Capacitor implosion risk analysis method and system based on time sequence characteristics

Abstract

The invention discloses a capacitor implosion risk analysis method and system based on time sequence characteristics, and relates to the technical field of risk analysis, and the method comprises the steps: collecting the current, temperature and equivalent series resistance data in the operation process of a capacitor, and constructing an operation parameter time sequence data set; first-order instant change features are extracted in each time window, and second-order time sequence features are obtained based on continuous window analysis; identifying abnormal change features and counting occurrence times, duration and multi-parameter joint modes to form an enhanced implosion risk representation vector; and classifying each time window based on implosion risk characteristics, judging implosion risks to be low, medium and high levels, and generating a detailed risk early warning sequence and corresponding processing suggestions. According to the method, multi-parameter and multi-time-dimension continuous risk analysis and graded early warning can be realized, the operation safety of the capacitor is effectively improved, implosion accidents are prevented, and the overall risk management and control efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of risk analysis technology, specifically to a method and system for analyzing the risk of capacitor implosion based on time-series characteristics. Background Technology

[0002] Capacitors are widely used in power systems and electronic equipment, and their safe operation has a significant impact on equipment reliability and system stability. With the continuous expansion of capacitor usage and increasingly complex operating conditions, traditional risk analysis methods relying on periodic manual inspections or single-indicator monitoring are no longer sufficient to meet the requirements of efficiency and real-time monitoring. Implosion accidents can be caused by abnormal changes in operating parameters such as current, temperature, and equivalent series resistance. Failure to detect these accidents in a timely manner will seriously affect equipment safety and may even lead to system shutdown or economic losses.

[0003] However, existing capacitor operation monitoring systems suffer from several problems. Different capacitors operate under varying conditions and have different parameter characteristics, leading to inconsistent response rates for the same monitoring indicators on different devices. Traditional risk analysis methods may suffer from delays, omissions, or misjudgments in anomaly identification and risk assessment, resulting in untimely risk warnings or wasted resources, thus reducing the effectiveness of implosion protection. Therefore, a time-series characteristic-based capacitor implosion risk analysis method and system are needed to address these issues. Summary of the Invention

[0004] To address the aforementioned technical problems, a method and system for analyzing capacitor implosion risk based on time-series characteristics are provided. This technical solution solves the problems mentioned in the background technology, such as the differences in operating conditions and parameter characteristics of different capacitors, the different response effects of the same monitoring indicators on different devices, and the potential for lag, missed judgments, or misjudgments in abnormal feature identification and risk assessment in traditional risk analysis methods, leading to untimely risk warnings or resource waste and reducing the effectiveness of implosion protection.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for analyzing the risk of capacitor implosion based on time-series characteristics, comprising: Collect operating parameter data during capacitor operation and construct an operating parameter time series dataset, wherein the operating parameter data includes current data, temperature data and equivalent series resistance data; Based on the time series dataset of running parameters, the instantaneous change features corresponding to each running parameter are extracted in each time window to obtain the first-order time series feature dataset. Based on the first-order time series feature dataset, the variation characteristics of instantaneous change features between continuous observation windows are analyzed to obtain the second-order time series feature dataset. Based on the second-order time series feature dataset, the second-order time series features corresponding to the temperature data, current data and equivalent series resistance data within the current time window are extracted to identify abnormal change features. Based on the abnormal change characteristics, the abnormal change characteristics are segmented and summarized in the time dimension, and the occurrence frequency and duration of the abnormal change characteristics in each segment are counted to obtain the implosion risk characteristic dataset. Based on the implosion risk feature dataset, the implosion risk of each time window is divided into different risk levels, and products of different risk levels are classified and sorted to obtain implosion risk warning results.

[0006] In an optional embodiment, the step of extracting the instantaneous change features corresponding to each running parameter within each time window based on the running parameter time series dataset to obtain a first-order time series feature dataset specifically includes: Obtain the current data, temperature data, and equivalent series resistance data corresponding to each time window in the runtime parameter time series dataset to form a window runtime parameter dataset; Time alignment processing is performed on the runtime parameter data in the window runtime parameter dataset to obtain runtime parameter time series data under a unified time scale, and runtime parameter time series data is constructed. Based on the time series data of the operating parameters, the changes in the operating parameters between adjacent sampling times are obtained, and a sequence of changes in the operating parameters is constructed. Acquire historical operational parameter data and calculate the frequency and average magnitude of abnormal changes at each time point to obtain quantitative indicators for weighting. The weighted change sequence is obtained by weighting the sequence of changes in operating parameters within the current time window based on quantitative indicators. Based on the weighted change sequence, the rate of change and magnitude of change of each operating parameter within the current time window are extracted. Within each time window, temperature time series data and current time series data are extracted from the running parameter window dataset, and the distance between the two sets of sequences is calculated according to the dynamic time warping method, which is used as the first instantaneous change feature. Within each time window, the equivalent series resistance time series data is extracted from the running parameter window dataset and subjected to Mann-Kendall trend test to obtain the statistical significance index value, which serves as the second instantaneous change feature. The rate of change and magnitude characteristics are combined with the first instantaneous change characteristics and the second instantaneous change characteristics to form the instantaneous change characteristic vector for each time window; The instantaneous change feature vectors of each time window are collected to construct the final first-order time series feature dataset.

[0007] In an optional embodiment, the step of analyzing the variation characteristics of instantaneous changes in features across consecutive observation windows based on a first-order time-series feature dataset to obtain a second-order time-series feature dataset specifically includes: Obtain the instantaneous change feature vector corresponding to each time window in the first-order time series feature dataset to form a time series feature set; Arrange the feature vectors of consecutive windows in chronological order to construct a continuous observation feature sequence; For each operating parameter in the continuous observation feature sequence, calculate the difference change between adjacent windows to form a second-order change sequence; Based on the second-order change sequence, the acceleration characteristics of each running parameter are obtained; Local pattern recognition is performed on the changing acceleration features to extract the extreme points and inflection points that appear within a continuous window, forming local jump features; In a continuous window, the number of continuous windows that keep the changing acceleration characteristics in the same direction is used as an indicator of trend stability. The acceleration characteristics, local jump characteristics, and trend stability indicators corresponding to each operating parameter are concatenated according to the dimensions of the operating parameters to form a second-order time series feature vector. The second-order time series feature vector is normalized to obtain the standardized second-order time series feature vector; Obtain the abrupt change amplitude of vectors in the standardized second-order time series feature vectors, and retain the feature vectors that exceed the preset threshold to form a key second-order time series feature subset; A subset of key second-order time-series features for each runtime parameter is collected to construct a second-order time-series feature dataset.

[0008] In an optional embodiment, the step of extracting second-order time-series features corresponding to temperature data, current data, and equivalent series resistance data within the current time window based on the second-order time-series feature dataset, and identifying abnormal change features, specifically includes: Based on the second-order time series feature dataset, the key second-order feature vectors corresponding to temperature, current and equivalent series resistance within the current time window are obtained. Extract the acceleration features, local jump features, and trend stability indices corresponding to each operational parameter in the key second-order feature vectors respectively; By pre-setting a historical time span, the historical distribution characteristics of the change acceleration features and local jump features corresponding to each operating parameter are statistically analyzed to obtain the parameter historical distribution model. The acceleration features and local jump features corresponding to each running parameter within the current time window are sorted and compared with similar features within a preset historical time span to determine the relative position of the current feature in the historical distribution model. Based on relative position, the deviation sort values ​​of each operating parameter within the current time window are obtained; Based on the trend stability index and deviation ranking value, the abnormal evolution characterization quantity is obtained; Based on the distribution characteristics of the abnormal evolution characteristics of each operating parameter within the current time window, candidate abnormal states are determined; Based on the abnormal candidate states, the corresponding second-order time series feature vectors are determined as abnormal change features; The formula for calculating the abnormal evolution characteristic quantity is as follows:

[0009] In the formula, For the first Each running parameter in the time window The anomalous evolutionary characterization quantity, For the first Each running parameter in the time window Deviation from sort value, For the first Each running parameter in the time window Trend stability indicators This represents the number of consecutive observation windows. Index for the current time window, For the first Each running parameter in the historical window Trend stability indicators.

