Film capacitor voltage monitoring method and system for intelligent sensor

Abstract

The invention relates to the technical field of intelligent sensing, and discloses a thin film capacitor voltage monitoring method and system for an intelligent sensor, and the method comprises the steps: obtaining initial voltage data; performing noise suppression and iterative state estimation according to the initial voltage data to obtain a filtering voltage sequence; calculating a voltage standard deviation according to the filtering voltage sequence, and if the voltage standard deviation exceeds a preset threshold value, determining a key voltage change trend; obtaining an associated parameter, and carrying out multi-parameter interaction analysis on the associated parameter and the key voltage change trend to obtain a parameter correlation vector; quantifying the contribution degree of each parameter according to the parameter correlation vector, and determining a high-contribution parameter set; according to the high-contribution parameter set and a filtering voltage sequence fitting model, obtaining a capacitor health state estimated value; and judging whether the pre-estimated value is lower than a safety threshold value or not and triggering a real-time alarm. According to the invention, monitoring accuracy and stability can be realized, and an efficient solution is provided for fault early warning of the capacitor.

Description

Technical Field

[0001] This invention relates to the field of intelligent sensing technology, and in particular to a method and system for monitoring the voltage of a thin-film capacitor for intelligent sensors. Background Technology

[0002] Currently, film capacitors, as key energy storage components, are widely used in precision control scenarios such as new energy, industrial frequency converters, and medical equipment because the stability of their operating state is crucial to the safety and reliability of the entire electronic system. Accurate and real-time monitoring of capacitor voltage is a core technology for ensuring the healthy operation of equipment and preventing sudden failures. Especially in complex industrial environments, intelligent sensor systems rely on precise perception of the state of internal components to make reliable decisions. Therefore, improving the monitoring capabilities of core components like film capacitors is a key link in promoting the intelligent development of industry.

[0003] In one existing technology, a simple voltage acquisition circuit is deployed to measure the terminal voltage of a capacitor at fixed time intervals, and the measured value is compared with preset upper and lower threshold values. When the voltage value exceeds this safe range, the system triggers an alarm. However, this method has a single monitoring dimension, relying solely on static voltage thresholds, and cannot capture the voltage fluctuation trends and subtle performance degradation characteristics of the capacitor under dynamic loads or temperature changes. Furthermore, this method ignores the interaction between voltage changes and other key operating parameters such as temperature and current, resulting in an inability to comprehensively assess the true health status of the capacitor. By the time an abnormal voltage is detected, performance degradation has often progressed to a more severe stage, and the system lacks the ability to provide early warnings.

[0004] In summary, existing technologies suffer from insufficient monitoring accuracy. Summary of the Invention

[0005] This invention provides a method and system for monitoring the voltage of a thin-film capacitor for smart sensors, in order to solve the problem of insufficient monitoring accuracy.

[0006] In a first aspect, to address the aforementioned technical problems, the present invention provides a method for monitoring the voltage of a thin-film capacitor for a smart sensor, comprising:

[0007] Obtain initial voltage data;

[0008] Based on the initial voltage data, noise suppression and iterative state estimation are performed to obtain the filtered voltage sequence;

[0009] Based on the filtered voltage sequence, the voltage standard deviation is calculated. If the voltage standard deviation exceeds a preset standard deviation threshold, event marking and sequence transformation are performed to obtain the key voltage change trend.

[0010] Obtain the associated parameters, and perform multi-parameter interactive analysis based on the key voltage change trend and the associated parameters to obtain the parameter correlation vector;

[0011] Based on the parameter correlation vector, the parameter contribution is quantified and determined to obtain a set of high-contribution parameters;

[0012] Based on the set of high contribution parameters and the filtered voltage sequence, feature vector fitting and state estimation are performed to obtain the estimated value of the capacitor's health status.

[0013] Based on the estimated health status of the capacitor, it is determined whether it is below a preset safety threshold. If so, an alarm is triggered, and a structured monitoring log is obtained.

[0014] Preferably, the step of performing noise suppression and iterative state estimation based on the initial voltage data to obtain the filtered voltage sequence includes:

[0015] Based on the initial voltage data, a smoothed voltage sequence is obtained by performing data smoothing preprocessing.

[0016] Based on the smoothed voltage sequence, iterative prediction and correction processing is performed to obtain the filtered voltage sequence.

[0017] Preferably, the step of calculating the voltage standard deviation based on the filtered voltage sequence, and if the voltage standard deviation exceeds a preset standard deviation threshold, performing event marking and sequence transformation to obtain the key voltage change trend, includes:

[0018] Based on the filtered voltage sequence, segmentation processing is performed to obtain voltage data within multiple time windows;

[0019] Calculate the voltage standard deviation for each time window based on the voltage data within the multiple time windows;

[0020] If the voltage standard deviation of any of the time windows exceeds a preset standard deviation threshold, the time window is marked as a dynamic change event, and a set of dynamic change events is obtained.

[0021] Based on the set of dynamic events, frequency domain transformation and energy percentage calculation are performed to obtain the key voltage change trend.

[0022] Preferably, the step of obtaining the correlation parameters, and performing multi-parameter interactive analysis based on the key voltage change trend and the correlation parameters to obtain a parameter correlation vector, includes:

[0023] Obtain relevant parameters from external monitoring systems;

[0024] Based on the key voltage change trend and the associated parameters, a multi-parameter interaction matrix is ​​constructed;

[0025] Based on the multi-parameter interaction matrix, principal component extraction is performed to obtain the dimensionality-reduced feature vector;

[0026] Based on the reduced-dimensionality feature vectors, a classification model is trained to obtain a classification label sequence;

[0027] Based on the classification label sequence, the correlation coefficient between the dimensionality-reduced feature vector and the key voltage change trend is calculated to obtain the parameter correlation vector.

