Steady-state feature extraction system for power equipment based on load perception

Abstract

The present invention relates to the technical field of load management of electric power equipment, and specifically to a steady-state feature extraction system for station electric power equipment based on load perception. The system comprises: a load data processing module obtains load data based on the operating status of the electric power equipment, extracts correlation features between parameters by analyzing the interaction effects between each type of load data, identifies the periodic features and abnormal fluctuations of the load data, and obtains a standard load data set. In the present invention, more accurate and comprehensive steady-state feature extraction of electric power equipment can be achieved by analyzing and processing multi-dimensional parameters of load data. At the same time, by calculating the interaction effects of load data, the correlation between parameters can be better identified and analyzed, ensuring data consistency and improving the accuracy of feature extraction. In data quantile analysis, by constructing data subsets of different quantiles and extracting parameter features under steady-state conditions, subtle differences in equipment under different steady-state conditions can be captured.

Description

Technical Field

[0001] The present invention relates to the technical field of load management of electric power equipment, and in particular to a steady-state feature extraction system for station electric equipment based on load perception. Background Art

[0002] The technical field of power equipment load management mainly involves the load monitoring, analysis and management of various equipment in the power system. By collecting and analyzing the real-time and historical data of power equipment, the safe, stable and efficient operation of the power system is ensured. The field includes real-time monitoring of power load, load forecasting, optimal scheduling and equipment health status assessment to avoid problems such as overload and unbalanced operation of power equipment.

[0003] Among them, the main purpose of the steady-state feature extraction system of power equipment in the station is to collect and analyze the load characteristics of the electrical equipment in the station to extract the characteristic parameters of the equipment under steady-state operation. These characteristic parameters can be used for equipment status monitoring, anomaly detection and equipment health management, thereby helping managers to better understand and control the operating status of power equipment, realize intelligent and precise management of power equipment, and ensure the safe, stable and efficient operation of key power facilities such as substations.

[0004] In the existing technology, subtle characteristic changes of equipment under complex steady-state conditions are easily overlooked in the status monitoring of power equipment. For example, in the case of frequent load fluctuations, it may be difficult to accurately distinguish normal fluctuations from potential fault warnings, resulting in false alarms or missed fault detection, which in turn affects the accuracy of maintenance decisions. Single-dimensional analysis methods can only capture surface changes in equipment parameters, which is not conducive to identifying potential abnormalities of equipment under different operating conditions, making it difficult to detect equipment degradation trends at an early stage. When processing dynamic data of power equipment, it is difficult for existing technologies to adjust according to real-time changes in equipment status at any time. For example, when load parameters of power equipment fluctuate due to external factors such as temperature changes or fluctuations in power demand, the model may not be able to effectively adapt to these changes, resulting in a lag in parameter adjustment, thereby reducing the real-time and accuracy of monitoring. It is not conducive to the system to detect potential equipment degradation or fault hazards in a timely manner, and increases the risk of sudden equipment failures. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings existing in the prior art and to propose a steady-state feature extraction system for station electrical equipment based on load perception.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: A station power equipment steady-state feature extraction system based on load perception includes:

[0007] The load data processing module obtains load data based on the operating status of the power equipment, extracts the correlation characteristics between parameters by analyzing the interaction effects between each type of load data, identifies the periodic characteristics and abnormal fluctuations of the load data, and obtains a standard load data set;

[0008] The steady-state characteristic quantile analysis module selects quantile values ​​as analysis targets according to the standard load data set, constructs data subsets for each quantile value, extracts voltage and current parameters in the quantile subset for modeling, analyzes the changes in regression coefficients at each quantile, and generates a steady-state characteristic performance matrix;

[0009] The load state clustering module extracts the key eigenvalues ​​of the load data under various operating conditions in the steady-state characteristic performance matrix, selects the initial position of the cluster center, groups the key eigenvalues ​​according to the similarity measurement, updates the cluster center for each group of key eigenvalues, and obtains the load state clustering result;

[0010] The adaptive feature extraction module builds a feature extraction model based on the load state clustering results and the standard load data set according to the current load signal and the current load data, refers to the key eigenvalue information of the steady-state feature performance matrix, judges the current feature change trend, dynamically adjusts the parameters of the feature extraction model, and obtains an adaptive steady-state feature extraction result.

[0011] As a further solution of the present invention, the step of obtaining the correlation features between the extraction parameters is specifically as follows:

[0012] Obtain load data based on the operating status of the power equipment, including equipment voltage, current, frequency, temperature and operating time parameters, normalize the data, map the values ​​of each type of parameter to the same interval, and perform standardization to generate a processed load data set;

[0013] Based on the processed load data set, the formula is used:

[0014] ;

[0015] Calculation parameters and Correlation coefficient , generate parameter correlation analysis results;

[0016] in, is a time sample, is the total number of samples, It's time Time load parameters The normalized data, It's time Time load parameters The normalized data, is the load parameter The average value of is the load parameter The average value of

[0017] Based on the parameter correlation analysis results, mutually related parameter pairs in the load data are extracted, the current load state is compared with the previous operation data of the electrical equipment, the characteristic changes of the related parameters are analyzed, and the correlation feature set between the load parameters is established.

[0018] As a further solution of the present invention, the steps of identifying the periodic characteristics and abnormal fluctuations of load data are specifically as follows:

[0019] According to the associated feature set between the load parameters, the continuous data of the equipment voltage, current, frequency, temperature and operating time parameters are sorted by time point, and the fluctuation of the parameters is gradually identified, and the change direction, amplitude and time interval of each parameter are analyzed to generate trend analysis results for each type of load data;

[0020] According to the trend analysis results of each type of load data, the previous load data is compared with the current data to identify the periodic changes in the data, record the start time and duration of the fluctuation, and generate a standard load data set.

