A method and device for evaluating the state of equipment in a medium and low voltage power distribution system

Abstract

The present application belongs to the technical field of medium and low voltage power distribution system equipment state evaluation, and particularly relates to a medium and low voltage power distribution system equipment state evaluation method and device. The method is applied to an edge intelligent terminal, and comprises: receiving multi-source operation monitoring data of power distribution equipment, and constructing an adaptive data window based on a window fluctuation index and an edge load occupancy rate; performing double-constraint abnormality identification and adaptive similarity deduplication on the window data; generating a window cleaning quality confidence based on an abnormality proportion, a redundancy elimination proportion and a missing repair proportion, and generating a feature confidence of equipment operation features; generating a corrected comprehensive state index based on a feature value, the feature confidence and a load influence factor, and determining an equipment state grade in combination with a historical stable baseline, the load influence factor and a hysteresis determination condition. The present scheme can improve the accuracy and stability of equipment state evaluation.

Description

Technical Field

[0001] This invention belongs to the technical field of condition assessment of medium and low voltage power distribution system equipment, and specifically relates to a method and apparatus for condition assessment of medium and low voltage power distribution system equipment. Background Technology

[0002] In medium- and low-voltage power distribution systems, distribution equipment such as distribution transformers, power cables, switchgear, ring main units, and prefabricated substations are typically equipped with online monitoring devices to collect operational monitoring data such as voltage, current, power, temperature, partial discharge, vibration, insulation status, switch operation status, and ambient temperature and humidity. Existing technologies usually first clean the operational monitoring data, then extract equipment operating characteristics based on the cleaned data, and determine the operating status of the power distribution equipment through preset thresholds, evaluation rules, or condition evaluation models. However, the accuracy and stability of the equipment condition evaluation results from existing technologies still need improvement. Summary of the Invention

[0003] The purpose of this invention is to provide a method and apparatus for evaluating the condition of equipment in medium and low voltage power distribution systems, thereby solving the problem of low accuracy in equipment condition evaluation results in the prior art.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for evaluating the condition of equipment in a medium- and low-voltage power distribution system, applied to an edge intelligent terminal, the method comprising: Receive multi-source operation monitoring data of power distribution equipment in medium and low voltage power distribution systems, repair missing data in the multi-source operation monitoring data, and generate missing data repair markers; An adaptive data window is constructed based on the window fluctuation index corresponding to the multi-source operation monitoring data and the edge load occupancy rate of the edge intelligent terminal. The window length and sliding step size of the adaptive data window are dynamically adjusted according to the window fluctuation index and the edge load occupancy rate. Double-constraint anomaly identification and adaptive similarity deduplication are performed on the data within the adaptive data window to obtain cleaned window data, anomaly identification results, and redundancy deduplication results. Among them, double-constraint anomaly identification distinguishes and processes abnormal data based on data direction deviation and time continuity deviation, while adaptive similarity deduplication judges the similarity of window data and selects the best to retain based on the candidate redundancy screening results. Based on the anomaly identification results, redundancy removal results, and missing data repair markers, the anomaly ratio, redundancy removal ratio, and missing data repair ratio corresponding to the adaptive data window are determined, and the window cleaning quality confidence is generated based on the anomaly ratio, redundancy removal ratio, and missing data repair ratio. Based on the extraction of equipment operation characteristics from the window data after cleaning, feature confidence levels corresponding to the equipment operation characteristics are generated according to the confidence level of window cleaning quality, forming a feature evaluation input that includes feature values ​​and feature confidence levels; Based on the feature evaluation input and the load influence factor corresponding to the power distribution equipment, a modified comprehensive state index is generated. The feature confidence level is used to correct the contribution of the corresponding feature value in the state evaluation, and the load influence factor is used to correct the impact of load changes on the state evaluation results. The equipment status level of the power distribution equipment is determined based on the revised comprehensive status index, the historical stability baseline corresponding to the power distribution equipment, the load influence factor, and the hysteresis judgment conditions.

[0005] By receiving multi-source operational monitoring data at the edge intelligent terminal and constructing an adaptive data window based on window fluctuation indicators and edge load occupancy, the data window can simultaneously adapt to fluctuations in on-site operating conditions and the computing power status of the edge side, solving the problem that fixed windows cannot simultaneously consider the accuracy of abnormal response and the constraints of computing resources. By performing dual-constraint anomaly identification and adaptive similarity deduplication on the adaptive data window, it is possible to distinguish abnormal data, operating condition switching data, and redundant data while cleaning the data, reducing the risk of mistakenly deleting normal operating condition change data and retaining low-value redundant data. By converting the anomaly ratio, redundancy removal ratio, and missing data repair ratio into window cleaning quality confidence, and further generating feature confidence corresponding to equipment operating characteristics, the data cleaning results can substantially participate in subsequent status evaluation, solving the problem of the lack of quality transfer relationship between traditional data cleaning and status evaluation. By combining feature evaluation input, load influence factors, historical stable baselines, and hysteresis judgment conditions to determine the equipment status level, it is possible to reduce the interference of normal load changes and short-term fluctuations on the status evaluation results, thereby improving the accuracy, stability, and on-site adaptability of status evaluation of medium and low voltage power distribution system equipment.

[0006] Furthermore, based on the window fluctuation index corresponding to the multi-source operation monitoring data and the edge load occupancy rate of the edge intelligent terminal, an adaptive data window is constructed, including: A window volatility index is generated based on the variance characteristics, volatility characteristics, and local change gradients of the data within the candidate window. The edge load utilization rate is generated based on the ratio of the amount of data to be processed by the edge intelligent terminal per unit time to the maximum amount of data that can be processed per unit time. Based on the window fluctuation index and edge load occupancy rate, the basic window length and basic sliding step size are adjusted to obtain the window length and sliding step size of the adaptive data window.

[0007] By generating a window volatility index based on the variance, volatility, and local change gradients of the data within the candidate window, and generating an edge load occupancy rate based on the ratio of the current amount of data to be processed to the maximum amount of data that can be processed per unit time, the basic window length and basic sliding step size are adjusted accordingly. This ensures that the construction process of the adaptive data window is simultaneously constrained by the degree of data volatility and the edge computing load. This scheme can improve the window's sensitivity to abnormal changes when the running data fluctuates significantly, and reduce the risk of redundant computation and processing congestion when the edge intelligent terminal load is high, thereby improving the real-time performance and resource utilization efficiency of edge data processing.

[0008] Furthermore, double-constraint anomaly detection is performed on the data within the adaptive data window, including: For each sampling point within the adaptive data window, calculate the data direction deviation amount, which characterizes the degree of directional deviation of the sampling point in the multidimensional running data space; Calculate the temporal continuity deviation, which characterizes the degree of deviation of a sampling point from its continuous evolution relative to its preceding and following adjacent sampling points; A fusion anomaly score is generated based on the deviation of data direction and the deviation of time continuity. Based on the fusion anomaly score and time continuity deviation, the sampling points are differentiated into strong anomaly data, anomaly data to be observed, continuous operating condition switching points, or normal data.

[0009] By calculating the data direction deviation of sampling points in the multidimensional operational data space and the temporal continuity deviation of sampling points relative to their preceding and following adjacent sampling points, and generating a fused anomaly score based on these two types of deviations, anomaly identification no longer relies solely on a single numerical mutation or spatial deviation judgment. This scheme can simultaneously evaluate the spatial anomaly degree and temporal evolution rationality of sampling points, thereby improving the accuracy of anomaly identification and reducing the risk of misjudgment due to a single anomaly criterion.

[0010] Furthermore, based on the fusion anomaly score and temporal continuity deviation, the sampling points are differentiated into strongly anomalous data, anomaly data to be observed, continuous operating condition switching points, or normal data, including: When the fusion anomaly score meets the strong anomaly condition and the time continuity deviation meets the continuity deviation condition, the sampling point is determined as strong anomaly data. When the fusion anomaly score meets the observed anomaly condition but does not meet the strong anomaly condition, the sampling point is determined as the observed anomaly data. When the data direction deviation meets the direction deviation condition but the time continuity deviation does not meet the continuity deviation condition, the sampling point is determined as the continuous operating condition switching point, and the sampling point is retained with reduced weight. When a sampling point does not meet the criteria for determining strong abnormal data, abnormal data to be observed, or continuous operating condition switching point, the sampling point will be determined as normal data.

[0011] By using fusion anomaly scores, temporal continuity deviations, and data direction deviations, sampling points are categorized into strongly anomalous data, anomaly data to be observed, continuous operating condition switching points, or normal data. Continuous operating condition switching points are then retained with reduced weight, enabling tiered and differentiated handling of anomalies. This scheme avoids mistakenly deleting continuously changing processes such as load switching, equipment start-up and shutdown, and short-term operating condition transitions as anomalous data, while simultaneously identifying and processing strongly anomalous data, thereby improving the authenticity of data cleaning results and the reliability of subsequent status assessments.

