Fault-tolerant analysis system based on similar data

By adopting a fault-tolerant analysis system based on similar data in the fault-tolerant analysis technology, using low-dimensional space mapping and topological change identification, the problem of insufficient flexibility and timeliness in the existing technology is solved, and more efficient fault identification and Responsiveness ensures system stability and data integrity.

CN118897750BActive Publication Date: 2024-11-29HUANENG LANCANG RIVER HYDROPOWER CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202411356844.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-27
Publication Date
2024-11-29
Estimated Expiration
2044-09-27

AI Technical Summary

Technical Problem

Existing fault-tolerant analysis technologies lack flexibility and timeliness in dynamic and complex data environments, making it difficult to effectively identify small or early abnormalities, resulting in delayed failure response time, waste of resources and system stability.

Method used

Using a fault-tolerant analysis system based on similar data, the real-time data flow is mapped through the similar feature recognition module to generate a fit score and identify key topological changes; the dynamic fault detection module recognizes abnormal structures and generates preliminary fault location; the fault impact The analysis module estimates the fault propagation path and range; the data recovery and isolation module perform collective correction and abnormal data point isolation to limit error propagation.

Benefits of technology

It realizes accurate identification of key topological changes in the data, monitors and warnings for potential faults in real time, improves the pertinence and speed of fault responses, effectively limits error propagation, and maintains system stability and data integrity.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118897750B_ABST
    Figure CN118897750B_ABST
Patent Text Reader

Abstract

The present invention relates to the technical field of fault-tolerant analysis, and specifically to a fault-tolerant analysis system based on similar data, wherein the system comprises: a similar feature recognition module maps data points to a low-dimensional space based on data points in a real-time data stream, and generates a fit score by calculating the positional relationship of the data points in the low-dimensional space. In the present invention, a fit score is generated by calculating the low-dimensional space mapping and positional relationship of data points in a real-time data stream, thereby achieving accurate identification of key topological changes in the data. This processing strategy can monitor and warn potential system failures in real time, which is superior to the delayed response mode in traditional technologies. Analyzing the specific location of the failure point and predicting its propagation path and range allows the system to carry out targeted prevention and rapid response. In addition, the collective correction of similar data points and the isolation strategy of abnormal data points effectively limit the propagation of errors and maintain the stability of the system and data integrity.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of fault-tolerant analysis, and in particular to a fault-tolerant analysis system based on similar data. Background Art

[0002] The field of fault-tolerant analysis technology involves methods and mechanisms for implementing error detection, error prevention, and error recovery in computer systems, networks, and other information processing systems. The core purpose of this technology is to ensure that the system can continue to operate even in the event of partial hardware or software failures, thereby improving the reliability and availability of the system. Fault-tolerant technology usually includes redundant design (such as multiple processors, disk mirroring), error detection and correction code (such as ECC memory), and fault recovery strategies (such as checkpoint / rollback mechanisms). In addition, fault-tolerant analysis also involves modeling and simulation of system behavior, by pre-identifying potential error scenarios and weaknesses to design a more complete system.

[0003] Fault-tolerance analysis systems are specific examples of applied fault-tolerance analysis techniques. These systems are specifically designed to monitor and assess the fault-tolerance capabilities of other systems. They analyze data in real time to identify conditions that could cause system failures, assess the system's ability to respond when encountering specific errors, and provide recommendations for improvement measures. The main uses of these systems include improving the stability and security of critical infrastructure, such as in financial services, healthcare, aerospace, and other industries that require high reliability. By implementing fault-tolerance analysis systems, organizations are able to reduce the risk of sudden failures and ensure service continuity and data integrity.

[0004] Existing technologies include redundant design, error detection and correction codes, and fault recovery strategies, but these technologies usually rely on relatively static and preset fault response mechanisms, such as checkpoints and rollbacks, which may not be flexible or timely enough in dynamic and complex data environments. Traditional methods have limited ability to identify minor or initial anomalies in highly variable data streams, resulting in delayed fault response times and inability to effectively prevent and mitigate the immediate development of faults. Existing fault recovery strategies lack predictions of fault propagation paths and scopes, resulting in possible waste of resources during the recovery process, such as over-repairing non-critical areas or ignoring potentially critical affected areas. This approach not only increases the system operation and maintenance costs, but may also affect the long-term stability and availability of the system. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings in the prior art and to propose a fault-tolerant analysis system based on similar data.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: The fault-tolerant analysis system based on similar data includes:

