A power grid monitoring method and device based on power grid main equipment quality risk index extraction, a terminal device, and a storage medium

Abstract

This invention discloses a power grid monitoring method, device, terminal equipment, and storage medium based on the extraction of quality risk indicators for main power grid equipment, belonging to the field of equipment monitoring. The method includes: slicing the process parameter sequence of the main power grid equipment to be monitored into several parameter sequence segments within a monitoring window; calculating the number of anomalies and relative deviation of each parameter sequence segment; performing least squares regression on the normalized sequence of the parameter sequence segments to obtain the trend slope; constructing a linear membership function for each segment based on the residual stationarity index calculated using the number of anomalies, relative deviation, trend slope, and exponentially weighted moving average; generating a current risk score trajectory by weighting and summing the risk scores within each monitoring window according to preset weights; and generating equipment quality risk indicators based on the current risk score trajectory and historical risk data corresponding to quality risk labels to monitor the main power grid equipment. By implementing this invention, the problem of low monitoring accuracy in existing technologies is solved.

Description

Technical Field

[0001] This invention relates to the field of power grid equipment monitoring technology, and in particular to a power grid monitoring method, device, terminal equipment and storage medium based on the extraction of quality risk indicators of main power grid equipment. Background Technology

[0002] With the digitalization of power grid main equipment manufacturing and quality control, enterprises are gradually building multi-source data systems and conducting remote quality supervision and process compliance verification, focusing on raw material inspection, key component inspection, process and procedure inspection, type and factory testing, and finished product warehousing. To ensure the operational reliability of power grid main equipment, quality risk monitoring extracts interpretable quality risk indicators from complex production data to control potential equipment quality hazards. Currently, power grid main equipment quality risk monitoring acquires equipment identification and production process data, calls the corresponding equipment's digital model to generate standard parameters, and then compares them with measured data step by step to output component-level judgment results, thereby improving the precision of remote construction.

[0003] However, current digital models rely on fixed standard parameters generated by the model as a comparison benchmark and use Boolean logic to judge whether a parameter is qualified or unqualified. This cannot cope with slight drifts in process parameters caused by the interaction of multiple factors during the production process, which can easily lead to false alarms and reduce the accuracy of monitoring the main equipment of the power grid. Summary of the Invention

[0004] This invention provides a power grid monitoring method, device, terminal equipment, and storage medium based on the extraction of quality risk indicators of main power grid equipment. It can effectively solve the problem that existing technologies cannot cope with slight drifts in process parameters caused by the linkage of multiple factors during production, which easily leads to false alarms and reduces the accuracy of main power grid equipment monitoring.

[0005] An embodiment of the present invention provides a power grid monitoring method based on the extraction of quality risk indicators of main power grid equipment, comprising: Acquire the process parameter sequence, corresponding quality risk labels, and historical risk data of the main power grid equipment to be monitored during the production process; The process parameter sequence is sliced ​​within a preset monitoring window to obtain several parameter sequence sub-segments. The number of anomalies and the relative deviation of the parameter sequence sub-segments within the monitoring window are calculated. The parameter sequence segments are normalized to obtain a normalized sequence. Least square regression is performed based on the normalized sequence to obtain the trend slope. An exponentially weighted moving average is then calculated based on the trend slope and the normalized sequence to obtain the residual stationarity index. Based on the number of anomalies, the relative deviation, the trend slope, and the residual stationarity index, a piecewise linear membership function is constructed for each segment. The piecewise linear membership function and a preset weight are used to perform a weighted summation to obtain the risk score within each monitoring window. The current risk score trajectory is generated based on the risk score within each monitoring window. Based on the historical risk data corresponding to the quality risk label, the equipment quality risk index is generated by time aggregation of the current risk score trajectory and the corresponding historical risk data. The main equipment of the power grid to be monitored is monitored according to the equipment quality risk indicators mentioned above; wherein, the equipment quality risk indicators include the maximum risk value, the proportion of exceeding the limit, and the long-term average risk level.

[0006] Furthermore, the process parameter sequence includes several time parameter data pairs, each time parameter data pair including a timestamp and the process parameter sample value at the current time corresponding to the timestamp; Based on the process parameter sequence, slices are generated within a preset monitoring window to obtain several parameter sequence sub-segments, including: Based on the preset time offset, the timestamps of each time parameter data pair are mapped to the target time axis, and a monitoring window is constructed on the target time axis according to the type of process parameter sampling value. Based on the preset slicing operator, all time parameter data pairs within each monitoring window are extracted to generate several parameter sequence segments.

[0007] Furthermore, the calculation of the number of anomalies and the relative deviation of the parameter sequence segment within the monitoring window includes: For each parameter sequence segment, a deviation coefficient for each process parameter sample value is calculated based on the process parameter sample value and a preset process parameter benchmark value; wherein, the preset process parameter benchmark value includes the standard value of the process parameter and the deviation range. The number of process parameter sample values ​​in each parameter sequence segment whose deviation coefficient is greater than a preset deviation threshold is taken as the number of abnormalities of the parameter sequence segment within the monitoring window. Calculate the median of the process parameter sample values ​​within each parameter sequence segment, and perform robust normalization of the process parameter sample values ​​and the median of the process parameter sample values ​​based on the median and absolute deviation method to obtain normalized sample values; The relative deviation is calculated based on the normalized sampled values, the standard values ​​of process parameters, and the deviation range.

[0008] Further, least squares regression is performed based on the normalized sequence to obtain the trend slope, including: Based on the mapped timestamp on the target time axis corresponding to each normalized sample value in the normalized sequence, a target data pair is constructed to characterize the pairing of the mapped timestamp and the normalized sample value. Based on the target data, a least squares regression objective function is constructed, and the trend slope is obtained by solving the least squares regression objective function.

[0009] Furthermore, based on the trend slope and the normalized sequence, an exponentially weighted moving average is calculated to obtain the residual stationarity index, including: The regression fit value is calculated based on the trend slope, the regression intercept corresponding to the trend slope, and the corresponding mapping timestamp. The regression residuals are calculated based on the regression fitted values ​​and the normalized sequence, and the standard deviation of the regression residuals is determined based on the mean of the regression residuals within the current parameter sequence segment. The stationarity index of the residuals is obtained by calculating the index of stationarity of the residuals through an exponentially weighted moving average based on the regression residuals and the standard deviation of the regression residuals.

[0010] Further, piecewise linear membership functions are constructed based on the number of anomalies, the relative deviation, the trend slope, and the residual stationarity index. These piecewise linear membership functions are then weighted and summed using preset weights to obtain a risk score within each monitoring window, including: Based on the number of anomalies, the relative deviation, the trend slope, and the residual stationarity index, membership functions for the number of anomalies, the relative deviation, the trend slope, and the residual stationarity index are constructed respectively based on their corresponding preset upper and lower limits. The risk score for each monitoring window is obtained by weighted summation based on the membership functions of the number of anomalies, relative deviation, trend slope, residual stability index, and preset weights. The preset weights are obtained by defuzzification calculation based on the preset triangular fuzzy function judgment matrix.

