A metering circuit breaker health monitoring method and device and metering circuit breaker
By determining the fault accumulation sequence and contact aging sequence of the circuit breaker, and using the LSTM model to train electrical data samples and adjust the weights, the problem of inaccurate prediction in the health monitoring of metered circuit breakers is solved, and accurate prediction of the aging state of the circuit breaker is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- TUV RHEINLAND SHANGHAI
- Filing Date
- 2026-03-10
- Publication Date
- 2026-06-16
AI Technical Summary
The accuracy of health monitoring prediction results for metering circuit breakers in the existing technology is low, mainly due to the lack of fault samples, which prevents the model from learning abnormal features, and there is also the problem of overfitting.
By determining the fault accumulation sequence and contact aging sequence of the circuit breaker, an LSTM model is used to train electrical data samples, the sample weights are adjusted to correct the importance, and a support vector machine is combined to optimize the error change trend, thereby achieving accurate prediction of the health status of the circuit breaker.
It improves the accuracy of circuit breaker aging prediction, solves the problems of overlearning and slow convergence speed in traditional methods, and realizes accurate monitoring of the health status of circuit breakers.
Smart Images
Figure CN122220872A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of electrical variable measurement technology, specifically to a method, device, and metering circuit breaker for health monitoring. Background Technology
[0002] A metering circuit breaker is an intelligent power device that integrates power protection and energy metering functions. It enables intelligent electricity metering and billing, power monitoring, and energy consumption management. Since the performance stability of circuit breakers directly affects the safety, economy, and reliability of the power system, health monitoring of metering circuit breakers allows for real-time assessment of the metering element's status, enabling early detection of anomalies or evaluation of component conditions.
[0003] With the integration of artificial intelligence and traditional electrical components, sensor-based edge computing is being used to replace traditional methods that rely solely on key parameter thresholds and are only applicable to small-scale scenarios, depending on experience-based judgment. Therefore, machine learning is currently commonly used to monitor circuit breaker data by training on samples.
[0004] Health monitoring of circuit breakers cannot rely on only one type of sample. Due to the long service life of metered circuit breakers, there are situations where they do not need to be replaced for extended periods. This results in a large amount of normal sample data collected from metered circuit breakers, while samples of faulty circuit breakers are lacking or even nonexistent. Consequently, the abnormal characteristics of circuit breakers cannot be learned during model training. Furthermore, overlearning can occur with normally functioning circuit breakers, leading to overfitting of the entire model. All of these factors contribute to the inaccurate health monitoring predictions of metered circuit breakers, causing them to deviate from the actual health status. Summary of the Invention
[0005] To address the problem of low accuracy in existing methods for predicting the health of metered circuit breakers using artificial intelligence, this invention aims to provide a method, apparatus, and metered circuit breaker for health monitoring. The specific technical solution adopted is as follows: This invention provides a method for health monitoring of metering circuit breakers, the method comprising: Based on electrical data samples, determine the fault accumulation sequence and contact aging sequence of the circuit breaker, and use the fault accumulation sequence to determine the importance of the target circuit breaker for circuit breakers with contact faults. The importance of the circuit breaker is corrected by using the fitting error between the fault accumulation sequence and the contact aging sequence. The corrected importance of the target circuit breaker and the contact aging value are used as its circuit breaker characteristics. The initial weights of electrical data samples are determined by utilizing the similarity of circuit breaker features between the target circuit breaker and the other circuit breakers. The electrical data samples are trained using an LSTM model based on the initial weights to obtain the error change trend of the electrical data samples. The training weights of the electrical data samples are updated using the error change trend until the target error rate of the LSTM model is reached; The current electrical data of the circuit breaker is input into a trained LSTM model to predict contact aging parameters for future time periods.
[0006] Further, determining the fault accumulation sequence and contact aging sequence of the circuit breaker based on the electrical data sample includes: Based on the electrical data samples and the corresponding fault log files, determine the cumulative average fault rate up to the corresponding electrical data sample according to the sample time sequence; The cumulative fault sequence of the circuit breaker is obtained by arranging and combining the cumulative average fault rate of each electrical data sample according to the time sequence of the samples. Determine the contact aging value corresponding to each electrical data sample, and obtain the contact aging sequence of the circuit breaker based on the contact aging values arranged and combined in the time sequence of the samples.
[0007] Furthermore, the determination of the target circuit breaker's importance for circuit breakers with contact faults using the fault accumulation sequence includes: The minimum fault accumulation sequence is obtained by extracting the minimum value of each element with the same index in the fault accumulation sequence of all circuit breakers. The importance of the target circuit breaker for contact faults can be obtained by using the Euclidean distance between the corresponding elements of the fault accumulation sequence of the target circuit breaker and the minimum fault accumulation sequence.
[0008] Furthermore, the method of correcting the circuit breaker importance using the fitting error between the fault accumulation sequence and the contact aging sequence includes: Determine the two-dimensional data points composed of corresponding elements of the fault accumulation sequence and contact aging sequence of the target circuit breaker; The fitting error of all electrical data samples of the target circuit breaker is obtained by inputting the two-dimensional data points into the preset fitting polynomial; The importance of the circuit breaker is corrected by using the fitting error of all electrical data samples of the target circuit breaker and the overall fitting error of all circuit breakers.
[0009] Furthermore, the step of determining the initial weights of electrical data samples by utilizing the circuit breaker feature similarity between the target circuit breaker and the remaining circuit breakers includes: Determine the differences in modified circuit breaker importance and contact aging sequence between the target circuit breaker and the other circuit breakers; The similarity of characteristics of each circuit breaker is obtained by correcting the differences in importance of the circuit breakers and the differences in the contact aging sequence. The ratio between the standard deviation and the mean of all circuit breaker feature similarities is taken as the coefficient of variation of the circuit breaker feature similarity set; The initial weights of electrical data samples are determined by using the coefficient of variation of the circuit breaker feature similarity set.