[0007] The similarity feature recognition module maps the data points in the real-time data stream to a low-dimensional space, generates a fit score by calculating the positional relationship of the data points in the low-dimensional space, and uses the fit score to identify key topological changes in the data, including connectivity, the appearance and disappearance of holes, and generates a record of topological feature changes;

[0008] The dynamic fault detection module receives the topology feature change record, identifies key abnormal structures in the topology map by mapping the changes in the data structure, including newly appeared holes or mutant clusters, generates preliminary fault location, detects the abnormal position and type appearing in the data according to the preliminary fault location, and generates a fault detection result;

[0009] The fault impact analysis module uses the fault detection results to analyze the location of the fault point in the data set, estimate the fault propagation path and range, analyze the affected data area, and generate a fault impact range prediction;

[0010] The data recovery and isolation module performs collective corrections on groups of similar data points based on the predicted scope of impact of the fault, isolates the detected abnormal behavior data points, limits error propagation, and establishes data isolation.

[0011] As a further solution of the present invention, the step of obtaining the fitness score is specifically as follows:

[0012] Collect data points in real-time data streams and use principal component analysis to map the data points from high-dimensional space to low-dimensional space using the formula:

[0013]

[0014] Calculate the mapped data point set;

[0015] in, represents the original data point matrix, P represents the projection matrix, Represents the set of data points after mapping;

[0016] Using the mapped data point set, the distance matrix between the data points is calculated using the distance calculation formula:

[0017]

[0018] Generate a distance matrix;

[0019] in, Represents data points With data points The distance between , Respectively represent the data points in dimensional coordinates, is the regularization parameter;

[0020] Based on the distance matrix, the goodness of fit score for each data point is calculated using the formula:

[0021]

[0022] Get the goodness of fit score;

[0023] in, Representative data points The fit score of To adjust the parameters, For data points and The distance between is a collection of goodness-of-fit scores.

[0024] As a further solution of the present invention, the step of obtaining the topological feature change record is specifically:

[0025] Based on the set of goodness-of-fit scores, the critical value of the score is determined by statistical methods to identify topological changes, using the formula:

[0026]

[0027] Calculate the critical value;

[0028] in, is the critical value, is the average value, is the standard deviation, is the adjustment factor, is the coefficient of variation;

[0029] Using the cutoff values, a step function analysis was performed on each score using the formula:

[0030]

[0031] Generate a set of step function responses;

[0032] in, For data points The step response of is a step function, , is the regulating factor, Score the goodness of fit of data points;

[0033] According to the step function response set, the topology change is counted using the formula:

[0034]

[0035] Calculate the total number of key topological changes and generate topological feature change records;

[0036] in, Indicates the total number of topology changes.

[0037] As a further solution of the present invention, the steps of obtaining the preliminary fault location are specifically as follows:

[0038] Receive the topology feature change record, analyze the change of data structure, and use the formula:

[0039]

[0040] Calculate the anomaly index of each data point and generate a set of change indices;

[0041] in, Representative The anomaly index of the data point, For data point The value of the dimension, is the weight of the dimension, , Adjusting for the effects of exponential growth;

[0042] According to the set of change indices, low-dimensional mapping of the data is performed to emphasize key abnormal structures, using the formula:

[0043]

[0044] Evaluate the topological changes after mapping and generate a key structure score after mapping;

[0045] in, is the key structure score after mapping, is the data point value after mapping, Adjust the parameters for the mapping, is the smoothing parameter;

[0046] Using the mapped key structure scores, the voids and mutation clusters that appear are identified using the formula:

[0047]

[0048] Determine the most critical abnormal structure and generate preliminary fault location results;

[0049] in, This is the preliminary fault location result. is the structural score, is the historical average score, is the stability adjustment factor.

[0050] As a further solution of the present invention, the step of obtaining the fault detection result is specifically:

[0051] Receive and analyze the preliminary fault location results, using the formula:

[0052]

[0053] Generate analysis results of abnormal locations;

[0054] in, is the analysis result of the abnormal position. is the fault location component, is the average value, is the standard deviation, is the smoothing constant, is the weight coefficient;

[0055] Using the analysis results of the abnormal position, the formula is adopted:

[0056]

[0057] Generate preliminary classification results of fault types;

[0058] in, is the preliminary classification result of the fault type. , To adjust the parameters, is the power adjustment parameter;

[0059] Based on the preliminary classification results of the fault types, the fault location and type are determined using the formula:

[0060]

[0061] Generate fault detection results;

[0062] in, is the fault detection result, is the significance threshold, is the significance adjustment parameter.

