, and feature extraction of low-frequency vibration signals, the key vibration characteristics during the operation of the circuit breaker can be comprehensively captured, including vibration amplitude fluctuations, low-frequency harmonic intensity, resonance frequency offset, and nonlinear characteristics. Combining data standardization processing methods and feature vector construction techniques, the transformation of vibration signals from raw data to high-precision characteristic parameters is realized, ensuring the consistency and reliability of the input data. These key technologies make the characterization of vibration signals more accurate, laying a data foundation for the accuracy of fault identification.

[0061] In the present invention, by using the support vector machine algorithm to train the fault mode recognition model and combining classification optimization and kernel function mapping techniques, the mapping relationship between the low-frequency vibration characteristics of the circuit breaker and different fault states can be accurately identified. At the same time, through the verification set, the classification accuracy and error level of the model are strictly verified to ensure the robustness and generalization ability of the fault recognition model. The overall technical solution effectively improves the real-time performance and accuracy of fault detection, providing technical support for the safe operation and precise maintenance of the circuit breaker. Description of the drawings

[0062] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only those of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0063] Figure 1 Schematic diagram of the circuit breaker fault detection method according to the embodiment of the present invention;

[0064] Figure 2 Schematic diagram of the circuit breaker fault detection system according to the embodiment of the present invention. Detailed implementation manners

[0065] The following will describe the present invention in detail with reference to the drawings and specific embodiments. At the same time, it should be noted here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments. For some well-known technologies, those skilled in the art can also adopt other alternative methods for implementation; and the drawing part is only for more specific description of the embodiments, and is not intended to specifically limit the present invention.

[0066] As shown Figure 1 in the figure, a method for detecting faults in a pump station circuit breaker includes the following steps:

[0067] S1: Install a low-frequency vibration sensor at a specified position of the pump station circuit breaker, collect vibration signals below 50 Hz during the operation of the circuit breaker, and form initial vibration data;

[0068] S2: Preprocess the initial vibration data collected in S1, including removing noise, filtering, and eliminating abnormal data, to obtain preprocessed vibration data;

[0069] S3: Analyze the preprocessed vibration data in S2, extract characteristic parameters related to the low-frequency vibration of the circuit breaker. The characteristic parameters include vibration amplitude fluctuation, low-frequency harmonic intensity, resonance frequency offset, and nonlinear characteristics; and form a feature vector with the extracted characteristic parameters;

[0070] S4: Use historical circuit breaker operation data and historical fault data to establish a feature pattern training set, and combine with the feature vector in S3. Adopt the support vector machine algorithm to train the feature pattern training set to generate a circuit breaker fault pattern recognition model;

[0071] S5: Convert the current vibration data into a real-time feature vector and input it into the fault pattern recognition model generated in S4 to determine whether there is an abnormality in the circuit breaker; when an abnormal pattern is detected, output a fault recognition result, record the corresponding characteristic parameters, and generate a fault diagnosis report.

[0072] S1 specifically includes:

[0073] S11: Select a vibration signal sensitive area as the installation position of the sensor. The sensitive area is the support frame of the circuit breaker or the fixed connection point of the switch operating mechanism to ensure that the low-frequency vibration signal can be effectively transmitted to the sensor;

[0074] S12: At the installation position determined in S11, use threaded fasteners to fix the low-frequency vibration sensor to the surface of the circuit breaker, and use silicone to fill the gap around the sensor to enhance the fixing stability and signal transmission efficiency of the sensor;

[0075] S13: Simulate vibration signals from 0 Hz to 50 Hz by connecting to a standard signal generator, adjust the sensitivity and output amplitude of the low-frequency vibration sensor to ensure that the error of the vibration signal output by the low-frequency vibration sensor within the target frequency range is controlled within ±1%;

[0076] S14: directly connect the vibration signal output by the low-frequency vibration sensor to a high-speed data recording device, and set the sampling frequency to 100 Hz to ensure accurate collection of vibration signals below 50 Hz; the collected vibration signal is continuously recorded in the form of amplitude and time series;

[0077] S15: Save the vibration signal collected by S14 in digital format as initial vibration data. The initial vibration data includes the amplitude, frequency and timestamp information of the vibration, which provides a basis for the subsequent processing and feature extraction of vibration data. The above steps ensure that the sensor can accurately capture low-frequency vibration signals by clarifying the installation position, adopting a reliable fixing method and calibration technology. The vibration signal is directly collected to a high-speed recording device and stored in a digital format to avoid loss and distortion in signal transmission, providing a high-quality vibration data basis for subsequent steps, and significantly improving the accuracy and reliability of fault detection.

