Circuit breaker state identification method and system based on real-time electrical signal
The method improves circuit breaker state recognition through multi-source electrical signal analysis and adaptive threshold denoising, enabling real-time fault warning and reducing maintenance costs by optimizing state identification parameters and thresholds.
Patent Information
- Application Number
- CN202411346683.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-26
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-09-26
AI Technical Summary
Existing circuit breaker status monitoring technologies rely on single parameters and fixed threshold judgments, leading to inefficiencies in real-time monitoring and misjudgment, and are unable to effectively process massive data for accurate fault warning.
A circuit breaker state identification method using multi-source electrical signal acquisition, adaptive threshold denoising, circuit breaker characteristic frequency extraction, time-varying parameter analysis, multi-dimensional clustering, support vector machine algorithm, and online incremental learning to construct an adaptive circuit breaker state classifier for real-time evaluation and fault warning.
Enhances the accuracy and reliability of circuit breaker state recognition, providing timely fault warnings and reducing maintenance costs by optimizing state identification parameters and thresholds, ensuring stable performance in complex electromagnetic environments.
Smart Images

Figure CN118861839B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data processing, and in particular to a circuit breaker state identification method and system based on real-time electrical signals. Background Art
[0002] As a key device in the power system, the reliability of the circuit breaker is directly related to the safe and stable operation of the entire power grid. At present, circuit breaker status monitoring mainly relies on regular maintenance and offline testing. Although this method can detect problems with the equipment in a timely manner, it cannot achieve real-time monitoring and early warning. With the development of smart grids, some online monitoring technologies have been applied to circuit breaker status assessment, such as mechanical property analysis and SF6 gas pressure monitoring. These methods evaluate the working status of the circuit breaker by collecting information such as the mechanical parameters and insulation medium parameters of the circuit breaker.
[0003] However, the existing circuit breaker status monitoring technology still has some shortcomings. First, most methods only focus on a single parameter or system, which makes it difficult to fully reflect the overall status of the circuit breaker. Second, traditional status assessment methods often use fixed threshold judgments, which lack adaptability and are prone to misjudgment or missed judgments. In addition, the existing technology is inefficient in processing massive monitoring data, making it difficult to achieve real-time status identification and fault warning. Summary of the invention
[0004] In view of this, an embodiment of the present invention provides a circuit breaker state identification method and system based on real-time electrical signals, which are used to improve the accuracy of circuit breaker state identification based on real-time electrical signals.
[0005] The present invention provides a circuit breaker state recognition method based on real-time electrical signals, comprising: performing multi-source electrical signal acquisition on a contact system, an arc extinguishing chamber and an operating mechanism of a circuit breaker to obtain an original state signal; performing adaptive threshold denoising and circuit breaker characteristic frequency extraction on the original state signal to obtain a circuit breaker characteristic signal spectrum, and performing time-varying parameter analysis on the circuit breaker characteristic signal spectrum to obtain a circuit breaker state feature vector; performing multi-dimensional clustering processing on the circuit breaker state feature vector to obtain a circuit breaker state space, and dynamically adjusting the boundaries of the circuit breaker state space to obtain an adaptive circuit breaker state judgment criterion with a state migration probability; and performing a support vector machine algorithm on the adaptive circuit breaker state feature vector to obtain a circuit breaker state space. The circuit breaker state judgment criterion and historical operation data are processed to obtain a circuit breaker state classifier; the circuit breaker electrical signal collected in real time is subjected to fast Fourier transform and wavelet packet decomposition to obtain time-frequency characteristics, and the time-frequency characteristics are input into the circuit breaker state classifier to obtain the working state evaluation result of the current circuit breaker, and the working state evaluation result is subjected to fuzzy comprehensive evaluation to obtain the circuit breaker health state index and fault warning level; the circuit breaker state classifier is subjected to online incremental learning and fault mode update, and the judgment threshold in the adaptive circuit breaker state judgment criterion is dynamically adjusted to obtain an updated circuit breaker state identification parameter set and an optimized state judgment threshold.
[0006] The present invention also provides a circuit breaker state identification system based on real-time electrical signals, comprising:
[0007] The acquisition module is used to collect multi-source electrical signals from the contact system, arc extinguishing chamber and operating mechanism of the circuit breaker to obtain the original state signal;
[0008] An extraction module, used for performing adaptive threshold denoising and circuit breaker characteristic frequency extraction on the original state signal to obtain a circuit breaker characteristic signal spectrum, and performing time-varying parameter analysis on the circuit breaker characteristic signal spectrum to obtain a circuit breaker state characteristic vector;
[0009] An adjustment module, configured to perform multi-dimensional clustering processing on the circuit breaker state feature vector to obtain a circuit breaker state space, and dynamically adjust the boundary of the circuit breaker state space to obtain an adaptive circuit breaker state determination criterion with a state transition probability;
[0010] A processing module, used for processing the adaptive circuit breaker state determination criteria and historical operation data by a support vector machine algorithm to obtain a circuit breaker state classifier;
[0011] An input module is used to perform fast Fourier transform and wavelet packet decomposition on the real-time collected circuit breaker electrical signal to obtain time-frequency features, and input the time-frequency features into the circuit breaker state classifier to obtain the current circuit breaker working state evaluation result, and perform fuzzy comprehensive evaluation on the working state evaluation result to obtain the circuit breaker health state index and fault warning level;
[0012] The updating module is used to perform online incremental learning and fault mode updating on the circuit breaker state classifier, and dynamically adjust the determination threshold in the adaptive circuit breaker state determination criterion to obtain an updated circuit breaker state identification parameter set and an optimized state determination threshold.
[0013] In the technical solution provided by the present invention, through multi-source electrical signal collection and analysis, comprehensive monitoring of the circuit breaker contact system, arc extinguishing chamber and operating mechanism is realized, the accuracy and reliability of state recognition are significantly improved, and the limitations of the traditional single parameter evaluation method are overcome; the adaptive threshold denoising and circuit breaker characteristic frequency extraction technology are adopted to effectively filter out environmental noise and irrelevant signals, enhance the anti-interference ability of the system, and ensure stable performance in complex electromagnetic environments; through multi-dimensional clustering processing and dynamic boundary adjustment, an adaptive circuit breaker state judgment criterion with state migration probability is constructed, so that the state space can be adaptively adjusted as the circuit breaker operating conditions change, overcome the limitations of the traditional fixed threshold judgment method, and improve the flexibility and adaptability of state judgment; the circuit breaker state classifier is constructed using the support vector machine algorithm, and the intelligent classification of the circuit breaker state is realized, which not only improves the accuracy of state recognition, but also enhances the system's recognition ability for new states; by real-time collected electrical The signal is subjected to fast Fourier transform and wavelet packet decomposition, combined with fuzzy comprehensive evaluation, which can provide the health status index and fault warning level of the circuit breaker in real time, providing an intuitive and reliable decision-making basis for operation and maintenance personnel; through online incremental learning and fault mode updating of the circuit breaker state classifier, the state identification parameter set and judgment threshold are continuously optimized, so that the system can adapt to the state changes of the circuit breaker during long-term operation and maintain continuous high-efficiency performance; based on the evaluation of contact life, arc extinguishing chamber performance and operating mechanism reliability, targeted maintenance suggestions and priority sorting can be provided, which provides strong support for the implementation of predictive maintenance, helps to reduce the operation and maintenance cost of the circuit breaker and extend its service life; through the comprehensive status evaluation of the contact system, arc extinguishing chamber and operating mechanism, a more comprehensive and accurate description of the health status of the circuit breaker can be given, providing a more scientific and reliable basis for operation and maintenance decisions, helping to improve the safety and reliability of power grid operation, and significantly improving the timeliness and effectiveness of circuit breaker fault warning. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0015] Figure 1 The present invention is a flowchart of a circuit breaker status identification method based on real-time electrical signals in an embodiment of the present invention.
[0016] Figure 2 Schematic diagram of a circuit breaker state identification system based on real-time electrical signals in an embodiment of the present invention. DETAILED DESCRIPTION
[0017] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0018] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.
[0019] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.
[0020] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , Figure 1 is a flow chart of a circuit breaker state identification method based on real-time electrical signals according to an embodiment of the present invention. Figure 1 As shown, the following steps are included:
[0021] S101, collecting multi-source electrical signals from the contact system, arc extinguishing chamber and operating mechanism of the circuit breaker to obtain original state signals;
[0022] S102, performing adaptive threshold denoising and circuit breaker characteristic frequency extraction on the original state signal to obtain a circuit breaker characteristic signal spectrum, and performing time-varying parameter analysis on the circuit breaker characteristic signal spectrum to obtain a circuit breaker state characteristic vector;
[0023] S103, performing multi-dimensional clustering processing on the circuit breaker state feature vector to obtain a circuit breaker state space, and dynamically adjusting the boundary of the circuit breaker state space to obtain an adaptive circuit breaker state determination criterion with state transition probability;
[0024] S104, processing the adaptive circuit breaker state determination criteria and historical operation data by using a support vector machine algorithm to obtain a circuit breaker state classifier;
[0025] S105, performing fast Fourier transform and wavelet packet decomposition on the real-time collected circuit breaker electrical signal to obtain time-frequency features, and inputting the time-frequency features into the circuit breaker state classifier to obtain the working state evaluation result of the current circuit breaker, and performing fuzzy comprehensive evaluation on the working state evaluation result to obtain the circuit breaker health state index and fault warning level;
[0026] S106: Perform online incremental learning and fault mode update on the circuit breaker state classifier, and dynamically adjust the determination threshold in the adaptive circuit breaker state determination criterion to obtain an updated circuit breaker state identification parameter set and an optimized state determination threshold.