[0028] Preferably, the step of quantifying and determining the parameter contribution based on the parameter correlation vector to obtain a set of high-contribution parameters includes:

[0029] Based on the parameter correlation vector, calculate the weight coefficients corresponding to the associated parameters to obtain the weight coefficient vector;

[0030] Based on the weight coefficient vector, if the weight coefficient is higher than the preset contribution threshold, it is determined to be a high contribution parameter, and a set of high contribution parameters is obtained.

[0031] Preferably, the step of performing feature vector fitting and state estimation based on the set of high-contribution parameters and the filtered voltage sequence to obtain a predicted value of the capacitor's health status includes:

[0032] Based on the set of high contribution parameters and the filtered voltage sequence, covariance calculation is performed to obtain the interactive feature vector;

[0033] Based on the interaction feature vector, regression fitting is performed to generate a performance degradation model;

[0034] Based on the performance degradation model, a weighted fusion calculation is performed on the set of high-contribution parameters to obtain the estimated value of the capacitor's health status.

[0035] Preferably, the step of determining whether the estimated health status of the capacitor is below a preset safety threshold, and if so, triggering an alarm to obtain a structured monitoring log, includes:

[0036] If the estimated health status of the capacitor is lower than a preset safety threshold, an abnormal status detection is performed to obtain an abnormal detection result.

[0037] Based on the anomaly detection results, a trigger command is sent to generate an alarm trigger status;

[0038] Based on the alarm trigger status, the initial voltage data is integrated to generate and update a structured monitoring log.

[0039] In a second aspect, the present invention provides a thin-film capacitor voltage monitoring system for a smart sensor, comprising:

[0040] The data acquisition module is used to acquire initial voltage data;

[0041] The signal processing module is used to perform noise suppression and iterative state estimation based on the initial voltage data to obtain a filtered voltage sequence;

[0042] The trend recognition module is used to calculate the voltage standard deviation based on the filtered voltage sequence. If the voltage standard deviation exceeds a preset standard deviation threshold, event marking and sequence transformation are performed to obtain the key voltage change trend.

[0043] The parameter analysis module is used to acquire related parameters, and perform multi-parameter interactive analysis based on the key voltage change trend and the related parameters to obtain a parameter correlation vector;

[0044] The contribution analysis module is used to quantify and determine the contribution of parameters based on the parameter correlation vector, and obtain a set of high-contribution parameters.

[0045] The state estimation module is used to perform feature vector fitting and state estimation based on the set of high contribution parameters and the filtered voltage sequence to obtain a predicted value of the capacitor's health state.

[0046] The alarm recording module is used to determine whether the capacitor's health status is below a preset safety threshold based on the estimated value. If so, an alarm is triggered to obtain a structured monitoring log.

[0047] Thirdly, the present invention also provides an electronic device including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the thin-film capacitor voltage monitoring method for a smart sensor as described in any one of the preceding claims.

[0048] Fourthly, the present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform the thin-film capacitor voltage monitoring method for smart sensors as described in any one of the preceding claims.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] (1) This invention acquires raw voltage signals by deploying a high-precision sensor array and combines iterative state estimation with Kalman filtering to suppress noise. This method first ensures the precision of the raw data, enabling the capture of subtle dynamic changes in the capacitor. Subsequently, filtering effectively separates noise interference under complex operating conditions, restoring the true voltage change trend. This method overcomes the shortcomings of existing technologies where data inaccuracy is caused by noise, significantly improving the accuracy and stability of voltage monitoring, and providing a reliable data foundation for subsequent health status assessment.

[0051] (2) This invention extracts key voltage change trends and integrates related parameters such as temperature and current for multi-parameter interactive analysis, quantifying the contribution of each parameter to screen out high-contribution parameters. This method not only focuses on the voltage itself but also delves deeper into the key driving factors affecting voltage changes and their weights, constructing a multi-dimensional comprehensive evaluation system. Compared with existing technologies that rely solely on a single static voltage threshold, this invention can more comprehensively and dynamically reveal the internal state changes of capacitors, thereby providing a deeper understanding of their performance degradation mechanisms.

[0052] (3) This invention utilizes selected high-contribution parameters to fit a performance degradation model, outputs a real-time estimated health status of the capacitor, and compares it with a safety threshold to trigger an early warning. This method shifts the monitoring target from the current voltage value to the future health trend. By establishing a degradation model associated with key influencing factors, it achieves early prediction of capacitor failures. This solves the problem that existing technologies cannot effectively predict health status and can only respond after a failure occurs, providing an efficient solution for predictive maintenance and greatly enhancing the operational safety and reliability of the system. Attached Figure Description

[0053] Figure 1 This is a schematic flowchart of a thin-film capacitor voltage monitoring method for smart sensors provided in the first embodiment of the present invention;

[0054] Figure 2 This is a schematic diagram of a thin-film capacitor voltage monitoring system for smart sensors provided in the second embodiment of the present invention. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] Reference Figure 1The first embodiment of the present invention provides a method for monitoring the voltage of a thin-film capacitor for a smart sensor, comprising the following steps:

[0057] S11, acquire initial voltage data;

[0058] S12, based on the initial voltage data, perform noise suppression and iterative state estimation to obtain the filtered voltage sequence;

[0059] S13, Calculate the voltage standard deviation based on the filtered voltage sequence. If the voltage standard deviation exceeds a preset standard deviation threshold, perform event marking and sequence transformation to obtain the key voltage change trend.

[0060] S14, obtain the correlation parameters, and perform multi-parameter interactive analysis based on the key voltage change trend and the correlation parameters to obtain the parameter correlation vector;

[0061] S15, Based on the parameter correlation vector, the parameter contribution is quantified and determined to obtain a set of high contribution parameters;

[0062] S16. Based on the set of high contribution parameters and the filtered voltage sequence, perform feature vector fitting and state estimation to obtain the estimated value of the capacitor's health status.

[0063] S17. Based on the estimated health status of the capacitor, determine whether it is below a preset safety threshold. If so, trigger an alarm and obtain a structured monitoring log.

