A Power Data Intelligent Monitoring System and Method Based on Multivariate Data Analysis

By constructing a many-to-many structured mapping table and fault feature fingerprint matching, the problem of unjoint analysis of multi-category data in the power system is solved, enabling efficient and accurate fault monitoring and coordinated control of power equipment.

CN123122015APending Publication Date: 2026-11-17JIANGSU PAINENG POWER TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611294223.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-25
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In existing technologies, data analysis in power systems fails to effectively combine multiple types of data, resulting in the inability to extract key correlation information in the early stages of faults. Single-dimensional threshold features suffer from homogenization interference, leading to misjudgment of rare faults and incorrect identification of common faults, which reduces the accuracy of similarity analysis of power data.

Method used

Based on multivariate data analysis, a many-to-many structured mapping table is constructed between fault types and emergency control actions. Through fault feature fingerprint matching and similarity analysis, real-time fault monitoring and coordinated control of power equipment are realized.

Benefits of technology

It improves the efficiency and accuracy of intelligent monitoring of power data, overcomes the homogeneity interference of single-dimensional threshold alarm features, reduces the probability of misjudging rare faults and common faults, and realizes high-precision fault identification and maintenance of power equipment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123122015A_ABST
    Figure CN123122015A_ABST
Patent Text Reader

Abstract

This invention discloses a power data intelligent monitoring system and method based on multivariate data analysis, belonging to the field of power data intelligent monitoring technology. The invention includes: S10: constructing a many-to-many structured mapping table between fault types and emergency control actions; S20: determining fault samples corresponding to various emergency control actions; S30: performing cold start completion processing on the many-to-many structured mapping table between fault types and emergency control actions; S40: performing similarity analysis on the determined fault feature fingerprints; S50: performing corresponding fault maintenance and linkage control processing on power equipment in real time through the time sequence of emergency control actions. This invention obtains fault feature fingerprints by horizontally concatenating the time-series evolution morphology submatrix and the index coupling strength submatrix in the same dimension, generating a unique full-feature digital profile of each type of fault, overcoming the homogeneity interference problem of single-dimensional threshold alarm features, and improving the intelligent monitoring accuracy of the system.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of intelligent power data monitoring technology, specifically to an intelligent power data monitoring system and method based on multivariate data analysis. Background Technology

[0002] With the continuous development of smart grid construction, the requirements for the safe and stable operation of power equipment are becoming increasingly stringent. Accurately assessing the operating status of power equipment is of great significance for ensuring the safe, reliable, and economical operation of the power system.

[0003] Currently, data mining and analysis in power systems are still in a state of single-category analysis, which only isolates and analyzes single-dimensional explicit threshold features such as current exceeding limits and temperature exceeding standards. This lack of combined analysis of multi-category data results in a large amount of critical correlation information in the early stages of faults being unable to be obtained through data analysis and mining. At the same time, single-dimensional threshold features suffer from homogenization interference problems due to the high degree of overlap between different fault alarm features. Furthermore, when analyzing power data, the differences in the local distribution density of the analysis samples are not taken into account, leading to abnormal misjudgments of rare faults and incorrect identification of common faults, thereby reducing the accuracy of similarity analysis of power data. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent power data monitoring system and method based on multivariate data analysis to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for intelligent monitoring of power data based on multivariate data analysis, characterized in that: the method includes: S10: Based on the historical dispatch and operation data of power equipment, obtain emergency control events of power equipment under different fault states, and construct a many-to-many structured mapping table between fault types and emergency control actions; S20: Based on the leading-edge sliding time window of the action time window corresponding to various emergency control actions, retrieve the set of all historical basic operation data segments corresponding to the power equipment to determine the fault samples corresponding to various emergency control actions. S30: Analyze the fault feature fingerprints corresponding to each fault sample based on the temporal evolution and index coupling of the fault samples, and match the fault feature fingerprints with the emergency control actions corresponding to the fault samples. Based on the matching results, perform cold start completion processing on the many-to-many structured mapping table between fault types and emergency control actions. S40: Based on the time series of basic operating data monitored in each monitoring interval, determine the fault feature fingerprint of the power equipment in each monitoring interval, and perform similarity analysis on the determined fault feature fingerprint. S50: Generates an emergency control action sequence corresponding to the real-time fault identifier of the power equipment based on a many-to-many structured mapping table, and performs corresponding fault maintenance and linkage control processing on the power equipment in real time through the emergency control action sequence.

[0006] Furthermore, S10 includes: S101: Obtain historical dispatch operation data of power equipment. The dispatch operation data includes basic operation data of power equipment, dispatch control data, fault event ledger data and external related data. Align the dispatch control data and fault event ledger data according to time sequence to determine emergency control events of power equipment under different states. The emergency control event refers to the complete time sequence record of the automatic dispatch system or dispatch personnel issuing dispatch control instructions, instruction transmission and equipment response within the entire cycle from the moment the power equipment fails to the completion of the fault handling closed loop. S102: Based on the fault type of power equipment, classify the determined emergency control events to obtain several subsets of emergency control events, and bind and match each fault type with the corresponding subset of emergency control events one by one. S103: Traverse the subset of emergency control events that match each fault type, extract the full-link emergency control action sequence corresponding to each emergency control event in the subset of emergency control events, select a fault type in turn, and synchronously traverse the adjacent device linkage operation data in the external related data monitored within the full cycle time window corresponding to each emergency control event in the subset of emergency control events that match the currently selected fault type, and extract the cross-device linkage emergency control action sequence. The full-link emergency control action sequence and cross-device linkage emergency control action sequence corresponding to the selected fault types are collected to obtain the target action sequence. The target action sequence is then processed, and a many-to-many structured mapping table between fault types and emergency control actions is constructed based on the processing results.

[0007] Furthermore, the specific method for processing the target action sequence is as follows: Repeated emergency control actions in the target action sequence are deduplicated. The deduplication criterion is a complete match of the emergency control action identifier string. The ratio between the total number of historical executions of each emergency control action in the target action sequence after deduplication and the total number of emergency control events stored in the selected emergency control event subset matching the fault type is calculated. Emergency control actions with a ratio lower than a set threshold are deleted. Then, the fault identifier corresponding to the selected fault type is associated with all emergency control action identifiers stored in the target action sequence after deduplication and deletion in a structured field representation. All classified fault types are traversed to construct a many-to-many structured mapping table between fault types and emergency control actions.

[0008] Furthermore, S20 includes: S201: In the many-to-many structured mapping table between fault types and emergency control actions, traverse all emergency control actions, and based on the action time window set corresponding to each emergency control action, retrieve the set of historical basic operation data segments stored in the leading-edge sliding time window of each action time window set to obtain several sets of historical basic operation data segments. S202: Bind the emergency control action to the set of all historical basic operation data segments obtained based on the emergency control action, and mark the sample set composed of all the bound historical basic operation data segments as the fault sample of the emergency control action. The basic operation data includes current harmonic distortion rate, voltage drop slope and partial discharge pulse count, wherein the current harmonic distortion rate is the core main indicator, and the voltage drop slope and partial discharge pulse count are auxiliary indicators.

