Method and system for detecting real-time service life of piezoresistor
By dividing the time window and selecting parameters through current impulse experiments on varistors, a prediction model was constructed, which solved the problem of accuracy in predicting the real-time remaining life of varistors and achieved more accurate life detection and early warning.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHENZHEN RUILONGYUAN ELECTRONICS CO LTD
- Filing Date
- 2026-02-09
- Publication Date
- 2026-05-15
AI Technical Summary
In existing technologies, the real-time remaining life prediction of varistors has a large deviation, making it difficult to accurately distinguish between long-term aging and degradation caused by instantaneous impact, resulting in inaccurate predictions.
By dividing the current impact experiment of the entire life cycle of the varistor into multiple time windows, the correlation parameters of each time window are obtained, the directly detectable parameters are screened, the mixed influence of parameters is analyzed using the support vector machine model, a prediction model is constructed, and weighted processing is performed to obtain the changing feature quantity, so as to realize real-time remaining lifetime detection.
This improves the accuracy and stability of real-time remaining life prediction for varistors, effectively monitors and warns of their usage status, reduces errors, and enhances the reliability of the prediction model.
Smart Images
Figure CN122043097A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the technical field of resistance lifetime testing, specifically to a real-time lifetime testing method and system for varistors. Background Technology
[0002] A varistor is a resistor whose resistance changes with voltage within a certain current and voltage range, or a resistor whose resistance is sensitive to voltage. It is a resistive device with nonlinear volt-ampere characteristics, mainly used for voltage clamping when a circuit is subjected to overvoltage, absorbing excess current to protect sensitive components. Traditional maintenance relies on periodic replacement or reactive failure, posing significant safety hazards or wasting resources. During engineering applications, it is necessary to monitor the lifespan of varistors. Therefore, real-time lifespan detection and early warning of varistors are crucial for predictive maintenance in critical areas such as power systems, rail transit, and communication base stations. A multi-feature fusion-based online lifespan prediction system for surge protectors needs to be designed to predict the remaining lifespan of surge protectors online in real time, achieving full lifecycle health management.
[0003] Whether it is long-term slow aging or instantaneous strong lightning strike, it may cause short-term or long-term changes in certain monitoring characteristics of varistors. Traditional multi-feature fusion models directly use models such as LSTM (Long Short Term Memory networks) to correlate static time series, and the predicted real-time remaining lifetime has a large deviation. Summary of the Invention
[0004] To address the technical problem of improving the accuracy of real-time remaining life prediction for varistors during use, the present invention aims to provide a real-time life detection method and system for varistors, the specific technical solution of which is as follows: In a first aspect, embodiments of the present invention provide a method for real-time lifetime detection of a varistor, the method comprising: The current surge test of the varistor throughout its entire life cycle was divided into multiple time windows, and several correlation parameters characterizing the deterioration of the resistor performance were obtained for each time window. The complex effects among various related parameters are analyzed to select the directly measured parameters during the resistance degradation process from among the multiple related parameters in each time window. Based on the correlation between the direct detection parameters and the indirect detection parameters of multiple related parameters, the correlation significance and reference factor of the two types of detection parameters are obtained. The direct detection parameters are weighted according to the correlation significance and reference factor to obtain the characteristic quantity of the varistor's tendency to deteriorate in different time windows. A prediction model is constructed based on the changing characteristic quantities, and the remaining lifetime of the target varistor currently in operation is detected using the prediction model.
[0005] In one optional embodiment, the current surge test of the varistor throughout its entire lifespan is divided into multiple time windows, and several correlation parameters characterizing the degradation of the resistor's performance in each time window are obtained, including: Based on the degradation characteristics of varistor in multi-level surge protection system, the experimental parameters for current impulse test are configured. The current surge test was initiated and multiple related parameters were obtained until the varistor was completely damaged. These parameters included varistor voltage, leakage current, nonlinear coefficient, series equivalent capacitance, parallel equivalent capacitance, component reactance, and component impedance.
[0006] In one alternative embodiment, the overlapping effects between various correlated parameters are analyzed to filter out directly detectable parameters measured by instruments during the resistance degradation process from multiple correlated parameters in each time window, including: Multiple correlation parameters for each time window are used as multidimensional feature samples and input into the support vector machine model; Based on the discrimination results formed by the support vector machine model in the feature space, the correlation parameters that have significant discrimination effects in each time window during the resistance degradation process are screened to obtain the direct detection parameters for the corresponding time window.
[0007] In one optional embodiment, the correlation significance and reference factor between the two types of detection parameters are obtained based on the correlation between the direct detection parameters and the indirect detection parameters of multiple related parameters, including: The correlation significance between direct and indirect detection parameters is obtained based on their relative distance in space and time. Based on the differences in the changes between direct and indirect detection parameters in adjacent time windows and the differences in the significance of their correlation, a reference factor is obtained to characterize the degree of influence of indirect detection parameters on direct detection parameters.
[0008] In one optional embodiment, the correlation significance between the direct detection parameters and the indirect detection parameters is obtained based on their relative distance in space and time, including: Based on the Euclidean distance between the current directly detected parameters and the current indirectly detected parameters in the topology diagram, the spatial parameter distance between the two types of current detected parameters is obtained. The topology diagram is a structural diagram formed by the connection relationship of each varistor in a multi-level surge protection system. Based on the time warping distance between the first parameter sequence to which the current directly detected parameter belongs and the second parameter sequence to which the current indirectly detected parameter belongs, the temporal correlation distance between the two types of current detected parameters is obtained. The correlation significance between the two types of current detection parameters is obtained by the ratio of spatial parameter distance to temporal correlation distance.
