A substation monitoring information feature extraction method and device

Abstract

The application discloses a substation monitoring information feature extraction method and device, the method comprises the following steps: a data preprocessing step, including: processing the analog signal, extracting the correlation features of the analog data through digital signal processing; extracting the alarm information text keywords; extracting, cleaning and time synchronization preprocessing of the data; a data classification step, including: classifying and forming the state features according to the correlation features of the analog data and the text keywords of the alarm information; a time sequence feature extraction step, including: processing the state features obtained in the data classification step according to the time sequence relationship to form the synchronous state features with time sequence information. The application extracts the basic features, state features and time sequence features of the substation monitoring information, provides correct and dimension-reduced high-quality data for data analysis from the substation, and provides an accurate and efficient data preprocessing method for analyzing and diagnosing the operation state and equipment working state by using the substation monitoring information.

Description

Technical Field

[0001] This invention relates to a method and apparatus for extracting features from substation monitoring information, belonging to the field of intelligent substation technology in power systems. Background Technology

[0002] After a substation becomes unmanned, its operation is centrally monitored by the control center. In the event of a fault or anomaly, grid operators must promptly and accurately determine the current grid operating status and abnormal components based on the content and sequence of monitoring information, and issue adjustment or incident handling instructions to ensure the safe and stable operation of the grid. The control center receives a massive amount of monitoring information, but lacks the technical means to summarize, classify, extract, correlate, and synthesize it. Relying entirely on manual analysis and diagnosis when faced with this massive amount of information is limited by the operators' abilities, experience, and physical and mental state. This results in insufficient time for information filtering, an inability to quickly and accurately perceive the situation on-site, and a serious risk of signal omissions and misjudgments. This affects the efficiency of handling abnormal faults and may even cause the fault's impact to expand, which is detrimental to the safe and stable operation of the grid.

[0003] Computer-aided analysis methods can be used to diagnose the operating status of substation equipment. Data collected from multiple devices in a substation contains elements and characteristics reflecting the operating status of the power system. However, directly applying this data for analysis is problematic because redundant, repetitive, erroneous, or missing data in the field results in a large amount of useless, repetitive, or low-information-content data consuming resources, reducing the accuracy and efficiency of judging the system's operating status, and affecting the practicality of intelligent auxiliary analysis functions. A possible measure is to pre-analyze and process the multi-source data from the field before diagnosis, extracting data features and simplifying information resources to enable rapid and accurate computer processing. Extracting these features or factors as diagnostic criteria and analyzing the correlation between features and results can provide the causes of faults or anomalies and corresponding contingency plans, offering a reference for operators. Summary of the Invention

[0004] The purpose of this invention is to address the analysis challenges arising from the increasing volume and dimensions of substation monitoring information. This invention proposes a feature extraction method and apparatus for substation monitoring information, enabling the effective diagnosis of substation monitoring data and the extraction of basic, status, and temporal features of the substation monitoring information.

[0005] The present invention specifically adopts the following technical solution: a method for extracting features from substation monitoring information, comprising:

[0006] The data preprocessing steps include: processing analog signals and extracting correlation features of analog data through digital signal processing; extracting alarm information text keywords; and performing data extraction, cleaning, and time synchronization preprocessing.

[0007] The data classification steps include: classifying and forming status features based on the correlation characteristics of preprocessed analog data and the text keywords of alarm information;

[0008] The temporal feature extraction step includes: processing the state features obtained from the data classification step according to the temporal relationship to form synchronous state features with temporal information.

[0009] As a preferred embodiment, the data extraction step in the data preprocessing step specifically includes: analyzing the correlation between the data information and the analysis requirements based on the physical characteristics associated with the analyzed information, and determining the data information attribute range, wherein the data information attribute range is divided into: event occurrence range, information reporting device range, data information attribute range, and time range.

[0010] As a preferred embodiment, the event occurrence range includes: determining the physical system impact range of the event based on the primary topology, secondary circuits, and communication circuits of the physical system reflected by the information; the event occurrence range that requires data analysis is divided into the physical event range corresponding to protection actions and the physical event range corresponding to fault and abnormal alarms.

[0011] As a preferred embodiment, the time range is determined based on the timeliness of the information and the cycle of physical system events, specifically including:

[0012] The start time of the time range corresponds to the protection logic and settings. The start time of abnormal alarm event analysis is: abnormal alarm event action time - abnormal diagnosis time - abnormal alarm pre-analysis time; the abnormal alarm pre-analysis time is set according to the pre-event analysis requirements of the alarm event. The start time of protection action analysis is: protection action time - protection time setting - protection action pre-analysis time; the protection action pre-analysis time is set according to the pre-fault analysis requirements of the protection action.

