Power equipment state evaluation method based on normal cloud model and gray fixed weight clustering
By fusing and cleaning power equipment condition monitoring data using a normal cloud model and grey weighted clustering method, the problem of evaluating multi-source data is solved, enabling accurate assessment of power equipment condition and anomaly detection, and supporting safe equipment operation.
Patent Information
- Application Number
- CN202511628986.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-07
- Publication Date
- 2026-04-07
AI Technical Summary
Existing power equipment condition assessment methods lack effective data fusion and cleaning techniques when processing multi-source, heterogeneous condition monitoring data, which limits the reliability and practicality of the assessment results and makes it impossible to respond quickly to changes in equipment condition.
The normal cloud model is used to effectively fuse and clean the power equipment condition monitoring data. The grey weighted clustering method is used to conduct in-depth analysis of the fused data. By constructing a key parameter system and an anomaly detection mechanism, the accurate assessment of the power equipment condition can be achieved.
It enables accurate assessment of the condition of power equipment, allowing for accurate classification and evaluation with limited data samples, timely detection of abnormal conditions, providing a theoretical basis for condition-based maintenance, and ensuring the safe operation of equipment.
Smart Images

Figure CN121808427A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the field of electric digital data processing, and particularly relates to a power equipment state evaluation method based on a normal cloud model and a grey fixed-weight clustering. BACKGROUND
[0002] With the continuous progress of smart grid technology, the state monitoring and evaluation of power equipment has become a key link to ensure the safe and stable operation of the power grid. The construction of smart grid not only promotes the rapid development of sensor technology and state monitoring technology, but also makes the power grid data volume show an explosive growth trend.
[0003] Under this background, the introduction of big data technology provides an unprecedented opportunity for power equipment state evaluation. By integrating shared power transmission and transformation equipment state monitoring data, as well as power grid operation data and meteorological environment data closely related to the state of the equipment, the power equipment state evaluation is gradually developing towards information integration and comprehensive analysis based on panoramic state monitoring.
[0004] However, in practical applications, the diversity of power equipment state monitoring parameters, the difference in acquisition methods and the variability of parameter formats have brought great challenges to data transmission, storage and analysis processing.
[0005] For the equipment state evaluation method, there are currently three major categories based on traditional detection technology, data analysis and expert system. Periodic preventive test and online monitoring in traditional detection technology have their own advantages and disadvantages; statistical analysis and machine learning algorithm based on data analysis require a large amount of high-quality data support and high algorithm requirements; the evaluation method based on expert system has certain problems in subjectivity and standardization.
[0006] The invention patent application with the application publication date of September 23, 2025 and the application publication number of CN 120688337 A discloses "a power equipment state evaluation method, system, device and storage medium for intelligent monitoring", which includes: based on the working characteristic parameters of the power equipment, a multi-physical field coupling calculation model is constructed to simulate the behavior of the power equipment under different working states; the expected output value is output, the physical field signal after aligning the expected output value and time is compared, the deviation is calculated, and the initial prediction residual is obtained; according to the initial prediction residual, the model parameters of the multi-physical field coupling calculation model are adjusted through the incremental correction algorithm, the power equipment state estimation value is corrected, and the initial prediction residual is updated; based on the updated prediction residual, the abnormal state of the power equipment is analyzed through the residual value, and the fault feature is extracted; the fault type of the power equipment is identified. The technical solution significantly improves the accuracy and real-time performance of the power equipment state evaluation through multi-physical field coupling calculation and incremental correction algorithm. However, it is based on the multi-physical field coupling calculation model to simulate the behavior of the power equipment under different working states, which lacks effective data fusion and cleaning means, resulting in certain limitations on the reliability and practicality of the evaluation results.
[0007] The invention patent with the grant announcement date of September 1, 2023 and the grant announcement number of CN 116499531 B discloses "a power equipment state evaluation method, device, electronic equipment and storage medium", which fully considers the correlation between oil temperature and power transformer load loss when evaluating the heat dissipation state of the power transformer, calculates the load loss of the power transformer as the evaluation reference object, realizes the targeted evaluation of the heat dissipation state, significantly improves the scientific and reasonable of the evaluation process, greatly reduces the incidence of invalid evaluation, thereby providing effective protection for the reliability of the evaluation results. At the same time, by combining the monitoring of the running cooperation state of the heat dissipation component to evaluate the heat dissipation state of the power transformer, the evaluation method is more in line with the actual situation, which reduces the incidence of evaluation errors, and is conducive to improving the accuracy of the heat dissipation state evaluation of the power transformer. Since this technical solution evaluates the heat dissipation state of the transformer according to the correlation between the oil temperature and the load loss of the power transformer, it is not applicable to other types of power equipment (mainly refers to the power equipment without oil cooling device or unrelated to oil cooling), which limits its application range.
[0008] In addition, although the existing power equipment state evaluation method can realize the monitoring and evaluation of the equipment state to a certain extent, it often has problems such as large amount of calculation, inaccurate evaluation results, and inability to quickly respond to changes in equipment state. Especially when dealing with multi-source and heterogeneous state monitoring data, these methods often lack effective data fusion and cleaning means, which limits the reliability and practicality of the evaluation results.