[0009] In an optional embodiment, a reference factor characterizing the influence of the indirect detection parameter on the direct detection parameter is obtained based on the differences in the direct and indirect detection parameters over adjacent time windows and the differences in their correlation significance. This factor includes: Based on the difference in the sustained response of the current indirect detection parameter associated with the current direct detection parameter within adjacent time windows, obtain the variation coefficient of the current indirect detection parameter within adjacent time windows. Based on the significance difference of the correlation significance between the current indirect detection parameter and the current direct detection parameter in adjacent time windows, and the historical standard deviation of the current indirect detection parameter, the change coefficient of the current indirect detection parameter in adjacent time windows is obtained. Based on the coefficient of variation and the coefficient of degree of variation, a reference factor is obtained to determine the degree of influence of the current indirect detection parameter on the current direct detection parameter.
[0010] In an optional embodiment, the directly detected parameters are weighted according to the correlation significance and a reference factor to obtain the characteristic quantity of the varistor's tendency to deteriorate in different time windows, including: Based on the correlation significance and reference factor, the reference weights of the directly detected parameters in the corresponding time window are obtained; Based on the reference weights of the directly detected parameters in each time window and the area of the curve peak, the change characteristic quantities under the corresponding time window are obtained.
[0011] In one optional embodiment, the reference weights of the direct detection parameters within the corresponding time window are obtained based on the association significance and the reference factor, including: The significance difference of each indirect detection parameter in adjacent time windows is obtained by comparing the significance difference of the association between each indirect detection parameter associated with the direct detection parameter in adjacent time windows. The indirect parameter weights of the corresponding indirect detection parameters are obtained by multiplying the significance difference of each indirect detection parameter with the corresponding reference factor. The reference weight of the direct detection parameter is obtained by averaging the weights of all indirect parameters in adjacent time windows.
[0012] In one optional embodiment, a prediction model is constructed based on the changing characteristic quantities, and the prediction model is used to detect the remaining lifetime of the target varistor currently in operation, including: The Euclidean distance between two changing features is used as the input sample for the support vector machine model. The support vector machine model is trained, and a prediction model is obtained based on the trained support vector machine model. The time-series data stream of various parameters monitored by the online monitoring terminal for the target varistor is input into the prediction model, and the real-time remaining lifetime of the target varistor is obtained based on the output of the prediction model.
[0013] Secondly, embodiments of the present invention also provide a real-time lifetime detection system for varistors. The detection system is a system corresponding to any method in the first aspect, and the system includes: The acquisition module is used to divide the current impact test of the varistor throughout its entire life cycle into multiple time windows, and acquire multiple correlation parameters characterizing the deterioration of the resistance performance in each time window. The filtering module is used to analyze the mixed effects between various related parameters, so as to filter out the directly detected parameters measured by the instrument during the resistance degradation process from multiple related parameters in each time window; The first acquisition module is used to obtain the correlation significance and reference factor of the two types of detection parameters based on the correlation between the direct detection parameters and the indirect detection parameters of multiple related parameters. The second acquisition module is used to weight the directly detected parameters according to the correlation significance and reference factor to obtain the characteristic quantity of the varistor as it tends to deteriorate in different time windows. The lifetime prediction module is used to build a prediction model based on the changing characteristic quantities, and to use the prediction model to detect the remaining lifetime of the target varistor currently in operation.
[0014] The present invention has the following beneficial effects: The technical solution of this invention divides the current surge test of a varistor throughout its entire lifespan into multiple time windows and obtains multiple correlation parameters characterizing the degradation of the resistor's performance in each time window. This ensures that the parameter changes within each time window primarily reflect the degradation mechanism of that stage. Within each time window, there are overlapping influences among the multiple correlation parameters of the varistor. By analyzing these correlation parameters in each time window, directly measured parameters obtained through instrumentation during the resistor degradation process are selected. After selecting the directly measured parameters, the correlation significance and reference factor between the directly measured parameters and the indirect measured parameters of the multiple correlation parameters are obtained, reducing the influence of irrelevant parameters on lifespan assessment. To prevent interference, the directly detected parameters are further weighted to obtain the characteristic quantities of the varistor's degradation trend within different time windows. The characteristic quantities constructed through adaptive weighting can more accurately characterize the true state of the varistor at different degradation stages. A prediction model is constructed based on the characteristic quantities, and the prediction model is used to detect the remaining life of the target varistor currently in operation. The prediction model can more accurately distinguish between transient impacts and continuous degradation under complex operating conditions, thereby improving the reliability and stability of the real-time remaining life prediction results, realizing effective monitoring and early warning of the varistor's operating status, and significantly improving the accuracy of real-time remaining life prediction during the use of the varistor. Attached Figure Description
[0015] 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.
[0016] Figure 1 A flowchart illustrating a real-time lifetime detection method for a varistor according to an embodiment of the present invention; Figure 2 The flowchart illustrates the calculation of association significance and reference factor according to one embodiment of the present invention. Figure 3 This is a flowchart illustrating the calculation of a variable characteristic quantity provided in one embodiment of the present invention; Figure 4 This is a schematic diagram of a real-time lifetime detection system for a varistor provided in one embodiment of the present invention. Detailed Implementation
[0017] 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 real-time lifetime detection method and system for varistors 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.
[0018] 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.