[0064] In step S11, it is necessary to obtain the initial voltage data.

[0065] In one specific implementation, this step involves deploying a sensor array consisting of multiple high-sensitivity capacitive sensors at key locations on the thin-film capacitor to acquire data. This sensor array can capture minute voltage changes on or inside the capacitor in real time. To ensure precise alignment and consistency of data over time, the acquisition system employs a multi-channel synchronous acquisition mechanism. All sensor channels in the array operate synchronously under a unified high-frequency clock signal. This approach controls the timestamp error of data sampling between different channels to within the nanosecond level, effectively avoiding data distortion caused by time deviations.

[0066] In one embodiment, the sensor array is calibrated using a standard voltage source within a temperature range of 0°C to 50°C, and the temperature-voltage compensation curve is fitted using the least squares method. During data acquisition, if the timestamp error between channels exceeds 1 μs, synchronous recalibration is triggered; if more than 10 sampling points of data are lost consecutively in a single channel, interpolation compensation of adjacent channel data is enabled.

[0067] For example, a sensor array may contain 16 sensing units uniformly distributed on the surface of a capacitor. Each unit has a sampling frequency set to 10 kHz. The acquisition system uses a high-performance 24-bit analog-to-digital converter (ADC) to convert the analog voltage signals captured by the sensors into digital signals. The final output of this step is a structured multidimensional time-series dataset, specifically represented as a data table. Each row represents a sampling time point, containing at least one timestamp column and 16 voltage value columns corresponding to the 16 sensing units. The voltage values ​​are typically in millivolts (mV). This initial voltage data accurately records the voltage distribution of the capacitor at each sampling instant, providing a high-fidelity raw data foundation for subsequent noise suppression and state analysis.

[0068] In step S12, noise suppression and iterative state estimation are performed based on the initial voltage data to obtain a filtered voltage sequence, including:

[0069] Based on the initial voltage data, a smoothed voltage sequence is obtained by performing data smoothing preprocessing.

[0070] Based on the smoothed voltage sequence, iterative prediction and correction processing is performed to obtain the filtered voltage sequence.

[0071] In one specific implementation, to extract a smoother and more stable voltage signal that more closely approximates the actual physical state, this step processes the noisy initial voltage data. First, based on the initial voltage data, a data smoothing preprocessing is performed to obtain a smoothed voltage sequence. Specifically, this data smoothing preprocessing employs a moving average filtering method, replacing the original data point at the center of the window with the average value of data points within a fixed-size window to suppress high-frequency random noise. The moving average window size is a key parameter, determined experimentally based on the characteristics of the signal's sampling frequency and the main noise frequencies. Preferably, the window size is set to 5, striking a balance between effectively smoothing noise and preserving signal details.

[0072] For example, for the voltage value sequence [100050, 100150, 99980, 100020, 99900] mV collected at five consecutive time points in channel 1, a moving average filter with a window size of 5 is applied. The midpoint 99980 mV will then be replaced with the average of these five points, i.e., (100050 + 100150 + 99980 + 100020 + 99900) / 5 = 100020 mV. The output smoothed voltage sequence of this processing is a low-noise time-series voltage sequence generated for each channel in the initial voltage data.

[0073] Subsequently, based on the smoothed voltage sequence, iterative prediction and correction processing is performed to obtain the filtered voltage sequence. Specifically, this processing is based on a capacitor dynamic system model, which is established based on circuit theory. The behavior of the capacitor is equivalent to a first-order or second-order RC circuit network composed of an ideal capacitor (C), an equivalent series resistance (ESR), and an equivalent parallel resistance (EPR). Its state-space expression describes the physical law of the capacitor voltage change over time. The specific parameters of this model (such as the values ​​of C, ESR, and EPR) are obtained from the datasheets provided by the capacitor manufacturer, or are precisely determined through system identification experiments in a calibration environment. The system identification experiment includes applying a known excitation signal (such as a step current) to the capacitor in a calibration environment, collecting voltage response data, and fitting the model parameters using parameter estimation methods such as the least squares method. Those skilled in the art can refer to standard system identification literature (such as Ljung's "System Identification: Theory and Applications") to implement this.

[0074] The prediction phase of this process predicts the voltage state at the next moment based on the current system state estimate and the dynamic system model. This prediction process is achieved through state transition equations: ,in, This represents the predicted prior state value at time k, whose main component is voltage. It is the optimal state estimate at time k-1. It is the state transition matrix. B represents the external control input at time k, such as the circuit's input current, and is the control input matrix. The state transition matrix A is determined by discretizing the RC time constant of the capacitor model; it quantifies the natural decay or change of the capacitor voltage from time k-1 to time k in the absence of external input.

[0075] The correction phase compares the actual measured value at the next moment (the corresponding value in the smoothed voltage sequence) with the prior state prediction value and calculates the deviation between them. The specific process for generating a correction amount is as follows: First, the system calculates an optimal correction gain, which is based on the system's prior state covariance and measurement noise covariance. The prior state covariance quantifies the uncertainty of the dynamic system model's prediction results, while the measurement noise covariance quantifies the uncertainty of the sensor measurements. This gain optimizes the weighting between the model prediction and the sensor measurements to minimize the error in the final state estimation. Then, the deviation calculated at the beginning of step one is multiplied by this optimal correction gain to obtain the final correction amount. Finally, this correction amount is added to the prior state prediction value to obtain a more accurate optimal state estimate at time k. The final output filtered voltage sequence of this step is an iteratively optimized time sequence generated for each channel that accurately reflects the dynamic behavior of the capacitor's true voltage.