[0009] Therefore, how to effectively integrate and clean the condition monitoring data of power equipment, and conduct in-depth analysis of the integrated data to achieve accurate assessment of the condition of power equipment, has become an urgent technical problem to be solved. Summary of the Invention
[0010] The purpose of this invention is to provide a power equipment condition assessment method based on a normal cloud model and grey weighted clustering. It effectively fuses and cleans power equipment condition monitoring data using a normal cloud model, and then uses grey weighted clustering to conduct in-depth analysis of the fused data, thereby achieving an accurate assessment of the power equipment's condition. Because the normal cloud model effectively handles the fuzziness and randomness of the data, the data fusion process is more consistent with reality; while the grey weighted clustering method can achieve accurate classification and assessment of equipment condition even with limited data samples. Therefore, the technical solution of this invention can perform fusion analysis and in-depth mining of condition monitoring data, enabling rapid detection of abnormal conditions and personalized condition evaluation, understanding the health status of key performance indicators, providing a basis for condition-based maintenance, and supporting the safe operation of equipment.
[0011] The technical solution of this invention is: to provide a power equipment condition assessment method based on a normal cloud model and grey weighted clustering, characterized by the following steps:
[0012] Step S10) Based on the research and analysis of current power equipment condition monitoring technology, establish a basic parameter system for condition assessment applicable to the overall and basic unit condition assessment of power equipment; remove parameters with low correlation to faults and defects, retain parameters with high correlation, and construct a key parameter system for condition assessment.
[0013] Step S20) Check the quality of the data in the key parameter system in step S10, and perform data cleaning if the data quality does not meet the requirements.
[0014] Step S30) Establish a data analysis model, use the correlation between vertical and horizontal data to judge abnormal changes in equipment status and its related relationships, and perform equipment anomaly detection by analyzing the correlation and change patterns of multidimensional status data;
[0015] Step S40) Based on the relevant characteristic parameters of the power transmission and transformation equipment, determine the characteristic parameters related to the key performance and their coupling relationship with the key performance status of the equipment, evaluate each key performance, screen out the key performance that has anomalies, and obtain the evaluation value of each key performance.
[0016] Specifically, in step S10, the key parameter system of power transmission and transformation equipment is divided into three layers: status information layer, sensor data layer, and application interaction layer. The status quantities of the status information layer include offline information, online monitoring, live detection, manual inspection, preventive experiment, operation information, ledger information, and environmental information. The interaction parameters of the application interaction layer include geographical information of scheduling step S10, operation information, meteorological information, and lightning monitoring information.
[0017] Furthermore, in step S10, key parameters are selected according to the following steps:
[0018] The basic parameter is quantized into a four-dimensional vector. The quantized vector of parameter A is represented as A = (A1, A2, A3, A4). Then: A1 represents the proportion of parameter A in the annual fault statistics; A2 represents the proportion of parameter A in the annual major / urgent defect statistics; A3 represents the proportion of parameter A in the annual general defect statistics; A4 represents the proportion of parameter A in the defect database of the production management system.
[0019] The weight vector H of the basic parameters is normalized to [0,1] to obtain the key parameter extraction method: the larger the weight, the stronger the correlation and the more representative the basic parameter. The weight is used as the basis for key parameter extraction, and the basic parameters with weights above 0.5 are selected as key parameters.
[0020] Furthermore, after selecting key parameters, the confidence and support between sequences are calculated using association rules, time series with similar rates of change or obvious correlations are extracted, parameters with correlation between sequences and their changes are extracted, and a key parameter system for state assessment is constructed.
[0021] Furthermore, the Apriori algorithm is used to calculate the confidence and support between sequences in the following steps:
[0022] (1) Scan the entire database T and calculate the number of times each item in T appears. In this process, generate a set C1 of first-order candidate options.
[0023] (2) Set a minimum support, delete frequent itemsets with less than the minimum support from C1, and retain the remaining frequent itemsets to generate a first-order frequent itemset L1.
[0024] (3) Use L1 to generate candidate second-order candidate itemsets C2, and then scan the database to count the itemsets in C2;
[0025] (4) Determine the frequent phase set L2 from the candidate set C2 based on the minimum support;
[0026] (5) Use L1 and L2 to generate candidate third-order candidate itemsets C3, and then scan the transaction database to count the itemsets in C3;
[0027] (6) Determine the frequent phase set L3 from the candidate set C3 based on the minimum support;
[0028] (7) Repeat this process until the number of frequent phase sets reaches the upper limit or the frequent phase sets are empty, at which point the algorithm ends.
[0029] Specifically, in step S20, the data includes cleanable data and non-cleanable data, and the cleanable data is cleaned; the data cleaning steps include:
[0030] Step S21) First, based on the characteristics of the power transmission and transformation equipment data, fit the corresponding time label for each type of data;
[0031] Step S22) For each state variable's unit time label, fit its data within a certain time period, analyze the residual sequence to determine whether there is abnormal data, and if there is abnormal data, determine whether to use outlier values or intervened data.
[0032] Step S23) Combined with April correlation analysis, the time series of state variables with strong correlation are analyzed simultaneously. Data cleaning is judged according to the abnormal patterns set between the series, and the characteristics of abnormal time series are extracted and matched one by one with the abnormal patterns.