[0011] Furthermore, the quality risk label includes a first label for indicating that the quality is acceptable and a second label for indicating that the quality is unacceptable: Based on the historical risk data corresponding to the aforementioned quality risk labels, and by performing time aggregation on the current risk score trajectory and the corresponding historical risk data, equipment quality risk indicators are generated, including: Historical risk data labeled with the first quality risk label is used as the first historical sample, the historical time interval corresponding to the first historical sample is used as the historical benchmark window, and the control limits are determined based on the historical benchmark window. The maximum risk score in the current risk score trajectory is taken as the maximum risk value; The ratio of the number of windows with risk scores greater than the control limit in the current risk score trajectory to the total number of monitoring windows is taken as the excess ratio. The average risk score in the current risk score trajectory is taken as the long-term average risk level; The maximum risk value, the proportion of exceeding limits, and the long-term average risk level are used as indicators of equipment quality risk.

[0012] As an improvement to the above solution, another embodiment of the present invention provides a power grid monitoring device based on the extraction of quality risk indicators of main power grid equipment, comprising: The power grid main equipment data acquisition module is used to acquire the process parameter sequence, corresponding quality risk labels, and historical risk data of the main equipment to be monitored during the production process. The first indicator calculation module is used to slice the process parameter sequence within a preset monitoring window to obtain several parameter sequence sub-segments, and calculate the number of anomalies and relative deviation of the parameter sequence sub-segments within the monitoring window. The second indicator calculation module is used to normalize the parameter sequence segments to obtain a normalized sequence, perform least squares regression based on the normalized sequence to obtain the trend slope, and perform exponential weighted moving average calculation based on the trend slope and the normalized sequence to obtain the residual stationarity index. The risk score calculation module is used to construct a piecewise linear membership function based on the number of anomalies, the relative deviation, the trend slope, and the residual stationarity index, respectively, and to perform a weighted summation based on the piecewise linear membership function and preset weights to obtain the risk score within each monitoring window, and to generate the current risk score trajectory based on the risk score within each monitoring window. The equipment quality risk index generation module is used to generate equipment quality risk indexes by using the historical risk data corresponding to the quality risk label as a benchmark and performing time aggregation based on the current risk score trajectory and the corresponding historical risk data. The power grid main equipment monitoring module is used to monitor the main equipment of the power grid to be monitored according to the equipment quality risk indicators; wherein, the equipment quality risk indicators include the maximum risk value, the proportion of exceeding the limit, and the long-term average risk level.

[0013] Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements a power grid monitoring method based on the extraction of power grid main equipment quality risk indicators as described in the above embodiments.

[0014] Another embodiment of the present invention provides a computer-readable storage medium, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the power grid monitoring method based on the extraction of power grid main equipment quality risk indicators as described in the above embodiment.

[0015] By implementing this invention, at least the following beneficial effects are achieved: This invention provides a power grid monitoring method, device, terminal equipment, and storage medium based on the extraction of quality risk indicators for main power grid equipment. The method calculates the number of anomalies and relative deviations based on process parameter sequences, statistically analyzing both explicit anomalies exceeding benchmarks and quantifying the overall parameter deviation, avoiding the omission of minor drifts that, while not exceeding limits, exhibit continuous deviations. By normalizing the sequence, it obtains trend slope and residual stationarity indicators, capturing the evolution of parameters over time rather than judging individual parameter values ​​in isolation, thus achieving more accurate perception of minor drifts. Furthermore, by constructing a piecewise linear membership function, it transforms the indicator values ​​into risk membership degrees, and then obtains continuous risk scores through weighted summation, replacing rigid Boolean judgments and allowing minor drifts to correspond to gradient risk scores, avoiding false alarms. Finally, using historical risk data corresponding to quality risk labels as a benchmark, it aggregates the current risk score trajectory with the corresponding historical risk data over time to generate the maximum risk value, exceedance ratio, and long-term average risk level, characterizing equipment quality risk from a holistic perspective rather than judging isolated processes, further reducing the probability of false alarms caused by minor drifts and significantly improving the accuracy of main power grid equipment monitoring. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a power grid monitoring method based on the extraction of quality risk indicators of main power grid equipment, according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of a power grid monitoring device based on the extraction of quality risk indicators of main power grid equipment, according to an embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] See Figure 1To address the problem that existing technologies cannot handle slight drifts in process parameters caused by the interaction of multiple factors during production, leading to false alarms and reduced accuracy in monitoring power grid main equipment, an embodiment of the present invention provides a flowchart illustrating a power grid monitoring method based on the extraction of quality risk indicators from power grid main equipment, comprising: S1. Obtain the process parameter sequence, corresponding quality risk label, and historical risk data of the main equipment of the power grid to be monitored during the production process; Specifically, the process parameter sequence refers to the set of continuous parameter data directly related to equipment quality collected by field sensors, test bench recording systems, etc., during the production of main power grid equipment. It includes several time parameter data pairs, reflecting the dynamic changes in the process. The quality risk label is a binary label used to identify whether there is a quality risk in the main power grid equipment or production batch. It consists of a first label (label value 0) indicating qualified quality and a second label (label value 1) indicating unqualified quality. The judgment criteria include records of unqualified factory tests and feedback on defects during grid acceptance. Historical risk data refers to the total amount of data accumulated during the historical production of the same model of main power grid equipment, including process parameter sequences, quality risk labels, risk score trajectories, and corresponding technical standard clause information.

[0019] In a preferred embodiment of the present invention, the process parameter sequence P includes several pairs of time parameter data. , Each time parameter data pair includes a timestamp and the sampled process parameter value at the current moment corresponding to the timestamp. For timestamps, For a moment The sampled values ​​of process parameters.

[0020] S2. Slice the process parameter sequence within a preset monitoring window to obtain several parameter sequence segments, and calculate the number of anomalies and relative deviation of the parameter sequence segments within the monitoring window; Specifically, the monitoring window represents the time interval used to segment the process parameter sequence, including fixed windows and event windows, facilitating the focus on calculating risk characteristics of local processes. A parameter sequence segment represents a local data fragment extracted from the process parameter sequence using a slicing operator, corresponding to a specific monitoring window, containing all time parameter data pairs within that window. The anomaly count represents the cumulative number of times the sampled process parameter values ​​within the monitoring window exceed the preset benchmark range, reflecting the frequency of parameter out-of-bounds errors. The relative deviation represents the degree of deviation of the overall level of the process parameters within the monitoring window from the standard value; after dimensionless processing, it facilitates cross-parameter comparisons.

[0021] Indicatively, the process parameter sequence is sliced ​​within a preset monitoring window (fixed window or event window) to obtain several parameter sequence sub-segments; for each sub-segment, the number of anomalies is calculated to count the number of out-of-bounds sampling values, the relative deviation is calculated to quantify the overall parameter offset, and core indicators reflecting parameter out-of-bounds characteristics and overall offset characteristics are extracted.

[0022] Preferably, the process parameter sequence includes several time parameter data pairs, and each time parameter data pair includes a timestamp and the process parameter sample value at the current time corresponding to the timestamp; Based on the process parameter sequence, slices are generated within a preset monitoring window to obtain several parameter sequence sub-segments, including: Based on the preset time offset, the timestamps of each time parameter data pair are mapped to the target time axis, and a monitoring window is constructed on the target time axis according to the type of process parameter sampling value. Based on the preset slicing operator, all time parameter data pairs within each monitoring window are extracted to generate several parameter sequence segments.