[0021] As a further solution of the present invention, the steps of extracting the voltage and current parameters in the quantile subset for modeling are specifically as follows:

[0022] Based on the load parameter data in the standard load data set, the abnormal fluctuation part is distinguished from the normal fluctuation data, and quantile division is performed, and the subset data are saved respectively to generate load data subsets of multiple quantiles;

[0023] According to the load data subsets of the multiple quantiles, the voltage and current parameters are divided into data by time period, the data within each quantile is gradually fitted, and the error is adjusted to construct a quantile model;

[0024] Based on the data within each quantile in the quantile model, the formula is used:

[0025] ;

[0026] Calculate regression coefficients , get the updated regression coefficient;

[0027] in, is the current regression coefficient, is the learning rate, is the total number of data points, Based on the current regression coefficient The predicted current value obtained is It is The actual current value of the data point, It is The voltage value of each data point.

[0028] As a further solution of the present invention, the step of obtaining the change of the analytical regression coefficient at each quantile is specifically as follows:

[0029] Based on the updated regression coefficient, according to the voltage and current data corresponding to each type of quantile, the data is recorded and arranged in chronological order, and the fluctuation of the regression coefficient under each type of quantile is compared and analyzed to generate the change data of the regression coefficient;

[0030] Based on the change data of the regression coefficient, the voltage and current correlation characteristics under each type of quantile are extracted, and a comparative analysis is performed, the characteristic differences between the quantiles are recorded, and a steady-state characteristic performance matrix is ​​generated.

[0031] As a further solution of the present invention, the step of obtaining the key feature values ​​into groups according to the similarity measurement is specifically as follows:

[0032] Based on the steady-state characteristic performance matrix, key characteristic values ​​in voltage, current, frequency, temperature and operating time load data are extracted, and key parameters are screened by obtaining operating data in each time period. Key parameters are characteristic values ​​that change significantly under load conditions, and the parameters are arranged in chronological order to generate a key characteristic data set under load conditions;

[0033] Based on the key characteristic data set under the load conditions described, the formula is used:

[0034] ;

[0035] Calculate features and The Euclidean distance between , get the similarity data between features;

[0036] in, , Representation characteristics and In the The value of the parameter dimension, Indicates the total number of parameter dimensions.

[0037] As a further solution of the present invention, the steps of obtaining the load status clustering result are specifically as follows:

[0038] According to the similarity data between the features, the voltage, current, frequency, temperature and operating time load parameters of the equipment are compared, the similarity of the feature values ​​is determined through a set similarity threshold, and the similarity matching feature values ​​are grouped into the same group to generate a feature grouping result;

[0039] According to the feature grouping results, the feature values ​​in each group are analyzed step by step, and the initial cluster center of each group is selected. By comparing the feature values ​​in each group, the position of the cluster center is adjusted, and the process is repeated until the cluster center is stable to obtain the load state clustering result.

[0040] As a further solution of the present invention, the step of obtaining the current feature change trend is specifically as follows:

[0041] Based on the load state clustering results and the standard load data set, voltage fluctuation, load current change rate, frequency deviation and temperature change rate are extracted, and the features are compared one by one with the feature values ​​in the previous data set in a time series manner, and a difference analysis is performed to generate a transient feature model of the current load state;

[0042] Based on the data of the transient characteristic model of the current load state, the formula is adopted:

[0043] ;

[0044] Calculate the given load signal , load fluctuation frequency and time lag effects Next, parameters The posterior distribution of , get the parameter value in the posterior distribution;

[0045] in, represents the load sensing parameter, Indicates the load signal characteristics of the current equipment. represents the frequency characteristics of load fluctuations, represents the time lag effect in load switching, represents a probability distribution;

[0046] Based on the parameter values ​​in the posterior distribution, the real-time monitored voltage fluctuations, load current change rate and other characteristics are called up, and compared and analyzed with the key characteristic parameters of the equipment in normal operation or stable state. The parameters in the model are adjusted according to the comparison results, and the adaptive steady-state feature extraction results are obtained by iteratively optimizing the feature model.

[0047] Compared with the prior art, the advantages and positive effects of the present invention are:

[0048] In the present invention, by analyzing and processing the multidimensional parameters of load data, more accurate and comprehensive steady-state feature extraction of power equipment can be achieved. At the same time, by calculating the interaction effect of load data, the correlation between parameters can be better identified and analyzed, data consistency can be ensured, and the accuracy of feature extraction can be improved. In data quantile analysis, by constructing data subsets of different quantiles and extracting parameter features under steady-state conditions, the subtle differences of equipment under different steady-state conditions can be captured. Not only the sensitivity of feature recognition is improved, but also the performance of equipment under different operating conditions can be more comprehensively reflected. By measuring the similarity and clustering analysis of multiple key feature values, the grouping and classification of equipment operating status can be achieved, and the health status of equipment under different operating conditions can be accurately identified, so that status monitoring and prediction can be performed more effectively. Combined with the dynamic feature modeling of load data, the parameters of the feature extraction model are adjusted in real time, and the ability to adapt is possessed, so that equipment monitoring can respond to changes in equipment status in a timely manner, and the accuracy of anomaly detection is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 is a system flow chart of the present invention;

[0050] Figure 2 A flow chart for extracting correlation features between parameters of the present invention;

[0051] Figure 3 A flow chart for identifying periodic characteristics and abnormal fluctuations of load data according to the present invention;

[0052] Figure 4 A flow chart for modeling the voltage and current parameters extracted from the quantile subsets of the present invention;

[0053] Figure 5 A flow chart of analyzing the change of regression coefficient under each quantile in the present invention;

[0054] Figure 6 A flow chart of the present invention for grouping key feature values ​​according to similarity measurement;