[0009] Let the current harmonic distortion rate, voltage drop slope, and partial discharge pulse count correspond to numbers 1, 2, and 3, respectively. The time window is slid to the left based on the left endpoint of the leading edge sliding time window. Duration, according to The set of historical basic runtime data segments corresponding to the duration of the sliding window. The duration of the sliding window is entirely within the normal steady-state operating range before the power equipment failure occurs. The sliding arithmetic mean values ​​corresponding to the current harmonic distortion rate, voltage drop slope, and partial discharge pulse count are calculated. , , and standard deviation , , ,according to The normal operation benchmark threshold corresponding to the indicator numbered i is calculated, where, This indicates the number corresponding to each indicator. This represents the confidence coefficient corresponding to indicator number i, with a value range of... , This represents the normal operating baseline threshold corresponding to indicator number i. Specifically, when and hour, The value is 30s. hour, The value is 60s. The basic operating data collected within the sliding window of the duration is under the normal operating conditions of the power equipment. In the early stage of insulation deterioration of the power equipment, partial discharge pulses are low-frequency and sparse events. Therefore, the 60-second sliding window can fully capture sporadic discharge pulses, avoid missing early weak insulation abnormalities, and thus ensure the sensitivity of fault feature identification.

[0010] Furthermore, the action time window refers to the trigger moment of the emergency control action. Starting from the point where the emergency control action ends. The continuous time interval of the endpoint, the action time window is The aforementioned leading-edge sliding time window refers to the period from the timestamp corresponding to the starting point of the latent feature to the moment when the emergency control action is triggered. The continuous time interval, the leading edge sliding time window is The latent feature starting point refers to the timestamp in the basic operation data of power equipment where a single core main indicator first exceeds the normal operation benchmark threshold for three consecutive sampling points or two auxiliary indicators simultaneously exceed the normal operation benchmark threshold for three consecutive sampling points. Within the leading-edge sliding time window, all latent and abrupt fault features before the control action is triggered can be completely captured, overcoming the problem that conventional fixed backtracking windows are prone to mixing in too much normal operation data, which is conducive to improving the accuracy of fault feature identification.

[0011] Furthermore, S30 includes: S301: Exploit The standardized formula normalizes the fault samples. The standardized formula is: ,in, This represents the historical normalized result corresponding to indicator number i. This represents the historical sampled value corresponding to indicator number i. It retrieves the normalized historical basic operational data segment set for each fault sample, and obtains the timestamp of the first time each indicator in each retrieved historical basic operational data segment set exceeds the normal operation benchmark threshold. Using the implicit feature starting point of the leading-edge sliding time window corresponding to each retrieved historical basic operational data segment set as the anchor point, it calculates the relative exceedance time series ratio of each indicator in each historical basic operational data segment set, thus obtaining the temporal evolution morphology sub-matrix B of the fault sample. The relative exceedance time series ratio is used to represent the ratio of the difference between the obtained timestamp and the timestamp corresponding to the anchor point to the time length corresponding to the leading-edge sliding time window. If indicator i does not exceed the normal operation benchmark threshold for the first time within the leading-edge sliding time window, the relative exceedance time series ratio corresponding to indicator i is assigned a value of 1. Where n represents the total number of normalized historical basic operational data segments stored in the fault sample, and the j-th column of the time-series evolution morphology sub-matrix corresponds to the historical basic operational data segment set numbered j. , This represents the relative time-series ratio of indicator number 3 in the historical basic running data segment set numbered n. The first, second, and third rows of the time-series evolution morphology sub-matrix correspond to indicators numbered 1, 2, and 3, respectively. S302: Divide each set of normalized historical basic operational data segments according to the indicator type to obtain three time series of historical basic operational data for different indicators. Calculate the Pearson correlation coefficient between any two time series of historical basic operational data in each set of historical basic operational data segments using the Pearson correlation coefficient formula to obtain the indicator coupling strength submatrix D of the fault sample. ,in, This represents the Pearson correlation coefficient between the historical basic operational data time series corresponding to indicator number 2 and indicator number 3 in the historical basic operational data segment set numbered n. The j-th column of the indicator coupling strength submatrix corresponds to the historical basic operational data segment set numbered j. The first, second, and third rows of the index coupling strength submatrix correspond to the coupling relationship between index number 1 and index number 2, index number 1 and index number 3, and index number 2 and index number 3, respectively. If any historical basic operational data time series involved in the calculation has no fluctuations throughout and causes the denominator of the Pearson correlation coefficient calculation formula to be 0, then the corresponding Pearson correlation coefficient is assigned a value of 0. S303: Based on feature dimension alignment rules, the temporal evolution morphology submatrix and the indicator coupling strength submatrix are horizontally concatenated in the same dimension to obtain the fault feature fingerprint of the fault sample. The fault feature fingerprint of the fault sample is matched with the emergency control action bound to the fault sample, and cold start completion is performed on the many-to-many structured mapping table between fault type and emergency control action according to the matching result. The smallest storage unit of the many-to-many structured mapping table after cold start completion is "fault identifier ↔ emergency control action identifier ↔ fault feature fingerprint". By using the relative limit exceedance time series ratio, the order of the first limit exceedance of different indicators and the time difference ratio of the fault anchor point are all encoded into the temporal evolution morphology submatrix, which can accurately record the complete process of fault from bud to outbreak. The entire time-series evolution path, through Pearson correlation coefficient, encodes the linkage and correlation strength between different indicators into the indicator coupling strength submatrix. This can accurately capture the linkage and coupling patterns between multiple indicators when a fault occurs. The fault feature fingerprint, composed of the time-series evolution morphology submatrix and the indicator coupling strength submatrix, generates a unique full-feature digital profile of each type of fault, overcoming the homogeneity interference problem of single-dimensional threshold alarm features such as current exceeding limits and temperature exceeding limits. At the same time, relying on the full-dimensional implicit feature encoding capability of the fault feature fingerprint, it is conducive to completing high-precision fault mode identification of power equipment fault types collected within the monitoring time window, especially suitable for special scenarios where power fault samples are unevenly distributed and similar fault features highly overlap.

