Key load early degradation identification method and system based on transient phase mutation characteristics

Abstract

This application relates to a method and system for early degradation identification of critical loads based on transient phase change characteristics. The method includes: acquiring a discrete-time signal sequence corresponding to a transient phase change event, and denoising the discrete-time signal sequence to obtain a transient signal sequence; performing data reliability assessment based on the transient signal sequence to obtain a data reliability judgment result; extracting features from the transient signal sequence based on the data reliability judgment result to obtain an enhanced feature vector; inputting the enhanced feature vector into a health benchmark model to perform equipment health status assessment to obtain a health status score; mapping the health status score to a corresponding health level based on a level threshold, and generating a degradation identification result based on the health level. This method can capture phase change characteristics during transient processes, effectively improving the accuracy of early degradation identification.

Description

Technical Field

[0001] This invention belongs to the field of power supply protection, and in particular relates to a method and system for early degradation identification of critical loads based on transient phase change characteristics. Background Technology

[0002] With the development of electrical equipment condition monitoring technology, online monitoring technology based on electrical quantity analysis has emerged, characterized by its non-intrusiveness and strong real-time performance. This has led to the current mainstream monitoring methods based on steady-state amplitude or energy characteristics. Traditional technologies primarily rely on continuously collecting voltage and current signals from critical loads and calculating their effective values, peak values, harmonic content, or power consumption—amplitude or energy domain characteristics—to determine equipment failure through threshold comparison or trend analysis. However, in complex power supply scenarios such as mountainous highways, current monitoring methods or traditional approaches have significant problems. First, the length and variable impedance of power lines lead to severe signal attenuation and distortion, resulting in a low signal-to-noise ratio for steady-state characteristics. Second, the operating conditions of critical loads are highly dynamic due to traffic flow; normal large power fluctuations can completely mask subtle changes in electrical characteristics caused by early latent defects such as contact degradation and insulation aging, leading to insufficient sensitivity and delayed early warning in the monitoring system. Traditional amplitude-based monitoring methods struggle to identify early latent degradation states that show no significant abnormalities in electrical amplitude before a fault occurs. Summary of the Invention

[0003] Therefore, it is necessary to provide a method and system for early degradation identification of critical loads based on transient phase change characteristics, which can focus on the phase change characteristics in the transient process, in order to address the above-mentioned technical problems.

[0004] Firstly, this application provides a method for identifying early degradation of critical loads based on transient phase change characteristics, including:

[0005] Obtain the discrete-time signal sequence corresponding to the transient phase change event, and perform noise reduction processing on the discrete-time signal sequence to obtain the transient signal sequence;

[0006] Based on transient signal sequences, data credibility assessment is performed to obtain data credibility judgment results;

[0007] Based on the data reliability judgment results, feature extraction is performed on the transient signal sequence to obtain an enhanced feature vector;

[0008] The enhanced feature vectors are input into the health benchmark model to assess the health status of the equipment and obtain a health status score; the health benchmark model is trained using historical enhanced feature vectors.

[0009] Based on the level threshold, the health status score is mapped to the corresponding health level, and based on the health level, the degradation identification result is generated.

[0010] Furthermore, based on the data reliability assessment results, feature extraction is performed on the transient signal sequence to obtain an enhanced feature vector, including:

[0011] Based on the data reliability judgment results, Hilbert transform is performed on the transient signal sequence to obtain the analytic signal, and based on the analytic signal, the instantaneous values ​​of the local phase are extracted to obtain the instantaneous phase sequence;

[0012] Based on instantaneous phase sequences and transient signal sequences, time-frequency domain feature extraction is performed to obtain multidimensional feature vectors;

[0013] A multidimensional feature vector is added to the end of the historical feature queue to obtain the real-time feature queue. Based on the real-time feature queue, trend analysis is performed on each feature in the multidimensional feature vector to obtain a time trend index.

[0014] By integrating multidimensional feature vectors and time trend indicators, an enhanced feature vector is obtained.

[0015] Furthermore, based on the instantaneous phase sequence and transient signal sequence, time-frequency domain feature extraction is performed to obtain a multi-dimensional feature vector, including:

[0016] Calculate the difference between each adjacent point in the instantaneous phase sequence to obtain the instantaneous phase difference sequence, and extract the maximum absolute value from the instantaneous phase difference sequence to obtain the maximum absolute amplitude of the phase jump;

[0017] Traverse the instantaneous phase difference sequence and calculate the number of absolute values ​​greater than the difference threshold to obtain the significant phase jump number;

[0018] A short-time Fourier transform is performed on the transient signal sequence to obtain the frequency domain phase spectrum, and the group delay spectrum is calculated based on the frequency domain phase spectrum.

[0019] Calculate the difference between the maximum and minimum values ​​in the group delay spectrum to obtain the distribution range of the group delay spectrum;

[0020] By integrating the maximum absolute amplitude of phase jumps, the number of significant phase jumps, and the distribution range of the group delay spectrum, a multidimensional eigenvector is obtained.

[0021] Furthermore, based on the real-time feature queue, trend analysis is performed on each feature in the multi-dimensional feature vector to obtain time trend indicators, including:

[0022] Based on the real-time feature queue, with the feature value as the dependent variable and the corresponding transient phase change event number as the independent variable, a linear fit is performed to obtain a trend line; based on the trend line, the trend slope value is obtained.

[0023] Based on a sliding window and a real-time feature queue, the arithmetic mean of the features within the sliding window is calculated to obtain the moving average.

[0024] Based on a sliding window and a real-time feature queue, the percentage of each feature whose value is greater than a mild threshold is calculated to obtain the excess percentage.

[0025] By integrating the trend slope value, moving average, and the percentage of exceeding the standard, a time trend indicator is obtained.

[0026] Furthermore, based on the transient signal sequence, data reliability assessment is performed to obtain data reliability judgment results, including:

[0027] Based on the transient signal sequence, the variance of the background signal is calculated to obtain the noise power, and the variance of the transient signal sequence within the time window corresponding to the transient phase change event is calculated to obtain the total power;

[0028] The signal-to-noise ratio is calculated based on the noise power and the total power.

[0029] The formula for the signal-to-noise ratio is:

[0030]

[0031] Wherein, SNR is the signal-to-noise ratio. For total power, Noise power;

[0032] Based on the type of transient phase burst event, a standard transient waveform template is extracted, and the standard transient waveform template is compared with the transient signal sequence to obtain the distortion rate index;

[0033] Based on the quality threshold, the signal-to-noise ratio and distortion rate are compared with the corresponding quality threshold to obtain the data credibility judgment result.

[0034] Furthermore, after generating the degradation identification result based on the health level, it also includes:

[0035] Time series smoothing is performed on the sequence of health status scores to obtain a smoothed health score sequence;

[0036] Based on a short-term sliding window, the number of times the score in the smoothed health score sequence exceeds the warning threshold is counted to obtain the number of times the limit is exceeded, and a warning identifier is obtained based on the warning rules and the number of times the limit is exceeded.