[0019] During the use of varistors, their lifespan degradation includes both long-term, slow material aging and sudden degradation caused by instantaneous high-current surges. These two types of behaviors change across multiple electrical parameters, and their interactions are intertwined and superimposed in both time series and parameter space. This makes it difficult for methods based on full lifespan or static parameter modeling to distinguish between transient anomalies and irreversible degradation, resulting in insufficient accuracy in real-time remaining lifetime prediction. For example, in existing technologies, multi-feature fusion models directly use models such as LSTM for static time series correlation. However, normal aging data and sudden failure data overlap to some extent in the feature space. Multi-feature fusion models are prone to discrimination errors in situations such as severe lightning strikes and severe device aging, leading to significant deviations in real-time remaining lifetime prediction.
[0020] The technical solution of this invention divides the continuous lifecycle of a device into a series of continuous analysis segments by defining an analysis window. Each analysis segment is used as a single time window for data analysis. Based on the changing trends of different parameters in different time windows, a more accurate real-time lifespan prediction model for resistors is constructed, improving the accuracy of real-time remaining lifespan prediction for varistors under operating conditions. The specific solution of the real-time lifespan detection method and system for varistors provided by this invention will be described in detail below with reference to the accompanying drawings.
[0021] Please see Figure 1 , Figure 1 The flowchart illustrates a real-time lifetime detection method for a varistor according to an embodiment of the present invention. This detection method can detect the remaining lifetime of the varistor in real time using a detection terminal. The detection terminal can be a detection board composed of a microcontroller or other computer equipment capable of running the detection method; no specific limitations are imposed here. The detection method includes: S11: Divide the current surge test of the varistor throughout its entire life cycle into multiple time windows, and obtain multiple correlation parameters characterizing the deterioration of the resistance performance in each time window.
[0022] Specifically, the degradation state of a varistor is not consistent throughout its entire lifespan, but exhibits different evolutionary characteristics at different stages. By dividing the lifespan into time windows (or sliding windows), the continuous lifetime evolution process can be decomposed into several relatively stable local states. This ensures that the changes in associated parameters within each time window primarily reflect the degradation mechanism of that stage, thus avoiding global modeling from masking local anomalies. The division points of the time windows can be determined based on the variation characteristics of multiple associated parameters, such as the slope change of the leakage current curve among multiple associated parameters; alternatively, the division can be based on a combination of the slope changes of curves from multiple types of associated parameters, as long as it can characterize the differences in the degree of degradation of the varistor based on each time window.
[0023] It is understandable that varistors have various application scenarios. For example, in a multi-stage surge protection system, there are multiple varistors, and damage to any one of them will severely affect the normal operation of the system. Since it is difficult to accumulate a large amount of lightning damage in real-world applications, surge current testing can be conducted on multi-stage surge protection systems in the laboratory. The surge current testing equipment can generate surge currents ranging from 200A to 100KA to simulate the surges experienced by the varistors. Based on the data collected during the current surge test, the entire experimental period is divided into multiple time windows. Several related parameters for each time window are stored, and each related parameter is related to the performance degradation of the varistor, such as the varistor's varistor voltage and leakage current.
[0024] In the data from the current impact test, some parameters (such as leakage current and series equivalent resistance) change drastically after the damage point appears. This is related to the changes in the internal structure of the varistor after it is completely damaged. Globally unified parameter analysis involves the integration of data correlations and ignores some minor errors that cause resistor damage. Therefore, this invention constructs a time window and analyzes the changing trends of parameters in different time windows to build a more accurate real-time life prediction model for resistors.
[0025] For example, step S11 includes sub-steps S11-1 to S11-2, which are described in detail below: S11-1: Based on the degradation characteristics of varistors in multi-stage surge protection systems, the experimental parameters for the current impulse test are configured. The loads varistors bear in practical applications exhibit significant engineering characteristics, such as surge current amplitude distribution, repetitive impact characteristics, and cumulative energy effects. If the experimental parameters do not match the actual operating conditions, it is difficult to accurately reproduce the degradation process of the varistor, resulting in a lack of representativeness in the subsequently collected data. In this embodiment of the invention, before conducting the current impulse test, the key parameters of the impulse test are configured based on the actual application environment and typical degradation mechanisms of the varistor in a multi-stage surge protection system, including the impulse current waveform, amplitude range, duration, number of impulses, and impulse interval. The varistor is subjected to an 8 / 20μs impulse current wave as specified in national and railway industry standards during the experiment.
[0026] S11-2: A current surge test is conducted to obtain multiple related parameters until the varistor is completely damaged. These parameters include varistor voltage, leakage current, nonlinear coefficient, series equivalent capacitance, parallel equivalent capacitance, component reactance, and component impedance. With each surge, the varistor absorbs energy, causing changes in its internal structure. Initially, the changes are gradual; as degradation accumulates, performance declines, and the changes in various related parameters become more moderate. When the degradation reaches its critical point, the next surge can fundamentally alter its internal structure, resulting in visible damage points. After this point, the varistor is completely damaged and loses all its application value.
[0027] During the experiment, various related parameters were collected. Varistor voltage and nonlinear coefficient reflect changes in the nonlinear characteristics of the material; leakage current reflects the development of internal defects and the formation of conductive channels; series equivalent capacitance, parallel equivalent capacitance, component reactance, and component impedance reflect changes in the equivalent circuit parameters of the device and the deterioration state of the internal microstructure. After data acquisition, the collected data were smoothed and denoised, and then the data from each iteration were integrated in tabular form. All parameters were organized and normalized using min-max normalization, and then the normalized variation trend of each parameter with the number of surges was plotted by superposition.