[0010] In an optional embodiment, the step of segmenting and summarizing the abnormal change features along the time dimension based on the abnormal change features, and counting the occurrence frequency and duration of the abnormal change features within each segment to obtain the implosion risk feature dataset, specifically includes: Obtain the abnormal change characteristics of the markers within the current time window and the consecutive time windows before and after it, and use the current time window and the consecutive time windows before and after it as the candidate range; Based on the abnormal change characteristics of each time window within the candidate range, the distribution of the abnormal change characteristics of each time window within the candidate range is determined. Based on the distribution of abnormal change characteristics within the candidate range, the candidate range is divided into several adaptive time periods, forming a set of adaptive time periods; For each adaptive time period, count the number of times the abnormal features appear within that time period. For the abnormal change characteristics within each adaptive time period, the duration of the abnormal characteristics within that time period is calculated. Within each adaptive time period, the abnormal change characteristics of each time window are divided into columns by running parameters and arranged into rows by time order to form a multi-parameter abnormal feature matrix within the time period; Based on the multi-parameter abnormal feature matrix within a time period, the simultaneous and overlapping occurrence patterns of abnormal features of each operating parameter within the same time period are extracted to form multi-parameter joint feature information. At the same time, the abnormal change features are arranged in chronological order to form a time series matrix. For the abnormal change feature sequence, the difference between the abnormal change feature vectors corresponding to adjacent time windows is obtained to form time evolution information; Based on time evolution information, the number of times each operating parameter changes abnormally and the number of times it decreases continuously within each time period are counted to obtain the abnormal evolution characterization quantity and form the abnormal evolution trend characteristics of each time period. Based on the frequency of occurrence, duration, multi-parameter joint characteristics, and abnormal evolution trend characteristics of abnormal change features in each time period, an enhanced implosion risk characterization vector is constructed for each time period to characterize the implosion risk characteristics of that time period. By aggregating the implosion risk characteristics of all time periods, a dataset of implosion risk characteristics for the current observation period is obtained.

[0011] In an optional embodiment, the step of classifying the implosion risk of each time window into different risk levels based on the implosion risk feature dataset, and classifying and sorting products of different risk levels to obtain implosion risk warning results, specifically includes: Based on the enhanced implosion risk characterization vectors corresponding to each time window within the current observation period in the implosion risk characteristic dataset, the relative distribution statistics of the occurrence frequency, duration of anomalies, and characterization of anomalies are performed, and they are divided into low, medium, and high intervals. For each time window, if the number of occurrences of abnormal change features, the duration of abnormality, and the amount of abnormal evolution characterization all fall into the low range, and the enhanced implosion risk characterization vector corresponding to that time window does not contain multi-parameter joint feature information, then the implosion risk corresponding to that time window is judged as safe. If the safety conditions are not met, and at least one of the occurrence frequency, duration or evolution characteristics of abnormal changes falls into the middle range, or if the enhanced implosion risk characterization vector corresponding to the time window contains multi-parameter joint feature information, then the implosion risk corresponding to the time window is judged as warning type. If neither the safety type nor the warning type conditions are met, that is, if at least one of the abnormal change characteristics, the abnormal duration, or the abnormal evolution characterization quantity falls into the high range, then the implosion risk corresponding to the time window is judged as critical type. In this case, the multi-parameter joint characteristic information is not a necessary condition. The time window deemed safe is marked as low-risk, and routine operations and standard testing strategies are implemented. The time window that is identified as a warning is marked as a medium-risk level, and key monitoring and early maintenance strategies are implemented. The time window that is determined to be critical is marked as high-risk, and emergency measures such as immediate handling, isolation, and shutdown are arranged. Arrange the implosion risk levels corresponding to each time window in chronological order to form an implosion risk warning sequence; Based on the implosion risk warning sequence, an implosion risk warning result is generated, which includes the risk level, time information, and corresponding processing operation for each time window.

[0012] Furthermore, a capacitor implosion risk analysis system based on time-series characteristics is proposed to implement the risk analysis method described above, including: The data acquisition module is used to acquire current, temperature and equivalent series resistance data of the capacitor during operation, and generate a time series dataset of operating parameters. The feature extraction module is used to process the time series data of the running parameters within each time window, extract the change rate features, amplitude features, dynamic time warping distance features and Mann-Kendall trend test features, and construct a first-order time series feature dataset. The second-order feature analysis module is used to perform continuous window difference calculation, acceleration extraction, local jump identification and trend stability analysis on the first-order time series feature dataset to obtain the second-order time series feature dataset. Anomaly identification module, which is used to extract key feature vectors based on second-order time series feature dataset, calculate deviation ranking value and abnormal evolution characterization quantity by combining historical distribution model, and identify abnormal change characteristics; The risk characterization module is used to segment and summarize the abnormal change characteristics in the time dimension, count the number of occurrences, duration and multi-parameter joint features, form an enhanced implosion risk characterization vector, and construct an implosion risk feature dataset. The risk warning module is used to classify the implosion risk of each time window into different levels and generate implosion risk warning results, including the risk level, time information and processing operations of the corresponding time window.

[0013] In an optional embodiment, the feature extraction module includes: A first-order feature calculation unit is used to calculate the rate of change and amplitude of the operating parameters within each time window, and to extract the first-order instantaneous change feature vectors of temperature, current and equivalent series resistance. A sequence distance calculation unit is used to calculate the temperature and current sequence distance based on a dynamic time warping method, and generate a first instantaneous change feature. A trend significance analysis unit is used to perform Mann-Kendall trend test on the equivalent series resistance time series data to obtain a second instantaneous change feature; The feature vector combination unit is used to combine the rate of change, magnitude features, and first and second instantaneous change features to form a first-order feature vector for each time window and to collect and construct a first-order time series feature dataset.

[0014] In an optional embodiment, the second-order feature analysis module includes: The difference calculation unit is used to calculate the first-order feature vector difference of a continuous window to form a second-order change sequence; An acceleration feature unit is used to extract the change acceleration features of each operating parameter. A local jump identification unit is used to identify extreme points and inflection points to form local jump features; A trend stability analysis unit is used to count the number of continuous windows in the same direction of the acceleration change feature to obtain a trend stability index. The second-order feature vector generation unit is used to concatenate the change acceleration features, local jump features and trend stability indicators according to the running parameter dimension to generate a standardized second-order time series feature vector, and select key feature subsets to construct a second-order time series feature dataset.

[0015] In an optional embodiment, the risk warning module includes: The risk statistics unit is used to perform relative distribution statistics on the occurrence frequency, duration and abnormal evolution characteristics of abnormal change features based on the enhanced implosion risk characterization vector, and to divide the region into low interval, medium interval and high interval. The risk determination unit is used to determine whether each time window is safe, warning, or critical, based on whether the indicator falls within the interval and the joint feature information of multiple parameters. A risk level marking unit is used to mark safe types as low risk, warning types as medium risk, and critical types as high risk, and to arrange corresponding handling strategies. The early warning generation unit is used to arrange the risk levels of each time window in chronological order, generate an implosion risk early warning sequence and a final early warning result, including time information, risk level and processing operation.

[0016] Compared with the prior art, the beneficial effects of the present invention are: This paper proposes a method and system for analyzing capacitor implosion risk based on time-series characteristics. It constructs a time-series dataset of operating parameters by collecting multi-parameter data such as current, temperature, and equivalent series resistance during capacitor operation. Within each time window, first-order instantaneous change features are extracted, and second-order time-series features are obtained based on continuous window analysis, enabling accurate identification of abnormal change features. Abnormal change features are segmented and summarized along the time dimension, and the frequency, duration, and multi-parameter joint patterns are statistically analyzed to form an enhanced implosion risk characterization vector. Based on the implosion risk features, each time window is classified into low, medium, and high-level risks, enabling risk assessment and early warning. This invention can achieve continuous risk analysis and graded early warning across multiple parameters and time dimensions, significantly improving the safety and reliability of capacitor operation, reducing the risk of potential implosion accidents, and improving risk management efficiency and automation. Attached Figure Description

[0017] Figure 1 This is a flowchart of a capacitor implosion risk analysis method based on time-series characteristics proposed in this invention; Figure 2 This is a flowchart of the first-order temporal feature extraction process in this invention; Figure 3 This is a flowchart of the second-order time series feature analysis and anomaly identification process in this invention; Figure 4 This is a system framework diagram of a capacitor implosion risk analysis system based on time-series characteristics proposed in this invention. Detailed Implementation

[0018] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0019] Reference Figure 1 - Figure 4 As shown in the figure, an embodiment of the present invention provides a capacitor implosion risk analysis method based on time-series characteristics, comprising: Collect operating parameter data during capacitor operation and construct a time series dataset of operating parameters. The operating parameter data includes current data, temperature data, and equivalent series resistance data. Based on the time series dataset of running parameters, the instantaneous change features corresponding to each running parameter are extracted in each time window to obtain the first-order time series feature dataset. Based on the first-order time series feature dataset, the variation characteristics of instantaneous change features between continuous observation windows are analyzed to obtain the second-order time series feature dataset. Based on the second-order time series feature dataset, the second-order time series features corresponding to the temperature data, current data and equivalent series resistance data within the current time window are extracted to identify abnormal change features. Based on the abnormal change characteristics, the abnormal change characteristics are segmented and summarized in the time dimension, and the occurrence frequency and duration of the abnormal change characteristics in each segment are counted to obtain the implosion risk characteristic dataset. Based on the implosion risk feature dataset, the implosion risk of each time window is divided into different risk levels, and products of different risk levels are classified and sorted to obtain implosion risk warning results.

