Abnormal data processing method and system based on control and allocation data fusion

Abstract

This invention provides an anomaly data processing method based on the fusion of control and distribution data, comprising: acquiring system models and operational data of the control cloud platform and the integrated operation and distribution platform; determining the correlation between the system models of the control cloud platform and the integrated operation and distribution platform based on pre-trained model objects; and performing data feature extraction and anomaly data processing on the operational data based on the correlation between the system models and the pre-trained neural network model. This method addresses the shortcomings of insufficient cross-professional information sharing in traditional methods and improves the quality of data governance by employing power grid data feature extraction and anomaly processing technologies.

Description

Technical Field

[0001] This invention belongs to the field of power system simulation, specifically relating to an abnormal data processing method and system based on the fusion of control and distribution data. Background Technology

[0002] While integrated automation systems for control and dispatch are widely used, data sharing and information fusion have not yet been achieved between various business systems such as control, production, marketing, and operation and maintenance. Data within each business system is isolated from other systems, with the same basic business data being repeatedly created in different systems and exhibiting significant discrepancies. This seriously hinders the development of the business itself and the application of the system. The marketing department is committed to the integration of marketing and distribution, enabling a high degree of data sharing between the marketing and production business systems. The control department is committed to the integration of control and dispatch, achieving information sharing in scheduling and monitoring. However, the data fusion capabilities of the entire business on both the control and dispatch sides remain insufficient. In practice, problems such as redundant maintenance and poor consistency, known as "information silos," persist, leading to difficulties in interaction and low data utilization over time. Therefore, the realism of simulation effects in large-scale joint anti-accident drills needs to be improved, urgently requiring simulation systems to achieve model and data fusion.

[0003] The construction of the ubiquitous power Internet of Things provides more data support for power grid dispatching and operation management. It requires the comprehensive utilization of various information resources and analysis methods, such as power generation, transmission, substation, load, fault, operation, and external environment. As the power grid's situational awareness capabilities continue to improve, the scenarios for simulation training are becoming increasingly rich, and the breadth, density, and accuracy of simulation data are constantly being improved and expanded. The information integration in the existing multi-level control simulation system is insufficient, and the ability of various units to collaboratively handle accidents is inadequate. It is necessary to build a targeted multi-level control joint training simulation system that is highly consistent with the power grid, covering the entire scope, process, and scenario, to conduct large-scale power grid operation analysis and evaluation, and to improve the ability of training control personnel to manage the large power grid for auxiliary decision-making and accident handling. Summary of the Invention

[0004] To address the current lack of data sharing and information integration among various business systems such as regulation, production, and marketing, where data within each system is isolated and often duplicated with significant discrepancies, severely hindering business development and system application, this invention provides an abnormal data processing method based on regulation and control data fusion, comprising:

[0005] Acquire system models and operational data from the control cloud platform and the integrated operation and distribution platform;

[0006] The correlation between the system models of the control cloud platform and the integrated operation and dispatch platform is determined based on the pre-trained model objects.

[0007] Based on the correlation of the system model and the pre-trained neural network model, data feature extraction and abnormal data processing are performed on the running data.

[0008] Preferably, the acquisition of the system model and operational data of the control cloud platform includes:

[0009] Obtain system models for container, device, topology, and external environment classes related to the control cloud platform;

[0010] Obtain the container, system, device, and topology system models for the integrated operation, distribution, and dispatch platform;

[0011] The operational data of the control cloud platform is obtained from the operational data center of the control cloud platform through the data acquisition service;

[0012] The operational data of the integrated operation, distribution, and dispatch platform is obtained through the data acquisition service on the distribution side.

[0013] Preferably, the training of the model object includes:

[0014] The historical control cloud platform data and historical operation and allocation integrated platform data input into the model object are processed through convolutional layers, pooling layers, and fully connected layers to extract keywords and perform periodic scanning, resulting in keyword matching rules, operational data, and other reference data arranged by name, description, and code.

[0015] The other comparative data include data with consistency issues, correlation issues, and model matching issues.

[0016] Preferably, the step of determining the association between the system models of the control cloud platform and the integrated operation and dispatch platform based on pre-trained model objects includes:

[0017] The consistency problem is eliminated by using the rule base of the big data platform, and the correlation problem and model matching problem are fed back to the big data platform.

[0018] The running data is uniformly encoded with a preset field width for ID.

[0019] Preferably, the step of extracting data features and processing abnormal data from the operational data based on the correlation relationships of the system model and the pre-trained neural network model includes:

[0020] Based on the correlation of the system model and the pre-trained neural network model, the system performs data feature extraction, routine data anomaly processing, accident data preprocessing, and defect data preprocessing on the operational data.