[0012] Furthermore, adaptive similarity deduplication is performed on the data within the adaptive data window, including: Generate a lightweight statistical signature vector based on the statistical characteristics of the data sequences to be compared; Candidate redundant pairs are filtered from multiple data sequences to be compared based on lightweight statistical signature vectors; Calculate linear and nonlinear correlation indices for candidate redundant pairs, and generate an adaptive hybrid similarity based on the linear and nonlinear correlation indices; Redundancy judgment is performed on candidate redundant pairs based on adaptive hybrid similarity; For data sequences identified as redundant, a retention priority is generated based on the missing data repair ratio, outlier ratio, local stationarity score, and sensor status reliability. The data is then selected for retention based on the retention priority.

[0013] By first generating a lightweight statistical signature vector based on the statistical characteristics of the data sequences to be compared, and then using this lightweight statistical signature vector to filter candidate redundant pairs, the computational overhead of full pairwise similarity comparison on the edge side is reduced. By calculating linear and nonlinear correlation indices for the candidate redundant pairs and generating an adaptive hybrid similarity, both linear and nonlinear redundancy relationships can be considered. Furthermore, by generating a retention priority based on the missing data repair ratio, outlier ratio, local stationarity score, and sensor state reliability, higher-quality data sequences can be prioritized for retention among redundant data. This scheme improves the accuracy of redundancy deduplication and the efficiency of edge-side processing, while avoiding the retention of low-quality data during the deduplication process.

[0014] Furthermore, a confidence level for window cleaning quality is generated based on the proportion of anomalies, the proportion of redundancy removal, and the proportion of missing data repair, including: The total number of sampling points, outliers, redundant removal points, and missing repair points within the adaptive data window are counted. The anomaly percentage is generated based on the ratio of the number of outliers to the total number of sampling points. The redundancy removal percentage is generated based on the ratio of the number of redundant removal points to the total number of sampling points. The missing repair percentage is generated based on the ratio of the number of missing repair points to the total number of sampling points; Based on the negative correlation between the proportion of anomalies, the proportion of redundancy removal, and the proportion of missing data repair, a confidence level for window cleaning quality is generated.

[0015] By separately counting the total number of sampling points, the number of outliers, the number of redundant points removed, and the number of missing points repaired, and generating outlier percentages, redundant point removal percentages, and missing point repair percentages, and then using the negative correlation among these three factors to generate a window cleaning quality confidence score, the data quality information during the cleaning process can be quantitatively expressed. This solution solves the problem of traditional data cleaning methods that only output cleaned data and do not reflect the quality of the cleaning process. It enables subsequent status evaluation to identify whether the data basis of a certain window is reliable, thereby reducing the adverse impact of low-quality windows on the judgment of equipment status levels.

[0016] Furthermore, based on the confidence level of the window cleaning quality, feature confidence levels corresponding to the equipment operating characteristics are generated, including: Based on the confidence level of window cleaning quality, the continuity and consistency of equipment operation characteristics between adjacent adaptive data windows, and the credibility of the data source corresponding to the equipment operation characteristics, feature confidence levels are generated. Specifically, when the variation of device operating characteristics between adjacent adaptive data windows increases, or the confidence of the data source corresponding to the device operating characteristics decreases, the confidence of the features corresponding to the device operating characteristics is reduced.

[0017] By generating feature confidence scores based on window cleaning quality confidence, the continuity and consistency of equipment operating characteristics between adjacent adaptive data windows, and the reliability of the data source corresponding to the equipment operating characteristics, each equipment operating characteristic not only has a feature value but also a confidence score attribute reflecting data quality and source reliability. This scheme can reduce the role of corresponding features in the status assessment when the feature variation amplitude increases abnormally or the data source reliability decreases, avoiding excessive influence of abnormally changing features or low-reliability data sources on the comprehensive status assessment, thereby improving the stability and interpretability of the status assessment results.

[0018] Furthermore, based on the feature evaluation inputs and the load impact factors corresponding to the power distribution equipment, a modified comprehensive state index is generated, including: A comprehensive state index is generated based on each feature value, the basic weight corresponding to each feature value, and the feature confidence level corresponding to each feature value. A load impact factor is generated based on the deviation of the current current, current power, and current voltage of the power distribution equipment from the historical stable baseline. The comprehensive state index is modified based on the load influence factor to obtain the modified comprehensive state index.

[0019] By generating a comprehensive state index based on various eigenvalues, basic weights, and eigenvalue confidence levels, the contribution of different equipment operating characteristics to the state assessment is constrained by the reliability of the characteristics. Furthermore, by generating load influence factors based on the deviations of current current, current power, and current voltage from historical stable baselines, and using these load influence factors to correct the comprehensive state index, the non-fault-related increases in the state index caused by normal load changes can be suppressed. This scheme reduces the risk of misjudging normal load increases as equipment failure deterioration under high-load operating conditions and improves the ability of the state assessment to identify actual fault changes.

[0020] Furthermore, following the step of determining the equipment status level of the power distribution equipment, the following steps are also included: Upload summary information, including window cleaning quality confidence, feature confidence, corrected comprehensive status index and equipment status level, to the cloud, and receive parameter update vectors returned by the cloud based on the summary information; Based on the parameter update vector, a local parallel trial run verification is performed in the edge intelligent terminal to obtain the verification results corresponding to the new parameter group and the verification results corresponding to the original parameter group. When the verification result corresponding to the new parameter group is better than the verification result corresponding to the original parameter group, and the difference between the two meets the solidification condition, the new parameter group will be solidified as the current operating parameters of the edge intelligent terminal. When the verification result corresponding to the new parameter group is not better than the verification result corresponding to the original parameter group, or the difference between the two does not meet the solidification conditions, the system will revert to the original parameter group and upload the reason for the revert to the cloud.

[0021] By uploading summary information such as window cleaning quality confidence, feature confidence, corrected comprehensive status indicators, and device status levels to the cloud, the cloud can generate parameter update vectors based on long-term operational results at the edge. Local parallel trial runs are then performed on the edge intelligent terminal based on these parameter update vectors to obtain verification results for both the new and original parameter sets. The new parameter set is only solidified when it meets the solidification conditions; otherwise, it reverts to the original parameter set. This achieves closed-loop coordination between cloud optimization and stable edge operation. This scheme avoids abrupt changes in edge status evaluation results caused by direct parameter replacement from the cloud, improving the security, controllability, and alarm stability of the parameter update process.

[0022] In a second aspect, the present invention provides a condition evaluation device for medium and low voltage power distribution system equipment, applied to an edge intelligent terminal, comprising: The data acquisition module is used to receive multi-source operation monitoring data of power distribution equipment in medium and low voltage power distribution systems, repair missing data in the multi-source operation monitoring data, and generate missing data repair markers. The window construction module is used to construct an adaptive data window based on the window fluctuation index corresponding to the multi-source operation monitoring data and the edge load occupancy rate of the edge intelligent terminal. The window length and sliding step size of the adaptive data window are dynamically adjusted according to the window fluctuation index and the edge load occupancy rate. The data cleaning module is used to perform dual-constraint anomaly identification and adaptive similarity deduplication on the data within the adaptive data window, resulting in cleaned window data, anomaly identification results, and redundancy deduplication results. Dual-constraint anomaly identification distinguishes and processes abnormal data based on deviations in data direction and time continuity. Adaptive similarity deduplication judges the similarity of window data and selects the best to retain based on the candidate redundancy screening results. The cleaning quality confidence generation module is used to determine the proportion of anomalies, the proportion of redundancy removal, and the proportion of missing data repair corresponding to the adaptive data window based on the anomaly identification results, the redundancy removal results, and the missing data repair markers, and to generate the window cleaning quality confidence based on the anomaly proportion, the redundancy removal proportion, and the missing data repair proportion. The feature evaluation input generation module is used to extract equipment operation features based on the window data after cleaning, and generate feature confidence levels corresponding to the equipment operation features according to the window cleaning quality confidence level, forming feature evaluation inputs including feature values ​​and feature confidence levels; The status index generation module is used to generate a modified comprehensive status index based on the feature evaluation input and the load influence factor corresponding to the power distribution equipment. The feature confidence is used to correct the contribution of the corresponding feature value in the status evaluation, and the load influence factor is used to correct the impact of load changes on the status evaluation results. The status level determination module is used to determine the equipment status level of the power distribution equipment based on the modified comprehensive status index, the historical stability baseline corresponding to the power distribution equipment, the load influence factor, and the hysteresis judgment condition. Attached Figure Description