[0076] In step S13, the voltage standard deviation is calculated based on the filtered voltage sequence. If the voltage standard deviation exceeds a preset standard deviation threshold, event marking and sequence transformation are performed to obtain the key voltage change trend, including:

[0077] Based on the filtered voltage sequence, segmentation processing is performed to obtain voltage data within multiple time windows;

[0078] Calculate the voltage standard deviation for each time window based on the voltage data within the multiple time windows;

[0079] If the voltage standard deviation of any of the time windows exceeds a preset standard deviation threshold, the time window is marked as a dynamic change event, and a set of dynamic change events is obtained.

[0080] Based on the set of dynamic events, frequency domain transformation and energy percentage calculation are performed to obtain the key voltage change trend.

[0081] In one specific implementation, this step aims to automatically identify significant dynamic change events from a smooth voltage sequence. This step first segments the filtered voltage sequence to obtain voltage data within multiple time windows. Specifically, this processing employs a sliding window method, sliding across the filtered voltage sequence with a fixed window length and step size to divide the continuous sequence data into a series of data segments, i.e., time windows. The window length is set based on the shortest duration of the event to be captured; for example, it can be set to 1 millisecond. The step size is typically smaller than the window length to ensure no event is missed.

[0082] Subsequently, based on the voltage data within the multiple time windows, the standard deviation of the voltage for each time window is calculated. For each time window containing N voltage data points, the standard deviation is calculated as follows: first, the average value of all voltage values ​​within the window is calculated; then, the sum of the squares of the differences between each voltage value and the average value is calculated; this sum is divided by N, and the square root is taken to obtain the standard deviation of the voltage for that time window. For example, assuming a normal operating time window A has a data sequence of [100010, 100025, 100015, 99990] mV, its calculated standard deviation is 12.75 mV. In another time window B where a load change occurs, the data sequence is [100100, 105200, 104800, 101000] mV, and its calculated standard deviation is 709.31 mV.

[0083] Next, if the voltage standard deviation of any of the time windows exceeds a preset standard deviation threshold, the time window is marked as a dynamic change event, resulting in a dynamic change event set. The standard deviation threshold here is a dynamic event discrimination standard deviation threshold, set based on the analysis of a large amount of historical data of the capacitor under normal and stable operating conditions (e.g., collecting at least 1000 hours of voltage data from normal operation and calculating its standard deviation distribution). It is typically set to the mean voltage standard deviation under stable operating conditions plus three times the standard deviation, and linearly corrected according to the actual application environment (e.g., temperature fluctuation range ±10℃). For example, it can be set to 50mV. Those skilled in the art can adjust this threshold according to capacitor specifications, operating environment, and monitoring requirements to balance the sensitivity and false alarm rate of event detection. The specific judgment process is as follows: if the calculated standard deviation is greater than the threshold, the corresponding time window data is marked as a dynamic change event and added to the dynamic change event set; if it is less than or equal to the threshold, it is determined to be a normal fluctuation. In the example above, the standard deviation of time window A, 12.75mV, is less than 50mV and is judged as normal fluctuation; while the standard deviation of time window B, 709.31mV, is much greater than 50mV, so it is marked as a dynamic change event and added to the dynamic change event set.

[0084] Finally, based on the set of dynamically changing events, frequency domain transformation and energy percentage calculation are performed to obtain the key voltage change trend. Specifically, a Fast Fourier Transform is applied to each event data segment in the set of dynamically changing events, converting it from a time-domain signal into a complex frequency-domain sequence containing amplitude and phase. Based on this, any frequency component in the frequency-domain sequence is calculated. The energy, which is determined by the square of its amplitude. To indicate, among which, Represents a specific, fixed frequency component. This indicates that after performing a Fast Fourier Transform on a time-domain signal, the frequency... The complex Fourier coefficients are obtained at the given point. Then, the total energy of the event data segment is obtained by summing the energy of all frequency components. Next, the proportion of energy of each major frequency component to the total energy is calculated, as follows: ,

[0085] Here, k is a specific, fixed index representing a particular frequency component that is currently being monitored. It is a frequency component The energy percentage, j represents a summation index used to iterate through all frequency components. Specifically, the first frequency component is taken. energy Add a second frequency component energy Then, the energy of each subsequent frequency component is added sequentially until the energy of the last frequency component is exhausted. The final result of this summation is the total energy of the signal. The dominant frequency components are those frequencies that account for a high percentage of energy and contribute significantly to the overall signal fluctuation.

[0086] For example, the analysis results might show that 60% of the total energy of the event signal is concentrated in the 50Hz low-frequency component related to load changes, while another 30% is concentrated in the 20kHz high-frequency component related to the inverter switching frequency. The final output key voltage change trend of this step is structured data, generating a record for the dynamic event of time window B, which includes the start and end times of the event, as well as the key frequency components (50Hz and 20kHz) and their corresponding energy percentages (60% and 30%).

[0087] In step S14, correlation parameters are obtained, and multi-parameter interactive analysis is performed based on the key voltage change trend and the correlation parameters to obtain a parameter correlation vector, including:

[0088] Obtain relevant parameters from external monitoring systems;

[0089] Based on the key voltage change trend and the associated parameters, a multi-parameter interaction matrix is ​​constructed;

[0090] Based on the multi-parameter interaction matrix, principal component extraction is performed to obtain the dimensionality-reduced feature vector;

[0091] Based on the reduced-dimensionality feature vectors, a classification model is trained to obtain a classification label sequence;

[0092] Based on the classification label sequence, the correlation coefficient between the dimensionality-reduced feature vector and the key voltage change trend is calculated to obtain the parameter correlation vector.

[0093] It should be noted that the key voltage change trend is used to identify the time period corresponding to the dynamic change event determined in step S13; when constructing the multi-parameter interaction matrix, only the voltage, temperature and current time series within these time periods are sliced ​​using the same sliding window parameters (window length and step size) as in step S13, and the mean and standard deviation of the data within each time window are calculated to form the multi-parameter interaction matrix; the energy proportion information obtained by frequency domain transformation is not directly used for matrix construction, but can be used for subsequent analysis or diagnosis.