[0037] The first and second differences of the smoothed health score series are calculated to obtain the trend severity index;

[0038] Based on the warning identifier and the trend severity indicator, the degradation identification results are adjusted to obtain the warning identification results, which are used to indicate the issuance of warnings.

[0039] Secondly, this application also provides a critical load early degradation identification system based on transient phase change characteristics, including:

[0040] The denoising module is used to acquire the discrete-time signal sequence corresponding to the transient phase change event, and to denoise the discrete-time signal sequence to obtain the transient signal sequence.

[0041] The evaluation module is used to evaluate the reliability of data based on transient signal sequences and obtain the data reliability judgment result.

[0042] The feature module is used to extract features from transient signal sequences based on data reliability judgment results, and obtain enhanced feature vectors.

[0043] The scoring module is used to input the enhanced feature vectors into the health benchmark model to assess the health status of the equipment and obtain a health status score; the health benchmark model is trained using historical enhanced feature vectors.

[0044] The identification module is used to map the health status score to the corresponding health level based on the level threshold, and generate the deterioration identification result based on the health level. The deterioration identification result is used to indicate the need for early warning.

[0045] Thirdly, this application also provides a computer device including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement any step of the method provided in the first aspect of this application.

[0046] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any step of the method provided in the first aspect of this application.

[0047] The aforementioned method and system for identifying early degradation of critical loads based on transient phase change characteristics acquires discrete-time signal sequences corresponding to transient phase change events, and denoises these sequences to obtain transient signal sequences. Based on these sequences, data reliability is assessed to obtain a data reliability judgment result. Based on this judgment result, features are extracted from the transient signal sequences to obtain enhanced feature vectors. These enhanced feature vectors are then input into a health benchmark model to assess equipment health status and obtain a health status score. The health benchmark model is trained using historical enhanced feature vectors. Based on a level threshold, the health status score is mapped to a corresponding health level, and based on this level, a degradation identification result is generated. This method can detect early physical connection degradation that is not yet reflected in the amplitude by capturing minute phase distortions in current or voltage signals during transient moments such as equipment start-up and shutdown, and load changes. It assesses the status from multiple physical perspectives, effectively improving the reliability and accuracy of degradation assessment. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a schematic diagram of the process of a critical load early degradation identification method based on transient phase change characteristics provided in an embodiment of the present invention;

[0050] Figure 2 This is a schematic diagram of the structure of a critical load early degradation identification system based on transient phase change characteristics, provided in an embodiment of the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0052] In one embodiment, such as Figure 1 As shown, a method for early degradation identification of critical loads based on transient phase change characteristics is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, and to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0053] Step 101: Obtain the discrete-time signal sequence of the corresponding transient phase change event, and perform noise reduction processing on the discrete-time signal sequence to obtain the transient signal sequence.

[0054] The discrete-time signal sequence corresponding to a transient phase change event refers to the raw digital signal collected from the sensors of the monitored equipment, which is time-aligned with a transient phase change event. A transient phase change event refers to a sudden, non-periodic change in the phase of voltage, current, or vibration signals during equipment operation, occurring within a very short time. It is usually a manifestation of some kind of fault or anomaly. The discrete-time signal sequence is a series of digital amplitude values ​​sampled at fixed time intervals. Denoising is a signal processing operation aimed at suppressing or eliminating background noise, interference, and other irrelevant components from the raw discrete-time signal sequence to highlight the signal components related to the event itself. The transient signal sequence refers to the discrete-time signal sequence after denoising, which retains the core signal characteristics related to the transient phase change event while significantly reducing the influence of noise, resulting in higher signal quality and providing a cleaner data foundation for subsequent analysis.

[0055] The terminal acquires the original digital signal sequence associated with a specific transient phase change event, performs denoising processing on the original sequence using techniques such as digital filters, and its core operation is to analyze the frequency components or time-frequency characteristics of the signal, identify and attenuate those frequency bands or components that belong to background noise, while retaining as many signal components as possible that represent the characteristics of the event itself, thereby improving the accuracy and reliability of feature extraction and state assessment in subsequent steps. Optionally, a power frequency notch filter is applied to remove fundamental power grid interference; a wavelet threshold denoising algorithm is used to suppress random noise; and phase calibration is performed to compensate for the inherent delay caused by the sensor and transmission channel.

[0056] Step 102: Based on the transient signal sequence, perform data credibility assessment to obtain the data credibility judgment result.

[0057] Specifically, data credibility assessment is an operation that analyzes and judges the data quality of transient signal sequences. It evaluates whether the sequence can truly and effectively reflect the characteristics of transient phase change events, rather than invalid data that is severely contaminated or distorted. The data credibility judgment result is an output containing judgment information, which can be a Boolean value, a confidence score, or a quality level label, used to guide feature extraction strategies.

[0058] The terminal calculates quality indicators by measuring the signal-to-noise ratio, distortion rate of the standard event waveform, and amplitude rationality. Then, it compares these indicators with preset quality thresholds and performs comprehensive logical judgments to check the reliability of the input data. This provides meta-information about the reliability of the data itself for subsequent analysis steps, preventing low-quality or invalid data from directly entering the core analysis process, thereby reducing the risk of false alarms from the source.

[0059] Step 103: Based on the data reliability judgment results, perform feature extraction on the transient signal sequence to obtain an enhanced feature vector.

[0060] Specifically, the goal of feature extraction is to calculate a set of numerical indicators from a transient signal sequence that can quantify the key characteristics of the signal. These indicators should effectively characterize the device status information related to the transient phase change event. An enhanced feature vector is a multi-dimensional numerical array, where each dimension represents a specific feature extracted from the signal. Enhancement occurs because the extraction process is guided or weighted by data reliability judgment results, making the features more reflective of the true patterns in reliable data.

[0061] Based on the data credibility judgment results, the terminal performs feature extraction operations on the transient signal sequence. This relies on signal processing and feature engineering methods, including calculating the signal's time-domain statistics, frequency-domain features, or time-frequency-domain features. The extraction process takes into account the data credibility judgment results. If the data credibility is low, features sensitive to noise are avoided, or the extracted feature values ​​are discounted. A continuous signal waveform is compressed and abstracted into a set of feature values ​​with clear physical or mathematical meaning, thereby converting the device status information into a standard input format that can be directly processed by downstream machine learning models.

[0062] Step 104: Input the enhanced feature vector into the health benchmark model to assess the health status of the equipment and obtain a health status score; the health benchmark model is trained using historical enhanced feature vectors.

[0063] Specifically, the health baseline model is a pre-trained mathematical model that learns feature patterns corresponding to healthy and unhealthy states of equipment from a large number of historical, labeled, enhanced feature vectors using machine learning methods. Equipment health status assessment is an operation that uses the health baseline model to quantitatively analyze the current state of the equipment. The health status score is a continuous numerical value or a probability value used to quantitatively represent the current relative health level of the equipment; a higher score generally indicates better health, while a lower score suggests deterioration.