[0028] At this point, multiple correlation parameters characterizing the deterioration of resistive performance in each time window have been obtained, and the process proceeds to step S12.
[0029] S12: Analyze the interplay between various related parameters to select the directly measured parameters during the resistance degradation process from among the multiple related parameters in each time window.
[0030] Specifically, as the number of current surges increases in the current surge experiment, surge events at different times become intertwined and merged. In order to reflect the short-term fluctuations and aging trends of the resistance performance, it is necessary to analyze the intertwined effects between various related references and determine the related parameters that need to be measured within each time window. These related parameters are denoted as direct detection parameters. Direct detection parameters are those parameters that can be directly measured by instruments during the feature selection process of the varistor in the model training phase. After obtaining the direct detection parameters for each time window, the corresponding parameters can be directly measured during the use of the varistor, thereby obtaining the real-time remaining life of the varistor. In the current surge experiment, the surge time is the sampling duration of a single surge waveform. The time window is constructed with 10 times the surge time as the window width and one-fifth of the window width as the sliding step size.
[0031] Within each window, the changes in characteristics related to aging and failure of the varistor are extracted, which can reflect the characteristics of short-term resistance performance fluctuations and aging trends. Experimental states are divided for all monitored parameters within each time window. During multiple impacts, the changes in several related parameters vary, and multiple related parameters interact with each other. The parameter signals between adjacent electric shock states are intertwined. Faced with complex multi-parameter fusion, a support vector machine model can be used to predict and classify the parameter signals within each time window, thereby obtaining the direct detection parameters for each time window.
[0032] For example, step S12 includes sub-steps S12-1 to S12-2, which are described in detail below: S12-1: Multiple correlation parameters from each time window are used as multi-dimensional feature samples and input into the support vector machine (SVM) model to train the SVM model. The training objective is to find the optimal classification hyperplane. Since the waveform changes of the correlation parameters are non-linear data, the hyperplane cannot be directly determined for SVM model training and classification. Therefore, features are extracted from the waveform changes of each detection parameter, and the hyperplane is determined based on the feature values, achieving linear segmentation of the resistance detection data.
[0033] S12-2: Based on the discrimination results formed by the support vector machine model in the feature space, the correlation parameters that have significant discriminative effects in each time window during the resistance degradation process are screened to obtain the direct detection parameters for the corresponding time window. Taking the t-th time window as an example, the correlation parameters are denoted as the change amplitude of the monitoring data at continuous moments during the electric shock change. Where m is a natural number greater than 0, based on the physical layout of the varistor array, a topology diagram of multiple varistors in a multi-level surge protection system is constructed, and several detection parameters of the varistor performance that can be directly measured by instruments in the current detection state are extracted. Let m and t be the direct detection parameters, where m and t are both natural numbers greater than 0.
[0034] At this point, the direct detection parameters for each time window have been obtained based on the above method, and we proceed to step S13.
[0035] S13: Based on the correlation between the direct detection parameters and the indirect detection parameters of multiple related parameters, obtain the correlation significance and reference factor of the two types of detection parameters.
[0036] Specifically, taking a multi-level surge protection system as an example, this system also has some issues that affect directly detected parameters. The changing parameters, which in turn affect the results of the target variable, are the indirect detection parameters (or indirect influence parameters). The impact of these indirect detection parameters varies significantly across different time windows. Therefore, random indirect detection parameters are used within the current time window. By analyzing the changes in the parameters and the distance patterns between indirect and direct detection parameters in the topology diagram, the correlation significance between any two types of detection parameters can be calculated. Correlation significance characterizes the degree of influence of indirect detection parameters on direct detection parameters during the degradation process of the varistor. Based on correlation significance, it can be determined which indirect detection parameters are highly correlated with direct detection parameters at the structural and evolutionary levels. The results are primarily used to screen and quantify the reliability of the correlation between different indirect detection parameters and direct detection parameters.
[0037] Indirect detection parameters and direct detection parameters have a high correlation significance, but their effects may differ significantly at different stages of degradation. Reference factors are used to characterize this stage-specific impact. Reference factors are used to characterize the actual influence of a given indirect detection parameter on changes in the direct detection parameter within the current time window.
[0038] For example, please refer to Figure 2 , Figure 2 The flowchart for calculating association significance and reference factor includes: S13-1: The correlation significance between direct and indirect detection parameters is obtained based on their relative distance in space and time. During the degradation of a varistor, although multiple electrical parameters change with device aging, different parameters exhibit varying degrees of sensitivity to degradation. This invention characterizes the structural relationship of different parameters in a multi-parameter feature space based on their relative spatial distance, reflecting the proximity of a direct detection parameter to a certain indirect detection parameter within the overall parameter distribution. If the distance between the two is small in the feature space, it indicates a high degree of consistency in their change patterns.
[0039] Temporal relative distance is used to characterize the degree of synchronization of different detection parameters during the degradation process, such as whether two types of parameters begin to change significantly within similar time windows, and whether their change rhythms, acceleration, or abrupt changes are consistent. By performing time warping or similarity calculations on the parameter sequences, interference caused by sampling offsets or local lags can be eliminated, yielding more realistic temporal evolution correlations. A computational model can be constructed based on the correlation significance and the relative distance between the two types of parameters in space and time, and the corresponding correlation significance can be derived from the output of the computational model.