[0023] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart of a method for evaluating the condition of equipment in a medium- and low-voltage power distribution system according to an embodiment of the present invention; Figure 2 This is a structural block diagram of a medium- and low-voltage power distribution system equipment condition evaluation device according to an embodiment of the present invention; Figure 3 This is a structural block diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0024] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0025] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0026] Example 1 like Figure 1 As shown, this embodiment provides a method for evaluating the condition of equipment in a medium- and low-voltage power distribution system. This method is applied to an edge intelligent terminal and includes the following steps: S1. Receive multi-source operation monitoring data of power distribution equipment in medium and low voltage power distribution system, repair missing data in multi-source operation monitoring data, and generate missing data repair markers; S2. Based on the window fluctuation index corresponding to the multi-source operation monitoring data and the edge load occupancy rate of the edge intelligent terminal, construct an adaptive data window, wherein the window length and sliding step size of the adaptive data window are dynamically adjusted according to the window fluctuation index and the edge load occupancy rate. S3. Perform dual-constraint anomaly identification and adaptive similarity deduplication on the data within the adaptive data window to obtain the cleaned window data, anomaly identification results, and redundancy deduplication results. Dual-constraint anomaly identification distinguishes and processes abnormal data based on data direction deviation and time continuity deviation. Adaptive similarity deduplication judges the similarity of the window data and selects the best to retain based on the candidate redundancy screening results. S4. Based on the anomaly identification results, redundancy removal results, and missing data repair markers, determine the anomaly ratio, redundancy removal ratio, and missing data repair ratio corresponding to the adaptive data window, and generate the window cleaning quality confidence score based on the anomaly ratio, redundancy removal ratio, and missing data repair ratio. S5. Extract equipment operation characteristics based on window data after cleaning, generate feature confidence levels corresponding to equipment operation characteristics based on window cleaning quality confidence levels, and form feature evaluation inputs including feature values ​​and feature confidence levels; S6. Based on the feature evaluation input and the load influence factor corresponding to the power distribution equipment, generate a modified comprehensive state index. The feature confidence level is used to correct the contribution of the corresponding feature value in the state evaluation, and the load influence factor is used to correct the impact of load changes on the state evaluation results. S7. Determine the equipment status level of the power distribution equipment based on the modified comprehensive status index, the historical stability baseline corresponding to the power distribution equipment, the load influence factor, and the hysteresis judgment conditions.

[0027] The above-mentioned solution addresses the challenges of multi-source heterogeneity, high noise levels, numerous missing and redundant data points, rapid changes in operating conditions, and limited computing power at the edge of medium- and low-voltage power distribution systems. Instead of simply outputting cleaned data, it continues to transmit anomaly identification results, redundancy removal results, and missing data repair markers from the data cleaning process to subsequent condition assessment processes. By generating window cleaning quality confidence scores and further forming feature confidence scores corresponding to equipment operating characteristics, the cleaning quality can contribute to the correction of condition indicators. Introducing load influence factors reduces the non-fault-related interference of normal load changes on equipment condition assessment. Combining historical stability baselines and hysteresis judgment conditions to determine equipment condition levels reduces frequent state jumps caused by fluctuations in a single window. Therefore, this solution improves the stability, reliability, and adaptability of condition assessment for medium- and low-voltage power distribution equipment under the condition of limited computing power at the edge intelligent terminal.

[0028] Specifically, multi-source operation monitoring data may include one or more of the following: voltage data, current data, power data, load data, equipment temperature data, partial discharge data, vibration data, insulation status data, switch operation status data, ambient temperature and humidity data, and monitoring device status data.

[0029] For any sampling time The operational monitoring data can be represented as:

[0030] in, Represents voltage data. Represents current data. Represents power data. This indicates the equipment temperature data. This represents partial discharge data. Representing vibration data, Indicates other operating status quantities or insulation status quantities.

[0031] The edge intelligent terminal performs time alignment, format unification, dimension normalization, and missing data repair on multi-source operation monitoring data, and saves the repair results in association with missing data repair markers, so that the confidence level of subsequent cleaning quality can distinguish between the original valid sampled values ​​and the missing data repaired values.

[0032] Furthermore, when constructing adaptive data windows, edge intelligent terminals comprehensively consider the degree of fluctuation in operating data and the current processing pressure of the edge intelligent terminal. Greater fluctuations in operating data indicate a stronger risk of short-term changes or anomalies within the window; in this case, the window length and sliding step size can be appropriately shortened to improve anomaly response accuracy. Higher edge load occupancy indicates more strain on the edge intelligent terminal's current computing resources; adjusting the window length and sliding step size can reduce redundant computation overhead. Therefore, the window partitioning process is constrained by both data state and computing power state, providing data units more adapted to the current operating conditions for subsequent anomaly identification, deduplication, and state evaluation.

[0033] Furthermore, dual-constraint anomaly identification is used to simultaneously examine the directional deviation of sampling points in the multidimensional data space and their continuity deviation in the time series. If a sampling point differs significantly in direction from other sampling points in the multidimensional space and also disrupts the continuous evolution relationship relative to its preceding and following adjacent sampling points, then this sampling point is more likely to be an isolated anomaly. If a sampling point has a large directional deviation but a low temporal continuity deviation, then this sampling point may correspond to load switching, equipment start-up and shutdown, or short-term operating condition transition, and should not be directly deleted. Through the above differentiation and processing, this scheme can reduce the risk of misjudging normal continuous operating condition transitions as anomalous data.

[0034] Furthermore, adaptive similarity deduplication is used to reduce redundant data in multi-source operational monitoring data. The edge intelligent terminal first filters candidate redundant pairs based on statistical characteristics, and then judges the similarity of these pairs. For data sequences determined to be redundant, the edge intelligent terminal selects which to retain based on data integrity, anomaly degree, stationarity, and sensor status reliability. This ensures that the redundancy removal process reduces computational and storage overhead while retaining higher-quality data sequences. The redundancy removal results are then used to calculate the confidence level of the window cleaning quality, making the deduplication process have a traceable impact on subsequent status evaluations.

[0035] Furthermore, the confidence level of window cleaning quality decreases as the proportion of anomalies, redundancy removal, and missing data repair increases. In other words, even if a window can be cleaned to form cleaned window data, if the original data of that window contains a large number of anomalies, redundancy removals, or missing data repairs, the reliability of the data foundation corresponding to that window remains low. By further converting the confidence level of window cleaning quality into feature confidence, the contribution of features generated by low-quality data can be reduced during subsequent state evaluation, avoiding excessive influence of low-quality windows on the equipment state level.

[0036] Furthermore, equipment operating characteristics can include general statistical characteristics and equipment-specific characteristics. General statistical characteristics can include mean characteristics, variance characteristics, and volatility characteristics. For distribution transformers, equipment-specific characteristics can include hot spot temperature characteristics, partial discharge pulse characteristics, vibration spectrum characteristics, and load rate characteristics; for power cables, equipment-specific characteristics can include transient high-frequency change characteristics, partial discharge energy distribution characteristics, insulation state change characteristics, and load current fluctuation characteristics; for switchgear, equipment-specific characteristics can include opening and closing action time characteristics, operating mechanism vibration characteristics, contact temperature rise characteristics, and partial discharge characteristics. The edge intelligent terminal represents each equipment operating characteristic as a combination of characteristic value and characteristic confidence level, enabling the condition evaluation process to simultaneously consider the magnitude of the characteristic value and the confidence level of the characteristic.

[0037] Furthermore, the revised comprehensive status index can be generated based on eigenvalues, eigenvalue confidence levels, eigenvalue basis weights, and load impact factors. The eigenvalue confidence level is used to adjust the contribution of the corresponding eigenvalue to the comprehensive evaluation, while the load impact factor describes the deviation of load-related quantities such as current, power, and voltage from the historical stable baseline. When the increase in operating quantities such as temperature, current, and power is mainly caused by an increase in normal load, the load impact factor can suppress the non-fault amplification of the comprehensive status index, thereby reducing false alarms under high load conditions.

[0038] Furthermore, the equipment status level can include levels such as normal, abnormal, and alarm. The edge intelligent terminal can determine a historical stable baseline based on historical window data showing high cleaning quality confidence, stable status, and no valid alarms, and generate a dynamic status threshold by combining this with current load influencing factors. Hysteresis determination conditions can include entry status thresholds, exit status thresholds, and consecutive window quantity conditions. By separating the entry and exit thresholds and requiring that the status level switch only occur after the corrected comprehensive status index meets the corresponding conditions for multiple consecutive windows, the problem of frequent upward or downward adjustments caused by short-term fluctuations can be reduced.

[0039] In one embodiment, after determining the equipment status level of the power distribution equipment, the edge intelligent terminal can also upload summary information including window cleaning quality confidence, feature confidence, corrected comprehensive status index and equipment status level to the cloud, and receive parameter update vectors returned by the cloud based on the summary information; perform local parallel trial operation verification in the edge intelligent terminal based on the parameter update vectors to obtain the verification results corresponding to the new parameter group and the original parameter group, and perform parameter solidification or parameter rollback on the new parameter group according to the verification results corresponding to the new parameter group and the original parameter group.

[0040] Specifically, the cloud can perform long-term statistical analysis based on summary information uploaded by multiple edge intelligent terminals or over multiple periods, generating parameter update vectors for optimizing window adjustment, anomaly detection, deduplication, status evaluation, dynamic thresholds, and load correction. After receiving the parameter update vectors, the edge intelligent terminals simultaneously run the original parameter group and the new parameter group within a local trial operation cycle, obtaining status evaluation results corresponding to the original parameter group and the new parameter group, respectively. The edge intelligent terminals then determine whether the new parameter group is suitable for permanent implementation based on the continuity of the status sequence, alarm stability, and consistency with manual review results, historical alarm results, or equipment maintenance results.