[0023] Specifically, time parameter data pairs are the basic building blocks of the process parameter sequence, consisting of a timestamp (acquisition time) and the corresponding sampled process parameter value (the measured value of the parameter at that time), ensuring the temporal correlation of the parameter data. The time offset is the difference between the actual trigger time and the theoretical trigger time of different data sources (such as sensors and recording systems), used to correct time deviations between data sources. The target time axis is a unified time reference axis; timestamps from all data sources are mapped to this axis after correction by the time offset, achieving time synchronization of data from different data sources. The type of process parameter sampled value is a classification of the process parameter's attributes, such as slowly varying parameters (temperature, pressure) and short-term abrupt change parameters (voltage at the moment of closing), with different parameter types corresponding to different monitoring window types. The slicing operator is a preset data analysis tool used to extract all time parameter data pairs within the monitoring window, achieving unified alignment and segmentation of data with different sampling rates and trigger calibers.

[0024] In a preferred embodiment of the present invention, each data pair includes a timestamp and a corresponding process parameter sampling value, such as temperature, pressure, humidity, and partial discharge amplitude, ensuring the structure and integrity of the data. First, the time offset of each data source is estimated, and the offset value is determined through a multi-data source synchronous calibration test. The original timestamp of each time parameter data pair is added to the corresponding time offset and mapped onto the target time axis, achieving time alignment of data from different data sources and eliminating the impact of time offset on subsequent analysis. Then, the monitoring window type is selected according to the type of process parameter sampling value. For slowly varying parameters such as temperature and pressure, a fixed window (e.g., 10 minutes in length) is constructed; for short-term abrupt changes such as the moment of valve closing or valve action, an event window (e.g., 2 minutes before and after the critical event) is constructed, ensuring that the monitoring window matches the parameter characteristics and improving the accuracy of feature extraction. Next, using a preset slicing operator, all time parameter data pairs within each monitoring window are extracted, generating several parameter sequence segments. Each segment includes the mapped timestamp within the window, the process parameter sampling value, and the identifier of its data source, achieving reproducible alignment of data with different sampling rates and providing standardized data fragments for subsequent feature calculation.

[0025] This embodiment achieves time synchronization across different data sources through time offset correction, solving the problem of inconsistent time bases across data sources in existing technologies and providing support for cross-stage data integration. A monitoring window is constructed based on parameter type, achieving precise matching between the monitoring window and parameter characteristics, improving the targeting of parameter sequence segments, and ensuring the effectiveness of subsequent feature calculations.

[0026] Preferably, the calculation of the number of anomalies and the relative deviation of the parameter sequence segment within the monitoring window includes: For each parameter sequence segment, a deviation coefficient for each process parameter sample value is calculated based on the process parameter sample value and a preset process parameter benchmark value; wherein, the preset process parameter benchmark value includes the standard value of the process parameter and the deviation range. The number of process parameter sample values ​​in each parameter sequence segment whose deviation coefficient is greater than a preset deviation threshold is taken as the number of abnormalities of the parameter sequence segment within the monitoring window. Calculate the median of the process parameter sample values ​​within each parameter sequence segment, and perform robust normalization of the process parameter sample values ​​and the median of the process parameter sample values ​​based on the median and absolute deviation method to obtain normalized sample values; The relative deviation is calculated based on the normalized sampled values, the standard values ​​of process parameters, and the deviation range.

[0027] Specifically, the process parameter baseline value is a standard reference value used to determine whether the sampled values ​​of the process parameters are qualified. This includes the standard value (nominal value) and the deviation range (allowable fluctuation range) of the process parameter, determined by technical standards, equipment drawings, or statistical results of long-term qualified samples. The deviation coefficient characterizes the degree to which a single process parameter sampled value deviates from the standard value; it is dimensionless and used to determine whether the sampled value exceeds the allowable range. The preset deviation threshold is the critical value for determining whether a sampled value is abnormal, usually set to 1. A deviation coefficient greater than this threshold indicates that the sampled value exceeds the allowable range. The median and absolute deviation method is a robust normalization method that standardizes the sampled values ​​by calculating the median and median absolute deviation of the parameter sequence segments, exhibiting strong resistance to noise interference. Robust normalization is a standardization process that maintains the stability of the normalization results even in the presence of outliers and noise, avoiding the impact of outliers on subsequent analysis. Normalized sampled values ​​are obtained after robust normalization, with a concentrated value range, eliminating dimensional and scale differences, and facilitating cross-parameter comparative analysis.

[0028] In a preferred embodiment of the present invention, the standard value of each process parameter is determined based on technical standards, equipment drawings, or the statistical mean of long-term qualified samples. and deviation range To form the baseline values ​​of process parameters This provides a standard basis for anomaly detection. For each parameter sequence segment, the deviation coefficient of each process parameter sample value is calculated. Deviation coefficient = |sampled value - standard value| / deviation range , This represents the sampled values ​​of process parameters; the number of sampled values ​​in the statistical sub-segment whose deviation coefficient is greater than a preset deviation threshold represents the number of anomalies within the monitoring window. Record as an abnormal event ,otherwise Number of anomalies monitored within the window: This reflects the frequency of parameter out-of-bounds errors. The median of all process parameter samples within the parameter sequence segment is calculated. Using the median and absolute deviation method, the median is subtracted from each sample value, and then divided by (median absolute deviation × 1.4826) to obtain the normalized sample value. S represents a parameter sequence segment, eliminating dimensions and noise interference to standardize the sampled values. Finally, based on the normalized sampled values, the standard values ​​of process parameters, and the deviation range, the relative deviation is calculated using the formula (|normalized sampled value mean - standard value| / (deviation range + ε), where ε is a numerically stable term of 0.001-0.01). This quantifies the overall offset of the parameter sequence segment S, facilitating cross-window and cross-parameter comparisons.

[0029] This embodiment calculates the number of anomalies and the relative deviation, respectively, and characterizes the risk features of the parameters from two dimensions: frequency and overall offset, making the risk feature extraction more comprehensive and accurate.

[0030] S3. Normalize the parameter sequence segments to obtain a normalized sequence. Perform least squares regression on the normalized sequence to obtain the trend slope. Calculate the residual stationarity index by performing an exponentially weighted moving average on the trend slope and the normalized sequence. Specifically, a normalized sequence refers to a sequence obtained by robustly normalizing a sub-segment of the parameter sequence, eliminating differences in dimensions and scales, and providing a unified scale for subsequent trend analysis. The trend slope represents the rate and direction of parameter change obtained through least squares regression, reflecting the temporal evolution trend of the process parameters. The residual stationarity index represents the result of an exponentially weighted moving average based on the regression residuals, quantifying the stability of parameter fluctuations deviating from the temporal trend.

[0031] To illustrate, the median and absolute deviation methods are used to robustly normalize the parameter sequence segments to eliminate dimensional and noise interference; a least squares regression model is constructed based on the normalized sequence to solve for the trend slope in order to capture the time-series change trend of the parameters; and an exponentially weighted moving average is used to calculate the residual stationarity index to quantify the stationarity of parameter fluctuations.

[0032] Preferably, the trend slope is obtained by performing least squares regression based on the normalized sequence, including: Based on the mapped timestamp on the target time axis corresponding to each normalized sample value in the normalized sequence, a target data pair is constructed to characterize the pairing of the mapped timestamp and the normalized sample value. Based on the target data, a least squares regression objective function is constructed, and the trend slope is obtained by solving the least squares regression objective function.