[0012] Furthermore, S40 includes: S401: The total duration corresponding to the sliding window is taken as the monitoring interval length of the power equipment power data. The time series of basic operation data monitored within the monitoring interval is taken as the sample to be analyzed. The fault feature fingerprint of the sample to be analyzed is calculated and used as the real-time fault feature fingerprint of the power equipment at the timestamp corresponding to the right end of the monitoring interval, denoted as fault feature fingerprint X. The calculation method of the fault feature fingerprint of the sample to be analyzed is the same as the calculation method of the fault feature fingerprint of the fault sample. S402: Traverse all fault feature fingerprints stored in the many-to-many structured mapping table, and record each fault feature fingerprint as fault feature fingerprint Y in turn. Flatten fault feature fingerprint Y and fault feature fingerprint X into one-dimensional feature vectors in descending order of the weights assigned to each indicator. S403: Calculate the two one-way reachability distances between fault feature fingerprint Y and fault feature fingerprint X, and take the arithmetic mean of the two calculated one-way reachability distances. Record the arithmetic mean as... According to the formula Calculate the similarity between fault feature fingerprint X and fault feature fingerprint Y, where, Representing an exponential function with base e and e=2.72, the two unidirectional reachability distances obtained through calculation can avoid the numerical bias caused by using only the distance in one direction. This ensures that the similarity obtained by taking the arithmetic mean of the two unidirectional reachability distances is not affected by two scenarios: sparse local distribution of fault samples and excessively dense high-density areas of fault samples. This reduces the probability of misjudging rare faults with sparse edges as anomalies or misjudging different faults with similar features in high-density areas as the same type of fault.

[0013] Furthermore, the specific calculation method for the two unidirectional reachability distances between the fault feature fingerprint Y and the fault feature fingerprint X is as follows: Euclidean distance between fault feature fingerprint X and fault feature fingerprint Y Perform calculations and sort all the calculated Euclidean distances in ascending order of their numerical values. Sort the samples to obtain the distance sorting sequence 1 corresponding to the fault feature fingerprint X; Will The distance from fault feature fingerprint X to fault feature fingerprint Y is defined as the unidirectional reachability distance, where k represents a positive integer and is less than the total number of fault feature fingerprints that the many-to-many structured mapping table can store. This represents the Euclidean distance of the k-th position in sorted sequence 1; Iterate through all fault fingerprints in the many-to-many structured mapping table except for fault fingerprint Y, and calculate the Euclidean distance between fault fingerprint Y and each of the remaining fault fingerprints traversed. All calculated Euclidean distances are processed in ascending order of data. Sort the data to obtain the distance sorting sequence 2 corresponding to the fault feature fingerprint Y. The distance that can be traveled unidirectionally from fault feature fingerprint Y to fault feature fingerprint X. represents the Euclidean distance at the k-th position in sorted sequence 2, and max indicates taking the sign of the maximum value.

[0014] Furthermore, S50 includes: S501: Traverse all fault feature fingerprints stored in the many-to-many structured mapping table, and calculate the similarity between each fault feature fingerprint and the fault feature fingerprint X in each traversal. S502: When all similarities are less than the preset similarity threshold, the fault identifier corresponding to the fault feature fingerprint X is marked as a new fault identifier, and the mapping record h consisting of the new fault identifier, the fault feature fingerprint corresponding to the new fault identifier, and the empty-position emergency control action identifier is automatically written into the many-to-many structured mapping table. S503: When there is a similarity greater than or equal to a preset similarity threshold, the fault identifier bound to the fault feature fingerprint corresponding to the maximum similarity is used as the real-time fault identifier of the power equipment. Specifically, when there are multiple identical maximum similarities, the cumulative number of fault samples bound to the fault feature fingerprint corresponding to each maximum similarity is stored in set G. If there is a unique maximum value in set G, the fault identifier bound to the fault feature fingerprint corresponding to the maximum value stored in set G is used as the real-time fault identifier of the power equipment. If there are multiple identical maximum values ​​in set G, the historical update frequency of the mapping record bound to the fault feature fingerprint corresponding to each maximum value in set G is obtained, and the fault identifier bound to the fault feature fingerprint corresponding to the maximum update frequency is used as the real-time fault identifier of the power equipment. All emergency control action identifiers bound to real-time fault identifiers are retrieved from the many-to-many structured mapping table. An emergency control action sequence is generated according to the order of priority from high to low for each emergency control action identifier. The generated emergency control action sequence is then sent to the edge controller of the power equipment. The edge controller performs corresponding fault maintenance and linkage control processing on the power equipment in real time according to the sent content. The cumulative number of fault samples corresponding to the fault feature fingerprints bound to the finally determined real-time fault identifiers in the many-to-many structured mapping table is incremented by 1. The priority of the emergency control action identifier is determined by the total number of fault samples in the fault feature fingerprints bound to the emergency control action. The larger the total number of fault samples, the higher the corresponding priority. When the total number of fault samples in the fault feature fingerprints bound to multiple emergency control action identifiers is completely equal, the shorter the historical single execution time of the emergency control action, the higher the corresponding priority.

[0015] A power data intelligent monitoring system based on multivariate data analysis, the system includes a mapping table construction module, a fault sample analysis module, a cold start completion processing module, a similarity analysis module, and an operation and maintenance and linkage control module; The mapping table construction module is used to construct a many-to-many structured mapping table between fault types and emergency control actions; The fault sample analysis module is used to analyze fault samples corresponding to various emergency control actions. The cold start completion processing module is used to perform cold start completion processing on the many-to-many structured mapping table between fault types and emergency control actions. The similarity analysis module is used to determine the fault feature fingerprints of power equipment in each monitoring interval and to perform similarity analysis on the determined fault feature fingerprints. The operation and maintenance and linkage control module is used to perform corresponding fault maintenance and linkage control processing on power equipment in real time through emergency control action sequence.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention initially constructs a many-to-many structured mapping table between fault types and emergency control actions by using scheduling control data, fault event ledger data, and external related data. Based on the fault feature fingerprints generated from the equipment's basic operating data, the initially constructed many-to-many structured mapping table is cold-started and completed. The cold-started and completed many-to-many structured mapping table uses "fault identifier ↔ emergency control action identifier ↔ fault feature fingerprint" as the smallest storage unit, representing the joint analysis of multiple types of data, which greatly improves the monitoring efficiency and accuracy of intelligent monitoring of power data.

[0017] 2. This invention accurately records the complete temporal evolution path of a fault from its inception to its outbreak through a temporal evolution morphology submatrix, and accurately captures the linkage and coupling patterns between multiple indicators when a fault occurs through an indicator coupling strength submatrix. By horizontally splicing the temporal evolution morphology submatrix and the indicator coupling strength submatrix in the same dimension, a fault feature fingerprint is obtained, generating a unique digital profile of the full characteristics of each type of fault. This overcomes the homogeneity interference problem of single-dimensional threshold alarm features such as current exceeding limits and temperature exceeding standards, and further improves the accuracy of intelligent monitoring of power data.