[0013] The end time of the time range corresponds to the event return time; the end analysis time of the abnormal alarm event is set to the abnormal alarm event return time; the end analysis time of the protection action is set to: the protection event return time + the post-event analysis time, or set to the protection group reset time.

[0014] In a preferred embodiment, the data cleaning step in the data preprocessing process specifically includes:

[0015] Data validity identification is used to: confirm whether the device is working properly, whether the data provided by the device is valid, and whether the provided data is visible;

[0016] Data consistency identification is used to: identify data by utilizing the multi-source redundancy characteristics of data;

[0017] Data redundancy processing is used to: remove redundant items from data information and reduce the dimensionality of transaction items;

[0018] Data noise processing is used to identify inaccurate data and perform noise reduction.

[0019] Data loss handling is used to: when equipment or communication loop malfunctions cause information loss due to channel failure, and to fill in missing or erroneous data for short-term missing data based on data consistency relationships, primary topology relationships, or secondary loop relationships; for long-term missing or erroneous data, it is determined that the data cannot participate in data analysis until the data is recovered or the abnormal alarm status that caused the data abnormality returns to normal.

[0020] In a preferred embodiment, the time synchronization preprocessing step of the data preprocessing step uses monitoring data information for time synchronization processing to distinguish the order of data information. Specifically, it includes: sorting data information according to the order of time tags; considering the data information acquisition response time index and time resolution index in the data analysis time range; and performing time synchronization processing on the time tags of data and monitoring information based on the actual occurrence time of physical events.

[0021] As a preferred embodiment, the data classification step specifically includes: classifying the data information according to the relevance of the mining content, mapping the data in the database to a certain category; such as distinguishing the cause of data changes and information generation as accidents, equipment failures, equipment anomalies or operation control, or classifying them according to the source of information generation as abnormal system operation state, primary equipment, secondary equipment or secondary circuit, communication equipment or circuit; and constructing initial feature classification rules based on the classification of data and information.

[0022] As a preferred embodiment, the temporal feature extraction step includes:

[0023] Data segments are extracted based on data status, specifically including: based on the actual monitored analog data, the extracted data segments can be divided into normal status data, out-of-limit status data, and fault status data;

[0024] Sort data segments in chronological order, specifically including: arranging data states according to the switching order of data in different states; and using time series combination methods to form a single data state sequence in chronological order.

[0025] Combining multivariate data to construct a multivariate data time-series state matrix specifically includes: dynamically expanding and combining multivariate data according to time label order to form a time-series state matrix M(m×n) of multivariate data; where n is the data dimension, m is the time dimension, and m is the matrix data element.ij For the corresponding data value j at time i; sort the multivariate data by time sequence, extract the time label sequence of the multivariate data as the time dimension scale of the multivariate data time-series state matrix, extract the state value of each data at each time scale, and fill the data value into the multivariate data time-series state matrix.

[0026] A state switching matrix is ​​constructed based on the obtained multivariate data time-series state matrix.

[0027] As a preferred embodiment, the method for constructing the state transition matrix is ​​as follows:

[0028] Based on the multivariate data time-series state matrix, a multivariate data state switching matrix C(l×n) can be constructed, where n is the data dimension, consistent with the multivariate data time-series state matrix, and l is the time dimension, with l = m-1, c ij Reflecting the changes in data state between adjacent time points; defined as follows:

[0029]

[0030] For Boolean variables, the elements of the state transition matrix are the algebraic differences between the Boolean values ​​at the current time and the Boolean values ​​at the previous time. The analog data processing method is as follows: if the difference between the change in the modulus at the current time and the change in the modulus at the previous time is greater than the change threshold value S of data j... j For a positive transition, the corresponding element in the state transition matrix is ​​1; for a negative transition, the corresponding element is -1; and for a transition less than the threshold value, the corresponding element is 0.

[0031] After the above transformation, the state switching matrix is ​​completely changed to Boolean elements. Data change information is extracted based on whether the matrix element is 0.

[0032] This invention also proposes a feature extraction device for substation monitoring information, comprising:

[0033] The data preprocessing module is used to perform the following: process analog signals, extract correlation features of analog data through digital signal processing; extract alarm information text keywords; and perform data extraction, cleaning, and time synchronization preprocessing.