[0033] Specifically, the target data pair consists of paired data composed of each normalized sample value in the normalized sequence and its corresponding target time axis mapped timestamp, ensuring the correlation between the normalized sample values ​​and time, and providing a foundation for time series trend analysis. The least squares regression objective function is a mathematical function that aims to minimize the sum of squared deviations between the normalized sample values ​​and the regression fitted values, used to construct a linear regression model of normalized sample values ​​and time. The trend slope is the slope parameter of the linear regression model, characterizing the rate and direction of change of the normalized sample values ​​over time. Positive numbers indicate an upward trend in the parameter, negative numbers indicate a downward trend, and the larger the absolute value, the more significant the trend.

[0034] In a preferred embodiment of the present invention, a least-squares regression objective function is constructed based on the target data pairs. The core of this function is to find the linear model that best reflects the relationship between the normalized sampled values ​​and time by minimizing the sum of squared deviations. The trend slope is obtained by solving for the partial derivatives of the objective function and setting them to zero. Analytical solution , This represents the normalized sample value of the i-th sequence at time point t. The intercept term of the i-th sequence is substituted into the statistical value of the target data pair to calculate the trend slope, thereby capturing the temporal change trend of the normalized sampled values.

[0035] This embodiment uses the least squares regression method to solve for the trend slope. The model is simple, computationally efficient, and can effectively capture the linear changing trend of parameters, making it suitable for time-series analysis of process parameters of main power grid equipment. The extraction of the trend slope provides core features of the time-series dimension for subsequent risk assessment, making up for the shortcomings of existing technologies that only focus on single-point parameter values ​​and ignore trend changes.

[0036] Preferably, the residual stationarity index is obtained by calculating an exponentially weighted moving average based on the trend slope and the normalized sequence, including: The regression fit value is calculated based on the trend slope, the regression intercept corresponding to the trend slope, and the corresponding mapping timestamp. The regression residuals are calculated based on the regression fitted values ​​and the normalized sequence, and the standard deviation of the regression residuals is determined based on the mean of the regression residuals within the current parameter sequence segment. The stationarity index of the residuals is obtained by calculating the index of stationarity of the residuals through an exponentially weighted moving average based on the regression residuals and the standard deviation of the regression residuals.

[0037] Specifically, the regression intercept is the intercept parameter of the least squares regression model, representing the baseline level of the normalized sequence near the starting point of time, and together with the trend slope, it constitutes the complete linear regression model. The regression fit value is the predicted value of the normalized sampled value calculated based on the regression intercept, trend slope, and mapping timestamp, reflecting the degree of fit of the linear regression model to the actual data. The regression residual is the difference between the normalized sampled value and the corresponding regression fit value, representing the degree to which the actual data deviates from the linear trend; the larger the residual, the more severe the deviation. The mean of the residual is the arithmetic mean of all regression residuals within the parameter sequence segment, reflecting the overall offset level of the residuals. The standard deviation of the residual is a statistic representing the dispersion of the regression residual, reflecting the severity of residual fluctuations; the larger the standard deviation, the more severe the fluctuations. The exponentially weighted moving average calculation represents a weighted smoothing process for the regression residuals, controlling the weight of the current residual and historical residuals through a smoothing coefficient, which can capture both short-term fluctuations and characterize long-term trends. The residual stationarity index is a quantitative indicator obtained by exponential weighted moving average and standardization. It characterizes the volatility stationarity of the normalized sequence deviating from the linear trend. The larger the value, the more drastic the volatility and the higher the quality risk.

[0038] In a preferred embodiment of the present invention, the exponentially weighted moving average of the residuals is calculated: , For residuals, Let S be the standard deviation of the residuals within the parameter sequence segment S. This is the smoothing coefficient, used to control the sensitivity of the exponentially weighted moving average to new residual information. The closer the value is to one, the better the smoothness of the current residual. The larger the weight in the value, the more sensitive the model is to short-term fluctuations; the closer the value is to zero, the greater the weight of the historical residuals, and the more it focuses on characterizing slow drift.

[0039] Based on the calculation of regression fit values ​​and regression residuals, a precise characterization of the deviation of parameter time-series trends is achieved, overcoming the shortcomings of existing technologies that only focus on the parameter values ​​themselves and ignore fluctuation characteristics. The introduction of the exponentially weighted moving average method can capture both short-term fluctuations in residuals and long-term trends, thus improving the ability to characterize parameter fluctuation features.

[0040] S4. Construct piecewise linear membership functions for each segment based on the number of anomalies, the relative deviation, the trend slope, and the residual stationarity index. Perform weighted summation based on the piecewise linear membership functions and preset weights to obtain the risk score within each monitoring window. Generate the current risk score trajectory based on the risk score within each monitoring window. Specifically, the piecewise linear membership function represents a function that maps risk indicator values ​​to membership values ​​in the [0,1] interval, achieving standardized quantification of risk indicators; a larger value indicates a higher risk. The risk score represents the comprehensive quantitative result obtained by weighted summation of the membership values ​​of each risk indicator within the monitoring window, reflecting the overall quality risk level within that window. The current risk score trajectory represents a continuous trajectory formed by sequentially connecting the risk scores of each window according to the time sequence of the monitoring window, reflecting the dynamic changes in quality risk during equipment production.

[0041] In a schematic manner, piecewise linear membership functions are constructed for the indicators of anomaly frequency, relative deviation, trend slope, and residual stationarity, and the values ​​of each indicator are standardized to the membership values ​​in the interval [0,1]. The membership values ​​are weighted and summed using preset weights to obtain the risk score for each monitoring window. The risk scores of each window are concatenated in chronological order to generate the current risk score trajectory, thereby realizing dynamic risk tracking.

[0042] Preferably, piecewise linear membership functions are constructed based on the number of anomalies, the relative deviation, the trend slope, and the residual stationarity index, respectively. A weighted sum is then performed based on the piecewise linear membership functions and preset weights to obtain a risk score within each monitoring window, including: Based on the number of anomalies, the relative deviation, the trend slope, and the residual stationarity index, membership functions for the number of anomalies, the relative deviation, the trend slope, and the residual stationarity index are constructed respectively based on their corresponding preset upper and lower limits. The risk score for each monitoring window is obtained by weighted summation based on the membership functions of the number of anomalies, relative deviation, trend slope, residual stability index, and preset weights. The preset weights are obtained by defuzzification calculation based on the preset triangular fuzzy function judgment matrix.

[0043] Specifically, the preset upper and lower limits represent the risk quantification boundaries set for each risk indicator, including a risk starting point (lower limit) and a risk upper limit (upper limit), determined by technical standards, process windows, and historical qualified sample statistics. The closer the membership function of the anomaly frequency is to 1, the higher the risk. The relative deviation membership function is a piecewise linear function mapping relative deviation to membership values ​​in the [0,1] interval; the closer the relative deviation is to or exceeds the preset upper limit, the closer the membership value is to 1, and the higher the risk. The trend slope membership function is a piecewise linear function mapping trend slope to membership values ​​in the [0,1] interval; the closer the absolute value of the trend slope is to or exceeds the preset upper limit, the closer the membership value is to 1, and the higher the risk. The residual stationarity index membership function is a piecewise linear function mapping residual stationarity index to membership values ​​in the [0,1] interval; the closer the residual stationarity index is to or exceeds the preset upper limit, the closer the membership value is to 1, and the higher the risk.