[0018] 3. This invention calculates the similarity between fault feature fingerprints by taking the arithmetic mean of two unidirectional reachable distances and then calculating the similarity based on the arithmetic mean. This similarity is not affected by two scenarios: sparse local distribution of fault samples and excessively dense high-density areas of fault samples. This reduces the probability of misjudging rare faults with sparse edges as anomalies or misjudging different faults with similar features in high-density areas as the same type of fault, thereby further improving the accuracy of intelligent monitoring of power data. Attached Figure Description

[0019] Figure 1 This is a schematic diagram illustrating the workflow of an intelligent power data monitoring method based on multivariate data analysis according to the present invention. Detailed Implementation

[0020] Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0021] like Figure 1 As shown, this invention provides a technical solution for intelligent monitoring of power data based on multivariate data analysis. The method includes: S10: Based on the historical dispatch and operation data of power equipment, obtain emergency control events of power equipment under different fault states, and construct a many-to-many structured mapping table between fault types and emergency control actions; S10 includes: S101: Acquire historical dispatch operation data of power equipment. Dispatch operation data includes basic operation data of power equipment, dispatch control data, fault event ledger data, and external correlation data. Align dispatch control data and fault event ledger data according to time sequence to determine emergency control events of power equipment under different states. Emergency control events refer to the complete time sequence record of dispatch control commands issued by the automatic dispatch system or dispatch personnel, command transmission, and equipment response within the entire cycle from the moment a fault occurs to the completion of the fault handling closed loop. Dispatch control data includes switch opening and closing status, protection device action signals, dispatch command records, load control command execution results, etc. Fault event ledger data includes historical fault ledgers of power equipment, full-process fault handling traceability records, and fault classification and source tracing reports. External correlation data includes linkage operation data of adjacent equipment, etc. S102: Based on the fault types of power equipment, the identified emergency control events are classified and processed to obtain several subsets of emergency control events. Each fault type is then bound and matched with its corresponding subset of emergency control events. All emergency control events collected under each fault type share the same fault triggering characteristics and operational impact attributes, and are adapted to the same emergency control handling logic. Fault triggering characteristics refer to the earliest quantifiable abnormal characteristics that can be captured by monitoring sensors when a power equipment fault occurs, specifically including winding temperature exceeding a set threshold and current harmonic distortion rate exceeding limits. Operational impact attributes refer to the quantifiable negative impacts on the operational stability of the power equipment itself after a fault occurs, specifically including the percentage increase in insulating oil dielectric loss and the percentage reduction in expected winding life. Emergency control handling logic refers to the logic applied to the fault type. A pre-defined, standardized economic handling procedure executed in a fixed sequence. For example, suppose emergency control event a corresponds to a winding temperature of 68℃ in phase A of main transformer No. 1, and emergency control event b corresponds to a winding temperature of 63℃ in phase A of main transformer No. 3. Both temperatures exceed the set threshold of 55℃. The fault triggering characteristics of emergency control events a and b are both winding temperatures exceeding the set threshold, and the corresponding operational impact attributes are both an increase in insulating oil dielectric loss value greater than or equal to 15% and a reduction in expected winding life greater than or equal to 20%. The emergency control handling logic for emergency control events a and b is both "execute load reduction to 80% of rated capacity → start cooling system → push alarm work order". Therefore, the fault type corresponding to emergency control events a and b is both winding overheating fault. Thus, emergency control events a and b are stored in the same subset of emergency control events. S103: Traverse the subset of emergency control events that match each fault type, extract the full-link emergency control action sequence corresponding to each emergency control event in the subset of emergency control events, select a fault type in turn, and synchronously traverse the adjacent device linkage operation data in the external related data monitored within the full cycle time window corresponding to each emergency control event in the subset of emergency control events that match the currently selected fault type, and extract the cross-device linkage emergency control action sequence. The system aggregates the end-to-end emergency control action sequences and cross-device linkage emergency control action sequences corresponding to the selected fault types to obtain the target action sequence. Duplicate emergency control actions in the target action sequence are deduplicated based on a complete match of the emergency control action identifier string. The ratio between the total historical execution count of each emergency control action in the deduplicated target action sequence and the total number of emergency control events stored in the subset of emergency control events matching the selected fault type is calculated. Emergency control actions with a ratio lower than a set threshold are deleted. Then, the fault identifier corresponding to the selected fault type is associated with all emergency control action identifiers stored in the target action sequence after deduplication and deletion using a structured field association representation. All classified fault types are traversed to construct a many-to-many structured mapping table between fault types and emergency control actions. The threshold value range is set to... The time frame for capturing the emergency control action sequence of the entire link is from the time of fault occurrence recorded in the emergency control event to the time of completion of the fault handling closed loop recorded in the emergency control event. The time of completion of the fault handling closed loop is the timestamp of the generation of the execution feedback message of the last emergency control action corresponding to the emergency control event. The many-to-many structured mapping table uses "fault identifier ↔ emergency control action identifier ↔ fault feature fingerprint" as the smallest storage unit, and supports the many-to-many relationship of one type of fault corresponding to multiple sets of emergency control actions and one type of emergency control action adapting to multiple types of faults. S20: Based on the leading-edge sliding time window of the action time window corresponding to various emergency control actions, retrieve the set of all historical basic operation data segments corresponding to the power equipment to determine the fault samples corresponding to various emergency control actions. S20 includes: S201: In the many-to-many structured mapping table between fault types and emergency control actions, traverse all emergency control actions, and based on the action time window set corresponding to each emergency control action, retrieve the set of historical basic operation data segments stored in the leading-edge sliding time window of each action time window set to obtain several sets of historical basic operation data segments. Action time window refers to the moment when an emergency control action is triggered. Starting from the point where the emergency control action ends. The continuous time interval of the endpoint, the action time window is The front-sliding time window refers to the time stamp corresponding to the starting point of the latent feature. When the emergency control action is triggered The continuous time interval, the leading edge sliding time window is Within the leading-edge sliding time window corresponding to each emergency control action, there is only one hidden feature starting point. This hidden feature starting point is the left endpoint of the leading-edge sliding time window. Apart from this left endpoint, there are no other hidden feature starting points that meet the rules for determining hidden feature starting