[0055] Figure 7 It is a flow chart of the load state clustering result of the present invention;

[0056] Figure 8 This is a flow chart of the present invention for determining the current feature change trend. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0058] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0059] See also Figure 1 ,The station power equipment steady-state feature extraction system based on load perception includes:

[0060] The load data processing module obtains load data based on the operating status of the power equipment, including multiple parameters such as equipment voltage, current, frequency, temperature, and operating time, normalizes the data, calculates the interaction effect between each type of load data, extracts the correlation characteristics between parameters, performs trend analysis on the load data, identifies the periodic characteristics and abnormal fluctuations of the load data, and obtains a standard load data set;

[0061] The steady-state characteristic quantile analysis module selects quantile values ​​as analysis targets based on the standard load data set and the marked abnormal fluctuation characteristics. The quantile values ​​include 10%, 50%, and 90% values. A data subset is constructed for each quantile value, and the voltage and current parameters in the quantile subset are extracted for modeling. The regression coefficient of each quantile model is calculated, and the changes in the regression coefficient under each quantile are analyzed to determine the differences in characteristic performance between quantiles and generate a steady-state characteristic performance matrix.

[0062] The load state clustering module extracts the key eigenvalues ​​of the load data under various operating conditions in the steady-state feature performance matrix. The key eigenvalues ​​are one or more of the equipment voltage, current, frequency, temperature or operating time parameters. The initial position of the cluster center is selected, and the Euclidean distance between the key eigenvalues ​​is calculated to determine the similarity. The key eigenvalues ​​are grouped according to the similarity measure, and the features with high similarity are grouped into the same group. The cluster center is updated for each group of key eigenvalues ​​until the cluster center is stable and no longer changes, and the load state clustering result is obtained;

[0063] The adaptive feature extraction module is based on the load state clustering results and the standard load data set. According to the current load signal characteristics, the load signal characteristics are the voltage fluctuation, load current change rate, frequency deviation, and temperature change rate of the equipment during operation, which can reflect the dynamic characteristics of the instantaneous operation state of the equipment. The module builds a feature extraction model with the current load data, refers to the key eigenvalue information of the steady-state feature performance matrix, calculates the parameter value in the posterior distribution, judges the current feature change trend, and dynamically adjusts the parameters of the feature extraction model to obtain the adaptive steady-state feature extraction results.

[0064] The standard load data set includes normalized voltage data set, current data set, frequency data set, temperature data set and operating time data set. The steady-state feature performance matrix includes voltage regression coefficient and current regression coefficient. The 10% quantile is used to identify the steady-state characteristics of the equipment under low voltage or low current conditions, the 50% quantile is used to represent the typical steady-state characteristics of the equipment, and the 90% quantile reflects the steady-state characteristics of the equipment under high voltage or high current conditions. The load state clustering results include high similarity feature groups, medium similarity feature groups and low similarity feature groups. The high similarity feature group refers to the stable load state of the equipment under similar working conditions, while the low similarity feature group refers to the characteristics with large differences in working conditions and significant load state fluctuations. The adaptive steady-state feature extraction results include the changing trend of the load signal characteristics and the corresponding posterior distribution parameters.

[0065] See also Figure 2 , the steps for obtaining the associated features between the extracted parameters are as follows:

[0066] Obtain load data based on the operating status of the power equipment, including equipment voltage, current, frequency, temperature and operating time parameters, normalize the data, map the values ​​of each type of parameter to the same interval, and perform standardization to generate a processed load data set;

[0067] First, the data at each time point is collected in real time through the equipment sensors, including voltage sensors, current transformers, temperature sensors, and frequency meters. The collected raw data is preliminarily processed by the data acquisition system. Subsequently, the maximum and minimum normalization method is used to standardize these data, and the data values ​​of each type of load parameter are compressed to the [0,1] interval. The specific operation includes determining the maximum and minimum values ​​in the collected historical data, which are used as the upper and lower limits of normalization, and finally generating normalized voltage, current, frequency and other data sets. Through this normalization process, the calculation errors caused by units and dimensions of different parameters are eliminated, so that subsequent data analysis can be compared and analyzed at the same scale. The normalized data set becomes the basis for analyzing the load status.

[0068] Based on the processed load data set, the formula is used:

[0069] ;

[0070] Calculation parameters and Correlation coefficient , generate parameter correlation analysis results;

[0071] in, Used to indicate load parameters and The linear correlation between the two, the range of the correlation coefficient value is [1,1]. The closer the value is to 1, the stronger the positive correlation between the two; close to 1 means negative correlation; close to 0 means no obvious linear relationship. is a time sample, representing the load data at a specific moment, is the total number of samples, indicating the number of time points, It's time Time load parameters The normalized data is collected by sensors and normalized. It's time Time load parameters The normalized data is collected by sensors and normalized. is the load parameter The average value is calculated as: , is the load parameter The average value is calculated as: .

[0072] If the voltage and current data of the device are normalized, the normalized data at three time points in a certain day are: voltage data : [0,0.5,1], current data : [0.2, 0.5, 0.8].

[0073] Voltage The average value of:

[0074] ;

[0075] Current The average value of:

[0076] ;

[0077] Calculate the numerator part:

[0078] ;

[0079] Calculate the denominator:

[0080] ;

[0081] in:

[0082] ;

[0083] ;

[0084] So the denominator is:

[0085] ;

[0086] Calculate the correlation coefficient:

[0087] ;

[0088] Result description:

[0089] The correlation coefficient is 1, indicating that in these three sets of normalized data, there is a completely positive correlation between voltage and current, indicating that the change trend of the load voltage is completely consistent with that of the current.