[0021] Preferably, data feature extraction is performed on the operational data, including:

[0022] Useless status indicator variables in the operational data are deleted through a big data platform.

[0023] Preferably, the operational data undergoes routine data anomaly processing, including:

[0024] Data conversion errors, data range errors, data anomalies, bus imbalance, line measurement imbalance, winding imbalance on both or three sides, and data lacking calculation components in the total index are identified through data generalization, normalization, or manual correction.

[0025] The data with conversion errors, data range errors, and data anomalies are eliminated by mathematical statistical algorithms, and then the rules are validated.

[0026] The state estimation algorithm is used to correct the data on bus imbalance, line measurement imbalance, and imbalance on both or three sides of the winding.

[0027] The data that lacks a calculation component in the total index will be fed back to the control cloud platform and the integrated operation and distribution platform for modification.

[0028] The data anomalies include: missing data and noise jumps.

[0029] Preferably, the operational data undergoes accident data preprocessing, including:

[0030] Accident data in the operational data is identified through data generalization, normalization, or manual correction, and the accident data is divided into critical fault data, serious fault data, general fault data, and notification fault data.

[0031] The critical fault data, serious fault data, general fault data, and notification fault data are segmented and words are removed using natural language processing methods to obtain vectorized fault data.

[0032] The vectorized fault data is tagged based on the tag samples of pre-acquired historical monitoring and alarm information.

[0033] Using a hierarchical clustering algorithm, the fault data after the labeling process is divided into line fault data, transformer fault data, feeder section fault data, microgrid fault data, and cascading fault data.

[0034] The fault data of the line, transformer, feeder section, microgrid, and cascading faults are matched with the faulty equipment in the power grid to obtain fault-related equipment data.

[0035] The fault-related equipment data is input into a pre-trained neural network model for learning, and classification results with different labels are obtained.

[0036] The label classification includes: fault type, fault severity, and fault location;

[0037] The neural network model is trained by taking labeled event samples extracted from historical monitoring and alarm information as input and the classification results of different labels as output.

[0038] Preferably, the step of inputting the fault-associated device data into a pre-trained neural network model for learning to obtain classification results with different labels includes:

[0039] The classification results are used as the basic probability allocation values ​​for fault analysis.

[0040] Preferably, the operational data undergoes defect data preprocessing, including:

[0041] Based on power grid fault-type operation data, alarm event-type operation data, equipment ledger data, and signal-related equipment data, defect identification is performed on the operation data. When the number of alarms occurring on the same type of equipment exceeds a preset value, it is identified as a family-related defect.

[0042] Based on the equipment model, production batch, alarm information, defect level, and defect cause of the same type of equipment with the aforementioned family-related defects, determine the frequency of the same production batch, the same type of alarm information, and the same defect cause, and record the defects.

[0043] The defect records are tagged based on the defect records and device types, and the defect records are stored in the defect database of the big data platform.

[0044] Based on the same concept, the present invention provides an abnormal data processing system based on the fusion of regulation and allocation data, including: an acquisition module, an association module, and a processing module;

[0045] The acquisition module is used to acquire the system model and operation data of the control cloud platform and the integrated operation and distribution platform;

[0046] The association module is used to determine the association between the system models of the control cloud platform and the integrated operation and distribution platform based on the pre-trained model objects.

[0047] The processing module is used to extract data features and process abnormal data from the running data based on the correlation of the system model and the pre-trained neural network model.

[0048] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0049] 1. This invention provides an abnormal data processing method based on the fusion of control and distribution data, comprising: acquiring system models and operational data of a control cloud platform and an integrated operation and distribution platform; determining the correlation between the system models of the control cloud platform and the integrated operation and distribution platform based on pre-trained model objects; and performing data feature extraction and abnormal data processing on the operational data based on the correlation between the system models and a pre-trained neural network model. This method solves the shortcomings of insufficient cross-professional information sharing in traditional methods, and improves the quality of data governance by employing power grid data feature extraction and abnormal processing technologies.

[0050] 2. This invention provides an abnormal data processing method and system based on the fusion of control and distribution data. By applying a neural network model through a big data platform, it realizes feature extraction and data fusion of accident sets at all levels of the power grid.

[0051] 3. This invention provides an abnormal data processing method and system based on the fusion of control and allocation data. Through the unified data fusion mechanism of the big data platform, it greatly reduces the maintenance costs and manual maintenance workload for exercise units, improves the consistency, accuracy and completeness of simulation system data, and realizes the on-demand and rapid preparation of simulation training scenario data.

