Extra-high voltage transformer substation handover test data fusion and abnormity early warning method
By cleaning, formatting, and normalizing the dimensions of multi-source heterogeneous test data from UHV substations, and combining this with a multidimensional association rule model and dynamic threshold range, the problems of insufficient data fusion and inadequate adaptability of early warning systems in existing technologies have been solved, enabling accurate analysis of equipment status and efficient identification of early faults.
Patent Information
- Application Number
- CN202511448673.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-11
- Publication Date
- 2026-01-23
AI Technical Summary
In existing technologies, there is a lack of systematic multi-source heterogeneous test data fusion mechanisms. Most platforms only support independent analysis of a certain type of test data, which makes it impossible to effectively mine the parameter correlation features between devices. Traditional fixed threshold judgment rules have low accuracy when facing large-scale, highly complex equipment systems and cannot adapt to the differences in different equipment types, operating environments and historical working conditions. Furthermore, existing early warning systems have not established a data feedback and self-learning mechanism that links with the actual on-site operating status, resulting in a high misjudgment rate and slow response.
By collecting heterogeneous raw data from multiple sources, cleaning, format unification, and dimension normalization are performed to generate a standardized experimental dataset. A pre-trained multidimensional association rule model is used to mine implicit association rules, and comparisons are made based on dynamic threshold intervals to calculate anomaly confidence. Combined with visual early warning signals and sample feedback mechanisms, iterative optimization is achieved.
It significantly enhances the comparability and coupling between different test parameters, improves the ability to identify early faults and edge anomalies and the timeliness of response, and constructs a closed-loop feedback mechanism that integrates visual early warning, manual review, sample feedback and model iterative optimization, thereby improving the stability and intelligence level of the system in long-term deployment.
Smart Images
Figure CN121388976A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of power system intelligent operation and maintenance, and particularly relates to a method for data fusion and abnormal early warning of handover test of an extra-high voltage substation. BACKGROUND
[0002] In the process of equipment handover test before the construction and operation of an extra-high voltage substation, various electrical performance and insulation state detection need to be carried out for key equipment such as transformers, GIS, circuit breakers and mutual inductors, including but not limited to insulation resistance test, dielectric loss factor test, partial discharge test, loop resistance test, SF6 gas detection, vibration characteristic analysis and ultrasonic detection.
[0003] The existing technology has the following problems: first, there is a lack of systematic multi-source heterogeneous test data fusion mechanism, and most platforms only support independent analysis of a certain type of test data, which makes it difficult to effectively mine the parameter correlation characteristics between devices. Second, the traditional fixed threshold judgment rule has low precision when facing large-scale and high-complexity device systems, and cannot adapt to the differences in different device types, operating environments and historical operating conditions, resulting in high misjudgment rate and slow response. Third, the existing early warning system is a static model, and there is no data feedback and self-learning mechanism linked to the actual operating state of the field. The model is difficult to optimize and upgrade as the device state evolves, and its accuracy and adaptability gradually decrease over time, which restricts its application in practical engineering. SUMMARY
[0004] The present application provides a method for data fusion and abnormal early warning of handover test of an extra-high voltage substation, which has an intelligent auxiliary analysis mechanism with multi-dimensional fusion and dynamic discrimination, and improves the abnormal early warning capability and operation decision efficiency in the handover test stage.
[0005] A method for data fusion and abnormal early warning of handover test of an extra-high voltage substation, comprising the following steps: S1: collecting multi-source heterogeneous raw data in the handover test of an extra-high voltage substation through a data interface; cleaning, unifying the format and dimension normalizing the multi-source heterogeneous raw data to generate a standardized test data set; S2: inputting the standardized test data set into a pre-trained multi-dimensional association rule model for analysis; the multi-dimensional association rule model mines and outputs the implicit association rules between different test parameters; based on the implicit association rules, the standardized test data set is fused and calculated to generate fusion data representing the overall health status of the entire substation equipment; S3: comparing the fusion data with a dynamic threshold interval generated based on a historical handover test data learning set; according to the comparison result, the abnormal confidence of the current handover test data is calculated; S4: according to the size of the abnormal confidence, start a different level of visual warning signal; at the same time, the standardized test data set of this handover test and its corresponding abnormal confidence are fed back to the historical handover test data learning set, and the multi-dimensional association rule model and the dynamic threshold interval are iteratively optimized.