[0094] In one specific implementation, this step aims to analyze the intrinsic relationship between voltage changes and other physical parameters. This step first acquires the relevant parameters from an external monitoring system. This external monitoring system refers to a set of dedicated sensors and their interface circuits integrated with the main voltage acquisition system and synchronized with the data time. These primarily include a temperature sensor for measuring the capacitor's operating temperature and a current sensor for measuring the current flowing through the capacitor.

[0095] Subsequently, a multi-parameter interaction matrix is ​​constructed based on the key voltage change trend and the associated parameters. Specifically, this construction process uses a sliding window with the same window length and step size as in step S13 for the time period in which the key voltage change trend occurs. It synchronously slices the voltage, temperature, and current time series, and calculates the mean and standard deviation of the data within each time window, thus obtaining a six-dimensional feature vector, for example: [mean voltage: 102775mV, standard deviation voltage: 2500mV, mean current: 10A, standard deviation current: 3A, mean temperature: 40.5℃, standard deviation temperature: 0.2℃]. Stacking the six-dimensional feature vectors generated from all time windows in chronological order constitutes the multi-parameter interaction matrix.

[0096] Next, principal component extraction is performed based on the multi-parameter interaction matrix to obtain the dimensionality-reduced eigenvectors. Specifically, this process first standardizes each column of the multi-parameter interaction matrix to have a mean of 0 and a standard deviation of 1, eliminating the influence of different physical dimensions. Then, the covariance matrix of the standardized matrix is ​​calculated, and eigenvalue decomposition is performed on the covariance matrix to obtain a set of eigenvalues ​​sorted from largest to smallest and their corresponding eigenvectors. The dimensionality after dimensionality reduction is determined by calculating the cumulative variance contribution rate, i.e., selecting the first k principal components such that the sum of these k eigenvalues ​​accounts for more than a preset 95% variance contribution rate threshold. Finally, matrix multiplication is performed between the standardized multi-parameter interaction matrix and the projection matrix composed of these k selected eigenvectors, and the result is the dimensionality-reduced eigenvector.

[0097] Then, based on the dimensionality-reduced feature vectors, a classification model is trained to obtain a classification label sequence. The classification model here is a support vector machine (SVM) model, which is trained offline using a large historical dataset containing numerous normal and abnormal operating conditions. The input features for training are the dimensionality-reduced feature vectors generated from the historical data, and the labels are determined by whether or not dynamic change events were marked when applied to the historical data in step S13. Model parameter selection is performed using a grid search combined with five-fold cross-validation to find the optimal kernel function (such as a radial basis function) and regularization parameter C. For example, the search range is set to C∈[0.1,10], and the kernel function is a radial basis function (γ=0.01), thereby maximizing the model's classification accuracy on the validation set. The trained model is used to classify the real-time generated dimensionality-reduced feature vectors and output the corresponding classification label sequence.

[0098] Finally, based on the classification label sequence, the correlation coefficient between the dimensionality-reduced feature vector and the key voltage change trend is calculated to obtain the parameter correlation vector. Specifically, this calculation process focuses on the time window marked as a dynamic change event by the classification label sequence. The Pearson correlation coefficient method is used to calculate the correlation between the dimensionality-reduced feature vector within this window and the voltage standard deviation calculated for the same time window in step S13, resulting in a correlation coefficient value within the range of [-1, 1].

[0099] In one implementation, firstly, the average value of all dimensionality-reduced eigenvectors (in a certain dimension) and the average value of all corresponding voltage standard deviations are calculated for each time window. Then, the deviation of the dimensionality-reduced eigenvector value for each time window from the aforementioned average value, and the deviation of its corresponding voltage standard deviation from the aforementioned average value, are calculated. Next, each pair of deviations is multiplied and summed to obtain the covariance sum of the deviations. Finally, this covariance sum is divided by the product of the standard deviations of the two variables to obtain a correlation coefficient value in the range [-1, 1]. The correlation coefficients calculated for all time windows marked as dynamically changing events collectively constitute the final output parameter correlation vector of this step.

[0100] It should be noted that the calculation of the parameter correlation vector focuses on the time window marked as a dynamic change event by the classification label sequence; for each time window, the Pearson correlation coefficient between the dimensionality-reduced feature vector and the voltage standard deviation calculated for the same time window in step S13 is calculated; the voltage standard deviation, as a quantitative indicator of voltage dynamic change, is used to evaluate the correlation between multi-parameter features and voltage fluctuations, thereby obtaining the parameter correlation vector.

[0101] In step S15, based on the parameter correlation vector, parameter contribution is quantified and determined to obtain a set of high-contribution parameters, including:

[0102] Based on the parameter correlation vector, calculate the weight coefficients corresponding to the associated parameters to obtain the weight coefficient vector;

[0103] Based on the weight coefficient vector, if the weight coefficient is higher than the preset contribution threshold, it is determined to be a high contribution parameter, and a set of high contribution parameters is obtained.

[0104] In one specific implementation, this step aims to select the core parameter that has the greatest impact on the capacitor's state from multiple related parameters. This step first calculates the weight coefficients corresponding to the related parameters based on the parameter correlation vector, obtaining a weight coefficient vector. Specifically, the calculation of these weight coefficients follows the results of the principal component analysis in step S14. After the principal component analysis performs eigenvalue decomposition on the covariance matrix of the multi-parameter interaction matrix, the eigenvector corresponding to the largest eigenvalue is selected; this eigenvector is the loading vector of the first principal component.

[0105] It should be noted that the load vector of the first principal component directly quantifies the projection weight of its corresponding original correlation parameter (such as mean temperature, standard deviation of current, etc.) in the direction of maximum data change. Therefore, this vector is directly used as the weight coefficient vector, and each element in the vector is the contribution quantification value of the corresponding correlation parameter.