[0052] 4. This invention provides an abnormal data processing method and system based on the fusion of control and allocation data. Through the preprocessing of accident-type and defect-type data, it automatically constructs a simulation accident set and an equipment defect library, which makes it easier for users to focus on key faults and defects, and provides better data service support for subsequent simulation training data analysis and mining. Attached Figure Description

[0053] Figure 1 A flowchart of the method provided by the present invention;

[0054] Figure 2 A flowchart of a method for implementing integrated control and allocation data fusion is provided in an embodiment of the present invention;

[0055] Figure 3 This is a system structure diagram provided for an embodiment of the present invention. Detailed Implementation

[0056] The embodiments of the present invention will be further described with reference to the accompanying drawings.

[0057] Example 1:

[0058] This invention provides an integrated data processing method and system for control and distribution. Addressing the problems of laborious and time-consuming creation of power system simulation scenarios and low efficiency in processing massive amounts of data, it employs integrated data fusion technology to achieve rapid, on-demand preparation of simulation training scenario data. Through an "event-driven, iterative" approach, it achieves feature extraction and data fusion of fault sets at various levels of the power grid. The unified cloud-based data fusion mechanism significantly reduces maintenance costs and manual maintenance workload for exercise units, improving the consistency, accuracy, and completeness of simulation system data. Figure 1 The method flowchart is provided below, with specific steps as follows:

[0059] Step 1: Obtain the system model and operational data of the control cloud platform and the integrated operation and distribution platform;

[0060] Step 2: Determine the relationship between the system models of the control cloud platform and the integrated operation and dispatch platform based on the pre-trained model objects;

[0061] Step 3: Based on the correlation of the system model and the pre-trained neural network model, perform data feature extraction and abnormal data processing on the running data;

[0062] Step 1: Obtain the system model and operational data of the control cloud platform and the integrated operation and distribution platform, and combine them with... Figure 2 A flowchart illustrating a method for integrating regulation and allocation data fusion is provided, specifically including:

[0063] 1. Model and data acquisition for the control system.

[0064] The control system model is obtained from the model data center on the control cloud platform through the model acquisition service. The model includes container classes, equipment classes, topology classes, and external environment classes. Container classes include power grids, substations, power plants, new energy power stations, bays, loads, DC systems, DC pole systems, DC grounding pole systems, and cross-sections. Equipment classes include generators, AC lines, towers, busbars, transformers, circuit breakers, disconnect switches, grounding switches, parallel capacitors, parallel reactors, static var generators, synchronous condensers, AC filters, DC lines, converter valves, converters, smoothing reactors, DC disconnect switches, DC grounding switches, DC wave traps, and DC filters. Topology classes include single-ended components, double-ended components, endpoint numbers, node numbers, and topology islands. External environment classes include reservoir information, meteorological models, and geographic information.

[0065] Operational data of the control system is obtained from the operation data center on the control cloud platform through data acquisition services. Operational data includes measurement data, fault data, planning and forecasting data, and alarm event data. Measurement data includes active and reactive power of the power grid, power plants, substations, AC lines, DC lines, generators, transformers, and loads, as well as bus voltage and frequency. Fault data includes equipment faults and over-limit data of circuit breakers and disconnectors. Planning and forecasting data includes load forecasts of the power grid, AC lines, transformers, and loads, and output plans of generators and power plants. Alarm event data includes remote signaling changes of circuit breakers and disconnectors, comprehensive intelligent alarm data, fault data, and equipment defect data. Meteorological data includes data on weather stations, power plants, substations, and typhoon paths.

[0066] By controlling the cloud model to obtain services, sub-models for a certain time period and several regions on the control side can be obtained.

[0067] 2. Model and data acquisition for the system.

[0068] The integrated operation, distribution, and dispatch platform obtains the distribution system model through the distribution-side model. The model includes container classes, system classes, equipment classes, and topology classes. Container classes include feeders, substations, power plants, new energy power stations, ring main units, adjustable temperature-controlled loads, and interruptible loads. System classes include microgrid systems, integrated energy systems, and precise load shedding systems. Equipment classes include feeder sections, poles, distribution buses, distribution transformers, circuit breakers, load switches, disconnect switches, grounding switches, capacitors, reactors, and static var generators. Topology classes include single-ended components, double-ended components, endpoint numbers, node numbers, and topology islands.

[0069] Operational data of the distribution system is obtained from the integrated operation, distribution, and dispatch platform through the distribution-side data acquisition service. This includes measurement data, fault data, planning and forecasting data, alarm event data, and equipment ledger data. Measurement data includes active and reactive power of feeders, substations, AC lines, generators, transformers, and user loads, as well as bus voltage. Fault data includes equipment faults and over-limit data of distribution circuit breakers and disconnectors. Planning and forecasting data includes wind and solar power forecasts for renewable energy plants, and generator and power plant output plans. Alarm event data includes remote signaling changes of circuit breakers and disconnectors, changes in the point of common coupling (PCC) point, equipment faults, equipment anomalies, and defects. Equipment ledger data includes equipment operating status, equipment lifecycle data, and historical equipment defect data.