[0033] Furthermore, the cleanable data in step S20 includes:
[0034] Erroneous data: Data that is spatially off-target or temporally isolated;
[0035] Missing values: Data is incomplete, resulting from communication interruption or database storage abnormalities.
[0036] Communication noise or external interference: a situation where the rate of change exceeds the normal range within a short period of time;
[0037] Bad data caused by short-term communication interruption: data changes very little or has a variance of 0 over a period of time;
[0038] Temporary horizontal migration data: A small amount of data migrates horizontally relative to the normal value within a certain period of time, and then returns to the normal state.
[0039] Furthermore, the unwashable data in step S20 includes:
[0040] High noise or high variance abnormal data: The variance of long-term measurement data is much greater than that of historical data. The anomaly is caused by damage to the acquisition sensor, hardware failure of the acquisition device, or low battery. The acquisition device has malfunctioned.
[0041] Sudden change amounts that strongly correlated sequences all show within a certain period: If sudden change amounts occur in the sequences simultaneously or with a lag of N periods, it indicates that the monitored parameter has been affected by the external environment, which may be external interference or equipment anomaly.
[0042] Horizontal migration of strongly correlated sequences after a certain period: If a permanent horizontal migration occurs in the sequences simultaneously or with a lag of N periods, it indicates that the equipment state is abnormal.
[0043] Accelerated upward trend of strongly correlated sequences after a certain period: If an accelerated upward trend occurs in the sequences simultaneously or with a lag of N periods, it indicates that the equipment state is abnormal.
[0044] Furthermore, anomaly detection is performed on the non-cleanable data, and the anomaly detection includes the following steps:
[0045] Step S31): For the historical data of each parameter, fit an autoregressive model and train a self-organizing step SOM to obtain the transition probability matrix {X1, X2... Xn} of a single parameter;
[0046] Step S32): Input the historical data of all parameters into the K-mean step S algorithm to obtain m normal clustering clusters;
[0047] Step S33): For the real-time data stream to be measured, convert the data stream of a single parameter into a transition probability sequence through the transition probability matrix, cluster the overall data stream, and determine whether the parameter data at each time point belongs to the m normal clustering clusters;
[0048] Step S34): Anomaly detection rules for the data stream to be measured:
[0049] If there are no 0 values in the transition probability sequences of all parameters, and all the data in the overall data stream belong to the m normal clustering clusters, it is determined that the data is normal.
[0050] If there are only a small number of 0 values in the transition probability sequences of only a few parameters, and there is a small amount of data in the data stream that does not belong to the m normal clustering clusters, it is determined that there is a small amount of noise in the data, and these noises belong to sensor anomalies that can be ignored.
[0051] When there is a period of 0 values in the transition probability sequences of k parameters (k < N), and a large segment of data in the data stream does not belong to the m normal clustering clusters, it indicates that the parameters of the device have anomalies and the device operating state is abnormal;
[0052] For the abnormal operating state of the device, based on the zero-crossing points of the transition probability sequences of the k parameters, determine the time when the parameter anomaly occurs.
[0053] Specifically, in step S40, the fuzzy comprehensive evaluation method is used to evaluate each key performance according to the following steps:
[0054] Step S41) Establish the factor set for the evaluation object;
[0055] Step S42) Establish a set of comments;
[0056] Step S43) Establish fuzzy evaluation matrices at each level;
[0057] Step S44) Combining the weight matrix and the fuzzy evaluation matrix, select the weighted average comprehensive evaluation to obtain the comprehensive evaluation result;
[0058] Step S45) Process the comprehensive evaluation results from step S44 to obtain the final evaluation results.
[0059] The power equipment condition assessment method based on normal cloud model and grey weighted clustering described in this invention effectively integrates and cleans power equipment condition monitoring data through normal cloud model, and uses grey weighted clustering method to conduct in-depth analysis of the integrated data to achieve accurate assessment of power equipment condition.
[0060] Compared with the prior art, the advantages of the present invention are:
[0061] 1. The technical solution of this invention effectively fuses and cleans power equipment condition monitoring data using a normal cloud model, and then uses a grey weighted clustering method to conduct in-depth analysis of the fused data. This allows for the fusion analysis and deep mining of multi-source power equipment condition monitoring data to achieve accurate assessment of the power equipment condition. Because the normal cloud model effectively handles the fuzziness and randomness of the data, the data fusion process is more consistent with reality; while the grey weighted clustering method can achieve accurate classification and assessment of equipment condition with limited data samples.
[0062] 2. The technical solution of the present invention performs data cleaning on washable data and anomaly detection on non-washable data, which can comprehensively, timely and accurately grasp the health status of key performance, provide a theoretical basis for condition-based maintenance, and provide strong support for ensuring the safe operation of equipment.