[0041] The above implementation method uses cloud-based statistical optimization, edge-side local parallel trial operation, parameter verification, parameter solidification or rollback to ensure that the cloud-based optimization results are verified for stability on the edge intelligent terminal side before taking effect. This avoids sudden changes in device status levels caused by direct replacement of cloud parameters, thereby improving the continuity and controllability of edge-side status evaluation results.

[0042] In one embodiment, receiving multi-source operation monitoring data of power distribution equipment in a medium- and low-voltage power distribution system, repairing missing data in the multi-source operation monitoring data, and generating missing data repair markers may include: receiving multi-source operation monitoring data of distribution transformers, power cables, switchgear, ring main units, or prefabricated substations; timestamping, binding equipment identifiers, binding monitoring point identifiers, and binding data source identifiers to the multi-source operation monitoring data to generate basic data records; repairing missing data in the multi-source operation monitoring data, and generating missing data repair markers based on the repair results, wherein the missing data repair markers are used to characterize whether the corresponding data is missing data that has been repaired.

[0043] Specifically, basic data records can include at least the equipment identifier, monitoring point identifier, sampling time, data type, data value, data source, and sensor status identifier. The equipment identifier and monitoring point identifier can determine the power distribution equipment to which the data belongs and the specific monitoring location; the sampling time can establish the time series relationship of the data; and the data source and sensor status identifier can provide a basis for subsequent calculations of data source reliability and feature confidence.

[0044] Furthermore, regarding missing moments Data values, if their adjacent times and All samples have valid values, and preliminary repair can be performed using linear interpolation.

[0045] in, This represents the data value after the missing information has been repaired. and These represent the valid sampled values ​​at adjacent times before and after the missing time point. For the original valid sampled points, the missing data repair marker can be denoted as m. i =0; For missing repair points, the missing repair marker can be denoted as m. i =1. The missing repair flag will continue to be used in subsequent calculations of the missing repair percentage and the confidence level of window cleaning quality.

[0046] The above scheme generates basic data records and missing data repair markers simultaneously during the data reception phase, enabling each sampled value to be associated with the equipment, monitoring point, time, source, and repair status. This solves the problem of repaired data being mistaken for the original true sampled value in subsequent evaluations, thereby improving the traceability of data quality evaluation and equipment status evaluation.

[0047] In one embodiment, constructing an adaptive data window based on the window fluctuation index corresponding to multi-source operation monitoring data and the edge load occupancy rate of the edge intelligent terminal may include: generating a window fluctuation index based on the variance characteristics, volatility characteristics, and local change gradient of the data within the candidate window; generating an edge load occupancy rate based on the ratio of the amount of data to be processed by the edge intelligent terminal per unit time to the maximum amount of data that can be processed per unit time; and adjusting the basic window length and basic sliding step size based on the window fluctuation index and the edge load occupancy rate to obtain the window length and sliding step size of the adaptive data window.

[0048] Specifically, for the nth candidate window W n It can calculate the window volatility index. :

[0049] in, This represents the variance characteristics after normalization. This represents the normalized volatility characteristic. This represents the normalized local gradient. , , This indicates the weight of the volatility indicator.

[0050] Edge load occupancy It can be represented as:

[0051] in, This indicates the amount of data to be processed or the current computing task on the edge intelligent terminal per unit of time. This indicates the maximum amount of data that an edge intelligent terminal can process or the maximum amount of computing tasks it can handle per unit of time.

[0052] Furthermore, let the basic window length be... The basic sliding step size is Then the window length of the nth adaptive data window It can be represented as:

[0053] in, This represents the window fluctuation adjustment coefficient. Indicates the edge load adjustment coefficient. and These represent the minimum and maximum allowed window lengths, respectively.

[0054] The sliding step size of the nth adaptive data window It can be represented as:

[0055] in, This represents the adjustment coefficient of the fluctuation to the sliding step size. This represents the adjustment coefficient of the load on the sliding step size. and These represent the minimum and maximum allowed sliding step sizes, respectively. `clip` restricts the calculation results to between the corresponding minimum and maximum values.

[0056] The above scheme incorporates both window fluctuation metrics and edge load occupancy into the window construction process, enabling the window length and sliding step size to adaptively adjust according to changes in operating conditions and edge computing power status. When operating data fluctuates significantly, it improves the accuracy of anomaly response; when edge load is high, it reduces overlapping computational load and processing pressure, thus balancing real-time status evaluation with edge-side resource constraints.

[0057] In one embodiment, performing dual-constraint anomaly identification on data within an adaptive data window may include: calculating, for each sampling point within the adaptive data window, a data direction deviation amount used to characterize the degree of directional deviation of the sampling point in the multidimensional operating data space; calculating, a time continuity deviation amount used to characterize the degree of continuous evolution deviation of the sampling point relative to its preceding and following adjacent sampling points; generating a fusion anomaly score based on the data direction deviation amount and the time continuity deviation amount; and identifying anomalies in the sampling points based on the fusion anomaly score.

[0058] Specifically, for the adaptive data window W n Any sampling point within It can calculate the deviation of data direction. :

[0059] Where N represents the number of sampling points participating in the comparison within the window. It represents a small positive quantity that prevents the denominator from being zero; This represents the Euclidean norm. The deviation in data direction. Used to characterize sampling points The degree of directional deviation relative to other sampling points in the multidimensional operating data space. , 1,1) means restricting the input value to [ Within the interval [1,1].

[0060] Furthermore, the deviation in time continuity can be calculated. :

[0061] in, It is used to determine whether the current sampling point disrupts the continuous evolution relationship between adjacent sampling points. This represents a small positive quantity to prevent the denominator from being zero. For sampling points at the window boundaries, calculations can be performed using adjacent points on one side, or these points can be excluded from the calculation of time continuity deviation.

[0062] Edge intelligent terminals can respectively and After normalization, we get and And generate a fusion anomaly score. :

[0063] in, and Indicates the fusion weight, and + =1.

[0064] The above scheme improves the ability of anomaly identification to distinguish between isolated anomalies and continuous operating condition changes by simultaneously introducing data direction deviation and time continuity deviation. This is achieved by introducing both data direction deviation and time continuity deviation.

[0065] In one embodiment, anomaly identification of sampling points based on fused anomaly scores may include: When the fusion anomaly score meets the strong anomaly condition and the time continuity deviation meets the continuity deviation condition, the sampling point is determined as strong anomaly data. When the fusion anomaly score meets the observed anomaly condition but does not meet the strong anomaly condition, the sampling point is determined as the observed anomaly data. When the data direction deviation meets the direction deviation condition but the time continuity deviation does not meet the continuity deviation condition, the sampling point is determined as the continuous operating condition switching point, and the sampling point is retained with reduced weight. When a sampling point does not meet the criteria for determining strong abnormal data, abnormal data to be observed, or continuous operating condition switching point, the sampling point will be determined as normal data.

[0066] Specifically, a strong anomaly threshold can be set. Observable abnormal threshold Time continuity deviates from the threshold and data direction deviation from threshold . This is used to determine whether the directional deviation of the sampling point in the multidimensional running data space meets the directional deviation condition.

[0067] when and At that time, sampling points can be Data identified as strongly anomalous is removed or isolated for storage. when At that time, sampling points can be Data identified as abnormal and to be observed is retained but with an additional abnormal weight label. when and At that time, sampling points can be Points identified as potential continuous operating condition switching points are not deleted directly, but retained with reduced weight; when At that time, sampling points can be The data is considered normal.

[0068] It should be noted that strong anomalies typically exhibit both spatial deviation and temporal continuity disruption. While continuous operating condition switching points may differ from historical states in data direction, the changes before and after them still maintain continuity. Directly removing continuous operating condition switching points could lead to the loss of true operating condition change information in subsequent condition assessments. Therefore, this embodiment adopts a reduced-weight retention approach, allowing them to participate in condition assessments but with a reduced contribution.

[0069] The above solution reduces the chance of accidental deletion of data during normal load switching, equipment start-up and shutdown, or short-term transition by setting differentiated processing rules for strong anomalies, anomalies to be observed, and continuous operating condition switching points. At the same time, it isolates or removes strong anomaly data, thereby improving the confidence level of subsequent cleaning quality and the reliability of status evaluation results.

[0070] In one embodiment, performing adaptive similarity deduplication on data within an adaptive data window may include: generating a lightweight statistical signature vector based on the statistical characteristics of the data sequences to be compared; filtering candidate redundant pairs from multiple data sequences to be compared based on the lightweight statistical signature vector; calculating linear correlation and nonlinear correlation indices for the candidate redundant pairs, and generating an adaptive hybrid similarity based on the linear correlation and nonlinear correlation indices; and performing redundancy judgment on the candidate redundant pairs based on the adaptive hybrid similarity.