[0010] Furthermore, determining the initial weights of electrical data samples using the coefficient of variation of the circuit breaker feature similarity set includes: By utilizing the fitting error between the fault accumulation sequence and the contact aging sequence, the importance of each electrical data sample to the contact fault sample is determined; By utilizing the importance of the samples and their corresponding contact aging values, the unit aging value of the electrical data samples is determined. The initial weights of electrical data samples are determined by using the coefficient of variation of the circuit breaker feature similarity set and the unit aging value of the electrical data samples.
[0011] Furthermore, the step of training the LSTM model on the electrical data samples based on the initial weights to obtain the error change trend of the electrical data samples includes: The initial error of the electrical data samples is obtained by training the LSTM model based on the initial weights. The training weights of the next electrical data samples are then updated using the initial error and the initial weights. Using the training weights of the next electrical data sample as the initial weights, the step of training the LSTM model on the electrical data sample based on the initial weights is executed repeatedly to obtain the initial error of the electrical data sample, thereby obtaining the error change trend of the electrical data sample.
[0012] Furthermore, updating the training weights of the electrical data samples using the error change trend until the target error rate of the LSTM model is reached includes: Based on a class of support vector machines, electrical data samples that are identified as outliers are determined, and the error difference value of the outliers is determined during multiple LSTM model training processes. The average value of the error difference and the number of samples with a decreasing error trend are determined by using the error difference values of the outlier samples. The consistency of outlier samples is determined by the average value of the error difference and the number of samples with a downward trend. The training weights of the electrical data samples are then updated based on the consistency of the outlier samples until the target error rate of the LSTM model is reached.
[0013] The present invention also provides a metering circuit breaker health monitoring device, the device being used to implement the metering circuit breaker health monitoring method as described in any of the preceding claims; the device comprising: The fault correlation analysis module is used to determine the fault accumulation sequence and contact aging sequence of the circuit breaker based on electrical data samples. It uses the fault accumulation sequence to determine the circuit breaker importance of the target circuit breaker for contact faults. It uses the fitting error between the fault accumulation sequence and the contact aging sequence to correct the circuit breaker importance. The corrected circuit breaker importance and contact aging value of the target circuit breaker are used as its circuit breaker characteristics. The sample weight adjustment module is used to determine the initial weights of electrical data samples by utilizing the circuit breaker feature similarity between the target circuit breaker and other circuit breakers; to train the LSTM model on the electrical data samples based on the initial weights to obtain the error change trend of the electrical data samples; and to update the training weights of the electrical data samples using the error change trend until the target error rate of the LSTM model is reached. The model health prediction module is used to input the current electrical data of the circuit breaker into the trained LSTM model to predict the contact aging parameters for future time periods.
[0014] The present invention also provides a metering circuit breaker, the metering circuit breaker including a processor, a memory, and a metering circuit breaker health monitoring program stored in the memory and executable by the processor, wherein when the metering circuit breaker health monitoring program is executed by the processor, it implements the steps of the metering circuit breaker health monitoring method as described in any of the preceding claims.
[0015] The present invention has the following beneficial effects: This invention solves the problem in the traditional training process by determining the weight of the samples. Because there are few or no aging circuit breaker samples, the features of aging samples cannot be learned, while the features of normal circuit breaker samples are overlearned.
[0016] By determining the weights of the samples, this approach addresses the problem that traditional methods for predicting aging rely solely on changes in samples from the most recent time period, while historical circuit breaker failures can also affect subsequent sample changes.
[0017] By adjusting the sample weights during the iteration process, the problem of relying solely on the weight changes of the initial samples or traditional samples to effectively change the training model of LSTM (Long Short-Term Memory) network is solved. Furthermore, by adjusting the sample error during the iteration process, the problem of determining the weights without changing the actual prediction results caused by traditional training is solved.
[0018] This solves the problems of inaccurate prediction and low convergence speed of traditional LSTM models, and improves the accuracy of circuit breaker aging prediction. Attached Figure Description
[0019] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0020] Figure 1 A flowchart illustrating the steps of a metering circuit breaker health monitoring method according to an embodiment of the present invention; Figure 2 This is a detailed flowchart of step S1 in a metering circuit breaker health monitoring method provided in an embodiment of the present invention. Figure 3 This is a detailed flowchart of step S2 in a metering circuit breaker health monitoring method provided in an embodiment of the present invention. Figure 4 This is a detailed flowchart of step S3 in a metering circuit breaker health monitoring method provided in an embodiment of the present invention. Figure 5 This is a detailed flowchart of step S4 in a metering circuit breaker health monitoring method provided in an embodiment of the present invention. Figure 6 This is a detailed flowchart of step S5 in a metering circuit breaker health monitoring method provided in an embodiment of the present invention. Figure 7 This is a schematic diagram of the hardware operating environment of the metering circuit breaker involved in the embodiments of the present invention; Figure 8 This is a schematic diagram of the frame structure of the metering circuit breaker health monitoring device involved in the embodiment of the present invention. Detailed Implementation
[0021] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a metering circuit breaker health monitoring method proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.
[0022] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0023] The specific scheme of the metering circuit breaker health monitoring method provided by the present invention will be described in detail below with reference to the accompanying drawings.
[0024] Example 1: For a metering circuit breaker health monitoring method provided by this invention, please refer to [link / reference]. Figure 1 The diagram illustrates a flowchart of a metering circuit breaker health monitoring method according to an embodiment of the present invention.