[0063] 3. The technical solution of this invention integrates and deeply mines multi-source data from power equipment condition monitoring, enabling rapid detection of abnormal states and personalized condition assessment. It can comprehensively, timely, and accurately grasp the health status of key performance indicators, providing a theoretical basis for condition-based maintenance and strong support for ensuring the safe operation of equipment. Attached Figure Description
[0064] Figure 1 This is a flowchart illustrating the steps of the power equipment condition assessment method of the present invention. Detailed Implementation
[0065] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0066] This invention provides a method for assessing the condition of power equipment based on a normal cloud model and grey weighted clustering. The method effectively fuses and cleans power equipment condition monitoring data using a normal cloud model, and then employs grey weighted clustering to conduct in-depth analysis of the fused data. This approach integrates and deeply mines multi-source power equipment condition monitoring data to achieve accurate assessment of the equipment's condition. Because the normal cloud model effectively handles the fuzziness and randomness of the data, the data fusion process more closely reflects reality; while the grey weighted clustering method enables accurate classification and assessment of equipment condition even with limited data samples.
[0067] Furthermore, the technical solution of this invention cleans washable data and detects anomalies in non-cleanable data, enabling a comprehensive, timely, and accurate understanding of the health status of key performance indicators. This provides a theoretical basis for condition-based maintenance and strong support for ensuring the safe operation of equipment.
[0068] Specifically, such as Figure 1 As shown in the figure, an embodiment of the power equipment condition assessment method based on normal cloud model and grey weighted clustering of the present invention includes the following steps:
[0069] Step S10: Based on the research and analysis of current power equipment condition monitoring technology, establish a basic parameter system for condition assessment applicable to the overall and basic unit condition assessment of power equipment; remove parameters with low correlation to faults and defects, retain parameters with strong correlation, and construct a key parameter system for condition assessment.
[0070] Step S20: Check the quality of the data in the key parameter system in step S10. If the data quality does not meet the requirements, perform data cleaning.
[0071] Step S30: Establish a data analysis model, use the correlation between vertical and horizontal data to judge abnormal changes in equipment status and its related relationships, and detect equipment anomalies by analyzing the correlation and change patterns of multidimensional status data.
[0072] Step S40: Based on the relevant characteristic parameters of the power transmission and transformation equipment, determine the characteristic parameters related to key performance and their coupling relationship with the key performance status of the equipment, evaluate each key performance, screen out the key performance that has anomalies, and obtain the evaluation value of each key performance.
[0073] In step S10, the key parameter system for power transmission and transformation equipment is divided into three layers: a status information layer, a sensor data layer, and an application interaction layer. The status quantities in the status information layer include offline information, online monitoring, live-line testing, manual inspection, preventative testing, operational information, ledger information, and environmental information. The interaction parameters in the application interaction layer include geographical information for scheduling step S10, operational information, meteorological information, and lightning monitoring information. The status quantities of power transmission and transformation equipment are classified and integrated according to components to form a hierarchical parameter system, which also includes various types of underlying sensor data. In this embodiment, the equipment parameter system is divided into three layers: a status information layer, a sensor data layer, and an application interaction layer. The components in the component layer are consistent with the status evaluation guidelines. Based on the data update speed of the status quantities, they can be divided into real-time data, near-real-time data, and static data, which serve as the basis for subsequently establishing the key parameter system.
[0074] Constructing a key parameter system requires investigating, collecting, and organizing data on defects, anomalies, and faults in power equipment. Due to differences in acquisition methods and data composition, status information is categorized based on its static, near-real-time, and real-time attributes. This embodiment uses mathematical models based on association rules and principal component analysis to extract key parameters, removing parameters with low correlation to faults and defects and retaining those with strong correlation. This constructs a key parameter system for status evaluation, addressing the ambiguity and uncertainty inherent in the status information of power transmission and transformation equipment, and enabling accurate and effective assessment of the operating status of power equipment. The overall construction method of the key parameter system for status monitoring is as follows:
[0075] Through a comprehensive survey and analysis of the principles, applications, and cost-effectiveness of current main power equipment condition monitoring technologies (including online monitoring, live testing, and inspection), a basic parameter system for condition monitoring applicable to the overall and basic unit condition assessment of power equipment is established.
[0076] For near real-time status information, a mathematical model based on principal component analysis is applied to extract key parameters of various components of power transmission equipment. Relevant information in the parameters is obtained through orthogonal transformation, and the contribution rate of each parameter to the principal components is used as the basis for key parameter extraction. Key parameters are selected according to the following steps:
[0077] Step S11: Quantize the basic parameter into a four-dimensional vector. Represent the quantized vector of parameter A as A = (A1, A2, A3, A4). Then: A1 represents the proportion of parameter A in the historical fault statistics; A2 represents the proportion of parameter A in the historical major / urgent defect statistics; A3 represents the proportion of parameter A in the historical general defect statistics; A4 represents the proportion of parameter A in the production management system defect database.
[0078] The status information data of power transmission and transformation equipment includes heterogeneous data from multiple sources such as online monitoring, live-line testing, power outage testing, manual inspection, power grid operation, and environmental meteorology. This data is integrated into a unified data information platform after standardization and normalization, providing strong data support for subsequent status assessment and fault early warning of power transmission and transformation equipment. However, due to factors such as short-term sensor failures, communication port anomalies, and equipment system disturbances, various outliers are unavoidable in the status information data. These outliers include cleanable and non-cleanable anomaly patterns.