[0027] It should be noted that the contact system is monitored by high-frequency resistance measurement technology to obtain the contact resistance time series data; the arc extinguishing chamber is monitored by transient voltage sampling technology to obtain the arc extinguishing voltage waveform data; at the same time, the current of the operating mechanism is collected to obtain the operating mechanism action current curve. The collection of these multi-source electrical signals provides a comprehensive data basis for subsequent state recognition. Next, the collected original state signal is subjected to adaptive threshold denoising to eliminate the interference of environmental noise and irrelevant signals. The adaptive threshold denoising technology can dynamically adjust the denoising threshold according to the characteristics of the signal, so as to more effectively retain useful information. Subsequently, the circuit breaker characteristic signal spectrum is obtained by the circuit breaker characteristic frequency extraction technology. By analyzing the frequency domain characteristics of the signal, the characteristic frequency component that best reflects the circuit breaker state is identified. The circuit breaker characteristic signal spectrum is subjected to time-varying parameter analysis, and the characteristic parameters reflecting the change of the circuit breaker state are calculated, thereby obtaining the circuit breaker state characteristic vector. This vector contains the key information describing the current state of the circuit breaker. Then, the circuit breaker state characteristic vector is subjected to multi-dimensional clustering processing to construct the circuit breaker state space. Multidimensional clustering technology can cluster similar state features together to form different state categories. The constructed circuit breaker state space is dynamically adjusted to take into account the dynamic change characteristics of the circuit breaker state. By analyzing the frequency and probability of state transition in historical data, an adaptive circuit breaker state judgment criterion with state migration probability is obtained. This criterion can be adaptively adjusted as the circuit breaker operating status changes, improving the accuracy and flexibility of state judgment. Then, the adaptive circuit breaker state judgment criterion and historical operating data are processed by the support vector machine algorithm to train the circuit breaker state classifier. The support vector machine algorithm achieves effective classification of different states of the circuit breaker by finding the optimal classification hyperplane. Fast Fourier transform and wavelet packet decomposition are performed on the real-time collected circuit breaker electrical signal to obtain time-frequency features. Fast Fourier transform provides the frequency domain information of the signal, while wavelet packet decomposition can capture the local time-frequency characteristics of the signal. These time-frequency features are input into the previously trained circuit breaker state classifier to obtain the current circuit breaker working state evaluation result. In order to further improve the reliability of the evaluation, the working status evaluation results are subjected to fuzzy comprehensive evaluation, and finally the circuit breaker health status index and fault warning level are obtained. Fuzzy comprehensive evaluation technology can handle the uncertainty in the evaluation process and provide more reasonable evaluation results. Finally, in order to maintain the continuous effectiveness of the system, the circuit breaker state classifier is subjected to online incremental learning and fault mode update. This means that the system can continuously absorb new operating data, optimize the classification model, and adapt to the long-term changes in the circuit breaker state. At the same time, the judgment threshold in the adaptive circuit breaker state judgment criterion is dynamically adjusted to obtain an updated circuit breaker state identification parameter set and an optimized state judgment threshold. This dynamic adjustment mechanism ensures that the system can always maintain efficient state recognition capabilities.
[0028] For example, during the operation of a circuit breaker, the contact resistance gradually increased from the initial 50μΩ to 65μΩ through high-frequency resistance measurement technology. At the same time, the transient voltage sampling of the arc extinguishing chamber showed that the arc duration was extended from 5ms to 7ms, and the operating mechanism action current curve also showed slight fluctuations. After these original state signals were processed by adaptive threshold denoising, the noise level was reduced by about 20%. Through the circuit breaker characteristic frequency extraction, 200Hz, 500Hz and 1kHz were identified as the key frequency components reflecting the circuit breaker state. Time-varying parameter analysis was performed on these frequency components, and the state feature vectors obtained included parameters such as the contact resistance change rate, arc duration growth rate and operating mechanism action current fluctuation amplitude. Multi-dimensional clustering processing divided these feature vectors into three state categories: normal, mild wear and moderate wear. The state classifier trained by the support vector machine algorithm classified these features, and combined with fuzzy comprehensive evaluation, it was finally concluded that the circuit breaker health status index was 0.82 (full score is 1), and the fault warning level was mild warning. Through this series of processing, operation and maintenance personnel can timely understand the health status of the circuit breaker, take corresponding maintenance measures, and effectively prevent the occurrence of potential failures.
[0029] By executing the above steps, through multi-source electrical signal acquisition and analysis, comprehensive monitoring of the circuit breaker contact system, arc extinguishing chamber and operating mechanism is achieved, the accuracy and reliability of state recognition are significantly improved, and the limitations of the traditional single parameter evaluation method are overcome; the adaptive threshold denoising and circuit breaker characteristic frequency extraction technology are used to effectively filter out environmental noise and irrelevant signals, enhance the anti-interference ability of the system, and ensure stable performance in complex electromagnetic environments; through multi-dimensional clustering processing and dynamic boundary adjustment, an adaptive circuit breaker state judgment criterion with state migration probability is constructed, so that the state space can be adaptively adjusted as the circuit breaker operating conditions change, overcome the limitations of the traditional fixed threshold judgment method, and improve the flexibility and adaptability of state judgment; the circuit breaker state classifier is constructed using the support vector machine algorithm to realize the intelligent classification of the circuit breaker state, which not only improves the accuracy of state recognition, but also enhances the system's ability to recognize new states; by real-time collected telecommunications The signal is subjected to fast Fourier transform and wavelet packet decomposition, and combined with fuzzy comprehensive evaluation, the health status index and fault warning level of the circuit breaker can be given in real time, providing an intuitive and reliable decision-making basis for operation and maintenance personnel; by performing online incremental learning and fault mode updating on the circuit breaker state classifier, the state recognition parameter set and judgment threshold are continuously optimized, so that the system can adapt to the state changes during the long-term operation of the circuit breaker and maintain continuous high-efficiency performance; based on the evaluation of contact life, arc extinguishing chamber performance and operating mechanism reliability, targeted maintenance suggestions and priority sorting can be provided, which provides strong support for the implementation of predictive maintenance, helps to reduce the operation and maintenance cost of the circuit breaker and prolong its service life; through the comprehensive status evaluation of the contact system, arc extinguishing chamber and operating mechanism, a more comprehensive and accurate description of the health status of the circuit breaker can be given, providing a more scientific and reliable basis for operation and maintenance decisions, helping to improve the safety and reliability of power grid operation, and significantly improving the timeliness and effectiveness of circuit breaker fault warning.
[0030] In a specific embodiment, the process of executing step S101 may specifically include the following steps:
[0031] (1) Perform high-frequency resistance measurement on the contact system of the circuit breaker to obtain the contact resistance time series data, and perform sliding window analysis on the contact resistance time series data to obtain the contact resistance change rate;
[0032] (2) Sampling the transient voltage of the arc extinguishing chamber of the circuit breaker to obtain arc extinguishing voltage waveform data, and performing peak detection and duration calculation on the arc extinguishing voltage waveform data to obtain arc extinguishing characteristic parameters;
[0033] (3) Collect current from the operating mechanism of the circuit breaker to obtain the operating mechanism action current curve, and perform integration operation on the operating mechanism action current curve to obtain the operating energy index;
[0034] (4) Evaluate the degree of wear of the contact system based on the contact resistance change rate to obtain a predicted value for the contact life. Estimate the contact replacement time based on the predicted value for the contact life to obtain contact maintenance recommendations.
[0035] (5) Analyze the performance of the arc extinguishing chamber according to the arc extinguishing characteristic parameters to obtain the arc extinguishing efficiency index, and classify the arc extinguishing chamber status according to the arc extinguishing efficiency index to obtain the arc extinguishing chamber working status evaluation result;
[0036] (6) Evaluate the reliability of the operating mechanism based on the operating energy index to obtain the operating mechanism reliability index, and predict the failure risk of the operating mechanism based on the operating mechanism reliability index to obtain the maintenance priority of the operating mechanism;
[0037] (7) Perform time synchronization processing on the contact resistance time series data, arc extinguishing voltage waveform data and operating mechanism action current curve to obtain a multi-source signal collaborative analysis data set;
[0038] (8) The multi-source signal collaborative analysis data set is decomposed at multiple scales by wavelet packet transform algorithm to obtain the circuit breaker state characteristic components, and the energy entropy of the circuit breaker state characteristic components is calculated to obtain the circuit breaker comprehensive state index;
[0039] (9) Based on the contact maintenance recommendations, arc extinguishing chamber working status assessment results and operating mechanism maintenance priorities, a comprehensive status assessment of the circuit breaker is performed to obtain the circuit breaker health status descriptor;
[0040] (10) Based on the circuit breaker comprehensive status indicators and the circuit breaker health status descriptor, the circuit breaker is globally judged to obtain the original status signal.