[0044] Specifically, the triangular fuzzy function judgment matrix is ​​a matrix constructed based on fuzzy hierarchical analysis, with elements being triangular fuzzy numbers used to characterize the relative importance of anomaly frequency, relative deviation, trend slope, and residual stationarity indicators. The preset weights represent deterministic weights obtained after defuzzifying the triangular fuzzy function judgment matrix, ensuring that the sum of the weights is 1, and are used for subsequent weighted calculation of risk scores.

[0045] Indicatively, based on a preset triangular fuzzy function judgment matrix, the preset weights are obtained after defuzzification using the centroid method and consistency verification, thus clarifying the relative importance of each risk indicator in the comprehensive evaluation.

[0046] In a preferred embodiment of the present invention, a hierarchical model is established comprising a target layer, a criterion layer, and sub-criterion layers. For any pair of indicators... , Construct the triangular fuzzy number judgment matrix: ,in, Let be the triangular fuzzy number of index a relative to index b. These represent the subjective preference boundaries for weaker, balanced, and stronger preferences, respectively. In this invention, firstly, based on practical experience in the quality management of power grid main equipment, several levels of linguistic scales are given, such as "equally important," "slightly important," "significantly important," and "strongly important," corresponding to a pre-defined set of triangular fuzzy number intervals. Then, domain experts are organized to score each indicator pair. The corresponding triangular fuzzy numbers are obtained by looking up the linguistic scales provided by the experts. Finally, the results from multiple experts are summarized using the minimum value at the lower bound, the arithmetic mean of the median, and the maximum value at the upper bound to obtain the final result. This process ensures that the boundaries of subjective preferences reflect both expert consensus and have clear quantitative basis.

[0047] In a preferred embodiment of the present invention, the fuzzy weights are obtained using geometric mean or extended analysis. The clear weights are obtained by using the center-of-gravity method to resolve fuzziness. In the formula, For the first after summarization The lower bound, median, and upper bound of the comprehensive triangular fuzzy weights of each quality risk indicator reflect the fuzzy range of the relative importance of the indicator at this level. For the same level The comprehensive triangular fuzzy weight components of each quality risk indicator are obtained. The weights of each indicator are then obtained by taking the centroid of the triangular fuzzy weights and normalizing them. This is used for subsequent weighted calculations of risk probability and risk trajectory. Consistency is then verified using the maximum eigenvalue method. In the formula, The largest eigenvalue corresponding to the judgment matrix at this level. This represents the number of quality risk indicators used in the comparison at this level. To the number of indicators The corresponding random consistency index is determined based on the standard reference table for the analytic hierarchy process (AHP). The resulting consistency ratio is... Used to measure the consistency of expert judgment matrices, when If consistency is considered acceptable, the corresponding weights can be permanently fixed in the model of this invention and managed according to version; if If the threshold is exceeded, the judgment matrix needs to be readjusted or reviewed.

[0048] In a preferred embodiment of the present invention, the preset weights include weights for the number of anomalies, relative deviation, trend slope, and residual stationarity. The membership values ​​of the four risk indicators are multiplied by their corresponding weights and then summed to obtain the risk score within each monitoring window, thus achieving comprehensive risk quantification. For each quality indicator... Constructing a piecewise linear membership function: ,in, Let be the value of the k-th quality risk indicator. The quality risk indicators in this invention are derived from the features and statistics obtained in the preceding steps, including but not limited to: relative deviation. Number of anomalies in the monitoring window Trend slope The exponentially weighted moving average of residuals. Each For a specific risk dimension, such as the degree of process parameter deviation, the intensity of trend deviation, the frequency of clause violations, or the degree of environmental condition deviation, the risk level of that dimension is reflected in real numbers. If necessary, a monotonic transformation can be used to pre-process it to the direction where "the larger the value, the higher the risk." The risk score is calculated using a weighted sum. .

[0049] In this invention, For the first The risk threshold value of each quality risk indicator indicates that the risk is considered low or the condition is normal at or below this value. This is the upper bound of the risk for this indicator, representing the risk level when... Approaching or exceeding this value should be considered a high-risk state. Both are determined jointly based on the acceptable range specified in the technical standards and the enterprise's process window: on the one hand, the nominal values, permissible deviations, and limit values ​​given in the standards or technical agreements serve as hard constraints; on the other hand, historical statistical results from long-term acceptable samples are combined to select soft constraints such as percentiles and process capability indices, so that... More closely reflects actual normal operating levels. The corresponding standard allows for an upper limit or an internal warning limit, thus balancing compliance and process feasibility.

[0050] For example, the pressure holding performance of GIS equipment can be assessed using quality risk indicators. Defined as the pressure drop rate per unit time. Relevant standards require that the 24-hour pressure drop rate not exceed a certain limit; however, in enterprise process management, a lower pressure drop rate is considered a normal control target. In this embodiment, the upper limit of the empirical control within the process window can be selected as... The upper limit specified in the standard is selected as the maximum allowable value. When measured Less than At that time, membership degree This indicates that the risk is negligible; when Between and When the membership degree is between these two values, it increases linearly according to the membership function formula, reflecting a gradual increase in risk; when... Exceed When the membership degree is truncated to 1, it indicates that the dimension is in a high-risk state.

[0051] S5. Based on the historical risk data corresponding to the quality risk label, aggregate the current risk score trajectory with the corresponding historical risk data over time to generate equipment quality risk indicators. Specifically, the equipment quality risk index is a quantitative indicator at the equipment level obtained by aggregating the current risk score trajectory over time. It includes the maximum risk value, the proportion of exceeding limits, and the long-term average risk level, and is used to comprehensively assess equipment quality risks.

[0052] Preferably, the quality risk label includes a first label for indicating that the quality is acceptable and a second label for indicating that the quality is unacceptable. Based on the historical risk data corresponding to the aforementioned quality risk labels, and by performing time aggregation on the current risk score trajectory and the corresponding historical risk data, equipment quality risk indicators are generated, including: Historical risk data labeled with the first quality risk label is used as the first historical sample, the historical time interval corresponding to the first historical sample is used as the historical benchmark window, and the control limits are determined based on the historical benchmark window. The maximum risk score in the current risk score trajectory is taken as the maximum risk value; The ratio of the number of windows with risk scores greater than the control limit in the current risk score trajectory to the total number of monitoring windows is taken as the excess ratio. The average risk score in the current risk score trajectory is taken as the long-term average risk level; The maximum risk value, the proportion of exceeding limits, and the long-term average risk level are used as indicators of equipment quality risk.

