Online user-transformer relation identification method and system based on multi-source data fusion

Through the online household change relationship identification method of multi-source data fusion, the problem of insufficient grid abnormal identification in the prior art is solved, the refined management of the power system is realized, and the stability of the power system and the user service level are improved.

CN119089249APending Publication Date: 2024-12-06YANTAI POWER SUPPLY COMPANY OF STATE GRID SHANDONG ELECTRIC POWER
View PDF 0 Cites 4 Cited by

Patent Information

Application Number
CN202411271472.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-11
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

The existing power system monitoring technology lacks sufficiently fine data processing methods when processing huge data sets, making it difficult to identify short-term variations and tiny grid abnormalities, resulting in insufficient fault prediction and untimely response, which affects the operating efficiency and energy allocation of the power system. Fairness.

Method used

The online household-variable relationship recognition method based on multi-source data fusion is adopted. By capturing the power frequency voltage curve and signal-to-noise ratio data, data formatting and noise filtering are carried out, time features are extracted, short-term variation is identified, and threshold analysis is set. Use grouping algorithms and supervised learning framework to generate a household-change relationship prediction model to realize real-time data analysis and identification.

Benefits of technology

Improve data quality, timely detect and predict power grid abnormalities, improve the accuracy and operational efficiency of household change relationship identification, enhance the accuracy of power distribution and system energy efficiency management, improve the overall stability of the power system and user services level.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119089249A_ABST
    Figure CN119089249A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of power system monitoring, in particular to an online user change relation identification method and system based on multi-source data fusion, and the method comprises the following steps: capturing a power frequency voltage curve, a power frequency synchronization sequence and signal-to-noise ratio data, carrying out the data formatting and noise filtering, carrying out the standardization processing of the filtered data, and carrying out the recognition of the user change relation. A standardized feature data set is generated. According to the method, the electric power data is captured and formatted, noise is filtered out, standardization processing is executed, the data quality and usability are improved, short-time variation is accurately recognized through a difference method, careful monitoring on dynamic changes of an electric power system is enhanced, and abnormal behaviors are revealed in time through deviation analysis of threshold value setting; effective sorting and classification of abnormal points are achieved, key support is provided for real-time monitoring and fault prediction of a power grid, the accuracy and efficiency of user-transformer relation recognition are improved through automatic classification and stability inspection, power distribution and energy efficiency management are optimized, and the stability of a power system and the user service level are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of power system monitoring, and in particular to an online household-to-transformer relationship identification method and system based on multi-source data fusion. Background Art

[0002] The field of power system monitoring technology involves real-time or non-real-time tracking, recording and analysis of various parameters in the power system to ensure stable operation and efficient management of the power system. The field includes the measurement of voltage, current, frequency and power, as well as more advanced functions such as fault detection, energy efficiency management and load forecasting. Power system monitoring technology also focuses on extracting useful information from huge amounts of data, using modern information technology and communication means to achieve intelligent control and automated management of the power grid. The development of this technology field has promoted the transition from traditional power grids to smart grids, which include highly automated power grids with advanced measurement equipment, communication networks and data processing systems.

[0003] Among them, the online household-transformer relationship identification method is a specific application in the field of power system monitoring technology. It uses data collected from power lines to determine the connection relationship between meters and transformers. The main purpose of this method is to improve the accuracy and efficiency of power distribution, reduce energy losses, and help power companies better manage and maintain power grid equipment. By automatically identifying household-transformer relationships, power companies can quickly locate specific power usage points and optimize load management and fault response, which is crucial to maintaining the stability of the power grid and improving user service levels.

[0004] Although existing technologies cover functions such as voltage, current, frequency and power measurement and fault detection, they lack sufficiently sophisticated data processing methods to identify short-term variations and minor grid anomalies when processing large data sets. This limitation leads to deficiencies in real-time data analysis and fault prediction, and cannot effectively identify and prevent minor distortions or anomalies that cause large-scale impacts. In terms of automated management and intelligent control, existing technologies often still rely on traditional threshold and rule response mechanisms and lack dynamic adaptive processing capabilities. This will result in insufficient or inaccurate responses when dealing with complex and changing grid environments, thereby affecting the operating efficiency of the power system and the fairness of energy distribution, leading to delayed fault responses, and increasing system recovery time and costs. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose an online household change relationship identification method and system based on multi-source data fusion.

[0006] In order to achieve the above object, the present invention adopts the following technical solution, which is an online household change relationship identification method based on multi-source data fusion, comprising the following steps:

[0007] S1: Capture power frequency voltage curve, power frequency synchronization sequence, signal-to-noise ratio data, format data and filter noise, perform standardization on the filtered data, and generate a standardized feature data set;

[0008] S2: Based on the standardized feature data set, extract the time features of multiple types of data, apply the difference method to identify the short-term variation in voltage and current, summarize the distortion features, and generate a distortion time series feature set;

[0009] S3: performing deviation analysis on the features in the distorted time series feature set, setting thresholds and identifying out-of-range data points, recording abnormal points and sorting their occurrence frequency and intensity, and generating an abnormal behavior feature point set;

[0010] S4: Based on the abnormal behavior feature point set, a grouping algorithm is applied to classify the data points, each category corresponds to the household change relationship in the substation area, the stability and repeatability of the classification results are verified, and a household change relationship grouping model is generated;

[0011] S5: using the household change relationship grouping model, selecting a feature set to train a prediction model in a supervised learning framework, performing cross-validation and optimizing output, and generating a household change relationship prediction model;

[0012] S6: deploy the household change relationship prediction model to perform real-time data analysis, perform classification prediction on newly accessed data, update the prediction results in real time, and continuously monitor the model performance to generate real-time household change recognition results.

[0013] As a further solution of the present invention, the standardized feature data set includes voltage time characteristics, current time characteristics, and short-time variation indicators; the distortion time series feature set includes variation starting point, variation ending point, and variation amplitude; the abnormal behavior feature point set includes data point deviation level, data point frequency interval, and data point intensity score; the household change relationship grouping model includes classification standards, grouping effect consistency indicators, and repeatability verification results; the household change relationship prediction model includes feature selection accuracy, prediction accuracy, and model response time; the real-time household change identification results include new data matching degree, prediction update frequency, and performance monitoring results.