[0041] Specifically, during the monitoring process of the contact system, the time series data of the contact resistance is obtained through high-frequency resistance measurement technology. High-frequency resistance measurement is a technology that continuously records resistance changes at a high sampling frequency in a short period of time. Its core is to capture the trend of gradual changes in contact resistance over time and usage. Next, this time series data is processed by sliding window analysis. Sliding window analysis sets a fixed-size time window and performs moving average or differential calculation on the resistance data in the window to obtain the rate of change of resistance over time, that is, the contact resistance change rate. This change rate can be used to evaluate the degree of wear of the contact. When the wear increases, the resistance change rate will increase significantly.
[0042] For example, suppose that during a measurement, the contact resistance gradually increases from 50μΩ to 60μΩ, the sampling frequency is 1kHz, and the sliding window size is 100 data points. Through sliding window analysis, the resistance change rate per second is 0.1μΩ / s. This change rate combined with historical data can be used to predict the life of the contact and estimate when the contact needs to be replaced based on the expected wear. In the performance monitoring of the arc extinguishing chamber, transient voltage sampling technology is used to obtain the voltage waveform data during the arc extinguishing process. By peak detection and duration calculation of the waveform data, the peak value of the arc extinguishing voltage and the duration of the arc can be extracted. These two arc extinguishing characteristic parameters are important bases for evaluating the working efficiency of the arc extinguishing chamber. The peak detection technology scans the highest point of the voltage waveform to determine the maximum voltage value when the arc occurs, and the duration calculation is used to determine the duration of the arc from triggering to extinguishing. Through these data, it is possible to effectively determine whether the arc extinguishing chamber is in a healthy state. For example, if the arc duration is extended to 7ms, which should be 5ms under normal circumstances, it may mean that the performance of the arc extinguishing chamber has deteriorated.
[0043] For monitoring of the operating mechanism, the current curve during the operation of the operating mechanism is collected by the current sensor, and then the current curve is integrated to calculate the operating energy index. Integration operation is an accumulation operation. By calculating the area under the current curve, the total energy consumed during the operation is obtained. The operating energy index is closely related to the mechanical action performance of the mechanism. Excessive energy consumption may mean increased resistance to the mechanism action, wear of mechanical parts or other hidden dangers. For example, assuming that the integral result of the action current in a certain operation is 500 joules (J), while it should be 450J under normal circumstances. This abnormal increase in energy indicates that the operating mechanism may need maintenance. After the data is processed, the contact resistance change rate, arc extinguishing characteristic parameters and operating energy indicators are used to evaluate the health status of their respective systems. According to the contact resistance change rate, the degree of wear of the contact system is evaluated, and the remaining life of the contact is predicted by this evaluation result, and then the time for the next contact replacement is calculated. The arc extinguishing characteristic parameters are used to analyze the working performance of the arc extinguishing chamber and obtain the arc extinguishing efficiency index. If the arc extinguishing efficiency is low, the evaluation result of the working status of the arc extinguishing chamber will indicate that the arc extinguishing chamber needs to be repaired. The operating energy index is used to evaluate the reliability of the operating mechanism, predict the failure risk based on the reliability index, and generate maintenance priorities.
[0044] After completing these independent evaluations, the contact resistance time series data, arc extinguishing voltage waveform data, and operating mechanism current curve are time synchronized. The purpose of this step is to ensure that multi-source signals can be analyzed under a unified time reference to obtain the correlation between multi-source signals during circuit breaker operation. After time synchronization, the wavelet packet transform algorithm is used to perform multi-scale decomposition on these synchronized data. Wavelet packet transform is a tool that can decompose signals into different frequency components. By decomposing multi-source signals step by step, different characteristic components reflecting the state of the circuit breaker can be extracted. Each characteristic component represents the state information of the circuit breaker at a certain frequency or time scale.
[0045] Then, the energy entropy is calculated for the decomposed characteristic components. It is an indicator to measure signal complexity, which is used to indicate the energy distribution of different frequency components in the system. Its calculation formula is:
[0046] ;
[0047] in, Indicates The energy proportion of the frequency component is is the total number of frequency components. Through energy entropy calculation, the complexity and stability of the circuit breaker state can be judged. The higher the energy entropy value, the more complex the system state is and the more uncertain factors exist; while the lower the energy entropy value, the simpler and more stable the system state is.
[0048] After all data analysis is completed, the comprehensive state descriptor of the circuit breaker is generated by combining the contact maintenance recommendations, the arc extinguishing chamber working state assessment results, and the maintenance priority of the operating mechanism. This descriptor is a multi-dimensional state representation that comprehensively reflects the current health status of each key component of the circuit breaker. Finally, by making a global state judgment on the comprehensive state descriptor and the circuit breaker comprehensive state indicator, the original state signal of the circuit breaker is obtained and provided to the operation and maintenance personnel as the final state reference.
[0049] For example, during a routine inspection of a circuit breaker, the system detected that the contact resistance gradually increased from the initial 50μΩ to 65μΩ through high-frequency resistance measurement technology. The sliding window analysis results showed that the resistance change rate was 1.2μΩ / s. Combined with the wear assessment model, the contact life was expected to be about 30 days. The arc duration obtained by sampling the transient voltage of the arc extinguishing chamber was 6.8ms, the voltage peak was 850V, and the integral calculation result of the operating mechanism action current was 480J. Based on these data, the system evaluated that the contacts needed to be replaced within 30 days, the arc extinguishing chamber was in a state of slight wear, and the operating mechanism still had a high reliability. This comprehensive evaluation result provides clear suggestions for the maintenance of the circuit breaker and avoids the occurrence of potential failures.
[0050] In a specific embodiment, the process of executing step S102 may specifically include the following steps:
[0051] (1) Perform wavelet threshold denoising on the original state signal to obtain a preliminary denoised signal, and calculate the signal-to-noise ratio of the preliminary denoised signal to obtain a signal-to-noise ratio index;
[0052] (2) Evaluate the noise level according to the signal-to-noise ratio index to obtain the adaptive threshold coefficient, and perform secondary wavelet threshold denoising on the preliminary denoised signal according to the adaptive threshold coefficient to obtain the optimized denoised signal;
[0053] (3) Performing fast Fourier transform on the optimized noise reduction signal to obtain the circuit breaker characteristic signal spectrum, and performing peak detection on the circuit breaker characteristic signal spectrum to obtain the circuit breaker characteristic frequency set;
[0054] (4) performing bandpass filtering on the optimized noise reduction signal according to the circuit breaker characteristic frequency set to obtain the circuit breaker characteristic signal, and performing Hilbert transform on the circuit breaker characteristic signal to obtain the analytical signal;
[0055] (5) Calculate the instantaneous frequency of the analytical signal to obtain a time-frequency distribution diagram, and estimate the energy density of the time-frequency distribution diagram to obtain a time-varying energy spectrum;
[0056] (6) Extract the time-frequency features of the circuit breaker state according to the time-varying energy spectrum to obtain the time-frequency feature vector, and perform principal component analysis on the time-frequency feature vector to obtain the reduced-dimensional feature vector;
[0057] (7) Calculate the statistical parameters of the reduced-dimensional feature vector to obtain a statistical feature set, and construct a circuit breaker state feature space based on the statistical feature set to obtain a circuit breaker state feature vector.
[0058] It should be noted that the original electrical signal is initially denoised using wavelet threshold denoising technology. This method uses wavelet transform to decompose the signal into different frequency bands, and then sets appropriate thresholds on each frequency band to eliminate noise components. The core of wavelet threshold denoising is that it can set denoising thresholds for signals at each scale, thereby effectively suppressing noise while retaining the main signal features. After the initial denoising is completed, the signal-to-noise ratio (SNR) indicator of the signal is calculated, which is an important parameter for evaluating signal quality.
[0059] Signal-to-Noise Ratio The calculation formula is:
[0060]
[0061] in, represents the signal power, represents the noise power. By calculating , the proportion of noise in the signal can be quantified to clarify the quality of the current signal. If the SNR is low, it means that there is a lot of noise in the signal and further optimization of denoising is needed. Next, the noise level is evaluated based on the signal-to-noise ratio indicator and the adaptive threshold coefficient is determined. The adaptive threshold coefficient dynamically adjusts the threshold of the signal at different frequency components, so that noise suppression and signal fidelity can be better balanced during the denoising process. Subsequently, this threshold coefficient is used to perform secondary wavelet threshold denoising on the preliminary denoised signal. Secondary denoising aims to process the residual noise in the signal more finely, thereby obtaining an optimized denoised signal.
[0062] The optimized noise reduction signal is subjected to a fast Fourier transform (FFT) to convert the signal from the time domain to the frequency domain and obtain the characteristic signal spectrum of the circuit breaker. Fast Fourier transform is an efficient algorithm that can decompose complex time domain signals into multiple frequency components and reveal the amplitude information of the signal at different frequencies. By analyzing these frequency components, the key frequency characteristics reflecting the state of the circuit breaker can be extracted. Next, these frequency components are peak detected to identify the most significant frequency in the signal and obtain the circuit breaker characteristic frequency set. After obtaining the characteristic frequency set, the optimized noise reduction signal is processed using bandpass filtering to retain these key frequency components. Bandpass filtering is a filtering technology that only allows signals within a certain frequency range to pass through while suppressing other frequency components. This filtering can effectively eliminate irrelevant noise and interference signals, thereby obtaining a purer circuit breaker characteristic signal.