[0071] Specifically, for the data sequence X to be compared, a lightweight statistical signature vector can be generated:

[0072] in, Represents the mean of the sequence. Represents the series variance. Represents the volatility of the sequence. This indicates the percentage of missing data repaired in the sequence.

[0073] For any two data sequences X and Y to be compared, when If the similarity is high, (X,Y) is included in the candidate redundant pair set; otherwise, it is directly determined that the two do not belong to the highly similar candidate pair and are not included in the fine similarity calculation. This indicates the candidate selection threshold.

[0074] Furthermore, for candidate redundant pairs (X,Y), the Pearson correlation coefficient can be calculated:

[0075] And calculate mutual information:

[0076] in, Describes the joint probability distribution. Represents the marginal probability distribution. It represents a small positive quantity that prevents abnormalities in the logarithmic term or denominator.

[0077] After normalizing the mutual information, we get:

[0078] in, This represents the maximum mutual information value in the current set of candidate redundant pairs. Further calculation of the adaptive hybrid weights is then performed.

[0079] And generate adaptive hybrid similarity:

[0080] When satisfied When this happens, it can be determined that data sequences X and Y constitute redundant data sequences. This represents the similarity threshold.

[0081] The above scheme first reduces the number of data sequences that need to be finely compared by using lightweight statistical signature vectors, and then identifies redundant data by adaptive fusion of linear correlation indicators and nonlinear correlation indicators. It can take into account both linear and nonlinear redundancy scenarios and reduce the computational overhead of full pairwise comparisons at edge intelligent terminals.

[0082] In one embodiment, the preferred retention in adaptive similarity deduplication may include: generating a retention priority for data sequences determined to be redundant based on the missing data repair ratio, outlier ratio, local stationarity score, and sensor state confidence; retaining data sequences with higher retention priority and deleting, compressing, or delaying the upload of data sequences with lower retention priority; and including the amount of data that is deleted, compressed, or delayed in the upload as part of the redundancy deduplication result.

[0083] Specifically, for a data sequence X, the retention priority Q(X) can be calculated:

[0084] in, Indicates the percentage of missing parts repaired. Indicates the proportion of outliers. Indicates the local stationarity score. Indicates the reliability of the sensor status. , , , This indicates the priority weight for retention. Edge intelligent terminals can retain data sequences with higher Q values ​​from redundant data sequences, and delete, compress, or delay the upload of data sequences with lower Q values.

[0085] It should be noted that data sequences with lower missing data repair rates, lower outlier rates, better local stability, and more reliable sensor status generally have higher data quality and better value for status assessment. Selecting data to retain based on retention priorities can avoid retaining low-quality data sequences during the deduplication process.

[0086] The above scheme introduces retention priority in redundancy deduplication, which not only reduces data redundancy and edge processing pressure, but also prioritizes the retention of more reliable data sequences. At the same time, the amount of data that is deleted, compressed, or delayed in uploading is included in the redundancy deduplication result, which is beneficial for subsequent accurate calculation of the redundancy removal ratio and the confidence level of window cleaning quality.

[0087] In one embodiment, generating a window cleaning quality confidence score based on the anomaly ratio, redundancy removal ratio, and missing data repair ratio may include: statistically analyzing the total number of sampling points, the number of anomalies, the number of redundant removal points, and the number of missing data repair points within the adaptive data window; generating the anomaly ratio based on the ratio of the number of anomalies to the total number of sampling points; generating the redundancy removal ratio based on the ratio of the number of redundant removal points to the total number of sampling points; generating the missing data repair ratio based on the ratio of the number of missing data repair points to the total number of sampling points; and generating a window cleaning quality confidence score based on the negative correlation between the anomaly ratio, the redundancy removal ratio, and the missing data repair ratio.

[0088] Specifically, let the adaptive data window W be... n The total number of sampling points is The sum of the number of strong outliers and the converted number of outliers to be observed is The number of redundant removal points is The number of missing repair points is ,in, The number of redundant deduplication points can be determined based on the redundancy removal results. When adaptive similarity deduplication is performed on a data sequence basis, the number of sampling points contained in the deleted, compressed, or delayed uploaded data sequence is converted into the number of redundant removal points. .

[0089] Then they can be calculated separately:

[0090]

[0091]

[0092] in, Indicates the percentage of abnormalities. This indicates the percentage of redundant items removed. This indicates the percentage of missing repairs. Edge intelligent terminals can generate window cleaning quality confidence scores based on these three ratios. :

[0093] in, , , These represent the influence coefficients of the percentage of anomalies, the percentage of redundant removals, and the percentage of missing repairs on the confidence level of the cleaning quality, respectively.

[0094] The above scheme maps the anomaly rate, redundancy removal rate, and missing data repair rate to a unified window cleaning quality confidence score, enabling the data quality information generated during the cleaning process to be quantitatively incorporated into subsequent status evaluations. The more anomalies, redundancy removals, or missing data repairs, the lower the window cleaning quality confidence score, thus reducing the impact of low-quality windows on equipment status evaluation results.

[0095] In one embodiment, generating feature confidence scores corresponding to device operation features based on window cleaning quality confidence scores may include: generating feature confidence scores based on window cleaning quality confidence scores, the continuity and consistency of device operation features between adjacent adaptive data windows, and the reliability of the data source corresponding to the device operation features; wherein, when the variation of device operation features between adjacent adaptive data windows increases, or the reliability of the data source corresponding to the device operation features decreases, the feature confidence scores corresponding to the device operation features are reduced.

[0096] Specifically, edge intelligent terminals can represent the operational characteristics of each device as a combination of feature values ​​and feature confidence levels:

[0097] in, This represents the feature evaluation input set corresponding to the nth adaptive data window. The feature value representing the operational characteristic of the k-th device. Let M represent the feature confidence of the k-th device operation feature in the n-th adaptive data window, and M represent the number of device operation features.

[0098] Feature confidence can be determined jointly by window cleaning quality confidence, feature continuity consistency, and the confidence of the corresponding sensor or data source:

[0099] in, This indicates the confidence level in the quality of window cleaning. This represents the continuity and consistency of the k-th feature within the adjacent windows. This represents the reliability of the sensor or data source corresponding to the k-th feature. Continuity consistency can be expressed as:

[0100] in, This represents the k-th feature value of the n-th adaptive data window. This represents the k-th feature value of the previous adaptive data window.

[0101] The above scheme further transfers the confidence level of window cleaning quality to the feature layer, so that each device operation feature not only has a feature value, but also a feature confidence level that reflects the quality of the data source, the quality of window cleaning, and the continuity of adjacent windows. Therefore, during condition evaluation, the feature contribution of abnormal jumps, poor sensor conditions, or low cleaning quality can be reduced, improving the stability and interpretability of the evaluation results.

[0102] In one embodiment, generating a modified comprehensive state index based on feature evaluation inputs and load impact factors corresponding to power distribution equipment may include: generating a comprehensive state index based on each feature value, the basic weight corresponding to each feature value, and the feature confidence level corresponding to each feature value; generating load impact factors based on the deviation of the current current, current power, and current voltage of the power distribution equipment from the historical stable baseline; and modifying the comprehensive state index based on the load impact factors to obtain the modified comprehensive state index.

[0103] Specifically, the edge intelligent terminal can first normalize the feature values ​​to obtain... A comprehensive state index is generated based on each feature value and the confidence level of each feature. :

[0104] in, This represents the basic weight of the k-th feature. This represents the feature confidence level of the k-th feature. This represents the k-th normalized eigenvalue.

[0105] Furthermore, load impact factor It can be represented as:

[0106] in, , , These represent the current, power, and voltage statistics for the current window, respectively. , , These represent the historical stable baseline values, , , These represent the rated current, rated power, and rated voltage, respectively. , , This indicates the load impact weight. Edge intelligent terminals can generate a modified comprehensive status index based on the load impact factor. :

[0107] in, This represents the load correction factor.

[0108] It should be noted that the load influence factor is used here to reduce the non-fault amplification of the state indicators caused by the load; the load influence factor is introduced into the dynamic state threshold to make the state level threshold adaptively adjust with the current load condition. The two factors act on the indicator side and the threshold side, respectively.

[0109] The above scheme reduces the interference of low-confidence features and normal load fluctuations on equipment condition evaluation by using feature confidence as a correction factor for condition indicators and further using load influence factors to correct the comprehensive condition indicators. Especially in high-load operation scenarios, when the increase in temperature, current, or power is mainly caused by load changes, the load influence factor can suppress the non-fault amplification of condition indicators, thereby reducing the risk of misjudgment.

[0110] In one embodiment, determining the equipment status level of the power distribution equipment based on the modified comprehensive status index, the historical stable baseline corresponding to the power distribution equipment, the load influence factor, and the hysteresis judgment condition may include: selecting a set of windows from the historical window data of the power distribution equipment whose cleaning quality confidence level meets the confidence condition, whose status is stable, and whose no effective alarms have occurred; generating a historical stable baseline and historical baseline dispersion based on the window set; generating a dynamic status threshold for distinguishing different equipment status levels based on the historical stable baseline, historical baseline dispersion, and load influence factor; and determining the equipment status level of the power distribution equipment based on the comparison result between the modified comprehensive status index and the dynamic status threshold.