[0014] As a further solution of the present invention, the power frequency voltage curve, the power frequency synchronization sequence, and the signal-to-noise ratio data are captured, and data formatting and noise filtering are performed, and the filtered data is subjected to standardization processing, and the steps of generating a standardized feature data set are specifically as follows:

[0015] S101: Capture power frequency voltage curve, power frequency synchronization sequence, signal-to-noise ratio data, perform data format correction, adjust data fields, and obtain formatted voltage synchronization data;

[0016] S102: according to the formatted voltage synchronization data, filter parameters are set to identify and eliminate noise, filter configuration for data processing is adjusted, and processed synchronization data is obtained;

[0017] S103: numerically adjusting the processed synchronous data, setting a standardized scale, applying numerical transformation to process the standardized data, and generating a standardized feature data set.

[0018] As a further solution of the present invention, based on the standardized feature data set, the time features of multiple types of data are extracted, the short-term variations in voltage and current are identified by using a differential method, the distortion features are summarized, and the steps of generating a distortion time series feature set are specifically as follows:

[0019] S201: According to the standardized characteristic data set, each data is timestamped through the voltage and current data sets, and the date and time of the data are recorded, including reading the original record of the data, extracting and attaching time information to each data point, and generating a time-stamped data set;

[0020] S202: performing differential analysis based on the time-stamped data set using a moving window difference method, calculating the voltage and current differences between consecutive data points point by point, identifying short-time variation points, marking and recording the short-time variation points, and generating a distortion record set;

[0021] S203: Arrange the distortion events in chronological order of occurrence through the distortion record set, record the voltage and current distortion conditions at each time point during the monitoring period, and generate a distortion time series feature set.

[0022] As a further solution of the present invention, the formula of the moving window difference method is as follows:

[0023]

[0024] Among them, Δy represents the short-term variation of voltage and current, α represents the weight of voltage change, β represents the weight of current change, γ represents the normalization factor, and t window Represents the time length of the moving window, V t and V t+1 Indicates the voltage reading within two seconds. t+1 and I t Indicates the current reading for two consecutive seconds.

[0025] As a further solution of the present invention, the steps of performing deviation analysis on the features in the distorted time series feature set, setting thresholds and identifying out-of-range data points, recording abnormal points and sorting the frequency and intensity of occurrence, and generating an abnormal behavior feature point set are specifically as follows:

[0026] S301: extracting each data point from the time series feature set, setting a monitoring threshold, comparing the value of each data point with the set threshold, marking the point exceeding the threshold as abnormal, identifying the abnormality through parameter comparison, and obtaining an abnormal data point set;

[0027] S302: According to the abnormal data point set, count the occurrence frequency of each abnormal point, measure the degree of deviation from the normal value, sort the data points according to the frequency and the degree of deviation, organize the data points into an ordered abnormal list, and obtain sorted abnormal data;

[0028] S303: Based on the sorted abnormal data, combined with the recorded frequency and deviation degree information, the characteristics of the abnormal points are reflected, and a set of abnormal behavior feature points is generated by integrating the information.

[0029] As a further solution of the present invention, according to the abnormal behavior feature point set, a grouping algorithm is applied to classify the data points, each category corresponds to the household change relationship in the substation area, and the stability and repeatability of the classification result are verified. The steps of generating a household change relationship grouping model are specifically as follows:

[0030] S401: based on the abnormal behavior feature point set, group the data points, analyze the behavior pattern of each data point, preliminarily classify the data, classify the representative household change relationship, and obtain the classified data set;

[0031] S402: Performing a stability test according to the classification data set, including performing consistency and repeatability tests on the categories in the classification data set, and generating a stability test record by comparing the results of multiple tests;

[0032] S403: Based on the stability test record, record the data points of each category and the corresponding household change relationship, determine the structure and classification accuracy of the model, and generate a household change relationship grouping model.

[0033] As a further solution of the present invention, the household change relationship grouping model is used to select a feature set to train a prediction model in a supervised learning framework, perform cross-validation and optimize the output, and generate a household change relationship prediction model in the following steps:

[0034] S501: According to the household change relationship grouping model, a feature set is selected from a data set, initialization parameters of the model are adjusted to match the target supervised learning framework requirements, and a preliminary training model is obtained by adjusting the training configuration and starting the model training;

[0035] S502: Based on the preliminary training model, perform cross-validation settings, determine the ratio and number of iterations of validation segmentation, perform cyclic testing on the model and evaluate the prediction performance, regulate the validation process, and generate a validation optimization model;

[0036] S503: Based on the verification optimization model, perform model tuning, including fine-tuning of learning rate and regularization parameters, control the output quality and performance of the model, refine the prediction capability of the model through parameter adjustment, and generate a household change relationship prediction model.

[0037] As a further solution of the present invention, the household change relationship prediction model is deployed to perform real-time data analysis, classify and predict the newly accessed data, update the prediction results in real time, and continuously monitor the model performance. The steps of generating real-time household change identification results are specifically as follows:

[0038] S601: Deploy the model according to the household change relationship prediction model, configure the data access interface, manage the real-time data stream by adjusting the receiving buffer size and the data access frequency, and obtain the data access configuration;

[0039] S602: Based on the data access configuration, adjust the prediction model and process the real-time data, set the data batch processing size and the data update interval, and iteratively update the prediction results by adjusting the parameters to obtain the real-time prediction update results;

[0040] S603: Based on the real-time prediction update results, performance monitoring parameters are set, including error tracking and adjustment of warning thresholds. By continuously monitoring the performance and response time of the model, the model is adjusted to maintain optimal performance, the model's responsiveness and prediction accuracy are optimized, and real-time household change identification results are generated.