[0063] Furthermore, the filtered characteristic signal is subjected to Hilbert transform to obtain the analytical form of the signal. Hilbert transform can extract instantaneous amplitude and instantaneous phase information from the original signal. By analyzing the signal, subtle changes in the circuit breaker electrical signal can be captured, which is crucial for analyzing the instantaneous frequency. By calculating the instantaneous frequency of the analytical signal, a time-frequency distribution diagram of the circuit breaker electrical signal can be generated. This distribution diagram shows how the frequency of the signal changes over time. Next, the energy density of the time-frequency distribution diagram is estimated to calculate the energy distribution of the signal in each time-frequency unit, and finally the time-varying energy spectrum is obtained. The time-varying energy spectrum provides the distribution of signal energy at different times and frequencies, which can help identify instantaneous characteristic changes during the operation of the circuit breaker.
[0064] According to the time-varying energy spectrum, the time-frequency feature vector reflecting the state of the circuit breaker can be extracted. The time-frequency feature vector is a multidimensional representation of the circuit breaker electrical signal, which contains the energy characteristics of different time and frequency components. In order to simplify the analysis and reduce the data dimension, principal component analysis (PCA) is performed on the time-frequency feature vector. Principal component analysis is a linear dimensionality reduction method that projects the data into a low-dimensional space and retains the most important feature dimensions in the data, thereby simplifying the data structure and reducing redundant information. The feature vector after dimensionality reduction contains the most critical information of the circuit breaker state. On this basis, the statistical parameters of these feature vectors can be calculated, such as mean, variance, skewness and kurtosis. These statistical parameters are used to describe the overall distribution of the signal and finally construct the state feature space of the circuit breaker. Through this feature space, the characteristic performance of the circuit breaker under different operating conditions can be accurately described, and the circuit breaker state feature vector can be obtained.
[0065] For example, during a certain operation, the original electrical signal of the circuit breaker collected a large noise interference. After the initial wavelet denoising, the signal-to-noise ratio was only 15dB, indicating that the noise interference was still serious. According to the signal-to-noise ratio, the adaptive threshold coefficient was set to 0.8, and secondary wavelet denoising was performed. After denoising, the signal-to-noise ratio was increased to 25dB. Through fast Fourier transform, it was found that the main frequencies of the circuit breaker signal were concentrated at 200Hz, 600Hz and 1.2kHz. These frequencies were band-pass filtered to remove other irrelevant frequency components. After that, the instantaneous frequency of the analytical signal was calculated through Hilbert transform, the time-frequency distribution diagram was generated, and the time-varying energy spectrum was calculated. Finally, through principal component analysis, three main feature dimensions were extracted, corresponding to the power increment, instantaneous frequency change and energy density change of the signal. By statistically calculating these feature dimensions, the state feature vector of the circuit breaker was constructed to represent the health status of the circuit breaker during this operation cycle.
[0066] In a specific embodiment, the process of executing step S103 may specifically include the following steps:
[0067] (1) Perform K-means cluster analysis on the circuit breaker state feature vector to obtain the initial state category, and calculate the silhouette coefficient of the initial state category to obtain the clustering effectiveness index;
[0068] (2) Evaluate the clustering results according to the clustering effectiveness index to obtain the optimal number of clusters, and perform hierarchical clustering on the circuit breaker state feature vectors according to the optimal number of clusters to obtain the target state category;
[0069] (3) Assign labels to the target state categories to obtain the circuit breaker state space, and extract boundary points from the circuit breaker state space to obtain the initial state boundary;
[0070] (4) Construct a space partition diagram of the circuit breaker state space according to the initial state boundary to obtain a state partition diagram, and perform boundary smoothing on the state partition diagram to obtain an optimized state boundary;
[0071] (5) Perform state transition frequency statistics on historical state data to obtain a state transition matrix, and calculate the transition probability between states based on the state transition matrix to obtain a state migration probability model;
[0072] (6) The optimized state boundary is dynamically adjusted according to the state transition probability model to obtain an adaptive state boundary, and a circuit breaker state determination rule is constructed according to the adaptive state boundary to obtain an adaptive circuit breaker state determination criterion.
[0073] Specifically, the K-means clustering algorithm is used to perform cluster analysis on the state feature vectors of the circuit breaker. K-means clustering is a commonly used unsupervised learning algorithm, and its goal is to assign feature vectors to different cluster centers according to feature similarity. Each feature vector is assigned to the cluster with the closest distance by calculating the Euclidean distance from each cluster center. The algorithm continuously updates the position of the cluster center until the clustering results converge and obtain a stable state category. At this point, each circuit breaker state feature vector is assigned to an initial state category, indicating the specific state of the circuit breaker at different operating times. Next, the initial clustering results are verified. In order to evaluate the effect of clustering, the silhouette coefficient is calculated as an indicator of clustering effectiveness. The silhouette coefficient is an indicator to measure the quality of clustering. Its value ranges from -1 to 1. The closer it is to 1, the better the clustering effect. The tightness within the cluster is high and the differences between different categories are large. Silhouette coefficient The calculation formula is:
[0074]
[0075] in, Indicates The average distance from a sample to other samples in its category, Represents the average distance from the sample to the nearest category. By calculating the silhouette coefficient, the clustering effect of each feature vector can be quantified and the rationality of the entire clustering model can be determined. According to the size of the silhouette coefficient, the initial clustering results are evaluated and the number of clusters with the best clustering effect is selected. The selection of the number of clusters directly affects the complexity and accuracy of the model. Therefore, the clustering results can be optimized through the clustering effectiveness index. After determining the optimal number of clusters, the hierarchical clustering method is used to re-cluster the circuit breaker state feature vectors. Hierarchical clustering is an algorithm that gradually merges or splits categories. Unlike K-means, it does not require the number of clusters to be set in advance. Through hierarchical clustering, the state categories can be divided more finely, and clustering results that are more in line with the operating status of the circuit breaker can be obtained.
[0076] After clustering is completed, labels are assigned to each target state category to form the state space of the circuit breaker. The state space is a multidimensional feature space, and each dimension represents the state characteristics of the circuit breaker under a specific operating condition. The construction of the state space allows the relationship between different state categories to be clearly expressed. In order to further optimize the state space, boundary points are extracted in the state space to identify the initial boundaries of the circuit breaker state. These boundary points are the dividing lines in the feature space, which define the distinction criteria for different state categories. In order to make the state space clearer, the initial state boundary is spatially segmented and a state partition diagram is constructed. The state partition diagram visualizes the various state categories in the state space and makes the boundaries between each category clearer. After the partition diagram is constructed, the boundary smoothing process is performed to eliminate the noise and irregular boundaries generated during the initial partitioning, so that the dividing lines of each state category are smoother and consistent with the state transition of the actual operation of the circuit breaker.
[0077] In order to make the state judgment dynamically adaptable, the state transition frequency statistics are then performed based on the historical operation data to generate a state transition matrix. The state transition matrix describes the frequency of the circuit breaker transferring from one state to another, reflecting the probability distribution of the state change of the circuit breaker in actual operation. Through the state transition matrix, a state migration probability model can be established to quantify the possibility of the circuit breaker transitioning from one state category to other state categories. Based on the state migration probability model, the optimized state boundary is dynamically adjusted to form an adaptive state boundary. The adaptive state boundary can be adjusted in real time according to the state change of the circuit breaker during operation, avoiding the use of fixed thresholds to judge the state, thereby improving the flexibility and accuracy of state judgment. Finally, the state judgment rule of the circuit breaker is constructed through the adaptive state boundary. This rule combines the characteristic vector and state space of the circuit breaker, and can accurately classify the current state of the circuit breaker during operation to form an adaptive circuit breaker state judgment criterion.
[0078] For example, in a circuit breaker operation monitoring, the system collected multiple characteristic data such as contact resistance, arc extinguishing voltage and operating mechanism current. Through K-means clustering, these characteristic vectors were divided into 5 initial state categories, and the average value of the silhouette coefficient was 0.75, indicating that the clustering effect was good. Then, according to the analysis results of the silhouette coefficient, the number of clusters was adjusted to 3, and the state was re-divided by hierarchical clustering. Finally, the operation state of the circuit breaker was divided into three categories: "normal", "mild wear" and "moderate wear". In the process of constructing the state space, the boundary points were extracted and the boundary between "normal" and "mild wear" was identified. In the generated state partition diagram, it was found that some boundaries were discontinuous. After boundary smoothing, the partition diagram became smoother. Subsequently, by analyzing the historical operation data of the circuit breaker and generating the state transition matrix, it was found that the transition probability of the circuit breaker from "mild wear" to "moderate wear" was 30%. Based on this, the state boundary was adjusted to form an adaptive state boundary model. Finally, the circuit breaker state judgment criterion was generated by combining the state transition probability model with the optimized state boundary.