[0106] For example, assuming the original six correlation parameters are the mean voltage, standard deviation of voltage, mean temperature, standard deviation of temperature, mean current, and standard deviation of current, the calculated weight coefficient vector may be [0.75, 0.10, 0.15, 0.05, 0.55, 0.30], where 0.75 is the weight coefficient corresponding to the mean voltage.

[0107] Subsequently, based on the weight coefficient vector, if the weight coefficient is higher than a preset contribution threshold, it is determined to be a high contribution parameter, thus obtaining a set of high contribution parameters. The contribution threshold here is officially called the high contribution parameter determination threshold. Its setting is based on historical data analysis combined with the Pareto principle, aiming to filter out a few key parameters that can explain most of the dynamic behavior of the system. For example, it can be set to 0.5. The specific process of judgment is as follows: the absolute value of each element in the weight coefficient vector is compared with the contribution threshold. If the absolute value of an element is greater than the threshold, its corresponding associated parameter is determined to be a high contribution parameter and included in the final output set of high contribution parameters for this step. In the aforementioned example, the absolute values ​​of the weight coefficients are [0.75, 0.10, 0.15, 0.05, 0.55, 0.30], where 0.75 and 0.55 are both greater than 0.5. Therefore, their corresponding parameters, the average voltage and average current, are determined to be high contribution parameters and included in the final output set of high contribution parameters for this step.

[0108] In step S16, based on the set of high-contribution parameters and the filtered voltage sequence, feature vector fitting and state estimation are performed to obtain a predicted value of the capacitor's health state, including:

[0109] Based on the set of high contribution parameters and the filtered voltage sequence, covariance calculation is performed to obtain the interactive feature vector;

[0110] Based on the interaction feature vector, regression fitting is performed to generate a performance degradation model;

[0111] Based on the performance degradation model, a weighted fusion calculation is performed on the set of high-contribution parameters to obtain the estimated value of the capacitor's health status.

[0112] In one specific implementation, this step aims to establish a model capable of predicting the long-term performance degradation trend of capacitors. This step first calculates the covariance based on the set of high-contribution parameters and the filtered voltage sequence to obtain an interactive feature vector. Specifically, this calculation process extracts the time series (e.g., the "mean voltage" sequence) of the high-contribution parameters determined in step S15 and the filtered voltage sequence. The specific calculation process is as follows: first, the average value of these two time series is calculated; then, for each time point, the deviation between the value at that point and the average value of the respective sequence is calculated; next, these two deviations are multiplied to obtain the deviation product for that time point; finally, the average of the deviation products at all time points is calculated, and this average value is the covariance between the two time series. This process is repeated for all high-contribution parameters, and the resulting covariance values ​​collectively constitute the interactive feature vector.

[0113] Subsequently, regression fitting is performed based on the interaction feature vector to generate a performance degradation model. This performance degradation model is a linear regression model. The training process uses historical datasets derived from accelerated aging experiments or long-term monitoring of capacitors. Each data point in this dataset contains a historical interaction feature vector calculated using the above method, and a corresponding true health status value normalized to the 0-1 range through offline precise measurements (such as capacitance and equivalent series resistance). The regression fitting process employs ordinary least squares, which mathematically searches for an optimal set of model coefficients and a bias term such that, for all data in the training set, the sum of the squared differences between the health status values ​​predicted by the model based on the interaction feature vector and the actual health status values ​​is minimized. The model parameter selection refers to determining this optimal set of coefficients and bias terms, which are fixed after training.

[0114] Finally, based on the performance degradation model, a weighted fusion calculation is performed on the set of high-contribution parameters to obtain the estimated health status of the capacitor. This calculation process involves inputting the real-time generated interaction feature vector into the trained performance degradation model. The operation performed within the model is the weighted fusion calculation, which multiplies each element (i.e., each covariance value) in the interaction feature vector by its corresponding model coefficient determined during training. Then, all products are summed, and the model's bias term is added to output the final prediction result.

[0115] In step S16, in addition to the methods described above, the following may also be included:

[0116] Based on the set of high contribution parameters and the filtered voltage sequence, time-domain feature extraction and standardization are performed to obtain a normalized feature vector;

[0117] Based on the normalized feature vector, regression fitting is performed to generate a performance degradation model;

[0118] Based on the performance degradation model, the normalized feature vector is weighted and fused to obtain the estimated health status of the capacitor.

[0119] In one specific implementation, this step aims to establish a model capable of predicting the long-term performance degradation trend of capacitors. This step first involves time-domain feature extraction and standardization based on the set of high-contribution parameters and the filtered voltage sequence to obtain a normalized feature vector. Specifically, this process extracts the time series of the high-contribution parameters determined in step S15 (e.g., the average voltage sequence and the average current sequence) and the filtered voltage sequence. For each time window (e.g., the same window length and step size as in step S13), calculate the single-parameter features, the statistics of each high-contribution parameter sequence within this time window, such as mean, standard deviation, and root mean square (RMS); and the correlation features, the statistics of the filtered voltage sequence within this time window, such as mean and standard deviation. Then, combine all the extracted feature values ​​into an initial feature vector. To ensure that features of different dimensions and orders of magnitude are comparable in the model, the initial feature vector is standardized using the Z-score method, which calculates the mean and standard deviation for each feature dimension based on the training dataset, then subtracts the mean from all values ​​of that dimension and divides by the standard deviation. The processed data has a mean of 0 and a standard deviation of 1 for each dimension, thus obtaining a normalized feature vector.

[0120] Subsequently, regression fitting is performed based on the normalized feature vector to generate a performance degradation model. This performance degradation model is a linear regression model, trained using historical datasets derived from accelerated aging experiments or long-term monitoring of capacitors. Each data point in this dataset contains a historical normalized feature vector calculated and standardized using the aforementioned method, and a corresponding true health status value normalized to the 0-1 range through offline precise measurements (such as capacitance or equivalent series resistance). The regression fitting process employs ordinary least squares, which uses mathematical calculations to find an optimal set of model coefficients and a bias term, minimizing the sum of squared differences between the health status values ​​predicted by the model based on the normalized feature vector and the true health status values ​​for all data in the training set.