[0040] Furthermore, sub-step S13-1 includes: The first step is to obtain the spatial parameter distance between the two types of current detection parameters based on the Euclidean distance between the current directly detected parameters and the current indirectly detected parameters in the topology diagram. The topology diagram is a structural diagram formed by the connection relationships of various varistors in a multi-level surge protection system. It can be derived from the circuit diagram representing the electrical connections of each varistor. In the topology diagram, if a single varistor is a detection unit, there are multiple directly detected parameters within the unit, and indirect influence parameters are generated between units through topological connections. For a given varistor, in addition to the several directly measured parameters, there are also some indirect influence parameters affecting the current varistor. For example, for the x-th directly detected data of varistor A, varistor B, which is connected to varistor A, also contains N types of detected data. Different data have different degrees of influence on varistor A. It is necessary to calculate the correlation between the j-th indirect detected parameter of varistor B and the x-th directly detected data of varistor A. Therefore, the position of the x-th directly detected parameter is denoted as... The j-th indirect detection parameter associated with the direct detection parameter is Then the spatial parameter distance is .
[0041] The second step is to obtain the temporal correlation distance between the first parameter sequence to which the current directly detected parameter belongs and the second parameter sequence to which the current indirectly detected parameter belongs, based on the time warping distance. Using the aforementioned current directly detected parameter as an example... The current indirect detection parameters are For example, the temporal correlation distance is .
[0042] The third step is to obtain the correlation significance between the two currently detected parameters based on the ratio of spatial parameter distance to temporal correlation distance. This can be based on the formula: Calculate the significance of the association between the j-th indirect detection parameter and the x-th direct detection parameter in the current time window. , Represents a very small positive number, which can be a preset minimum value, for example... The value can be specifically set to 0.1 to prevent... The problem of a denominator of 0 caused by a value of 0. Temporal correlation distance reflects the degree of matching between two sequences over time; the smaller the value, the higher the degree of matching, indicating that the changes in the two current detection parameters are more similar. This indicator is used to screen global common-mode interference features that are less affected by spatial distance but have high temporal correlation. When calculating the correlation significance based on the above formula, given a fixed Euclidean distance between the direct and indirect detection parameters, a smaller temporal correlation distance results in a higher correlation significance. Therefore, it can be used to screen parameter data that are spatially similar but have significant changes in temporal correlation.
[0043] It should be noted that, based on the characteristic that indirect detection parameters are obtained by calculating directly detection parameters during the current impact process, the higher the correlation significance, the stronger the possibility of coexistence between the two parameters. Since there are significant changes in the coexisting parameter signals, they need to be given special attention during the model construction process. Therefore, joint analysis of the degree of change between different time windows is required for the two types of detection parameters that are correlated.
[0044] S13-2: Based on the differences in the changes and correlation significance between directly and indirectly detected parameters in adjacent time windows, a reference factor is obtained to characterize the degree of influence of the indirect detection parameter on the directly detected parameter. In practical applications, even if an indirect detection parameter has a high correlation significance with a directly detected parameter, its role may still differ in different stages of degradation. For example, some parameters are sensitive in the early stages of degradation but tend to saturate in the later stages; some parameters only change drastically near the failure stage. Therefore, it is necessary to characterize the stage-specific influence capability based on the reference factor.
[0045] When the correlation significance of a parameter changes drastically between adjacent time windows and its own historical fluctuations are small, it indicates that the parameter is extremely sensitive to minor anomalies and needs to be given higher weight. Therefore, the performance differences of all indirect detection parameters that affect the current x-th direct detection parameter in two adjacent time windows are calculated as reference factors affecting the x-th direct detection parameter in the current time window.
[0046] The calculation steps for the reference factor will be further described below, including: The first step is to obtain the variation coefficient of the current indirect detection parameter in adjacent time windows based on the difference in the sustained response state of the current directly detected parameter and the associated indirect detection parameter within adjacent time windows. The adjacent time windows are respectively... and , This represents the duration of the j-th indirect detection parameter that is associated with the x-th direct detection parameter within the i-th time window; This represents the duration of the j-th indirect detection parameter that is associated with the x-th direct detection parameter in the (i+1)-th time window, taking the maximum of the two. The coefficient of variation is based on ratios. It is concluded that To represent a very small positive number, you can set... The value is 0.01 to prevent calculation errors. The larger the ratio, the greater the difference in duration between two adjacent electric shock change detection time windows, which is used to amplify the difference in electric shock changes between different electric shock change time windows.
[0047] The second step involves obtaining the variation coefficient of the current indirect detection parameter within adjacent time windows based on the significance difference between the correlation significance of the current directly detected parameter and the current indirect detection parameter within adjacent time windows, and the historical standard deviation of the current indirect detection parameter. The standard deviation of the j-th indirect detection parameter in the overall historical database is the historical standard deviation, denoted as . The coefficient of variation is , Similarly, representing a very small positive number, this coefficient of variation reflects the information difference of the j-th indirect detection parameter between adjacent electric shock change detection time windows. The larger the ratio, the higher the degree of variation of the indirect detection parameter in different time windows, and the greater the influence of the indirect detection parameter on the direct detection parameter.
[0048] The third step is to obtain a reference factor for the degree of influence of the current indirect detection parameter on the current direct detection parameter based on the coefficient of variation and the coefficient of degree of variation. The reference factor can be calculated based on the product of the two coefficients, using the formula: Calculate the reference factors for x direct detection parameters and j-th indirect detection parameter within time window i. Based on the above method, reference factors for each direct detection parameter and its associated indirect detection parameter can be obtained, and all reference factors can be labeled and stored accordingly.