[0079] In a specific embodiment, the process of executing step S104 may specifically include the following steps:
[0080] (1) The adaptive circuit breaker state judgment criterion is parameterized to obtain a feature descriptor, and feature extraction is performed on the historical operation data based on the feature descriptor to obtain a training data set;
[0081] (2) Perform data normalization on the training data set to obtain standardized training data, and perform principal component analysis on the standardized training data to obtain a reduced-dimensional feature set;
[0082] (3) Constructing a kernel function based on the reduced-dimensional feature set to obtain a nonlinear mapping function, and using the nonlinear mapping function to transform the standardized training data into a high-dimensional space to obtain a transformed feature space;
[0083] (4) Constructing the maximum margin hyperplane in the transformed feature space to obtain the initial classification boundary, and then soft-optimizing the initial classification boundary to obtain the optimized classification hyperplane;
[0084] (5) Select support vectors based on the optimized classification hyperplane to obtain the key sample set, and construct a decision function based on the key sample set to obtain a multi-category classifier;
[0085] (6) The multi-category classifier is cross-validated to obtain the classification performance index, and the classifier parameters are fine-tuned according to the classification performance index to obtain the circuit breaker state classifier.
[0086] Specifically, the state judgment criteria of the adaptive circuit breaker are parameterized to generate feature descriptors. These feature descriptors are obtained by extracting features from electrical signals and other key parameters collected during the operation of the circuit breaker. They cover multi-dimensional state information such as the change rate of contact resistance, the voltage fluctuation characteristics of the arc extinguishing chamber, and the operating current of the operating mechanism. Based on these descriptors, the corresponding features can be extracted from the historical operation data to construct a training data set. The training data set is a typical feature set of the circuit breaker under different operating conditions, which is used for the training and optimization of the subsequent classification model. In order to ensure the consistency of the data and the stability of the calculation, the training data set is then normalized. The purpose of normalization is to scale the feature data of each dimension to the same order of magnitude to avoid excessive impact on model training due to the large value range of some features. The common method of normalization is to scale the data according to its maximum and minimum values so that the value range of each feature is between 0 and 1. After normalization, the data distribution is more uniform, which is convenient for subsequent principal component analysis (PCA).
[0087] After completing data normalization, principal component analysis is performed to simplify the data in the form of a reduced-dimensional feature set. PCA projects high-dimensional data into a low-dimensional subspace through linear transformation while retaining the main information in the data. It can effectively reduce feature redundancy and improve the computational efficiency of subsequent classification models. The goal of PCA is to find a new coordinate axis system that maximizes the variance of the data in this new coordinate axis system, thereby explaining most of the variability of the data. Through PCA, a reduced-dimensional feature set can be obtained, which retains most of the useful information in the original data while reducing the dimension. Based on the reduced-dimensional feature set, the kernel function is then constructed. The kernel function transforms the reduced-dimensional data into a high-dimensional space through nonlinear mapping, in which the linear separability of the data is enhanced. Commonly used kernel functions include polynomial kernels, Gaussian kernels, etc., which can map the original feature space into a more complex high-dimensional feature space through nonlinear transformation. In this space, the data distribution is clearer, which is conducive to the construction of classification models. In the transformed high-dimensional feature space, the maximum margin hyperplane is constructed based on the principle of support vector machine (SVM). The maximum margin hyperplane refers to finding an optimal interface between different types of data, so that the distance (i.e., the interval) from the interface to the closest point of the two types of data is maximized. The purpose of the maximum margin is to make the classification model have the strongest ability to distinguish data and have good generalization performance. In order to further optimize the classification effect, soft margin optimization technology is used. Soft margin optimization allows some data points to exist on the "wrong" side of the hyperplane to improve the fault tolerance of the model, especially when there is noise or outliers in the data, soft margin optimization can prevent overfitting.
[0088] After soft margin optimization, the final optimized classification hyperplane is obtained. On this basis, support vectors are selected to construct key sample sets. Support vectors are sample points located near the boundary of the hyperplane, which play a decisive role in the construction of the hyperplane. Based on these key sample points, a decision function is further constructed. The decision function is used to predict the category of a new sample. It determines the category to which the sample belongs by calculating the similarity between the new sample and the support vector and combining the position of the hyperplane. The accuracy and stability of the decision function directly affect the performance of the final classifier. In order to verify the performance of the classifier, the multi-category classifier is evaluated using cross-validation technology. Cross-validation is to divide the data set into multiple subsets, which are used for training and validating the model respectively, and then evaluate the performance of the classifier on different data. Through cross-validation, the situation where the model performs well on the training set but poorly on the test set can be avoided, ensuring that the classifier has good generalization ability. According to the results of cross-validation, classification performance indicators such as accuracy, recall and F1 value are calculated. These indicators can quantify the performance of the classifier on different categories and help discover the advantages and disadvantages of the model.
[0089] Finally, the classifier parameters are fine-tuned according to the classification performance indicators to further optimize the performance of the model. By adjusting the parameters of the kernel function or the weight of the regularization term, the generalization ability of the model can be improved and the classification error can be reduced. Finally, after parameter adjustment, a stable and accurate circuit breaker status classifier is constructed, which is used for real-time monitoring and classification of the operating status of circuit breakers in practical applications.
[0090] For example, in the monitoring of a circuit breaker operation, the feature vector in the electrical signal was extracted through the adaptive state judgment criterion. After normalization and PCA dimension reduction, the main feature dimensions obtained included the contact resistance change rate and the arc chamber voltage peak change. Subsequently, the Gaussian kernel function was used to map these dimension-reduced data to a high-dimensional space, and an initial classification hyperplane was constructed. After soft margin optimization, the position of the classification hyperplane was adjusted so that data points with less electrical noise influence could be classified more accurately. The cross-validation results showed that the accuracy of the classifier was 92%, and after parameter fine-tuning, the accuracy was increased to 95%. In the end, this circuit breaker state classifier can accurately distinguish between the three states of "normal", "mild fault" and "serious fault".
[0091] In a specific embodiment, the process of executing step S105 may specifically include the following steps:
[0092] (1) Perform fast Fourier transform on the real-time collected circuit breaker electrical signal to obtain spectrum data, and extract characteristic frequencies from the spectrum data to obtain a frequency domain feature set;
[0093] (2) Perform wavelet packet decomposition on the real-time collected circuit breaker electrical signal to obtain multi-scale coefficients, and calculate the energy of the multi-scale coefficients to obtain a time-frequency energy distribution diagram;
[0094] (3) Perform feature fusion based on the frequency domain feature set and the time-frequency energy distribution diagram to obtain a comprehensive time-frequency feature vector, and normalize the comprehensive time-frequency feature vector to obtain a standardized feature input;
[0095] (4) Input the standardized features into the circuit breaker state classifier to obtain the circuit breaker state probability distribution, and perform maximum likelihood estimation on the circuit breaker state probability distribution to obtain the preliminary working state assessment result;
[0096] (5) Conduct confidence analysis on the preliminary working status assessment results to obtain the status assessment reliability index, and revise the preliminary working status assessment results based on the status assessment reliability index to obtain the revised working status assessment results;
[0097] (6) Construct a fuzzy evaluation index system to obtain a fuzzy evaluation factor set, and perform a fuzzy comprehensive evaluation on the revised working status evaluation results based on the fuzzy evaluation factor set to obtain a fuzzy subset of the circuit breaker health status;
[0098] (7) Defuzzify the fuzzy subset of the circuit breaker health status using the centroid method to obtain the circuit breaker health status index. Set the warning threshold based on the circuit breaker health status index to obtain the fault warning level.
[0099] Specifically, the real-time collected circuit breaker electrical signal is subjected to fast Fourier transform (FFT). The function of fast Fourier transform is to convert the time domain signal of the circuit breaker into a frequency domain signal to obtain spectrum data. The spectrum data shows the energy distribution of different frequency components in the signal, which is of great significance for identifying vibration, noise and other anomalies during the operation of the circuit breaker. By extracting the characteristic frequency of the spectrum data, the significant frequency components appearing in the circuit breaker can be captured. These frequencies may represent certain specific mechanical or electrical characteristics, such as the frequency reflection of contact switching, arc extinguishing process and other behaviors. The extracted characteristic frequencies constitute the frequency domain feature set, which provides basic data for subsequent signal analysis. Then, the same electrical signal is subjected to wavelet packet decomposition. Wavelet packet decomposition is a time-frequency analysis method that subdivides the signal at different time scales. Compared with the global frequency domain analysis of Fourier transform, wavelet packet decomposition can obtain the frequency change of the signal at different time scales, thereby performing multi-level analysis of the circuit breaker signal. Through wavelet packet decomposition, the coefficients of the signal at different scales can be obtained, and these coefficients can represent the local characteristics of the signal at each time scale. By calculating the energy of these multi-scale coefficients, we can get the time-frequency energy distribution diagram. The time-frequency energy distribution diagram shows the energy distribution of the signal at different times and can reflect the energy change trend of the circuit breaker at different time points.