[0064] The terminal takes the enhanced feature vector as input and feeds it into a pre-trained health baseline model. Internally, the model calculates and infers from the complex mapping relationship between learned features and health status, comparing and matching the current event signal represented by the feature vector with historical normal / abnormal patterns stored in the model's memory. This outputs an objective, quantitative health status score, transforming the abstract concept of device status into a monitorable and comparable numerical indicator. It can also calculate the deviation of each dimension indicator from the health baseline; when any dimension indicator exceeds an adaptive threshold, it is directly determined as early degradation of critical load.

[0065] Step 105: Based on the level threshold, map the health status score to the corresponding health level, and generate the deterioration identification result based on the health level.

[0066] The health status threshold is a set of predefined numerical boundaries used to divide health status intervals. Optionally, a score above 90 can be defined as healthy, 80-90 as attenuated, and below 80 as abnormal. The health status level is a discrete status label based on the health status score and the threshold, which is more intuitive than continuous scores and facilitates the development of different maintenance strategies. The degradation identification result is a clear conclusion generated based on the health status level, which can be an indicator or a brief description indicating whether equipment degradation has occurred and the severity of the degradation.

[0067] The terminal compares the health status score with a preset level threshold, determines which threshold range the score falls into, and maps it to the corresponding health level. Based on the health level, it generates the final degradation identification result. For example, if the level is warning, the identification result may be that degradation has been detected, and a level one warning is initiated. The quantified score and level are transformed into a clear and executable decision instruction, which directly serves the early warning system for predictive maintenance of equipment, completing the closed loop from data analysis to operation and maintenance actions.

[0068] This embodiment provides a method for identifying early degradation of critical loads based on transient phase change characteristics. The method acquires discrete-time signal sequences corresponding to transient phase change events and denoises these sequences to obtain transient signal sequences. Based on these sequences, data reliability is assessed to obtain a data reliability judgment result. Based on this judgment result, features are extracted from the transient signal sequences to obtain enhanced feature vectors. These enhanced feature vectors are then input into a health benchmark model to assess equipment health status and obtain a health status score. The health benchmark model is trained using historical enhanced feature vectors. Based on a level threshold, the health status score is mapped to a corresponding health level, and based on this level, a degradation identification result is generated. Through these methods, the method can capture minute phase distortions in current or voltage signals during transient events such as equipment start-up and shutdown, and load changes, revealing early physical connection degradation not yet reflected in amplitude. This multi-physical perspective assessment effectively improves the reliability and accuracy of degradation evaluation.

[0069] In one embodiment, based on the data reliability judgment result, feature extraction is performed on the transient signal sequence to obtain an enhanced feature vector, including:

[0070] Step 201: Based on the data reliability judgment result, perform Hilbert transform on the transient signal sequence to obtain the analytic signal, and extract the instantaneous value of the local phase based on the analytic signal to obtain the instantaneous phase sequence.

[0071] The Hilbert transform is a mathematical transformation used to process real-valued signals. For a discrete-time signal, its Hilbert transform result is a new real signal of the same length. This new signal is the convolution result of the original signal and a specific impulse response. Physically, it can be understood as a -90-degree phase shift of all frequency components of the original signal. An analytic signal is a complex signal composed of the original real signal as its real part and the Hilbert transform result of that signal as its imaginary part. It is an extended representation of a purely real signal in the complex domain. The instantaneous value of the local phase refers to the instantaneous state of an oscillating signal at a specific point in time, which can be described by a rotating vector. The angle between this vector and the positive real axis is the instantaneous value of the local phase at that point in time, a quantity that changes with time. The instantaneous phase sequence is a new sequence obtained by calculating the phase angle of the analytic signal point by point. Each element in this sequence corresponds to the instantaneous value of the local phase of the original signal at the corresponding sampling time.

[0072] The terminal takes a transient signal sequence s(t) as input. For each sampling point of this sequence, a digital filter with Hilbert transform is applied to generate a new sequence H(t) of the same length. H(t) is the Hilbert transform result of s(t). Taking the original signal sequence s(t) as the real part and its Hilbert transform result H(t) as the imaginary part, the two are combined to form a complex sequence Z(t) = s(t) + j*H(t). The complex sequence Z(t) is the analytic signal. For each complex point Z(t) of the analytic signal Z(t),... i ), calculate its phase angle φ(t) i The calculation uses the four-quadrant arctangent function to accurately obtain the angle between the complex point and the positive real axis in the complex plane. The range is usually between (-π, π] or (0, 2π] radians. The phase angles of all time points obtained are arranged into a new sequence in chronological order to obtain the instantaneous phase sequence.

[0073] Step 202: Based on the instantaneous phase sequence and transient signal sequence, perform time-frequency domain feature extraction to obtain a multi-dimensional feature vector.

[0074] Specifically, time-frequency domain feature extraction is a collection of mathematical operations designed to calculate numerical indicators from a signal that reflect both its frequency component characteristics and the time-varying patterns of these frequency components. A multidimensional feature vector is a one-dimensional array containing multiple numerical elements, where each element is a scalar value calculated from an instantaneous phase sequence or transient signal sequence using a specific formula, characterizing a certain aspect of the signal's time-frequency properties. Together, they constitute a digital description of a signal event.

[0075] The terminal calculates the instantaneous phase sequence and transient signal sequence according to a predefined set of mathematical formulas. It performs a first-order difference operation on the instantaneous phase sequence to obtain a difference sequence in which the phase changes over time. From this difference sequence, it can find the maximum value, count the number of points exceeding a certain threshold, etc., reflecting the severity and frequency of phase changes. It performs a short-time Fourier transform on the transient signal sequence to obtain its spectrum over time. From this spectrum, it can calculate indicators such as the distribution range of the group delay spectrum, reflecting the concentration and diffusion characteristics of signal energy in the time-frequency plane. All the calculated scalar values ​​representing different characteristics are collected and arranged in a fixed order to form an ordered numerical list, which is a multidimensional feature vector.

[0076] Step 203: Add a multi-dimensional feature vector to the end of the historical feature queue to obtain the real-time feature queue. Based on the real-time feature queue, perform trend analysis on each feature in the multi-dimensional feature vector to obtain the time trend index.

[0077] Specifically, the historical feature queue is a data structure of fixed or variable length, stored chronologically, containing multi-dimensional feature vectors corresponding to several past events. The real-time feature queue is the updated queue formed by adding the latest multi-dimensional feature vector to the end of the historical feature queue, representing a time-series window of the device's status characteristics up to the current moment. Trend analysis is an analytical method for time-series data, aiming to quantify the direction and pattern of change exhibited by the sequence through mathematical modeling or statistical calculations, such as linear growth, decay, or periodic fluctuations. Time trend indicators are quantitative results obtained after performing trend analysis on a time-arranged subsequence formed by a specific feature dimension in the real-time feature queue; the result is usually one or a set of scalar values.