[0090] Based on the results of parameter correlation analysis, mutually correlated parameter pairs in the load data are extracted, the current load state is compared with the previous operation data of the power equipment, the characteristic changes of the associated parameters are analyzed, and the associated characteristic set between the load parameters is established;

[0091] Based on the calculated correlation coefficient , firstly, the normalized data of parameters such as voltage, current, frequency, temperature and running time are analyzed for correlation. Through correlation analysis methods (such as Pearson correlation coefficient analysis method), each group of load parameters is compared in pairs to obtain the correlation between them. In the specific operation, the change trend of these parameters such as voltage, current, temperature, frequency and so on in different time periods is analyzed to generate a correlation matrix. In the process of correlation feature extraction, the normalized data set is first called to extract the correlation features of each pair of parameters according to the time series. For example, the voltage data and the current data are matched to observe the trend of their change over time, and the correlation between them is further compared through the correlation analysis tool to extract the parameter pairs with high correlation. At the same time, the normalized data of the temperature and frequency of the equipment are called to compare the change patterns of these data under different load conditions, analyze whether the temperature increase is accompanied by frequency fluctuations, and obtain their fluctuation synchronization. Through the correlation analysis method, highly correlated parameter pairs are extracted and recorded in the system, marked as highly correlated parameter groups. In this process, the historical operation data of the equipment can be used as a comparative reference to analyze the change relationship between the parameters under the current load state. Ultimately, the acquired correlation features can help further optimize the equipment's operating status monitoring and provide early warning basis for abnormal fluctuations in equipment load changes.

[0092] See also Figure 3 ,The specific steps for identifying the periodic characteristics and abnormal fluctuations of load data are:

[0093] According to the associated feature set between load parameters, the continuous data of equipment parameters such as voltage, current, frequency, temperature and operating time are sorted by time points, and the fluctuation of parameters is gradually identified. The change direction, amplitude and time interval of each parameter are analyzed to generate trend analysis results for each type of load data;

[0094] First, continuous data of parameters such as equipment voltage, current, frequency, temperature and operating time are obtained. Then, based on the time series information of these parameters, the time series analysis method is called to gradually identify the fluctuation of each parameter. In this process, the load data is sorted by time point, and the change direction, change amplitude and time interval of each load parameter are analyzed, including the analysis of the synchronization of voltage and current fluctuations, and whether the changes in temperature and frequency have significant correlation. When executing the operation, by comparing the differences in data within continuous time periods, the key change nodes are extracted, and the trend of load data within a specific time period is recorded. Finally, the trend analysis results of each type of load data are obtained, providing important data support for the subsequent analysis of the long-term stability of equipment operation.

[0095] Based on the trend analysis results of each type of load data, compare the previous load data with the current data, identify the periodic changes in the data, record the start time and duration of the fluctuation, and generate a standard load data set;

[0096] First, historical load data is called for comparison with current data. The periodic detection method is used to gradually identify periodically changing load parameters by extracting repetitive fluctuation patterns that appear in the load data. During operation, the system compares data in multiple time periods and focuses on whether parameters such as voltage, current, and frequency show similar fluctuation patterns within a fixed time interval. At the same time, the anomaly detection algorithm is called to analyze sudden fluctuations during equipment operation, filter out data segments with large changes in a short period of time, and record the start time and duration of these abnormal fluctuations. After execution, the identification results of periodic characteristics and abnormal fluctuations are integrated into a standard load data set, which is used to support daily monitoring of equipment loads and provide key references for load optimization and management.

[0097] See also Figure 4 , the specific steps for extracting the voltage and current parameters in the quantile subset for modeling are:

[0098] Based on the load parameter data in the standard load data set, the abnormal fluctuation part is distinguished from the normal fluctuation data, and quantile division is performed, and the subset data is saved separately to generate load data subsets of multiple quantiles;

[0099] First, a data set containing load parameters such as voltage, current, frequency and temperature is extracted from the standard load data set, and the marked abnormal fluctuation part is screened. The abnormal fluctuation period is marked as a high fluctuation load area and processed separately from the normal fluctuation data in the standard load data. On this basis, the 10%, 50% and 90% quantile values ​​are selected respectively. For each quantile value, the voltage and current data are divided into quantiles through data processing. Each quantile value will classify different parts of the data to form a corresponding load data subset. During the execution process, each subset will be stored according to the corresponding quantile value mark and serve as the basis for subsequent analysis, and finally multiple quantile data subsets containing different load states are obtained.

[0100] According to the load data subsets of multiple quantiles, the voltage and current parameters are divided into data by time period, the data within each quantile is gradually fitted, and the error is adjusted to build a quantile model;

[0101] First, extract the time series data of voltage and current from each quantile subset, and pass these data into the model. The model will process them separately according to the quantile interval of the data. In specific operations, the system will divide the data by time period, and perform data fitting on the voltage and current parameters of different time periods. In this process, a step-by-step fitting method is adopted to gradually optimize the fitting of the data within each quantile. In detail, the system will start from the local trend of the data and use the linear fitting algorithm to gradually fit the fluctuations of voltage and current. First, the linear relationship in a short time period is processed, and then the time window is expanded to ensure the accuracy of the fitting effect within the entire quantile interval. In order to ensure the accuracy of the modeling process, the system will also make adjustments based on the fitting error. If the fitting error of a part of the data exceeds the set threshold, the system will process the part more carefully, or select other fitting methods, such as polynomial fitting or spline curve fitting, until the error meets the requirements. Finally, the system generates the corresponding voltage and current model, which records the corresponding relationship between voltage and current, as well as the parameters adjusted during the fitting process. After the execution of the operation, these models will be stored in the form of data files, and the models record the fitting parameters of voltage and current at each quantile.