[0053] Specifically, the first label is a quality risk label with a value of 0, used to indicate that the main equipment of the power grid or the production batch is of acceptable quality. The criterion for judgment is that no quality problems such as failure to pass factory testing or defects in grid connection acceptance occur within the specified observation period. The second label is a quality risk label with a value of 1, used to indicate that the main equipment of the power grid or the production batch is unacceptable quality. The criterion for judgment is the occurrence of any quality problem such as failure to pass factory testing or defects in grid connection acceptance. The first historical sample is the sample data in the historical risk data where the quality risk label is the first label 0, i.e., the relevant data of historically acceptable equipment, used to construct a baseline for normal operation. The historical baseline window is the historical time interval corresponding to the first historical sample, used to characterize the risk characteristics under normal production conditions. The control limit is a critical value set based on the statistical measure of risk scores within the historical baseline window, used to determine whether the current risk score is abnormal, and the calculation formula is... , The historical baseline window H risk score is the average. Standard deviation, A control coefficient of 1.5-3.0 is used to determine the leniency of the control limits based on the target false alarm rate. The higher the value, the wider the control limit and the less sensitive it is to short-term fluctuations. The maximum risk value is the maximum risk score of all monitoring windows in the current risk score trajectory, reflecting the highest risk level in the equipment production process. The exceedance ratio is the ratio of the number of monitoring windows with risk scores greater than the control limit to the total number of windows in the current risk score trajectory, reflecting the frequency of abnormal risks occurring in the equipment production process. The long-term average risk level is the arithmetic mean of the risk scores of all monitoring windows in the current risk score trajectory, reflecting the overall average risk level in the equipment production process.

[0054] In a preferred embodiment of the present invention, the upper control limit of the risk score is obtained. Risk scores over m consecutive time windows All meet When a situation is considered to indicate that the current production status has significantly deviated from historical normal levels in a statistical sense, the system will output a warning and provide the corresponding chain of evidence. Its terms and conditions are mapped for on-site handling and review. For the residual sequence The aggregated residual statistics, such as the mean or sum of absolute values ​​of the residuals within a window, reflect the magnitude of the overall fit deviation and are important references for comprehensively judging trend anomalies. When no consecutive m windows satisfy the condition... At this time, the system will determine the current window as being in a normal or slightly risky state, without triggering a formal warning, but will only [set the window status to normal]. The corresponding evidence chain is recorded for subsequent trend analysis and threshold calibration. In this way, the evolution of quality risks in the main power grid equipment can be continuously monitored while keeping the false alarm rate under control.

[0055] S6. Monitor the main equipment of the power grid to be monitored according to the equipment quality risk indicators; wherein, the equipment quality risk indicators include the maximum risk value, the proportion of exceeding the limit, and the long-term average risk level.

[0056] To illustrate, based on the generated equipment quality risk indicators, a comprehensive assessment of the quality risks of main power grid equipment is conducted to determine whether there are potential quality hazards in the equipment, providing a basis for decision-making in quality control and intervention.

[0057] In a preferred embodiment of the present invention, taking the production quality risk monitoring of a certain 110kV GIS equipment as an example, the process parameter sequence (including SF6 gas pressure, shell temperature, bolt tightening torque, etc.) during the production process of the equipment is first collected. There are no records of non-compliance in the factory test for this batch of equipment, and the quality risk label is set to 0 (first label). Historical risk data, including process parameters and risk labels, of 50 qualified 110kV GIS equipment are retrieved. A fixed monitoring window length of 10 minutes is set, and the process parameter sequence is sliced ​​to obtain 20 parameter sequence sub-segments. The preset standard value of SF6 gas pressure (process parameter standard value) is 0.6MPa, with a deviation range of 0.06MPa. The deviation coefficient of the pressure sampling values ​​within each sub-segment is calculated, and the number of sampling values ​​with a deviation coefficient greater than 1 is counted as the number of anomalies. The number of anomalies in a certain sub-segment is 3. The median of the pressure sampling values ​​in this sub-segment is calculated, and after robust normalization, a normalized sequence is obtained. The relative deviation is calculated to be 0.4 based on the standard value and the deviation range. A least squares regression model was constructed based on the normalized sequence, and the trend slope was found to be 0.005. The mean and standard deviation of the regression residuals were calculated, and a smoothing coefficient λ=0.2 was taken. An exponentially weighted moving average was calculated on the residuals, and the residual stationarity index was found to be 1.2. A triangular fuzzy function judgment matrix was constructed, and three domain experts scored the data. After summarization and defuzzification, the preset weights were obtained: anomaly frequency weight 0.3, relative deviation weight 0.3, trend slope weight 0.2, and residual stationarity index weight 0.2. The consistency test CR=0.08≤0.1, indicating that the weights were effective. Piecewise linear membership functions were constructed for each indicator: a membership value of 0.3 for 3 anomalies, 0.14 for a relative deviation of 0.4, 0.25 for a trend slope of 0.005, and 0.1 for a residual stationarity index of 1.2. The weighted summation yielded the risk score for this window: 0.3×0.3+0.3×0.14+0.2×0.25+0.2×0.1=0.192. The risk scores of 20 windows were concatenated chronologically to generate the current risk score trajectory. Using the risk data of 50 qualified devices from the past as the historical baseline window, the control limit UCL was calculated to be 0.35. The maximum risk score in the current risk score trajectory was 0.28 (maximum risk value), the number of windows with risk scores exceeding the control limit was 0, the exceedance rate was 0%, and the average risk score of all windows was 0.15 (long-term average risk level). Finally, based on the equipment quality risk indicators, it was determined that the 110kV GIS equipment had a low quality risk and met the production quality requirements.

[0058] By implementing this embodiment, the number of anomalies and relative deviations are calculated based on the process parameter sequence. This not only statistically analyzes explicit anomalies exceeding the baseline but also quantifies the overall parameter deviation, avoiding the omission of minor drifts that, while not exceeding limits, exhibit continuous deviations. By normalizing the sequence, trend slope and residual stationarity indicators are obtained to capture the evolution of parameters over time, rather than judging individual parameter values ​​in isolation, thus achieving more accurate perception of minor drifts. Furthermore, by constructing a piecewise linear membership function, the indicator values ​​are transformed into risk membership degrees, and a continuous risk score is obtained through weighted summation, replacing rigid Boolean judgments. This allows minor drifts to correspond to gradient risk scores, avoiding false alarms. Finally, using historical risk data corresponding to quality risk labels as a benchmark, the current risk score trajectory is aggregated with the corresponding historical risk data over time to generate the maximum risk value, exceedance ratio, and long-term average risk level. This provides a holistic view of equipment quality risk, rather than isolated judgments for each process, further reducing the probability of false alarms caused by minor drifts and significantly improving the accuracy of monitoring main power grid equipment.

[0059] See Figure 2 This is a schematic diagram of the structure of a power grid monitoring device based on the extraction of quality risk indicators of main power grid equipment, according to an embodiment of the present invention, comprising: The power grid main equipment data acquisition module is used to acquire the process parameter sequence, corresponding quality risk labels, and historical risk data of the main equipment to be monitored during the production process. The first indicator calculation module is used to slice the process parameter sequence within a preset monitoring window to obtain several parameter sequence sub-segments, and calculate the number of anomalies and relative deviation of the parameter sequence sub-segments within the monitoring window. The second indicator calculation module is used to normalize the parameter sequence segments to obtain a normalized sequence, perform least squares regression based on the normalized sequence to obtain the trend slope, and perform exponential weighted moving average calculation based on the trend slope and the normalized sequence to obtain the residual stationarity index. The risk score calculation module is used to construct a piecewise linear membership function based on the number of anomalies, the relative deviation, the trend slope, and the residual stationarity index, respectively, and to perform a weighted summation based on the piecewise linear membership function and preset weights to obtain the risk score within each monitoring window, and to generate the current risk score trajectory based on the risk score within each monitoring window. The equipment quality risk index generation module is used to generate equipment quality risk indexes by using the historical risk data corresponding to the quality risk label as a benchmark and performing time aggregation based on the current risk score trajectory and the corresponding historical risk data. The power grid main equipment monitoring module is used to monitor the main equipment of the power grid to be monitored according to the equipment quality risk indicators; wherein, the equipment quality risk indicators include the maximum risk value, the proportion of exceeding the limit, and the long-term average risk level.