[0041] An online household change relationship identification system based on multi-source data fusion, the online household change relationship identification system based on multi-source data fusion is used to execute the online household change relationship identification method based on multi-source data fusion, the system comprises:

[0042] The data capture and formatting module captures the power frequency voltage curve, power frequency synchronization sequence, and signal-to-noise ratio data, performs data standardization, removes random noise in the data, performs data alignment, and generates a standardized feature data set;

[0043] The time feature extraction module extracts time series data from the standardized feature data set, performs time window segmentation on the voltage and current data, uses differential technology to identify short-term variations in multiple windows, records and marks the variations, and obtains a distortion time series feature set;

[0044] The abnormal behavior analysis module statistically analyzes the distorted time series feature set, identifies abnormal data points, sets an abnormal detection threshold, and sorts and records the abnormal data points according to the frequency and intensity of occurrence to form an abnormal behavior feature point set;

[0045] The household change relationship grouping module uses the abnormal behavior feature point set to classify the data and divide the data into groups, where the groups represent household change relationship categories, and the classified data is stability verified to generate a household change relationship grouping model;

[0046] The prediction model deployment module is based on the household change relationship grouping model, applies the supervised learning method to train the prediction model, evaluates the model performance through cross-validation, optimizes the parameters, deploys the model on the real-time data analysis platform, continuously updates the data prediction, and constructs real-time household change identification results.

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

[0048] In the present invention, by capturing, formatting, filtering out noise and performing standardized processing on power data, the data quality and availability are effectively improved, making it more accurate to extract key features from the data. Further, by identifying short-term variations through differential methods, the dynamic changes of the power system can be monitored at a more subtle level. By using threshold settings for deviation analysis, not only can abnormal behaviors be discovered in a timely manner, but also abnormal points can be sorted and classified, thereby providing more accurate support for real-time monitoring and fault prediction of the power grid. Through automated classification and stability testing, the accuracy and operational efficiency of identifying household-to-transformer relationships are improved, thereby directly enhancing the accuracy of power distribution and the energy efficiency management of the system. The introduction of such refined management methods improves the overall stability of the power system and the level of user service. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a schematic diagram of the workflow of the present invention;

[0050] Figure 2 This is a detailed flow chart of S1 of the present invention;

[0051] Figure 3 This is a detailed flow chart of S2 of the present invention;

[0052] Figure 4 This is a detailed flow chart of S3 of the present invention;

[0053] Figure 5 This is a detailed flow chart of S4 of the present invention;

[0054] Figure 6 This is a detailed flow chart of S5 of the present invention;

[0055] Figure 7 This is a detailed flow chart of S6 of the present invention;

[0056] Figure 8 It is a system flow chart of the present invention. 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 present invention provides a technical solution, an online household change relationship identification method based on multi-source data fusion, comprising the following steps:

[0060] S1: Capture power frequency voltage curve, power frequency synchronization sequence, signal-to-noise ratio data, format data and filter noise, perform standardization on the filtered data, and generate a standardized feature data set;

[0061] S2: Based on the standardized feature data set, the time features of each data type are extracted, the short-term variations in voltage and current are identified using the differential method, the variations are recorded as distortion features, and the distortion features are summarized to generate a distortion time series feature set;

[0062] S3: Perform deviation analysis on the features in the distorted time series feature set, set a threshold to determine if the data point exceeds the normal range, record the abnormal points, sort the frequency and intensity of the abnormal points, and generate an abnormal behavior feature point set;

[0063] S4: Based on the abnormal behavior feature point set, execute the grouping algorithm to classify the data points. Each category represents the target household change relationship of the substation area. Perform stability test on the classification results, verify the consistency and repeatability of the grouping, and generate a household change relationship grouping model;

[0064] S5: Make the user change relationship grouping model, set the supervised learning framework to select the matching feature set to train the prediction model, cross-validate the model and test the prediction performance, perform model tuning and optimize the result output, and generate the user change relationship prediction model;

[0065] S6: Deploy the household change relationship prediction model to perform real-time data analysis, classify and predict the newly accessed data, update the prediction results in real time, and continuously monitor the model performance to generate real-time household change identification results.

[0066] The standardized feature data set includes voltage time characteristics, current time characteristics, and short-time variation indicators. The distortion time series feature set includes variation starting point, variation ending point, and variation amplitude. The abnormal behavior feature point set includes data point deviation level, data point frequency interval, and data point intensity score. The household change relationship grouping model includes classification standards, grouping effect consistency indicators, and repeatability verification results. The household change relationship prediction model includes feature selection accuracy, prediction accuracy, and model response time. The real-time household change identification results include new data matching degree, prediction update frequency, and performance monitoring results.

[0067] Specifically, if Figure 2 As shown, the power frequency voltage curve, power frequency synchronization sequence, and signal-to-noise ratio data are captured, and data formatting and noise filtering are performed. The filtered data is standardized to generate a standardized feature data set. Specifically, the steps are as follows:

[0068] S101: Capture the power frequency voltage curve, power frequency synchronization sequence, signal-to-noise ratio data, perform data format correction, adjust the data field, and obtain the formatted voltage synchronization data. The execution process is as follows;

[0069] Capture power frequency voltage curve, power frequency synchronization sequence, and signal-to-noise ratio data. First, perform time synchronization verification on the captured raw data to ensure that the data timing is accurate. Use the preset data format correction tool to standardize the format of each data, including unified data encoding, adjustment of sampling frequency, and correction of time scale deviation. Specifically, it includes removing outliers from voltage curve data, interpolating and completing lost frame data in power frequency synchronization sequence, and performing sliding average processing on signal-to-noise ratio data to smooth short-term fluctuations. After completing data preprocessing, adjust data fields, such as merging related data columns and splitting different data sources, to obtain formatted voltage synchronization data. This data set will be directly used in subsequent advanced data analysis and feature extraction steps.

[0070] S102: According to the formatted voltage synchronization data, filter parameters are set to identify and eliminate noise, and the filter configuration for data processing is adjusted to obtain the processed synchronization data. The execution flow is as follows;

[0071] According to the formatted voltage synchronization data, the filtering parameters are first set, including selecting the appropriate filter type (such as low-pass, high-pass or band-pass filter) and determining its cut-off frequency to adapt to different signal-to-noise ratio conditions and data characteristics, and noise identification is performed. By analyzing the spectral characteristics of the data, non-power frequency noise sources such as random electromagnetic interference and the electronic noise of the equipment itself are identified. Using the set filtering parameters, the filter configuration of data processing is adjusted. Through the iterative optimization process, it is ensured that the impact of noise on the synchronization data is minimized while keeping the core characteristics of the data intact. The process includes adjusting the filter parameters multiple times, testing the filtering effect, and evaluating the quality of the filtered data through visual and statistical methods to obtain the processed synchronization data. After removing most of the interference noise, the data will be more suitable for accurate data analysis and monitoring of the power grid status.