[0006] Optionally, the S1 comprises: S11: collecting multi-source heterogeneous original data in the handover test of the ultra-high voltage substation through a data interface, the multi-source heterogeneous original data comprising at least three of insulation resistance test data, dielectric loss factor test data, partial discharge test data, loop resistance test data, SF6 gas test data, vibration characteristic test data and ultrasonic detection data; S12: cleaning and format unification processing of the multi-source heterogeneous original data, wherein the cleaning processing comprises filling of missing values and elimination of abnormal values in the multi-source heterogeneous original data, and the format unification processing comprises converting the multi-source heterogeneous original data of different sources into unified CSV or JSON format to generate an intermediate standardized data set; S13: dimensionless normalization processing of the intermediate standardized data set, adopting a Z-score standardization method to convert test parameter values of different dimensions in the intermediate standardized data set to a unified dimensionless value interval to generate a final standardized test data set.
[0007] Optionally, the filling of the missing values in the multi-source heterogeneous original data is specifically: filling the missing values in the multi-source heterogeneous original data with the average value or the median value of the historical data of the same test equipment.
[0008] Optionally, the S2 comprises: S21: loading and calling a pre-trained multi-dimensional association rule model with the standardized test data set as input, wherein the pre-trained multi-dimensional association rule model is obtained by training a historical handover test data learning set through an FP-Growth algorithm, and is used for analyzing concurrent relationships and dependent relationships between multiple test parameters in the standardized test data set; S22: the pre-trained multi-dimensional association rule model scans and identifies the input standardized test data set, mines and outputs implicit association rules contained therein, and the implicit association rules represent the association relationships between different test parameters in the form of conditional probability of “antecedent→consequent”; S23: Assigning dynamic weights to each test parameter in the standardized test data set based on the importance and correlation strength between parameters revealed by the implicit association rules, and performing nonlinear combination calculation on the weighted parameter values to finally generate fusion data representing the overall health status of the entire substation equipment.
[0009] Optionally, the specific way of assigning dynamic weights to each test parameter in the standardized test data set is: dynamically calculating and assigning weight values according to the total number of times each parameter appears as a "post-term" in the implicit association rules or its average confidence, wherein the parameter with higher frequency or higher average confidence is assigned a larger weight.
[0010] Optionally, S3 includes: S31: Calling the dynamic threshold interval generated based on the historical handover test data learning set, wherein the dynamic threshold interval is defined by calculating the mean and standard deviation of the fusion data of all normal samples in the historical handover test data learning set, and the range of three times the standard deviation of the mean value is defined as the dynamic threshold interval; S32: Comparing the fusion data generated in S2 with the dynamic threshold interval, calculating the deviation degree of the fusion data relative to the center value of the dynamic threshold interval, wherein the deviation degree is the ratio of the absolute distance between the fusion data and the center value of the dynamic threshold interval to half the width of the dynamic threshold interval; S33: According to the calculated deviation degree, using a preset Gaussian distribution probability function to map the deviation degree to an abnormal confidence between 0 and 1, wherein the larger the deviation degree, the closer the abnormal confidence to 1, and the smaller the deviation degree, the closer the abnormal confidence to 0.
[0011] Optionally, the generation of the dynamic threshold interval is dynamically updated, and the update mechanism is: whenever a new confirmed normal standardized test data set is fed back and added to the historical handover test data learning set, the mean and standard deviation of the fusion data of all normal samples are automatically recalculated, and the dynamic threshold interval is updated.