[0049] At this point, the correlation significance and reference factor between each direct detection parameter and its associated indirect detection parameter have been obtained, and we proceed to step S14.
[0050] S14: The direct detection parameters are weighted according to the correlation significance and reference factor to obtain the characteristic quantity of the varistor's tendency to deteriorate in different time windows.
[0051] Specifically, in the degradation monitoring of varistors, although directly measured parameters (such as varistor voltage and leakage current) can be directly measured by instruments, their original values still have the problem that the sensitivity of a single parameter to the degradation stage is inconsistent, and the auxiliary characterization role of indirect measured parameters is not reflected. Therefore, it is necessary to perform weighted processing by comprehensively considering correlation significance and reference factors.
[0052] In the weighted processing, the significance of the association serves as both a correlation filter and a basic weight constraint. A higher significance indicates a more reliable characterization of the degradation of the direct detection parameter by the indirect detection parameter. Based on the reference factor, the actual impact of the indirect detection parameter on the change of the direct detection parameter within the current time window can be determined. Therefore, within each time window, using the direct detection parameter as the basic quantity, the significance of the association of each related indirect detection parameter and the reference factor are used as weighting factors to perform a weighted fusion process on the direct detection parameter, thereby obtaining a comprehensive characteristic quantity reflecting the change in degradation status within that time window.
[0053] For example, please refer to Figure 3 , Figure 3 The flowchart for calculating the changing characteristic quantities includes: S14-1: Based on the correlation significance and reference factor, obtain the reference weight of the direct detection parameter in the corresponding time window. By combining the correlation significance and reference factor to weight the direct detection parameter, indirect detection parameters that have a real degradation correlation with the direct detection parameter and have a significant impact within the current time window are introduced into the characterization of the direct detection parameter. This yields factors that can comprehensively reflect the varistor's emphasis in different time windows, and the degree of emphasis of these factors is obtained through the reference weight of the direct detection parameter in the corresponding time window.
[0054] The calculation of the reference weights specifically includes: The first step is to obtain the significance difference of each indirect detection parameter in adjacent time windows based on the difference in the significance of the association between the directly detected parameters and adjacent time windows. Taking the calculation of the significance difference between adjacent time windows as an example, the significance difference is denoted as... This value reflects the dynamic characteristics of changes between time windows. For example, in some cases, small changes in the time window may have a significant impact on the directly detected parameters, or changes in the time window may cause significant changes in the correlation significance of the directly detected parameters. Calculating the difference can capture the impact of such changes in the detection time window, helping the soft instrument model to better adapt to the actual electric shock environment.
[0055] The second step involves obtaining the indirect parameter weights for each indirect detection parameter by multiplying the significance difference of each parameter by the corresponding reference factor. Based on the product of the significance difference and the corresponding reference factor, the actual weight within that time window can be determined and recorded as the indirect parameter weight.
[0056] The third step is to obtain the reference weights of the direct detection parameters based on the average weights of all indirect parameters in adjacent time windows. This is achieved using the formula: Calculate the reference weight of the x-th direct detection parameter in the current i-th time window when using it to build the prediction model. , This represents the number of indirectly detected parameters within time window i.
[0057] S14-2: Based on the reference weights of the directly detected parameters and the area of the curve peaks in each time window, the characteristic quantities of change under the corresponding time window are obtained. As the current impact experiment progresses, the changing trends of the detected parameters differ in different time windows due to changes in the number of electric shocks or the increasing severity of the shocks. Therefore, the characteristic quantities of change of all monitored parameters in different windows are calculated using the formula: Calculate the characteristic quantity of the change of the u-th electric shock under the i-th time window. ; This represents the reference weight of the x-th direct detection parameter in the current i-th time window when it is used to construct the prediction model; This represents the area of the curve peak for the i-th time window, the u-th electric shock, and the x-th directly detected parameter; x takes values from 1 to n, where n represents the number of directly detected parameters. It can be understood that this embodiment of the invention uses the area of the curve peak to quantify the intensity and duration of parameter changes, thereby accurately determining the characteristic quantities of each directly detected parameter's change in different time windows. It can be understood that the area of the curve peak is the integral area of the transient response waveform of the parameter (such as the voltage / current change waveform over time) in a single current impact event, characterizing the energy absorption or thermal accumulation characteristics of a single impact. During a current impact event, the varistor undergoes a short-duration, intense nonlinear conduction process, characterized by a sudden change in terminal voltage and current within a very short time; significant energy dissipation and temperature rise occur inside the component; it gradually recovers after the impact, but irreversible material and structural damage occurs. This process is not determined by a single instantaneous peak value, but is dominated by the cumulative effect of energy input throughout the impact process. Therefore, the cumulative magnitude of voltage / current changes over time is measured; considering both the magnitude and duration of the change, the area of the curve obtained by integrating the area has stronger stability and representativeness compared to the instantaneous peak value.
[0058] It should be noted that, based on the above formula, the calculation is performed within the i-th time window, for the u-th electric shock process, by calculating the area of the parameter change waveform in the corresponding band of each directly detected parameter to obtain the cumulative intensity of the parameter change during the electric shock process; and by combining the parameter weights of each directly detected parameter in the lifetime prediction model, the cumulative intensity of the change of each parameter is weighted and fused to obtain the window-level change characteristic quantity that characterizes the intensity of the varistor degradation change within the time window.