[0078] S2 specifically includes:

[0079] S21: using a noise removal method based on wavelet transform, the initial vibration data is decomposed into multiple layers of wavelet coefficients, and threshold filtering is performed on the high-frequency coefficient part to remove the background noise caused by environmental interference and retain the low-frequency signal component;

[0080] S22: Use a FIR (finite impulse response) bandpass filter with a frequency range of 0.5 Hz to 50 Hz to filter the vibration data processed by S21 to remove non-target frequency components with frequencies below 0.5 Hz and above 50 Hz, ensuring that the retained vibration signal is concentrated in the frequency band related to the low-frequency vibration of the circuit breaker;

[0081] S23: Apply a statistical method to perform anomaly detection on the vibration signal processed by S22, calculate the mean and standard deviation of the signal amplitude, determine the data points whose amplitude exceeds the range of two times the standard deviation of the mean as abnormal data, and remove the abnormal data to ensure the authenticity and continuity of the vibration data;

[0082] S24: The vibration signal obtained after removing abnormal data in S23 is smoothed using a sliding average algorithm to eliminate small spikes caused by signal fluctuations, further improving the stability and accuracy of the data; the above steps remove environmental noise through wavelet transform to effectively reduce the impact of background interference on low-frequency vibration signals; use a bandpass filter to accurately extract the vibration signal of the target frequency band to enhance the effective components of the signal; by removing abnormal data and smoothing, the continuity and credibility of the vibration data are improved, providing high-quality preprocessing data for subsequent feature extraction, significantly improving the accuracy and robustness of fault detection.

[0083] S3 specifically includes:

[0084] S31: Segment the vibration data preprocessed in S2 according to time windows, calculate the amplitude range (maximum value minus minimum value) of the vibration signal within each segment, and calculate the mean and standard deviation of the amplitude ranges of all time windows, which are used as the characteristic parameters of the vibration amplitude fluctuation;

[0085] The mean of the vibration amplitude fluctuation, and the formula is: ;

[0086] The standard deviation of the vibration amplitude fluctuation, and the formula is: , where and respectively represent the maximum and minimum values of the vibration signal within the th time window; represents the total number of time windows; represents the mean of the vibration amplitude fluctuation; represents the standard deviation of the vibration amplitude fluctuation;

[0087] S32: Perform a fast Fourier transform on the vibration signal preprocessed in S2, extract the frequency components in the spectrum, and calculate the sum of the squared amplitudes of the frequency components below 10 Hz, which is used as the characteristic parameter of the low-frequency harmonic intensity to characterize the low-frequency energy characteristics of the circuit breaker vibration signal; the specific calculation formula is: , where represents the low-frequency harmonic intensity; represents the frequency corresponding to the vibration signal amplitude;

[0088] S33: Based on the spectrum analysis in S32, determine the frequency point with the largest amplitude in the vibration signal as the resonance frequency; compare this resonance frequency with the reference resonance frequency when the circuit breaker is operating normally, and calculate the frequency offset, which is used as the characteristic parameter of the resonance frequency offset; the calculation formula for the offset of the resonance frequency is: , where represents the offset of the resonance frequency; represents the current resonance frequency, that is, the frequency point with the largest amplitude in the vibration signal; represents the reference resonance frequency;

[0089] S34: Use the Hilbert-Huang transform (HHT) to perform nonlinear analysis on the vibration signal preprocessed in S2, decompose it into several intrinsic mode functions IMF, and calculate the IMF energy ratio and kurtosis, which are used as the characteristic parameters of the nonlinear characteristics of the vibration signal; the formula for the IMF energy ratio is: ; the formula for the IMF kurtosis is: ; where represents the energy ratio of the th IMF; represents the The kurtosis of an IMF; and respectively represent the start time and end time of the signal; is the signal duration; is the total number of IMFs obtained by decomposition;