[0025] The method includes: Step S1: Determine the fault accumulation sequence and contact aging sequence of the circuit breaker based on the electrical data sample, and use the fault accumulation sequence to determine the importance of the target circuit breaker for circuit breakers with contact faults. Because LSTM (Long Short-Term Memory) models can handle time-series data well, and can predict future circuit breaker states based on the most recent circuit breaker state for metered health monitoring, LSTM is relatively accurate for health prediction in circuit breaker health monitoring.
[0026] The system collects electrical quantity data such as current and voltage from circuit breakers to obtain time-series electrical data (electrical time-series data). Based on this data, a window size of 30 (adjustable) is selected to divide the historical electrical data of the circuit breakers into multiple time-series samples. This process is repeated for each circuit breaker to obtain multiple time-series related electrical data samples (referred to as "samples"). During data collection, the system analyzes the circuit breaker's fault log file to determine if a fault occurred in the corresponding sample. This can be represented by 0 or 1, for example, 0 indicating no fault and 1 indicating a fault. The proportion of fault occurrences to the total sample data is used as the fault rate for that sample. It should be noted that at this point, only the existence of a fault is recorded, and the type of fault cannot be determined.
[0027] This section uses LSTM to predict the aging condition of circuit breaker contacts. Traditionally, the assessment of contact aging relies primarily on the physical structural characteristics of the circuit breaker, such as tripping turns and the open / closed state of the switch, to determine whether contact replacement is necessary, thus enabling health monitoring of this component. This traditional method can determine the aging degree of each sample and quantify it numerically (e.g., setting a range of 1%-100%) as the contact aging value. Alternatively, the contact aging value for each sample can be obtained through expert evaluation. .
[0028] Therefore, each sample can be considered to include monitoring data such as current and voltage collected by the sensor, as well as the corresponding contact aging value.
[0029] The contact aging values of the circuit breaker are different at different time periods of its operation. In the early stage, the overall performance of the circuit breaker is good because it has just started to be used, so the contact aging values are relatively small. As the operating time increases, the contacts will show a certain degree of aging. The aging value can be used to determine whether the component or the circuit breaker needs to be replaced, which is very consistent with the characteristics of the LSTM model.
[0030] However, the following problems exist: Most circuit breaker contacts do not typically reach their normal service life. Instead, they fail due to the interplay of faults among the circuit breaker's components, necessitating contact replacement. Therefore, if a circuit breaker has a history of failures, it is more likely to fail again in the future, accelerating its aging process. Thus, the weight of a sample is not necessarily greater the closer it is to the current time (i.e., the short-term memory of LSTM). If a circuit breaker experienced a failure during its historical operation, even though the time elapsed is considerable, it still holds significant reference value for the current sample.
[0031] Specifically, please refer to Figure 2 Step S1, determining the fault accumulation sequence and contact aging sequence of the circuit breaker based on the electrical data sample, includes: Step S11: Based on the electrical data samples and the corresponding fault log files, determine the cumulative average fault rate up to the corresponding electrical data sample according to the sample time sequence. Step S12: Based on the cumulative average failure rate of each electrical data sample, the failure cumulative sequence of the circuit breaker is obtained by arranging and combining the samples according to the time sequence. Step S13: Determine the contact aging value corresponding to each electrical data sample, and obtain the contact aging sequence of the circuit breaker based on the contact aging value arranged and combined according to the sample time sequence.
[0032] For multiple time-series samples of the circuit breaker, the cumulative fault value of the time-series samples is calculated. If the circuit breaker frequently experiences faults, its contacts are relatively more susceptible to the effects of other faults in the circuit breaker, leading to contact abnormalities. The cumulative fault value of the samples is obtained. . This represents the mean of the failure rates of all samples before the sample to be analyzed for the circuit breaker (i.e., the cumulative average failure rate up to the corresponding electrical data sample), thereby determining the cumulative failure rate of each sample.
[0033] Since the increase in sample failure rate with increasing usage time is normal, the cumulative failure rate also shows the same trend. Therefore, we choose to number the time sequence corresponding to each sample of the circuit breaker, with the horizontal axis representing the sample sequence number and the vertical axis representing the corresponding cumulative failure rate. Thus, the sample can be represented as follows: ki represents any i-th sample from any k-th circuit breaker. This represents the cumulative failure rate of the sample, i.e., the cumulative average failure rate of the sample.
[0034] The above analysis was performed on all samples of all circuit breakers. Scatter plot data of these samples were obtained (the horizontal axis represents the sample order). The vertical axis represents fault accumulation. The minimum fault accumulation sequence is obtained based on the scatter data of the samples. (This refers to the entire sample, i.e., all samples from all circuit breakers.) Minimum Fault Cumulative Sequence This reflects the minimum cumulative change in faults exhibited by these circuit breakers during normal use. The methods used to obtain this data specifically include: Select the sample with the lowest failure rate in the same scatter plot order (i.e., fix the horizontal axis and select the sample with the lowest failure rate on the vertical axis), then iterate through the horizontal axis to obtain the corresponding sequence, which is used as the sequence. This sequence is used as a baseline sequence for fault changes over time. The same analytical method is used to obtain a baseline aging sequence for the contact aging values corresponding to each electrical data sample. It should be noted that, in addition to the minimum fault accumulation sequence here... and baseline aging sequence There are also the individual fault accumulation sequences for each circuit breaker k. and contact aging sequence ,sequence and sequence All data are derived from multiple electrical data samples of circuit breaker k arranged in a time sequence.
[0035] Furthermore, specifically, step S1, which uses the fault accumulation sequence to determine the importance of the target circuit breaker for circuit breaker faults involving contact failures, includes: The minimum fault accumulation sequence is obtained by extracting the minimum value of each element with the same index in the fault accumulation sequence of all circuit breakers. The importance of the target circuit breaker for contact faults can be obtained by using the Euclidean distance between the corresponding elements of the fault accumulation sequence of the target circuit breaker and the minimum fault accumulation sequence.