[0100] By fusing the frequency domain feature set with the time-frequency energy distribution diagram, a comprehensive time-frequency feature vector can be obtained. The feature fusion process includes merging the two types of features according to certain rules, such as weighted average or cascade, so that the fused feature vector contains feature information in both the time domain and the frequency domain. The comprehensive time-frequency feature vector can fully reflect the operating status of the circuit breaker at different times and frequencies, and provide input data for subsequent state classification. After obtaining the comprehensive time-frequency feature vector, it needs to be normalized to obtain a standardized feature input. The purpose of normalization is to eliminate the magnitude differences between different features so that all feature values are scaled to a relatively consistent range, usually [0,1] or [-1, 1]. Through normalization, it is possible to avoid the model's over-reliance on or neglect of certain features due to differences in feature magnitude.
[0101] The standardized feature input is passed into the pre-built circuit breaker state classifier, and the classifier can output the state probability distribution of the circuit breaker based on the input data. The state probability distribution reflects the possibility of the circuit breaker being in different working states, such as "normal", "mild fault" or "serious fault". In order to improve the accuracy of classification, maximum likelihood estimation is performed based on the state probability distribution to obtain preliminary working state assessment results. Maximum likelihood estimation is a statistical method that aims to infer the most likely state result by maximizing the likelihood function of observed data. After the preliminary working state assessment results are obtained, confidence analysis is required. Confidence analysis is used to evaluate the reliability of the classification results, that is, the prediction credibility of the classifier for given input data. By calculating the reliability index of the state assessment, the confidence of the current state classification result can be judged. If the confidence of a certain state is low, the preliminary assessment results may need to be corrected. According to the reliability index of the state assessment, appropriate correction operations are performed to finally obtain more accurate working state assessment results.
[0102] In order to further improve the evaluation results, a fuzzy evaluation index system is constructed. By introducing fuzzy set theory, fuzzy evaluation maps the working state of the circuit breaker into multiple fuzzy evaluation factors, such as "good condition", "mild wear" and "serious fault". According to the fuzzy evaluation factor set, the corrected working state evaluation results can be fuzzy comprehensively evaluated to obtain the fuzzy subset of the health state of the circuit breaker. The principle of fuzzy comprehensive evaluation is to calculate the contribution of each factor to the circuit breaker state according to the weights of multiple fuzzy evaluation factors to form a final comprehensive judgment. After obtaining the fuzzy subset of the health state, the centroid method is used to defuzzify the fuzzy set. The centroid method is a classic method for converting fuzzy sets into clear values. It determines the health state index of the circuit breaker by calculating the centroid position of the fuzzy set. The health state index is a numerical value that quantifies the overall health level of the circuit breaker and can intuitively reflect the current state of the circuit breaker. Finally, the warning threshold is set based on the health state index to divide different fault warning levels. These warning levels can help operation and maintenance personnel identify potential fault risks in advance and take corresponding maintenance measures according to the warning level.
[0103] For example, during the operation of a circuit breaker, the real-time collected electrical signal is subjected to FFT to obtain spectrum data, and the extracted characteristic frequencies include significant components of 50Hz and 100Hz. These frequencies correspond to the switching operation and arc extinction process of the circuit breaker. Next, through wavelet packet decomposition, coefficients on multiple time scales are obtained, and the time-frequency energy distribution diagram is calculated, showing that the energy suddenly increases in a certain period of time, indicating that there may be an abnormal arc. These frequency domain features and time-frequency energy distribution information are feature fused to obtain a comprehensive time-frequency feature vector, which is normalized and input into the circuit breaker state classifier. The classifier predicts that the probability of the circuit breaker being in a "mild fault" state is 70% and the confidence level is 85%. According to the results of the maximum likelihood estimation, the preliminary working state assessment shows that the circuit breaker does have a mild fault. Subsequently, based on the fuzzy evaluation index system, the fuzzy factors of "mild wear" and "severe wear" are further combined to finally obtain the health state fuzzy subset. The health state index is 0.65 after defuzzification by the centroid method. According to the set warning threshold, this index indicates that the circuit breaker needs preventive maintenance and the fault warning level is "medium".
[0104] In a specific embodiment, the process of executing step S106 may specifically include the following steps:
[0105] (1) Feature extraction is performed on the newly collected circuit breaker operation data to obtain an incremental learning sample set, and the incremental learning sample set is merged with the original training data set to obtain an extended data set;
[0106] (2) Performing data balancing on the extended data set to obtain a balanced sample set, and updating the parameters of the circuit breaker state classifier based on the balanced sample set to obtain an updated circuit breaker state classifier;
[0107] (3) Evaluate the performance of the updated circuit breaker state classifier to obtain the classifier performance index, and dynamically adjust the classifier structure according to the classifier performance index to obtain the optimized circuit breaker state classifier;
[0108] (4) Perform feature analysis on the newly emerged fault mode to obtain a new fault feature descriptor, and add the new fault feature descriptor to the existing fault mode library to obtain an extended fault mode library;
[0109] (5) updating the adaptive circuit breaker state determination criterion according to the extended fault mode library to obtain a revised circuit breaker state determination criterion, and performing sensitivity analysis on the determination threshold in the revised circuit breaker state determination criterion to obtain the threshold adjustment range;
[0110] (6) Dynamically adjusting the threshold of the modified circuit breaker state determination criterion according to the threshold adjustment range to obtain an optimized state determination threshold, and combining the optimized state determination threshold with the optimized circuit breaker state classifier to obtain an updated circuit breaker state identification parameter set.
[0111] Specifically, features are extracted from the newly collected circuit breaker operation data to generate an incremental learning sample set. Feature extraction includes identifying key parameters in the circuit breaker operation process from electrical signals, such as the contact resistance change rate, arc extinguishing voltage peak, etc. These features represent different aspects of the circuit breaker operation state, and after extraction, a new sample set is formed. The incremental learning sample set is merged with the original training data set to obtain an extended data set, which contains the historical data of the circuit breaker operation and the newly collected operation data. The extended data set is subjected to data balancing. Data balancing is to solve the problem of data imbalance between different state categories. Usually, in the fault detection process, the amount of data in the normal state is much larger than the amount of data in the fault state. Therefore, it is necessary to balance the data by undersampling or oversampling methods so that the amount of data in each state category is roughly equal. This can prevent the classifier from being overly biased towards the normal state and ignoring the identification of the fault state. After data balancing, a balanced sample set is obtained. The balanced sample set is used to update the parameters of the circuit breaker state classifier. Parameter updating refers to readjusting the model parameters in the classifier according to the new data, such as the hyperplane parameters in the support vector machine or the weights in the neural network. Through parameter updating, the classifier can better adapt to the latest operating data, thereby improving the prediction ability of the circuit breaker status and obtaining an updated circuit breaker status classifier.
[0112] In order to evaluate the effect of the updated classifier, a performance evaluation is performed on it and the performance indicators of the classifier are calculated. These indicators usually include accuracy, precision, recall and F1 value, which can quantify the performance of the classifier in different states. According to the performance evaluation results, the structure of the classifier is dynamically adjusted. The dynamic adjustment methods include increasing or decreasing the number of classification layers, adjusting the regularization parameters or updating the kernel function, so that the classifier can still maintain efficient state recognition capabilities under complex circuit breaker operating conditions. After the adjustment is completed, the optimized circuit breaker state classifier is obtained. At the same time, for the newly emerging fault mode, the feature descriptor of the new fault is extracted through feature analysis. For example, a new fault frequency peak or abnormal operating energy consumption is found during a certain operation. This information can be extracted through spectrum analysis and time-frequency analysis to form a new fault feature descriptor. The new fault feature descriptor is added to the existing fault mode library. The expansion of the fault mode library can cover more fault types and manifestations, providing more abundant reference data for the classifier. According to the expanded fault mode library, the adaptive circuit breaker state judgment criterion is updated. The state judgment criterion is used to evaluate the circuit breaker state in real time and determine its working state in combination with the latest fault mode. The update of the judgment criteria is mainly reflected in the introduction of new failure modes and the adjustment of the judgment criteria of the original state categories. Subsequently, a sensitivity analysis is performed on the judgment thresholds in the updated judgment criteria. Sensitivity analysis is used to evaluate the impact of different threshold settings on the state classification results, and the adjustment range of the threshold is determined through experiments to ensure the applicability and flexibility of the judgment criteria in the new operating environment.
[0113] Based on the threshold adjustment range, the modified circuit breaker state judgment criterion is dynamically adjusted to obtain the optimized state judgment threshold. The dynamic threshold adjustment enables the judgment criterion to be adaptively adjusted according to the real-time changes in the circuit breaker operation, ensuring the accuracy and flexibility of the state assessment. Finally, the optimized state judgment threshold is combined with the optimized circuit breaker state classifier to form an updated circuit breaker state identification parameter set. The parameter set includes classifier model parameters, fault mode feature descriptors, judgment criteria, and state thresholds, which can more accurately and dynamically identify the operating state of the circuit breaker.