[0090] S35: Arrange the characteristic parameters extracted in S31 to S34 in a fixed order, including vibration amplitude fluctuation, low-frequency harmonic intensity, resonance frequency offset, and nonlinear characteristics, to form a characteristic vector in four dimensions, with each dimension corresponding to a characteristic parameter, as the characteristic expression of the low-frequency vibration signal of the circuit breaker.

[0091] S35 specifically includes:

[0092] S351: Normalize all the characteristic parameters extracted in S31 to S34, convert the original value of each characteristic parameter into a relative value, and unify its range to a fixed interval; during the normalization process, by comparing the minimum value and maximum value of the characteristic parameter, map the actual value of the characteristic parameter to the standard range to ensure that the dimensions of different characteristic parameters are consistent for subsequent operations; S352: Assign the normalized characteristic parameters to the four dimensions of the characteristic vector in a fixed order; the first dimension represents the comprehensive situation of vibration amplitude fluctuation, calculated by combining the mean and standard deviation of the vibration amplitude fluctuation; the second dimension represents the low-frequency harmonic intensity, directly assigned by the normalized low-frequency harmonic intensity characteristic parameter; the third dimension represents the resonance frequency offset, assigned by the normalized resonance frequency offset characteristic; the fourth dimension is used to express the nonlinear characteristics of the vibration signal, determined by weighting the normalized energy ratio and kurtosis parameters;

[0093] S353: Arrange the characteristic parameters in the above four dimensions in a predetermined order to form a four-dimensional characteristic vector, with each dimension of the characteristic vector representing vibration amplitude fluctuation, low-frequency harmonic intensity, resonance frequency offset, and nonlinear characteristics respectively; and perform a consistency check on the constructed characteristic vector to verify whether the normalized range of the characteristic parameters meets the standard and whether the parameters in each dimension are correctly mapped to the corresponding characteristic categories to ensure the accuracy of the characteristic vector;

[0094] The calculation steps for forming the characteristic vector in four dimensions are as follows:

[0095] First, normalize each characteristic parameter, and the formula is as follows: , where represents the original value of the characteristic parameter; and are respectively the minimum value and maximum value of the characteristic parameter in the sample dataset; represents the normalized characteristic value;

[0096] Then, the normalized characteristic parameters are assigned to the four dimensions of the feature vector in a specific order, specifically including: the first dimension is , representing the mean of the vibration amplitude fluctuation and the standard deviation average, and the expression is: ; the second dimension is , representing the low-frequency harmonic intensity, and the expression is: ; where is the normalized low-frequency harmonic intensity;

[0097] The third dimension is , representing the resonance frequency offset , and the expression is: ; where is the normalized vibration frequency offset;

[0098] The fourth dimension is , representing the comprehensive index of the nonlinear characteristics, and is obtained by calculating the weighted average of the energy ratio and the kurtosis , and the formula is: , where and are the weights of the energy ratio and the kurtosis, satisfying ; is the normalized energy ratio; is the normalized kurtosis;

[0099] Next, arrange the four feature dimensions in order to form a four-dimensional feature vector , expressed as: ; finally, verify the constructed feature vector to ensure that there are no omissions or redundancies between the feature dimensions; the verification methods include checking whether the normalization range conforms to , and whether the features of each dimension are correctly mapped to the corresponding parameters.