[0070] By using the service obtained from the configuration side model, you can obtain sub-models of certain feeders on the configuration side for a certain time period.

[0071] Step 2: Determine the correlation between the system models of the control cloud platform and the integrated operation, distribution, and dispatch platform based on the pre-trained model objects, specifically including:

[0072] 3. Generate the relationships between model objects.

[0073] According to the requirements of the simulation training exercise, the simulation instructor inputs the time period and simulation range, and uses the model acquisition services of 1-3 and 2-3 to obtain the model and data of a specific part of a historical section.

[0074] Keyword extraction and periodic scanning are performed on the system model and data.

[0075] Through the knowledge discovery module and the intelligent inference engine module, the important key fields of the data in each system are sorted by keywords. Usually, the first keyword is the name, the second keyword is the description, and the third keyword is the code.

[0076] By adjusting the matching rules through an adaptive learning module (neural network model), object association is achieved, forming a corresponding relationship reference library (integrated data). The adaptive learning module can automatically adjust the fuzzy matching rules, perform a second match on data that does not match the first time, achieve boundary model matching, and improve the matching success rate.

[0077] Model object anomaly handling. Anomalies mainly include consistency issues, relationship issues, and model matching issues. Consistency issues include non-standard naming, empty key fields, substations not finding their corresponding regions, and feeder equipment not finding its feeder. Relationship issues include incorrect voltage levels, identical start and end substations, incorrect substation and region affiliation, windings not finding their corresponding transformers, and users not finding their corresponding distribution transformers. Model matching issues include name matching failures, winding main transformer matching errors, and distribution transformer user matching errors.

[0078] For abnormal issues, the model class processing module can handle them. This module can automatically resolve issues such as non-standard naming and incorrect voltage level matching. Other unresolved issues are pushed to the maintenance personnel of the two source systems, the control cloud platform and the integrated operation and dispatch platform, via email for modification. After the modification is completed, the model verification module performs rule verification.

[0079] Unified encoding for object IDs. Based on the needs of business integration, the previous 18-bit field width encoding has been expanded to a 28-bit field width encoding, improving the unified management of IDs for various business models.

[0080] Complete the automatic construction of a unified object model consistent with the control side and the allocation side, and generate data object association relationships.

[0081] Step 3: Based on the correlation relationships of the system model and the pre-trained neural network model, perform data feature extraction and anomaly processing on the running data, specifically including:

[0082] 4. Perform data feature extraction and anomaly processing on data from the unified object model.

[0083] Having completed the fusion of the two system models through the first three steps, the next step is to fuse the data between the two systems. Data fusion includes the following steps: data feature extraction, routine data anomaly handling, correlation data anomaly handling, and accident data anomaly handling.

[0084] Data feature extraction. The main purpose of data feature extraction is to remove redundant attributes from feature data to the greatest extent possible, reduce the difficulty of data analysis and processing, minimize the number of insignificant and useless state indicator variables, reduce system storage requirements, and improve processing efficiency.

[0085] Routine data anomaly handling includes data conversion errors, data range errors, and general data anomalies. Data conversion errors include incorrect data units and incorrect meter base codes. Data range errors include missing data and data noise fluctuations. Data anomalies can be identified through data generalization, normalization, and manual correction. The routine data anomaly handling module (using mathematical statistics algorithms) handles these issues. This module automatically resolves small-scale data missing and noise fluctuation problems. Other unresolved issues are pushed to the maintenance personnel of the two source systems—the control cloud platform and the integrated operation and distribution platform—via email for modification. After modification, the data verification module performs rule verification.

[0086] Related data anomaly handling. This includes bus imbalance, line measurement imbalance, imbalance on two or three sides of windings, and missing calculation components for total added indicators. This is handled by the related data anomaly handling module. This module uses a state estimation algorithm service to automatically correct unbalanced measurements. However, the unresolved issue of missing calculation components is pushed via email to the maintenance personnel of the two source systems: the control cloud platform and the integrated operation and dispatch platform, for modification. After modification, the data verification module performs rule verification.

[0087] Complete the feature extraction and anomaly handling of the data.

[0088] 5. Preprocessing and training matching of accident-related data to construct a dataset.