[0012] Optionally, S4 includes: S41: According to the size of the abnormal confidence calculated in step S3, starting visual warning signals of different colors and levels on the monitoring interface, wherein no warning signal is sent when the abnormal confidence is less than 0.3, a blue warning signal is started when the abnormal confidence is between 0.3 and 0.6, a yellow warning signal is started when the abnormal confidence is between 0.6 and 0.8, and a red warning signal is started when the abnormal confidence is greater than 0.8. S42: The standardized test data set of this handover test and its corresponding abnormal confidence are taken as new samples, and after rechecking the actual state of the equipment confirmed by the operation and maintenance personnel, the new samples confirmed as normal state are added to the historical handover test data learning set; S43: using the updated historical handover test data learning set, retraining the multi-dimensional association rule model and recalculating the dynamic threshold interval, completing the iterative optimization of the multi-dimensional association rule model and the dynamic threshold interval, and using the optimized model and threshold interval for the next handover test data analysis.
[0013] Optionally, the visual warning signal includes a color warning window popped up on the man-machine interface of the monitoring system, a flashing device icon and a generated warning level text description.
[0014] Optionally, the yellow warning signal or the red warning signal is started at the same time, a warning log record is automatically generated, and the warning log record is pushed to the mobile terminal of the related operation and maintenance personnel through the message queue.
[0015] The beneficial effects of the present application are: The present application synchronously collects at least three types of multi-source heterogeneous raw data from a plurality of handover test devices (such as insulation resistance testers, dielectric loss testers, SF6 gas analyzers, etc.), and sequentially completes missing value filling, abnormal value elimination, format unification and Z-score dimensionless processing, and finally generates a standardized test data set with uniform structure and consistent scale. The processing procedure significantly enhances the comparability and coupling between different test parameters, lays a stable foundation for subsequent fusion modeling, and solves the problem of fragmented multi-type test result data in the prior art, which is difficult to analyze comprehensively.
[0016] The present application, by introducing a collaborative calculation strategy based on a multi-dimensional association rule model and a dynamic threshold interval, dynamically generates weights and constructs a nonlinear combined fusion data based on the mining of implicit dependency relationships between standardized parameters, which is used to accurately reflect the overall health status of the equipment. Combined with the dynamic threshold interval constructed based on the historical normal samples, the deviation degree calculation and abnormal confidence mapping of the fusion data can be realized, thereby effectively avoiding the misjudgment caused by the limitation of a single parameter in the traditional method, and improving the identification ability and response time of early faults and marginal abnormalities.
[0017] This invention constructs a closed-loop feedback mechanism integrating visual early warning, manual review, sample feedback, and model iterative optimization. Based on outputting visual early warning signals of different color levels, the system automatically filters newly added normal samples and updates the historical handover test data learning set, combined with expert system or manual review and annotation. When the number of new samples reaches a preset threshold, the system triggers a full reconstruction and optimization update of the multidimensional association rule model and the dynamic threshold interval, effectively achieving adaptive matching between the model and the evolution of on-site working conditions, and improving the system's stability, scalability, and intelligence level in long-term deployment. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0019] Fig. 1 This is a schematic diagram of the method flow according to an embodiment of the present invention; Fig. 2 This is a schematic diagram of the S3 process in an embodiment of the present invention. Detailed Implementation
[0020] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should also be noted that, to make the embodiments more comprehensive, the following embodiments are the best and preferred embodiments, and those skilled in the art can use other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.
[0021] like Figs. 1-2 As shown, a method for data fusion and anomaly early warning in ultra-high voltage substation handover tests includes the following steps: S1: Collect multi-source heterogeneous raw data from UHV substation handover tests via data interface; clean, standardize the format, and normalize the dimensions of the multi-source heterogeneous raw data to generate a standardized test data set, specifically: S11: First, use data interfaces to collect multi-source heterogeneous raw data in real time from various test equipment at the UHV substation handover test site. The multi-source heterogeneous raw data includes at least three of the following: insulation resistance test data, dielectric loss factor test data, partial discharge test data, loop resistance test data, SF6 gas test data, vibration characteristic test data, and ultrasonic test data.
[0022] The specific data collection method is as follows: The corresponding multi-source heterogeneous raw data is collected from the insulation resistance tester and loop resistance tester through the RS-485 serial interface; The corresponding multi-source heterogeneous raw data is collected from the medium loss factor tester and the partial discharge detection system through the IEC 61850 protocol conversion interface; The corresponding multi-source heterogeneous raw data is collected from the SF6 gas component analyzer, vibration sensor and ultrasonic detector through the OPC UA protocol interface.