[0100] In S4, generating the breaker fault mode recognition model specifically includes:

[0101] S41: Collect the historical operation data and historical fault data of the breaker in the normal operation state and known fault states. The collected data includes the original records of the vibration signals and the corresponding operation condition labels, and the operation condition labels clearly mark the normal state and different fault types;

[0102] S42: Preprocess the historical data collected in S41. According to the same denoising, filtering, and outlier removal methods as in S2, ensure that the quality of the historical data is consistent with the real-time data, and ensure that the data can be used for subsequent feature extraction and model training;

[0103] S43: Process the historical data after S42 according to the method of extracting feature parameters in S3, extract parameters such as vibration amplitude fluctuation, low-frequency harmonic intensity, resonance frequency offset, and nonlinear characteristics, generate a feature vector consistent with S35, and correspond the feature vector with the operating condition label one by one to form a feature pattern training set;

[0104] S44: Divide the feature pattern training set generated in S43 into a training set and a validation set according to a ratio of 8:2, where 80% of the data is used to train the support vector machine model and 20% of the data is used to test the model performance;

[0105] S45: Input the feature vector extracted in S43 into the support vector machine algorithm, and use the data in the training set and its corresponding operating condition label to train the support vector machine model so that it can identify the mapping relationship between the specified feature pattern and the circuit breaker fault state;

[0106] S46: After completing the model training, verify the classification accuracy of the support vector machine model through the validation set. If the accuracy reaches the preset requirement, save the model as the circuit breaker fault mode recognition model for subsequent real-time data fault detection.

[0107] S45 specifically includes:

[0108] S451: Define the training parameters of the support vector machine model, including the kernel function, penalty coefficient, and tolerance parameter; among them, the kernel function is used to map the input feature vector to a high-dimensional space; the penalty coefficient is used to balance the model's tolerance for training errors and the model complexity; the tolerance parameter determines the convergence accuracy of the model during optimization;

[0109] S452: Extract the feature vector from the training set, and arrange the feature vectors of all samples in the order of samples to form an input data matrix; at the same time, extract the operating condition label corresponding to each sample to construct an operating state label vector; where the input data matrix represents the feature information of the sample, and the operating state label vector represents the operating state to which each sample belongs;

[0110] S453: Set the training objective of the support vector machine model so that the model can find a classification hyperplane in the high-dimensional space, and this classification hyperplane can maximize the interval of the classification boundary while reducing classification errors; for this purpose, the training process considers both the distribution of input features and the classification requirements of sample labels;

[0111] S454: During the training process, use the kernel function to map the input data from the original feature space to the high-dimensional feature space; and in the high-dimensional space, through iterative optimization of the model's weight and bias parameters, construct an optimal classification hyperplane for distinguishing different operating states;

[0112] S455: Save the support vector machine model parameters obtained by iterative optimization, including the model weights, biases, and support vector sets, for identifying the mapping relationship between the input feature vectors and the breaker fault states.

[0113] The calculation steps for training the breaker fault mode recognition model are as follows:

[0114] Construct the input feature matrix and label vector: Arrange the feature vectors in the training set according to the samples to form the input feature matrix, defined as , where each feature vector contains four feature dimensions; at the same time, extract the operating condition labels of each sample in the training set to form the label vector , where is the corresponding operating state label, taking values of positive / negative classification or multi-class categories;

[0115] Optimize the support vector machine objective function: Use the input feature matrix and the label vector to train the model according to the following optimization objective function: ; Constraint conditions: ; ; , where, is the weight vector of the classification hyperplane, is the bias, is the feature vector after kernel function mapping, is the slack variable, is the penalty coefficient;

[0116] Kernel function mapping and classification hyperplane construction: Use the kernel function to map the input feature vectors to a high-dimensional feature space, and the kernel function is defined as: ; Calculate the classification hyperplane parameters and according to the optimization results to form the classification decision function: ; The above steps ensure the training efficiency and accuracy of the model by clarifying the initialization parameters and optimization objectives of the support vector machine; construct the corresponding relationship between the input features and the operating condition labels to provide a clear data input-output logic for the model; enhance the classification ability of the model in the complex feature space through kernel function mapping, and the finally generated support vector machine model can accurately identify the correlation between the specific feature patterns of the breaker and the fault states, providing a reliable classification basis for subsequent real-time detection.

[0117] S46 specifically includes:

[0118] S461: Input the feature vectors in the validation set divided in S44 into the trained support vector machine model one by one, classify each input feature vector according to the classification decision function, and output the predicted operating condition labels.