[0102] Based on the data within each quantile in the quantile model, the formula is used:

[0103] ;

[0104] Calculate regression coefficients , get the updated regression coefficient;

[0105] in, represents the linear relationship between voltage and current, is the current regression coefficient, used for iterative update, is the learning rate, which is the step size parameter in the gradient descent method. It is used to control the step size of each update and is usually selected by experience, such as 0.01, 0.001, etc. A larger learning rate may cause the model to skip the optimal solution, while a smaller learning rate helps to gradually converge to the optimal solution. is the sample mean operation, is the total number of data points, indicating the number of observed data, such as the number of voltage and current data points collected in a specific period of time, obtained through real-time monitoring data of the equipment. Based on the current regression coefficient The predicted current value is calculated by the formula: The predicted value is calculated by regression model based on the current voltage data. The estimated current value, It is The actual current value of each data point comes from the load monitoring data of the equipment, reflecting the real current value under load conditions. It is The voltage value of each data point is collected in real time during the operation of the equipment and represents the actual observed value of the load voltage.

[0106] For example, three sets of historical load data are obtained from the station's electrical equipment: voltage data :[210,220,230]V, current data :[4.5,5.0,5.5]A, set the initial regression coefficient , learning rate , total number of samples , calculate the predicted current value:

[0107] ;

[0108] ;

[0109] ;

[0110] Calculation error:

[0111] ;

[0112] ;

[0113] ;

[0114] Compute the gradient and update the regression coefficients:

[0115] ;

[0116] ;

[0117] ;

[0118] The updated regression coefficient by gradient descent is 1.36, indicating a strong positive linear relationship between voltage and current. Specifically, under this load condition, the current increases by about 1.36A for every 1V increase in voltage.

[0119] See also Figure 5 , the steps to obtain the changes in the regression coefficient at each quantile are as follows:

[0120] Based on the updated regression coefficient, according to the voltage and current data corresponding to each quantile, the data is recorded and arranged in chronological order, and the fluctuation of the regression coefficient under each quantile is compared and analyzed to generate the change data of the regression coefficient;

[0121] Based on the calculated regression coefficient of each quantile model, first obtain the voltage and current data corresponding to the 10%, 50%, and 90% quantiles. For different load conditions, record these data, arrange them in chronological order, and observe their fluctuations in different time periods one by one by comparing the regression coefficients in each quantile model. During the execution process, compare and analyze the performance differences of the regression coefficients under each quantile, and check whether the regression coefficients under each quantile remain stable or whether there are large fluctuations. For the data in different time periods, focus on identifying the significant changes in the regression coefficients under certain load conditions, mark the time points when the regression coefficients fluctuate greatly, and record the relevant voltage and current parameters. This process helps to analyze the changing characteristics of the relationship between voltage and current under different load levels and determine which load conditions will cause significant fluctuations in the regression coefficients. By completing these steps, the changing rules of the regression coefficients under different quantiles are obtained.

[0122] Based on the change data of the regression coefficient, the voltage and current correlation characteristics under each quantile are extracted, and a comparative analysis is performed to record the characteristic differences between quantiles and generate a steady-state characteristic performance matrix;

[0123] First, from the regression coefficient change data obtained from the previous analysis, the voltage and current correlation characteristics at 10%, 50%, and 90% quantiles are extracted, and the parameter performance at different time periods at each quantile is compared. During the execution process, the voltage and current data corresponding to the quantiles are normalized one by one to ensure that the data between different quantiles are comparable, and then the voltage and current change trends at different quantiles are correlated and analyzed. By analyzing the voltage and current fluctuations at each quantile, the differences in these fluctuations under different load conditions are observed, and the characteristic change points between each quantile are marked and recorded, especially the impact of load status on voltage and current changes. In the process of constructing the steady-state characteristic performance matrix, the voltage and current data of each quantile are first classified and divided into multiple time intervals in chronological order to ensure that these data are continuous and consistent in different time periods. Subsequently, in each time period, the key parameters of voltage and current are extracted, such as fluctuation range, trend direction, and parameter stability. By classifying these key parameters, the voltage and current performance of each quantile is organized into a matrix form, each column represents the characteristics of a quantile, and each row represents the voltage and current performance in different time periods. Finally, these characteristic data are stored, and after the matrix is ​​constructed, they are used for subsequent equipment operation monitoring and load status analysis to support the extraction of steady-state characteristics and long-term monitoring.

[0124] See also Figure 6 , the specific steps for obtaining the key feature values ​​into groups according to the similarity measure are:

[0125] Based on the steady-state characteristic performance matrix, the key characteristic values ​​in the voltage, current, frequency, temperature and operating time load data are extracted. By obtaining the operating data in each time period, the key parameters are screened. The key parameters are the characteristic values ​​that change significantly under load conditions, and the parameters are arranged in chronological order to generate the key characteristic data set under load conditions;

[0126] First, based on the parameters such as voltage, current, frequency, temperature and operating time in the steady-state characteristic performance matrix, the key characteristic values ​​that best reflect the operating status of the equipment under different load conditions are screened out. During the execution process, the key characteristic data in each time period is first obtained from the matrix, and the representative operating parameters are selected. For example, under load conditions with large voltage and current changes, the fluctuation range of these characteristic values ​​is recorded and arranged in chronological order. By comparing the changing trends of these parameters under different load conditions, the key characteristics that can reflect the load status of the equipment are found. During the execution process, the various parameters are grouped under different operating conditions to ensure that the correlation between these characteristic values ​​can be accurately recorded. Then, for each group of characteristic data, the position of the initial clustering center is determined. Through data analysis methods, Kmeans or other clustering algorithms are used to select the initial clustering center. In the specific operation, the initial center point of each type of feature is first determined. For example, the moment when the voltage and current fluctuations are the largest during the operation of the equipment is selected as the initial center, and this center position is used as the starting point for subsequent clustering. Subsequently, other similar data are assigned to the group closest to the center point.