[0023] After the device connection is established through the above interfaces, all the raw data output by the above devices is acquired item by item, and is input as multi-source heterogeneous raw data to the subsequent preprocessing process.
[0024] S12: After the collection is completed, first, the multi-source heterogeneous raw data is executed cleaning processing, specifically including two aspects of missing value filling and outlier rejection: Missing value filling: for the parameter items with missing values, the historical data of the parameter in the same test device is automatically searched, and the average value or median is calculated to fill the missing values; Outlier rejection: the Relyada criterion and box plot method are used to identify and reject outliers for each type of data, to ensure the stability and reliability of the retained data.
[0025] After the cleaning processing is completed, the format uniform processing is executed. First, a unified field name mapping table is constructed to cover all test data fields, and the multi-source heterogeneous raw data of different sources is converted into a structured uniform file format, specifically including: Mapping all data according to the standard field name in the field name mapping table; Uniformly converted into a CSV format file or a JSON format object with consistent structure; The above results are summarized as an intermediate standardized data set, which is the output after cleaning and format uniform processing.
[0026] S13: The intermediate standardized data set is subjected to dimensionless normalization processing to generate a standardized test data set After obtaining the intermediate standardized data set, all test parameters are processed by a unified dimensionless normalization method, which aims to eliminate the interference of different dimensions and scales on subsequent analysis and improve the consistency of data input. The dimensionless normalization method includes Z-score standardization method and Min-Max normalization method, and the user can select the appropriate method according to the scene.
[0027] When the Z-score standardization method is used, the normalization processing is based on the statistical values in the historical handover test data learning set, and the calculation formula is as follows: ; wherein, is the original value in the intermediate normalized data set, is the mean value of the parameter in the historical handover test data learning set, is the standard deviation thereof, is the value after Z-score normalization.
[0028] When the Min-Max normalization method is adopted, the maximum and minimum values of each parameter are calculated, and the normalization processing is performed according to the following formula: wherein, is the original value in the intermediate normalized data set, and are the minimum and maximum values of the parameter in the historical handover test data learning set, respectively, is the normalized value.
[0029] Regardless of the method used, all normalized data are summarized into a standardized test data set, which is the final output of the present step S1 and provides a unified, dimensionless and structured data input basis for subsequent multi-dimensional association rule model analysis.
[0030] S2: input the standardized test data set into a pre-trained multi-dimensional association rule model for analysis; the multi-dimensional association rule model mines and outputs the implicit association rules between different test parameters; based on the implicit association rules, the standardized test data set is fused and calculated to generate fusion data representing the overall health status of the entire substation equipment, specifically: S21: the standardized test data set output in S1 is taken as input data, and a pre-trained multi-dimensional association rule model that has completed training is loaded and called. The pre-trained multi-dimensional association rule model is obtained by training the historical handover test data learning set by the FP-Growth algorithm, and the training samples are the normal state samples in the historical handover test data learning set.
[0031] The training process includes the following operations: First, set the minimum support threshold and the minimum confidence threshold; Perform the FP-Growth algorithm on the standardized test parameters in the historical handover test data learning set to generate all frequent item sets; Construct all possible association rules based on the generated frequent item sets; Retain the association rules that meet the minimum support and minimum confidence requirements; Store all high-quality association rules finally generated in the multi-dimensional association rule model, and construct an index structure to speed up the subsequent rule matching process.
[0032] After the model is loaded, the model will be used to mine concurrent relationships and dependencies between various experimental parameters in the standardized experimental data set of the current input.
[0033] S22: After the pre-trained multi-dimensional association rule model is loaded, it scans and pattern recognizes the input standardized experimental data set, identifies combinations that meet the trained frequent patterns, and outputs implicit association rules.