[0119] S462: Record the actual operating condition labels and the prediction results of the support vector machine model for each sample in the validation set, form a correspondence table of actual labels and predicted labels for subsequent calculation of classification accuracy and error level.

[0120] S463: Statistically analyze the label correspondence table, calculate the ratio of the number of samples with correct model classification in the validation set to the total number of samples in the validation set as the classification accuracy of the model. The calculation formula is: , where represents the classification accuracy; represents the number of samples with correct classification; represents the total number of samples in the validation set;

[0121] S464: Compare the classification accuracy calculated in S463 with the preset classification accuracy threshold . When the classification accuracy is greater than or equal to the classification accuracy threshold , it is determined that the preset requirements are met, and the model that meets the requirements is saved as the circuit breaker fault mode recognition model; otherwise, return to S45 to re-adjust the model parameters and perform re-training; by setting the classification accuracy threshold, ensure that the accuracy of the model reaches the expected goal, and perform re-training optimization when the requirements are not met to improve the robustness and reliability of the model; the finally generated fault mode recognition model has high accuracy and low error level, providing a reliable classification basis for subsequent real-time fault detection.

[0122] S5 specifically includes:

[0123] S51: Real-time collect vibration data and perform preprocessing according to the same preprocessing method as in S2; subsequently, according to the method in S3, extract four characteristic parameters of vibration amplitude fluctuation, low-frequency harmonic intensity, resonance frequency offset, and nonlinear characteristics from the preprocessed real-time vibration data to construct a real-time feature vector.

[0124] S52: Input the constructed real-time feature vector into the fault mode recognition model generated in S4, and use its classification decision function to determine whether the current vibration data belongs to the defined fault mode; if the output classification result is a fault mode, mark that there is an abnormality in the circuit breaker.

[0125] S53: Record the characteristic parameters extracted in S51 to clarify the specific characteristics of the abnormal vibration signal, including the vibration amplitude fluctuation range, low-frequency harmonic intensity value, resonance frequency offset, and nonlinear characteristic index, for subsequent fault cause analysis;

[0126] S54: Based on the abnormal characteristic parameters recorded in S53 and the model classification results, generate a fault diagnosis report, which includes the specific values of the abnormal characteristic parameters, the classification results of the fault modes, and the fault types; and store the diagnosis report for subsequent maintenance and reference; the above steps quickly determine whether there is an abnormality in the circuit breaker through the classification function of the fault mode recognition model; record the abnormal characteristic parameters and generate a diagnosis report, providing detailed basis for fault location and maintenance, realizing real-time and accurate detection and record management of circuit breaker faults, and significantly improving maintenance efficiency and system operation reliability.

[0127] As Figure 2 shown, a fault detection system for a pumping station circuit breaker is used to implement the above-mentioned fault detection method for a pumping station circuit breaker, and includes the following modules:

[0128] Vibration signal acquisition module: A low-frequency vibration sensor installed at a specified part of the pumping station circuit breaker, used to collect vibration signals below 50 Hz in real time, and transmit the collected vibration signals to the signal processing module in a digital form;

[0129] Signal processing module: Connected to the vibration signal acquisition module, used to preprocess the collected vibration signals, and the preprocessing includes removing noise, filtering, and removing abnormal data to generate processed vibration data;

[0130] Feature extraction module: Connected to the signal processing module, used to analyze the preprocessed vibration data, extract the characteristic parameters of the low-frequency vibration of the circuit breaker, including vibration amplitude fluctuation, low-frequency harmonic intensity, resonance frequency offset, and nonlinear characteristics, and construct the extracted characteristic parameters into a feature vector;

[0131] Fault mode recognition module: Connected to the feature extraction module, used to receive the feature vector, and train it in combination with the operation data and historical fault data of the historical circuit breaker, and then generate a circuit breaker fault mode recognition model, used to determine whether the current feature vector belongs to a preset fault mode, and output a fault detection result;

[0132] Data recording and diagnosis module: Connected to the fault mode recognition module, used to record the fault detection result output by the fault mode recognition module and generate a fault diagnosis report, which includes the specific values of the abnormal characteristic parameters, the classification results of the fault modes, and the fault types.