points. The hidden feature starting point refers to the timestamp in the basic operating data of power equipment when a single core main indicator exceeds the normal operating benchmark threshold for the first time in three consecutive sampling points or when two auxiliary indicators exceed the normal operating benchmark threshold for the first time in three consecutive sampling points. S202: Bind the emergency control action with the set of all historical basic operation data segments obtained based on the emergency control action, and mark the sample set composed of all the bound historical basic operation data segments as the fault sample of the emergency control action. The basic operation data includes current harmonic distortion rate, voltage drop slope and partial discharge pulse count. Among them, current harmonic distortion rate is the core main indicator, and voltage drop slope and partial discharge pulse count are auxiliary indicators. Voltage drop slope is used to represent the rate of change of the bus voltage of the power equipment in a short period of time. Current harmonic distortion rate refers to the ratio of the total effective value of each harmonic component to the effective value of the fundamental component in the periodic non-sinusoidal current waveform. Partial discharge pulse count refers to the total number of effective partial discharge pulses with amplitudes exceeding the preset discharge amplitude threshold collected by the high-frequency partial discharge sensor within the monitoring time window. Let the current harmonic distortion rate, voltage drop slope, and partial discharge pulse count correspond to numbers 1, 2, and 3, respectively. The time window is slid to the left based on the left endpoint of the leading edge sliding time window. Duration, according to The set of historical basic runtime data segments corresponding to the duration of the sliding window. The duration of the sliding window is entirely within the normal steady-state operating range before the power equipment failure occurs. The sliding arithmetic mean values ​​corresponding to the current harmonic distortion rate, voltage drop slope, and partial discharge pulse count are calculated. , , and standard deviation , , ,according to The normal operation benchmark threshold corresponding to the indicator numbered i is calculated, where, This indicates the number corresponding to each indicator. This represents the confidence coefficient corresponding to indicator number i, with a value range of... , This represents the normal operating baseline threshold corresponding to indicator number i. Specifically, when and hour, The value is 30s. hour, The value is 60s. The basic operational data collected within the sliding window of the specified duration is under the normal operating conditions of the power equipment. S30: Analyze the fault feature fingerprints corresponding to each fault sample based on the temporal evolution and index coupling of the fault samples, and match the fault feature fingerprints with the emergency control actions corresponding to the fault samples. Based on the matching results, perform cold start completion processing on the many-to-many structured mapping table between fault types and emergency control actions. S30 includes: S301: Exploit The standardized formula normalizes the fault samples. The standardized formula is: ,in, This represents the historical normalized result corresponding to indicator number i. This represents the historical sampled value corresponding to indicator number i. It retrieves the normalized historical basic operational data segment set for each fault sample, and obtains the timestamp of the first time each indicator in each retrieved historical basic operational data segment set exceeds the normal operation benchmark threshold. Using the implicit feature starting point of the leading-edge sliding time window corresponding to each retrieved historical basic operational data segment set as the anchor point, it calculates the relative exceedance time series ratio of each indicator in each historical basic operational data segment set, obtaining the temporal evolution morphology submatrix B of the fault sample. The relative exceedance time series ratio represents the ratio of the difference between the obtained timestamp and the timestamp corresponding to the anchor point to the time length corresponding to the leading-edge sliding time window. If indicator i does not exceed the normal operation benchmark threshold for the first time within the leading-edge sliding time window, the relative exceedance time series ratio corresponding to indicator i is assigned a value of 1. Where n represents the total number of normalized historical basic operational data segments stored in the fault sample, and the j-th column of the time-series evolution morphology sub-matrix corresponds to the historical basic operational data segment set numbered j. , This represents the relative time-series ratio of indicator number 3 in the historical basic running data segment set numbered n. The first, second, and third rows of the time-series evolution morphology sub-matrix correspond to indicators numbered 1, 2, and 3, respectively. S302: Divide each set of normalized historical basic operational data segments according to the indicator type to obtain three time series of historical basic operational data for different indicators. Calculate the Pearson correlation coefficient between any two time series of historical basic operational data in each set of historical basic operational data segments using the Pearson correlation coefficient formula to obtain the indicator coupling strength submatrix D of the fault sample. ,in, This represents the Pearson correlation coefficient between the historical basic operational data time series corresponding to indicator number 2 and indicator number 3 in the historical basic operational data segment set numbered n. The j-th column of the indicator coupling strength submatrix corresponds to the historical basic operational data segment set numbered j. The first, second, and third rows of the index coupling strength submatrix correspond to the coupling relationship between index number 1 and index number 2, index number 1 and index number 3, and index number 2 and index number 3, respectively. If any historical basic operational data time series involved in the calculation has no fluctuations throughout and causes the denominator of the Pearson correlation coefficient calculation formula to be 0, then the corresponding Pearson correlation coefficient is assigned a value of 0. S303: Based on the feature dimension alignment rule, the temporal evolution morphology submatrix and the index coupling strength submatrix are horizontally concatenated in the same dimension to obtain the fault feature fingerprint of the fault sample. The first to nth columns of the matrix corresponding to the fault feature fingerprint are the temporal evolution morphology feature regions. arrive The index coupling strength feature area is listed, and the fault feature fingerprint of the fault sample is matched with the emergency control action bound to the fault sample. Based on the matching result, the many-to-many structured mapping table between the fault type and the emergency control action is cold-started and completed. The smallest storage unit of the many-to-many structured mapping table after cold-starting and completion is "fault identifier ↔ emergency control action identifier ↔ fault feature fingerprint". Cold-starting and completion refers to the completion operation of the three-element binding "fault identifier ↔ emergency control action identifier ↔ fault feature fingerprint" when the mapping table initially constructed only by scheduling control data, fault event ledger data and external related data has missing records and incomplete feature dimensions. It uses the fault feature fingerprint generated by the same source calculation of the historical basic operation data of power equipment to complete the completion operation of the three-element binding "fault identifier ↔ emergency control action identifier ↔ fault feature fingerprint". S40: Based on the time series of basic operating data monitored in each monitoring interval, determine the fault feature fingerprint of the power equipment in each monitoring interval, and perform similarity analysis on the determined fault feature fingerprint. S40 includes: S401: The total duration corresponding to the sliding window is taken as the monitoring interval length of the power equipment power data. The time series of basic operation data monitored within the monitoring interval is taken as the sample to be analyzed. The fault feature fingerprint of the sample to be analyzed is calculated and used as the real-time fault feature fingerprint of the power equipment at the timestamp corresponding to the right end of the monitoring