[0121] Finally, based on the performance degradation model, the normalized feature vector is weighted and fused to obtain the estimated health status of the capacitor. This calculation process involves inputting the real-time generated and standardized feature vector into the trained performance degradation model. The operation performed inside the model is the weighted fusion calculation, which multiplies each element in the feature vector by its corresponding model coefficient determined during training, then sums all the products and adds the model's bias term to output the final prediction result.

[0122] In step S17, based on the estimated health status of the capacitor, it is determined whether it is below a preset safety threshold. If so, an alarm is triggered, and a structured monitoring log is obtained, including:

[0123] If the estimated health status of the capacitor is lower than a preset safety threshold, an abnormal status detection is performed to obtain an abnormal detection result.

[0124] Based on the anomaly detection results, a trigger command is sent to generate an alarm trigger status;

[0125] Based on the alarm trigger status, the initial voltage data is integrated to generate and update a structured monitoring log.

[0126] In one specific implementation, this step is the final output of the system, designed to respond based on the health status prediction result. This step first performs anomaly detection if the predicted health status of the capacitor is lower than a preset safety threshold, obtaining an anomaly detection result. This safety threshold is a capacitor health status safety threshold, set based on the lifespan termination parameters provided by the capacitor manufacturer; for example, it can be set to 0.9. When the predicted health status is lower than this threshold, the system uses an anomaly detection algorithm based on statistical process control to process the predicted value.

[0127] The specific process of processing the estimated value using this algorithm is as follows: First, the system extracts the mean and standard deviation of the estimated health status from a historical database representing the healthy operation stages of the capacitor. This historical database originates from a series of estimated value data confirmed to be in a healthy state, collected and stored through continuous monitoring during the calibration phase of the capacitor's initial operation. Then, the deviation between the current estimated health status and the historical benchmark mean is calculated, and this deviation is divided by the historical benchmark standard deviation to obtain a Z-score. If the absolute value of the calculated Z-score exceeds a preset Z-score threshold (exemplarily set to 2), a statistically significant anomaly is determined, and the output anomaly detection result of this step is identified as "abnormal".

[0128] It should be noted that the preset Z-score threshold is based on the confidence interval theory in statistics, which aims to distinguish between normal data fluctuations and statistically significant deviations. For example, the threshold is set to 2, which corresponds to a confidence level of about 95%. That is, when the estimated value deviates from the historical mean by more than two standard deviations, the system has a 95% confidence that the deviation is not caused by random noise.

[0129] Subsequently, based on the anomaly detection result, a trigger command is sent to generate an alarm trigger status. When the anomaly detection result is "abnormal," the system uses a message queue mechanism to asynchronously send a data packet containing a timestamp, anomaly type, and a current health status estimate to the real-time alarm module as a trigger command. The output alarm trigger status of this step is set to "triggered."

[0130] Finally, based on the alarm trigger status, the initial voltage data is fused to generate and update a structured monitoring log. When the alarm trigger status is "triggered," the fusion process involves the system creating a structured data object containing multiple preset fields, including a timestamp, alarm trigger status, health status estimate, and key voltage statistics. These fields are set to ensure that the log records the minimum complete set of information required for fault diagnosis: the timestamp is used to locate the time of the fault; the alarm trigger status is used to explain the reason for the record; the health status estimate is used to quantify the severity of the fault; and the key voltage statistics (such as the average voltage extracted from the initial voltage data) provide a snapshot of the original operating conditions at the moment of the fault, facilitating subsequent in-depth root cause analysis.

[0131] The system fills the corresponding fields of the data object with the current timestamp, the "triggered" alarm status, the current health status estimate, and key statistics (such as the average voltage) extracted from the initial voltage data corresponding to that time point, thus forming a complete log record. This record is appended to the system monitoring log database, thereby completing the final output of this step: the update of the structured monitoring log.

[0132] It should be noted that the historical data (used for threshold setting and model training) must cover the entire life cycle of the capacitor, including no-load, rated load, overload (±20%), and temperature cycling (-40℃~85℃) scenarios, with a data volume of no less than 10. 6 Sample points are marked with their health status (based on a capacity decay rate of ≤5% as the health standard).

[0133] In summary, this invention significantly improves the accuracy and stability of capacitor health status monitoring through the synergistic optimization of multi-parameter fusion and noise suppression, providing an efficient solution for capacitor fault early warning in complex environments.

[0134] Reference Figure 2 The second embodiment of the present invention provides a thin-film capacitor voltage monitoring system for smart sensors, comprising:

[0135] The data acquisition module is used to acquire initial voltage data;

[0136] The signal processing module is used to perform noise suppression and iterative state estimation based on the initial voltage data to obtain a filtered voltage sequence;

[0137] The trend recognition module is used to calculate the voltage standard deviation based on the filtered voltage sequence. If the voltage standard deviation exceeds a preset standard deviation threshold, event marking and sequence transformation are performed to obtain the key voltage change trend.

[0138] The parameter analysis module is used to acquire related parameters, and perform multi-parameter interactive analysis based on the key voltage change trend and the related parameters to obtain a parameter correlation vector;

[0139] The contribution analysis module is used to quantify and determine the contribution of parameters based on the parameter correlation vector, and obtain a set of high-contribution parameters.

[0140] The state estimation module is used to perform feature vector fitting and state estimation based on the set of high contribution parameters and the filtered voltage sequence to obtain a predicted value of the capacitor's health state.

[0141] The alarm recording module is used to determine whether the capacitor's health status is below a preset safety threshold based on the estimated value. If so, an alarm is triggered to obtain a structured monitoring log.