[0036] Based on the minimum fault accumulation sequence obtained from the above embodiments, if circuit breaker k (as the target circuit breaker) is in normal operation over time, then the fault accumulation of circuit breaker k is similar to the minimum fault accumulation sequence described above. There is a high degree of consistency. If the fault accumulation sequence of circuit breaker k... Relative to the minimum fault accumulation sequence The presence of a clear upward shift indicates that the sample data for circuit breaker k shows a more pronounced effect on contact faults. Therefore, the importance of this circuit breaker relative to contact faults is determined, denoted as the circuit breaker importance. : In the formula, Represents a sequence The cumulative average failure rate of the i-th element (corresponding to the i-th sample). Represents a sequence The cumulative average failure rate of the i-th element with the same index in the table. The difference represents the cumulative average failure rate of the same i-th element, which can be the Euclidean distance between the two; N indicates that there are N elements in both sequences.
[0037] Step S2: Correct the importance of the circuit breaker by using the fitting error between the fault accumulation sequence and the contact aging sequence, and use the corrected circuit breaker importance and contact aging value of the target circuit breaker as its circuit breaker characteristics. Specifically, please refer to Figure 3 Step S2, which uses the fitting error between the fault accumulation sequence and the contact aging sequence to correct the importance of the circuit breaker, includes: Step S21: Determine the two-dimensional data points composed of the corresponding elements of the fault accumulation sequence and contact aging sequence of the target circuit breaker; Step S22: Input the two-dimensional data points into the preset fitting polynomial to obtain the fitting error of all electrical data samples of the target circuit breaker; Step S23: Correct the importance of the circuit breaker by using the fitting error of all electrical data samples of the target circuit breaker and the overall fitting error of all circuit breakers.
[0038] Based on the above embodiments, the circuit breaker faults described above only consider the changes in the fault itself and do not combine them with the impact of circuit breaker faults on contact aging. Therefore, the importance of the circuit breaker needs to be revised.
[0039] For example, the degree and type of circuit breaker failure have different effects on contact aging. The most obvious effect is the generation of electric arcs in the circuit breaker contacts. The presence of electric arcs leads to a large amount of discharge in a short period of time, resulting in high voltage or voltage change rate, which in turn leads to circuit breaker failure. Other faults have a weaker impact on circuit breaker contact aging.
[0040] Therefore, the circuit breaker k Substituting this into the fitting polynomial c of all samples for circuit breaker k, here... This indicates the sequence of circuit breakers k. and sequence Each pair of elements with the same index in the two-dimensional data points is input into the fitting polynomial c to obtain the corresponding fitting error. (The method for calculating the error in polynomial fitting is an existing technique, such as obtaining the corresponding polynomial through the least squares method after determining the order of a polynomial.) This fitting error The error can be either mean square error or root mean square error. If the fitting error increases significantly, it indicates that there is a significant difference between the fault performance and aging performance of the circuit breaker k and its normal performance. This means that the aging characteristics of the circuit breaker contacts are more obvious, and the correlation between the faults and contact aging is stronger. Therefore, the importance of the circuit breaker in this sample is also greater.
[0041] Then, based on the minimum fault sequence f and the benchmark aging sequence Y, the overall fitting error corresponding to all samples of all circuit breakers k is obtained by fitting. This reflects the impact of faults occurring during the normal operation of the circuit breaker on aging. The above-mentioned polynomial fitting is also used to obtain the corresponding overall fitting error. .
[0042] Obtain the corrected importance of the circuit breaker (corrected circuit breaker importance) : In the formula, To correct the importance of the k-th circuit breaker, Let k be the importance of the target circuit breaker, i.e., the importance of the circuit breaker before correction. This represents the activation function (i.e., the Rectified Linear Unit). This represents the fitting error of the target circuit breaker k. This indicates that the overall fitting error of all circuit breakers is combined, and the +1 is to prevent the activation function from being zero.
[0043] To avoid the situation where there are too few aging samples, and because many aging samples with a certain degree of aging appear as normal samples, it is necessary to further analyze whether the samples of each circuit breaker have the same characteristics as the samples of the other circuit breakers.
[0044] Step S3: Determine the initial weights of the electrical data samples by utilizing the circuit breaker feature similarity between the target circuit breaker and the other circuit breakers. Specifically, please refer to Figure 4 Step S3 includes: Step S31: Determine the differences in modified circuit breaker importance and contact aging sequence between the target circuit breaker and the other circuit breakers; Step S32: Obtain the characteristic similarity of each circuit breaker by correcting the differences in the importance of the circuit breakers and the differences in the contact aging sequence. Step S33: The ratio between the standard deviation and the mean of all circuit breaker feature similarities is used as the coefficient of variation of the circuit breaker feature similarity set. Step S34: Use the coefficient of variation of the circuit breaker feature similarity set to determine the initial weights of the electrical data samples.
[0045] Based on the above embodiments, each circuit breaker k is passed through This indicates the importance of the modified circuit breaker k for the target circuit breaker. and sequence The aging values of each contact in the circuit breaker are used as the circuit breaker characteristics of circuit breaker k. The similarity between circuit breaker characteristics is compared: This indicates the similarity of circuit breaker characteristics between any two circuit breakers of the kth and lth denots. () represents the positive correlation normalization function with a range of [0,1]. To correct the importance of the target circuit breaker k, The difference between the two values represents the corrected importance of circuit breaker l, and the absolute value of the corrected importance difference is taken. It can represent both the contact aging sequence of the target circuit breaker k and each element in that sequence, i.e., the contact aging value. Let l represent the contact aging sequence and its individual elements, and calculate the Euclidean distance between the two contact aging sequences.