[0034] The data classification module is used to perform the following: classify and form status features based on the correlation characteristics of preprocessed analog data and the text keywords of alarm information;

[0035] The temporal feature extraction module is used to perform the following: process the state features obtained from the data classification step according to the temporal relationship to form synchronous state features with temporal information.

[0036] The beneficial effects achieved by this invention are as follows: First, addressing the analysis challenges arising from the ever-increasing volume and dimensionality of substation monitoring information, this invention proposes a method and apparatus for feature extraction from substation monitoring information. This method performs validity diagnosis on substation monitoring data, extracting fundamental, status, and temporal features of the substation monitoring information. It provides accurate, dimensionality-reduced, high-quality data for data analysis originating from substations, and offers an accurate and efficient data preprocessing method for analyzing and diagnosing operational and equipment working conditions using substation monitoring information. Second, the data preprocessing method proposed in this invention reduces the burden of data analysis, providing operators with the necessary steps for analyzing and diagnosing results in a timely manner. Attached Figure Description

[0037] Figure 1 This is a flowchart of a method for extracting features from substation monitoring information according to the present invention. Detailed Implementation

[0038] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0039] Example 1: Taking the data analysis of transformer body faults in a 110kV internal bridge connection substation as an example, the substation monitoring information feature extraction method of the present invention is as follows.

[0040] (1) Data Information Source

[0041] The data and information elements used for data analysis in substations include substation models and graphic files, protection functions, protection logic principles, setting parameters, monitoring data, alarm information, etc.

[0042] The sources of data information used for data analysis are as follows:

[0043] 1) Based on the primary equipment topology within the station, protection devices whose operating range overlaps with that of the transformer protection can be extracted. In this example, the transformer protection overlaps with the operating ranges of the 110kV line protection, 110kV busbar protection, and 10kV sectional protection.

[0044] 2) Bay Wiring Diagram: Obtain the current transformers corresponding to the voltage and current signals of the protection equipment, as well as the circuit breakers controlled by the protection equipment. In this example, the voltage and current signals of the transformer differential protection and backup protection are collected from different current transformers at the same acquisition point. Simultaneously, the 110kV bus voltage transformer signal is collected from the same acquisition point as the 110kV line protection, and the 10kV bus voltage transformer signal is collected from the same acquisition point as the 10kV capacitor, 10kV feeder, and 10kV station service transformer protection. The circuit breakers controlled by the transformer protection also overlap with the aforementioned protections.

[0045] 3) In this example, the transformer protection, 110kV line protection, and 110kV bus protection share the same communication transmission path between the monitoring system and the remote control center.

[0046] 4) Protection Model: Based on the transformer protection instruction manual, extract a list of items such as equipment functions, setting parameters, analog inputs, digital inputs, digital outputs, communication network configuration, monitoring data, and alarm information; protection settings can be used to clarify the protection range of the transformer protection.

[0047] 5) Real-time monitoring: Extract monitoring data and alarm information of all protection devices that intersect or overlap with the transformer protection operating range, signal acquisition, and communication path in real time.

[0048] (2) Data preprocessing

[0049] 1) Data extraction

[0050] (a) Scope of the event

[0051] Based on the analysis requirements, the event range of transformer body failure can be defined according to the main wiring diagram, which is an internal transformer failure.

[0052] (b) Scope of information source equipment

[0053] Based on the fact that the protected object is a transformer, it can be determined that abnormal operating events involve the transformer protection and control devices. The range of abnormal events in the secondary circuit is determined based on the transformer bay wiring diagram. The range of abnormal events in the communication circuit is determined based on the communication network structure diagram of the transformer bay. The range of abnormal operation of secondary equipment is determined to be the transformer protection equipment itself. The secondary equipment affected by transformer body faults includes all local protection devices that are sensitive to transformer body faults, as shown in the table.

[0054] Table 1. Equipment Scope and Sources of Transformer Body Fault Monitoring Information

[0055]

[0056]

[0057] (c) Scope of data information attributes

[0058] Analog data for transformer body faults includes AC analog quantities such as voltage and current, or calculated analog quantities such as differential current and harmonic current. Data attributes may include time, RMS value, and phase. Attributes for oil temperature monitoring data may include time and actual value. Attributes for alarm information corresponding to protection actions and abnormal alarms may include time, device, alarm level, and alarm content.