[0111] Specifically, the edge intelligent terminal can select a set of windows that historically have high confidence in cleaning quality, stable status, and no effective alarms for the power distribution equipment d, and calculate the historical stable baseline mean of the equipment. and historical baseline dispersion Based on historical stable baselines, historical baseline dispersion, and current load impact factors, dynamic state thresholds can be generated.

[0112]

[0113] in, This represents a dynamic threshold between normal and abnormal states. This represents a dynamic threshold between abnormal and alarm states. , , , This represents the threshold adjustment parameter, and , This is to ensure that the dynamic threshold between abnormal and alarm states is higher than the dynamic threshold between normal and abnormal states.

[0114] Edge intelligent terminals can correct comprehensive status indicators The equipment status level of the power distribution equipment is determined by comparing it with a dynamic status threshold.

[0115] The above scheme establishes a historical stable baseline by utilizing historical window data with high cleaning quality and stable status, enabling the dynamic status threshold to reflect the historical operating characteristics of specific power distribution equipment. At the same time, by introducing load influence factors into the dynamic status threshold, the threshold can be adaptively adjusted according to the current operating conditions, thereby improving the adaptability of status level determination under different equipment, different load levels, and different operating stages.

[0116] In one embodiment, determining the equipment status level of the power distribution equipment based on the comparison result between the modified comprehensive status index and the dynamic status threshold may include: setting an entry status threshold and an exit status threshold for switching between adjacent equipment status levels, wherein the entry status threshold is higher than the exit status threshold; when the modified comprehensive status index meets the entry status threshold for multiple consecutive adaptive data windows, the equipment status level of the power distribution equipment is increased; when the modified comprehensive status index meets the exit status threshold for multiple consecutive adaptive data windows, the equipment status level of the power distribution equipment is decreased.

[0117] Specifically, for the transition between normal and abnormal states, an abnormal state threshold can be set. and exit exception threshold ,in To switch between abnormal and alarm states, an alarm threshold can be set. and exit alarm threshold ,in When adjusting the comprehensive state index K1 consecutive adaptive data windows greater than When this happens, the device status can be changed from normal to abnormal; when K2 consecutive adaptive data windows greater than When this happens, the device status can be upgraded from abnormal to alarm; when K3 consecutive adaptive data windows below When this happens, the device status can be downgraded from alarm to abnormal; when K4 consecutive adaptive data windows below When this happens, the device status can be downgraded from abnormal to normal. K1, K2, K3, and K4 represent the threshold number of consecutive windows.

[0118] It should be noted that different thresholds are used for adjusting the status level upwards and downwards, and combined with the number of consecutive windows, a hysteresis interval can be formed. When the overall status index fluctuates briefly near the threshold, the equipment status level will not switch repeatedly immediately; the status level adjustment is only performed when multiple consecutive windows meet the entry or exit conditions.

[0119] The above solution, through the combination of entry state threshold, exit state threshold and continuous window quantity conditions, reduces frequent state level jumps caused by short-term data fluctuations, instantaneous load disturbances or single window anomalies, thereby improving alarm stability and operational reliability.

[0120] In one embodiment, after determining the equipment status level of the power distribution equipment, a fault association score can be generated based on the feature value and feature confidence level; the fault type corresponding to the power distribution equipment can be determined according to the fault association score; when the fault association scores corresponding to multiple fault types meet the proximity condition, a composite risk warning is output.

[0121] Specifically, for the m-th type of fault, a fault association score can be calculated. :

[0122] in, This represents the correlation coefficient between the m-th type of fault and the k-th feature. This represents the feature confidence level of the k-th feature. This represents the k-th normalized feature value. When the fault association score of a certain fault type is the highest and exceeds the fault mapping threshold, this fault type can be used as the preferred fault mapping result; when the fault association scores of multiple fault types are close, a composite risk warning can be output instead of forcing a single fault classification.

[0123] Furthermore, for distribution transformers, if the hot spot temperature characteristic has a high confidence level and continues to rise, a thermal fault risk can be output; if the partial discharge pulse characteristic and vibration spectrum characteristic are both abnormal, and the corresponding characteristic confidence levels are high, a combined risk of discharge fault and mechanical abnormality can be output. For power cables, if the high-frequency abrupt change characteristic rises briefly but the deviation in time continuity is low, it can be preferentially identified as an operating condition disturbance; if the high-frequency abrupt change characteristic and energy distribution characteristic are continuously abnormal, and the characteristic confidence levels are high, an insulation defect or partial discharge evolution risk can be output.

[0124] The above solution incorporates feature confidence into the fault type mapping, ensuring that fault attribution depends not only on the magnitude of feature values ​​but also on the reliability of those values. For scenarios where multiple fault types have similar scores, outputting composite risk warnings can avoid premature and one-sided classification to a single fault type, thus improving the robustness of operational recommendations.

[0125] In one embodiment, uploading summary information to the cloud, including window cleaning quality confidence, feature confidence, corrected comprehensive status index and device status level, may include: uploading summary information to the cloud according to a preset period, wherein the summary information also includes at least one of window fluctuation index, edge load occupancy rate, anomaly ratio, redundancy removal ratio, missing repair ratio, dynamic status threshold, fault type and alarm record.

[0126] Specifically, edge intelligent terminals can periodically upload locally computed summary information to the cloud. The summary information does not necessarily contain all the original high-frequency data, but can include statistical information reflecting window construction, data cleaning, feature evaluation, state classification, and fault mapping results. For example, the summary information may include window fluctuation indicators. Edge load occupancy Abnormal proportion Redundancy removal ratio , percentage of missing repairs Confidence level of window cleaning quality Feature confidence Correcting the comprehensive status index Dynamic status thresholds, device status levels, fault types, and alarm records.

[0127] Furthermore, the cloud can perform long-term statistical analysis based on summary information uploaded from multiple periods, multiple devices, or multiple edge intelligent terminals to identify the relationship between parameter settings and the stability of state evaluation, and to provide a basis for the generation of subsequent parameter update vectors. Since summary information is uploaded instead of the full original data, communication load and cloud storage pressure can be reduced.

[0128] The above solution enables the cloud to obtain key statistical information such as edge data cleaning quality, status evaluation results, and alarm stability by periodically uploading summary information, providing a data foundation for cloud-edge closed-loop parameter correction. At the same time, the summary uploading method reduces the need for high-frequency raw data transmission, improving the communication efficiency and scalability of the system in medium and low voltage power distribution field applications.

[0129] In one embodiment, receiving a parameter update vector transmitted back from the cloud based on summary information may include: receiving a parameter update vector generated by the cloud based on summary information from multiple periods, wherein the parameter update vector includes at least one of window adjustment parameters, anomaly identification parameters, deduplication parameters, state evaluation weights, dynamic threshold parameters, and load correction parameters.

[0130] Specifically, the cloud can generate parameter update vectors based on summary information from multiple periods. :

[0131] in, It can be used as a window adjustment parameter; It can be used as an anomaly detection parameter; It can be used as a deduplication parameter; It can be used as a weight for state evaluation; It can be used as a dynamic threshold parameter; These parameters can be used as load correction parameters. The above parameter update vectors can be generated based on long-term statistical results from the cloud, and can be used to improve the subsequent window construction, anomaly identification, deduplication, state evaluation, and dynamic threshold grading processes of edge intelligent terminals.

[0132] Furthermore, the parameter update vector does not necessarily need to include all parameters each time. When the cloud optimizes only for a certain type of device, a certain type of operating condition, or a certain type of abnormal scenario, the parameter update vector can include only the parameters related to that optimization objective. After receiving the parameter update vector, the edge intelligent terminal can use it as a new parameter group to enter a local parallel trial run for verification.

[0133] The above solution transforms long-term statistical results in the cloud into parameter update vectors, enabling edge intelligent terminals to obtain optimization results formed by cross-cycle, cross-device, or cross-regional data statistics. At the same time, by limiting the parameter update content to parameters related to window adjustment, anomaly identification, deduplication, status evaluation, dynamic threshold, and load correction, it can maintain consistency between the main link of cloud optimization and edge-side status evaluation.

[0134] In one embodiment, performing local parallel trial operation verification based on parameter update vectors in an edge intelligent terminal may include: generating an original state sequence based on the original parameter set and generating a new state sequence based on the new parameter set corresponding to the parameter update vectors during the trial operation period; generating verification results corresponding to the new parameter set and verification results corresponding to the original parameter set based on the continuity of the state sequences, alarm stability, and consistency with manual review results, historical alarm results, or equipment maintenance results of the original and new state sequences.

[0135] Specifically, the edge intelligent terminal can use the new parameter set for local parallel trial operation without affecting the current output state evaluation results. The original parameter set is used to generate the original state sequence, and the new parameter set is used to generate the new state sequence. The edge intelligent terminal can calculate the verification score G for each of the two state sequences separately.