[0034] The implicit association rule is expressed in the form of "antecedent → consequent", and both the "antecedent" and the "consequent" are composed of one or more experimental parameters and their value intervals. The specific rule structure output is as follows: For example: {parameter A ∈ interval a, parameter B ∈ interval b} → {parameter C ∈ interval c}; Wherein: The "antecedent" represents a group of experimental parameters and their value intervals that are simultaneously satisfied; The "consequent" represents another group of experimental parameters and their value intervals that are highly likely to appear simultaneously under the condition that the above antecedent is true; All rules are accompanied by two core indicators: support and confidence.
[0035] The definitions of support and confidence are as follows: Support (Support) is the ratio of the number of samples that meet the antecedent and consequent conditions to the total number of samples; Confidence (Confidence) is the ratio of the number of samples that meet the antecedent and consequent conditions to the number of samples that only meet the antecedent condition.
[0036] That is: ; After generating implicit association rules, they are screened to eliminate low-quality rules that have no actual analysis value. The screening process includes two stages: lift calculation and threshold judgment.
[0037] The calculation method of lift is as follows: Lift is used to measure the independence and reinforcement relationship between the antecedent and the consequent, and is defined as the ratio of the confidence of the current rule to the support of the consequent: ; Wherein, the calculation method of the support of the consequent is: ; Therefore, the complete lift formula is: ; All the implicit association rules with confidence higher than the pre-set confidence threshold and lift greater than 1 are reserved for subsequent fusion calculation to ensure the robustness and reliability of the fusion results.
[0038] S23: After obtaining the final set of implicit association rules, the generation process of the fusion data is performed, which is divided into two stages: Dynamic weight allocation: for all test parameters in the standardized test data set, the weight value of each test parameter is dynamically calculated according to the total number of times it appears as a "consequent" in the implicit association rules and the average confidence of the rules in that position. The specific method is as follows: Count the number of times each parameter appears as a "consequent" in all retained implicit association rules; Calculate the average confidence of the rules corresponding to the parameter as a "consequent"; Standardize the fusion of these two indicators to generate the dynamic weight of the parameter; The parameters with higher frequency and higher confidence will be assigned a larger dynamic weight.
[0039] Nonlinear combination calculation: combine the above dynamic weight allocation results with the parameter values in the standardized test data set and input them into a pre-set neural network combiner or support vector machine regression model. The neural network combiner or support vector machine regression model has been trained based on historical data and has the ability to map weighted parameter values to health status scores.
[0040] The model will automatically perform the following processing: Feature fusion of all weighted parameter values; Perform nonlinear function transformation; Output a fusion data representing the overall health status of the UHV substation equipment.
[0041] The final generated fusion data will be used as the output result of this step S2 for subsequent steps of dynamic threshold comparison and abnormal confidence calculation.
[0042] S3: Compare the fusion data with the dynamic threshold interval generated based on the historical handover test data learning set; according to the comparison result, calculate the abnormal confidence of the current handover test data, specifically: S31: First, call the dynamic threshold interval constructed based on the historical handover test data learning set. The historical handover test data learning set consists of multiple handover test samples confirmed to be normal by artificial or confirmation, each sample containing a standardized test data set and its corresponding fusion data.
[0043] In order to construct the dynamic threshold interval, statistical analysis is performed on the fusion data of all normal samples in the historical handover test data learning set, the specific operation is as follows: Calculate the mean of all normal sample fusion data ; Calculate the standard deviation of all normal sample fusion data ; Define the numerical interval centered at with positive and negative as the boundaries of the current dynamic threshold interval, that is: ; To ensure dynamic adaptability, the dynamic threshold interval has a real-time updating mechanism. Whenever a new sample verified as normal (including its standardized test data set and fusion data) is received and added to the historical handover test data learning set, the above and values are automatically recalculated to dynamically refresh the latest dynamic threshold interval.
[0044] S32: After obtaining the fusion data calculated in step S2, compare the fusion data with the current dynamic threshold interval, and calculate the deviation of the fusion data relative to the center value of the dynamic threshold interval. The deviation is used to quantify the difference between the current fusion data and the normal interval, and the specific calculation method is as follows: First, define the center value of the dynamic threshold interval as the mean ; Define the upper limit of the dynamic threshold interval as ; The formula for calculating the deviation is: ; In the above formula: The numerator is the absolute distance between the fusion data and the center value of the dynamic threshold interval (i.e. the mean ); The denominator is half the width of the dynamic threshold interval (i.e. ), ensuring that the deviation is a dimensionless quantity.