[0059] (d) Time range

[0060] In this example, the pre-analysis time for abnormal alarms is set to 1 second, and the start time for the analysis of abnormal alarm events is:

[0061] Anomaly alarm event action time - anomaly diagnosis time - 1s

[0062] In this example, the protection action pre-analysis time is consistent with the fault recording pre-recording time, such as 100ms. The protection action start analysis time is:

[0063] Protection action timing - Protection time setpoint - 100ms

[0064] The end analysis time for abnormal alarm events is set to the return time of the abnormal alarm event, and the end analysis time for protection actions is set to the reset time of the entire protection group.

[0065] 2) Data cleaning

[0066] Data cleaning is mainly used for processing analog data from monitoring.

[0067] (a) Data validity identification:

[0068] For devices that malfunction or have communication problems, and cannot provide valid data, the data can be considered erroneous and will not be included in the data analysis.

[0069] For the analysis of data and monitoring information related to the example transformer protection, the equipment and monitoring information that need to be verified are shown in the table. Only when the equipment is operating normally and communication is normal can the data information released by the equipment be confirmed as valid.

[0070] Table 2. Requirements for the Effectiveness of Transformer Body Fault Monitoring Information

[0071]

[0072] (b) Data consistency identification:

[0073] Based on the example 110kV internal bridge main wiring primary topology, the following data relationships can be determined for the transformer protection analog quantities:

[0074] a) Transformer high-voltage side voltage = 110kV bus voltage

[0075] b) Transformer high-voltage side current = 110kV incoming line current ± 110kV bridge side current

[0076] c) Transformer low-voltage side voltage = 10kV bus voltage

[0077] d) Transformer low-voltage side current = 10kV feeder + 10kV capacitor + 10kV station service transformer ± 10kV segment current

[0078] Based on the data from the secondary signal acquisition points, the following data relationships can be determined:

[0079] The same analog signal acquired at the same acquisition point should be consistent, including the current and voltage on each side.

[0080] Based on the working principle of a transformer, the following data relationships exist under normal operating conditions:

[0081] a) The effective values ​​of the voltages on each side of the transformer are in a fixed proportional relationship; the phase relationship between the voltages on each side is fixed.

[0082] b) The effective values ​​of the currents on each side of the transformer are in a fixed proportional relationship; the phase relationship between the currents on each side is fixed.

[0083] Based on the above relationships, if any data items in the device are inconsistent with the above relationships, the data can be marked as data errors, as shown in the table.

[0084] Table 3. Results of Inconsistent Fault Data Analysis of Transformer Body

[0085]

[0086] Inconsistent data detected can be compared with equipment monitoring information to diagnose whether the equipment's monitoring of the working and communication status, as well as its monitoring of the secondary circuits, is accurate and timely.

[0087] (c) Data redundancy processing: Based on the response time of alarm information, the debouncing time in the example can be selected as 5s.

[0088] (d) Data noise processing: Examples include averaging the effective values, phase values, etc., to smooth the data.

[0089] (e) Data missing handling: In the example, missing data can be filled in based on data consistency. For data that cannot be filled in using consistent data, such as missing data at a specific time, linear interpolation or quadratic interpolation algorithms can be used to calculate the data at that time.

[0090] 3) Time synchronization preprocessing

[0091] The time resolution of alarm information for transformer protection is ≤2ms. When comparing alarm information, alarm information from different devices with a time interval of less than 2ms can be regarded as occurring at the same time.

[0092] The response time of analog data in the station monitoring system (from I / O input to data communication gateway output) is ≤2s. Therefore, when comparing analog data and alarm information, the time when the analog data corresponds to the alarm information at a certain moment should follow the following rules:

[0093] Analog data release time - 2s ≈ Alarm information release time - Alarm information detection time - 2ms.

[0094] (3) Feature extraction

[0095] Analog data feature identification is mainly used for data features that cannot be determined or are insufficient to constitute an alarm event by a single device. For example, the basic analog data features of transformer protection under abnormal or fault conditions may include: analog quantity abrupt changes, analog quantity exceeding limits, etc.

[0096] Feature extraction is performed to meet the needs of event analysis, such as analyzing a transformer failure and its correct isolation. The monitored data and alarm information are preprocessed to obtain basic features of the substation data and information during this process. Examples are shown in the table below:

[0097] Table 4 Basic Characteristics of Transformer Body Fault Data

[0098]

[0099]

[0100] The data characteristics corresponding to transformer body faults can be classified into: analog data characteristics, protection action information, protection intermediate information, etc. The table encodes the above basic characteristics of transformer body faults by serial number and organizes them according to the logical relationship of the basic characteristics and the order of appearance and return. It is assumed that the feature is true as 1 and the feature returns as 0. For example, 12(1) means that the feature with serial number 12 is true and 12(0) means that the feature with serial number 12 returns. The multi-data time-series state matrix of transformer body faults can be initially constructed, and the state switching matrix is ​​formed on this basis.