[0020] Furthermore, based on the runtime parameter time series dataset, the instantaneous change features corresponding to each runtime parameter are extracted within each time window to obtain a first-order time series feature dataset, specifically including: Obtain the current data, temperature data, and equivalent series resistance data corresponding to each time window in the runtime parameter time series dataset to form a window runtime parameter dataset; Time alignment processing is performed on the runtime parameter data in the window runtime parameter dataset to obtain runtime parameter time series data under a unified time scale, and runtime parameter time series data is constructed. Based on the time series data of the operating parameters, the changes in the operating parameters between adjacent sampling times are obtained, and a sequence of changes in the operating parameters is constructed. Acquire historical operational parameter data and calculate the frequency and average magnitude of abnormal changes at each time point to obtain quantitative indicators for weighting. The weighted change sequence is obtained by weighting the sequence of changes in operating parameters within the current time window based on quantitative indicators. Specifically, after constructing the sequence of operational parameter changes, historical operational parameter data is introduced to characterize the significance of these changes. To this end, for each operational parameter, the change between adjacent sampling times is first calculated in the historical operational data. Using the normal fluctuation range formed by the parameter during long-term operation as a reference, changes significantly exceeding this range are marked as abnormal changes. Based on this, for each time position or time period, the number of times the operational parameter is marked as an abnormal change in the historical data is counted to reflect the frequency of abnormal changes. Simultaneously, the magnitudes of all abnormal changes within that time position or time period are summarized, and their average level is calculated to reflect the typical intensity of abnormal changes. Subsequently, the number of occurrences of abnormal changes and the average magnitude of abnormal changes are combined to construct a quantitative index characterizing the sensitivity to anomalies at that time position. The number of occurrences of abnormal changes reflects the probability level of anomalies at that time position, while the average magnitude of abnormal changes reflects the severity of changes when anomalies occur. When a given time point exhibits a high frequency of anomalies and a generally large magnitude of anomaly changes, the corresponding quantitative indicator value is higher; conversely, when the frequency of anomalies is low and the magnitude of changes is small, the corresponding quantitative indicator value is lower. For example, in historical data, using 5-minute intervals as the statistical unit, if 10 anomalies occur within a given time point, and the average magnitude of these anomalies is 0.15, then the frequency of anomalies at that time point is recorded as 10, and the average magnitude of the anomalies is recorded as 0.15. Combining these two values ​​yields a quantitative indicator reflecting the anomaly sensitivity at that time point, which can be represented by a value of 1.5. Conversely, if only 2 anomalies occur within another time point, and the average magnitude of the anomalies is 0.05, the corresponding quantitative indicator value is 0.1. In this way, the frequency and severity of anomalies at different time points are uniformly mapped to the same numerical scale.

[0021] During the analysis within the current time window, each change in the sequence of operational parameter changes is adjusted based on its corresponding time position's quantification index. When the quantification index for a change's corresponding time position is large, the change is considered more likely to be related to an abnormal state, and the original change is amplified. Conversely, when the quantification index is small, the change is relatively compressed. For example, if a change is 0.08 and its corresponding time position's quantification index is 1.5, the weighted change is amplified to 0.12; while if another change is also 0.08 but its corresponding time position's quantification index is 0.1, the weighted change is only 0.01. Through this processing, the original sequence of operational parameter changes is transformed into a weighted sequence reflecting historical anomaly characteristics, providing a foundation for the subsequent extraction of change rate and magnitude characteristics.

[0022] Based on the weighted change sequence, the rate of change and magnitude of change of each operating parameter within the current time window are extracted. Within each time window, temperature time series data and current time series data are extracted from the running parameter window dataset, and the distance between the two sets of sequences is calculated according to the dynamic time warping method, which is used as the first instantaneous change feature. Within each time window, the equivalent series resistance time series data is extracted from the running parameter window dataset and subjected to Mann-Kendall trend test to obtain the statistical significance index value, which serves as the second instantaneous change feature. The rate of change and magnitude characteristics are combined with the first instantaneous change characteristics and the second instantaneous change characteristics to form the instantaneous change characteristic vector for each time window; The instantaneous change feature vectors of each time window are collected to construct the final first-order time series feature dataset.

[0023] Specifically, after obtaining the weighted change sequence, the variation characteristics of each operating parameter within the current time window are characterized. First, based on the weighted change sequence, the changes of each operating parameter within the current time window are statistically analyzed. The average level of the weighted change within the time window is used to characterize the rate of change of the operating parameter within that time window. Simultaneously, the difference between the maximum and minimum values ​​of the weighted change within the time window is used to characterize the amplitude of the change of the operating parameter within that time window, thus obtaining the rate of change and amplitude characteristics reflecting the speed and intensity of change. Building upon this, to analyze the consistency relationship between different operating parameters during the time evolution process, temperature and current time series data are extracted separately within each time window. Dynamic time warping methods are employed, such as nonlinear time alignment calculations to accumulate distance between the two sets of sequences, eliminating the influence of sampling start point offset and local time scaling, thereby obtaining the similarity between the temperature and current sequences during the time evolution process. When the cumulative distance is small, such as 0.05, it indicates that the two trends are highly consistent; when the cumulative distance is large, such as 0.2, it indicates that there is a significant difference in the rhythm or magnitude of change between the two, and this cumulative distance is taken as the first instantaneous change feature.

[0024] Simultaneously, time-series data of equivalent series resistance are extracted within the same time window. The Mann-Kendall trend test method is used, for example, to calculate a trend significance index by statistically analyzing the magnitude relationship between sampling points in the time series. This index reflects the monotonically increasing or decreasing trend of the equivalent series resistance within the current time window. When the trend significance index value is close to 0, such as 0.01, it indicates that the sequence change is close to random fluctuation with no obvious trend; when the trend significance index value is large, such as 2.3, it indicates that the equivalent series resistance shows a significant increasing trend; when the trend significance index value is negative, such as -1.8, it indicates a significant decreasing trend. This index value serves as the second instantaneous change feature in the subsequent feature vector construction. Subsequently, the change rate characteristics and change amplitude characteristics corresponding to each operating parameter within the current time window are combined with the sequence cumulative distance characteristics obtained based on the dynamic time warping method and the trend significance characteristics obtained based on the Mann-Kendall trend test to form the instantaneous change feature vector for the current time window. By collecting the instantaneous change feature vectors corresponding to each time window in chronological order, a first-order time-series feature dataset covering the entire observation period is finally constructed, providing a foundation for the analysis of the change characteristics between subsequent continuous observation windows.

[0025] Furthermore, based on the first-order time-series feature dataset, the variation characteristics of instantaneous changes between continuous observation windows are analyzed to obtain the second-order time-series feature dataset, which specifically includes: Obtain the instantaneous change feature vector corresponding to each time window in the first-order time series feature dataset to form a time series feature set; Arrange the feature vectors of consecutive windows in chronological order to construct a continuous observation feature sequence; For each operating parameter in the continuous observation feature sequence, calculate the difference change between adjacent windows to form a second-order change sequence; Based on the second-order change sequence, the acceleration characteristics of each running parameter are obtained; Specifically, firstly, the instantaneous change feature vectors corresponding to each time window are obtained from the first-order time-series feature dataset and arranged in chronological order to form a continuous time-series feature set, which reflects the temporal evolution characteristics of each operating parameter within the observation period. Then, the feature vectors of the continuous windows are aligned sequentially in chronological order to construct a continuous observation feature sequence, enabling analysis of the changing trends of each operating parameter in the time dimension. For each operating parameter in the continuous observation feature sequence, the change difference component is calculated between adjacent time windows. For example, for a certain operating parameter, if the change in window t is 0.12 and the change in window t+1 is 0.15, then the difference change is 0.03. This method yields a second-order change sequence covering the entire observation period. The second-order change sequence reflects the change rate of the operating parameter between continuous windows, thus revealing potential anomalous evolutionary behavior. Based on the second-order change sequence, the change acceleration feature of each operating parameter is further calculated to measure the degree of acceleration or deceleration of the change rate within the continuous time window. Acceleration can be positive or negative. A positive value indicates an increasing rate of change. For example, if the differential changes of a certain operating parameter over three consecutive windows are 0.02, 0.03, and 0.05, it means the change is accelerating, and the acceleration is positive. A negative value indicates a decreasing rate of change. For example, differential changes of 0.05, 0.03, and 0.01 indicate a gradual slowdown, and the acceleration is negative. In this way, the acceleration characteristics of each operating parameter within each time window are extracted, while retaining its positive or negative information, providing a foundation for subsequent local jump identification, trend stability analysis, and anomaly evolution characterization.