[0059] At this point, the change characteristics under each time window have been obtained based on the above method, and we proceed to step S15.
[0060] S15: Construct a prediction model based on the changing characteristic quantities, and use the prediction model to detect the remaining life of the target varistor currently in operation.
[0061] Specifically, the variation characteristic quantity is a comprehensive degradation characterization quantity obtained within different time windows, based on direct detection parameters and incorporating the influence of indirect detection parameters. During the model building phase, samples of the variation characteristic quantity corresponding to different lifetime stages of the varistor are collected and combined with its known lifetime state or failure time to complete the training of the prediction model. This establishes a nonlinear mapping relationship between the variation characteristic quantity and the real-time remaining lifetime. Since the variation characteristic quantity already comprehensively reflects the degradation information of multiple parameters, the prediction model does not need to handle complex parameter coupling problems and can obtain stable prediction results. To meet the adaptive use requirements of time windows, a prediction model can be built based on a support vector machine model to select the real-time remaining lifetime of the varistor that fits the changes within the current time window.
[0062] For example, step S15 includes sub-steps S15-1 to S15-2, which are described in detail below: S15-1: The Euclidean distance between two changing features is used as the input sample to train the support vector machine (SVM) model. The prediction model is then obtained based on the trained SVM model. The Euclidean distance quantifies the degree of difference in degradation state between different time windows. A common kernel function in SVM models for handling nonlinear data is the Radial Basis Function (RBF), i.e.: Where x and y are the features of the input sample. It is the Euclidean distance between the features of the input samples. It is an adjustable parameter used to control the density of the Gaussian distribution; using the radial basis kernel, the feature quantities of the same detection marker with different direct detection parameters are mapped to a high-dimensional space, and a hyperplane for classification is found in the new space. In this embodiment of the invention, the change feature quantities of all monitored parameters within a time window are used. The feature vector is used to calculate the distribution kernel function among several parameters from the input samples. By continuously updating and iterating the training, the optimal parameters are output after certain conditions are met or the maximum number of iterations is reached. This determines the hyperplane margin for support vector machine classification, resulting in a prediction model for real-time lifetime prediction of varistors.
[0063] S15-2: Input the time-series data stream of various parameters monitored by the online monitoring terminal for the target varistor into the prediction model, and obtain the real-time remaining lifetime of the target varistor based on the output of the prediction model. A prediction model for real-time lifetime analysis of varistors is constructed according to the above steps. The system continuously receives time-series data streams of several parameters from the online monitoring device, constructs a time window t for the latest detection data of the current resistance, analyzes the data features within the new window, and uses the trained prediction model to classify and predict the current data. Simultaneously, the prediction results are compared with the actual device performance of the varistor collected at the next time step, and the results are fed back into the model. The newly generated feedback data is merged with historical training data. By continuously optimizing the selection of device features, the prediction model's real-time prediction capability for varistor lifetime is improved.
[0064] Based on the same technical concept as the detection method, this invention also provides a real-time lifetime detection system for varistors. The detection system is the same as any of the systems described above. Please refer to [link to relevant documentation]. Figure 4 , Figure 4 The diagram shows the structure of the detection system, which includes a division acquisition module 41, a screening processing module 42, a first acquisition module 43, a second acquisition module 44, and a lifetime prediction module 45.
[0065] The division and acquisition module 41 is used to divide the current impact test of the varistor throughout its entire life cycle into multiple time windows, and acquire multiple correlation parameters characterizing the deterioration of the resistance performance in each time window.
[0066] The screening and processing module 42 is used to analyze the mixed effects between various related parameters in order to screen out the directly measured parameters during the resistance degradation process from multiple related parameters in each time window.
[0067] The first acquisition module 43 is used to obtain the correlation significance and reference factor of the two types of detection parameters based on the correlation between the direct detection parameters and the indirect detection parameters of multiple related parameters.
[0068] The second acquisition module 44 is used to weight the direct detection parameters according to the correlation significance and reference factor to obtain the characteristic quantity of the varistor's tendency to deteriorate in different time windows.
[0069] The lifetime prediction module 45 is used to construct a prediction model based on the changing characteristic quantities, and to use the prediction model to detect the remaining lifetime of the target varistor currently in operation.
[0070] 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.
[0071] 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.
Claims
1. A method for real-time lifetime detection of a varistor, characterized in that, The method includes: The current surge test of the varistor throughout its entire life cycle was divided into multiple time windows, and several correlation parameters characterizing the deterioration of the resistor performance were obtained for each time window. The complex effects among various related parameters are analyzed to select the directly measured parameters during the resistance degradation process from among the multiple related parameters in each time window. Based on the correlation between the direct detection parameters and the indirect detection parameters of the multiple associated parameters, the correlation significance and reference factor of the two types of detection parameters are obtained. The direct detection parameters are weighted according to the correlation significance and the reference factor to obtain the characteristic quantity of the varistor's tendency to deteriorate in different time windows; A prediction model is constructed based on the changing characteristic quantities, and the remaining lifetime of the target varistor currently in operation is detected using the prediction model.
2. The real-time lifetime detection method for a varistor according to claim 1, characterized in that, The current surge test of the varistor throughout its entire life cycle was divided into multiple time windows, and several correlation parameters characterizing the degradation of the resistor's performance in each time window were obtained, including: Based on the degradation characteristics of the varistor in the multi-level surge protection system, the experimental parameters for the current impulse test are configured. The current surge test is initiated and multiple related parameters are obtained until the varistor is completely damaged. These multiple related parameters include varistor voltage, leakage current, nonlinear coefficient, series equivalent capacitance, parallel equivalent capacitance, component reactance, and component impedance.