[0078] The terminal appends the multidimensional feature vector to the end of the historical feature queue that stores past features, forming an updated real-time feature queue. It then performs trend analysis. For each feature dimension in the real-time feature queue, it arranges N feature values ​​in chronological order to form a single-feature time series. The terminal performs mathematical analysis on the series, such as linear fitting to obtain the trend slope, calculating the mean within the sliding window, or counting the proportion exceeding a certain threshold. This analysis is performed on each feature in the multidimensional feature vector to obtain a set of corresponding time trend indicators.

[0079] Step 204: Integrate the multidimensional feature vector and time trend indicators to obtain the enhanced feature vector.

[0080] Integration is a data merging operation that combines or connects two or more data sets according to their dimensions to form a new, higher-dimensional single data set. An enhanced feature vector is a completely new, expanded feature vector formed by integrating two data sets: a multi-dimensional feature vector and a time trend indicator.

[0081] The terminal connects the multidimensional feature vector and the time trend indicator. In terms of data structure, the new indicator representing the trend is added as a new feature dimension to the static multidimensional feature vector, thus forming a longer feature array.

[0082] This embodiment integrates static features describing the current attributes of an event with dynamic trend features describing the long-term evolution of the equipment into a unified data representation, providing the most comprehensive and information-rich input for downstream health assessment models, thereby significantly improving the accuracy and timeliness of equipment degradation early warning.

[0083] In one embodiment, time-frequency domain feature extraction is performed based on the instantaneous phase sequence and transient signal sequence to obtain a multi-dimensional feature vector, including:

[0084] Step 301: Calculate the difference between each adjacent point in the instantaneous phase sequence to obtain the instantaneous phase difference sequence, and extract the maximum absolute value from the instantaneous phase difference sequence to obtain the maximum absolute amplitude of the phase jump.

[0085] The instantaneous phase difference sequence is a new sequence obtained by subtracting the value of each sampling point in the instantaneous phase sequence from the value of its preceding sampling point, quantifying the change in instantaneous phase between consecutive time points. The maximum absolute amplitude of the phase jump is a scalar value obtained by calculating the absolute values ​​of all differences in the instantaneous phase difference sequence and finding the maximum value. It represents the maximum abrupt change in instantaneous phase between adjacent sampling points during this event.

[0086] The terminal denotes the input instantaneous phase sequence of length N as: Perform a first-order difference operation on the sequence, for each element from the 2nd point to the Nth point in the sequence. (i ranges from 2 to N), calculate its relationship with the previous point. The difference yields a new sequence of length N-1. That is, the instantaneous phase difference sequence. For each element in the instantaneous phase difference sequence, calculate its absolute value to obtain an absolute value sequence. Traverse the absolute value sequence and find the element with the largest value. The maximum value is the maximum absolute amplitude of the phase jump.

[0087] Step 302: Traverse the instantaneous phase difference sequence and calculate the number of absolute values ​​greater than the difference threshold to obtain the significant phase jump number.

[0088] Specifically, the differential threshold is a pre-set numerical threshold used to determine whether a phase change is significant. It is a positive constant value; only when the absolute value of the phase difference exceeds this threshold is a significant phase jump considered to have occurred. The significant phase jump number is a scalar characteristic value, a counting result, equal to the total number of differences in the instantaneous phase difference sequence whose absolute value exceeds the differential threshold. It characterizes the number of significant phase abrupt changes with amplitudes exceeding a certain level that occur within a signal segment.

[0089] The terminal sets a fixed difference threshold. Starting from the first difference value, it checks the absolute value of each difference in turn. For each instantaneous phase difference sequence value, it determines whether it is greater than the set threshold. If it is greater, it is considered that a significant phase jump has occurred at the corresponding time point, and a counter is incremented. After traversing the entire difference sequence, the final value of the counter is the number of significant phase jumps.

[0090] Step 303: Perform a short-time Fourier transform on the transient signal sequence to obtain the frequency domain phase spectrum, and calculate the group delay spectrum based on the frequency domain phase spectrum.

[0091] Specifically, the Short-Time Fourier Transform (SFT) is a time-frequency analysis method that divides a long time-domain signal into many small segments along the time axis using a sliding window. Each segment is then subjected to a Fourier Transform, yielding a two-dimensional representation of the signal's frequency components over time. The frequency-domain phase spectrum refers to the complex sequence resulting from the Fourier Transform. The argument of this complex sequence as a function of frequency is called the phase spectrum. In the context of the SFT, this results in a series of time-varying phase spectra, one for each time point. The group delay spectrum is a function describing the time delay characteristics of different frequency components of a signal during propagation or within a system. In signal processing, it is defined as the negative derivative of the phase spectrum with respect to frequency, reflecting the arrival time or time delay of each frequency component relative to a reference point.

[0092] The terminal selects a time window and a sliding step size, and slides the window along the signal sequence from left to right. At each window position, the signal segment within the window is multiplied by the window function, and then a standard Fourier transform is performed on the windowed segment to obtain a complex spectrum at that moment, resulting in a two-dimensional complex matrix where rows correspond to frequencies and columns correspond to time. For each complex element in the complex matrix obtained by the short-time Fourier transform, its phase angle is calculated to obtain a matrix of the same dimension representing the frequency domain phase spectrum. For each column of the frequency domain phase spectrum matrix, the rate of change of phase with respect to frequency is calculated along the frequency axis. For the phase spectrum data at the current time point, the phase difference between adjacent frequency points is calculated, divided by the corresponding frequency resolution, and the result is negative. This calculation is performed over the entire frequency range to obtain the curve of group delay versus frequency at that time point, i.e., the group delay spectrum. Optionally, the focus can be on the core time period of the event, or the group delay spectrum over the entire time dimension can be averaged, or the result of a specific moment can be selected.

[0093] Step 304: Calculate the difference between the maximum and minimum values ​​in the group delay spectrum to obtain the distribution range of the group delay spectrum.

[0094] Among them, the group delay spectrum distribution range is a scalar characteristic value calculated from the group delay spectrum. Its value is the difference between the maximum and minimum values ​​of the spectrum over the entire frequency range of interest, which quantifies the range of variation of the group delay value.

[0095] The terminal determines the frequency range to be analyzed. Within the determined frequency range, it iterates through all values ​​of the group delay spectrum, finds the largest and smallest group delay values, and subtracts the smallest value from the largest value to obtain the difference, which is the distribution range of the group delay spectrum.

[0096] Step 305: Integrate the maximum absolute amplitude of the phase jump, the number of significant phase jumps, and the distribution range of the group delay spectrum to obtain a multidimensional feature vector.

[0097] In this embodiment, "integration" specifically refers to the operation of combining multiple independent scalar feature values ​​in a predetermined order to form an ordered list. The multidimensional feature vector is a one-dimensional array that sequentially contains three calculated feature values: the maximum absolute amplitude of the phase jump, the number of significant phase jumps, and the distribution range of the group delay spectrum.

[0098] The terminal obtains the maximum absolute amplitude of the phase jump, the number of significant phase jumps, and the distribution range of the group delay spectrum. It then arranges these three values ​​in a pre-agreed fixed order to form a list or array containing three elements.