[0046] Fixed circuit breaker to be analyzed By combining the characteristics of each circuit breaker, the circuit breaker can be obtained. The set of feature similarities with all other circuit breakers. The ratio between the standard deviation and the mean of all circuit breaker feature similarities is calculated as the coefficient of variation for this set. ,Should The larger the value, the more significant the feature difference between circuit breaker k and the other circuit breakers. This indicates that the number of features of circuit breaker k is seriously insufficient, and the weight of the circuit breaker (electrical data) samples needs to be adjusted to avoid ignoring such samples with certain aging phenomena and a small number of features during LSTM training.
[0047] More specifically, step S34 includes: By utilizing the fitting error between the fault accumulation sequence and the contact aging sequence, the importance of each electrical data sample to the contact fault sample is determined; By utilizing the importance of the samples and their corresponding contact aging values, the unit aging value of the electrical data samples is determined. The initial weights of electrical data samples are determined by using the coefficient of variation of the circuit breaker feature similarity set and the unit aging value of the electrical data samples.
[0048] Based on the above embodiments, the following were obtained: Let represent the modified circuit breaker importance of the k-th circuit breaker, which represents the importance of all samples of the k-th circuit breaker. For each sample, the importance of sample i is calculated (denoted as sample importance). : This indicates the importance of the target circuit breaker. Indicates the activation function; This indicates that the kth circuit breaker... The importance of a sample reflects how well that sample performs in relation to the fault. This represents the fitting error of the target circuit breaker k (which combines all samples of that circuit breaker). The fitting error of sample i using target circuit breaker k is also obtained through the polynomial fitting described above.
[0049] From this, the contact aging sequence of circuit breaker k can be obtained. and fault accumulation sequence The importance sequence of samples between Contact aging sequence analysis was performed using sample importance sequence analysis. and fault accumulation sequence The consistency of the sample importance sequence is crucial. If a sample in the sequence maintains consistency with the contact aging sequence of the circuit breaker, it indicates that the fault is more likely to be directly related to contact aging. For example, during opening or closing, the arc duration is prolonged, and the arc energy is significantly increased, manifested as a higher peak arc voltage and a longer current duration. Because contact aging leads to unstable contact separation speed during opening or increased contact bounce during closing, the arc is difficult to extinguish quickly, and the increased arc energy further erodes the contacts, accelerating contact aging.
[0050] Because the importance sequence of samples is represented by fault frequency and historical fault accumulation, and only records the presence or absence of faults without analyzing the fault types of the samples, and because circuit breaker contact aging often cannot be effectively represented by fault indicators (e.g., significant anomalies in the circuit breaker contacts are required for detection), even if the circuit breaker currently appears fault-free, it may still exhibit a certain degree of aging. This results in a lower importance for circuit breaker samples, but higher aging values, making these samples seem less representative of the circuit breaker's aging process. Therefore, it is necessary to calculate the unit aging value for the (electrical data) samples. : This represents the contact aging value of the i-th sample of circuit breaker k (in the contact aging sequence). middle), This indicates that the kth circuit breaker... The importance of each sample is determined by... To prevent the denominator from being 0.
[0051] Obtain the unit aging value for each sample of each circuit breaker to obtain the total unit aging value. For unit aging value Anomaly detection using ocsvm (One-Class Support Vector Machine) was performed to obtain circuit breaker samples exhibiting abnormal behavior, i.e., anomalous samples (existing technology). The anomalousness of each sample within the circuit breaker was then analyzed. Clearly, different operating states of the circuit breaker itself can affect its aging process. Therefore, it is necessary to analyze the importance of each circuit breaker in the LSTM training process and the differences in anomaly detection results among different circuit breakers.
[0052] This yields the initial weight value for each sample i. : The coefficient of variation represents the set of similarity features of circuit breakers. This indicates the unit aging value of the (electrical data) sample.
[0053] The initial weights of the samples calculated above The sample weights are used as the initial weights for training the LSTM.
[0054] Step S4: Train the LSTM model on the electrical data samples based on the initial weights to obtain the error change trend of the electrical data samples; Specifically, please refer to Figure 5 Step S4 includes: Step S41: Train the LSTM model on the electrical data samples based on the initial weights to obtain the initial error of the electrical data samples, and use the initial error and initial weights to update the training weights of the next electrical data samples. Step S42: Using the training weights of the next electrical data sample as the initial weights, the step of training the LSTM model on the electrical data sample based on the initial weights to obtain the initial error of the electrical data sample is executed repeatedly to obtain the error change trend of the electrical data sample.
[0055] In this embodiment, because important samples may be forgotten during training, their impact on the model gradually decreases as the number of iterations increases, even when they reappear in subsequent LSTM training. This slows down the model's convergence rate but does not actually change the training results. Therefore, it is necessary to dynamically adjust the training weights of the samples.
[0056] The specific process is as follows: Based on the initial weights in the above embodiments The sample is used for training to obtain the initial error of the sample. .
[0057] The initial error and initial weights are used to update the training weights for the next sample, thus obtaining the training weights for the next sample. .
[0058] Repeat the above process N times for training. Obtain the error, weights, and corresponding labeling results for each sample during each training iteration. Let u be the u-th round, represented by data pairs. This represents the training weight of the previous sample in each training iteration. The label is represented by -1 and 1, where label=-1 represents a normal sample and label=1 represents an abnormal sample. This label will be further explained in the following examples.
[0059] After N iterations, the error trend is obtained and analyzed. This error trend can be represented by subtracting the previous error from the previous one, and the mean of the differences is calculated as the error trend. Simultaneously record the positive and negative signs, denoted as . .
[0060] Step S5: Update the training weights of the electrical data samples using the error change trend until the target error rate of the LSTM model is reached. Step S6: Input the current electrical data of the circuit breaker into the trained LSTM model to predict the contact aging parameters for future time periods.