[0026] Local pattern recognition is performed on the changing acceleration features to extract the extreme points and inflection points that appear within a continuous window, forming local jump features; In a continuous window, the number of continuous windows that keep the changing acceleration characteristics in the same direction is used as an indicator of trend stability. The acceleration characteristics, local jump characteristics, and trend stability indicators corresponding to each operating parameter are concatenated according to the dimensions of the operating parameters to form a second-order time series feature vector. The second-order time series feature vector is normalized to obtain the standardized second-order time series feature vector; Obtain the abrupt change amplitude of vectors in the standardized second-order time series feature vectors, and retain the feature vectors that exceed the preset threshold to form a key second-order time series feature subset; A subset of key second-order time-series features for each runtime parameter is collected to construct a second-order time-series feature dataset.

[0027] Understandably, after obtaining the acceleration characteristics of each operating parameter, local pattern recognition is performed to extract extreme points and inflection points within a continuous window, thus forming local jump features. Specifically, by observing the acceleration sequence, when the acceleration suddenly changes from a positive value to a negative value or from a negative value to a positive value, that position is determined to be an inflection point; when the acceleration reaches its maximum or minimum value within a certain window range, it is determined to be a local extreme point. For example, for the current parameter, the acceleration changes in five consecutive windows are 0.02, 0.05, 0.08, 0.04, and 0.01, where 0.08 is a local extreme value and the direction of change slows down from increasing, then the inflection point corresponding to this window can be recorded as a local jump feature. Simultaneously, the number of consecutive windows where the acceleration characteristics remain in the same direction is counted within the continuous window; this number is used as a trend stability indicator to reflect the persistence of the rate of change trend. For example, if the acceleration of the temperature parameter is 0.01, 0.02, 0.03, and 0.04 in a continuous window, and the number of consecutive positive windows is 4, then the trend stability index is 4. If the acceleration subsequently becomes negative, the count is reset. The larger the trend stability index, the more stable the trend of the parameter's change during that time period. Subsequently, the acceleration feature, local jump feature, and trend stability index corresponding to each operating parameter are concatenated according to the parameter dimension to form a second-order time-series feature vector for each time window. The vector is then normalized to ensure that the feature values ​​of different parameters are on the same order of magnitude, resulting in a standardized second-order time-series feature vector. For example, the acceleration range is normalized to [-1, 1], local jump markers are represented by 0 or 1, and the trend stability index is normalized by dividing by the maximum possible number of windows. Next, the jump amplitude of each vector in the standardized second-order time-series feature vector is filtered, and feature vectors exceeding a preset threshold are retained to form a key second-order time-series feature subset. For example, when the jump amplitude threshold is set to 0.5, if the normalized acceleration jump amplitude of a certain window feature vector is 0.6, then the vector is retained; if the jump amplitude is 0.3, it is discarded. Finally, the key second-order time-series feature subsets of each running parameter are collected to construct a second-order time-series feature dataset covering all parameters and time windows, providing a foundation for subsequent anomaly identification and implosion risk characterization.

[0028] Furthermore, based on the second-order time-series feature dataset, second-order time-series features corresponding to temperature data, current data, and equivalent series resistance data within the current time window are extracted to identify anomalous change features, specifically including: Based on the second-order time series feature dataset, the key second-order feature vectors corresponding to temperature, current and equivalent series resistance within the current time window are obtained. Extract the acceleration features, local jump features, and trend stability indices corresponding to each operational parameter in the key second-order feature vectors respectively; By pre-setting a historical time span, the historical distribution characteristics of the change acceleration features and local jump features corresponding to each operating parameter are statistically analyzed to obtain the parameter historical distribution model. The acceleration features and local jump features corresponding to each running parameter within the current time window are sorted and compared with similar features within a preset historical time span to determine the relative position of the current feature in the historical distribution model. Specifically, after obtaining the second-order time-series feature dataset, the first step is to extract key second-order feature vectors corresponding to temperature, current, and equivalent series resistance for the current time window. These vectors contain the acceleration features, local jump features, and trend stability indices for each operating parameter. Subsequently, the key second-order feature vectors are split according to the operating parameter dimension, and the acceleration features, local jump markers, and trend stability indices for each parameter are extracted separately. The acceleration features characterize the degree to which the rate of change of the operating parameter accelerates or decelerates; the local jump features identify local extrema or abrupt changes within the time window; and the trend stability indices reflect the duration or stability of the change trend within consecutive windows, used for subsequent anomaly evolution analysis. For example, for the current parameter, the current window acceleration feature is 0.04 (positive value, indicating an accelerated rate of change), the local jump marker is 1 (indicating a local extrema or inflection point in this window), and the trend stability indices are 5 (indicating that the change trend remains positive for 5 consecutive windows). To determine whether the features of the current window are abnormal, the distribution of historical data needs to be referenced. Based on a preset historical time span—the time range used to construct the historical distribution model of parameters—this can include fixed time lengths (e.g., the past 1 hour, past 6 hours, past 24 hours), a fixed number of continuous time windows (e.g., the first 50 time windows), or variable sliding windows. This ensures sufficient historical operational data is covered for statistical feature distribution, such as the second-order feature sequence of the same parameter over the past 24 hours. The distribution characteristics of acceleration and local jump features for each operational parameter are statistically analyzed, including mean, standard deviation, maximum, minimum, and frequency of occurrence, thus obtaining the historical distribution model of the parameters. For example, the mean acceleration feature of the temperature parameter over the past 24 hours is 0.01, the standard deviation is 0.02, and local jump features occur an average of 2 times per hour. This model can reflect the typical change behavior of the parameter within the historical time span, providing a reference for identifying anomalies.

[0029] Subsequently, the acceleration characteristics and local jump characteristics of each operating parameter within the current time window are ranked and compared with the corresponding characteristics in the historical distribution model to determine the relative position of the current feature in the historical distribution model. For example, if the current temperature acceleration characteristic is 0.05, and it is in the top 10% range in the historical distribution model, it indicates that the change is significantly higher than the historical normal level; if the current local jump characteristic is marked as 1, and this type of jump occurs only 0.5 times per hour on average in the historical distribution model, it indicates that the jump frequency is abnormal. Through this ranking and comparison, the degree to which the change characteristics of each operating parameter in the current time window deviate from the historical normal level can be quantified, providing a basis for subsequent anomaly evolution characterization and implosion risk assessment.

[0030] Based on relative position, the deviation sort values ​​of each operating parameter within the current time window are obtained; Based on the trend stability index and deviation ranking value, the abnormal evolution characterization quantity is obtained; Based on the distribution characteristics of the abnormal evolution characteristics of each operating parameter within the current time window, candidate abnormal states are determined; Based on the abnormal candidate states, the corresponding second-order time series feature vectors are determined as abnormal change features; Specifically, within the current time window, for each operating parameter's acceleration and local jump characteristics, the feature value sequence of that window is first extracted, and then compared with the corresponding feature sequence within the historical time span. The comparison method involves sorting the historical feature values ​​in ascending order and recording the position of the current window feature within the historical sequence, forming the feature's relative position in the historical distribution. This relative position reflects the current feature's percentage ranking in historical data, clearly showing the feature's strength or weakness within the historical sequence. In this way, the relative position information of each operating parameter within the current window can be obtained, forming a complete set of relative positions, providing a basis for subsequent calculations of deviation ranking values ​​and anomalous evolution characterization quantities. For example, in the past 50 time windows, the acceleration feature value of a certain temperature parameter, sorted from low to high, is ranked 40th in the current window, indicating that its acceleration is higher than most historical windows; simultaneously, the local jump characteristic of the current parameter is ranked 12th in the same window, indicating a relatively low jump amplitude, while the acceleration feature of the equivalent series resistance parameter is ranked 35th, indicating that its change level is at a medium to high level. This comprehensive sorting example can visually demonstrate the relative position of each running parameter in the historical distribution within the current time window, providing a basis for further quantifying the degree of anomalies.

[0031] After obtaining the deviation ranking values ​​of each operating parameter within the current time window, the abnormal evolution characterization quantity is first calculated in conjunction with the trend stability index. Specifically, for each operating parameter, the trend stability index for the current window is obtained, and the cumulative sum of trend stability over the past few consecutive windows is calculated. Then, the trend stability of the current window is divided by the cumulative value of the historical windows, and multiplied by the deviation ranking value to obtain the comprehensive abnormal evolution characterization quantity. This value reflects both the degree to which the current window deviates from the historical distribution and the prominence of the current trend relative to the most recent historical trend. For example, if the deviation ranking value of the current parameter in the current window is 0.85, the current trend stability index is 4, and the sum of trend stability over the past five windows is 10, then the abnormal evolution characterization quantity is 0.34, indicating that the current window shows a significant deviation but the trend is relatively stable.