[0114] For example, during a certain operation, the newly collected operation data of the circuit breaker showed that its operating energy consumption peak had a significant fluctuation compared with the previous one. Through feature extraction, an incremental learning sample set was generated and merged with the original data set to form an extended data set. Since there were fewer fault samples in the newly collected data, the data balancing process oversampled the extended data set so that the amount of data for each fault state was balanced. The classifier updated its parameters through this extended data set, and the accuracy of the classifier increased from the original 92% to 95%. Next, after performance evaluation, the classifier performed well in the F1 value of "mild fault" and "serious fault" classification. To further optimize the classifier, its structure was dynamically adjusted, the number of layers of the classifier was increased, and the kernel function was optimized so that it could better capture the fault mode under the new operating conditions. In the fault mode analysis, a new fault feature, that is, a new arc voltage fluctuation feature, was extracted and added to the fault mode library. Finally, the updated state judgment criterion was dynamically threshold adjusted. The adjusted judgment criterion was more sensitive to the new fault mode, and combined with the optimized classifier, an updated state recognition parameter set was formed.
[0115] The embodiment of the present invention also provides a circuit breaker status identification system based on real-time electrical signals, such as Figure 2 As shown, the circuit breaker state identification system based on real-time electrical signals specifically includes:
[0116] The acquisition module 201 is used to acquire multi-source electrical signals from the contact system, arc extinguishing chamber and operating mechanism of the circuit breaker to obtain original state signals;
[0117] An extraction module 202 is used to perform adaptive threshold denoising and circuit breaker characteristic frequency extraction on the original state signal to obtain a circuit breaker characteristic signal spectrum, and perform time-varying parameter analysis on the circuit breaker characteristic signal spectrum to obtain a circuit breaker state characteristic vector;
[0118] An adjustment module 203 is used to perform multi-dimensional clustering processing on the circuit breaker state feature vector to obtain a circuit breaker state space, and dynamically adjust the boundary of the circuit breaker state space to obtain an adaptive circuit breaker state determination criterion with a state transition probability;
[0119] A processing module 204, configured to process the adaptive circuit breaker state determination criteria and historical operation data by using a support vector machine algorithm to obtain a circuit breaker state classifier;
[0120] An input module 205 is used to perform fast Fourier transform and wavelet packet decomposition on the real-time collected circuit breaker electrical signal to obtain time-frequency features, and input the time-frequency features into the circuit breaker state classifier to obtain the current circuit breaker working state evaluation result, and perform fuzzy comprehensive evaluation on the working state evaluation result to obtain the circuit breaker health state index and fault warning level;
[0121] The updating module 206 is used to perform online incremental learning and fault mode update on the circuit breaker state classifier, and dynamically adjust the determination threshold in the adaptive circuit breaker state determination criterion to obtain an updated circuit breaker state identification parameter set and an optimized state determination threshold.
[0122] Through the coordinated work of the above modules and through multi-source electrical signal acquisition and analysis, comprehensive monitoring of the circuit breaker contact system, arc extinguishing chamber and operating mechanism is achieved, which significantly improves the accuracy and reliability of state recognition and overcomes the limitations of the traditional single parameter evaluation method; the adaptive threshold denoising and circuit breaker characteristic frequency extraction technology are used to effectively filter out environmental noise and irrelevant signals, enhance the system's anti-interference ability, and ensure stable performance in complex electromagnetic environments; through multi-dimensional clustering processing and dynamic boundary adjustment, an adaptive circuit breaker state judgment criterion with state migration probability is constructed, so that the state space can be adaptively adjusted as the circuit breaker operating conditions change, overcome the limitations of the traditional fixed threshold judgment method, and improve the flexibility and adaptability of state judgment; the circuit breaker state classifier is constructed using the support vector machine algorithm to realize the intelligent classification of the circuit breaker state, which not only improves the accuracy of state recognition, but also enhances the system's ability to recognize new states; The electrical signal is subjected to fast Fourier transform and wavelet packet decomposition, combined with fuzzy comprehensive evaluation, which can provide the health status index and fault warning level of the circuit breaker in real time, providing an intuitive and reliable decision-making basis for operation and maintenance personnel; through online incremental learning and fault mode updating of the circuit breaker state classifier, the state identification parameter set and judgment threshold are continuously optimized, so that the system can adapt to the state changes of the circuit breaker during long-term operation and maintain continuous high-efficiency performance; based on the evaluation of contact life, arc extinguishing chamber performance and operating mechanism reliability, targeted maintenance suggestions and priority sorting can be provided, providing strong support for the implementation of predictive maintenance, helping to reduce the operation and maintenance costs of circuit breakers and extend their service life; through the comprehensive status evaluation of the contact system, arc extinguishing chamber and operating mechanism, a more comprehensive and accurate description of the health status of the circuit breaker can be given, providing a more scientific and reliable basis for operation and maintenance decisions, helping to improve the safety and reliability of power grid operation, and significantly improving the timeliness and effectiveness of circuit breaker fault warnings.
[0123] The above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the embodiments, a person skilled in the art should understand that the specific implementation modes of the present invention can still be modified or replaced by equivalents, and any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be included in the scope of the claims of the present invention.
Claims
1. A circuit breaker state identification method based on real-time electrical signals, characterized in that: include: Collect multi-source electrical signals from the circuit breaker's contact system, arc extinguishing chamber and operating mechanism to obtain original status signals; Performing adaptive threshold denoising and circuit breaker characteristic frequency extraction on the original state signal to obtain a circuit breaker characteristic signal spectrum, and performing time-varying parameter analysis on the circuit breaker characteristic signal spectrum to obtain a circuit breaker state characteristic vector; Performing multi-dimensional clustering processing on the circuit breaker state feature vector to obtain a circuit breaker state space, and dynamically adjusting the boundary of the circuit breaker state space to obtain an adaptive circuit breaker state determination criterion with state transition probability; Processing the adaptive circuit breaker state determination criteria and historical operation data by a support vector machine algorithm to obtain a circuit breaker state classifier; Performing fast Fourier transform and wavelet packet decomposition on the real-time collected circuit breaker electrical signal to obtain time-frequency features, and inputting the time-frequency features into the circuit breaker state classifier to obtain the current circuit breaker working state evaluation result, and performing fuzzy comprehensive evaluation on the working state evaluation result to obtain the circuit breaker health state index and fault warning level; The circuit breaker state classifier is subjected to online incremental learning and fault mode updating, and the determination threshold in the adaptive circuit breaker state determination criterion is dynamically adjusted to obtain an updated circuit breaker state identification parameter set and an optimized state determination threshold.
2. The circuit breaker state identification method based on real-time electrical signals according to claim 1, characterized in that: The multi-source electrical signal acquisition of the contact system, arc extinguishing chamber and operating mechanism of the circuit breaker to obtain the original state signal includes: Performing high-frequency resistance measurement on the contact system of the circuit breaker to obtain contact resistance time series data, and performing sliding window analysis on the contact resistance time series data to obtain the contact resistance change rate; Performing transient voltage sampling on the arc extinguishing chamber of the circuit breaker to obtain arc extinguishing voltage waveform data, and performing peak value detection and duration calculation on the arc extinguishing voltage waveform data to obtain arc extinguishing characteristic parameters; The current of the operating mechanism of the circuit breaker is collected to obtain an operating mechanism action current curve, and the operating mechanism action current curve is integrated to obtain an operating energy index; Evaluate the degree of wear of the contact system according to the contact resistance change rate to obtain a predicted value of contact life, and estimate the contact replacement time according to the predicted value of contact life to obtain contact maintenance suggestions; Performing a performance analysis on the arc extinguishing chamber according to the arc extinguishing characteristic parameters to obtain an arc extinguishing efficiency index, and classifying the arc extinguishing chamber state according to the arc extinguishing efficiency index to obtain an arc extinguishing chamber working state evaluation result; Performing an action reliability assessment on the operating mechanism according to the operating energy index to obtain an operating mechanism reliability index, and performing a failure risk prediction on the operating mechanism according to the operating mechanism reliability index to obtain an operating mechanism maintenance priority; Performing time synchronization processing on the contact resistance time series data, arc extinguishing voltage waveform data and operating mechanism action current curve to obtain a multi-source signal collaborative analysis data set; Performing multi-scale decomposition on the multi-source signal collaborative analysis data set by wavelet packet transform algorithm to obtain circuit breaker state characteristic components, and performing energy entropy calculation on the circuit breaker state characteristic components to obtain circuit breaker comprehensive state indicators; Performing a comprehensive status assessment on the circuit breaker according to the contact maintenance suggestion, the arc extinguishing chamber working status assessment result and the operating mechanism maintenance priority to obtain a circuit breaker health status descriptor; According to the circuit breaker comprehensive status indicator and the circuit breaker health status descriptor, a global status judgment is performed on the circuit breaker to obtain an original status signal.