[0127] Based on the key characteristic data set under load conditions, the formula is used:

[0128] ;

[0129] Calculate features and The Euclidean distance between , get the similarity data between features;

[0130] in, It is used to measure the similarity between two eigenvalues ​​in the multidimensional feature space. The smaller the distance, the more similar the features are; the larger the distance, the more significant the difference in features is. , Representation characteristics and In the The parameter dimension here can be voltage, current, frequency, temperature or running time, which is monitored and recorded in real time by the device sensor under different load conditions. For example, the voltage parameter is obtained by the voltage sensor, and the temperature parameter is monitored by sensors such as thermocouples. Indicates the total number of parameter dimensions, which refers to how many key parameters are considered in the calculation. Common parameters include voltage, current, frequency, temperature, and operating time. The number and type of these parameters are set according to the specific load sensing application scenario.

[0131] For example, two sets of historical data were obtained from the station's electrical equipment and , these two sets of data correspond to the voltage, temperature and operating time parameters under different load conditions:

[0132] Group The parameters are: (voltage, unit: V), (Temperature, unit: °C), (Running time, unit: hours).

[0133] Group The parameters are: (voltage, unit: V), (Temperature, unit: °C), (Running time, unit: hours).

[0134] Calculate the square of the voltage difference:

[0135] ;

[0136] Calculate the square of the temperature difference:

[0137] ;

[0138] Compute the square of the difference in running times:

[0139] ;

[0140] Add up all the differences and take the square root:

[0141] ;

[0142] The result is the Euclidean distance It shows that the characteristics of the three parameters of voltage, temperature and running time are and Features There are certain differences between them.

[0143] See also Figure 7 , the steps to obtain the load state clustering results are as follows:

[0144] According to the similarity data between features, the voltage, current, frequency, temperature and operating time load parameters of the equipment are compared, and the similarity of the feature values ​​is judged through the set similarity threshold, and the similarity matching feature values ​​are classified into the same group to generate the feature grouping result;

[0145] Based on the Euclidean distance calculated in the previous step First, a similarity threshold is selected to determine the grouping criteria. Assuming that the selected threshold is 5, the similarity between each pair of eigenvalues ​​is checked one by one through the threshold criteria during execution. If the similarity between two eigenvalues ​​is high (that is, their Euclidean distance is less than or equal to the threshold 5), they are classified into the same group. In order to perform this operation, the data of load parameters such as voltage, current, frequency, temperature and operating time of the equipment need to be compared one by one, and then the similarity degree of the eigenvalues ​​is evaluated by the similarity measurement method. The data sorting and screening in the grouping process depends on the historical operating data obtained under the operating conditions of the equipment. These data come from actual collection in different time periods and load conditions. Through the similarity grouping operation, the task of grouping eigenvalues ​​with high similarity into the same group is finally completed, and the eigenvalues ​​in each group show a relatively consistent form of expression under the load state.

[0146] According to the characteristic grouping results, the characteristic values ​​in each group are analyzed step by step, and the initial cluster center of each group is selected. By comparing the characteristic values ​​in each group, the position of the cluster center is adjusted. The process is repeated until the cluster center is stable, and the load status clustering result is obtained;

[0147] The initial cluster center can usually be determined based on the median or average value of all features in each group. In order to update the cluster center of each group, all eigenvalues ​​are first scanned within each group to obtain the data information of the eigenvalues, and then a new center point is selected based on this information. When performing the update operation, the position of the center point is readjusted by comparing the eigenvalues ​​in each group with the current cluster center to ensure that the new center can better represent all features in the group. To this end, it is necessary to observe the changing trend of each group of features, set a new center point based on its fluctuation range, and repeat this process until the center point of each group no longer changes significantly. Through multiple center update operations, a stable cluster center position is finally obtained, and the load state characteristics of each group can be accurately described.

[0148] See also Figure 8 , the specific steps for determining the current feature change trend are:

[0149] Based on the load state clustering results and the standard load data set, voltage fluctuation, load current change rate, frequency deviation and temperature change rate are extracted, and the features are compared one by one with the feature values ​​in the previous data set in a time series manner, and difference analysis is performed to generate a transient feature model of the current load state;

[0150] According to the current load signal characteristics, such as voltage fluctuation, load current change rate, frequency deviation and temperature change rate during equipment operation, these characteristics are compared and analyzed with the clustered load state results through the previously acquired real-time load data. During the execution process, the voltage and frequency deviation values, load current change rate and temperature fluctuation rate obtained by the monitoring equipment are used to call the clustered load data set for cross-matching, and these dynamic characteristics are analyzed one by one according to the real-time change trend of the equipment. By comparing historical data, the difference between the current load state and the historical steady-state characteristics is identified. Then, based on these differences, the feature extraction model is constructed. First, by analyzing the key eigenvalues ​​in the dynamic feature and steady-state feature performance matrix, the input parameters of the model are selected, including the instantaneous change rate of voltage, current, frequency and temperature. Next, the multi-layer structure of the model is designed to decompose the load state into several dimensions, each dimension corresponding to a key feature. During the construction process, each feature dimension is dynamically updated through the monitoring feedback of real-time data. Subsequently, the difference evaluation method is used to set the weight and priority of each feature. Through multiple iterative adjustments, the model can accurately reflect the instantaneous state of the equipment. Once completed, the feature extraction model is able to capture transient operational changes of the equipment and provide accurate feature value output for subsequent analysis and monitoring.