[0089] Accident data preprocessing. For various fault-related data from the control and distribution sides, accident data is extracted and classified, categorizing fault signals into four types: critical, severe, general, and notification. Accident data anomaly processing module performs information preprocessing on the accident-related data. This module can perform word segmentation, stop word removal, and other functions, and can use natural language processing techniques to vectorize the data.

[0090] Accident data is tagged. First, tagged event samples are extracted from historical monitoring and alarm information. Second, taking keyword-based circuit breaker tripping as an example, an alarm information set within a time window before and after a certain information with the keyword "circuit breaker tripping" is extracted as a milestone. When certain rules are met, various tagged monitoring and alarm events are formed, and finally, a sample library that can be used for training is constructed.

[0091] Fault set data clustering. Based on hierarchical clustering algorithm, the composition of fault sets is classified, and faults such as line faults, transformer faults, feeder segment faults, microgrid faults, and cascading faults are extracted, providing a basis for realizing refined fault set management.

[0092] Fault data is matched with power grid objects. The fault object matching module establishes the association between fault class data and model objects, generating signal-associated device data (fault-associated device data).

[0093] A neural network model is used for sample training to achieve automatic matching of fault sets. Monitoring data (labeled event samples extracted from historical monitoring and alarm information) is used as the training dataset, and fault type, fault severity, and fault location are used as labels. The data is input into a deep convolutional neural network for learning and training. The classification results of different labels are used as basic probability assignment values ​​to achieve the analysis of simulated fault sets.

[0094] 6. Preprocess defective data and build a defect database.

[0095] Defect data preprocessing. This includes defect identification, defect recording, and defect storage.

[0096] First, defect identification is performed. Using the defect processing module, defect data can be automatically identified. This module scans valid alarm data over a period of time based on power grid fault operation data, alarm event operation data, equipment ledger data, and signal-related equipment data. When the number of devices of the same model that have a certain typical alarm signal exceeds a set value, a family defect is automatically identified.

[0097] Secondly, defect records are made. By defining matching rules such as equipment model, production batch, alarm information, defect level, and defect cause, the frequency of similar alarm information or the same defect cause issued by the same batch of equipment is comprehensively analyzed. The defect processing module records suspected family defect records and stores them in the database.

[0098] Defects are objectified, tagged, and then stored in a database to build a defect database.

[0099] 7. Complete the integration of control and distribution data and anomaly handling.

[0100] The beneficial effects of this invention are as follows: The designed integrated data fusion method based on big data regulation and allocation, employing integrated data fusion technology, can solve the shortcomings of insufficient cross-professional information sharing in traditional methods. The unified coding of object IDs enables unified management of system models and data, achieving better data sharing. The use of power grid anomaly data feature extraction improves the quality of data governance. Through the learning and training of a neural network model on an equipment defect analysis library, automatic accident matching is achieved, facilitating users to focus on key faults and defects, providing better data service support for subsequent data analysis and mining, and improving the efficiency of simulation training scenario preparation.

[0101] Traditional control systems typically model only the power generation and transmission system, equating the distribution network to load. Integrated operation, distribution, and dispatch systems model only the distribution and consumption system, equating the transmission network to power generation. Existing simulation training systems lack sufficient information integration. There is a need to build a targeted, multi-level, joint training simulation system that maintains high consistency with simulation operations across the entire scope, process, and scenario. There is an urgent need to unify and integrate simulation models and data from the control and distribution / consumption sides through data fusion technology to improve the realism, accuracy, and consistency of simulation training operations.

[0102] This technology will improve the quality of data management and control, while also significantly enhancing the collaborative capabilities of various departments after an accident, improving the close cooperation between different production processes, shortening the fault repair cycle, improving the efficiency of accident handling, and reducing additional economic losses caused by low efficiency in joint fault handling.

[0103] Taking into account the integrated characteristics of the power grid and the business needs of dispatching and operation management, the system is implemented through the structured design of general data objects for power dispatching. Starting with object ID encoding rules, object model rules and object modeling methods are formulated, and service publishing is achieved through microservices.

[0104] In the preparation process for simulation training, lesson plan preparation is relatively time-consuming and labor-intensive. At this point, it is advisable to fully utilize past accident data to develop contingency plans, forming a set of general power accident examples. Through feature extraction algorithms, dispatching and maintenance personnel assign weighted scores to matters of concern. Combined with the impact of power grid accidents, information is fused to form a set of typical power accident examples. Based on the actual characteristics of various business applications, a standardized data flow for accident handling is designed, connecting all production links in dispatching and distribution, to meet the business needs of simulation training for personnel at all levels.