[0060] Preferably, the process parameter sequence includes several time parameter data pairs, each time parameter data pair including a timestamp and a corresponding process parameter sample value; The first indicator calculation module is used to slice the process parameter sequence within a preset monitoring window to obtain several parameter sequence sub-segments, including: Based on the preset time offset, the timestamps of each time parameter data pair are mapped to the target time axis, and a monitoring window is constructed on the target time axis according to the type of process parameter sampling value. Based on the preset slicing operator, all time parameter data pairs within each monitoring window are extracted to generate several parameter sequence segments.

[0061] The first indicator calculation module is also used to calculate the number of anomalies and the relative deviation of the parameter sequence segment within the monitoring window, including: For each parameter sequence segment, a deviation coefficient for each process parameter sample value is calculated based on the process parameter sample value and a preset process parameter benchmark value; wherein, the preset process parameter benchmark value includes the standard value of the process parameter and the deviation range. The number of process parameter sample values ​​in each parameter sequence segment whose deviation coefficient is greater than a preset deviation threshold is taken as the number of abnormalities of the parameter sequence segment within the monitoring window. Calculate the median of the process parameter sample values ​​within each parameter sequence segment, and perform robust normalization of the process parameter sample values ​​and the median of the process parameter sample values ​​based on the median and absolute deviation method to obtain normalized sample values; The relative deviation is calculated based on the normalized sampled values, the standard values ​​of process parameters, and the deviation range.

[0062] The second indicator calculation module is used to perform least squares regression based on the normalized sequence to obtain the trend slope, including: Based on the mapped timestamp on the target time axis corresponding to each normalized sample value in the normalized sequence, a target data pair is constructed to characterize the pairing of the mapped timestamp and the normalized sample value. Based on the target data, a least squares regression objective function is constructed, and the trend slope is obtained by solving the least squares regression objective function.

[0063] The second indicator calculation module is also used to calculate the residual stationarity index by performing an exponentially weighted moving average based on the trend slope and the normalized sequence, including: The regression fit value is calculated based on the trend slope, the regression intercept corresponding to the trend slope, and the corresponding mapping timestamp. The regression residuals are calculated based on the regression fitted values ​​and the normalized sequence, and the standard deviation of the regression residuals is determined based on the mean of the regression residuals within the current parameter sequence segment. The stationarity index of the residuals is obtained by calculating the index of stationarity of the residuals through an exponentially weighted moving average based on the regression residuals and the standard deviation of the regression residuals.

[0064] The risk score calculation module is used to construct piecewise linear membership functions for each segment based on the number of anomalies, the relative deviation, the trend slope, and the residual stationarity index. It then performs a weighted summation based on the piecewise linear membership functions and preset weights to obtain the risk score within each monitoring window, including: Based on the number of anomalies, the relative deviation, the trend slope, and the residual stationarity index, membership functions for the number of anomalies, the relative deviation, the trend slope, and the residual stationarity index are constructed respectively based on their corresponding preset upper and lower limits. The risk score for each monitoring window is obtained by weighted summation based on the membership functions of the number of anomalies, relative deviation, trend slope, residual stability index, and preset weights. The preset weights are obtained by defuzzification calculation based on the preset triangular fuzzy function judgment matrix.

[0065] The quality risk label includes a first label for indicating that the quality is acceptable and a second label for indicating that the quality is unacceptable. The equipment quality risk index generation module is used to generate equipment quality risk indicators based on the historical risk data corresponding to the quality risk label, by performing time aggregation on the current risk score trajectory and the corresponding historical risk data, including: Historical risk data labeled with the first quality risk label is used as the first historical sample, the historical time interval corresponding to the first historical sample is used as the historical benchmark window, and the control limits are determined based on the historical benchmark window. The maximum risk score in the current risk score trajectory is taken as the maximum risk value; The ratio of the number of windows with risk scores greater than the control limit in the current risk score trajectory to the total number of monitoring windows is taken as the excess ratio. The average risk score in the current risk score trajectory is taken as the long-term average risk level; The maximum risk value, the proportion of exceeding limits, and the long-term average risk level are used as indicators of equipment quality risk.

[0066] This invention provides a power grid monitoring device based on the extraction of quality risk indicators for main power grid equipment. It can calculate the number of anomalies and relative deviations according to process parameter sequences, statistically analyzing both explicit anomalies exceeding the benchmark and quantifying the overall parameter deviation, avoiding the omission of minor drifts that, while not exceeding limits, exhibit continuous deviations. By normalizing the sequence, it obtains trend slope and residual stationarity indicators, capturing the evolution of parameters over time rather than judging individual parameter values ​​in isolation, thus achieving more accurate perception of minor drifts. Furthermore, by constructing a piecewise linear membership function, it transforms the indicator values ​​into risk membership degrees, and then obtains continuous risk scores through weighted summation, replacing rigid Boolean judgments. This allows minor drifts to correspond to gradient risk scores, avoiding false alarms. Finally, using historical risk data corresponding to quality risk labels as a benchmark, it aggregates the current risk score trajectory with the corresponding historical risk data over time, generating the maximum risk value, the proportion exceeding limits, and the long-term average risk level. This provides a holistic view of equipment quality risk, rather than isolated judgments for each process, further reducing the probability of false alarms caused by minor drifts and significantly improving the accuracy of main power grid equipment monitoring.

[0067] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0068] Those skilled in the art will understand that, for convenience and brevity, the specific working process of the device described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0069] Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements a power grid monitoring method based on the extraction of power grid main equipment quality risk indicators as described in the above embodiments. The terminal device can be a desktop computer, laptop, handheld computer, cloud server, or other computing device. The terminal device may include, but is not limited to, a processor and a memory.

[0070] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0071] The memory can be used to store the computer program. The processor implements various functions of the terminal device by running or executing the computer program stored in the memory and calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function, etc.; the data storage area may store data created based on the use of the mobile phone, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device or other volatile solid-state storage device.

[0072] Another embodiment of the present invention provides a computer-readable storage medium, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the power grid monitoring method based on the extraction of power grid main equipment quality risk indicators as described in the above embodiment.