[0099] This embodiment uses multidimensional feature vectors to concisely describe the core characteristics of the current transient phase change event in the time and frequency domains, providing a standardized and information-intensive input for subsequent time trend analysis and final health status assessment, and realizing the transformation from the original waveform to a digital feature description that can be interpreted by machines.

[0100] In one embodiment, based on a real-time feature queue, trend analysis is performed on each feature in the multi-dimensional feature vector to obtain time trend indicators, including:

[0101] Step 401: Based on the real-time feature queue, with the feature value as the dependent variable and the corresponding transient phase change event number as the independent variable, perform linear fitting to obtain a trend line; based on the trend line, obtain the trend slope value.

[0102] The real-time feature queue is a data structure arranged chronologically, storing multi-dimensional feature vectors corresponding to each transient phase change event within a historical timeframe from the current moment. This represents the recent historical record of device feature evolution over time. The transient phase change event numbers are a monotonically increasing sequence of integers, uniquely identifying each event in chronological order. Number 1 represents the first recorded event, number 2 the second, and so on, serving as a proxy variable for time sequence. Linear fitting is a mathematical statistical method used to find a straight line that minimizes the overall deviation from a set of discrete data points; this line is called the best-fit line. The trend line is the best-fit line obtained through linear fitting, visually representing the approximate linear trend of the analyzed feature values ​​as a function of the event sequence. The trend slope is a key parameter extracted from the trend line; mathematically, it represents the slope of the fitted line, its magnitude indicating the average rate of change of the feature value with respect to the event number, with the sign indicating the direction of change.

[0103] For a specific feature to be analyzed in the real-time feature queue, the terminal extracts the corresponding values ​​of that feature across all historical events, forming a list. It then obtains the numbers of the transient phase change events corresponding to the feature values, forming another list. The terminal uses the least squares method to perform linear fitting on the data points. The core operation of the least squares method is to find a straight line y = a*x + b that minimizes the sum of the squares of the vertical distances from all data points to this line. This involves solving a system of equations about the slope a and the intercept b to minimize the sum of squared errors. After the calculation, the determined parameters a and b uniquely define a straight line, the trend line. From the obtained linear equation, the coefficient a is directly extracted; this is the trend slope value, which physically represents how many units the feature value is expected to increase or decrease on average with each new event.

[0104] Step 402: Based on the sliding window and the real-time feature queue, calculate the arithmetic mean of the features in the sliding window to obtain the moving average.

[0105] Specifically, a sliding window is a fixed-size time frame used for local data analysis, starting from the end of the real-time feature queue and covering a specified number of historical events backward. The moving average is a statistic calculated for a specific feature in the real-time feature queue; its value is the arithmetic mean of the feature's values ​​over the most recent events covered by the sliding window, reflecting the feature's average level over a short period.

[0106] The terminal sets the sliding window size to K. In the real-time feature queue, it finds the latest K feature vectors. If the total length of the queue is less than K, it takes the entire queue. From the K feature vectors, it extracts the value of the specific feature to be analyzed, resulting in a list containing K values. It sums all the values ​​in the list and then divides by the number of values ​​K to obtain the moving average.

[0107] Step 403: Based on the sliding window and the real-time feature queue, calculate the percentage of each feature whose value is greater than the mild threshold to obtain the excess percentage.

[0108] Specifically, the mild threshold is a pre-defined boundary value for a specific feature, indicating a need for attention or a slight anomaly. It is a sensitive threshold used to provide early warning of potential problems. The excess percentage is a proportion calculated for a specific feature in the real-time feature queue. It represents the percentage of events in a sliding window where the value of that feature exceeds its corresponding mild threshold, out of the total number of events in the window. It quantifies the frequency of slight anomalies in that feature in the recent period.

[0109] The terminal determines the size K of the sliding window and extracts a list of values ​​for the specific feature to be analyzed from the last K events in the real-time feature queue. It then obtains a preset mild threshold for that feature and initializes a counter to 0. Next, it iterates through each value in the list, checking if it exceeds the threshold. If it does, the counter is incremented by 1. After completing the iteration, the counter value C is obtained, representing the number of events exceeding the threshold within window K. The excess percentage is calculated as C / K, resulting in a value between 0 and 1. This measures the frequency with which a feature value deviates from its normal or good range in the recent past. A consistently high excess percentage indicates that the feature frequently touches or exceeds the attention threshold, even if each excess may be small. It suggests an unstable or continuously slightly deteriorating state pattern and is an effective indicator for capturing intermittent problems and assessing state stability.

[0110] Step 404: Integrate the trend slope value, moving average value, and excess percentage to obtain the time trend indicator.

[0111] Integration specifically refers to the operation of combining three different trend indicators for the same feature—namely, the trend slope value, the moving average, and the percentage of exceeding the standard—to form a triple describing the dynamic behavior of that feature. A time trend indicator, for a given feature, is a data unit containing three elements: the trend slope value, the moving average, and the percentage of exceeding the standard. For each feature in a multidimensional feature vector, a set of such indicators is generated. Arranging all the feature triples in the order of the features constitutes a complete set of time trend indicators.

[0112] Suppose we are currently processing the j-th feature in a multidimensional feature vector. The terminal obtains the trend slope, moving average, and overshoot percentage of this feature. These three values ​​are combined into an ordered tuple or a small vector in a fixed, predefined order. This triplet is the time trend indicator of the j-th feature. This calculation is repeated for each feature in the multidimensional feature vector. All triplets are connected or concatenated according to the original feature order to form a longer vector. This long vector is the time trend indicator containing all feature trend information.

[0113] This embodiment enables the health assessment model to not only observe the current instantaneous value of a feature, but also to gain insight into the historical behavioral patterns of each feature, thereby making a more comprehensive and forward-looking judgment and improving the pertinence and reliability of degradation identification.

[0114] In one embodiment, based on the transient signal sequence, data credibility assessment is performed to obtain a data credibility judgment result, including:

[0115] Step 501: Based on the transient signal sequence, calculate the variance of the background signal to obtain the noise power, and calculate the variance of the transient signal sequence within the time window corresponding to the transient phase change event to obtain the total power.

[0116] In this context, the background signal refers to the relatively stable segment of the transient signal sequence that does not belong to the transient phase change event itself, representing the background noise or environmental interference under normal equipment operation. Noise power is a quantitative indicator characterizing the intensity of background noise; its value is the variance of the background signal. Variance measures the degree to which the signal value fluctuates around its mean; the larger the variance, the stronger the background noise energy. The time window is a specific time interval of signal length intercepted around the moment the transient phase change event occurs, containing the main part of the event. Total power is a quantitative indicator characterizing the overall signal strength of the transient phase change event; its value is the variance of the transient signal sequence within the time window, reflecting the total energy of the event signal.

[0117] The terminal selects a relatively stable segment of the transient signal sequence, far from the event occurrence point, as the background signal segment. This segment should represent the sensor's noise floor. The variance of the background signal is calculated by first calculating the average value of all sampling points in the segment, then calculating the square of the difference between each sampling point value and the mean, and finally calculating the average of these squared values. The result is the noise power. Based on the start and end times or peak position of the transient phase change event, a time window covering the entire event process is determined. From the complete transient signal sequence, a signal segment within the time window is extracted, and the variance of the extracted signal segment is calculated using the same method as the noise power calculation. The result is the total power.