3. The circuit breaker state identification method based on real-time electrical signals according to claim 1, characterized in that: The method of performing adaptive threshold denoising and circuit breaker characteristic frequency extraction on the original state signal to obtain a circuit breaker characteristic signal spectrum, and performing time-varying parameter analysis on the circuit breaker characteristic signal spectrum to obtain a circuit breaker state characteristic vector comprises: Performing wavelet threshold denoising on the original state signal to obtain a preliminary denoised signal, and performing signal-to-noise ratio calculation on the preliminary denoised signal to obtain a signal-to-noise ratio index; The noise level is evaluated according to the signal-to-noise ratio index to obtain an adaptive threshold coefficient, and a secondary wavelet threshold denoising is performed on the preliminary noise reduction signal according to the adaptive threshold coefficient to obtain an optimized noise reduction signal; Performing fast Fourier transform on the optimized noise reduction signal to obtain a circuit breaker characteristic signal spectrum, and performing peak detection on the circuit breaker characteristic signal spectrum to obtain a circuit breaker characteristic frequency set; performing bandpass filtering on the optimized noise reduction signal according to the circuit breaker characteristic frequency set to obtain a circuit breaker characteristic signal, and performing Hilbert transform on the circuit breaker characteristic signal to obtain an analytical signal; Performing instantaneous frequency calculation on the analytical signal to obtain a time-frequency distribution diagram, and performing energy density estimation on the time-frequency distribution diagram to obtain a time-varying energy spectrum; Extracting time-frequency features of the circuit breaker state according to the time-varying energy spectrum to obtain a time-frequency feature vector, and performing principal component analysis on the time-frequency feature vector to obtain a reduced-dimensional feature vector; Statistical parameters are calculated for the dimension-reduced feature vector to obtain a statistical feature set, and a circuit breaker state feature space is constructed according to the statistical feature set to obtain a circuit breaker state feature vector.
4. The circuit breaker state identification method based on real-time electrical signals according to claim 1, characterized in that: The multi-dimensional clustering process is performed on the circuit breaker state feature vector to obtain a circuit breaker state space, and the circuit breaker state space is dynamically adjusted to obtain an adaptive circuit breaker state determination criterion with a state transition probability, including: Performing K-means cluster analysis on the circuit breaker state feature vector to obtain an initial state category, and performing silhouette coefficient calculation on the initial state category to obtain a clustering effectiveness index; The clustering results are evaluated according to the clustering effectiveness index to obtain an optimal number of clusters, and the circuit breaker state feature vectors are hierarchically clustered according to the optimal number of clusters to obtain a target state category; Assigning labels to the target state categories to obtain a circuit breaker state space, and extracting boundary points from the circuit breaker state space to obtain an initial state boundary; Constructing a space segmentation diagram of the circuit breaker state space according to the initial state boundary to obtain a state partition diagram, and performing boundary smoothing processing on the state partition diagram to obtain an optimized state boundary; Perform state transition frequency statistics on historical state data to obtain a state transition matrix, and calculate the state transition probability based on the state transition matrix to obtain a state migration probability model; The optimized state boundary is dynamically adjusted according to the state transition probability model to obtain an adaptive state boundary, and a circuit breaker state determination rule is constructed according to the adaptive state boundary to obtain an adaptive circuit breaker state determination criterion.
5. The circuit breaker state identification method based on real-time electrical signals according to claim 1, characterized in that: The method of processing the adaptive circuit breaker state determination criteria and historical operation data by a support vector machine algorithm to obtain a circuit breaker state classifier includes: Parameterizing the adaptive circuit breaker state determination criterion to obtain a feature descriptor, and extracting features from historical operation data according to the feature descriptor to obtain a training data set; Performing data normalization processing on the training data set to obtain standardized training data, and performing principal component analysis on the standardized training data to obtain a reduced dimension feature set; Constructing a kernel function according to the dimension reduction feature set to obtain a nonlinear mapping function, and using the nonlinear mapping function to perform a high-dimensional space transformation on the standardized training data to obtain a transformed feature space; Constructing a maximum margin hyperplane in the transformed feature space to obtain an initial classification boundary, and performing soft margin optimization on the initial classification boundary to obtain an optimized classification hyperplane; Selecting support vectors according to the optimized classification hyperplane to obtain a key sample set, and constructing a decision function based on the key sample set to obtain a multi-category classifier; The multi-category classifier is cross-validated to obtain a classification performance index, and classifier parameters are fine-tuned according to the classification performance index to obtain a circuit breaker state classifier.
6. The circuit breaker state identification method based on real-time electrical signals according to claim 1, characterized in that: The circuit breaker electrical signal collected in real time is subjected to fast Fourier transform and wavelet packet decomposition to obtain time-frequency features, and the time-frequency features are input into the circuit breaker state classifier to obtain the working state evaluation result of the current circuit breaker, and the working state evaluation result is subjected to fuzzy comprehensive evaluation to obtain the circuit breaker health state index and fault warning level, including: Performing fast Fourier transform on the real-time collected circuit breaker electrical signal to obtain spectrum data, and extracting characteristic frequencies from the spectrum data to obtain a frequency domain feature set; Performing wavelet packet decomposition on the real-time collected circuit breaker electrical signal to obtain multi-scale coefficients, and performing energy calculation on the multi-scale coefficients to obtain a time-frequency energy distribution diagram; Performing feature fusion according to the frequency domain feature set and the time-frequency energy distribution diagram to obtain a comprehensive time-frequency feature vector, and normalizing the comprehensive time-frequency feature vector to obtain a standardized feature input; Inputting the standardized features into a circuit breaker state classifier to obtain a circuit breaker state probability distribution, and performing maximum likelihood estimation on the circuit breaker state probability distribution to obtain a preliminary working state assessment result; Performing a confidence analysis on the preliminary working status assessment result to obtain a status assessment reliability index, and revising the preliminary working status assessment result according to the status assessment reliability index to obtain a revised working status assessment result; Constructing a fuzzy evaluation index system to obtain a fuzzy evaluation factor set, and performing a fuzzy comprehensive evaluation on the revised working status evaluation result according to the fuzzy evaluation factor set to obtain a fuzzy subset of the circuit breaker health status; The fuzzy subset of the circuit breaker health state is defuzzified using the centroid method to obtain a circuit breaker health state index, and a warning threshold is set according to the circuit breaker health state index to obtain a fault warning level.
7. The circuit breaker state identification method based on real-time electrical signals according to claim 6, characterized in that: The circuit breaker state classifier is subjected to online incremental learning and fault mode updating, and the determination threshold in the adaptive circuit breaker state determination criterion is dynamically adjusted to obtain an updated circuit breaker state identification parameter set and an optimized state determination threshold, including: Performing feature extraction on the newly collected circuit breaker operation data to obtain an incremental learning sample set, and merging the incremental learning sample set with the original training data set to obtain an extended data set; Performing data balancing processing on the extended data set to obtain a balanced sample set, and updating parameters of a circuit breaker state classifier according to the balanced sample set to obtain an updated circuit breaker state classifier; Performing performance evaluation on the updated circuit breaker state classifier to obtain a classifier performance index, and dynamically adjusting the classifier structure according to the classifier performance index to obtain an optimized circuit breaker state classifier; Performing feature analysis on the newly appeared fault mode to obtain a new fault feature descriptor, and adding the new fault feature descriptor to the existing fault mode library to obtain an extended fault mode library; The adaptive circuit breaker state determination criterion is updated according to the extended fault mode library to obtain a revised circuit breaker state determination criterion, and a sensitivity analysis is performed on the determination threshold in the revised circuit breaker state determination criterion to obtain a threshold adjustment range; The modified circuit breaker state determination criterion is dynamically threshold-adjusted according to the threshold adjustment range to obtain an optimized state determination threshold, and the optimized state determination threshold is combined with the optimized circuit breaker state classifier to obtain an updated circuit breaker state identification parameter set.
8. A circuit breaker state identification system based on real-time electrical signals, used to execute the circuit breaker state identification method based on real-time electrical signals according to any one of claims 1 to 7, characterized in that: include: The acquisition module is used to collect multi-source electrical signals from the contact system, arc extinguishing chamber and operating mechanism of the circuit breaker to obtain the original state signal; An extraction module, used for performing adaptive threshold denoising and circuit breaker characteristic frequency extraction on the original state signal to obtain a circuit breaker characteristic signal spectrum, and performing time-varying parameter analysis on the circuit breaker characteristic signal spectrum to obtain a circuit breaker state characteristic vector; An adjustment module, configured to perform multi-dimensional clustering processing on the circuit breaker state feature vector to obtain a circuit breaker state space, and dynamically adjust the boundary of the circuit breaker state space to obtain an adaptive circuit breaker state determination criterion with a state transition probability; A processing module, used for processing the adaptive circuit breaker state determination criteria and historical operation data by a support vector machine algorithm to obtain a circuit breaker state classifier; An input module is used to perform fast Fourier transform and wavelet packet decomposition on the real-time collected circuit breaker electrical signal to obtain time-frequency features, and input the time-frequency features into the circuit breaker state classifier to obtain the current circuit breaker working state evaluation result, and perform fuzzy comprehensive evaluation on the working state evaluation result to obtain the circuit breaker health state index and fault warning level; The updating module is used to perform online incremental learning and fault mode updating on the circuit breaker state classifier, and dynamically adjust the determination threshold in the adaptive circuit breaker state determination criterion to obtain an updated circuit breaker state identification parameter set and an optimized state determination threshold.
Citation Information
Patent Citations
Circuit breaker fault diagnosis method for sound and vibration signal fusion processing
CN107796602A
Medium-voltage vacuum circuit breaker mechanical fault diagnosis method based on multi-source data fusion
CN110929763A