[0105] Through the data object association expert module, we achieved object association between the control side and the distribution side. At the same time, by utilizing the control cloud object ID encoding rules, we carried out object modeling standardization, realizing organic and persistent association of data objects. Using feature extraction technology, we constructed a set of typical power accident examples for simulation training, providing technical support for the rapid compilation of simulation training courseware.

[0106] Example 2:

[0107] This invention provides an abnormal data processing system based on the fusion of regulation and allocation data, comprising: an acquisition module, a correlation module, and a processing module, combined with... Figure 3 The system architecture diagram is introduced;

[0108] The acquisition module is used to acquire the system model and operation data of the control cloud platform and the integrated operation and distribution platform;

[0109] The association module is used to determine the association between the system models of the control cloud platform and the integrated operation and distribution platform based on the pre-trained model objects.

[0110] The processing module is used to extract data features and process abnormal data from the running data based on the correlation of the system model and the pre-trained neural network model.

[0111] The acquisition module includes: a control system model submodule, an application system model submodule, a control data submodule, and an application data submodule;

[0112] The control system model submodule is used to obtain system models of container classes, device classes, topology classes, and external environment classes related to the control cloud platform;

[0113] The system model submodule is used to obtain the container class, system class, equipment class and topology class system models of the integrated operation, distribution and dispatch platform;

[0114] The control data submodule is used to obtain the operation data of the control cloud platform from the operation data center of the control cloud platform through the data acquisition service;

[0115] The allocation data submodule is used to obtain the operation data of the operation, allocation and dispatch integrated platform from the allocation side data acquisition service.

[0116] The association module includes: a comparison data submodule;

[0117] The reference data submodule is used to extract keywords and perform periodic scanning on the historical control cloud platform data and historical operation and allocation integrated platform data input into the model object through convolutional layers, pooling layers, and fully connected layers to obtain keyword matching rules, operational data, and other reference data arranged by name, description, and code.

[0118] The other comparative data include data with consistency issues, correlation issues, and model matching issues.

[0119] The processing module includes: an extraction processing submodule;

[0120] The extraction and processing submodule performs data feature extraction, routine data anomaly processing, accident data preprocessing, and defect data preprocessing on the running data based on the correlation of the system model and the pre-trained neural network model.

[0121] The extraction and processing submodule includes: an extraction unit, a routine anomaly processing unit, an accident processing unit, and a defect processing unit;

[0122] The extraction unit is used to delete useless state indicator variables of the running data through a big data platform;

[0123] The routine anomaly handling unit is used to perform routine data anomaly handling on the running data;

[0124] The accident handling unit is used to perform accident data preprocessing on the operational data;

[0125] The defect processing unit is used to perform defect-type data preprocessing on the running data.

[0126] The conventional anomaly handling unit includes: a data error subunit, a verification subunit, a correction subunit, and a feedback subunit;

[0127] The data error subunit is used to identify data conversion errors, data range errors, data anomalies, bus imbalance, line measurement imbalance, winding imbalance on both sides or three sides, and data lacking calculation components in the total index through data generalization, normalization, or manual correction.

[0128] The verification subunit is used to perform anomaly elimination processing on the data with data conversion errors, data range errors, and data anomalies through mathematical statistical algorithms, and then perform rule verification after processing.

[0129] The correction subunit is used to correct the data of bus imbalance, line measurement imbalance, and winding two- or three-sided imbalance using a state estimation algorithm.

[0130] The feedback subunit is used to feed back the data of missing calculation components of the total index to the control cloud platform and the integrated operation and distribution platform for modification.

[0131] The data anomalies include: missing data and noise jumps.

[0132] The incident processing unit includes: an incident data subunit, a word segmentation and removal subunit, a tagging subunit, a classification subunit, a matching subunit, and a learning subunit;

[0133] The accident data subunit is used to identify accident data in the operational data through data generalization, normalization or manual correction, and to classify the accident data into critical fault data, serious fault data, general fault data and notification fault data.

[0134] The word segmentation and word removal subunit is used to segment and remove words from the critical fault data, serious fault data, general fault data and notification fault data using natural language processing methods to obtain vectorized fault data.

[0135] The tagging subunit is used to tag the vectorized fault data based on tag samples of pre-acquired historical monitoring and alarm information.

[0136] The classification subunit is used to classify the labeled fault data into line fault data, transformer fault data, feeder segment fault data, microgrid fault data, and cascading fault data using a hierarchical clustering algorithm.

[0137] The matching subunit is used to perform fault association matching between the line fault data, transformer fault data, feeder section fault data, microgrid fault data and cascading fault data and the power grid fault equipment to obtain fault-associated equipment data.