[0032] Within the current time window, the abnormal evolution characteristics of each operating parameter are first statistically analyzed, including deviation ranking values ​​and trend stability indicators. Then, the distribution of these values ​​within the window is analyzed. The distribution characteristics mainly include two aspects: first, the concentration or dispersion of the degree of anomaly, i.e., whether all abnormal evolution characteristics of each parameter are in the high value range, or only a few parameters are abnormal; second, the consistency of the abnormal trend, i.e., whether the acceleration directions of multiple parameters remain in the same direction, thus determining whether a joint anomaly pattern exists. In single-parameter analysis, if the abnormal evolution characteristic of a certain operating parameter exceeds a preset threshold, the parameter is considered to have a significant deviation within the current window, and the window can be marked as a candidate anomaly state. In multi-parameter joint analysis, if the abnormal evolution characteristic of two or more parameters is simultaneously high, and the acceleration directions are consistent, it indicates that multiple parameters may have jointly deviated from their historical distribution. In this case, the window can be determined as a higher-priority candidate anomaly state. This approach considers not only individual parameter anomalies but also the identification of multi-parameter joint anomaly patterns. Ultimately, the second-order time-series feature vectors that meet the criteria for anomaly candidate states are identified as anomalous change features, providing a foundation for subsequent risk characterization. For example, in the current time window, the current parameter acceleration deviation ranking value is 0.85 (normalized to the 0-1 range, indicating a level higher than 85% of historical data), and the trend stability index is 4 (acceleration direction is consistent across four consecutive windows); the temperature parameter deviation ranking value is 0.65, and the trend stability index is 3; the equivalent series resistance deviation ranking value is 0.90, and the trend stability index is 5. Comprehensive analysis shows that the second-order features of these three operating parameters are all at a relatively high level in their historical distribution, and the direction of change in acceleration remains consistent within consecutive windows, indicating significant deviations and stable trends. Therefore, the second-order feature vector corresponding to this time window is identified as an anomalous change feature and recorded for subsequent implosion risk characterization.

[0033] The formula for calculating the anomalous evolution characterization quantity is as follows:

[0034] In the formula, For the first Each running parameter in the time window The anomalous evolutionary characterization quantity, For the first Each running parameter in the time window Deviation from sort value, For the first Each running parameter in the time window Trend stability indicators This represents the number of consecutive observation windows. Index for the current time window, For the first Each running parameter in the historical window Trend stability indicators.

[0035] Understandably, within the current time window, for each operating parameter, its ranking position deviating from the historical distribution is first obtained. This is then combined with the length of time the acceleration direction of the current window remains consistently consistent to calculate an anomaly evolution characterization. This characterization is designed considering two aspects: firstly, it reflects the anomaly intensity of the current window, i.e., the degree to which the parameter change is high in the historical sequence. The higher the deviation, the more significantly the current change is higher than in most historical cases, potentially indicating rapid or abrupt changes in internal current, temperature, or equivalent series resistance. Secondly, it incorporates the stability of the change trend. If the acceleration direction remains consistently in the same direction, it indicates that the anomaly change is not instantaneous noise but a continuous process with high reliability.

[0036] Furthermore, based on the abnormal change characteristics, the abnormal change characteristics are segmented and summarized along the time dimension, and the frequency and duration of the abnormal change characteristics within each segment are counted to obtain the implosion risk characteristic dataset, which specifically includes: Obtain the abnormal change characteristics of the markers within the current time window and the consecutive time windows before and after it, and use the current time window and the consecutive time windows before and after it as the candidate range; Based on the abnormal change characteristics of each time window within the candidate range, the distribution of the abnormal change characteristics of each time window within the candidate range is determined. Based on the distribution of abnormal change characteristics within the candidate range, the candidate range is divided into several adaptive time periods, forming a set of adaptive time periods; For each adaptive time period, count the number of times the abnormal features appear within that time period. For the abnormal change characteristics within each adaptive time period, the duration of the abnormal characteristics within that time period is calculated. Within each adaptive time period, the abnormal change characteristics of each time window are divided into columns by running parameters and arranged into rows by time order to form a multi-parameter abnormal feature matrix within the time period; Specifically, when analyzing the current time window, the abnormal change features marked within this window and several consecutive time windows before and after it are first used as candidate ranges to ensure that potentially continuing or spreading abnormal patterns are captured. Then, a distribution analysis is performed on the abnormal change features of each time window within the candidate range, including the distribution of abnormal features among various parameters and the degree of concentration or dispersion of abnormal features over time. Based on these distributions, the candidate range is divided into several adaptive time periods. The length of each time period can be automatically adjusted according to the concentration of abnormal features, forming an adaptive time period set to more precisely describe the evolution of abnormal changes over time. Specifically, during the division, the abnormal change features of each window within the candidate range are scanned in chronological order, and the number of consecutively occurring abnormal windows is counted. When the number of consecutive abnormal windows reaches a preset threshold (e.g., three consecutive windows showing at least one abnormal parameter), these windows are merged into the start and end interval of an adaptive time period. If a brief period of no abnormality occurs between consecutive abnormal windows (e.g., one window without an anomaly), it can be incorporated into the current time period according to the allowed interval rules. If consecutive empty windows exceed the threshold, they are used as the starting point of the next time period. This method allows the adaptive time period length to be automatically adjusted according to the anomaly distribution, making the anomaly features within the time period as concentrated as possible, which facilitates subsequent statistics and analysis.

[0037] Within each adaptive time period, the frequency of occurrence of abnormal change features is counted to quantify the frequency of abnormal events within that time period; simultaneously, the duration of abnormal features is counted to reflect the stability and potential impact of the abnormal state within that period. To visually demonstrate the correlation between multi-parameter anomalies within the time period, the abnormal change features of each time window are divided into columns according to the operating parameters and arranged into rows according to the time window order, constructing a multi-parameter anomaly feature matrix for the time period. Each row of the matrix corresponds to a time window, each column corresponds to an operating parameter, and each matrix element records the abnormal change feature state of that parameter within that time window, thus forming a matrix layout that visually reflects the simultaneous occurrence pattern of anomalies of various parameters. This reveals the simultaneous occurrence pattern and cross-influence of abnormal changes among different parameters. For example, within a candidate range, the abnormal change of the current parameter occurs 3 times within 5 consecutive windows, the temperature parameter occurs 4 times within the same range, and the equivalent series resistance occurs 2 times. Based on the continuous anomaly window merging rule, windows 1-4 are divided into the first adaptive time period, where anomaly features are concentrated and occur frequently. Window 5 has no anomalies, but anomalies occur in windows 6-7, so windows 6-7 are divided into the second time period, where anomaly features are more dispersed. By constructing a multi-parameter anomaly feature matrix, it can be intuitively observed that current and temperature parameter anomalies occur simultaneously in the first segment, while the anomalies in the second segment are mostly single-parameter events, providing a quantitative basis for subsequent implosion risk characterization.

[0038] Based on the multi-parameter abnormal feature matrix within a time period, the simultaneous and overlapping occurrence patterns of abnormal features of each operating parameter within the same time period are extracted to form multi-parameter joint feature information. At the same time, the abnormal change features are arranged in chronological order to form a time series matrix. For the abnormal change feature sequence, the difference between the abnormal change feature vectors corresponding to adjacent time windows is obtained to form time evolution information; Based on time evolution information, the number of times each operating parameter changes abnormally and the number of times it decreases continuously within each time period are counted to obtain the abnormal evolution characterization quantity and form the abnormal evolution trend characteristics of each time period. Based on the frequency of occurrence, duration, multi-parameter joint characteristics, and abnormal evolution trend characteristics of abnormal change features in each time period, an enhanced implosion risk characterization vector is constructed for each time period to characterize the implosion risk characteristics of that time period. By aggregating the implosion risk characteristics of all time periods, a dataset of implosion risk characteristics for the current observation period is obtained.