[0136] in, Represents the continuity score of the state sequence. This indicates the consistency score with manual review results, historical alarm results, or equipment maintenance results. This indicates the alarm stability score. , , This indicates the verification weight. The state sequence continuity score can be used to evaluate whether the state level changes are smooth; the consistency score can be used to evaluate whether the state evaluation results are consistent with existing review or maintenance information; the alarm stability score can be used to evaluate whether there are frequent jitters or abnormal increases or decreases in alarm output.

[0137] It should be noted that while the new parameter set generated in the cloud may be superior in the long-term statistical sense, it may still cause abrupt changes in the state evaluation results under the specific device, sensor status, and local operating conditions of the edge intelligent terminal. Therefore, parallel trial operation and stability verification are needed at the edge before determining whether to solidify the parameter set.

[0138] The above solution generates both the original state sequence and the new state sequence simultaneously within the edge intelligent terminal, and verifies them based on continuity, consistency, and alarm stability. This allows the parameter update process to have an on-site adaptation and verification step, which can reduce the impact of cloud parameter updates on the stability of edge-side state evaluation.

[0139] In one embodiment, performing parameter fixing or parameter rollback on the new parameter group based on the verification results corresponding to the new parameter group and the original parameter group may include: when the verification result corresponding to the new parameter group is better than the verification result corresponding to the original parameter group, and the difference between the two meets the fixing condition, fixing the new parameter group as the current operating parameters of the edge intelligent terminal; when the verification result corresponding to the new parameter group is not better than the verification result corresponding to the original parameter group, or the difference between the two does not meet the fixing condition, rolling back to the original parameter group and uploading the rollback reason to the cloud.

[0140] Specifically, the verification score corresponding to the new parameter set can be recorded as follows: The verification score corresponding to the original parameter set is recorded as .when Greater than ,and and The difference exceeds the curing threshold. At this time, the edge intelligent terminal will solidify the new parameter set as the current operating parameters, enabling subsequent window construction, anomaly detection, deduplication, state evaluation, and state classification processes to be executed based on the new parameter set. Not greater than ,or and The difference did not exceed the curing threshold. At this time, the edge intelligent terminal does not fix the new parameter set, but continues to use the original parameter set and uploads the reason for the rollback to the cloud.

[0141] Furthermore, rollback reasons can include information such as reduced continuity of state sequences due to the new parameter set, decreased alarm stability, and insufficient consistency with manual review results or historical alarm results. After receiving the rollback reasons in the cloud, they can be used as input for subsequent long-term statistics and parameter re-optimization.

[0142] The above solution, by setting clear parameters for fixation and rollback, ensures that cloud-based parameter updates do not directly replace edge-side operating parameters, but rather take effect only after performance comparison at the edge. This leverages the long-term statistical optimization capabilities of the cloud while maintaining the continuity, stability, and field adaptability of edge intelligent terminal status evaluation results.

[0143] Example 2 like Figure 2 As shown, based on the same inventive concept as the above embodiments, the present invention also provides a medium- and low-voltage power distribution system equipment condition evaluation device, applied to an edge intelligent terminal, comprising: The data acquisition module is used to receive multi-source operation monitoring data of power distribution equipment in medium and low voltage power distribution systems, repair missing data in the multi-source operation monitoring data, and generate missing data repair markers. The window construction module is used to construct an adaptive data window based on the window fluctuation index corresponding to the multi-source operation monitoring data and the edge load occupancy rate of the edge intelligent terminal. The window length and sliding step size of the adaptive data window are dynamically adjusted according to the window fluctuation index and the edge load occupancy rate. The data cleaning module is used to perform dual-constraint anomaly identification and adaptive similarity deduplication on the data within the adaptive data window, resulting in cleaned window data, anomaly identification results, and redundancy deduplication results. Dual-constraint anomaly identification distinguishes and processes abnormal data based on deviations in data direction and time continuity. Adaptive similarity deduplication judges the similarity of window data and selects the best to retain based on the candidate redundancy screening results. The cleaning quality confidence generation module is used to determine the proportion of anomalies, the proportion of redundancy removal, and the proportion of missing data repair corresponding to the adaptive data window based on the anomaly identification results, the redundancy removal results, and the missing data repair markers, and to generate the window cleaning quality confidence based on the anomaly proportion, the redundancy removal proportion, and the missing data repair proportion. The feature evaluation input generation module is used to extract equipment operation features based on the window data after cleaning, and generate feature confidence levels corresponding to the equipment operation features according to the window cleaning quality confidence level, forming feature evaluation inputs including feature values ​​and feature confidence levels; The status index generation module is used to generate a modified comprehensive status index based on the feature evaluation input and the load influence factor corresponding to the power distribution equipment. The feature confidence is used to correct the contribution of the corresponding feature value in the status evaluation, and the load influence factor is used to correct the impact of load changes on the status evaluation results. The status level determination module is used to determine the equipment status level of the power distribution equipment based on the modified comprehensive status index, the historical stability baseline corresponding to the power distribution equipment, the load influence factor, and the hysteresis judgment condition.

[0144] Example 3 like Figure 3 As shown, the present invention also provides an electronic device 100 for implementing a method for evaluating the condition of equipment in a medium- and low-voltage power distribution system; The electronic device 100 includes a memory 101, at least one processor 102, a computer program 103 stored in the memory 101 and executable on at least one processor 102, and at least one communication bus 104.

[0145] The memory 101 can be used to store computer program 103. The processor 102 implements the steps of the medium and low voltage power distribution system equipment condition evaluation method of Embodiment 1 by running or executing the computer program stored in the memory 101 and calling the data stored in the memory 101.

[0146] The memory 101 may primarily include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created based on the use of the electronic device 100 (such as audio data), etc. In addition, the memory 101 may include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other non-volatile solid-state storage device.

[0147] At least one processor 102 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 102 may be a microprocessor or any conventional processor. Processor 102 is the control center of electronic device 100, connecting various parts of electronic device 100 via various interfaces and lines.

[0148] The memory 101 in the electronic device 100 stores multiple instructions to implement a method for evaluating the status of equipment in a medium- and low-voltage power distribution system, and the processor 102 can execute multiple instructions to achieve the following: Receive multi-source operation monitoring data of power distribution equipment in medium and low voltage power distribution systems, repair missing data in the multi-source operation monitoring data, and generate missing data repair markers; An adaptive data window is constructed based on the window fluctuation index corresponding to the multi-source operation monitoring data and the edge load occupancy rate of the edge intelligent terminal. The window length and sliding step size of the adaptive data window are dynamically adjusted according to the window fluctuation index and the edge load occupancy rate. Double-constraint anomaly identification and adaptive similarity deduplication are performed on the data within the adaptive data window to obtain cleaned window data, anomaly identification results, and redundancy deduplication results. Among them, double-constraint anomaly identification distinguishes and processes abnormal data based on data direction deviation and time continuity deviation, while adaptive similarity deduplication judges the similarity of window data and selects the best to retain based on the candidate redundancy screening results. Based on the anomaly identification results, redundancy removal results, and missing data repair markers, the anomaly ratio, redundancy removal ratio, and missing data repair ratio corresponding to the adaptive data window are determined, and the window cleaning quality confidence is generated based on the anomaly ratio, redundancy removal ratio, and missing data repair ratio. Based on the extraction of equipment operation characteristics from the window data after cleaning, feature confidence levels corresponding to the equipment operation characteristics are generated according to the confidence level of window cleaning quality, forming a feature evaluation input that includes feature values ​​and feature confidence levels; Based on the feature evaluation input and the load influence factor corresponding to the power distribution equipment, a modified comprehensive state index is generated. The feature confidence level is used to correct the contribution of the corresponding feature value in the state evaluation, and the load influence factor is used to correct the impact of load changes on the state evaluation results. The equipment status level of the power distribution equipment is determined based on the revised comprehensive status index, the historical stability baseline corresponding to the power distribution equipment, the load influence factor, and the hysteresis judgment conditions.

[0149] Example 4 If the modules / units integrated in the electronic device 100 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, and read-only memory (ROM).

[0150] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0151] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0152] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0153] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0154] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0155] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for evaluating the condition of equipment in a medium- and low-voltage power distribution system, applied to edge intelligent terminals, characterized in that, The methods include: Receive multi-source operation monitoring data of power distribution equipment in medium and low voltage power distribution systems, repair missing data in the multi-source operation monitoring data, and generate missing data repair markers; An adaptive data window is constructed based on the window fluctuation index corresponding to the multi-source operation monitoring data and the edge load occupancy rate of the edge intelligent terminal. The window length and sliding step size of the adaptive data window are dynamically adjusted according to the window fluctuation index and the edge load occupancy rate. Double-constraint anomaly identification and adaptive similarity deduplication are performed on the data within the adaptive data window to obtain cleaned window data, anomaly identification results, and redundancy deduplication results. Among them, double-constraint anomaly identification distinguishes and processes abnormal data based on data direction deviation and time continuity deviation, while adaptive similarity deduplication judges the similarity of window data and selects the best to retain based on the candidate redundancy screening results. Based on the anomaly identification results, redundancy removal results, and missing data repair markers, the anomaly ratio, redundancy removal ratio, and missing data repair ratio corresponding to the adaptive data window are determined, and the window cleaning quality confidence is generated based on the anomaly ratio, redundancy removal ratio, and missing data repair ratio. Based on the extraction of equipment operation characteristics from the window data after cleaning, feature confidence levels corresponding to the equipment operation characteristics are generated according to the confidence level of window cleaning quality, forming a feature evaluation input that includes feature values ​​and feature confidence levels; Based on the feature evaluation input and the load influence factor corresponding to the power distribution equipment, a modified comprehensive state index is generated. The feature confidence level is used to correct the contribution of the corresponding feature value in the state evaluation, and the load influence factor is used to correct the impact of load changes on the state evaluation results. The equipment status level of the power distribution equipment is determined based on the revised comprehensive status index, the historical stability baseline corresponding to the power distribution equipment, the load influence factor, and the hysteresis judgment conditions.