[0138] The learning subunit is used to input the fault-related equipment data into a pre-trained neural network model for learning, and obtain classification results with different labels;

[0139] The label classification includes: fault type, fault severity, and fault location;

[0140] The neural network model is trained by taking labeled event samples extracted from historical monitoring and alarm information as input and the classification results of different labels as output.

[0141] The extraction and processing submodule includes: a fault analysis unit;

[0142] The fault analysis unit is used to perform fault analysis using the classification results as the basic probability allocation values.

[0143] The defect processing unit includes: a defect identification subunit, a defect recording subunit, and a storage subunit;

[0144] The defect identification subunit identifies defects in the operation data based on power grid fault operation data, alarm event operation data, equipment ledger data, and signal-related equipment data. When the number of alarms occurring in the same type of equipment exceeds a preset value, it is identified as a family defect.

[0145] The defect recording subunit determines the frequency of the same production batch, the same type of alarm information, and the same defect cause based on the equipment model, production batch, alarm information, defect level, and defect cause of the same type of equipment with the family-related defects, and records the defects accordingly.

[0146] The storage subunit performs tagging based on the defect record and device type, and stores the defect record in the defect database of the big data platform.

[0147] 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.

[0148] 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 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0149] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0150] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0151] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of the claims of the present invention pending approval.

Claims

1. An abnormal data processing method based on the fusion of regulation and allocation data, characterized in that, include: Acquire system models and operational data from the control cloud platform and the integrated operation and distribution platform; The correlation between the system models of the control cloud platform and the integrated operation and dispatch platform is determined based on the pre-trained model objects. Based on the correlation of the system model and the pre-trained neural network model, data feature extraction and abnormal data processing are performed on the running data. The training of the model object includes: The historical control cloud platform data and historical operation and allocation integrated platform data input into the model object are processed through convolutional layers, pooling layers and fully connected layers to extract keywords and perform periodic scanning to obtain keyword matching rules, operational data and other reference data arranged by name, description and code; The other comparative data includes data with consistency issues, correlation issues, and model matching issues; The process of extracting data features and processing abnormal data from the operational data based on the correlation relationships of the system model and the pre-trained neural network model includes: Based on the correlation of the system model and the pre-trained neural network model, data feature extraction, routine data anomaly processing, accident data preprocessing, and defect data preprocessing are performed on the operational data. Perform routine data anomaly handling on the aforementioned operational data, including: Data conversion errors, data range errors, data anomalies, bus imbalance, line measurement imbalance, winding imbalance on both or three sides, and data lacking calculation components in the total index are identified through data generalization, normalization, or manual correction. The data with conversion errors, data range errors, and data anomalies are eliminated by mathematical statistical algorithms, and then the rules are validated. The state estimation algorithm is used to correct the data on bus imbalance, line measurement imbalance, and imbalance on both or three sides of the winding. The data that lacks a calculation component in the total index will be fed back to the control cloud platform and the integrated operation and distribution platform for modification. The data anomalies include: missing data and noise jumps; The operational data undergoes accident data preprocessing, including: Accident data in the operational data is identified through data generalization, normalization, or manual correction, and the accident data is divided into critical fault data, serious fault data, general fault data, and notification fault data. The critical fault data, serious fault data, general fault data, and notification fault data are segmented and words are removed using natural language processing methods to obtain vectorized fault data. The vectorized fault data is tagged based on the tag samples of pre-acquired historical monitoring and alarm information. Using a hierarchical clustering algorithm, the fault data after the labeling process is divided into line fault data, transformer fault data, feeder section fault data, microgrid fault data, and cascading fault data. The fault data of the line, transformer, feeder section, microgrid, and cascading faults are matched with the faulty equipment in the power grid to obtain fault-related equipment data. The fault-related equipment data is input into a pre-trained neural network model for learning, and classification results with different labels are obtained. The label classification includes: fault type, fault severity, and fault location; The neural network model is trained by taking labeled event samples extracted from historical monitoring and alarm information as input and the classification results of different labels as output.

2. The method as described in claim 1, characterized in that, The acquisition of the system model and operational data of the control cloud platform includes: Obtain system models for container, device, topology, and external environment classes related to the control cloud platform; Obtain the container, system, device, and topology system models for the integrated operation, distribution, and dispatch platform; The operational data of the control cloud platform is obtained from the operational data center of the control cloud platform through the data acquisition service; The operational data of the integrated operation, distribution, and dispatch platform is obtained through the data acquisition service on the distribution side.

3. The method as described in claim 1, characterized in that, The process of determining the correlation between the system models of the control cloud platform and the integrated operation and dispatch platform based on the pre-trained model object includes: The consistency problem is eliminated by using the rule base of the big data platform, and the correlation problem and model matching problem are fed back to the big data platform. The running data is uniformly encoded with a preset field width for ID.