3. The real-time lifetime detection method for a varistor according to claim 1, characterized in that, The complex influences among various correlated parameters are analyzed to select the directly measured parameters during the resistance degradation process from multiple correlated parameters in each time window, including: Multiple correlation parameters for each time window are used as multidimensional feature samples and input into the support vector machine model; Based on the discrimination results formed by the support vector machine model in the feature space, the correlation parameters that have significant discrimination effects in each time window during the resistance degradation process are screened to obtain the direct detection parameters for the corresponding time window.
4. The real-time lifetime detection method for a varistor according to claim 1, characterized in that, Based on the correlation between the direct detection parameters and the indirect detection parameters of the multiple associated parameters, the correlation significance and reference factor of the two types of detection parameters are obtained, including: The correlation significance between the direct detection parameters and the indirect detection parameters is obtained based on their relative distance in space and time. Based on the differences in the direct detection parameters and the indirect detection parameters in adjacent time windows and the differences in their correlation significance, a reference factor is obtained to characterize the degree of influence of the indirect detection parameters on the direct detection parameters.
5. The real-time lifetime detection method for a varistor according to claim 4, characterized in that, The correlation significance between the direct detection parameter and the indirect detection parameter is obtained based on their relative distance in space and time, including: Based on the Euclidean distance between the current directly detected parameters and the current indirectly detected parameters in the topology diagram, the spatial parameter distance between the two types of current detected parameters is obtained. The topology diagram is a structural diagram formed by the connection relationship of each varistor in a multi-level surge protection system. Based on the time warping distance between the first parameter sequence to which the current directly detected parameter belongs and the second parameter sequence to which the current indirect detected parameter belongs, the temporal correlation distance between the two types of current detected parameters is obtained. The correlation significance of the two types of current detection parameters is obtained based on the ratio of the spatial parameter distance to the temporal correlation distance.
6. The real-time lifetime detection method for a varistor according to claim 4, characterized in that, Based on the differences in variation characteristics and correlation significance between the direct detection parameter and the indirect detection parameter in adjacent time windows, reference factors characterizing the degree of influence of the indirect detection parameter on the direct detection parameter are obtained, including: Based on the difference in the sustained response of the current indirect detection parameter associated with the current direct detection parameter within adjacent time windows, obtain the variation coefficient of the current indirect detection parameter within adjacent time windows. Based on the significance difference of the correlation significance between the current indirect detection parameter and the current direct detection parameter in adjacent time windows, and the historical standard deviation of the current indirect detection parameter, the change coefficient of the current indirect detection parameter in adjacent time windows is obtained. Based on the variation difference coefficient and the variation degree coefficient, a reference factor is obtained to determine the degree of influence of the current indirect detection parameter on the current direct detection parameter.
7. The real-time lifetime detection method for a varistor according to claim 1, characterized in that, The direct detection parameters are weighted according to the correlation significance and the reference factor to obtain the characteristic quantities of the varistor's tendency to deteriorate in different time windows, including: Based on the correlation significance and the reference factor, the reference weight of the direct detection parameter in the corresponding time window is obtained; Based on the reference weights of the direct detection parameters in each time window and the area of the curve peak, the change characteristic quantities under the corresponding time window are obtained.
8. The real-time lifetime detection method for a varistor according to claim 7, characterized in that, Based on the correlation significance and the reference factor, the reference weights of the direct detection parameters in the corresponding time window are obtained, including: The significance difference of each indirect detection parameter in adjacent time windows is obtained based on the difference in the significance of the association of each indirect detection parameter associated with the direct detection parameter in adjacent time windows. The indirect parameter weights of the corresponding indirect detection parameters are obtained by multiplying the significance difference of each indirect detection parameter with the corresponding reference factor. The reference weight of the direct detection parameter is obtained by averaging the weights of all indirect parameters in adjacent time windows.
9. The real-time lifetime detection method for a varistor according to claim 1, characterized in that, A prediction model is constructed based on the changing characteristic quantities, and the remaining lifetime of the target varistor currently in operation is detected using the prediction model, including: The Euclidean distance between two changing feature quantities is used as the input sample of the support vector machine model to train the support vector machine model, and the prediction model is obtained based on the trained support vector machine model. The time-series data stream of various parameters monitored by the online monitoring terminal for the target varistor is input into the prediction model, and the real-time remaining lifetime of the target varistor is obtained based on the output of the prediction model.
10. A real-time life detection system for a varistor, characterized in that, The detection system is the system corresponding to any one of the methods described in claims 1-9, and the system comprises: The acquisition module is used to divide the current impact test of the varistor throughout its entire life cycle into multiple time windows, and acquire multiple correlation parameters characterizing the deterioration of the resistance performance in each time window. The filtering module is used to analyze the mixed effects between various related parameters, so as to filter out the directly detected parameters measured by the instrument during the resistance degradation process from multiple related parameters in each time window; The first obtaining module is used to obtain the correlation significance and reference factor of the two types of detection parameters based on the correlation between the direct detection parameters and the indirect detection parameters of the multiple correlation parameters. The second acquisition module is used to weight the direct detection parameters according to the correlation significance and the reference factor to obtain the change characteristic of the varistor tending to deteriorate in different time windows. The lifetime prediction module is used to construct a prediction model based on the changing characteristic quantity, and to use the prediction model to detect the remaining lifetime of the target varistor currently in operation.