[0101] Table 5. Timing State Matrix of Multivariate Data for Transformer Body Faults

[0102]

[0103]

[0104] Table 6 Transformer Body Fault Status Switching Matrix

[0105]

[0106] By applying the method described in this invention, the basic features, status features, and time-series features of the substation monitoring information can be extracted in a structured representation.

[0107] Example 2: The present invention also proposes a feature extraction device for substation monitoring information, comprising:

[0108] The data preprocessing module is used to perform the following: process analog signals, extract correlation features of analog data through digital signal processing; extract alarm information text keywords; and perform data extraction, cleaning, and time synchronization preprocessing.

[0109] The data classification module is used to perform the following: classify and form status features based on the correlation characteristics of preprocessed analog data and the text keywords of alarm information;

[0110] The temporal feature extraction module is used to perform the following: process the state features obtained from the data classification step according to the temporal relationship to form synchronous state features with temporal information.

[0111] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0112] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method of feature extraction of substation monitoring information, characterized by, The method comprises the following steps: a data preprocessing step, comprising: processing analog signals, extracting associated features of analog data through digital signal processing, extracting text keywords of alarm information, and extracting, cleaning and time synchronizing the data; a data classification step, comprising: classifying and forming state features according to the associated features of the preprocessed analog data and the text keywords of the alarm information; a time sequence feature extraction step, comprising: processing the state features obtained in the data classification step according to the time sequence relationship to form synchronized state features with time sequence information; the data extraction in the data preprocessing step specifically comprises: analyzing the relevance of data information and analysis requirements according to the physical characteristics associated with the analyzed information, determining the data information attribute range, and sequentially dividing the data information attribute range into: event occurrence range, information reporting device range, data information attribute range and time range; the time sequence feature extraction step comprises: extracting data segments according to data states, specifically comprising: extracting data segments according to actual monitored analog data, and dividing the extracted data segments into normal state data, out-of-limit state data and fault state data; sorting data segments in chronological order, specifically comprising: arranging data states according to the switching sequence of data in different states; and forming a single data state sequence in chronological order by using a time sequence combination method; The combination multi-element data constructs a multi-element data time sequence state matrix, specifically including: the multi-element data is dynamically expanded and combined according to the time label order to form a multi-element data time sequence state matrix M(m n); wherein n is the data dimension, m is the time dimension, and the matrix data element m ij is the j data value at the i moment; the multi-element data is sorted according to the time sequence, the time label sequence of the multi-element data is extracted as the time dimension scale of the multi-element data time sequence state matrix, and the state value of each data at each time scale is extracted and filled into the data value of the multi-element data time sequence state matrix. constructing a state switching matrix based on the obtained multivariate data time sequence state matrix; the construction method of the state switching matrix is: According to the multi-dimensional data time sequence state matrix, a multi-dimensional data state switching matrix C(l n) is constructed, n is the data dimension, consistent with the multi-dimensional data time sequence state matrix, l is the time dimension, l=m-1, c ij reflects the change of data state at adjacent time; defined as follows: ； For the Boolean variable, the state switching matrix element is the algebraic difference value between the current time Boolean value and the previous time Boolean value; the analog quantity data processing method is: if the change difference value between the current time modulus value and the previous time modulus value is greater than the change threshold value S of j data j , the positive jump is 1; the negative jump is-1; less than the threshold value, the corresponding element is 0; After the above conversion, the state switching matrix is all changed into Boolean elements, and the data change information is extracted according to whether the matrix elements are 0.

2. The method of claim 1, wherein, The event occurrence range includes: determining the physical system influence range of the event according to the primary topology, secondary loop and communication loop of the physical system reflected by the information; and the event occurrence range requiring data analysis is divided into a physical event range corresponding to a protection action and a physical event range corresponding to a fault abnormal alarm.