[0072] S103: numerically adjust the processed synchronous data, set the standardized scale, apply numerical transformation to process the standardized data, and generate the standardized feature data set. The execution flow is as follows;

[0073] The processed synchronized data is numerically adjusted, and the standardized scale is set according to the actual distribution characteristics of the data. The steps involve calculating the mean and standard deviation of the data, and then using statistical parameters to convert the data to a standard normal distribution to facilitate comparison and analysis across data sets. The standardized data is processed by applying numerical transformations, including linear or nonlinear mapping of the data to ensure that the converted data better reflects the operating status of the original power grid. During the standardization process, the skewness and kurtosis of the data are also checked, and appropriate mathematical transformations, such as logarithmic transformation or Box-Cox transformation, are performed when necessary to improve the statistical properties of the data. After the processing is completed, a standardized feature data set is generated, which will include all necessary power grid operation characteristics and a unified format to facilitate subsequent machine learning training or in-depth data analysis.

[0074] Specifically, if Figure 3 As shown in the figure, based on the standardized feature data set, the time features of multi-type data are extracted, the difference method is applied to identify the short-term variation in voltage and current, the distortion features are summarized, and the steps of generating the distortion time series feature set are as follows:

[0075] S201: According to the standardized characteristic data set, each data is timestamped through the voltage and current data sets, and the date and time of the data are recorded, including reading the original record of the data, extracting and attaching the time information to each data point, and the execution flow of generating the time-stamped data set is as follows;

[0076] According to the standardized characteristic data set, each data in the voltage and current data set is first extracted and timestamped, including parsing the time field in the data set to ensure that the time information of each data point is accurate, synchronizing the receiving or recording time of each data with the server time to correct the time error caused by system delay, and continuing to automatically read the original data records through the program, including the log files of sensors, the database output of the monitoring system, or the real-time data read directly from the device. The data is deeply analyzed to extract key time elements, such as year, month, day, hour, minute, and second, and the elements are attached to each data point to form a complete time tag and generate a time-stamped data set. Each data point in this data set will include an accurate time tag to ensure that the specific time of each data point can be accurately traced back in subsequent analysis, which is helpful for time series analysis, event association, and pattern recognition.

[0077] S202: According to the time-stamped data set, a moving window difference method is used to perform a difference analysis, the voltage and current differences between consecutive data points are calculated point by point, and short-time variation points are identified, and the short-time variation points are marked and recorded to generate a distortion record set. The execution process is as follows;

[0078] The formula for the moving window difference method is as follows:

[0079]

[0080] Among them, Δy represents the short-term variation of voltage and current, α represents the weight of voltage change, β represents the weight of current change, γ represents the normalization factor, and t window Represents the time length of the moving window, V t and V t+1 Indicates the voltage reading within two seconds. t+1 and I t Indicates the current reading for two consecutive seconds.

[0081] This formula is used to identify short-term variations in voltage and current in the power grid, where voltage and current data are collected in real time through the power grid monitoring system, and data points are collected once every second.

[0082] Detailed explanation of the formula and the process of formula calculation and derivation:

[0083] V t and V t+1 Represents the voltage reading within two consecutive seconds. The voltage at time t is set to 110V, and the voltage at time t+1 is 115V;

[0084] I t and I t+1 Represents the current reading within two consecutive seconds. The current at time t is set to 5A, and the current at time t+1 is 5.5A;

[0085] α and β are the weights of voltage and current changes. Based on the grid operation stability standard and previous data analysis, α=0.6 and β=0.4 are set. These two weights are obtained through sensitivity analysis of historical data and reflect the proportion of the impact of voltage and current changes on the overall system.

[0086] γ is a factor used for normalization and is set to 1.2 to ensure consistency of the results regardless of changes in the window size;

[0087] t window is the size of the moving window. Since the data is collected once per second, t is set here. window = 1 second to adapt to the rapidly changing power grid environment;

[0088] Enter specific values ​​for calculation:

[0089]

[0090] The calculated result Δy≈2.67 represents the variation of voltage and current within a one-second time window. This is the difference obtained based on real-time monitoring data, which reflects the instability of the power grid during this period. It can identify and record short-term variations in the power grid, and provide data support for further analysis and response measures. The larger the value, the more significant the short-term variation, which requires further attention and possible intervention.

[0091] S203: Arrange the distortion events in the order of occurrence through the distortion record set, record the voltage and current distortion at each time point during the monitoring period, and generate the distortion time series feature set. The execution process is as follows;

[0092] Through the distortion record set, the records including voltage and current distortion data are analyzed in detail, and the distortion events are arranged in the chronological order of occurrence. This requires access to the timestamp information in each event record and sorting algorithm processing to ensure that the data is presented in chronological order. The key execution steps include verifying the accuracy of the timestamp, handling any possible timestamp errors or omissions, and filling in data gaps. The voltage and current distortion conditions at each time point during the monitoring period are recorded, including the amplitude, duration and scope of the distortion. The collection and analysis of data will support fault diagnosis and system health monitoring, and generate a distortion time series feature set. The feature set not only reflects a single distortion event, but also forms a continuous, time-related panoramic view of the grid operation distortion, providing data support for the optimization, regulation and preventive maintenance of the grid.