[0063] As a further solution of the present invention, the step of obtaining the fault impact range prediction is specifically as follows:

[0064] The fault detection results are analyzed to determine the location of the fault point in the data set using the formula:

[0065]

[0066] Calculate the location index of each fault point and generate fault location analysis results;

[0067] in, is the fault location analysis result, is the fault value, is the standard deviation, is the average value, To adjust the parameters;

[0068] Use the fault location analysis results to estimate the fault propagation path and range using the formula:

[0069]

[0070] Calculate the potential paths for fault propagation and generate fault propagation prediction results;

[0071] in, is the fault propagation prediction result, , Adjust parameters for sensitivity;

[0072] Based on the fault propagation prediction results, the affected data area is analyzed using the formula:

[0073]

[0074] Calculate the impact level of each affected area and generate a prediction of the impact range of the fault;

[0075] in, To predict the impact range of the fault, is the value predicted for fault propagation, Adjust the parameters for influence.

[0076] As a further solution of the present invention, the steps of establishing data isolation are specifically:

[0077] Starting from the fault impact range prediction, the affected data point groups are identified using the formula:

[0078]

[0079] Calculate the influence weight of each data point and generate the influence weight distribution result;

[0080] in, To influence the weight distribution results, is the data point value, is the average value of the data points, is the standard deviation, is the weight factor, is the error reduction factor;

[0081] The impact weight distribution result is used to collectively correct the affected data points and prepare for the isolation of abnormal behavior data points, using the formula:

[0082]

[0083] Adjust the values ​​of data points to reduce error propagation and generate collective correction results;

[0084] in, To collectively correct the results, is the original data point value, is the influence weight, To adjust the parameters;

[0085] The collective correction result is used to isolate the detected abnormal behavior data points, using the formula:

[0086]

[0087] Isolate abnormal data points, limit error propagation, and establish data isolation;

[0088] in, For data isolation, is the corrected data point value, is the isolation threshold, To adjust the parameters of data isolation sensitivity, is the indicator function.

[0089] Compared with the prior art, the advantages and positive effects of the present invention are:

[0090] In the present invention, a fit score is generated by calculating the low-dimensional spatial mapping and positional relationship of data points in a real-time data stream, thereby achieving accurate identification of key topological changes in the data. This processing strategy can monitor and warn of potential system failures in real time, which is superior to the delayed response method in traditional technologies. Analyzing the specific location of the fault point and predicting its propagation path and range allows the system to carry out targeted prevention and rapid response. In addition, the collective correction of similar data points and the isolation strategy of abnormal data points effectively limit the propagation of errors, maintain the stability of the system and data integrity, not only reducing the downtime of the system, but also improving the efficiency and accuracy of the system in handling faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] Figure 1 is a system flow chart of the present invention;

[0092] Figure 2 A flow chart of the steps for obtaining the fitness score of the present invention;

[0093] Figure 3 A flowchart of the steps for obtaining the topological feature change record of the present invention;

[0094] Figure 4 A flowchart of the steps for obtaining the initial fault location of the present invention;

[0095] Figure 5 A flowchart of the steps for obtaining the fault detection result of the present invention;

[0096] Figure 6 A flowchart of the steps for obtaining the prediction of the impact range of a fault according to the present invention;

[0097] Figure 7 This is a flow chart of the steps of establishing data isolation according to the present invention. DETAILED DESCRIPTION

[0098] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0099] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0100] Embodiment 1

[0101] See also Figure 1 ,The fault-tolerant analysis system based on similar data includes:

[0102] The similarity feature recognition module maps the data points in the real-time data stream to a low-dimensional space, generates a fit score by calculating the positional relationship of the data points in the low-dimensional space, and uses the fit score to identify key topological changes in the data, including connectivity, the appearance and disappearance of holes, and generates a record of topological feature changes;

[0103] The dynamic fault detection module receives the topology feature change records, identifies the key abnormal structures in the topology map by mapping the changes in the data structure, including the newly appeared holes or mutant clusters, generates preliminary fault location, detects the abnormal positions and types appearing in the data based on the preliminary fault location, and generates fault detection results;

[0104] The fault impact analysis module uses the fault detection results to analyze the location of the fault point in the data set, estimate the fault propagation path and range, analyze the affected data area, and generate a prediction of the fault impact range;

[0105] The data recovery and isolation module predicts the scope of fault impact, performs collective corrections on groups of similar data points, isolates the detected data points of abnormal behavior, limits the propagation of errors, and establishes data isolation.

[0106] The fault detection results specifically include newly emerged voids and mutant clusters. The fault impact range prediction specifically includes the fault location, fault propagation path, and affected area. Data isolation specifically refers to data point group correction, abnormal data point isolation, and error propagation limitation.