[0045] The larger the deviation, the more the fusion data deviates from the historical normal sample interval, indicating that the current device may have a greater risk of abnormality.
[0046] S33: After obtaining the deviation, input the deviation into the preset probability mapping function to generate the corresponding abnormal confidence. The present application provides two different mapping methods, which are suitable for different working conditions and support flexible switching: Gaussian distribution probability mapping function: adopt the standard Gaussian function form to nonlinearly map the deviation to the abnormal confidence, and its formula is as follows: ; Wherein: exp is the natural exponential function; the greater the deviation degree, the smaller the exponential term, and the closer the anomaly confidence to 1; the smaller the deviation degree, the closer the anomaly confidence to 0.
[0047] The segmented linear function mapping mode: in order to balance the engineering practicability and calculation efficiency, a set of segmented linear functions is also provided to directly map the deviation degree to the anomaly confidence, and the mapping rules are as follows: When the deviation degree < 1: ; When the deviation degree is greater than or equal to 1: ; Wherein, the "maximum expected deviation degree" is a preset upper limit value for defining a reasonable range of deviation degree, and ensuring that the mapping value is always between [0, 1].
[0048] The finally generated anomaly confidence is taken as the output of S3, which provides data basis for subsequent visual warning signal level determination and model feedback.
[0049] S4: according to the size of the anomaly confidence, start different levels of visual warning signals; at the same time, the standardized test data set of this handover test and its corresponding anomaly confidence are fed back to the historical handover test data learning set, and the multi-dimensional association rule model and the dynamic threshold interval are iteratively optimized, which is specifically: S41: after the calculation of the anomaly confidence, the corresponding visual warning signal is generated on the visual human-computer interface of the monitoring system according to the numerical value. The specific triggering rules of the warning signal are as follows: When the anomaly confidence is less than 0.3, no warning signal is sent, and the device is considered to be in a healthy state; When the anomaly confidence is between 0.3 and 0.6, start the blue warning signal; When the anomaly confidence is between 0.6 and 0.8, start the yellow warning signal; When the anomaly confidence is greater than 0.8, start the red warning signal.
[0050] The visual warning signal includes the following three types of content: Pop up the warning window of the corresponding color on the human-computer interface of the monitoring system, which contains the current device name, measurement point number, anomaly confidence value and current warning level; Highlight the icon of the corresponding device in the device topology diagram with the corresponding color and perform a flashing animation to enhance the visual recognition; Generate the corresponding warning level text description in the device information panel, and the text content includes "blue warning", "yellow warning" or "red warning".
[0051] In addition, to ensure the real-time response of operation and maintenance, the following operations are automatically performed when the yellow or red warning signal is started: A standardized format warning log record is generated, including device number, occurrence time, abnormal confidence, warning level, fusion data value, dynamic threshold upper and lower limit value; The warning log is pushed to the mobile terminal of the relevant operation and maintenance personnel through the message queue middleware (such as Kafka or RabbitMQ) deployed by the enterprise, realizing real-time transmission of abnormal information.
[0052] S42: To enhance the accuracy and controllability of learning data, the standardized test data set of this handover test and its corresponding abnormal confidence are used as new samples, which are rechecked and confirmed with the actual equipment state.
[0053] The rechecking method includes: Calling the integrated expert system for preliminary state identification to determine whether the fusion data has typical normal characteristics; If the expert system cannot make a clear determination, experienced operation and maintenance personnel perform manual confirmation, combining on-site inspection results, other channel monitoring data, and historical operation conditions for comprehensive judgment; Finally, the new sample is marked as "normal" or "abnormal" state.
[0054] Only when the new sample is clearly marked as "normal" state, it is included in the historical handover test data learning set as a reference data source for subsequent modeling and updating.