[0118] Step 502: Calculate the signal-to-noise ratio based on the noise power and total power.

[0119] The formula for the signal-to-noise ratio is:

[0120]

[0121] Wherein, SNR is the signal-to-noise ratio. For total power, This represents noise power.

[0122] Specifically, the signal-to-noise ratio (SNR) is a core indicator used to measure signal quality. It quantifies the ratio of useful signal power to background noise power in a transient phase change event and is expressed in decibels. A higher SNR means a clear signal with minimal noise impact, while a lower SNR means the signal is severely contaminated by noise.

[0123] The terminal subtracts the noise power from the total power to obtain the power estimate of the useful signal of the event. The noise energy is separated from the total energy, and the energy of the event signal is retained. The useful signal power (Ps) is divided by the noise power (Pn) to obtain a power ratio (Ps / Pn). The power ratio is then substituted into the formula SNR=10*log10(Ps / Pn) for calculation. The result is the signal-to-noise ratio, in decibels.

[0124] Step 503: Based on the type of transient phase burst event, extract the standard transient waveform template, and compare the standard transient waveform template with the transient signal sequence to obtain the distortion rate index.

[0125] Specifically, the type of transient phase burst event is a classification of transient events. Examples may include partial discharge, switching operations, and arcing to ground. Different types of events have typical waveform characteristics known theoretically or empirically. A standard transient waveform template is an idealized or typical signal waveform sequence corresponding to a type of transient phase burst event. It can be derived from theoretical simulations, the average of historical typical samples, or standard specifications, representing the regulated waveform that such events should have. The distortion rate index is a scalar value that quantifies the degree of difference between the actually acquired transient signal sequence and the standard transient waveform template. The greater the difference, the higher the distortion rate, indicating a more severe deviation between the actual waveform and the ideal form.

[0126] The terminal determines the type of the current transient phase change event through preliminary waveform analysis, event source information, or other auxiliary criteria. Based on the type, it retrieves the corresponding standard transient waveform template from a preset template library. This template should be aligned with the signal sequence to be compared in terms of time length and sampling rate. The denoised transient signal sequence is compared with the extracted standard transient waveform template. Optionally, the operation method includes calculating the residual by subtracting the amplitudes of the two sequences at the same time point to obtain the residual sequence; calculating the distortion rate index by calculating the ratio of the energy of the residual sequence to the energy of the template sequence, or by using the correlation coefficient and subtracting the correlation coefficient from 1 as a measure of the degree of distortion to obtain a specific value, namely the distortion rate index.

[0127] Step 504: Based on the quality threshold, compare the signal-to-noise ratio and distortion rate index with the corresponding quality threshold to obtain the data credibility judgment result.

[0128] The quality thresholds are a set of pre-defined limits used to judge whether the signal-to-noise ratio (SNR) and distortion rate are acceptable, including the minimum acceptable SNR and the maximum acceptable distortion rate. The data reliability judgment result is a binary judgment conclusion derived by comparing the SNR and distortion rate with their respective quality thresholds.

[0129] Terminal: The signal-to-noise ratio (SNR) is compared with the preset minimum acceptable SNR threshold, and the distortion rate is compared with the preset maximum acceptable distortion rate threshold. Based on the comparison results, logical synthesis is performed. For example, the transient signal sequence data quality is determined to be qualified only when the SNR is not lower than its minimum threshold and the distortion rate is not higher than its maximum threshold. Otherwise, the data quality is determined to be unreliable. The conclusion of the comprehensive judgment is the data reliability judgment result.

[0130] This embodiment uses quantitative evaluation of two dimensions, signal purity and waveform fidelity, and provides a clear final judgment on whether the data is reliable based on preset objective standards. This serves as the core basis for deciding whether to use the data or with what weight, thereby effectively filtering out low-quality data at the source of the process and improving the robustness of degradation identification.

[0131] In one embodiment, after generating the degradation identification result based on the health level, the method further includes:

[0132] Step 601: Perform time series smoothing on the sequence of health status scores to obtain a smoothed health score sequence.

[0133] The health status score sequence is a chronologically ordered list, where each element is a health status score obtained by the sensor at different points in history, reflecting the evolution of the device's health status over time. Time series smoothing is a signal processing technique used to reduce random fluctuations, short-term noise, or outliers in a data sequence while preserving its main long-term trends and periodic components. The aim is to make the data sequence smoother and easier to observe its inherent trends. The smoothed health score sequence is a new sequence obtained by processing the health status score sequence through time series smoothing. Short-term fluctuations and irregular jumps in the new sequence are suppressed, resulting in a smoother and more stable overall trajectory.

[0134] The terminal selects a specific time series smoothing method, such as simple moving average, weighted moving average, exponential smoothing, or low-pass filtering. It chooses a simple moving average method with a window length of L. Starting from the first score in the original health status score sequence, it processes each point in the sequence sequentially. For the i-th score in the sequence, it calculates the arithmetic mean of the L scores before and after it. This calculated mean is used as the smoothed value for the corresponding position in the new sequence. For each point in the original score sequence, it calculates a corresponding smoothed value. All the calculated smoothed values ​​are arranged in the original time order to form a smoothed health score sequence.

[0135] Step 602: Based on a short-term sliding window, count the number of times the score in the smoothed health score sequence exceeds the warning threshold to obtain the number of times the limit is exceeded, and obtain the warning identifier based on the warning rules and the number of times the limit is exceeded.

[0136] Specifically, a short-time sliding window is a fixed-length time interval used for local statistical analysis of a smoothed health score sequence. The window slides along the sequence, focusing only on the most recent segment of smoothed score data within the window at a time. The warning threshold is a pre-set numerical limit used to determine whether a health score has entered a state requiring vigilance; when a health score falls below this threshold, it is considered abnormal or in a warning zone. The number of times the limit is exceeded is a statistical value representing the total number of times the score value in the smoothed health score sequence exceeds the warning threshold within the recent time period covered by the short-time sliding window. Warning rules are a set of predefined logical judgment conditions used to transform the statistical value into a clear warning state identifier. The warning identifier is a discrete state identifier obtained after applying the warning rules to judge the number of times the limit is exceeded.

[0137] The terminal sets the length of the short-term sliding window to W. At the end of the smooth health score sequence, the W most recent score values ​​are taken. The W values ​​are traversed, and each value is compared with the warning threshold. Whenever a score value exceeds the warning threshold, a counter is incremented by 1. After the traversal is completed, the value of the counter is the number of times the limit is exceeded. The number of times the limit is exceeded is input into the preset warning rule for logical judgment. For example, the rule is defined as: if the number of times the limit is exceeded is greater than N, then it is judged whether the condition is true. According to the judgment result, the corresponding warning identifier is generated. If the condition is true, the warning identifier is set to warning; if it is not true, it is set to normal.