[0079] The data that can be cleaned includes:
[0080] Erroneous data: Data that is spatially off-target or temporally isolated;
[0081] Missing values: Data is incomplete, resulting from communication interruption or database storage abnormalities.
[0082] Communication noise or external interference: a situation where the rate of change exceeds the normal range within a short period of time;
[0083] Bad data caused by short-term communication interruption: data changes very little or has a variance of 0 over a period of time;
[0084] Temporary horizontal migration data: A small amount of data migrates horizontally relative to the normal value within a certain period of time, and then returns to the normal state.
[0085] Cleanable data includes: high noise or high variance abnormal data: the variance of long-term measurement data is much greater than that of historical data, the abnormality is caused by damage to the acquisition sensor, hardware failure of the acquisition device or low battery, and the acquisition device has malfunctioned.
[0086] The mutation amount that appears in highly correlated sequences within a certain period of time: The mutation amount that appears simultaneously or with a lag of N periods indicates that the monitoring parameter has been affected by external factors, which may be due to external interference or equipment malfunction.
[0087] A highly correlated sequence exhibits horizontal migration after a certain period of time: a permanent horizontal migration of a sequence occurring simultaneously or with a lag of N periods indicates an abnormality in the equipment status;
[0088] A highly correlated sequence shows an accelerated upward trend after a certain period of time: if a sequence shows an accelerated upward trend simultaneously or with a lag of N periods, it indicates that the equipment status is abnormal.
[0089] The data cleaning process for the above-mentioned cleanable data includes the following steps:
[0090] Step S21: First, based on the characteristics of the power transmission and transformation equipment data, fit the corresponding time label for each type of data;
[0091] Step S22: For the unit time tags of each state quantity, fit the data within a certain time period, analyze the residual sequence to determine whether there is abnormal data. If there is abnormal data, determine whether to use the outlier or the intervened data.
[0092] Step S23: Combine the Aprior correlation analysis, analyze the time series of state quantities with strong correlations simultaneously, make a judgment on data cleaning according to the set abnormal patterns between the sequences, extract the characteristics of the abnormal time series, and correspond them to the abnormal patterns one by one.
[0093] For the above uncleanable data, perform anomaly detection. The basic principle of anomaly detection is to establish a data analysis model using a large amount of normal state data, and use the correlations of vertical (time) and horizontal (different parameters) data to judge the abnormal changes of the device state and its associated relationships. The main goal is to change the traditional state analysis threshold judgment method, and perform device anomaly detection by analyzing the correlation relationships and change rules of multi-dimensional state data. Its anomaly detection steps include:
[0094] Step S31: For the historical data of each parameter, fit an autoregressive model and train the self-organizing step SOM to obtain the transition probability matrix {X1, X2...Xn} of a single parameter.
[0095] Step S32: Input the historical data of all parameters into the K-mean step S algorithm to obtain m normal clustering clusters.
[0096] Step S33: For the measured real-time data stream, convert the data stream of a single parameter into a transition probability sequence through the transition probability matrix, cluster the overall data stream, and judge whether the parameter data at each time point belongs to the m normal clustering clusters.
[0097] Step S34: Anomaly detection rules for the measured data stream:
[0098] A. If there are no 0 values in the transition probability sequences of each parameter, and all the data in the overall data stream belong to the m normal clustering clusters, it is judged that the data is normal.
[0099] B. If there are only a small number of 0 values in the transition probability sequences of only a few parameters, and there are a small number of data in the data stream that do not belong to the m normal clustering clusters, it is judged that there is a small amount of noise in the data, and these noises belong to sensor anomalies that can be ignored.
[0100] C. When there is a segment of 0 values in the transition probability sequences of k parameters (k < N), and a large segment of data in the data stream does not belong to the m normal clustering clusters, it means that the parameters of the device have an anomaly and the device running state is abnormal.
[0101] For abnormal operating states in C, the time of occurrence of the abnormality of the parameters is determined based on the zero-crossing points of the transition probability sequence of k parameters.
[0102] The main steps of the K-means S-clustering method are as follows:
[0103] 1) Randomly select k objects from n data objects as the initial cluster centers;
[0104] 2) Calculate the average value for each cluster and use this average value to represent the corresponding cluster;
[0105] 3) Assign each object to the nearest cluster based on its distance from the cluster centers;
[0106] 4) Proceed to step two and recalculate the average value for each cluster.
[0107] This process is repeated until a certain criterion function no longer changes significantly or the clustered objects no longer change.
[0108] The general K-means step S algorithm uses the squared error criterion function, where E is the sum of the root mean square errors of all objects in the dataset and their corresponding cluster centers, p is the given data object, and mi is the mean of the ci cluster (p and m are both multidimensional).
[0109] The value of k is iterated using the Calin step, Ski-Haraba step, and Sz criterion:
[0110] Where step S and step SB represent the inter-class variance, step S and step SW represent the intra-class variance, and VRCk represents the complexity. The larger VRCk is, the better the clustering effect.
[0111] In step S40, key performance characteristics include three main categories: electrical performance, insulation performance, and mechanical performance. Characteristic parameter sets are established for different key performance characteristics of different equipment.