3. The method of claim 1, wherein, The time range is determined according to the time limit of the information and the cycle of the physical system event, and specifically comprises: The starting time of the time range corresponds to the protection logic and the setting value, the starting analysis time of the abnormal alarm event is: the action time of the abnormal alarm event - the abnormal diagnosis time - the abnormal alarm pre-analysis time, the abnormal alarm pre-analysis time is set according to the pre-event analysis requirement of the alarm event, and the starting analysis time of the protection action is: the protection action time - the protection time setting value - the protection action pre-analysis time, the protection action pre-analysis time is set according to the pre-fault analysis requirement of the protection action; The ending time of the time range corresponds to the event return time; the ending analysis time of the abnormal alarm event is set as the return time of the abnormal alarm event; and the ending analysis time of the protection action is set as: the return time of the protection event + the post-event analysis time, or as the protection whole group reset time.

4. The method of claim 1, wherein, The data cleaning in the data preprocessing step specifically comprises: data validity identification, which is used to confirm whether the device is working normally, whether the data provided by the device is valid, and whether the provided data is visible; data consistency identification, which is used to identify data by using the multi-source redundancy features of the data. Data redundancy processing is used to eliminate redundant items in data information and reduce dimension of transaction items. Data noise processing is used to identify inaccurate data and perform noise reduction processing. Data missing processing is used to fill in missing or erroneous data according to consistency relationship of data, relationship between data, primary topology relationship or secondary loop relationship when channel failure causes information loss due to device abnormality or communication loop abnormality. Long-time missing or erroneous data cannot participate in data analysis until data recovery or abnormal alarm state recovery.

5. The method of claim 1, wherein, The time synchronization preprocessing of data in the data preprocessing step adopts monitoring data information to distinguish the sequence of data information, specifically including: data information is sorted according to the sequence of time tags; data analysis time range considers data information acquisition response time index and time resolution index; the time tag of data and monitoring information is processed according to the actual occurrence time of physical events.

6. The substation monitoring information feature extraction method of claim 1, wherein, The data classification step specifically includes: classifying data information according to relevance of mining content, mapping data in the database to a certain category in a given category; distinguishing the cause of data change and information generation as an accident, device failure, device abnormality or operation control, or distinguishing the cause of data change and information generation as an abnormal system operation state, primary device, secondary device or secondary loop, communication device or loop; constructing initial feature classification rules according to the classification of data and information.

7. A substation monitoring information feature extraction device characterized by comprising: It includes: A data preprocessing module is used to process analog signals and extract correlation features of analog data through digital signal processing. Extracting alarm information text keywords; extracting, cleaning and time synchronization preprocessing of data; A data classification module is used to classify and form state features according to the correlation features of preprocessed analog data and the text keywords of alarm information. A time sequence feature extraction module is used to process state features obtained by the data classification step according to time sequence relationship to form synchronous state features with time sequence information. In the data preprocessing module, data extraction specifically includes: analyzing the relevance of data information and analysis requirements according to the physical properties associated with the analyzed information, determining the data information attribute range, and the data information attribute range is divided into: event occurrence range, information reporting device range, data information attribute range, and time range. The time sequence feature extraction module includes: Data segments are extracted according to data state, specifically including: according to the actual monitored analog data, the extracted data segments are divided into normal state data, out-of-limit state data and fault state data. Data segments are sorted in time sequence, specifically including: arranging data state according to the switching sequence of data in different states; using time sequence combination method to form single data state sequence in time sequence; The combination multi-element data constructs a multi-element data time sequence state matrix, specifically including: the multi-element data is dynamically expanded and combined according to the time label order to form a multi-element data time sequence state matrix M(m n); wherein n is the data dimension, m is the time dimension, and the matrix data element m ij is the j data value at the i moment; the multi-element data is sorted according to the time sequence, the time label sequence of the multi-element data is extracted as the time dimension scale of the multi-element data time sequence state matrix, and the state value of each data at each time scale is extracted and filled into the data value of the multi-element data time sequence state matrix. A state switching matrix is constructed based on the obtained multivariate data time sequence state matrix. The construction method of the state switching matrix is: According to the multi-dimensional data time sequence state matrix, a multi-dimensional data state switching matrix C(l n) is constructed, n is the data dimension, consistent with the multi-dimensional data time sequence state matrix, l is the time dimension, l=m-1, c ij The change of data state at adjacent time is embodied; the definition is as follows: ； For the Boolean variable, the state switching matrix element is the algebraic difference value between the current time Boolean value and the previous time Boolean value; the analog quantity data processing method is: if the change difference value between the current time modulus value and the previous time modulus value is greater than the change threshold value S of j data j , the positive jump is 1, the negative jump is-1, and the value less than the threshold value is 0; After the above conversion, the state switching matrix is all changed into Boolean elements, and data change information is extracted according to whether the matrix elements are 0.

Images