[0138] Step 603: Calculate the first and second differences of the smoothed health score sequence to obtain the trend severity index.

[0139] Specifically, for a time series, its first-order differencing sequence is a new sequence obtained by calculating the difference between each element in the original sequence and its preceding element. It describes the amount of change between adjacent points in the original sequence, i.e., the instantaneous rate of change. Second-order differencing is a new sequence obtained by further differencing the first-order differencing sequence. It describes the amount of change in the first-order differencing sequence, i.e., the rate of change of the original sequence's rate of change, which can be understood as the acceleration of change. The trend severity index is a quantitative value calculated or combined from the first-order and second-order differencing information of a smoothed health score sequence. It is used to characterize the severity of the deterioration trend in the health status score and can be a combination of the latest first-order and second-order differencing values.

[0140] The terminal starts with the second score in the smoothed health score sequence, subtracting the previous score from the next to obtain a first-order difference sequence. This sequence reflects the rate of change of the score between adjacent time points. The same difference operation is then performed on the first-order difference sequence, starting from the second element, to calculate the second-order difference sequence, which reflects the rate of change of the score itself, i.e., acceleration. Information is extracted from the most recent first-order and second-order difference values ​​to form a trend severity index. For example, the latest first-order difference value, i.e., the last one in the sequence, is taken as the current rate of change, and the latest second-order difference value is taken as the current acceleration. The two values ​​are output as a pair as the trend severity index.

[0141] Step 604: Based on the warning identifier and the trend severity index, adjust the degradation identification result to obtain the warning identification result, which is used to indicate the issuance of a warning.

[0142] The adjustment process refers to revising, confirming, or reinforcing preliminary conclusions drawn from single-scores and fixed-level thresholds based on newly added information reflecting recent dynamics and trends. The early warning identification result is a final, decisive judgment derived from a comprehensive analysis of multiple factors, including the device's current health level, recent frequency of anomalies, and the severity of deterioration trends. It clearly indicates whether an early warning is needed and the possible warning level, directly driving the early warning system's actions.

[0143] The terminal generates the final warning identification result by combining the warning identifier and the trend severity indicator according to the preset adjustment logic rules. For example, if the degradation identification result itself indicates severe degradation, the warning identification result is directly confirmed as the highest level warning, regardless of other indicators. If the degradation identification result indicates slight degradation or caution, but the warning identifier shows a warning, and the trend severity indicator shows a rapid deterioration or acceleration, the warning identification result may be upgraded by one level. If the degradation identification result indicates normal, but the warning identifier is a warning and the trend is severe, the result may be adjusted to a warning. If the degradation identification result indicates slight abnormality, but the warning identifier is normal and the trend is stable or improving, the result may be downgraded or maintained. The judgment generated according to the adjustment logic is the warning identification result, and it is output to drive subsequent warning actions.

[0144] This embodiment organically integrates static level determination based on single-point scoring, statistical early warning based on recent frequency, and trend early warning based on dynamic changes, significantly improving the intelligence and accuracy of early warning: it avoids false alarms caused by fluctuations in a single score and prevents missed alarms due to slow but continuous deterioration; it can respond quickly to sudden serious faults and provide early warnings for gradual degradation, thereby generating more reliable and reasonable degradation identification results.

[0145] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0146] Based on the same inventive concept, this application also provides a critical load early degradation identification system based on transient phase change characteristics for implementing the aforementioned critical load early degradation identification method based on transient phase change characteristics. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more critical load early degradation identification system embodiments based on transient phase change characteristics provided below can be found in the limitations of the critical load early degradation identification method based on transient phase change characteristics described above, and will not be repeated here.

[0147] In one exemplary embodiment, such as Figure 2 As shown, a critical load early degradation identification system 700 based on transient phase change characteristics is provided, comprising:

[0148] The denoising module 701 is used to acquire the discrete-time signal sequence corresponding to the transient phase change event, and to denoise the discrete-time signal sequence to obtain the transient signal sequence.

[0149] Evaluation module 702 is used to evaluate the reliability of data based on transient signal sequences and obtain the data reliability judgment result;

[0150] Feature module 703 is used to extract features from transient signal sequences based on data reliability judgment results to obtain enhanced feature vectors;

[0151] The scoring module 704 is used to input the enhanced feature vector into the health benchmark model to evaluate the health status of the equipment and obtain a health status score; the health benchmark model is trained using historical enhanced feature vectors.

[0152] The identification module 705 is used to map the health status score to the corresponding health level based on the level threshold, and generate a deterioration identification result based on the health level. The deterioration identification result is used to indicate the need for early warning.

[0153] Furthermore, feature module 703 is also used for:

[0154] Based on the data reliability judgment results, Hilbert transform is performed on the transient signal sequence to obtain the analytic signal, and based on the analytic signal, the instantaneous values ​​of the local phase are extracted to obtain the instantaneous phase sequence;

[0155] Based on instantaneous phase sequences and transient signal sequences, time-frequency domain feature extraction is performed to obtain multidimensional feature vectors;

[0156] A multidimensional feature vector is added to the end of the historical feature queue to obtain the real-time feature queue. Based on the real-time feature queue, trend analysis is performed on each feature in the multidimensional feature vector to obtain a time trend index.

[0157] By integrating multidimensional feature vectors and time trend indicators, an enhanced feature vector is obtained.

[0158] Furthermore, feature module 703 is also used for:

[0159] Calculate the difference between each adjacent point in the instantaneous phase sequence to obtain the instantaneous phase difference sequence, and extract the maximum absolute value from the instantaneous phase difference sequence to obtain the maximum absolute amplitude of the phase jump;

[0160] Traverse the instantaneous phase difference sequence and calculate the number of absolute values ​​greater than the difference threshold to obtain the significant phase jump number;

[0161] A short-time Fourier transform is performed on the transient signal sequence to obtain the frequency domain phase spectrum, and the group delay spectrum is calculated based on the frequency domain phase spectrum.

[0162] Calculate the difference between the maximum and minimum values ​​in the group delay spectrum to obtain the distribution range of the group delay spectrum;

[0163] By integrating the maximum absolute amplitude of phase jumps, the number of significant phase jumps, and the distribution range of the group delay spectrum, a multidimensional eigenvector is obtained.

[0164] Furthermore, feature module 703 is also used for:

[0165] Based on the real-time feature queue, with the feature value as the dependent variable and the corresponding transient phase change event number as the independent variable, a linear fit is performed to obtain a trend line; based on the trend line, the trend slope value is obtained.

[0166] Based on a sliding window and a real-time feature queue, the arithmetic mean of the features within the sliding window is calculated to obtain the moving average.

[0167] Based on a sliding window and a real-time feature queue, the percentage of each feature whose value is greater than a mild threshold is calculated to obtain the excess percentage.

[0168] By integrating the trend slope value, moving average, and the percentage of exceeding the standard, a time trend indicator is obtained.