[0039] Specifically, within each adaptive time period, the simultaneous and overlapping patterns of abnormal features of each operating parameter are first analyzed based on a multi-parameter anomaly feature matrix. A simultaneous occurrence pattern refers to two or more parameters exhibiting anomalies simultaneously within the same time window. For example, if current and temperature both exhibit anomalies in the same window, it is recorded as "current-temperature simultaneous occurrence"; if three parameters exhibit anomalies simultaneously, it is recorded as "current-temperature-equivalent series resistance simultaneous occurrence". An overlapping occurrence pattern refers to different parameters exhibiting anomalies alternately in different time windows within the same time period. For example, if current exhibits anomalies in windows 1 and 3, and temperature exhibits anomalies in windows 2 and 4, it forms a current-temperature overlapping occurrence pattern. The frequency of simultaneous and overlapping occurrence patterns is statistically analyzed throughout the entire time period to generate multi-parameter joint feature information. Simultaneously, the anomaly change features within the time period are arranged chronologically to form a time series matrix, providing a foundation for subsequent trend analysis. Subsequently, the anomaly change feature vectors of adjacent time windows in the time series matrix are differentially calculated to obtain temporal evolution information, quantifying the magnitude and direction of anomaly feature changes over time. Based on this temporal evolution information, the number of consecutive increases and decreases in the abnormal changes of each operating parameter within each time period is further statistically analyzed to obtain anomaly evolution characterization quantities, reflecting the evolution trend and stability of the abnormal state within the time period. For example, if the current parameter has 4 consecutive windows of increase and 1 consecutive window of decrease within a time period, the abnormal evolution trend of the current parameter during that time period shows a significant upward trend; if the temperature parameter has 3 consecutive windows of increase and 2 consecutive windows of decrease, the abnormal evolution trend is relatively volatile.

[0040] By combining the frequency, duration, multi-parameter joint features, and anomaly evolution trends of abnormal characteristics within a time period, an enhanced implosion risk characterization vector is formed for that time period, quantifying the potential implosion risk level within that segment. Finally, the enhanced characterization vectors from all time periods are aggregated to obtain a complete implosion risk characteristic dataset for the current observation period, providing foundational data for subsequent risk level classification and early warning generation. For example, within a time period, current and temperature parameters simultaneously exhibit anomalies three times, and temperature and equivalent series resistance exhibit anomalies twice; the current parameter anomaly characteristics continuously increase in windows 1-4, while the temperature parameter anomaly characteristics fluctuate continuously in windows 2-5; combining the frequency and duration of occurrence, an enhanced implosion risk characterization vector for that time period can be generated. After aggregating the characterization vectors from all time periods, an implosion risk characteristic dataset for the entire observation period is formed.

[0041] Furthermore, based on the implosion risk characteristic dataset, the implosion risk for each time window is divided into different risk levels, and products of different risk levels are classified and sorted to obtain implosion risk warning results, specifically including: Based on the enhanced implosion risk characterization vectors corresponding to each time window within the current observation period in the implosion risk characteristic dataset, the relative distribution statistics of the occurrence frequency, duration of anomalies, and characterization of anomalies are performed, and they are divided into low, medium, and high intervals. For each time window, if the number of occurrences of abnormal change features, the duration of abnormality, and the amount of abnormal evolution characterization all fall into the low range, and the enhanced implosion risk characterization vector corresponding to that time window does not contain multi-parameter joint feature information, then the implosion risk corresponding to that time window is judged as safe. If the safety conditions are not met, and at least one of the occurrence frequency, duration or evolution characteristics of abnormal changes falls into the middle range, or if the enhanced implosion risk characterization vector corresponding to the time window contains multi-parameter joint feature information, then the implosion risk corresponding to the time window is judged as warning type. If neither the safety type nor the warning type conditions are met, that is, if at least one of the abnormal change characteristics, the abnormal duration, or the abnormal evolution characterization quantity falls into the high range, then the implosion risk corresponding to the time window is judged as critical type. In this case, the multi-parameter joint characteristic information is not a necessary condition. Understandably, within the current observation period, the enhanced implosion risk characterization vector for each time window is first statistically analyzed, calculating the frequency of occurrence, duration, and evolutionary characterization of anomalous changes. To visually reflect the deviation of each indicator, these statistical values ​​are compared with the distribution of the corresponding indicators throughout the entire observation period, dividing them into low, medium, and high intervals. The specific division can be determined by calculating the mean and standard deviation over the entire period; for example, an occurrence frequency below the mean is considered a low interval, above the mean plus one standard deviation is considered a high interval, and the middle portion is considered a medium interval. A similar method is used to divide the duration and evolutionary characterization of anomalous changes to ensure comparability of indicators across time windows. Subsequently, risk candidate analysis is performed for each time window, incorporating multi-parameter joint feature information. Multi-parameter joint features refer to the simultaneous occurrence of anomalous features of two or more operating parameters within the same time window, or the cross-correlation of anomalous features among different parameters. For example, if the second-order characteristics of current and temperature parameters simultaneously exhibit significant jumps within a given window, and the trends are consistent, then the window is considered to contain a multi-parameter joint anomaly pattern; if only a single parameter anomaly exists, it is not included in the joint feature calculation. In this way, the anomaly concentration and potential risk coupling of each time window can be quantified.

[0042] Based on the above analysis, the risk intensity of each window can be preliminarily classified: when the frequency, duration, and abnormal evolution characteristics all fall into the low range, and no multi-parameter joint features appear within the window, the window can be considered a low-intensity anomaly window and a safe candidate; if at least one of the indicators falls into the medium range, or there is a multi-parameter joint anomaly pattern, the anomaly intensity of the window is judged to be at a medium level and a warning candidate; if at least one of the frequency, duration, or abnormal evolution characteristics falls into the high range, the window anomaly intensity is extremely high and a critical candidate. In this way, a quantitative anomaly level indicator can be established for each time window, while providing a data foundation for the subsequent generation of implosion risk sequences and early warnings. For example, in the 7th window of the observation period, the current parameter shows 3 anomalies (low range), the temperature parameter duration is 5 minutes (medium range), the equivalent series resistance anomaly evolution characteristic is 0.42 (high range), and both current and temperature show anomalies simultaneously, then the window is marked as a warning candidate, reflecting the superposition effect of single indicators and multi-parameter joint anomalies.

[0043] The time window deemed safe is marked as low-risk, and routine operations and standard testing strategies are implemented. The time window that is identified as a warning is marked as a medium-risk level, and key monitoring and early maintenance strategies are implemented. The time window that is determined to be critical is marked as high-risk, and emergency measures such as immediate handling, isolation, and shutdown are arranged. Arrange the implosion risk levels corresponding to each time window in chronological order to form an implosion risk warning sequence; Based on the implosion risk warning sequence, an implosion risk warning result is generated, which includes the risk level, time information, and corresponding handling operations for each time window.

[0044] Specifically, after constructing the implosion risk characteristic dataset, the enhanced implosion risk characterization vector for each time window within the current observation period is analyzed. Combining the frequency of occurrence, duration, and relative distribution of anomalous evolution characterization quantities, the time windows are divided into low, medium, and high risk levels. For low-risk windows, the second-order acceleration changes of current, temperature, and equivalent series resistance all remain near their historical averages, with few local jump points and stable trends. No multi-parameter joint anomaly modes appear, therefore, the window is deemed safe. At this point, routine monitoring measures are implemented, such as recording current fluctuations, temperature rise, and ESR changes at a predetermined sampling frequency, and periodic temperature and current trend checks are performed to ensure stable capacitor operation.

[0045] For medium-risk windows, at least one parameter exhibits a moderate second-order acceleration fluctuation, or two parameters simultaneously show deviation trends, and multiple parameter joint anomaly patterns occasionally occur, which is classified as a warning type. In such cases, the sampling frequency of current and temperature during that time period can be increased, and load regulation or local cooling measures can be initiated to intervene in advance regarding potential temperature rises and current anomalies, preventing further risk escalation. For example, in a certain observation window, if the second-order acceleration of the temperature parameter reaches above the historical median, and the current parameter experiences a local jump, with a joint anomaly pattern showing simultaneous deviations in both parameters, it is marked as medium-risk, and the scheduling system automatically initiates an advance cooling strategy.

[0046] For high-risk windows, multiple parameters simultaneously exhibit high deviations in second-order acceleration, accompanied by continuous local jump points and a clear trend of stability. The second-order eigenvector shows a concentrated occurrence of multi-parameter joint anomaly patterns, classifying it as a critical situation. In this case, immediate emergency measures should be taken, such as stopping the charging and discharging operation of the capacitor unit, isolating the abnormal unit and triggering an alarm, and simultaneously activating the emergency cooling system for modules with rapidly rising temperatures. For example, if both current and ESR show high-amplitude accelerated changes within the 8th time window, and the temperature maintains an upward trend for three consecutive windows, with multiple parameter joint anomalies occurring simultaneously, this is marked as high-risk. The system should immediately disconnect the circuit and notify maintenance personnel for handling.