[0151] Based on the data of the transient characteristic model of the current load state, the formula is used:

[0152] ;

[0153] Calculate the given load signal , load fluctuation frequency and time lag effects Next, parameters The posterior distribution of , get the parameter value in the posterior distribution;

[0154] in, Indicates load sensing parameters, reflecting the operating characteristics of the equipment. The parameters may include voltage, current, frequency and other steady-state characteristic information of multiple equipment during operation, which are monitored and recorded by the equipment's sensors. Indicates the load signal characteristics of the current equipment, such as voltage fluctuation, load current change rate, etc. These characteristics are extracted from the power-consuming equipment in real time through load sensing, and can reflect the transient operating status of the equipment at the current time point. The frequency characteristic of load fluctuation is used to describe the frequency of load fluctuation during the operation of the equipment. The load fluctuation frequency can be obtained by monitoring the rate of load change and periodic changes. The specific method is as follows: First, during the operation of the equipment, the load signal The frequency sensor can be used to obtain the periodic fluctuation frequency of the signal. Assuming that the load voltage fluctuation period is 10 seconds, the frequency If there is no frequency sensor, data analysis can also be used to perform Fourier transform on the load signal data within a certain time range to find the most significant frequency component, which is the load fluctuation frequency. It represents the time lag effect in load conversion, reflecting the time required for the equipment to transition from one load state to another during the load conversion process. The time lag effect is calculated by the response speed of the equipment load state change. The calculation method is to obtain it based on the difference between the starting point of the load change and the time point of the equipment steady-state response. For example, if the load of the equipment starts to change at 10 seconds and reaches a new steady state at 15 seconds, the time lag effect is Second, Symbols representing probability distributions, indicating that all probability calculations in the formula are related to load signals and parameters Related, here, represents the likelihood function, given the parameters , load fluctuation frequency and time lag effects When the load signal The probability of occurrence is calculated by comparing historical data with the current load signal of the equipment, and the historical records of the operation of the station's electrical equipment are used to estimate these likelihood values. represents the joint prior distribution, describing the load fluctuation frequency Lower Parameters This joint distribution is usually based on the long-term historical operation data of the equipment. The characteristics of the equipment and the load fluctuation are extracted through long-term equipment operation monitoring data. The prior value can be established through statistical data of multiple equipment load fluctuations. Indicates load signal The evidence value, or marginal likelihood, indicates the load signal observed when all parameters are taken The overall probability of the load state is obtained by statistical methods from the steady-state characteristic performance matrix and the load state standard data set.

[0155] For example, to monitor the load perception of the station's electrical equipment, the current real-time monitoring data is: load signal (voltage) and (current), load fluctuation frequency , the period of voltage fluctuation obtained by the frequency sensor is 10 seconds, so , time lag of load transfer , indicating that it takes 3 seconds for the equipment load to reach a new steady state after the load changes.

[0156] By analyzing the historical data of the equipment and the current load status, the following probabilities are obtained: , indicating that the probability of observing the load signal under the current load fluctuation frequency and time lag is 0.78, and the joint prior distribution , indicating that the equipment operating parameters are under the current load fluctuation frequency The probability is 0.68, and the evidence value , indicating the load signal in all cases The probability of occurrence is 0.85.

[0157] Substituting the values ​​into the formula:

[0158] ;

[0159] The results show that the posterior distribution value of the current equipment is 0.624, which combines the load fluctuation frequency and time lag effects. In addition, if the posterior distribution value under the historical steady state is between 0.6-0.7, it means that the current system is still within the steady state range and the equipment is operating normally. If the value is much lower than the benchmark value, it may mean that the equipment has abnormal fluctuations or unstable load conversion under the current load state, which requires further analysis and adjustment. According to the size of the posterior distribution value, the change trend of the equipment characteristics can be judged as follows: if the posterior distribution value gradually increases, it means that the load state tends to be steady and the equipment characteristics are constantly stabilizing; if the posterior distribution value continues to decrease, especially far below the benchmark value, it means that the equipment characteristics fluctuate greatly and the load state deviates from the steady state, which may indicate equipment failure or unstable operation.

[0160] Based on the parameter values ​​in the posterior distribution, the voltage fluctuation, load current change rate and other characteristics monitored in real time are called and compared with the key characteristic parameters of the equipment in normal operation or stable state. The parameters in the model are adjusted according to the comparison results, and the adaptive steady-state feature extraction results are obtained by iteratively optimizing the feature model.

[0161] Dynamically adjust the parameters of the feature extraction model, and compare the real-time data extracted by the monitoring equipment with the historical steady-state feature values ​​based on the currently extracted voltage fluctuations, load current change rate, frequency deviation, and temperature change rate. During the execution process, the model parameters are gradually adjusted so that it can accurately reflect the current load state of the equipment. In the specific operation, the fluctuation amplitude of voltage and current is monitored in time series, and the real-time fluctuation is compared with the steady-state value in the feature extraction model, and the corresponding parameters in the model are modified according to the comparison results. Combined with the current frequency deviation and temperature change rate of the equipment, the feature model is iteratively optimized multiple times until the model can accurately reflect the current operating state of the equipment. After completing the parameter adjustment, the model can adapt to the instantaneous feature changes of the equipment and generate adaptive steady-state feature extraction results for further load state monitoring and equipment operation evaluation.