[0112] Based on the key performance characteristics of power transmission and transformation equipment, including electrical performance (insulation degradation and aging), mechanical performance, and overload capacity, relevant characteristic parameters for condition evaluation are identified. Through big data mining and analysis methods such as statistical analysis, classification identification, principal component analysis, and correlation analysis, characteristic parameters related to key performance and their coupling relationship with the key performance status of the equipment are determined. Dynamic maintenance of the indicator system is achieved through regular or irregular data analysis.
[0113] Based on the feature parameter sets corresponding to different key performance indicators, each key performance indicator is evaluated using methods such as fuzzy comprehensive evaluation, clustering, matter-element theory, and high-dimensional random matrices. Key performance indicators exhibiting anomalies are identified, and evaluation values for each key performance indicator are derived. This embodiment uses the fuzzy comprehensive evaluation method to evaluate each key performance indicator according to the following steps:
[0114] Let U = {u1, u2, Lun} be the set of evaluation factors (i.e., the set of evaluation indicators), V = {v1, v2, Lvn} be the set of comments (i.e., the set of evaluation levels), and R be the fuzzy matrix for evaluation.
[0115] Where Ri = (ri1, ri2, Lrin) (i = 1, 2, ..., n) is a single-factor fuzzy evaluation relative to evaluation factor ui, which is a fuzzy subset on the comment set V, and rij is the membership degree of the comment vj (j = 1, 2, ..., m) given relative to the ui-th evaluation factor. The weight set W = (w1, w2, Lwn) is a fuzzy subset on factor U, reflecting the importance of each evaluation factor, and is a quantitative representation of the importance of each factor in the set. Its rationality will directly affect the rationality of the evaluation result. Where wi is the weight of evaluation factor ui, and wi ≥ 0.
[0116] The comprehensive evaluation problem involves transforming the fuzzy set W of the evaluation factor set U into a fuzzy set B on the universe of discourse, the comment set V, through the fuzzy relation R. The expression is B = A & R, where & is a generalized fuzzy operator. There are many ways to choose it, such as the main factor prominence type, the main factor determination type, the weighted average model, etc.
[0117] In step S40, the key performance indicators are evaluated using the fuzzy comprehensive evaluation method according to the following steps:
[0118] Step S41: Establish the factor set for the evaluation object;
[0119] Step S42: Create a set of comments;
[0120] Step S43: Establish fuzzy evaluation matrices at each level;
[0121] Step S44 combines the weight matrix and the fuzzy evaluation matrix to select a weighted average comprehensive evaluation to obtain the comprehensive evaluation result;
[0122] Step S45 processes the comprehensive evaluation results from step S44 to obtain the final evaluation result.
[0123] When evaluating the health status of transformer windings and bushings in this embodiment, the process of applying the fuzzy comprehensive evaluation method is as follows:
[0124] (1) Establish the factor set for the evaluation object:
[0125] The L2 level indicators for transformers include five categories: oil chromatography, electrical testing, oiling testing, operating condition information, inspection and defects, i.e., U = {u1, u2, u3, u4, u5}. Further decomposition and selection can yield the final indicator layer. For example, oil chromatography can be decomposed into three indicators: hydrogen, acetylene, and ethylene, i.e., u1 = {u11, u12, u13}.
[0126] (2) Establish a comment collection:
[0127] The operating status of power distribution equipment is divided into four categories: "normal", "caution", "abnormal", and "serious". The evaluation set is: V = {normal, caution, abnormal, serious} = {v1, v2, v3, v4}.
[0128] (3) Establish fuzzy evaluation matrices at each level:
[0129] Let ui be the i-th indicator under a certain first-level indicator to evaluate the equipment, and rij be the membership degree of state vj in the evaluation set (j = 1, 2, 3, 4). Then the membership degree set Ri = {ri1, ri2, ri3, ri4} can represent the evaluation result according to indicator ui. Then all the indicators under this project constitute its fuzzy evaluation matrix.
[0130] Taking oil chromatography U1 as an example, the evaluation matrix of the test indicators is as follows:
[0131] (4) A comprehensive evaluation result can be obtained by performing composite operations:
[0132] Combining the weight matrix and the fuzzy evaluation matrix, a weighted average comprehensive evaluation is selected, namely:
[0133] This algorithm considers the impact of major factors on transformer condition while retaining all information about individual factors, making it more consistent with reality.
[0134] Taking oil chromatography test U1 as an example, its comprehensive evaluation results are as follows:
[0135] The comprehensive evaluation results of the transformer's health status, obtained by combining all L2-level indicators, are as follows:
[0136] (5) Processing of evaluation results:
[0137] After obtaining the evaluation value bj (j = 1, 2, 3, 4) through fuzzy comprehensive evaluation, the evaluation results must be processed to obtain the final evaluation result. The maximum membership method can be used, taking the elements of the comment set corresponding to the maximum evaluation value bmax = max{bj|j = 1, 2, 3, 4} as the evaluation result; alternatively, the fuzzy distribution method can be used, directly treating bj as the evaluation result, so that the evaluator has a comprehensive understanding of the state of the transformer to be evaluated.