[0093] Specifically, if Figure 4As shown in the figure, the deviation analysis is performed on the features in the distorted time series feature set, the threshold is set and the out-of-range data points are identified, the abnormal points are recorded and the frequency and intensity are sorted, and the steps for generating the abnormal behavior feature point set are as follows:

[0094] S301: extract each data point from the time series feature set, set a monitoring threshold, compare the value of each data point with the set threshold, mark the point exceeding the threshold as abnormal, identify the abnormality through parameter comparison, and obtain the execution process of the abnormal data point set as follows;

[0095] Each data point is extracted from the time series feature set, and detailed data analysis is performed on each point to determine whether it conforms to the expected grid behavior. First, the monitoring threshold is set. The threshold is set based on historical data analysis, industry standards, and expected grid performance. For example, upper and lower thresholds are set for voltage values ​​to capture voltage conditions that are too high or too low. A specially developed data processing program is used to compare the value of each data point with the set threshold. During the implementation process, it is completed through an automated script, which checks the data point by point to see if it exceeds the threshold and marks the points that exceed the threshold as abnormal. Through parameter comparison, not only a single anomaly is identified, but also the context in which the anomaly occurs is analyzed, such as whether it is a continuous anomaly or an isolated event. The resulting set of abnormal data points will be recorded and used for further analysis and fault diagnosis to facilitate subsequent maintenance and preventive measures.

[0096] S302: According to the abnormal data point set, count the occurrence frequency of each abnormal point, measure the degree of deviation from the normal value, sort the data points according to the frequency and the degree of deviation, and organize the data points into an ordered abnormal list. The execution process of obtaining the sorted abnormal data is as follows;

[0097] The frequency of occurrence of anomalies is calculated according to the formula:

[0098]

[0099] Calculate the abnormal frequency, where F i Represents the frequency of occurrence of the i-th abnormal point, N i represents the number of times the i-th outlier appears in the data set, and N represents the total number of data points in the data set;

[0100] In order to count the frequency of outliers, we first need to count the number of times each outlier appears in the data set. i , which is done by traversing each data point in the data set and checking whether each point meets the definition of an outlier. The number of occurrences of each outlier point is divided by the total number of data points N in the data set to obtain the frequency of occurrence of each outlier point F i ,This process ensures that the frequency calculation is based on actual ,observations rather than preset values;

[0101] Suppose the data set has a total of 1000 data points, and a specific outlier appears 50 times. Then the frequency of occurrence of the outlier is F. i The calculation is as follows:

[0102]

[0103] This means that the frequency of occurrence of this outlier is 5%.

[0104] S303: According to the sorted abnormal data, combined with the recorded frequency and deviation degree information, the characteristics of the abnormal points are reflected, and the execution process of generating the abnormal behavior feature point set by integrating the information is as follows;

[0105] According to the sorted abnormal data, the frequency and deviation degree information of each abnormal point record are comprehensively considered, and the characteristics of the abnormal points are further analyzed. The process involves advanced data integration technology. Through data fusion and information extraction, the interpretation ability and practical value of abnormal data are enhanced. For example, associating high-frequency abnormal points with large deviation degrees can reveal the most critical potential problems in the power grid. It not only reflects the specific situation of each abnormal point, but also shows its relative importance in the operation of the power grid. The integrated information will be used to generate a set of abnormal behavior feature points, which will provide key data support for power grid operation stability analysis, fault prediction and system optimization.

[0106] Specifically, if Figure 5 As shown in the figure, according to the abnormal behavior feature point set, the data points are classified by applying the grouping algorithm. Each category corresponds to the household change relationship in the substation area. The stability and repeatability of the classification results are verified. The specific steps of generating the household change relationship grouping model are as follows:

[0107] S401: Based on the abnormal behavior feature point set, the data points are grouped, and the behavior pattern of each data point is analyzed to preliminarily classify the data. The classification represents the household change relationship. The execution process of obtaining the classification data set is as follows;

[0108] The data points are grouped and the behavior patterns are analyzed according to the formula:

[0109]

[0110] Calculate the classification identity of each group, where C j represents the classification identifier of the jth group, p i represents the weight of the i-th feature, x ij represents the eigenvalue of the i-th data point in the j-th group, and m is the total number of features;

[0111] The grouping classification of data points is determined by weighted eigenvalues. First, each feature x of the data point is determined. ijThe features come from the quantization results of the data preprocessing stage, and a weight p is set for each feature. i , the weights are based on the relevance of the features’ impact on the classification results and their statistical significance, which can be determined through correlation analysis or previous studies;

[0112] Referring to a simple case, there are three features with weights p1=0.5, p2=0.3, p3=0.2, and the feature value of a data point group j is x 1j =0.8,x 2j =0.6,x 3j =0.9, then the classification mark C of the data point group j The calculation is as follows:

[0113] C j =0.5×0.8+0.3×0.6+0.2×0.9=0.4+0.18+0.18=0.76

[0114] This calculation shows that by weighting the value of the differentiating features, the C j =0.76 can represent the comprehensive feature expression of this data point group in classification.

[0115] S402: Perform stability test according to the classification data set, including testing the consistency and repeatability of the categories in the classification data set, and generate stability test records by comparing the results of multiple tests. The execution process is as follows;

[0116] Based on the classified data set, a stability test is performed. The steps are to ensure that the classification results of the data set are sufficiently reliable and consistent, providing a solid foundation for the next application. The categories in the classified data set are tested for consistency, and the data points in the same category are checked to see if they have similar characteristics. This is achieved by calculating the standard deviation and coefficient of variation of the data points within the class. A repeatability test is performed, that is, the classification operation is performed multiple times on the same data set to check the consistency of the classification results, including re-running the classification algorithm with different initialization seeds or parameter settings, and evaluating the sensitivity of the algorithm to the initial conditions. By comparing the results of multiple tests, the stability of the categories and the robustness of the algorithm are analyzed. The test and analysis will form a detailed stability test record, which lists in detail the configuration, results and comparative analysis of each test. This information is crucial for subsequent data application and model adjustment.

[0117] S403: Based on the stability test record, record the data points of each category and the corresponding household change relationship, determine the structure and classification accuracy of the model, and generate the household change relationship grouping model. The execution process is as follows;

[0118] Based on the stability test records, further in-depth analysis is conducted on the data points of each category and the corresponding household change relationships. First, representative data points in each category are identified, and then the points are matched with known household change relationships, so as to establish the association between each category and the specific household change characteristics and determine the structure of the model, including determining the hierarchy and connection method of each category in the model, as well as the logical relationship between categories. The structure is set based on statistical analysis of the data and actual application needs to ensure that the classification model not only accurately reflects the classification situation in the real world, but also effectively supports subsequent data processing and decision-making. Through comprehensive stability test records and household change relationship analysis, a household change relationship grouping model is generated. This model will intuitively display the classification accuracy and data distribution between different household changes, and provide strong data support for market analysis, customer service improvement and resource allocation.