interval, denoted as fault feature fingerprint X. The calculation method of the fault feature fingerprint of the sample to be analyzed is the same as the calculation method of the fault feature fingerprint of the fault sample. S402: Traverse all fault feature fingerprints stored in the many-to-many structured mapping table, and denot each fault feature fingerprint as fault feature fingerprint Y. Flatten fault feature fingerprints Y and X into one-dimensional feature vectors according to the weights assigned to each indicator in descending order. The weights assigned to each indicator are determined based on the fault warning priority and are complementary and equal. Specifically, the core indicator, current harmonic distortion rate, is assigned the highest weight, while the auxiliary indicators, voltage drop slope and partial discharge pulse count, are assigned progressively decreasing weights. The flattening operation starts from the first element of the first row of the matrix corresponding to the fault feature fingerprint and traverses row by row. The last element of the first row of the matrix corresponding to the fault feature fingerprint is adjacent to the first element of the second row, and the last element of the second row is adjacent to the first element of the third row, resulting in a feature vector of dimension X. One-dimensional eigenvectors; S403: Euclidean distance between fault feature fingerprint X and fault feature fingerprint Y Perform calculations and sort all the calculated Euclidean distances in ascending order of their numerical values. Sort the samples to obtain the distance sorting sequence 1 corresponding to the fault feature fingerprint X; Will The distance from fault feature fingerprint X to fault feature fingerprint Y is defined as the unidirectional reachability distance, where k represents a positive integer and is less than the total number of fault feature fingerprints that the many-to-many structured mapping table can store. Specifically, the value of k can be set to a range of... Because the historical dispatch and operation data of power equipment, as well as the accumulated fault event ledger data, are sufficient, the number of valid "fault identifier ↔ emergency control action identifier ↔ fault feature fingerprint" ternary mapping records stored in the many-to-many structured mapping table after cold start completion is no less than 10, which fully meets the minimum sample size requirement for bidirectional and unidirectional reachability distance similarity calculation. The Euclidean distance at the k-th position in sorted sequence 1 is equivalent to the k-th nearest neighbor reference distance of fault feature fingerprint X. S404: Traverse all fault feature fingerprints in the many-to-many structured mapping table except for fault feature fingerprint Y, and calculate the Euclidean distance between fault feature fingerprint Y and each of the remaining fault feature fingerprints traversed. All calculated Euclidean distances are processed in ascending order of data. Sort the data to obtain the distance sorting sequence 2 corresponding to the fault feature fingerprint Y. The distance that can be traveled unidirectionally from fault feature fingerprint Y to fault feature fingerprint X. The Euclidean distance at the k-th position in the sorted sequence 2 is equivalent to the k-th nearest neighbor reference distance of the fault feature fingerprint Y, and max indicates taking the sign of the maximum value. The two one-way reachability distances between fault feature fingerprint Y and fault feature fingerprint X are calculated, and the arithmetic mean of the two calculated one-way reachability distances is denoted as . According to the formula Calculate the similarity between fault feature fingerprint X and fault feature fingerprint Y, where, This represents an exponential function with base e and The similarity score ranges from 1 to 10. The closer the similarity value is to 1, the more similar the fault feature fingerprint X and the fault feature fingerprint Y are. S50: Generates an emergency control action sequence corresponding to the real-time fault identifier of the power equipment based on a many-to-many structured mapping table, and performs corresponding fault maintenance and linkage control processing on the power equipment in real time through the emergency control action sequence; The S50 includes: S501: Traverse all fault feature fingerprints stored in the many-to-many structured mapping table, and calculate the similarity between each fault feature fingerprint and the fault feature fingerprint X in each traversal. S502: When all similarities are less than the preset similarity threshold, the fault identifier corresponding to the fault feature fingerprint X is marked as a new fault identifier, and the mapping record h consisting of the new fault identifier, the fault feature fingerprint corresponding to the new fault identifier, and the empty-position emergency control action identifier is automatically written into the many-to-many structured mapping table. After the entire closed-loop process of on-site handling of a new fault is completed, the full-link emergency control action sequence corresponding to the emergency control event of the new fault is extracted. All emergency control action identifiers in the full-link emergency control action sequence are generated into independent mapping records according to the execution order. Each mapping record is only bound to one emergency control action identifier in the full-link emergency control action sequence. All newly generated mapping records are associated with the fault identifier of the new fault. First, the mapping record h in the many-to-many structured mapping table is deleted. Then, all newly generated mapping records are written into the many-to-many structured mapping table in batches to complete the cold start completion of the full mapping records of the new fault. The "empty placeholder emergency control action identifier" refers to a preset placeholder temporarily filled in the emergency control action identifier field of the many-to-many structured mapping table when a new fault has not been handled on-site and there is no valid historical emergency control action mapping relationship for the new fault in the many-to-many structured mapping table. Valid historical emergency control actions refer to emergency control actions that are not temporarily filled empty placeholders and can be directly issued, stored in the many-to-many structured mapping table. The range of values ​​for the similarity threshold is as follows: ; S503: When there is a similarity greater than or equal to a preset similarity threshold, the fault identifier bound to the fault feature fingerprint corresponding to the maximum similarity is used as the real-time fault identifier of the power equipment. Specifically, when there are multiple identical maximum similarities, the cumulative number of fault samples bound to the fault feature fingerprint corresponding to each maximum similarity is stored in set G. If there is a unique maximum value in set G, the fault identifier bound to the fault feature fingerprint corresponding to the maximum value stored in set G is used as the real-time fault identifier of the power equipment. If there are multiple identical maximum values ​​in set G, the historical update frequency of the mapping record bound to the fault feature fingerprint corresponding to each maximum value in set G is obtained, and the fault identifier bound to the fault feature fingerprint corresponding to the maximum update frequency is used as the real-time fault identifier of the power equipment. All emergency control action identifiers bound to real-time fault identifiers are retrieved from the many-to-many structured mapping table. An emergency control action sequence is generated according to the order of priority from highest to lowest for each emergency control action identifier. This sequence is then sent to the edge controller of the power equipment. The edge controller performs corresponding fault maintenance and linkage control processing on the power equipment in real time based on the sent content. The cumulative number of fault samples corresponding to the fault feature fingerprints bound to the finally determined real-time fault identifiers in the many-to-many structured mapping table is incremented by 1. The priority of emergency control action identifiers is determined by the total number of fault samples in the fault feature fingerprints bound to the emergency control action. The larger the total number of fault samples, the higher the corresponding priority. When the total number of fault samples in the fault feature fingerprints bound to multiple emergency control action identifiers is exactly equal, the shorter the historical single execution time of the emergency control action, the higher the corresponding priority.