2. The method for evaluating the condition of medium and low voltage power distribution system equipment according to claim 1, characterized in that, Based on the window fluctuation index corresponding to multi-source operation monitoring data and the edge load occupancy rate of edge intelligent terminals, an adaptive data window is constructed, including: A window volatility index is generated based on the variance characteristics, volatility characteristics, and local change gradients of the data within the candidate window. The edge load utilization rate is generated based on the ratio of the amount of data to be processed by the edge intelligent terminal per unit time to the maximum amount of data that can be processed per unit time. Based on the window fluctuation index and edge load occupancy rate, the basic window length and basic sliding step size are adjusted to obtain the window length and sliding step size of the adaptive data window.

3. The method for evaluating the condition of medium and low voltage power distribution system equipment according to claim 1, characterized in that, Perform double-constraint anomaly detection on the data within the adaptive data window, including: For each sampling point within the adaptive data window, calculate the data direction deviation amount, which characterizes the degree of directional deviation of the sampling point in the multidimensional running data space; Calculate the temporal continuity deviation, which characterizes the degree of deviation of a sampling point from its continuous evolution relative to its preceding and following adjacent sampling points; A fusion anomaly score is generated based on the deviation of data direction and the deviation of time continuity. Based on the fusion anomaly score and time continuity deviation, the sampling points are differentiated into strong anomaly data, anomaly data to be observed, continuous operating condition switching points, or normal data.

4. The method for evaluating the condition of medium and low voltage power distribution system equipment according to claim 3, characterized in that, Based on the fusion anomaly score and temporal continuity deviation, the sampling points are differentiated into strongly anomalous data, anomaly data to be observed, continuous operating condition switching points, or normal data, including: When the fusion anomaly score meets the strong anomaly condition and the time continuity deviation meets the continuity deviation condition, the sampling point is determined as strong anomaly data. When the fusion anomaly score meets the observed anomaly condition but does not meet the strong anomaly condition, the sampling point is determined as the observed anomaly data. When the data direction deviation meets the direction deviation condition but the time continuity deviation does not meet the continuity deviation condition, the sampling point is determined as the continuous operating condition switching point, and the sampling point is retained with reduced weight. When a sampling point does not meet the criteria for determining strong abnormal data, abnormal data to be observed, or continuous operating condition switching point, the sampling point will be determined as normal data.

5. The method for evaluating the condition of medium and low voltage power distribution system equipment according to claim 1, characterized in that, Perform adaptive similarity deduplication on the data within the adaptive data window, including: Generate a lightweight statistical signature vector based on the statistical characteristics of the data sequences to be compared; Candidate redundant pairs are filtered from multiple data sequences to be compared based on lightweight statistical signature vectors; Calculate linear and nonlinear correlation indices for candidate redundant pairs, and generate an adaptive hybrid similarity based on the linear and nonlinear correlation indices; Redundancy judgment is performed on candidate redundant pairs based on adaptive hybrid similarity; For data sequences identified as redundant, a retention priority is generated based on the missing data repair ratio, outlier ratio, local stationarity score, and sensor status reliability. The data is then selected for retention based on the retention priority.

6. The method for evaluating the condition of medium and low voltage power distribution system equipment according to claim 1, characterized in that, A confidence level for window cleaning quality is generated based on the proportion of anomalies, the proportion of redundancy removal, and the proportion of missing data repair, including: The total number of sampling points, outliers, redundant removal points, and missing repair points within the adaptive data window are counted. The anomaly percentage is generated based on the ratio of the number of outliers to the total number of sampling points. The redundancy removal percentage is generated based on the ratio of the number of redundant removal points to the total number of sampling points. The missing repair percentage is generated based on the ratio of the number of missing repair points to the total number of sampling points; Based on the negative correlation between the proportion of anomalies, the proportion of redundancy removal, and the proportion of missing data repair, a confidence level for window cleaning quality is generated.

7. The method for evaluating the condition of medium and low voltage power distribution system equipment according to claim 1, characterized in that, Based on the confidence level of the window cleaning quality, feature confidence levels corresponding to the equipment operating characteristics are generated, including: Based on the confidence level of window cleaning quality, the continuity and consistency of equipment operation characteristics between adjacent adaptive data windows, and the credibility of the data source corresponding to the equipment operation characteristics, feature confidence levels are generated. Specifically, when the variation of device operating characteristics between adjacent adaptive data windows increases, or the confidence of the data source corresponding to the device operating characteristics decreases, the confidence of the features corresponding to the device operating characteristics is reduced.

8. The method for evaluating the condition of medium and low voltage power distribution system equipment according to claim 1, characterized in that, Based on the feature evaluation inputs and the load impact factors corresponding to the power distribution equipment, a modified comprehensive state index is generated, including: A comprehensive state index is generated based on each feature value, the basic weight corresponding to each feature value, and the feature confidence level corresponding to each feature value. A load impact factor is generated based on the deviation of the current current, current power, and current voltage of the power distribution equipment from the historical stable baseline. The comprehensive state index is modified based on the load influence factor to obtain the modified comprehensive state index.

9. The method for evaluating the condition of medium and low voltage power distribution system equipment according to claim 1, characterized in that, Following the step of determining the equipment status level of power distribution equipment, the following steps are also included: Upload summary information, including window cleaning quality confidence, feature confidence, corrected comprehensive status index and equipment status level, to the cloud, and receive parameter update vectors returned by the cloud based on the summary information; Based on the parameter update vector, a local parallel trial run verification is performed in the edge intelligent terminal to obtain the verification results corresponding to the new parameter group and the verification results corresponding to the original parameter group. When the verification result corresponding to the new parameter group is better than the verification result corresponding to the original parameter group, and the difference between the two meets the solidification condition, the new parameter group will be solidified as the current operating parameters of the edge intelligent terminal. When the verification result corresponding to the new parameter group is not better than the verification result corresponding to the original parameter group, or the difference between the two does not meet the solidification conditions, the system will revert to the original parameter group and upload the reason for the revert to the cloud.

10. A condition evaluation device for medium and low voltage power distribution system equipment, applied to an edge intelligent terminal, characterized in that, include: The data acquisition module is used to receive multi-source operation monitoring data of power distribution equipment in medium and low voltage power distribution systems, repair missing data in the multi-source operation monitoring data, and generate missing data repair markers. The window construction module is used to construct an adaptive data window based on the window fluctuation index corresponding to the multi-source operation monitoring data and the edge load occupancy rate of the edge intelligent terminal. The window length and sliding step size of the adaptive data window are dynamically adjusted according to the window fluctuation index and the edge load occupancy rate. The data cleaning module is used to perform dual-constraint anomaly identification and adaptive similarity deduplication on the data within the adaptive data window, resulting in cleaned window data, anomaly identification results, and redundancy deduplication results. Dual-constraint anomaly identification distinguishes and processes abnormal data based on deviations in data direction and time continuity. Adaptive similarity deduplication judges the similarity of window data and selects the best to retain based on the candidate redundancy screening results. The cleaning quality confidence generation module is used to determine the proportion of anomalies, the proportion of redundancy removal, and the proportion of missing data repair corresponding to the adaptive data window based on the anomaly identification results, the redundancy removal results, and the missing data repair markers, and to generate the window cleaning quality confidence based on the anomaly proportion, the redundancy removal proportion, and the missing data repair proportion. The feature evaluation input generation module is used to extract equipment operation features based on the window data after cleaning, and generate feature confidence levels corresponding to the equipment operation features according to the window cleaning quality confidence level, forming feature evaluation inputs including feature values ​​and feature confidence levels; The status index generation module is used to generate a modified comprehensive status index based on the feature evaluation input and the load influence factor corresponding to the power distribution equipment. The feature confidence is used to correct the contribution of the corresponding feature value in the status evaluation, and the load influence factor is used to correct the impact of load changes on the status evaluation results. The status level determination module is used to determine the equipment status level of the power distribution equipment based on the modified comprehensive status index, the historical stability baseline corresponding to the power distribution equipment, the load influence factor, and the hysteresis judgment condition.

Images