[0107] See also Figure 2 , the steps to obtain the fit score are as follows:

[0108] Collect data points in real-time data streams and use principal component analysis to map the data points from high-dimensional space to low-dimensional space using the formula:

[0109]

[0110] Calculate the mapped data point set;

[0111] in, represents the original data point matrix, P represents the projection matrix, Represents the set of data points after mapping;

[0112] Use the mapped data point set to calculate the distance matrix between data points, using the distance calculation formula:

[0113]

[0114] Generate a distance matrix;

[0115] in, Represents data points With data points The distance between , Respectively represent the data points in dimensional coordinates, is the regularization parameter;

[0116] According to the distance matrix, the fitness score of each data point is calculated using the formula:

[0117]

[0118] Get the goodness of fit score;

[0119] in, Representative data points The fit score of To adjust the parameters, For data points and The distance between is a collection of goodness-of-fit scores.

[0120] Detailed explanation and example of the formula for step 1

[0121] formula:

[0122]

[0123] Parameter explanation:

[0124] : The original data point matrix.

[0125] : The projection matrix from high-dimensional space to low-dimensional space.

[0126] Z: The set of data points after mapping.

[0127] Calculation process and derivation:

[0128] Assume that the original data matrix is a matrix (3 data points, 5 features per data point), and want to reduce the data to 2 features. Projection matrix will be a Matrix. Example and It can be:

[0129]

[0130]

[0131] Perform matrix multiplication to get :

[0132]

[0133] Represents the set of data points after dimensionality reduction, where each data point now has only 2 features.

[0134] Detailed explanation and example of the formula for step 2

[0135] formula:

[0136]

[0137] Parameter explanation:

[0138] : Data Points With data points The distance between.

[0139] : Data Points and In the Dimensional coordinates.

[0140] : Regularization parameter.

[0141] Calculation process and derivation:

[0142] Assume that the result of step 1 above is used To calculate the distance, select , which is a common regularization choice used to prevent the denominator from being zero and improve the stability of the calculation.

[0143] Calculate the distance between data point 0 and data point 1:

[0144]

[0145] This distance Represents the relative position relationship between data point 0 and data point 1 in the space after dimensionality reduction.

[0146] Detailed explanation and example of the formula for step 3

[0147] formula:

[0148]

[0149] Parameter explanation:

[0150] : Data Points The fit score of .

[0151] : Adjustment parameters.

[0152] : Data Points and The distance between.

[0153] Calculation process and derivation:

[0154] Using the above calculation ,set up and (These values ​​are chosen to control the strength of the effect of distance and the flexibility of the calculation.) Compute the goodness of fit score for data point 0:

[0155]

[0156]

[0157]

[0158]

[0159] Fit score Indicates the relative fit between data point 0 and other points, where a higher value indicates a better fit.

[0160] See also Figure 3 , the specific steps for obtaining topological feature change records are:

[0161] Based on the goodness-of-fit score set, the critical value of the score is determined by statistical methods to identify topological changes, using the formula:

[0162]

[0163] Calculate the critical value;

[0164] in, is the critical value, is the average value, is the standard deviation, is the adjustment factor, is the coefficient of variation;

[0165] Using the critical value, a step function analysis was performed on each score using the formula:

[0166]

[0167] Generate a set of step function responses;

[0168] in, For data points The step response of is a step function, , is the regulating factor, Score the goodness of fit of data points;

[0169] According to the step function response set, the topology changes are counted using the formula:

[0170]

[0171] Calculate the total number of key topological changes and generate topological feature change records;

[0172] in, Indicates the total number of topology changes.

[0173] Step 1: Set the cutoff value for the goodness of fit score

[0174] formula:

[0175]

[0176] Parameter details and derivation process:

[0177] : Dataset If we assume that the data set ,but .

[0178] : Dataset The standard deviation is calculated as .

[0179] : Coefficient, used to adjust the offset of the critical value from the mean value. Here, it is assumed .

[0180] : Another adjustment factor, here assumed to be 0.5.

[0181] : Dataset The coefficient of variation, .

[0182] Calculation example:

[0183] Suppose there is a data set , calculated as follows:

[0184] calculate

[0185] calculate

[0186]

[0187]

[0188] Step 2: Calculate the step function response

[0189] formula:

[0190]

[0191] Parameter explanation and derivation process:

[0192] : step function, if but ,otherwise .

[0193] , : Nonlinear adjustment factor, here it is assumed .