[0047] The risk levels of all time windows are arranged chronologically to form an implosion risk warning sequence, visually demonstrating the risk evolution of the capacitor within the observation period. The implosion risk warning results generated based on this sequence include the risk level, time information, and corresponding handling actions for each time window, providing clear guidance for operation and maintenance decisions and safety management. For example, the first three windows are low risk, requiring routine monitoring; windows 4 to 6 are medium risk, requiring early intervention; and windows 7 and 8 are high risk, requiring immediate isolation and emergency cooling. The entire sequence clearly shows the evolution trend of risk from low to high.

[0048] Furthermore, a capacitor implosion risk analysis system based on time-series characteristics is proposed to implement the risk analysis method described above, including: The data acquisition module is used to collect current, temperature and equivalent series resistance data of the capacitor during operation, and generate a time series dataset of operating parameters. The feature extraction module is used to process the time series data of the running parameters within each time window, extract the change rate features, amplitude features, dynamic time warping distance features and Mann-Kendall trend test features, and construct a first-order time series feature dataset. The second-order feature analysis module is used to perform continuous window difference calculation, acceleration extraction, local jump identification, and trend stability analysis on the first-order time series feature dataset to obtain the second-order time series feature dataset. The anomaly identification module is used to extract key feature vectors based on the second-order time series feature dataset, and combine them with the historical distribution model to calculate the deviation ranking value and the anomaly evolution characterization quantity to identify abnormal change characteristics. The risk characterization module is used to segment and summarize the abnormal change characteristics in the time dimension, count the number of occurrences, duration and multi-parameter joint features, form an enhanced implosion risk characterization vector, and construct an implosion risk feature dataset. The risk warning module is used to classify the implosion risk of each time window into different levels and generate implosion risk warning results, including the risk level, time information and processing operations for the corresponding time window.

[0049] Furthermore, the feature extraction module includes: The first-order feature calculation unit is used to calculate the rate of change and magnitude of the operating parameters within each time window, and extract the first-order instantaneous change feature vectors of temperature, current and equivalent series resistance. The sequence distance calculation unit is used to calculate the temperature and current sequence distance based on the dynamic time warping method, and generate the first instantaneous change feature. The trend significance analysis unit is used to perform Mann-Kendall trend test on the equivalent series resistance time series data to obtain the second instantaneous change feature; The feature vector combination unit is used to combine the rate of change, magnitude features, and first and second instantaneous change features to form a first-order feature vector for each time window and to collect and construct a first-order time series feature dataset.

[0050] Furthermore, the second-order feature analysis module includes: The difference calculation unit is used to calculate the first-order feature vector difference of a continuous window, forming a second-order transformation sequence; Acceleration feature unit, used to extract the acceleration features of changes in various operating parameters; The local jump identification unit is used to identify extreme points and inflection points, forming local jump features; The trend stability analysis unit is used to count the number of continuous windows in the same direction of the acceleration characteristic of change, and to obtain the trend stability index. The second-order feature vector generation unit is used to concatenate the change acceleration features, local jump features, and trend stability indicators according to the running parameter dimension to generate standardized second-order time series feature vectors, and select key feature subsets to construct a second-order time series feature dataset.

[0051] Furthermore, the risk warning module includes: The risk statistics unit is used to perform relative distribution statistics on the occurrence frequency, duration and abnormal evolution characteristics of abnormal change features based on the enhanced implosion risk characterization vector, and to divide the region into low, medium and high intervals. The risk assessment unit is used to determine whether each time window is safe, warning, or critical, based on whether the indicators fall within the range and the combined characteristic information of multiple parameters. The risk level marking unit is used to mark safe types as low risk, warning types as medium risk, and critical types as high risk, and to arrange corresponding handling strategies. The early warning generation unit is used to arrange the risk levels of each time window in chronological order, generate an implosion risk early warning sequence and the final early warning result, including time information, risk level and handling operations.

[0052] In summary, the advantages of this invention are as follows: by constructing a time-series feature set of capacitor operating parameters and extracting first-order instantaneous change features and second-order change features, the dynamic performance of the target capacitor under different operating conditions can be analyzed, revealing which abnormal changes may lead to implosion risk and determining whether preventive measures need to be taken to reduce the risk; at the same time, by segmenting and summarizing the abnormal change features and conducting joint analysis of multiple parameters, high-risk time windows can be identified more accurately, the implosion risk classification and early warning strategy can be optimized, safety accidents can be avoided, and the reliability of capacitor operation and the efficiency of risk management can be improved.

[0053] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A method for analyzing the risk of capacitor implosion based on time-series characteristics, characterized in that, include: Collect operating parameter data during capacitor operation and construct an operating parameter time series dataset, wherein the operating parameter data includes current data, temperature data and equivalent series resistance data; Based on the time series dataset of running parameters, the instantaneous change features corresponding to each running parameter are extracted in each time window to obtain the first-order time series feature dataset. Based on the first-order time series feature dataset, the variation characteristics of instantaneous change features between continuous observation windows are analyzed to obtain the second-order time series feature dataset. Based on the second-order time series feature dataset, the second-order time series features corresponding to the temperature data, current data and equivalent series resistance data within the current time window are extracted to identify abnormal change features. Based on the abnormal change characteristics, the abnormal change characteristics are segmented and summarized in the time dimension, and the occurrence frequency and duration of the abnormal change characteristics in each segment are counted to obtain the implosion risk characteristic dataset. Based on the implosion risk feature dataset, the implosion risk of each time window is divided into different risk levels, and products of different risk levels are classified and sorted to obtain implosion risk warning results.

2. The capacitor implosion risk analysis method based on time-series characteristics according to claim 1, characterized in that, The step involves extracting the instantaneous change features corresponding to each running parameter within each time window based on the time series dataset of running parameters, thereby obtaining a first-order time series feature dataset, specifically including: Obtain the current data, temperature data, and equivalent series resistance data corresponding to each time window in the runtime parameter time series dataset to form a window runtime parameter dataset; Time alignment processing is performed on the runtime parameter data in the window runtime parameter dataset to obtain runtime parameter time series data under a unified time scale, and runtime parameter time series data is constructed. Based on the time series data of the operating parameters, the changes in the operating parameters between adjacent sampling times are obtained, and a sequence of changes in the operating parameters is constructed. Acquire historical operational parameter data and calculate the frequency and average magnitude of abnormal changes at each time point to obtain quantitative indicators for weighting. The weighted change sequence is obtained by weighting the sequence of changes in operating parameters within the current time window based on quantitative indicators. Based on the weighted change sequence, the rate of change and magnitude of change of each operating parameter within the current time window are extracted. Within each time window, temperature time series data and current time series data are extracted from the running parameter window dataset, and the distance between the two sets of sequences is calculated according to the dynamic time warping method, which is used as the first instantaneous change feature. Within each time window, the equivalent series resistance time series data is extracted from the running parameter window dataset and subjected to Mann-Kendall trend test to obtain the statistical significance index value, which serves as the second instantaneous change feature. The rate of change and magnitude characteristics are combined with the first instantaneous change characteristics and the second instantaneous change characteristics to form the instantaneous change characteristic vector for each time window; The instantaneous change feature vectors of each time window are collected to construct the final first-order time series feature dataset.

3. The capacitor implosion risk analysis method based on time-series characteristics according to claim 2, characterized in that, The second-order time-series feature dataset is obtained by analyzing the variation characteristics of instantaneous changes between continuous observation windows based on the first-order time-series feature dataset, specifically including: Obtain the instantaneous change feature vector corresponding to each time window in the first-order time series feature dataset to form a time series feature set; Arrange the feature vectors of consecutive windows in chronological order to construct a continuous observation feature sequence; For each operating parameter in the continuous observation feature sequence, calculate the difference change between adjacent windows to form a second-order change sequence; Based on the second-order change sequence, the acceleration characteristics of each running parameter are obtained; Local pattern recognition is performed on the changing acceleration features to extract the extreme points and inflection points that appear within a continuous window, forming local jump features; In a continuous window, the number of continuous windows that keep the changing acceleration characteristics in the same direction is used as an indicator of trend stability. The acceleration characteristics, local jump characteristics, and trend stability indicators corresponding to each operating parameter are concatenated according to the dimensions of the operating parameters to form a second-order time series feature vector. The second-order time series feature vector is normalized to obtain the standardized second-order time series feature vector; Obtain the abrupt change amplitude of vectors in the standardized second-order time series feature vectors, and retain the feature vectors that exceed the preset threshold to form a key second-order time series feature subset; A subset of key second-order time-series features for each runtime parameter is collected to construct a second-order time-series feature dataset.