[0061] Specifically, please refer to Figure 6 Step S5 includes: Step S51: Based on a support vector machine, determine the electrical data samples as anomalous samples, and determine the error difference value of the anomalous samples in multiple LSTM model training processes. Step S52: Determine the average value of the error difference and the number of samples with a decreasing error trend when the error change trend is decreasing by using the error difference values of the abnormal samples. Step S53: Determine the consistency of outlier samples by using the average value of the error difference and the number of samples with a downward trend, and update the training weights of electrical data samples by using the consistency of outlier samples until the target error rate of the LSTM model is reached.
[0062] Based on the anomaly detection results of ocsvm in the above embodiments, abnormal samples are obtained. These abnormal samples reflect a smaller number of circuit breakers with more significant actual aging characteristics. These samples are labeled, including -1 (normal sample) and 1 (abnormal sample).
[0063] Taking anomalous samples as an example, for all samples labeled with label=1 (anomalous samples), we determine whether the training process has received attention by analyzing whether the error of such samples shows a decreasing trend. If it decreases, it indicates that such samples have been valued during training; if it increases, they have been ignored. However, the following phenomenon may occur: the actual training is not affected by the aforementioned anomalous samples, but rather by the additional attention paid to normal samples. The LSTM training process on normal samples may cause some anomalous samples to also show a decreasing trend, resulting in misclassification. Therefore, it is necessary to analyze the consistency of these anomalous samples. : Positive and negative signs are denoted as , Samples that show a downward trend in error change are denoted as downward trend samples; This represents the number of downward trend samples obtained by statistically analyzing all outlier samples. Indicates the first There are 10 abnormal samples, where M represents the number of all abnormal samples with label=1. Here, all abnormal samples can be any circuit breaker, that is, all abnormal samples of all circuit breakers. This represents the mean of the error variation trend of all abnormal samples with label=1, i.e., the average value of the error difference. norm() represents the normalization function, with a value range of [0,1].
[0064] A larger q value for outlier samples indicates better consistency, while a smaller q value indicates worse consistency. Worse consistency necessitates reducing the weight of normal samples and increasing the weight of outlier samples. These weights are also known as sample weights or training weights.
[0065] Similarly, the consistency q of normal samples can also be obtained.
[0066] For any given sample, the training weights that need to be adjusted next are: In the formula, This indicates the training weights for the next sample. This represents the training weights of the previous sample in each training iteration. =1 or -1, This indicates the consistency between abnormal and normal samples (label=1 indicates an abnormal sample, label=-1 indicates a normal sample). This represents the error of each training iteration of the sample. Indicates positive and negative signs, including This indicates samples that show a decreasing trend in error variation. This indicates samples that show an upward trend in the error change trend.
[0067] The subsequent training process repeats the above process, realizing the determination of the weight for each sample in each training iteration.
[0068] Training of the LSTM model is stopped when the model error is less than 1% (target error rate, which can be adjusted). The monitoring and prediction for circuit breakers are as follows: The system collects electrical quantity data of the circuit breaker for the current time period (current electrical data), inputs this data into the LSTM model, and outputs the aging parameters of the circuit breaker contacts for future time periods, thereby realizing the health monitoring and prediction of the circuit breaker.
[0069] This invention solves the problem in the traditional training process by determining the weight of the samples. Because there are few or no aging circuit breaker samples, the features of aging samples cannot be learned, while the features of normal circuit breaker samples are overlearned.
[0070] By determining the weights of the samples, this approach addresses the problem that traditional methods for predicting aging rely solely on changes in samples from the most recent time period, while historical circuit breaker failures can also affect subsequent sample changes.
[0071] By adjusting the sample weights during the iteration process, the problem of relying solely on the weight changes of the initial samples or traditional samples to effectively change the training model of LSTM (Long Short-Term Memory) network is solved. Furthermore, by adjusting the sample error during the iteration process, the problem of determining the weights without changing the actual prediction results caused by traditional training is solved.
[0072] This solves the problems of inaccurate prediction and low convergence speed of traditional LSTM models, and improves the accuracy of circuit breaker aging prediction.
[0073] Example 2: This invention also proposes a metering circuit breaker. For example... Figure 7 As shown, Figure 7 This is a schematic diagram of the hardware operating environment of the metering circuit breaker involved in the embodiments of the present invention.
[0074] like Figure 7 As shown, the metering circuit breaker may include: a processor 1001, such as a CPU; a network interface 1004; a user interface 1003; a memory 1005; and a communication bus 1002. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display or an input unit such as a control panel; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be a high-speed RAM or a stable, non-volatile memory, such as a disk drive. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001. The memory 1005, as a computer storage medium, may include a metering circuit breaker health monitoring program.
[0075] Those skilled in the art will understand that Figure 7 The hardware structure shown does not constitute a limitation on the device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.
[0076] Continue to refer to Figure 7 , Figure 7 The memory 1005, which is a computer-readable storage medium, may include an operating device, a user interface module, a network communication module, and a metering circuit breaker health monitoring program.
[0077] exist Figure 7In this embodiment, the network communication module is mainly used to connect to the server and can communicate with the server for data; while the processor 1001 can call the metering circuit breaker health monitoring program stored in the memory 1005 and execute the steps in the above embodiments.
[0078] Based on the hardware structure of the metering circuit breaker described above, various embodiments of the metering circuit breaker health monitoring method of the present invention are implemented.