[0133] The present invention encompasses any alternatives, modifications, equivalent methods, and solutions that are made within the spirit and scope of the present invention. For the purpose of enabling the public to have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention. However, those skilled in the art can fully understand the present invention even without the description of these details. Additionally, well-known methods, processes, procedures, components, and circuits, etc. are not described in detail in order to avoid unnecessary confusion to the essence of the present invention.

[0134] The above description is only a preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art, several improvements and refinements can be made without departing from the principle of the present invention, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A pump station circuit breaker fault detection method, characterized in that: The following steps are involved: S1: Install low-frequency vibration sensors at designated locations of the pump station circuit breakers to collect vibration signals below 50 Hz during the operation of the circuit breakers to form initial vibration data; S2: preprocessing the initial vibration data collected by S1, including removing noise, filtering and eliminating abnormal data, to obtain preprocessed vibration data; S3: Analyze the vibration data preprocessed by S2, extract characteristic parameters related to the low-frequency vibration of the circuit breaker, the characteristic parameters include vibration amplitude fluctuation, low-frequency harmonic intensity, resonance frequency offset and nonlinear characteristics; and construct the extracted characteristic parameters into a characteristic vector; The S3 specifically includes: S31: Divide the vibration data preprocessed by S2 into segments according to time windows, calculate the amplitude range of the vibration signal in each segment, and calculate the mean and standard deviation of the amplitude range of all time windows as characteristic parameters of vibration amplitude fluctuation; S32: Perform fast Fourier transform on the vibration signal preprocessed by S2, extract the frequency components in the spectrum, and calculate the square sum of the amplitudes of each frequency component below 10 Hz as the characteristic parameter of the low-frequency harmonic intensity; S33: Based on the spectrum analysis of S32, determine the frequency point with the largest amplitude in the vibration signal as the resonance frequency; compare the resonance frequency with the reference resonance frequency when the circuit breaker is working normally, and calculate the frequency offset as the characteristic parameter of the resonance frequency offset; S34: Use Hilbert-Huang transform to perform nonlinear analysis on the vibration signal preprocessed by S2, decompose it to obtain several intrinsic mode functions IMF, and calculate the IMF energy proportion and kurtosis as characteristic parameters of the nonlinear characteristics of the vibration signal; S35: Arrange the characteristic parameters extracted from S31 to S34 in a fixed order, including vibration amplitude fluctuation, low-frequency harmonic intensity, resonance frequency shift and nonlinear characteristics, to form a characteristic vector of four dimensions, each dimension corresponding to a characteristic parameter; S4: Establish a feature pattern training set using historical circuit breaker operation data and historical fault data, and combine it with the feature vector in S3 to train the feature pattern training set using the support vector machine algorithm to generate a circuit breaker fault pattern recognition model; S5: Convert the current vibration data into a real-time feature vector and input it into the fault pattern recognition model generated by S4 to determine whether the circuit breaker is abnormal; when an abnormal pattern is detected, output the fault recognition result and record the corresponding feature parameters, and generate a fault diagnosis report.

2. A pump station circuit breaker fault detection method according to claim 1, characterized in that: The S1 specifically includes: S11: Selecting a vibration signal sensitive area as the installation position of the sensor, wherein the sensitive area is a support frame of the circuit breaker or a fixed connection point of a switch operating mechanism; S12: At the installation position determined in S11, use threaded fasteners to fix the low-frequency vibration sensor to the surface of the circuit breaker, and use silicone to fill the gap around the sensor; S13: Connect a standard signal generator to simulate a vibration signal of 0 Hz to 50 Hz, adjust the sensitivity and output amplitude of the low-frequency vibration sensor, and ensure that the error of the vibration signal output by the low-frequency vibration sensor within the target frequency range is controlled within ±1%; S14: directly connect the vibration signal output by the low-frequency vibration sensor to a high-speed data recording device, and set the sampling frequency to 100 Hz; the collected vibration signal is continuously recorded in the form of amplitude and time series; S15: The vibration signal collected in S14 is saved in a digital format as initial vibration data.