[0022] A power data intelligent monitoring system based on multivariate data analysis includes a mapping table construction module, a fault sample analysis module, a cold start completion processing module, a similarity analysis module, and an operation and maintenance and linkage control module. The mapping table construction module is used to build a many-to-many structured mapping table between fault types and emergency control actions; The fault sample analysis module is used to analyze fault samples corresponding to various emergency control actions; The cold start completion processing module is used to perform cold start completion processing on the many-to-many structured mapping table between fault types and emergency control actions; The similarity analysis module is used to determine the fault feature fingerprints of power equipment in each monitoring interval and to perform similarity analysis on the determined fault feature fingerprints. The operation and maintenance and linkage control module is used to perform corresponding fault maintenance and linkage control processing on power equipment in real time through emergency control action sequence.

[0023] Example 1: Suppose the power equipment is a 10kV intelligent ring main unit, and the fault that occurs in the 10kV intelligent ring main unit is a phase-to-phase short circuit fault. Then the fault type corresponding to the 10kV intelligent ring main unit is a phase-to-phase short circuit fault. Let subset C be the subset of emergency control events that are bound and matched with phase-to-phase short-circuit faults. Subset C is composed of all emergency control events triggered by phase-to-phase short-circuit faults. Extract the full-link emergency control action sequence corresponding to each emergency control event in subset C. The full-link emergency control action sequence is as follows: After the fault triggers the protection action, the protection device determines that it is a phase-to-phase short-circuit fault. The system automatically issues command B_001 to disconnect the circuit breakers on both sides of the faulty ring network cabinet. After the system verifies the fault isolation status, it issues command B_002 to put the standby ring network cabinet into operation to transfer the load to the non-faulty area. After the standby ring network cabinet completes the load transfer and the non-faulty area restores normal power supply, command B_003 is issued to block all reclosing devices in the faulty area to prevent secondary impact caused by fault re-connection. Finally, command B_004 is executed to upload a complete fault handling message with full time stamp to the distribution master station, and the fault handling is closed loop. Synchronously traverse the adjacent device linkage operation data in the external correlation data monitored within the full cycle time window corresponding to each emergency control event in the subset C of emergency control events, and extract the cross-device linkage emergency control action sequence corresponding to the fault as follows: After the B_001 instruction is executed, the B_005 action is triggered to block the overload protection of the adjacent 10kV intelligent distribution transformer terminal. Then, the B_006 action is triggered to trip the upstream incoming circuit breaker of the faulty ring network cabinet to complete the full link isolation on the power supply side. For a phase-to-phase short-circuit fault, let its corresponding unique fault identifier ID be A_001. All emergency control actions corresponding to this fault include: disconnecting the circuit breakers on both sides of the faulty ring main unit B_001, putting the standby ring main unit into the load transfer B_002, blocking the reclosing of the faulty area B_003, uploading the fault message to the distribution master station B_004, triggering the overload protection blocking of the adjacent 10kV intelligent distribution transformer terminal B_005, and tripping the upstream incoming circuit breaker of the faulty ring main unit B_006. In the many-to-many structured mapping table of fault types and emergency control actions, six independent mapping records are inserted one by one to complete the structured field binding of this type of fault with all corresponding emergency control actions. That is, fault identifier A_001 establishes a structured binding relationship with B_001, B_002, B_003, B_004, B_005, and B_006 respectively. The six independent mapping records are as follows: Mapping record 1: Fault identifier A_001 ↔ Emergency control action identifier B_001; Mapping record 2: Fault identifier A_001 ↔ Emergency control action identifier B_002; Mapping record 3: Fault identifier A_001 ↔ Emergency control action identifier B_003; Mapping record 4: Fault identifier A_001 ↔ Emergency control action identifier B_004; Mapping record 5: Fault identifier A_001 ↔ Cross-device linkage emergency control action identifier B_005; Mapping record 6: Fault identifier A_001 ↔ Cross-device linkage emergency control action identifier B_006; That is, the smallest storage unit of a many-to-many structured mapping table is "fault identifier ↔ emergency control action identifier".

[0024] Example 2: Let the set of historical basic operational data segments retrieved be historical basic operational data segment set j, and the timestamps of indicators numbered 1, 2, and 3 in historical basic operational data segment set j for the first time exceeding the normal operation benchmark threshold be respectively... , , and , , All within the leading-edge sliding time window Within, the leading-edge sliding time window corresponding to the historical basic operational data segment set j is... The anchor point is ; The relative over-limit time series ratios for indicators numbered 1, 2, and 3 are respectively: ; ; .

[0025] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent monitoring of power data based on multivariate data analysis, characterized in that: The method includes: S10: Based on historical dispatch and operation data of power equipment, construct a many-to-many structured mapping table between fault types and emergency control actions; S20: Based on the leading-edge sliding time window of the action time window corresponding to various emergency control actions, retrieve the set of all historical basic operation data segments corresponding to the power equipment to determine the fault samples corresponding to various emergency control actions. S30: Based on the temporal evolution and index coupling of fault samples, analyze the fault feature fingerprints corresponding to each fault sample, and perform cold start completion processing on the many-to-many structured mapping table between fault types and emergency control actions. S40: Perform similarity analysis on the fault feature fingerprints of the time series of basic operating data monitored within each monitoring interval; S50: Construct emergency control action sequence and use it to perform corresponding fault maintenance and linkage control processing on power equipment in real time. 2.The power data intelligent monitoring method based on multi-element data analysis according to claim 1, characterized in that: S10 includes: S101: Obtain historical dispatch operation data of power equipment. The dispatch operation data includes basic operation data of power equipment, dispatch control data, fault event ledger data and external related data. Align the dispatch control data and fault event ledger data according to time sequence to determine emergency control events of power equipment under different states. The emergency control event refers to the complete time sequence record of the automatic dispatch system or dispatch personnel issuing dispatch control instructions, instruction transmission and equipment response within the entire cycle from the moment the power equipment fails to the completion of the fault handling closed loop. S102: Based on the fault type of power equipment, classify the determined emergency control events to obtain several subsets of emergency control events, and bind and match each fault type with the corresponding subset of emergency control events one by one. S103: Traverse the subset of emergency control events that match each fault type, extract the full-link emergency control action sequence corresponding to each emergency control event in the subset of emergency control events, select a fault type in turn, and synchronously traverse the adjacent device linkage operation data in the external related data monitored within the full cycle time window corresponding to each emergency control event in the subset of emergency control events that match the currently selected fault type, and extract the cross-device linkage emergency control action sequence. The full-link emergency control action sequence and cross-device linkage emergency control action sequence corresponding to the selected fault types are collected to obtain the target action sequence. The target action sequence is then processed, and a many-to-many structured mapping table between fault types and emergency control actions is constructed based on the processing results. 3.The power data intelligent monitoring method based on multivariate data analysis of claim 2, wherein: The specific method for processing the target action sequence is as follows: Repeated emergency control actions in the target action sequence are deduplicated. The ratio between the total number of historical executions of each emergency control action in the deduplicated target action sequence and the total number of emergency control events stored in the selected emergency control event subset matching the fault type is calculated. Emergency control actions with a ratio lower than a set threshold are deleted. Then, the fault identifier corresponding to the selected fault type is associated with all emergency control action identifiers stored in the target action sequence after deduplication and deletion in a structured field representation. All classified fault types are traversed to construct a many-to-many structured mapping table between fault types and emergency control actions.