4. The capacitor implosion risk analysis method based on time-series characteristics according to claim 3, characterized in that, The second-order time-series feature dataset extracts second-order time-series features corresponding to temperature data, current data, and equivalent series resistance data within the current time window, and identifies abnormal change features, specifically including: Based on the second-order time series feature dataset, the key second-order feature vectors corresponding to temperature, current and equivalent series resistance within the current time window are obtained. Extract the acceleration features, local jump features, and trend stability indices corresponding to each operational parameter in the key second-order feature vectors respectively; By pre-setting a historical time span, the historical distribution characteristics of the change acceleration features and local jump features corresponding to each operating parameter are statistically analyzed to obtain the parameter historical distribution model. The acceleration features and local jump features corresponding to each running parameter within the current time window are sorted and compared with similar features within a preset historical time span to determine the relative position of the current feature in the historical distribution model. Based on relative position, the deviation sort values ​​of each operating parameter within the current time window are obtained; Based on the trend stability index and deviation ranking value, the abnormal evolution characterization quantity is obtained; Based on the distribution characteristics of the abnormal evolution characteristics of each operating parameter within the current time window, candidate abnormal states are determined; Based on the abnormal candidate states, the corresponding second-order time series feature vectors are determined as abnormal change features; The formula for calculating the abnormal evolution characteristic quantity is as follows: In the formula, For the first Each running parameter in the time window The anomalous evolutionary characterization quantity, For the first Each running parameter in the time window Deviation from sort value, For the first Each running parameter in the time window Trend stability indicators This represents the number of consecutive observation windows. Index for the current time window, For the first Each running parameter in the historical window Trend stability indicators.

5. The capacitor implosion risk analysis method based on time-series characteristics according to claim 4, characterized in that, The method involves segmenting and summarizing the abnormal change characteristics along the time dimension, and counting the frequency and duration of the abnormal change characteristics within each segment to obtain the implosion risk characteristic dataset, which specifically includes: Obtain the abnormal change characteristics of the markers within the current time window and the consecutive time windows before and after it, and use the current time window and the consecutive time windows before and after it as the candidate range; Based on the abnormal change characteristics of each time window within the candidate range, the distribution of the abnormal change characteristics of each time window within the candidate range is determined. Based on the distribution of abnormal change characteristics within the candidate range, the candidate range is divided into several adaptive time periods, forming a set of adaptive time periods; For each adaptive time period, count the number of times the abnormal features appear within that time period. For the abnormal change characteristics within each adaptive time period, the duration of the abnormal characteristics within that time period is calculated. Within each adaptive time period, the abnormal change characteristics of each time window are divided into columns by running parameters and arranged into rows by time order to form a multi-parameter abnormal feature matrix within the time period; Based on the multi-parameter abnormal feature matrix within a time period, the simultaneous and overlapping occurrence patterns of abnormal features of each operating parameter within the same time period are extracted to form multi-parameter joint feature information. At the same time, the abnormal change features are arranged in chronological order to form a time series matrix. For the abnormal change feature sequence, the difference between the abnormal change feature vectors corresponding to adjacent time windows is obtained to form time evolution information; Based on time evolution information, the number of times each operating parameter changes abnormally and the number of times it decreases continuously within each time period are counted to obtain the abnormal evolution characterization quantity and form the abnormal evolution trend characteristics of each time period. Based on the frequency of occurrence, duration, multi-parameter joint characteristics, and abnormal evolution trend characteristics of abnormal change features in each time period, an enhanced implosion risk characterization vector is constructed for each time period to characterize the implosion risk characteristics of that time period. By aggregating the implosion risk characteristics of all time periods, a dataset of implosion risk characteristics for the current observation period is obtained.

6. The capacitor implosion risk analysis method based on time-series characteristics according to claim 5, characterized in that, The aforementioned dataset based on implosion risk characteristics classifies the implosion risk of each time window into different risk levels, and then classifies and sorts products of different risk levels to obtain implosion risk warning results, specifically including: Based on the enhanced implosion risk characterization vectors corresponding to each time window within the current implosion risk characteristic dataset, the relative distribution statistics of the occurrence frequency, duration of anomalies, and characterization of anomalies are performed, and these are divided into low, medium, and high intervals. For each time window, if the number of occurrences of abnormal change features, the duration of abnormality, and the amount of abnormal evolution characterization all fall into the low range, and the enhanced implosion risk characterization vector corresponding to that time window does not contain multi-parameter joint feature information, then the implosion risk corresponding to that time window is judged as safe. If the safety conditions are not met, and at least one of the occurrence frequency, duration or evolution characteristics of abnormal changes falls into the middle range, or if the enhanced implosion risk characterization vector corresponding to the time window contains multi-parameter joint feature information, then the implosion risk corresponding to the time window is judged as warning type. If neither the safety type nor the warning type conditions are met, that is, if at least one of the abnormal change characteristics, the abnormal duration, or the abnormal evolution characterization quantity falls into the high range, then the implosion risk corresponding to the time window is judged as critical type. In this case, the multi-parameter joint characteristic information is not a necessary condition. The time window deemed safe is marked as low-risk, and routine operations and standard testing strategies are implemented. The time window that is identified as a warning is marked as a medium-risk level, and key monitoring and early maintenance strategies are implemented. The time window that is determined to be critical is marked as high-risk, and emergency measures such as immediate handling, isolation, and shutdown are arranged. Arrange the implosion risk levels corresponding to each time window in chronological order to form an implosion risk warning sequence; Based on the implosion risk warning sequence, an implosion risk warning result is generated, which includes the risk level, time information, and corresponding processing operation for each time window.

7. A capacitor implosion risk analysis system based on time-series characteristics, used to implement the risk analysis method as described in any one of claims 1-6, characterized in that, include: The data acquisition module is used to acquire current, temperature and equivalent series resistance data of the capacitor during operation, and generate a time series dataset of operating parameters. The feature extraction module is used to process the time series data of the running parameters within each time window, extract the change rate features, amplitude features, dynamic time warping distance features and Mann-Kendall trend test features, and construct a first-order time series feature dataset. The second-order feature analysis module is used to perform continuous window difference calculation, acceleration extraction, local jump identification and trend stability analysis on the first-order time series feature dataset to obtain the second-order time series feature dataset. Anomaly identification module, which is used to extract key feature vectors based on second-order time series feature dataset, calculate deviation ranking value and abnormal evolution characterization quantity by combining historical distribution model, and identify abnormal change characteristics; The risk characterization module is used to segment and summarize the abnormal change characteristics in the time dimension, count the number of occurrences, duration and multi-parameter joint features, form an enhanced implosion risk characterization vector, and construct an implosion risk feature dataset. The risk warning module is used to classify the implosion risk of each time window into different levels and generate implosion risk warning results, including the risk level, time information and processing operations of the corresponding time window.

8. The capacitor implosion risk analysis system based on time-series characteristics according to claim 7, characterized in that, The feature extraction module includes: A first-order feature calculation unit is used to calculate the rate of change and amplitude of the operating parameters within each time window, and to extract the first-order instantaneous change feature vectors of temperature, current and equivalent series resistance. A sequence distance calculation unit is used to calculate the temperature and current sequence distance based on a dynamic time warping method, and generate a first instantaneous change feature. A trend significance analysis unit is used to perform Mann-Kendall trend test on the equivalent series resistance time series data to obtain a second instantaneous change feature; The feature vector combination unit is used to combine the rate of change, magnitude features, and first and second instantaneous change features to form a first-order feature vector for each time window and to collect and construct a first-order time series feature dataset.

9. A capacitor implosion risk analysis system based on time-series characteristics according to claim 7, characterized in that, The second-order feature analysis module includes: The difference calculation unit is used to calculate the first-order feature vector difference of a continuous window to form a second-order change sequence; An acceleration feature unit is used to extract the change acceleration features of each operating parameter. A local jump identification unit is used to identify extreme points and inflection points to form local jump features; A trend stability analysis unit is used to count the number of continuous windows in the same direction of the acceleration change feature to obtain a trend stability index. The second-order feature vector generation unit is used to concatenate the change acceleration features, local jump features and trend stability indicators according to the running parameter dimension to generate a standardized second-order time series feature vector, and select key feature subsets to construct a second-order time series feature dataset.

10. A capacitor implosion risk analysis system based on time-series characteristics according to claim 7, characterized in that, The risk warning module includes: The risk statistics unit is used to perform relative distribution statistics on the occurrence frequency, duration and abnormal evolution characteristics of abnormal change features based on the enhanced implosion risk characterization vector, and to divide the region into low interval, medium interval and high interval. The risk determination unit is used to determine whether each time window is safe, warning, or critical, based on whether the indicator falls within the interval and the joint feature information of multiple parameters. A risk level marking unit is used to mark safe types as low risk, warning types as medium risk, and critical types as high risk, and to arrange corresponding handling strategies. The early warning generation unit is used to arrange the risk levels of each time window in chronological order, generate an implosion risk early warning sequence and a final early warning result, including time information, risk level and processing operation.

Images