[0194] Calculation example:

[0195] Continue to use the previous data and consider , calculate the step response:

[0196]

[0197] Step 3: Generate topological feature change records

[0198] formula:

[0199]

[0200] Parameter explanation and derivation process:

[0201] :if It is 1 if The value is 0, indicating whether there is a significant change.

[0202] Calculation example:

[0203] , indicating that no significant topological changes have occurred.

[0204] See also Figure 4 ,The specific steps for obtaining the initial fault location are:

[0205] Receive topology feature change records, analyze data structure changes, and use the formula:

[0206]

[0207] Calculate the anomaly index of each data point and generate a set of change indices;

[0208] in, Representative The anomaly index of the data point, For data point The value of the dimension, is the weight of the dimension, , Adjusting for the effects of exponential growth;

[0209] According to the set of change indexes, low-dimensional mapping of data is performed to emphasize key abnormal structures, using the formula:

[0210]

[0211] Evaluate the topological changes after mapping and generate a key structure score after mapping;

[0212] in, is the key structure score after mapping, is the data point value after mapping, Adjust the parameters for the mapping, is the smoothing parameter;

[0213] Using the mapped key structure scores, we identify the voids and mutation clusters that appear, using the formula:

[0214]

[0215] Determine the most critical abnormal structure and generate preliminary fault location results;

[0216] in, This is the preliminary fault location result. is the structural score, is the historical average score, is the stability adjustment factor.

[0217] Formula and detailed explanation of step 1:

[0218] formula:

[0219]

[0220] Parameter details:

[0221] : No. The significant change index of a data point is used to quantify the degree of abnormality of the data point relative to the average behavior.

[0222] : No. The data point in The value of the dimension.

[0223] : No. The dimension weight parameter determines the importance of each dimension in the calculation.

[0224] , : Parameters used to adjust the exponential growth of data and control the abnormal intensity The rate of change.

[0225] Assumed parameter values:

[0226] (All dimensions have equal weight)

[0227]

[0228]

[0229] (Assuming the value of the data point in a certain dimension)

[0230] Calculation process:

[0231] calculate Exponential growth adjustment section:

[0232] Will The direct impact is added with the exponential growth adjustment: 3.198

[0233] Applying the logic function yields :

[0234]

[0235] Result interpretation:

[0236] Indicates data points have high anomaly indices and deviate significantly from normal data behavior.

[0237] Formula and detailed explanation of step 2:

[0238] formula:

[0239]

[0240] Parameter details:

[0241] B: The key structure score after mapping, which reflects the degree of key anomaly of the data structure after mapping.

[0242] : The significant change index obtained from step 1.

[0243] : Smoothing parameter, used to avoid numerical instability caused by a too small denominator.

[0244] : Mapping adjustment parameters and mapped data point values.

[0245] Assumed parameter values:

[0246]

[0247] (Assuming all mapping dimensions have the same weight)

[0248] (Assuming the data point value after mapping)

[0249] Calculation process:

[0250] Calculate the impact of the adjustment on the mapped data point value:

[0251] The calculation formula of the key structure score after applying the mapping is:

[0252] Result interpretation:

[0253] Indicates that there is a moderate degree of critical anomaly structure in the mapped data structure.

[0254] Formula and detailed explanation of step 3:

[0255] formula:

[0256]

[0257] Parameter details:

[0258] : Preliminary fault location results, used to identify the most significant abnormal structure.

[0259] : The structure score obtained from step 2.

[0260] : Historical average score of the structure, used for baseline comparison.

[0261] : Stability adjustment factor.

[0262] Assumed parameter values:

[0263]

[0264]

[0265] Calculation process:

[0266] Calculate the adjusted structure score:

[0267] Result solution:

[0268] It means that the identified abnormal structure shows a high degree of abnormality after considering historical data and stability adjustment, which indicates a significant failure point.

[0269] See also Figure 5 , the specific steps for obtaining the fault detection results are:

[0270] Receive and analyze the preliminary fault location results, using the formula:

[0271]

[0272] Generate analysis results of abnormal locations;

[0273] in, is the analysis result of the abnormal position. is the fault location component, is the average value, is the standard deviation, is the smoothing constant, is the weight coefficient;

[0274] Using the analysis results of the abnormal position, the formula is adopted:

[0275]

[0276] Generate preliminary classification results of fault types;

[0277] in, is the preliminary classification result of the fault type. , To adjust the parameters, is the power adjustment parameter;

[0278] Based on the preliminary classification results of the fault type, the fault location and type are determined using the formula:

[0279]

[0280] Generate fault detection results;

[0281] in, is the fault detection result, is the significance threshold, is the significance adjustment parameter.