[0079] In addition, the present invention also provides a metering circuit breaker health monitoring device, please refer to... Figure 8 The metering circuit breaker health monitoring device includes: The fault correlation analysis module A10 is used to determine the fault accumulation sequence and contact aging sequence of the circuit breaker based on electrical data samples. It uses the fault accumulation sequence to determine the circuit breaker importance of the target circuit breaker for contact faults. It uses the fitting error between the fault accumulation sequence and the contact aging sequence to correct the circuit breaker importance. The corrected circuit breaker importance and contact aging value of the target circuit breaker are used as its circuit breaker characteristics. The sample weight adjustment module A20 is used to determine the initial weights of electrical data samples by utilizing the circuit breaker feature similarity between the target circuit breaker and other circuit breakers; to train the electrical data samples using an LSTM model based on the initial weights, thereby obtaining the error change trend of the electrical data samples; and to update the training weights of the electrical data samples using the error change trend until the target error rate of the LSTM model is reached. The model health prediction module A30 is used to input the current electrical data of the circuit breaker into the trained LSTM model to predict the contact aging parameters for future time periods.
[0080] Furthermore, the fault correlation analysis module A10 is also used for: Based on the electrical data samples and the corresponding fault log files, determine the cumulative average fault rate up to the corresponding electrical data sample according to the sample time sequence; The cumulative fault sequence of the circuit breaker is obtained by arranging and combining the cumulative average fault rate of each electrical data sample according to the time sequence of the samples. Determine the contact aging value corresponding to each electrical data sample, and obtain the contact aging sequence of the circuit breaker based on the contact aging values arranged and combined in the time sequence of the samples.
[0081] Furthermore, the fault correlation analysis module A10 is also used for: The minimum fault accumulation sequence is obtained by extracting the minimum value of each element with the same index in the fault accumulation sequence of all circuit breakers. The importance of the target circuit breaker for contact faults can be obtained by using the Euclidean distance between the corresponding elements of the fault accumulation sequence of the target circuit breaker and the minimum fault accumulation sequence.
[0082] Furthermore, the fault correlation analysis module A10 is also used for: Determine the two-dimensional data points composed of corresponding elements of the fault accumulation sequence and contact aging sequence of the target circuit breaker; The fitting error of all electrical data samples of the target circuit breaker is obtained by inputting the two-dimensional data points into the preset fitting polynomial; The importance of the circuit breaker is corrected by using the fitting error of all electrical data samples of the target circuit breaker and the overall fitting error of all circuit breakers.
[0083] Furthermore, the sample weight adjustment module A20 is also used for: Determine the differences in modified circuit breaker importance and contact aging sequence between the target circuit breaker and the other circuit breakers; The similarity of characteristics of each circuit breaker is obtained by correcting the differences in importance of the circuit breakers and the differences in the contact aging sequence. The ratio between the standard deviation and the mean of all circuit breaker feature similarities is taken as the coefficient of variation of the circuit breaker feature similarity set; The initial weights of electrical data samples are determined by using the coefficient of variation of the circuit breaker feature similarity set.
[0084] Furthermore, the sample weight adjustment module A20 is also used for: By utilizing the fitting error between the fault accumulation sequence and the contact aging sequence, the importance of each electrical data sample to the contact fault sample is determined; By utilizing the importance of the samples and their corresponding contact aging values, the unit aging value of the electrical data samples is determined. The initial weights of electrical data samples are determined by using the coefficient of variation of the circuit breaker feature similarity set and the unit aging value of the electrical data samples.
[0085] Furthermore, the sample weight adjustment module A20 is also used for: The initial error of the electrical data samples is obtained by training the LSTM model based on the initial weights. The training weights of the next electrical data samples are then updated using the initial error and the initial weights. Using the training weights of the next electrical data sample as the initial weights, the step of training the LSTM model on the electrical data sample based on the initial weights is executed repeatedly to obtain the initial error of the electrical data sample, thereby obtaining the error change trend of the electrical data sample.
[0086] Furthermore, the sample weight adjustment module A20 is also used for: Based on a class of support vector machines, electrical data samples that are identified as outliers are determined, and the error difference value of the outliers is determined during multiple LSTM model training processes. The average value of the error difference and the number of samples with a decreasing error trend are determined by using the error difference values of the outlier samples. The consistency of outlier samples is determined by the average value of the error difference and the number of samples with a downward trend. The training weights of the electrical data samples are then updated based on the consistency of the outlier samples until the target error rate of the LSTM model is reached.
[0087] The specific implementation of the metering circuit breaker health monitoring device of the present invention is basically the same as the embodiments of the metering circuit breaker health monitoring method described above, and will not be repeated here.
[0088] Furthermore, the present invention also provides a computer-readable storage medium. The computer-readable storage medium stores a metering circuit breaker health monitoring program, wherein when executed by a processor, the metering circuit breaker health monitoring program implements the steps of the metering circuit breaker health monitoring method described above.
[0089] The method implemented when the metering circuit breaker health monitoring program is executed can be referred to in various embodiments of the metering circuit breaker health monitoring method of the present invention, and will not be repeated here.
[0090] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0091] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.
[0092] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0093] The above description is only a preferred embodiment of the present invention and does not limit the scope of protection of the present invention. All equivalent structural / method transformations made under the inventive concept of the present invention using the contents of the present invention specification and drawings, or direct / indirect applications in other related technical fields, are included within the scope of protection of the present invention.
Claims
1. A method for health monitoring of metering circuit breakers, characterized in that, The method includes: Based on electrical data samples, determine the fault accumulation sequence and contact aging sequence of the circuit breaker, and use the fault accumulation sequence to determine the importance of the target circuit breaker for circuit breakers with contact faults. The importance of the circuit breaker is corrected by using the fitting error between the fault accumulation sequence and the contact aging sequence. The corrected importance of the target circuit breaker and the contact aging value are used as its circuit breaker characteristics. The initial weights of electrical data samples are determined by utilizing the similarity of circuit breaker features between the target circuit breaker and the other circuit breakers. The electrical data samples are trained using an LSTM model based on the initial weights to obtain the error change trend of the electrical data samples. The training weights of the electrical data samples are updated using the error change trend until the target error rate of the LSTM model is reached; The current electrical data of the circuit breaker is input into a trained LSTM model to predict contact aging parameters for future time periods.