[0162] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A station power equipment steady-state feature extraction system based on load perception, characterized in that: The system comprises: The load data processing module obtains load data based on the operating status of the power equipment, extracts the correlation characteristics between parameters by analyzing the interaction effects between each type of load data, identifies the periodic characteristics and abnormal fluctuations of the load data, and obtains a standard load data set; The steady-state characteristic quantile analysis module selects quantile values ​​as analysis targets according to the standard load data set, constructs data subsets for each quantile value, extracts voltage and current parameters in the quantile subset for modeling, analyzes the changes in regression coefficients at each quantile, and generates a steady-state characteristic performance matrix; The load state clustering module extracts the key eigenvalues ​​of the load data under various operating conditions in the steady-state characteristic performance matrix, selects the initial position of the cluster center, groups the key eigenvalues ​​according to the similarity measurement, updates the cluster center for each group of key eigenvalues, and obtains the load state clustering result; The adaptive feature extraction module builds a feature extraction model based on the load state clustering result and the standard load data set according to the current load signal characteristics and the current load data, refers to the key eigenvalue information of the steady-state feature performance matrix, determines the current feature change trend, dynamically adjusts the parameters of the feature extraction model, and obtains the adaptive steady-state feature extraction result; Based on the load parameter data in the standard load data set, the abnormal fluctuation part is distinguished from the normal fluctuation data, and quantile division is performed, and the subset data are saved respectively to generate load data subsets of multiple quantiles; According to the load data subsets of the multiple quantiles, the voltage and current parameters are divided into data by time period, the data within each quantile is gradually fitted, and the error is adjusted to construct a quantile model; Based on the data within each quantile in the quantile model, the formula is used: ； Calculate regression coefficients , get the updated regression coefficient; in, is the current regression coefficient, is the learning rate, is the total number of data points, Based on the current regression coefficient The predicted current value obtained is is the actual current value of the kth data point, is the voltage value of the kth data point; Based on the updated regression coefficient, according to the voltage and current data corresponding to each type of quantile, the data is recorded and arranged in chronological order, and the fluctuation of the regression coefficient under each type of quantile is compared and analyzed to generate the change data of the regression coefficient; Based on the change data of the regression coefficient, the voltage and current correlation characteristics under each type of quantile are extracted, and a comparative analysis is performed, the characteristic differences between the quantiles are recorded, and a steady-state characteristic performance matrix is ​​generated.

2. The station power equipment steady-state feature extraction system based on load perception according to claim 1 is characterized in that: The steps of obtaining the correlation features between the extraction parameters are specifically as follows: Obtain load data based on the operating status of the power equipment, including equipment voltage, current, frequency, temperature and operating time parameters, normalize the data, map the values ​​of each type of parameter to the same interval, and perform standardization to generate a processed load data set; Based on the processed load data set, the formula is used: ； Calculate the correlation coefficient between parameters i and j , generate parameter correlation analysis results; Where t is the time sample, n is the total number of samples, is the normalized data of load parameter i at time t, is the normalized data of load parameter j at time t, is the average value of load parameter i, is the average value of load parameter j; Based on the parameter correlation analysis results, mutually related parameter pairs in the load data are extracted, the current load state is compared with the previous operation data of the electrical equipment, the characteristic changes of the related parameters are analyzed, and the correlation feature set between the load parameters is established.

3. The station power equipment steady-state feature extraction system based on load perception according to claim 2 is characterized in that: The steps of obtaining the periodic characteristics and abnormal fluctuations of the load data are specifically as follows: According to the associated feature set between the load parameters, the continuous data of the equipment voltage, current, frequency, temperature and operating time parameters are sorted by time point, and the fluctuation of the parameters is gradually identified, and the change direction, amplitude and time interval of each parameter are analyzed to generate trend analysis results for each type of load data; According to the trend analysis results of each type of load data, the previous load data is compared with the current data to identify the periodic changes in the data, record the start time and duration of the fluctuation, and generate a standard load data set.

4. The station power equipment steady-state feature extraction system based on load perception according to claim 1 is characterized in that: The step of obtaining the key feature values ​​into groups according to the similarity measurement is specifically as follows: Based on the steady-state characteristic performance matrix, key characteristic values ​​in the voltage, current, frequency, temperature and operating time load data are extracted, and key parameters are screened by obtaining the operating data in each time period. The key parameters are characteristic values ​​that change significantly under load conditions, and the parameters are arranged in chronological order to generate a key characteristic data set under load conditions; Based on the key characteristic data set under the load conditions described, the formula is used: ； Calculate the Euclidean distance between features p and q , get the similarity data between features; in, , It represents the value of features p and q in the rth parameter dimension, and m represents the total number of parameter dimensions.

5. The station power equipment steady-state feature extraction system based on load perception according to claim 4 is characterized in that: The steps for obtaining the load status clustering result are specifically as follows: According to the similarity data between the features, the voltage, current, frequency, temperature and operating time load parameters of the equipment are compared, the similarity of the feature values ​​is determined through a set similarity threshold, and the similarity matching feature values ​​are grouped into the same group to generate a feature grouping result; According to the feature grouping results, the feature values ​​in each group are analyzed step by step, and the initial cluster center of each group is selected. By comparing the feature values ​​in each group, the position of the cluster center is adjusted, and the process is repeated until the cluster center is stable to obtain the load state clustering result.

6. The station power equipment steady-state feature extraction system based on load perception according to claim 5 is characterized in that: The steps for determining the current feature change trend are specifically as follows: Based on the load state clustering results and the standard load data set, voltage fluctuation, load current change rate, frequency deviation and temperature change rate are extracted, and the features are compared one by one with the feature values ​​in the previous data set in a time series manner, and a difference analysis is performed to generate a transient feature model of the current load state; Based on the data of the transient characteristic model of the current load state, the formula is adopted: ； Calculate the load signal F and load fluctuation frequency and time lag effects Next, parameters The posterior distribution of , get the parameter value in the posterior distribution; in, represents the load sensing parameter, F represents the load signal characteristics of the current device, represents the frequency characteristics of load fluctuations, represents the time lag effect in load conversion, Q represents the probability distribution; Based on the parameter values ​​in the posterior distribution, the real-time monitored voltage fluctuations, load current change rate and other characteristics are called up, and compared and analyzed with the key characteristic parameters of the equipment in normal operation or stable state. The parameters in the model are adjusted according to the comparison results, and the adaptive steady-state feature extraction results are obtained by iteratively optimizing the feature model.

Images