[0138] Specifically, the key inventive point of this technical solution is as follows:
[0139] 1. The normal cloud model is used to effectively fuse and clean power equipment condition monitoring data. The grey weighted clustering method is then used for in-depth analysis of the fused data. This process of fusing and deeply mining multi-source power equipment condition monitoring data aims to achieve accurate assessment of the power equipment's condition. Because the normal cloud model effectively handles the fuzziness and randomness of the data, the data fusion process is more consistent with reality; while the grey weighted clustering method can achieve accurate classification and assessment of equipment condition even with limited data samples.
[0140] 2. Data cleaning is performed on washable data, and anomaly detection is performed on non-washable data. This allows for a comprehensive, timely, and accurate understanding of the health status of key performance indicators, providing a theoretical basis for condition-based maintenance and strong support for ensuring the safe operation of equipment.
[0141] 3. Data cleaning modes for cleanable abnormal data, anomaly detection for non-cleanable data, and basic principles of anomaly detection.
[0142] Furthermore, the technical solution of the present invention has the following further improvements or optimizations:
[0143] 1. Further explore the value of user handbooks, technical files, and other data; establish user fact tags, combined tags, and predictive tags; create user profiles; and centrally and comprehensively display important user basic information, tag information, electricity inspection records, electricity test report records, and other information to help provide better customer service.
[0144] 2. Based on electrical test data and in accordance with relevant standards, establish a mapping relationship between electrical test indicators and equipment faults. Combine this with machine learning models to provide early warnings of potential faults, indicate the causes of faults, and prepare for emergency response.
[0145] In summary, the technical solution of this invention effectively fuses and cleans power equipment condition monitoring data using a normal cloud model, and employs a grey weighted clustering method to conduct in-depth analysis of the fused data, thereby achieving an accurate assessment of the power equipment's condition. It effectively handles the fuzziness and randomness of the data, making the data fusion process more consistent with reality; it can achieve accurate classification and assessment of equipment condition even with limited data samples. Therefore, the technical solution of this invention can perform fusion analysis and in-depth mining of condition monitoring data, enabling rapid detection of abnormal conditions and personalized condition evaluation, understanding the health status of key performance indicators, providing a basis for condition-based maintenance, and supporting the safe operation of equipment.
[0146] This invention can be widely used in the fields of equipment status assessment, operation scheduling, and equipment management of power equipment.
Claims
1. A power equipment condition assessment method based on normal cloud model and grey weighted clustering, characterized by: Includes the following steps: Step S10) Based on the research and analysis of current power equipment condition monitoring technology, establish a basic parameter system for condition assessment applicable to the overall and basic unit condition assessment of power equipment; Remove parameters that are less correlated with faults and defects, retain parameters that are highly correlated, and construct a key parameter system for condition assessment; Step S20) Check the quality of the data in the key parameter system in step S10, and perform data cleaning if the data quality does not meet the requirements. Step S30) Establish a data analysis model, use the correlation between vertical and horizontal data to judge abnormal changes in equipment status and its related relationships, and perform equipment anomaly detection by analyzing the correlation and change patterns of multidimensional status data; Step S40) Based on the relevant characteristic parameters of the power transmission and transformation equipment, determine the characteristic parameters related to the key performance and their coupling relationship with the key performance status of the equipment, evaluate each key performance, screen out the key performance that has anomalies, and obtain the evaluation value of each key performance.
2. The power equipment condition assessment method based on normal cloud model and grey weighted clustering as described in claim 1, characterized in that... In step S10, select the key parameters according to the following steps: The basic parameter is quantized into a four-dimensional vector. The quantized vector of parameter A is represented as A = (A1, A2, A3, A4). Then: A1 represents the proportion of parameter A in the annual fault statistics; A2 represents the proportion of parameter A in the annual major / urgent defect statistics; A3 represents the proportion of parameter A in the annual general defect statistics; A4 represents the proportion of parameter A in the defect database of the production management system. The weight vector H of the basic parameters is normalized to [0,1] to obtain the key parameter extraction method: the larger the weight, the stronger the correlation and the more representative the basic parameter. The weight is used as the basis for key parameter extraction, and the basic parameters with weights above 0.5 are selected as key parameters.
3. The power equipment condition assessment method based on normal cloud model and grey weighted clustering as described in claim 2, characterized in that... After selecting key parameters, the confidence and support between sequences are calculated using association rules. Time series with similar rates of change or obvious correlations are extracted. Parameters that are correlated in the sequences themselves or whose changes are correlated are extracted to construct a key parameter system for state assessment.
4. The power equipment condition assessment method based on normal cloud model and grey weighted clustering as described in claim 3, characterized in that: The confidence and support between sequences are calculated using the Apriori algorithm in the following steps: (1) Scan the entire database T and calculate the number of times each item in T appears. In this process, generate a set C1 of first-order candidate options. (2) Set a minimum support, delete frequent itemsets with less than the minimum support from C1, and retain the remaining frequent itemsets to generate a first-order frequent itemset L1. (3) Use L1 to generate candidate second-order candidate itemsets C2, and then scan the database to count the itemsets in C2; (4) Determine the frequent phase set L2 from the candidate set C2 based on the minimum support; (5) Use L1 and L2 to generate candidate third-order candidate itemsets C3, and then scan the transaction database to count the itemsets in C3; (6) Determine the frequent phase set L3 from the candidate set C3 based on the minimum support; (7) Repeat this process until the number of frequent phase sets reaches the upper limit or the frequent phase sets are empty, at which point the algorithm ends.