2. The method for health monitoring of metering circuit breakers according to claim 1, characterized in that, The determination of the circuit breaker's fault accumulation sequence and contact aging sequence based on electrical data samples includes: Based on the electrical data samples and the corresponding fault log files, determine the cumulative average fault rate up to the corresponding electrical data sample according to the sample time sequence; The cumulative fault sequence of the circuit breaker is obtained by arranging and combining the cumulative average fault rate of each electrical data sample according to the time sequence of the samples. Determine the contact aging value corresponding to each electrical data sample, and obtain the contact aging sequence of the circuit breaker based on the contact aging values arranged and combined in the time sequence of the samples.
3. The method for health monitoring of metering circuit breakers according to claim 1, characterized in that, The method of determining the importance of a target circuit breaker for contact faults using a fault accumulation sequence includes: The minimum fault accumulation sequence is obtained by extracting the minimum value of each element with the same index in the fault accumulation sequence of all circuit breakers. The importance of the target circuit breaker for contact faults can be obtained by using the Euclidean distance between the corresponding elements of the fault accumulation sequence of the target circuit breaker and the minimum fault accumulation sequence.
4. The method for health monitoring of metering circuit breakers according to claim 1, characterized in that, The method of correcting the importance of circuit breakers by utilizing the fitting error between the fault accumulation sequence and the contact aging sequence includes: Determine the two-dimensional data points composed of corresponding elements of the fault accumulation sequence and contact aging sequence of the target circuit breaker; The fitting error of all electrical data samples of the target circuit breaker is obtained by inputting the two-dimensional data points into the preset fitting polynomial; The importance of the circuit breaker is corrected by using the fitting error of all electrical data samples of the target circuit breaker and the overall fitting error of all circuit breakers.
5. The method for health monitoring of metering circuit breakers according to claim 1, characterized in that, The method of determining the initial weights of electrical data samples based on the similarity of circuit breaker features between the target circuit breaker and the other circuit breakers includes: Determine the differences in modified circuit breaker importance and contact aging sequence between the target circuit breaker and the other circuit breakers; The similarity of characteristics of each circuit breaker is obtained by correcting the differences in importance of the circuit breakers and the differences in the contact aging sequence. The ratio between the standard deviation and the mean of all circuit breaker feature similarities is taken as the coefficient of variation of the circuit breaker feature similarity set; The initial weights of electrical data samples are determined by using the coefficient of variation of the circuit breaker feature similarity set.
6. The method for health monitoring of metering circuit breakers according to claim 5, characterized in that, The method of determining the initial weights of electrical data samples using the coefficient of variation of the circuit breaker feature similarity set includes: By utilizing the fitting error between the fault accumulation sequence and the contact aging sequence, the importance of each electrical data sample to the contact fault sample is determined; By utilizing the importance of the samples and their corresponding contact aging values, the unit aging value of the electrical data samples is determined. The initial weights of electrical data samples are determined by using the coefficient of variation of the circuit breaker feature similarity set and the unit aging value of the electrical data samples.
7. The method for health monitoring of metering circuit breakers according to claim 1, characterized in that, The step of training an LSTM model on electrical data samples based on initial weights to obtain the error change trend of the electrical data samples includes: The initial error of the electrical data samples is obtained by training the LSTM model based on the initial weights. The training weights of the next electrical data samples are then updated using the initial error and the initial weights. Using the training weights of the next electrical data sample as the initial weights, the step of training the LSTM model on the electrical data sample based on the initial weights is executed repeatedly to obtain the initial error of the electrical data sample, thereby obtaining the error change trend of the electrical data sample.
8. The method for health monitoring of metering circuit breakers according to claim 1, characterized in that, The step of updating the training weights of electrical data samples using error change trends until the target error rate of the LSTM model is reached includes: Based on a class of support vector machines, electrical data samples that are identified as outliers are determined, and the error difference value of the outliers is determined during multiple LSTM model training processes. The average value of the error difference and the number of samples with a decreasing error trend are determined by using the error difference values of the outlier samples. The consistency of outlier samples is determined by the average value of the error difference and the number of samples with a downward trend. The training weights of the electrical data samples are then updated based on the consistency of the outlier samples until the target error rate of the LSTM model is reached.
9. A metering-type circuit breaker health monitoring device, characterized in that, The apparatus is used to implement the metering-type circuit breaker health monitoring method as described in any one of claims 1 to 8; the apparatus includes: The fault correlation analysis module is used to determine the fault accumulation sequence and contact aging sequence of the circuit breaker based on electrical data samples. It uses the fault accumulation sequence to determine the circuit breaker importance of the target circuit breaker for contact faults. It uses the fitting error between the fault accumulation sequence and the contact aging sequence to correct the circuit breaker importance. The corrected circuit breaker importance and contact aging value of the target circuit breaker are used as its circuit breaker characteristics. The sample weight adjustment module is used to determine the initial weights of electrical data samples by utilizing the circuit breaker feature similarity between the target circuit breaker and other circuit breakers; to train the LSTM model on the electrical data samples based on the initial weights to obtain the error change trend of the electrical data samples; and to update the training weights of the electrical data samples using the error change trend until the target error rate of the LSTM model is reached. The model health prediction module is used to input the current electrical data of the circuit breaker into the trained LSTM model to predict the contact aging parameters for future time periods.
10. A metering circuit breaker, characterized in that, The metering circuit breaker includes a processor, a memory, and a metering circuit breaker health monitoring program stored in the memory and executable by the processor, wherein when the metering circuit breaker health monitoring program is executed by the processor, it implements the steps of the metering circuit breaker health monitoring method as described in any one of claims 1 to 8.