[0119] Specifically, if Figure 6 As shown in FIG. 1 , the user-to-user relationship grouping model is used, a feature set is selected in a supervised learning framework to train a prediction model, cross-validation is performed and the output is optimized, and the steps for generating a user-to-user relationship prediction model are as follows:

[0120] S501: According to the household change relationship grouping model, a feature set is selected from the data set, the initialization parameters of the model are adjusted to match the target supervised learning framework requirements, and the execution process of the preliminary training model is obtained by adjusting the training configuration and starting the model training.

[0121] According to the household change relationship grouping model, the feature set most relevant to the household change relationship is carefully selected from the data set. The feature set can significantly affect the performance and accuracy of the model. In the process of feature selection, the correlation between each feature and the household change relationship is analyzed, and those with high correlation are prioritized. The initialization parameters of the model are adjusted, including the learning rate, batch size, and activation function type, to ensure that the model can better adapt to the target supervised learning framework requirements. The model training is started by fine-tuning the training configuration, such as modifying the loss function and the optimizer selection. During the training process, the learning progress and performance of the model are monitored in real time to ensure that the training process is stable and effective, and a preliminary training model is obtained. The model preliminarily adapts to the prediction needs of household change relationships and lays the foundation for further optimization and application.

[0122] S502: Based on the preliminary training model, cross-validation settings are performed, the ratio and number of iterations of validation segmentation are determined, the model is cyclically tested and the prediction performance is evaluated, the validation process is regulated, and the execution flow of generating a validation optimization model is as follows;

[0123] Based on the preliminary training model, set the parameters of cross-validation, including determining the ratio of validation splits and the number of iterations. Cross-validation is an important step in model training. By dividing the data into multiple groups, alternately using a part of them as the test set and the rest as the training set, the generalization ability of the model is evaluated. Determine the appropriate split ratio and number of iterations to ensure that each data point has a chance to be verified and the model does not overfit to the training data. During the cross-validation process, the model is tested in a loop, and the model parameters are adjusted each iteration to evaluate its predictive performance. Through testing and regulating the validation process, the parameters and structure of the model are refined to make it more accurately meet the needs of the prediction task, and a validation optimization model is generated. This model shows excellent stability and prediction accuracy after multiple validations.

[0124] S503: Based on the validation optimization model, the model is tuned, including fine-tuning the learning rate and regularization parameters, controlling the output quality and performance of the model, and refining the prediction ability of the model through parameter adjustment. The execution process of generating the household change relationship prediction model is as follows;

[0125] Based on the validation optimization model, the model is further tuned, focusing on adjusting the learning rate and regularization parameters. These two parameters have a significant impact on the output quality and performance of the model. By fine-tuning the learning rate, the convergence speed and stability of the model training process can be controlled. Adjusting the regularization parameters helps reduce the overfitting phenomenon of the model and enhance its generalization ability. Through careful parameter adjustment, the output quality of the model is controlled to ensure that the model can show good prediction performance on different data sets. Through fine adjustments, the prediction ability of the model is refined, and finally a household change relationship prediction model is generated. After multi-level optimization and adjustment, the model can effectively predict and analyze household change relationships and provide strong data support for decision support.

[0126] Specifically, if Figure 7 As shown in the figure, the household change relationship prediction model is deployed to perform real-time data analysis, classify and predict the newly accessed data, update the prediction results in real time, and continuously monitor the model performance. The specific steps for generating real-time household change identification results are as follows:

[0127] S601: According to the household change relationship prediction model, the model is deployed, the data access interface is configured, and the real-time data stream is managed by adjusting the receiving buffer size and the data access frequency. The execution process of the data access configuration is as follows;

[0128] The receive buffer size and data access frequency are adjusted according to the formula:

[0129] B new =B old ×(1+β×Δr)

[0130] Perform configuration update, where Bnew represents the new cache size, B old represents the old cache size, β is the adjustment factor, and Δr represents the change ratio of data access frequency;

[0131] The purpose of adjusting the receiving buffer size is to optimize the management of data flow and ensure the continuity of data processing. In the process, β is the adjustment factor, which is used to control the sensitivity of buffer size adjustment, and Δr represents the change ratio of data access frequency relative to the original setting. By adjusting these two parameters, the buffer size can be dynamically adjusted to adapt to the change of data flow.

[0132] Set the initial cache size to 100MB (B old ), the data access frequency increases by 20% (Δr = 0.20), and the adjustment factor β is set to 0.5, then the new cache size B new The calculation is as follows:

[0133] B new =B old ×(1+β×Δr)=100×(1+0.5×0.20)=100×1.10=110MB

[0134] This calculation shows that by adaptively increasing the cache capacity with reference to the increase in data access frequency, data flow and processing latency can be managed more effectively.

[0135] S602: Based on the data access configuration, the prediction model is adjusted and real-time data is processed. The data batch processing size and data update interval are set. The prediction results are iteratively updated by adjusting the parameters. The execution process of obtaining the real-time prediction update results is as follows;

[0136] Based on the data access configuration, adjust the prediction model and process real-time data. First, determine the interface and data format for the model to receive real-time data to ensure smooth access and correct parsing of data. Adjust the input layer of the model to adapt to the real-time data characteristics, such as data scale and type, to ensure that the model can accurately process the data flowing in immediately. Set the data batch processing size. The parameters have a direct impact on the model processing speed and response time. The appropriate batch size can balance the utilization of computing resources and real-time requirements. Set the data update interval to ensure that the model can update the prediction results at the set frequency and adapt to the real-time changes of the data flow. During the model operation, adjust the parameters, such as adjusting the learning rate or using a dynamic adjustment strategy, to make subtle adjustments to the model's prediction logic to adapt to the continuously changing data characteristics. Adjust the model performance indicators based on monitoring, such as prediction accuracy and response time, and optimize the model configuration through a real-time feedback loop. Iterate and update the prediction results. Each data update will re-evaluate the model output to ensure the timeliness and accuracy of the output results. This dynamic update mechanism allows the model to continuously adapt to the new data environment, improve the reliability and accuracy of the prediction, and obtain real-time prediction update results. The results provide immediate support for subsequent decisions and enhance the overall responsiveness and efficiency of the system.