[0169] Furthermore, the evaluation module 702 is used for:

[0170] Based on the transient signal sequence, the variance of the background signal is calculated to obtain the noise power, and the variance of the transient signal sequence within the time window corresponding to the transient phase change event is calculated to obtain the total power;

[0171] The signal-to-noise ratio is calculated based on the noise power and the total power.

[0172] The formula for the signal-to-noise ratio is:

[0173]

[0174] Wherein, SNR is the signal-to-noise ratio. For total power, Noise power;

[0175] Based on the type of transient phase burst event, a standard transient waveform template is extracted, and the standard transient waveform template is compared with the transient signal sequence to obtain the distortion rate index;

[0176] Based on the quality threshold, the signal-to-noise ratio and distortion rate are compared with the corresponding quality threshold to obtain the data credibility judgment result.

[0177] Furthermore, the system also includes an early warning module for:

[0178] Time series smoothing is performed on the sequence of health status scores to obtain a smoothed health score sequence;

[0179] Based on a short-term sliding window, the number of times the score in the smoothed health score sequence exceeds the warning threshold is counted to obtain the number of times the limit is exceeded, and a warning identifier is obtained based on the warning rules and the number of times the limit is exceeded.

[0180] The first and second differences of the smoothed health score series are calculated to obtain the trend severity index;

[0181] Based on the warning identifier and the trend severity indicator, the degradation identification results are adjusted to obtain the warning identification results, which are used to indicate the issuance of warnings.

[0182] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of a critical load early degradation identification method based on transient phase change characteristics as described above.

[0183] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0184] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0185] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for identifying early degradation of critical loads based on transient phase change characteristics, characterized in that, The method includes: Obtain the discrete-time signal sequence corresponding to the transient phase change event, and perform noise reduction processing on the discrete-time signal sequence to obtain the transient signal sequence; Based on the transient signal sequence, data credibility assessment is performed to obtain data credibility judgment results; Based on the data reliability judgment result, feature extraction is performed on the transient signal sequence to obtain an enhanced feature vector; The enhanced feature vectors are input into the health benchmark model to assess the health status of the equipment and obtain a health status score; the health benchmark model is trained using the historical enhanced feature vectors. Based on the level threshold, the health status score is mapped to the corresponding health level, and based on the health level, a deterioration identification result is generated.

2. The method according to claim 1, characterized in that, Based on the data reliability judgment result, feature extraction is performed on the transient signal sequence to obtain an enhanced feature vector, including: Based on the data reliability judgment result, the transient signal sequence is subjected to Hilbert transform to obtain an analytical signal, and based on the analytical signal, the instantaneous values ​​of the local phase are extracted to obtain an instantaneous phase sequence; Based on the instantaneous phase sequence and the transient signal sequence, time-frequency domain feature extraction is performed to obtain a multi-dimensional feature vector; The multidimensional feature vector is added to the end of the historical feature queue to obtain the real-time feature queue. Based on the real-time feature queue, trend analysis is performed on each feature in the multidimensional feature vector to obtain a time trend index. The enhanced feature vector is obtained by integrating the multidimensional feature vector and the time trend index.

3. The method according to claim 2, characterized in that, The step involves extracting time-frequency domain features based on the instantaneous phase sequence and the transient signal sequence to obtain a multi-dimensional feature vector, including: Calculate the difference between each adjacent point in the instantaneous phase sequence to obtain the instantaneous phase difference sequence, and extract the maximum absolute value from the instantaneous phase difference sequence to obtain the maximum absolute amplitude of the phase jump; By traversing the instantaneous phase difference sequence, the number of absolute values ​​greater than the difference threshold is calculated to obtain the significant phase jump number; The transient signal sequence is subjected to a short-time Fourier transform to obtain the frequency domain phase spectrum, and the group delay spectrum is calculated based on the frequency domain phase spectrum. Calculate the difference between the maximum and minimum values ​​in the group delay spectrum to obtain the distribution range of the group delay spectrum; The multidimensional feature vector is obtained by integrating the maximum absolute amplitude of the phase jump, the number of significant phase jumps, and the distribution range of the group delay spectrum.

4. The method according to claim 2, characterized in that, The step of performing trend analysis on each feature in the multidimensional feature vector based on the real-time feature queue to obtain time trend indicators includes: Based on the real-time feature queue, with the value of the feature as the dependent variable and the number of the corresponding transient phase change event as the independent variable, a linear fit is performed to obtain a trend line; based on the trend line, the trend slope value is obtained. Based on the sliding window and the real-time feature queue, the arithmetic mean of the features in the sliding window is calculated to obtain the moving average. Based on the sliding window and the real-time feature queue, the percentage of each feature whose value is greater than the mild threshold is calculated to obtain the excess percentage. The time trend indicator is obtained by integrating the trend slope value, the moving average value, and the percentage of exceeding the standard.

5. The method according to claim 1, characterized in that, The process of evaluating data reliability based on the transient signal sequence to obtain a data reliability judgment result includes: Based on the transient signal sequence, the variance of the background signal is calculated to obtain the noise power, and the variance of the transient signal sequence within the time window corresponding to the transient phase change event is calculated to obtain the total power; Based on the noise power and the total power, the signal-to-noise ratio is calculated. The formula for the signal-to-noise ratio is as follows: Wherein, SNR is the signal-to-noise ratio. For total power, Noise power; Based on the type of the transient phase burst event, a standard transient waveform template is extracted, and the standard transient waveform template is compared with the transient signal sequence to obtain the distortion rate index; Based on the quality threshold, the signal-to-noise ratio and the distortion rate index are compared with the corresponding quality threshold to obtain the data reliability judgment result.

6. The method according to claim 1, characterized in that, After generating the degradation identification result based on the health level, the process further includes: The sequence of health status scores is smoothed over time to obtain a smoothed health score sequence. Based on a short-term sliding window, the number of times the score in the smoothed health score sequence exceeds the warning threshold is counted to obtain the number of times the threshold is exceeded, and a warning identifier is obtained based on the warning rule and the number of times the threshold is exceeded; The first and second differences of the smoothed health score sequence are calculated to obtain the trend severity index; Based on the warning identifier and the trend severity index, the degradation identification result is adjusted to obtain the warning identification result, which is used to indicate that a warning should be issued.

7. A critical load early degradation identification system based on transient phase change characteristics, characterized in that, The system includes: The denoising module is used to acquire the discrete-time signal sequence corresponding to the transient phase change event, and to denoise the discrete-time signal sequence to obtain the transient signal sequence. The evaluation module is used to evaluate the data credibility based on the transient signal sequence and obtain the data credibility judgment result. The feature module is used to extract features from the transient signal sequence based on the data reliability judgment result to obtain an enhanced feature vector; The scoring module is used to input the enhanced feature vector into the health benchmark model to assess the health status of the equipment and obtain a health status score; the health benchmark model is trained using the historical enhanced feature vectors. The identification module is used to map the health status score to the corresponding health level based on the level threshold, and generate a deterioration identification result based on the health level. The deterioration identification result is used to indicate the need for early warning.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images