[0282] Detailed explanation and derivation of the formula for step 1

[0283] formula:

[0284]

[0285] Parameter interpretation and calculation process:

[0286] : Fault location index, assumed to be a value directly obtained from the sensor or system.

[0287] : The average value is calculated as .

[0288] : The standard deviation of .

[0289] : Smoothing constant, usually set to a small value to avoid the denominator being zero, assumed to be 0.1.

[0290] : Weight coefficient, increases the impact of abnormalities, here it is assumed to be 0.05.

[0291] Calculation example:

[0292] Assume there are three fault location index values ,

[0293]

[0294]

[0295]

[0296] For each calculate ,

[0297]

[0298]

[0299]

[0300]

[0301]

[0302]

[0303]

[0304] This sum It indicates the weighted change taking into account the abnormal position, reflecting the comprehensive impact of the abnormal degree.

[0305] Detailed explanation and derivation of the formula for step 2

[0306] formula:

[0307]

[0308] Parameter interpretation and calculation process:

[0309] : Detailed analysis results calculated in step 1.

[0310] : Sensitivity adjustment parameter, controlling the rate of exponential decay, assumed to be 0.5.

[0311] : Parameter that affects the growth rate of the denominator, assumed to be 0.2.

[0312] : Power adjustment parameter, increasing nonlinear response, here assumed to be 2.

[0313] Calculation example:

[0314] Continue using the result from step 1 ,

[0315] For each calculate ,

[0316]

[0317]

[0318]

[0319]

[0320] This sum It indicates that the classification effect of the fault type is taken into account, indicating that under a given abnormal location analysis, the fault types at different locations have different performance intensities.

[0321] Detailed explanation and derivation of the formula for step 3

[0322] formula:

[0323]

[0324] Parameter interpretation and calculation process:

[0325] : The preliminary classification result of the fault type calculated in step 2.

[0326] : Fault significance threshold, used to determine whether a fault is significant, assumed to be 0.5.

[0327] : Parameter for adjusting the significance effect, assumed to be 0.1.

[0328] Calculation example:

[0329] Use the result of step 2 ,

[0330] For each calculate ,

[0331]

[0332]

[0333]

[0334]

[0335] This maximum value Indicates the most significant fault type, that is, in this set of data, the most significant fault type has the greatest impact relative to the threshold.

[0336] See also Figure 6 ,The specific steps for obtaining the fault impact range prediction are:

[0337] Analyze the fault detection results and determine the location of the fault point in the data set using the formula:

[0338]

[0339] Calculate the location index of each fault point and generate fault location analysis results;

[0340] in, is the fault location analysis result, is the fault value, is the standard deviation, is the average value, To adjust the parameters;

[0341] Use the fault location analysis results to estimate the fault propagation path and range using the formula:

[0342]

[0343] Calculate the potential paths for fault propagation and generate fault propagation prediction results;

[0344] in, is the fault propagation prediction result, Adjust parameters for sensitivity;

[0345] Based on the fault propagation prediction results, the affected data area is analyzed using the formula:

[0346]

[0347] Calculate the impact level of each affected area and generate a prediction of the impact range of the fault;

[0348] in, To predict the impact range of the fault, is the value predicted for fault propagation, Adjust the parameters for influence.

[0349] Formula for step 1:

[0350]

[0351] Parameter explanation and assumed values:

[0352] : Fault value, assumed to be the voltage drop value recorded by a sensor in the power grid, in volts (V). For example: .

[0353] : The standard deviation of is calculated, assuming .

[0354] : The average value is calculated, assuming that .

[0355] : Adjustment coefficient, assumed to be .

[0356] Calculation process:

[0357] Calculate each Contributions:

[0358]

[0359] The same calculation

[0360] 2. Sum to get .

[0361] explain:

[0362] A larger value indicates that the fault point has a greater impact on the system.

[0363] Formula for step 2:

[0364]

[0365] Parameter explanation and assumed values:

[0366] : Adjustment coefficient, assumed to be .

[0367] Calculation process:

[0368] exist Select the maximum value, for example .

[0369] calculate :

[0370]

[0371] explain:

[0372] A larger value indicates that the propagation path is the dominant fault path.

[0373] Formula for step 3:

[0374]

[0375] Parameter explanation and assumed values:

[0376] : Adjustment coefficient, assumed to be .

[0377] Calculation process:

[0378] For each Apply the formula, assuming there is only one value .

[0379] calculate :

[0380]

[0381] explain:

[0382] The numerical value indicates the degree of impact, with higher values ​​indicating a larger affected area.