4. The method of claim 3, wherein the method is characterized by: S20 includes: S201: In the many-to-many structured mapping table between fault types and emergency control actions, traverse all emergency control actions, and based on the action time window set corresponding to each emergency control action, retrieve the set of historical basic operation data segments stored in the leading-edge sliding time window of each action time window set to obtain several sets of historical basic operation data segments. S202: Bind the emergency control action to the set of all historical basic operation data segments obtained based on the emergency control action, and mark the sample set composed of all the bound historical basic operation data segments as the fault sample of the emergency control action. The basic operation data includes current harmonic distortion rate, voltage drop slope and partial discharge pulse count, wherein the current harmonic distortion rate is the core main indicator, and the voltage drop slope and partial discharge pulse count are auxiliary indicators.

5. The method of claim 4, wherein the method is based on multi-variate data analysis. The action time window refers to the continuous time interval from the trigger time of the emergency control action to the end time of the execution of the emergency control action. The leading-edge sliding time window refers to the continuous time interval from the timestamp corresponding to the latent feature start point to the trigger time of the emergency control action. The latent feature start point refers to the timestamp in the basic operation data of power equipment where a single core main indicator first exceeds the normal operation benchmark threshold for three consecutive sampling points or two auxiliary indicators simultaneously exceed the normal operation benchmark threshold for three consecutive sampling points for the first time.

6. The method of claim 5, wherein the method further comprises: S30 includes: S301: Normalize the fault sample, retrieve the set of historical basic operating data segments after normalization in each fault sample, obtain the timestamp of the first time that each indicator in each retrieved set of historical basic operating data segments exceeds the normal operating benchmark threshold, take the latent feature starting point of the leading-edge sliding time window corresponding to each retrieved set of historical basic operating data segments as the anchor point, calculate the relative over-limit time series ratio of each indicator in each set of historical basic operating data segments, and obtain the time series evolution morphology sub-matrix of the fault sample. The relative over-limit time series ratio is used to represent the ratio of the difference between the obtained timestamp and the timestamp corresponding to the anchor point to the time length corresponding to the leading-edge sliding time window. S302: Divide each set of normalized historical basic operation data segments according to the index type to obtain three time series of historical basic operation data with different indicators. Calculate the Pearson correlation coefficient between any two time series of historical basic operation data in each set of historical basic operation data segments to obtain the index coupling strength submatrix of the fault sample. S303: Based on the feature dimension alignment rule, the temporal evolution morphology submatrix and the index coupling strength submatrix are horizontally concatenated in the same dimension to obtain the fault feature fingerprint of the fault sample. The fault feature fingerprint of the fault sample is matched with the emergency control action bound to the fault sample. Based on the matching result, cold start completion is performed on the many-to-many structured mapping table between fault type and emergency control action.

7. The method of claim 6, wherein the method further comprises: S40 includes: S401: The total duration corresponding to the sliding window is taken as the monitoring interval length of the power equipment power data. The time series of basic operation data monitored within the monitoring interval is taken as the sample to be analyzed. The fault feature fingerprint of the sample to be analyzed is calculated and used as the real-time fault feature fingerprint of the power equipment at the right end of the monitoring interval, denoted as fault feature fingerprint X. S402: Traverse all fault feature fingerprints stored in the many-to-many structured mapping table, and record each fault feature fingerprint as fault feature fingerprint Y in turn. Flatten fault feature fingerprint Y and fault feature fingerprint X into one-dimensional feature vectors in descending order of the weights assigned to each indicator. S403: Calculate two one-way reachable distances between the fault feature fingerprint Y and the fault feature fingerprint X, take the arithmetic mean of the two calculated one-way reachable distances, and denote the arithmetic mean as , and calculate the similarity between the fault feature fingerprint X and the fault feature fingerprint Y according to the formula .

8. The intelligent monitoring method for power data based on multivariate data analysis according to claim 7, characterized in that: The specific calculation method for the two unidirectional reachability distances between the fault feature fingerprint Y and the fault feature fingerprint X is as follows: Euclidean distance between fault feature fingerprint X and fault feature fingerprint Y Perform calculations and sort all the calculated Euclidean distances in ascending order of their numerical values. Sort the samples to obtain the distance sorting sequence 1 corresponding to the fault feature fingerprint X; Will The distance from fault feature fingerprint X to fault feature fingerprint Y is defined as the unidirectional reachability distance, where k represents a positive integer and is less than the total number of fault feature fingerprints that the many-to-many structured mapping table can store. This represents the Euclidean distance of the k-th position in sorted sequence 1; Iterate through all fault fingerprints in the many-to-many structured mapping table except for fault fingerprint Y, and calculate the Euclidean distance between fault fingerprint Y and each of the remaining fault fingerprints traversed. All calculated Euclidean distances are processed in ascending order of data. Sort the data to obtain the distance sorting sequence 2 corresponding to the fault feature fingerprint Y. The distance that can be traveled unidirectionally from fault feature fingerprint Y to fault feature fingerprint X. This represents the Euclidean distance at position k in sorted sequence 2.

9. The intelligent monitoring method for power data based on multivariate data analysis according to claim 8, characterized in that: The S50 includes: S501: Traverse all fault feature fingerprints stored in the many-to-many structured mapping table, and calculate the similarity between each fault feature fingerprint and the fault feature fingerprint X in each traversal. S502: When all similarities are less than the preset similarity threshold, the fault identifier corresponding to the fault feature fingerprint X is marked as a new fault identifier, and the mapping record consisting of the new fault identifier, the fault feature fingerprint corresponding to the new fault identifier, and the empty-position emergency control action identifier is automatically written into the many-to-many structured mapping table. S503: When there is a similarity greater than or equal to a preset similarity threshold, the fault identifier bound to the fault feature fingerprint corresponding to the maximum similarity will be used as the real-time fault identifier of the power equipment. All emergency control action identifiers bound to real-time fault identifiers are retrieved from the many-to-many structured mapping table. Emergency control action sequences are generated in descending order of priority of each emergency control action identifier. The generated emergency control action sequence is then sent to the edge controller of the power equipment. The edge controller performs corresponding fault maintenance and linkage control processing on the power equipment in real time according to the sent content.

10. A power data intelligent monitoring system based on multivariate data analysis, applied to any one of the power data intelligent monitoring methods based on multivariate data analysis according to claims 1-9, characterized in that: The system includes a mapping table construction module, a fault sample analysis module, a cold start completion processing module, a similarity analysis module, and an operation and maintenance and linkage control module. The mapping table construction module is used to construct a many-to-many structured mapping table between fault types and emergency control actions; The fault sample analysis module is used to analyze fault samples corresponding to various emergency control actions. The cold start completion processing module is used to perform cold start completion processing on the many-to-many structured mapping table between fault types and emergency control actions. The similarity analysis module is used to determine the fault feature fingerprints of power equipment in each monitoring interval and to perform similarity analysis on the determined fault feature fingerprints. The operation and maintenance and linkage control module is used to perform corresponding fault maintenance and linkage control processing on power equipment in real time through emergency control action sequence.