3. A pump station circuit breaker fault detection method according to claim 1, characterized in that: The S2 specifically includes: S21: using a noise removal method based on wavelet transform, the initial vibration data is decomposed into multiple layers of wavelet coefficients, and threshold filtering is performed on the high-frequency coefficient part to remove the background noise caused by environmental interference and retain the low-frequency signal component; S22: Filter the vibration data processed by S21 using an FIR bandpass filter with a frequency range of 0.5 Hz to 50 Hz to remove non-target frequency components with frequencies lower than 0.5 Hz and higher than 50 Hz; S23: Apply a statistical method to perform abnormality detection on the vibration signal processed by S22, calculate the mean and standard deviation of the signal amplitude, determine the data points whose amplitude exceeds the range of two times the standard deviation of the mean as abnormal data, and eliminate the abnormal data; S24: The vibration signal obtained after removing the abnormal data in S23 is smoothed by a sliding average algorithm to eliminate small peaks caused by signal fluctuations.

4. A pump station circuit breaker fault detection method according to claim 1, characterized in that: The S35 specifically includes: S351: performing normalization processing on all characteristic parameters extracted in S31 to S34, converting the original value of each characteristic parameter into a relative value so that its range is unified into a fixed interval; S352: in a fixed order, the normalized characteristic parameters are assigned to the four dimensions of the characteristic vector; the first dimension represents the comprehensive situation of the vibration amplitude fluctuation, which is calculated by combining the mean and standard deviation of the vibration amplitude fluctuation; the second dimension represents the low-frequency harmonic intensity, which is directly assigned by the normalized low-frequency harmonic intensity characteristic parameter; the third dimension represents the resonance frequency offset, which is assigned by the normalized resonance frequency offset characteristic; the fourth dimension is used to express the nonlinear characteristics of the vibration signal, which is determined by weighting the normalized energy proportion and kurtosis parameter; S353: Arrange the characteristic parameters of the above four dimensions in a predetermined order to form a four-dimensional characteristic vector, where each dimension of the characteristic vector represents the vibration amplitude fluctuation, the low-frequency harmonic intensity, the resonance frequency offset and the nonlinear characteristic.

5. A pump station circuit breaker fault detection method according to claim 1, characterized in that: The S4 specifically includes: S41: Collecting historical operation data and historical fault data of the circuit breaker in normal operation and known fault states, wherein the collected historical operation data and historical fault data include original records of vibration signals and corresponding operation condition labels, and the operation condition labels clearly mark normal states and different fault types; S42: pre-process the historical data collected in S41, using the same denoising, filtering and outlier removal methods as S2 to ensure that the quality of the historical data is consistent with the real-time data; S43: The historical data processed by S42 is processed according to the method of extracting characteristic parameters in S3, and vibration amplitude fluctuation, low-frequency harmonic intensity, resonance frequency offset and nonlinear characteristic parameters are extracted to generate a characteristic vector consistent with S35, and the characteristic vector is matched with the operating condition label one by one to form a characteristic pattern training set; S44: Divide the feature pattern training set generated by S43 into a training set and a validation set in a ratio of 8:2, where 80% of the data is used to train the support vector machine model and 20% of the data is used to test the model performance; S45: inputting the feature vector extracted in S43 into a support vector machine algorithm, and training the support vector machine model using the data of the training set and its corresponding operating condition labels, so that the support vector machine model can identify the mapping relationship between the specified feature pattern and the circuit breaker fault state; S46: After completing the model training, the classification accuracy of the support vector machine model is verified through the verification set. If the accuracy meets the preset requirements, the model is saved as a circuit breaker fault mode recognition model.