[0383] See also Figure 7 , the specific steps to establish data isolation are:

[0384] Starting from the prediction of the scope of the fault impact, the affected data point groups are identified using the formula:

[0385]

[0386] Calculate the influence weight of each data point and generate the influence weight distribution result;

[0387] in, To influence the weight distribution results, is the data point value, is the average value of the data points, is the standard deviation, is the weight factor, is the error reduction factor;

[0388] The impact weight distribution results are used to collectively correct the affected data points and prepare for the isolation of abnormal behavior data points using the formula:

[0389]

[0390] Adjust the values ​​of data points to reduce error propagation and generate collective correction results;

[0391] in, To collectively correct the results, is the original data point value, is the influence weight, To adjust the parameters;

[0392] The collective correction results are used to isolate the detected abnormal behavior data points using the formula:

[0393]

[0394] Isolate abnormal data points, limit error propagation, and establish data isolation;

[0395] in, For data isolation, is the corrected data point value, is the isolation threshold, To adjust the parameters of data isolation sensitivity, is the indicator function.

[0396] Formula 1: Calculation of impact weight distribution results

[0397]

[0398] Parameter Description:

[0399] : No. The value of the data point.

[0400] : The mean of the data points.

[0401] : Standard deviation of the data points.

[0402] : No. The weight of the data point.

[0403] : Error reduction coefficient, controlling the impact of outliers.

[0404] Calculation process and examples:

[0405] Suppose there are three data points with values ,average value , standard deviation , weight , error reduction factor .

[0406] Compute the deviation of each data point from the mean:

[0407]

[0408] Calculate the weight adjustment for each data point:

[0409]

[0410]

[0411] Calculate the probability density function using the normal distribution formula:

[0412]

[0413] Calculate influence weight distribution :

[0414]

[0415] Result interpretation:

[0416] result It represents the weighted average deviation probability density of the entire data set under given parameters.

[0417] Formula 2: Calculation of collective correction results

[0418]

[0419] Parameter Description:

[0420] : No. The value of the data point.

[0421] : No. The influence weight of each data point is obtained by Formula 1.

[0422] : Adjustment coefficient, used to balance the impact weight.

[0423] Calculation process and examples:

[0424] Continue using previous data points and influence weight . Set the adjustment factor .

[0425] Compute the collective correction for each data point:

[0426]

[0427] Calculate the total collective correction result :

[0428]

[0429] Result interpretation:

[0430] result Represents the collective correction result of the entire dataset, taking into account the influence weight of each data point and its bias adjustment.

[0431] Formula 3: Data isolation result calculation

[0432]

[0433] Parameter Description:

[0434] : No. The collective correction result of the data points is obtained by formula 2.

[0435] : Isolation threshold, which determines when to isolate a data point.

[0436] : Adjustment factor that controls the sensitivity of the isolation process.

[0437] : Indicator function, used to determine whether the isolation conditions are met.

[0438] Calculation process and examples:

[0439] Use the previous correction results and settings .

[0440] Determine and calculate data isolation:

[0441]

[0442] Calculate the total data isolation result :

[0443]

[0444] Result interpretation:

[0445] result Represents the total effect of successfully isolated data points, where data points greater than the threshold are effectively isolated, reducing the risk of error propagation.

[0446] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A fault-tolerant analysis system based on similar data, characterized in that: The system comprises: The similarity feature recognition module maps the data points in the real-time data stream to a low-dimensional space, generates a fit score by calculating the positional relationship of the data points in the low-dimensional space, and uses the fit score to identify key topological changes in the data, including connectivity, the appearance and disappearance of holes, and generates a record of topological feature changes; The dynamic fault detection module receives the topology feature change record, identifies key abnormal structures in the topology map by mapping the changes in the data structure, including newly appeared holes or mutant clusters, generates preliminary fault location, detects the abnormal position and type appearing in the data according to the preliminary fault location, and generates a fault detection result; The steps for obtaining the preliminary fault location are specifically as follows: Receive the topology feature change record, analyze the change of data structure, and use the formula: Calculate the anomaly index of each data point and generate a set of change indices; in, Representative The anomaly index of the data point, For data point The value of the dimension, is the weight of the dimension, , Adjusting for the effects of exponential growth; According to the set of change indices, low-dimensional mapping of the data is performed to emphasize key abnormal structures, using the formula: Evaluate the topological changes after mapping and generate a key structure score after mapping; in, is the key structure score after mapping, is the data point value after mapping, Adjust the parameters for the mapping, is the smoothing parameter; Using the mapped key structure scores, the voids and mutation clusters that appear are identified using the formula: Determine the most critical abnormal structure and generate preliminary fault location results; in, This is the preliminary fault location result. is the structural score, is the historical average score, is the stability adjustment factor; The steps for obtaining the fault detection result are specifically as follows: Receive and analyze the preliminary fault location results, using the formula: Generate analysis results of abnormal locations; in, is the analysis result of the abnormal position. is the fault location component, is the average value, is the standard deviation, is the smoothing constant, is the weight coefficient; Using the analysis results of the abnormal position, the formula is adopted: Generate preliminary classification results of fault types; in, is the preliminary classification result of the fault type. , To adjust the parameters, is the power adjustment parameter; Based on the preliminary classification results of the fault types, the fault location and type are determined using the formula: Generate fault detection results; in, is the fault detection result, is the significance adjustment parameter; The fault impact analysis module uses the fault detection results to analyze the location of the fault point in the data set, estimate the fault propagation path and range, analyze the affected data area, and generate a fault impact range prediction; The data recovery and isolation module performs collective corrections on groups of similar data points based on the predicted scope of impact of the fault, isolates the detected abnormal behavior data points, limits error propagation, and establishes data isolation.