[0137] S603: Based on the real-time prediction update results, performance monitoring parameters are set, including error tracking and adjustment of warning thresholds, and the model is adjusted and maintained to maintain optimal performance by continuously monitoring the performance and response time of the model, thereby optimizing the model's response capability and prediction accuracy, and generating the real-time household change recognition results. The execution process is as follows;

[0138] Error tracking and adjustment of warning thresholds are based on the formula:

[0139]

[0140] Calculate the real-time error and adjust the threshold according to the formula:

[0141] T new =T old +α×(E t -T old )

[0142] Among them, E t represents the actual error at time t, y t represents the actual output at time t, represents the predicted output at time t, T new represents the new warning threshold, T old represents the old warning threshold, α is the adjustment rate, a constant less than 1;

[0143] In order to monitor the performance of the model and respond to output deviations in a timely manner, the absolute error E tMeasure the prediction accuracy at each time point by calculating the difference between the actual output yt and the predicted output The absolute value of the difference between the two is then used to update the warning threshold using an adjustment formula, allowing the model to adapt to changes or deviations;

[0144] Assume that the actual output at a certain time t is 100, the predicted output is 95, the old warning threshold is 10, and the adjustment rate α is set to 0.1;

[0145] Calculation error E t :

[0146]

[0147] Adjust the warning threshold T new :

[0148] T new =T old +α×(E t -T old )=10+0.1×(5-10)=10-0.5=9.5

[0149] This result shows that by adjusting the thresholds in real time, the model can more accurately reflect the current performance and warning needs.

[0150] like Figure 8 As shown, the online household change relationship identification system based on multi-source data fusion includes:

[0151] The data capture and formatting module captures the power frequency voltage curve, power frequency synchronization sequence, and signal-to-noise ratio data, performs data standardization, removes random noise in the data, performs data alignment, and generates a standardized feature data set;

[0152] The time feature extraction module extracts time series data from the standardized feature data set, divides the voltage and current data into time windows, uses differential technology to identify short-term variations in multiple windows, records and marks the variations, and obtains the distorted time series feature set;

[0153] The abnormal behavior analysis module statistically analyzes the distorted time series feature set, identifies abnormal data points, sets the abnormal detection threshold, and sorts and records the abnormal data points according to the frequency and intensity of occurrence to form an abnormal behavior feature point set;

[0154] The household change relationship grouping module uses the abnormal behavior feature point set to classify the data and divide the data into groups. The groups represent the household change relationship categories. The classified data is stability verified to generate a household change relationship grouping model.

[0155] The prediction model deployment module is based on the household change relationship grouping model, applies supervised learning methods to train the prediction model, evaluates the model performance through cross-validation, optimizes parameters, deploys the model on the real-time data analysis platform, continuously updates data predictions, and constructs real-time household change identification results.

[0156] 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. An online household change relationship identification method based on multi-source data fusion, characterized in that: The following steps are involved: Capture power frequency voltage curve, power frequency synchronization sequence, signal-to-noise ratio data, format data and filter noise, perform standardization on the filtered data, and generate a standardized feature data set; Based on the standardized feature data set, time features of multiple types of data are extracted, short-term variations in voltage and current are identified by using a differential method, distortion features are summarized, and a distortion time series feature set is generated; Performing deviation analysis on the features in the distorted time series feature set, setting thresholds and identifying out-of-range data points, recording abnormal points and sorting their occurrence frequency and intensity, and generating an abnormal behavior feature point set; According to the abnormal behavior feature point set, a grouping algorithm is applied to classify the data points, each category corresponds to the household change relationship in the substation area, the stability and repeatability of the classification results are verified, and a household change relationship grouping model is generated; Using the household change relationship grouping model, selecting a feature set to train a prediction model in a supervised learning framework, performing cross-validation and optimizing output, and generating a household change relationship prediction model; The household change relationship prediction model is deployed to perform real-time data analysis, classify and predict newly accessed data, update the prediction results in real time, and continuously monitor the model performance to generate real-time household change identification results.

2. The online household change relationship identification method based on multi-source data fusion according to claim 1 is characterized in that: The standardized feature data set includes voltage time features, current time features, and short-time variation indicators. The distortion time series feature set includes variation starting point, variation ending point, and variation amplitude. The abnormal behavior feature point set includes data point deviation level, data point frequency interval, and data point intensity score. The household change relationship grouping model includes classification standards, grouping effect consistency indicators, and repeatability verification results. The household change relationship prediction model includes feature selection accuracy, prediction accuracy, and model response time. The real-time household change identification results include new data matching degree, prediction update frequency, and performance monitoring results.

3. The online household change relationship identification method based on multi-source data fusion according to claim 1 is characterized in that: Capture the power frequency voltage curve, power frequency synchronization sequence, signal-to-noise ratio data, format the data and filter the noise, perform standardization on the filtered data, and generate a standardized feature data set in the following steps: Capture power frequency voltage curve, power frequency synchronization sequence, signal-to-noise ratio data, perform data format correction, adjust data fields, and obtain formatted voltage synchronization data; According to the formatted voltage synchronization data, filter parameters are set to identify and eliminate noise, filter configuration for data processing is adjusted, and processed synchronization data is obtained; The processed synchronous data is numerically adjusted, a standardized scale is set, and a numerical conversion is applied to process the standardized data to generate a standardized feature data set.

4. The online household change relationship identification method based on multi-source data fusion according to claim 1 is characterized in that: Based on the standardized feature data set, the time features of multiple types of data are extracted, the short-term variations in voltage and current are identified by using a differential method, the distortion features are summarized, and the steps of generating a distortion time series feature set are specifically as follows: According to the standardized characteristic data set, each data is timestamped through the voltage and current data sets to record the date and time of the data, including reading the original record of the data, extracting and attaching time information to each data point, and generating a time-stamped data set; According to the time-stamped data set, a moving window difference method is used to perform a difference analysis, the voltage and current differences between consecutive data points are calculated point by point, and short-time variation points are identified, the short-time variation points are marked and recorded, and a distortion record set is generated; Through the distortion record set, the distortion events are arranged in the order of occurrence, the voltage and current distortion conditions at each time point during the monitoring period are recorded, and a distortion time series feature set is generated.