4. The method as described in claim 1, characterized in that, Data feature extraction of the operational data includes: Useless status indicator variables in the operational data are deleted through a big data platform.

5. The method as described in claim 1, characterized in that, The step involves inputting the fault-related device data into a pre-trained neural network model for learning to obtain classification results with different labels, followed by: The classification results are used as the basic probability allocation values ​​for fault analysis.

6. The method as described in claim 5, characterized in that, The operational data undergoes defect data preprocessing, including: Based on power grid fault-type operation data, alarm event-type operation data, equipment ledger data, and signal-related equipment data, defect identification is performed on the operation data. When the number of alarms occurring on the same type of equipment exceeds a preset value, it is identified as a family-related defect. Based on the equipment model, production batch, alarm information, defect level, and defect cause of the same type of equipment with the aforementioned family-related defects, determine the frequency of the same production batch, the same type of alarm information, and the same defect cause, and record the defects. The defect records are tagged based on the defect records and device types, and the defect records are stored in the defect database of the big data platform.

7. An abnormal data processing system based on the fusion of regulation and allocation data, characterized in that, include: The module consists of an acquisition module, a relationship module, and a processing module. The acquisition module is used to acquire the system model and operation data of the control cloud platform and the integrated operation and distribution platform; The association module is used to determine the association between the system models of the control cloud platform and the integrated operation and distribution platform based on the pre-trained model objects. The processing module is used to extract data features and process abnormal data from the running data based on the correlation of the system model and the pre-trained neural network model. The association module includes: a comparison data submodule; The reference data submodule is used to extract keywords and perform periodic scanning on the historical control cloud platform data and historical operation and allocation integrated platform data input into the model object through convolutional layers, pooling layers and fully connected layers to obtain keyword matching rules, running data and other reference data arranged by name, description and code; The other comparative data includes data with consistency issues, correlation issues, and model matching issues; The processing module includes: an extraction processing submodule; The extraction and processing submodule performs data feature extraction, routine data anomaly processing, accident data preprocessing, and defect data preprocessing on the running data based on the correlation relationship of the system model and the pre-trained neural network model. The extraction and processing submodule includes: an extraction unit, a routine anomaly processing unit, an accident processing unit, and a defect processing unit; The extraction unit is used to delete useless state indicator variables of the running data through a big data platform; The routine anomaly handling unit is used to perform routine data anomaly handling on the running data; The accident handling unit is used to perform accident data preprocessing on the operational data; The defect processing unit is used to perform defect-type data preprocessing on the running data; The conventional anomaly handling unit includes: a data error subunit, a verification subunit, a correction subunit, and a feedback subunit; The data error subunit is used to identify data conversion errors, data range errors, data anomalies, bus imbalance, line measurement imbalance, winding imbalance on both sides or three sides, and data lacking calculation components in the total index through data generalization, normalization, or manual correction. The verification subunit is used to perform anomaly elimination processing on the data with data conversion errors, data range errors, and data anomalies through mathematical statistical algorithms, and then perform rule verification after processing. The correction subunit is used to correct the data of bus imbalance, line measurement imbalance, and winding two- or three-sided imbalance using a state estimation algorithm. The feedback subunit is used to feed back the data of missing calculation components of the total index to the control cloud platform and the integrated operation and distribution platform for modification. The data anomalies include: missing data and noise jumps; The incident processing unit includes: an incident data subunit, a word segmentation and removal subunit, a tagging subunit, a classification subunit, a matching subunit, and a learning subunit; The accident data subunit is used to identify accident data in the operational data through data generalization, normalization or manual correction, and to classify the accident data into critical fault data, serious fault data, general fault data and notification fault data. The word segmentation and word removal subunit is used to segment and remove words from the critical fault data, serious fault data, general fault data and notification fault data using natural language processing methods to obtain vectorized fault data. The tagging subunit is used to tag the vectorized fault data based on tag samples of pre-acquired historical monitoring and alarm information. The classification subunit is used to classify the labeled fault data into line fault data, transformer fault data, feeder segment fault data, microgrid fault data, and cascading fault data using a hierarchical clustering algorithm. The matching subunit is used to perform fault association matching between the line fault data, transformer fault data, feeder section fault data, microgrid fault data and cascading fault data and the power grid fault equipment to obtain fault-associated equipment data. The learning subunit is used to input the fault-related equipment data into a pre-trained neural network model for learning, and obtain classification results with different labels; The label classification includes: fault type, fault severity, and fault location; The neural network model is trained by taking labeled event samples extracted from historical monitoring and alarm information as input and the classification results of different labels as output.

Images