2. The fault-tolerant analysis system based on similar data according to claim 1, characterized in that: The steps for obtaining the fitness score are specifically as follows: Collect data points in real-time data streams and use principal component analysis to map the data points from high-dimensional space to low-dimensional space using the formula: Calculate the mapped data point set; in, represents the original data point matrix, P represents the projection matrix, Represents the set of data points after mapping; Using the mapped data point set, the distance matrix between the data points is calculated using the distance calculation formula: Generate a distance matrix; in, Represents data points With data points The distance between , Respectively represent the data points in dimensional coordinates, is the regularization parameter; Based on the distance matrix, the goodness of fit score for each data point is calculated using the formula: Get the goodness of fit score; in, Representative data points The fit score of , To adjust the parameters, For data points and The distance between is a collection of goodness-of-fit scores.

3. The fault-tolerant analysis system based on similar data according to claim 2, characterized in that: The steps for obtaining the topological feature change record are specifically as follows: Based on the set of goodness-of-fit scores, the critical value of the score is determined by statistical methods to identify topological changes, using the formula: Calculate the critical value; in, is the critical value, is the average value, is the standard deviation, is the adjustment factor, is the coefficient of variation; Using the cutoff values, a step function analysis was performed on each score using the formula: Generate a set of step function responses; in, For data points The step response of is a step function, , is the regulating factor, Score the goodness of fit of data points; According to the step function response set, the topology change is counted using the formula: Calculate the total number of key topological changes and generate topological feature change records; in, Indicates the total number of topology changes.

4. The fault-tolerant analysis system based on similar data according to claim 1, characterized in that: The steps for obtaining the prediction of the fault impact range are specifically as follows: The fault detection results are analyzed to determine the location of the fault point in the data set using the formula: Calculate the location index of each fault point and generate fault location analysis results; in, is the fault location analysis result, is the fault value, is the standard deviation, is the average value, To adjust the parameters; Use the fault location analysis results to estimate the fault propagation path and range using the formula: Calculate the potential paths for fault propagation and generate fault propagation prediction results; in, is the fault propagation prediction result, , Adjust parameters for sensitivity; Based on the fault propagation prediction results, the affected data area is analyzed using the formula: Calculate the impact level of each affected area and generate a prediction of the impact range of the fault; in, To predict the impact range of the fault, is the value predicted for fault propagation, Adjust the parameters for influence.

5. The fault-tolerant analysis system based on similar data according to claim 4, characterized in that: The steps of establishing data isolation are specifically as follows: Starting from the fault impact range prediction, the affected data point groups are identified using the formula: Calculate the influence weight of each data point and generate the influence weight distribution result; in, To influence the weight distribution results, is the data point value, is the average value of the data points, is the standard deviation, is the weight factor, is the error reduction factor; The impact weight distribution result is used to collectively correct the affected data points and prepare for the isolation of abnormal behavior data points, using the formula: Adjust the values ​​of data points to reduce error propagation and generate collective correction results; in, To collectively correct the results, is the original data point value, is the influence weight, To adjust the parameters; The collective correction result is used to isolate the detected abnormal behavior data points, using the formula: Isolate abnormal data points, limit error propagation, and establish data isolation; in, For data isolation, is the corrected data point value, is the isolation threshold, To adjust the parameters of data isolation sensitivity, is the indicator function.

Citation Information

Patent Citations

  • Method and device for fault-tolerant, time-controlled real-time communication

    AT509700A1

  • Online identification method and system for real-time topological relation of acquired voltage

    CN115144679A