6. A pump station circuit breaker fault detection method according to claim 5, characterized in that: The S45 specifically includes: S451: define training parameters of the support vector machine model, including kernel function, penalty coefficient and tolerance parameter; S452: extracting feature vectors from the training set, and arranging the feature vectors of all samples into an input data matrix in the order of the samples; at the same time, extracting the operating condition label corresponding to each sample to construct an operating state label vector; S453: Setting a training objective of the support vector machine model so that the model can find a classification hyperplane in the high-dimensional space, which can maximize the interval of the classification boundary and reduce the classification error; S454: During the training process, the kernel function is used to map the input data from the original feature space to the high-dimensional feature space; and in the high-dimensional space, an optimal classification hyperplane is constructed by iteratively optimizing the weights and bias parameters of the model to distinguish different operating states; S455: Save the support vector machine model obtained by iterative optimization, which is used to identify the mapping relationship between the input feature vector and the circuit breaker fault state.

7. A pump station circuit breaker fault detection method according to claim 6, characterized in that: The S46 specifically includes: S461: input the feature vectors in the validation set divided in S44 into the trained support vector machine model one by one, classify each input feature vector according to the classification decision function, and output the predicted operating condition label; S462: Record the actual operating condition label of each sample in the validation set and the prediction result of the support vector machine model to form a correspondence table between the actual label and the predicted label; S463: Perform statistics on the label correspondence table and calculate the ratio of the number of samples correctly classified by the model in the validation set to the total number of samples in the validation set as the classification accuracy of the model. The calculation formula is: ,in, Indicates classification accuracy; Indicates the number of samples classified correctly; Indicates the total number of samples in the validation set; S464: Compare the classification accuracy calculated in S463 with the preset classification accuracy threshold When compared, the classification accuracy Greater than or equal to the classification accuracy threshold , it is determined that the preset requirements are met, and the model that meets the requirements is saved as the circuit breaker fault mode recognition model; otherwise, return to S45 to readjust the model parameters and retrain.

8. A pump station circuit breaker fault detection method according to claim 1, characterized in that: The S5 specifically includes: S51: collecting vibration data in real time and preprocessing it according to the same preprocessing method as in S2; then extracting four characteristic parameters of vibration amplitude fluctuation, low-frequency harmonic intensity, resonance frequency offset and nonlinear characteristics from the preprocessed real-time vibration data according to the method in S3, and constructing a real-time characteristic vector; S52: input the constructed real-time feature vector into the fault mode recognition model generated by S4, and use its classification decision function to determine whether the current vibration data belongs to the defined fault mode; if the output classification result is a fault mode, mark the circuit breaker as abnormal; S53: recording the characteristic parameters extracted in S51, and clarifying the specific characteristics of the abnormal vibration signal, including the vibration amplitude fluctuation range, low-frequency harmonic intensity value, resonance frequency offset and nonlinear characteristic index; S54: Based on the abnormal characteristic parameters and model classification results recorded in S53, a fault diagnosis report is generated, wherein the report includes specific values ​​of the abnormal characteristic parameters, classification results of the fault mode, and the fault type.

9. A pump station circuit breaker fault detection system, used to implement a pump station circuit breaker fault detection method as claimed in any one of claims 1 to 8, characterized in that: Includes the following modules: Vibration signal acquisition module: A low-frequency vibration sensor installed at a designated location of the pump station circuit breaker, used to collect vibration signals below 50 Hz in real time and transmit the collected vibration signals in digital form to the signal processing module; Signal processing module: connected to the vibration signal acquisition module, used for preprocessing the collected vibration signal, wherein the preprocessing includes noise removal, filtering and elimination of abnormal data to generate processed vibration data; Feature extraction module: connected with the signal processing module, used to analyze the pre-processed vibration data, extract the characteristic parameters of the low-frequency vibration of the circuit breaker, including vibration amplitude fluctuation, low-frequency harmonic intensity, resonance frequency offset and nonlinear characteristics, and construct the extracted characteristic parameters into a feature vector; Fault pattern recognition module: connected to the feature extraction module, used to receive feature vectors, and train with historical circuit breaker operation data and historical fault data to generate a circuit breaker fault pattern recognition model, which is used to determine whether the current feature vector belongs to the preset fault mode and output the fault detection result; Data recording and diagnosis module: connected to the fault mode recognition module, used to record the fault detection results output by the fault mode recognition module and generate a fault diagnosis report, which includes the specific values ​​of the abnormal characteristic parameters, the classification results of the fault mode and the fault type.

Images