5. The online household change relationship identification method based on multi-source data fusion according to claim 4 is characterized in that: The formula of the moving window difference method is as follows: Among them, Δy represents the short-term variation of voltage and current, α represents the weight of voltage change, β represents the weight of current change, γ represents the normalization factor, and t window Represents the time length of the moving window, V t and V t+1 Indicates the voltage reading within two seconds. t+1 and I t Indicates the current reading for two consecutive seconds.

6. The online household change relationship identification method based on multi-source data fusion according to claim 1 is characterized in that: The steps of performing deviation analysis on the features in the distorted time series feature set, setting thresholds and identifying out-of-range data points, recording abnormal points and sorting the frequency and intensity of occurrence, and generating an abnormal behavior feature point set are as follows: Extracting each data point from the time series feature set, setting a monitoring threshold, comparing the value of each data point with the set threshold, marking points exceeding the threshold as abnormal, identifying the abnormality through parameter comparison, and obtaining an abnormal data point set; According to the abnormal data point set, the occurrence frequency of each abnormal point is counted, the degree of deviation from the normal value is measured, the data points are sorted according to the frequency and the degree of deviation, the data points are sorted into an ordered abnormal list, and the sorted abnormal data is obtained; According to the sorted abnormal data, combined with the recorded frequency and deviation degree information, the characteristics of the abnormal points are reflected, and a set of abnormal behavior feature points is generated by integrating the information.

7. The online household change relationship identification method based on multi-source data fusion according to claim 1 is characterized in that: According to the abnormal behavior feature point set, the data points are classified by applying a grouping algorithm, each category corresponds to the household change relationship in the substation area, the stability and repeatability of the classification results are verified, and the steps of generating the household change relationship grouping model are specifically as follows: Based on the abnormal behavior feature point set, the data points are grouped, and the behavior pattern of each data point is analyzed to preliminarily classify the data, the classification represents the household change relationship, and obtain the classification data set; Performing a stability test according to the classification data set, including performing consistency and repeatability tests on the categories in the classification data set, and generating a stability test record by comparing the results of multiple tests; Based on the stability test records, the data points of each category and the corresponding household change relationships are recorded, the structure and classification accuracy of the model are determined, and a household change relationship grouping model is generated.

8. The online household change relationship identification method based on multi-source data fusion according to claim 1 is characterized in that: Using the household change relationship grouping model, selecting a feature set to train a prediction model in a supervised learning framework, performing cross validation and optimizing the output, and generating a household change relationship prediction model are specifically as follows: According to the household change relationship grouping model, a feature set is selected from the data set, and the initialization parameters of the model are adjusted to match the target supervised learning framework requirements. The training configuration is adjusted and the model training is started to obtain a preliminary training model; Based on the preliminary training model, cross-validation settings are performed, the ratio and number of iterations of validation segmentation are determined, the model is cyclically tested and the prediction performance is evaluated, the validation process is regulated, and a validation optimization model is generated; Based on the validation optimization model, model tuning is performed, including fine-tuning of the learning rate and regularization parameters, to control the output quality and performance of the model. By adjusting the parameters, the prediction capability of the model is refined to generate a household change relationship prediction model.

9. The online household change relationship identification method based on multi-source data fusion according to claim 1 is characterized in that: The steps of deploying the household change relationship prediction model to perform real-time data analysis, classify and predict newly accessed data, update the prediction results in real time, and continuously monitor the model performance to generate real-time household change identification results are as follows: According to the household change relationship prediction model, the model is deployed, the data access interface is configured, and the real-time data stream is managed by adjusting the receiving buffer size and the data access frequency to obtain the data access configuration; Based on the data access configuration, adjust the prediction model and process real-time data, set the data batch processing size and data update interval, and iteratively update the prediction results by adjusting the parameters to obtain real-time prediction update results; Based on the real-time prediction update results, performance monitoring parameters are set, including error tracking and adjustment of warning thresholds. By continuously monitoring the performance and response time of the model, the model is adjusted and maintained to maintain optimal performance, the model's responsiveness and prediction accuracy are optimized, and real-time household change identification results are generated.

10. The online household change relationship identification system based on multi-source data fusion is characterized by: According to the online household change relationship identification method based on multi-source data fusion according to any one of claims 1 to 9, the system comprises: The data capture and formatting module captures the power frequency voltage curve, power frequency synchronization sequence, and signal-to-noise ratio data, performs data standardization, removes random noise in the data, performs data alignment, and generates a standardized feature data set; The time feature extraction module extracts time series data from the standardized feature data set, performs time window segmentation on the voltage and current data, uses differential technology to identify short-term variations in multiple windows, records and marks the variations, and obtains a distortion time series feature set; The abnormal behavior analysis module statistically analyzes the distorted time series feature set, identifies abnormal data points, sets an abnormal detection threshold, and sorts and records the abnormal data points according to the frequency and intensity of occurrence to form an abnormal behavior feature point set; The household change relationship grouping module uses the abnormal behavior feature point set to classify the data and divide the data into groups, where the groups represent household change relationship categories, and the classified data is stability verified to generate a household change relationship grouping model; The prediction model deployment module is based on the household change relationship grouping model, applies the supervised learning method to train the prediction model, evaluates the model performance through cross-validation, optimizes the parameters, deploys the model on the real-time data analysis platform, continuously updates the data prediction, and constructs real-time household change identification results.

Citation Information

Cited By

  • Face recognition application method and system combined with intelligent glasses

    CN119828856A

  • A face recognition application method and system combined with smart glasses

    CN119828856B

  • Intelligent home environment monitoring management system based on cloud computing

    CN120257109A

  • Intelligent sample plot data acquisition and analysis system and method for public welfare forest ecological monitoring

    CN122435443A