[0055] S43: After completing the rechecking of the new sample and updating the historical handover test data learning set, automatically detect whether the number of new samples reaches the preset batch processing threshold. If the condition is met, automatically trigger the model iteration optimization process, which includes the following two operations: Re-train the multi-dimensional association rule model: Use the updated historical handover test data learning set as complete training data; Use the same FP-Growth algorithm as the initial modeling to re-execute frequent item set mining and implicit association rule generation; Maintain the same minimum support threshold and minimum confidence threshold as the initial training phase; Completely replace the original model to ensure that the model always adapts to the latest test data distribution.
[0056] Recalculate the dynamic threshold interval: Recalculate the mean and standard deviation of all fusion data in the updated learning set; Based on Reconstructing new dynamic threshold interval; Replacing original threshold upper and lower limits to ensure that the "normal" interval is always synchronized with the evolution of the device state in subsequent analysis.
[0057] The above iterative optimization process is triggered automatically, and the triggering condition is that the cumulative number of new samples confirmed as normal state in the historical handover test data learning set reaches the preset batch processing threshold (such as 20 or 50), thereby maintaining robustness while having continuous learning ability.
[0058] The optimized multi-dimensional association rule model and dynamic threshold interval will be used for the fusion analysis and abnormal identification of the next round of handover test data.
[0059] The present application covers any alternative, modification, equivalent method and scheme made on the essence and scope of the present application. In order to make the public have a thorough understanding of the present application, specific details are described in the following preferred embodiments of the present application, and the present application can also be fully understood without the description of these details for those skilled in the art. In addition, in order to avoid unnecessary confusion to the essence of the present application, well-known methods, processes, procedures, elements and circuits, etc. are not described in detail.
[0060] The above is only the preferred embodiment of the present application, and it should be pointed out that for ordinary skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, and these improvements and refinements should be considered as the protection scope of the present application.
Claims
1. A method for data fusion and anomaly early warning in ultra-high voltage substation handover tests, characterized in that, Includes the following steps: S1: Collect multi-source heterogeneous raw data from the UHV substation handover test through the data interface; clean, unify the format and normalize the dimensions of the multi-source heterogeneous raw data to generate a standardized test data set; S2: Input the standardized test data set into a pre-trained multidimensional association rule model for analysis; the multidimensional association rule model mines and outputs implicit association rules between different test parameters; based on the implicit association rules, perform fusion calculation on the standardized test data set to generate fusion data characterizing the overall health status of the entire substation equipment; S3: Compare the fused data with the dynamic threshold range generated based on the historical handover test data learning set; calculate the anomaly confidence level of the current handover test data based on the comparison results; S4: Based on the magnitude of the anomaly confidence level, activate different levels of visual early warning signals; simultaneously, use the standardized test data set of this handover test and its corresponding anomaly confidence level as new samples, feed them back to the historical handover test data learning set, and iteratively optimize the multidimensional association rule model and dynamic threshold interval.
2. The method for data fusion and anomaly early warning of UHV substation handover test according to claim 1, characterized in that, S1 includes: S11: Collect multi-source heterogeneous raw data from the UHV substation handover test through the data interface. The multi-source heterogeneous raw data includes at least three of the following: insulation resistance test data, dielectric loss factor test data, partial discharge test data, loop resistance test data, SF6 gas test data, vibration characteristic test data, and ultrasonic detection data. S12: Clean and unify the format of the multi-source heterogeneous raw data. The cleaning process includes filling in missing values and removing outliers in the multi-source heterogeneous raw data. The format unification process includes converting the multi-source heterogeneous raw data from different sources into a unified CSV or JSON format to generate an intermediate standardized data set. S13: Perform dimensional normalization processing on the intermediate standardized data set. Use the Z-score normalization method to transform the experimental parameter values of different dimensions in the intermediate standardized data set to a unified dimensionless numerical range, thereby generating the final standardized experimental data set.
3. The method for data fusion and anomaly early warning of UHV substation handover test according to claim 2, characterized in that, The process of filling missing values in the multi-source heterogeneous raw data specifically involves filling missing values in the multi-source heterogeneous raw data with the average or median of historical data from the same experimental equipment.