5. The power equipment condition assessment method based on normal cloud model and grey weighted clustering as described in claim 1, characterized in that... In step S20, the data includes washable data and non-washable data; Data cleaning is performed on cleanable data. The data cleaning steps include: Step S21) First, based on the characteristics of the power transmission and transformation equipment data, fit the corresponding time label for each type of data; Step S22) For each state variable's unit time label, fit its data within a certain time period, analyze the residual sequence to determine whether there is abnormal data, and if there is abnormal data, determine whether to use outlier values or intervened data. Step S23) Combined with April correlation analysis, the time series of state variables with strong correlation are analyzed simultaneously. Data cleaning is judged according to the abnormal patterns set between the series, and the characteristics of abnormal time series are extracted and matched one by one with the abnormal patterns.
6. The power equipment condition assessment method based on normal cloud model and grey weighted clustering as described in claim 5, characterized in that: The washable data includes: Erroneous data: Data that is spatially off-target or temporally isolated; Missing values: Data is incomplete, resulting from communication interruption or database storage abnormalities. Communication noise or external interference: a situation where the rate of change exceeds the normal range within a short period of time; Bad data caused by short-term communication interruption: data changes very little or has a variance of 0 over a period of time; Temporary horizontal migration data: A small amount of data migrates horizontally relative to the normal value within a certain period of time, and then returns to the normal state.
7. The power equipment condition assessment method based on normal cloud model and grey weighted clustering as described in claim 5, characterized in that: The unwashable data includes: High noise or high variance abnormal data: The variance of long-term measurement data is much greater than that of historical data. The anomaly is caused by damage to the acquisition sensor, hardware failure of the acquisition device, or low battery. The acquisition device has malfunctioned. The mutation amount that appears in highly correlated sequences within a certain period of time: The mutation amount that appears simultaneously or with a lag of N periods indicates that the monitoring parameter has been affected by external factors, which may be due to external interference or equipment malfunction. A highly correlated sequence exhibits horizontal migration after a certain period of time: a permanent horizontal migration of a sequence occurring simultaneously or with a lag of N periods indicates an abnormality in the equipment status; A highly correlated sequence shows an accelerated upward trend after a certain period of time: if a sequence shows an accelerated upward trend simultaneously or with a lag of N periods, it indicates that the equipment status is abnormal.
8. The power equipment condition assessment method based on normal cloud model and grey weighted clustering as described in claim 7, characterized in that: Anomaly detection is performed on unwashable data, and the anomaly detection includes the following steps: Step S31: For the historical data of each parameter, fit an autoregressive model and train a self-organizing step (SOM) to obtain the transition probability matrix {X1,X2Xn} for a single parameter. Step S32 inputs the historical data of all parameters into the K-means step S algorithm to obtain m normal clusters; Step S33: For the real-time data stream to be tested, the data stream of a single parameter is converted into a transition probability sequence through a transition probability matrix, the overall data stream is clustered, and it is determined whether the parameter data at each time point belongs to one of the m normal clusters. Step S34: Anomaly detection rules for the data stream under test: If none of the transition probability sequences of each parameter have a value of 0, and all the data in the overall data stream belong to m normal clusters, then it is determined that there is no anomaly in the data; If only a few parameters have a small number of 0 values in their transition probability sequences, and a small number of data in the data stream do not belong to the m normal clusters, then it is determined that there is a small amount of noise in the data, and this noise is a negligible sensor anomaly. If the transition probability sequence of k parameters contains a segment of 0 values, and a large segment of data in the data stream does not belong to m normal clusters, then the parameters of the device are abnormal and the device is in an abnormal operating state. When equipment operates abnormally, the time of the abnormality is determined based on the zero-crossing points of the transition probability sequence of k parameters.
9. The power equipment condition assessment method based on normal cloud model and grey weighted clustering as described in claim 8, characterized in that... In step S40, the key performance indicators are evaluated using the fuzzy comprehensive evaluation method according to the following steps: Step S41: Establish the factor set for the evaluation object; Step S42: Create a set of comments; Step S43: Establish fuzzy evaluation matrices at each level; Step S44 combines the weight matrix and the fuzzy evaluation matrix to select a weighted average comprehensive evaluation to obtain the comprehensive evaluation result; Step S45 processes the comprehensive evaluation results from step S44 to obtain the final evaluation result.
10. The power equipment condition assessment method based on normal cloud model and grey weighted clustering as described in claim 1, characterized in that: The power equipment condition assessment method based on normal cloud model and grey weighted clustering effectively integrates and cleans power equipment condition monitoring data through normal cloud model, and uses grey weighted clustering method to conduct in-depth analysis of the integrated data to achieve accurate assessment of power equipment condition.
Citation Information
Patent Citations
A method, apparatus, electronic device, and storage medium for assessing the condition of power equipment.
CN116499531B
Power equipment state evaluation method, system and equipment for intelligent monitoring and storage medium
CN120688337A