[0073] The storage medium is a computer-readable storage medium, and the computer program is stored in the computer-readable storage medium. When the computer program is executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0074] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A power grid monitoring method based on the extraction of quality risk indicators from main power grid equipment, characterized in that, include: Acquire the process parameter sequence, corresponding quality risk labels, and historical risk data of the main power grid equipment to be monitored during the production process; The process parameter sequence is sliced ​​within a preset monitoring window to obtain several parameter sequence sub-segments. The number of anomalies and the relative deviation of the parameter sequence sub-segments within the monitoring window are calculated. The parameter sequence segments are normalized to obtain a normalized sequence. Least square regression is performed based on the normalized sequence to obtain the trend slope. An exponentially weighted moving average is then calculated based on the trend slope and the normalized sequence to obtain the residual stationarity index. Based on the number of anomalies, the relative deviation, the trend slope, and the residual stationarity index, a piecewise linear membership function is constructed for each segment. The piecewise linear membership function and a preset weight are used to perform a weighted summation to obtain the risk score within each monitoring window. The current risk score trajectory is generated based on the risk score within each monitoring window. Based on the historical risk data corresponding to the quality risk label, the equipment quality risk index is generated by time aggregation of the current risk score trajectory and the corresponding historical risk data. The main equipment of the power grid to be monitored is monitored according to the equipment quality risk indicators mentioned above; wherein, the equipment quality risk indicators include the maximum risk value, the proportion of exceeding the limit, and the long-term average risk level.

2. The power grid monitoring method based on the extraction of quality risk indicators of main power grid equipment as described in claim 1, characterized in that, The process parameter sequence includes several time parameter data pairs, and each time parameter data pair includes a timestamp and a corresponding process parameter sample value; Based on the process parameter sequence, slices are generated within a preset monitoring window to obtain several parameter sequence sub-segments, including: Based on the preset time offset, the timestamps of each time parameter data pair are mapped to the target time axis, and a monitoring window is constructed on the target time axis according to the type of process parameter sampling value. Based on the preset slicing operator, all time parameter data pairs within each monitoring window are extracted to generate several parameter sequence segments.

3. The power grid monitoring method based on the extraction of quality risk indicators of main power grid equipment as described in claim 2, characterized in that, The calculation of the number of anomalies and the relative deviation of the parameter sequence segments within the monitoring window includes: For each parameter sequence segment, a deviation coefficient for each process parameter sample value is calculated based on the process parameter sample value and a preset process parameter benchmark value; wherein, the preset process parameter benchmark value includes the standard value of the process parameter and the deviation range. The number of process parameter sample values ​​in each parameter sequence segment whose deviation coefficient is greater than a preset deviation threshold is taken as the number of abnormalities of the parameter sequence segment within the monitoring window. Calculate the median of the process parameter sample values ​​within each parameter sequence segment, and perform robust normalization of the process parameter sample values ​​and the median of the process parameter sample values ​​based on the median and absolute deviation method to obtain normalized sample values; The relative deviation is calculated based on the normalized sampled values, the standard values ​​of process parameters, and the deviation range.

4. The power grid monitoring method based on the extraction of quality risk indicators of main power grid equipment as described in claim 3, characterized in that, The trend slope is obtained by performing least squares regression on the normalized sequence, including: Based on the mapped timestamp on the target time axis corresponding to each normalized sample value in the normalized sequence, a target data pair is constructed to characterize the pairing of the mapped timestamp and the normalized sample value. Based on the target data, a least squares regression objective function is constructed, and the trend slope is obtained by solving the least squares regression objective function.

5. A power grid monitoring method based on the extraction of quality risk indicators of main power grid equipment as described in claim 4, characterized in that, Based on the trend slope and the normalized sequence, an exponentially weighted moving average is calculated to obtain the residual stationarity index, including: The regression fit value is calculated based on the trend slope, the regression intercept corresponding to the trend slope, and the corresponding mapping timestamp. The regression residuals are calculated based on the regression fitted values ​​and the normalized sequence, and the standard deviation of the regression residuals is determined based on the mean of the regression residuals within the current parameter sequence segment. The stationarity index of the residuals is obtained by calculating the index of stationarity of the residuals through an exponentially weighted moving average based on the regression residuals and the standard deviation of the regression residuals.

6. The power grid monitoring method based on the extraction of quality risk indicators of main power grid equipment as described in claim 1, characterized in that, Based on the number of anomalies, the relative deviation, the trend slope, and the residual stationarity index, piecewise linear membership functions are constructed respectively. A weighted sum is then performed based on these piecewise linear membership functions and preset weights to obtain the risk score within each monitoring window, including: Based on the number of anomalies, the relative deviation, the trend slope, and the residual stationarity index, membership functions for the number of anomalies, the relative deviation, the trend slope, and the residual stationarity index are constructed respectively based on their corresponding preset upper and lower limits. The risk score for each monitoring window is obtained by weighted summation based on the membership functions of the number of anomalies, relative deviation, trend slope, residual stability index, and preset weights. The preset weights are obtained by defuzzification calculation based on the preset triangular fuzzy function judgment matrix.

7. The power grid monitoring method based on the extraction of quality risk indicators of main power grid equipment as described in claim 1, characterized in that, The quality risk label includes a first label for indicating that the quality is acceptable and a second label for indicating that the quality is unacceptable. Based on the historical risk data corresponding to the aforementioned quality risk labels, and by performing time aggregation on the current risk score trajectory and the corresponding historical risk data, equipment quality risk indicators are generated, including: Historical risk data labeled with the first quality risk label is used as the first historical sample, the historical time interval corresponding to the first historical sample is used as the historical benchmark window, and the control limits are determined based on the historical benchmark window. The maximum risk score in the current risk score trajectory is taken as the maximum risk value; The ratio of the number of windows with risk scores greater than the control limit in the current risk score trajectory to the total number of monitoring windows is taken as the excess ratio. The average risk score in the current risk score trajectory is taken as the long-term average risk level; The maximum risk value, the proportion of exceeding limits, and the long-term average risk level are used as indicators of equipment quality risk.

8. A power grid monitoring device based on the extraction of quality risk indicators from main power grid equipment, characterized in that, include: The power grid main equipment data acquisition module is used to acquire the process parameter sequence, corresponding quality risk labels, and historical risk data of the main equipment to be monitored during the production process. The first indicator calculation module is used to slice the process parameter sequence within a preset monitoring window to obtain several parameter sequence sub-segments, and calculate the number of anomalies and relative deviation of the parameter sequence sub-segments within the monitoring window. The second indicator calculation module is used to normalize the parameter sequence segments to obtain a normalized sequence, perform least squares regression based on the normalized sequence to obtain the trend slope, and perform exponential weighted moving average calculation based on the trend slope and the normalized sequence to obtain the residual stationarity index. The risk score calculation module is used to construct a piecewise linear membership function based on the number of anomalies, the relative deviation, the trend slope, and the residual stationarity index, respectively, and to perform a weighted summation based on the piecewise linear membership function and preset weights to obtain the risk score within each monitoring window, and to generate the current risk score trajectory based on the risk score within each monitoring window. The equipment quality risk index generation module is used to generate equipment quality risk indexes by using the historical risk data corresponding to the quality risk label as a benchmark and performing time aggregation based on the current risk score trajectory and the corresponding historical risk data. The power grid main equipment monitoring module is used to monitor the main equipment of the power grid to be monitored according to the equipment quality risk indicators; wherein, the equipment quality risk indicators include the maximum risk value, the proportion of exceeding the limit, and the long-term average risk level.

9. A terminal device, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement a power grid monitoring method based on the extraction of quality risk indicators of main power grid equipment as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform a power grid monitoring method based on the extraction of power grid main equipment quality risk indicators as described in any one of claims 1 to 7.

Images