4. The method for data fusion and anomaly early warning of UHV substation handover test according to claim 3, characterized in that, S2 includes: S21: Using the standardized experimental dataset as input, load and call the pre-trained multidimensional association rule model, wherein the pre-trained multidimensional association rule model is obtained by training the historical handover experimental data learning set through the FP-Growth algorithm, and is used to analyze the concurrency and dependency relationships between multiple experimental parameters in the standardized experimental dataset; S22: The pre-trained multidimensional association rule model scans and performs pattern recognition on the input standardized experimental data set, mines and outputs the implicit association rules contained therein, and the implicit association rules represent the association relationship between different experimental parameters in the form of "preceding term → subsequent term" conditional probability. S23: Based on the importance and correlation strength between parameters revealed by the implicit association rules, dynamic weights are assigned to each test parameter in the standardized test data set, and nonlinear combination calculations are performed on the weighted parameter values to finally generate fused data that characterizes the overall health status of the entire substation equipment.
5. The method for data fusion and anomaly early warning of UHV substation handover test according to claim 4, characterized in that, The specific method for assigning dynamic weights to each experimental parameter in the standardized experimental data set is as follows: based on the total number of times each parameter appears as a "subject" in the implicit association rule or its average confidence level, its weight value is dynamically calculated and assigned, wherein the parameter with higher frequency of occurrence or higher average confidence level is assigned a greater weight.
6. The method for data fusion and anomaly early warning of UHV substation handover test according to claim 5, characterized in that, S3 includes: S31: Call the dynamic threshold interval generated based on the historical handover test data learning set. The dynamic threshold interval is defined by calculating the mean and standard deviation of the fusion data of all normal samples in the historical handover test data learning set, and defining the range of the mean plus or minus three times the standard deviation as the dynamic threshold interval. S32: Compare the fused data generated in S2 with the dynamic threshold interval, and calculate the deviation of the fused data from the center value of the dynamic threshold interval. The deviation is the ratio of the absolute distance between the fused data and the center value of the dynamic threshold interval to half the width of the dynamic threshold interval. S33: Based on the calculated deviation, the deviation is mapped to an anomaly confidence level between 0 and 1 using a preset Gaussian distribution probability function, wherein the larger the deviation, the closer the anomaly confidence level is to 1, and the smaller the deviation, the closer the anomaly confidence level is to 0.
7. The method for data fusion and anomaly early warning of UHV substation handover test according to claim 6, characterized in that, The dynamic threshold interval is generated and updated dynamically. The update mechanism is as follows: whenever a new set of standardized test data that is confirmed to be normal is added to the historical handover test data learning set through feedback, the mean and standard deviation of the fusion data of all normal samples are automatically recalculated and the dynamic threshold interval is updated.
8. The method for data fusion and anomaly early warning of UHV substation handover test according to claim 7, characterized in that, S4 includes: S41: Based on the anomaly confidence level calculated in step S3, activate visual warning signals of different colors and levels on the monitoring interface. When the anomaly confidence level is less than 0.3, no warning signal is issued; when the anomaly confidence level is between 0.3 and 0.6, a blue warning signal is activated; when the anomaly confidence level is between 0.6 and 0.8, a yellow warning signal is activated; and when the anomaly confidence level is greater than 0.8, a red warning signal is activated. S42: The standardized test data set of this handover test and its corresponding anomaly confidence level are used as new samples. After verifying the actual equipment status confirmed by the operation and maintenance personnel on site, the new samples that are confirmed to be in normal status are added to the historical handover test data learning set. S43: Using the updated historical handover test data learning set, retrain the multidimensional association rule model and recalculate the dynamic threshold interval respectively, complete the iterative optimization of the multidimensional association rule model and the dynamic threshold interval, and use the optimized model and threshold interval for the next handover test data analysis.
9. The method for data fusion and anomaly early warning of UHV substation handover test according to claim 8, characterized in that, The visual warning signals include a colored warning window that pops up on the human-machine interface of the monitoring system, a flashing device icon, and a generated text description of the warning level.
10. The method for data fusion and anomaly early warning of UHV substation handover test according to claim 8, characterized in that, Upon activation of the yellow or red warning signal, an warning log record is automatically generated and pushed to the mobile terminal of the relevant maintenance personnel via a message queue.