[0142] It should be noted that the thin-film capacitor voltage monitoring system for smart sensors provided in this embodiment of the invention is used to execute all the process steps of the thin-film capacitor voltage monitoring method for smart sensors described in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0143] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a thin-film capacitor voltage monitoring program for a smart sensor. When the processor executes the computer program, it implements the steps in the various embodiments of the thin-film capacitor voltage monitoring method for smart sensors described above, for example... Figure 1 The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above system embodiments, such as the signal processing module.

[0144] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0145] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0146] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0147] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0148] If the modules / units integrated into the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or system capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0149] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0150] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for thin film capacitor voltage monitoring for smart sensors, characterized by, The method comprises the following steps: acquiring initial voltage data; performing noise suppression and iterative state estimation according to the initial voltage data to obtain a filtered voltage sequence; calculating a voltage standard deviation according to the filtered voltage sequence, and if the voltage standard deviation exceeds a preset standard deviation threshold, performing event marking and sequence transformation to obtain a key voltage change trend; acquiring associated parameters, performing multi-parameter interaction analysis according to the key voltage change trend and the associated parameters to obtain a parameter correlation vector; performing parameter contribution quantification and determination according to the parameter correlation vector to obtain a high-contribution parameter set; performing feature vector fitting and state estimation according to the high-contribution parameter set and the filtered voltage sequence to obtain a capacitor health state estimate value; judging whether the capacitor health state estimate value is lower than a preset safety threshold, and if yes, triggering an alarm to obtain a structured monitoring log.

2. The thin film capacitor voltage monitoring method for smart sensors according to claim 1, wherein, The method of performing noise suppression and iterative state estimation according to the initial voltage data to obtain a filtered voltage sequence comprises the following steps: performing data smoothing preprocessing according to the initial voltage data to obtain a smoothed voltage sequence; performing iterative prediction and correction processing according to the smoothed voltage sequence to obtain a filtered voltage sequence.

3. The thin film capacitor voltage monitoring method for smart sensors according to claim 1, wherein, The method of calculating a voltage standard deviation according to the filtered voltage sequence, and if the voltage standard deviation exceeds a preset standard deviation threshold, performing event marking and sequence transformation to obtain a key voltage change trend comprises the following steps: performing segmentation processing according to the filtered voltage sequence to obtain voltage data in a plurality of time windows; calculating a voltage standard deviation of each time window according to the voltage data in the plurality of time windows; if the voltage standard deviation of any time window exceeds a preset standard deviation threshold, marking the time window as a dynamic change event to obtain a dynamic change event set; performing frequency domain transformation and energy proportion calculation according to the dynamic change event set to obtain a key voltage change trend.

4. The thin film capacitor voltage monitoring method for smart sensors of claim 1, wherein, The method of acquiring associated parameters, performing multi-parameter interaction analysis according to the key voltage change trend and the associated parameters to obtain a parameter correlation vector comprises the following steps: acquiring associated parameters from an external monitoring system; constructing a multi-parameter interaction matrix according to the key voltage change trend and the associated parameters; performing principal component extraction according to the multi-parameter interaction matrix to obtain a reduced-dimension feature vector; performing classification model training according to the reduced-dimension feature vector to obtain a classification label sequence; calculating a correlation coefficient of the reduced-dimension feature vector and the key voltage change trend according to the classification label sequence to obtain a parameter correlation vector.

5. The method for thin film capacitor voltage monitoring for smart sensors as claimed in claim 1 wherein, The method of performing parameter contribution quantification and determination according to the parameter correlation vector to obtain a high-contribution parameter set comprises the following steps: calculating a weight coefficient corresponding to the associated parameters according to the parameter correlation vector to obtain a weight coefficient vector; if the weight coefficient is higher than a preset contribution threshold, determining it as a high-contribution parameter to obtain a high-contribution parameter set.

6. The thin film capacitor voltage monitoring method for smart sensors according to claim 2, wherein, The method of performing feature vector fitting and state estimation according to the high-contribution parameter set and the filtered voltage sequence to obtain a capacitor health state estimate value comprises the following steps: According to the high contribution parameter set and the filtered voltage sequence, covariance calculation is performed to obtain an interaction feature vector; According to the interaction feature vector, regression fitting is performed to generate a performance degradation model; According to the performance degradation model, weighted fusion calculation is performed on the high contribution parameter set to obtain a capacitor health state estimate value.

7. The method for thin film capacitor voltage monitoring of smart sensors according to claim 1, wherein, According to the capacitor health state estimate value, it is judged whether it is lower than a preset safety threshold, if yes, alarm triggering is performed to obtain a structured monitoring log, including: If the capacitor health state estimate value is lower than the preset safety threshold, abnormal state detection is performed to obtain an abnormal detection result; According to the abnormal detection result, a trigger instruction is sent to generate an alarm triggering state; According to the alarm triggering state, the initial voltage data is fused to generate and update the structured monitoring log.

8. A thin film capacitor voltage monitoring system for a smart sensor, characterized by, Including: A data acquisition module is configured to acquire initial voltage data; A signal processing module is configured to perform noise suppression and iterative state estimation on the initial voltage data to obtain a filtered voltage sequence; A trend identification module is configured to calculate a voltage standard deviation according to the filtered voltage sequence, and if the voltage standard deviation exceeds a preset standard deviation threshold, event marking and sequence transformation are performed to obtain a key voltage change trend; A parameter analysis module is configured to obtain associated parameters, and perform multi-parameter interaction analysis on the key voltage change trend and the associated parameters to obtain a parameter correlation vector; A contribution analysis module is configured to perform parameter contribution quantification and determination according to the parameter correlation vector to obtain a high contribution parameter set; A state estimation module is configured to perform feature vector fitting and state estimation on the high contribution parameter set and the filtered voltage sequence to obtain a capacitor health state estimate value; An alarm recording module is configured to judge whether the capacitor health state estimate value is lower than a preset safety threshold, if yes, alarm triggering is performed to obtain a structured monitoring log.

Images