A line loss abnormality analysis operation monitoring system based on big data

Abstract

This invention discloses a big data-based line loss anomaly analysis and operation monitoring system, relating to the field of power system technology. It includes a region division unit, a data acquisition and analysis unit, a virtual model building unit, and a platform sharing and management unit. The region division unit is electrically connected to the data acquisition and analysis unit and the virtual model building unit. The virtual model building unit is also electrically connected to the platform sharing and management unit. In the process of line loss anomaly analysis and monitoring, this invention pre-divides the power grid region and builds a virtual simulation model based on the divided region. Within the constructed simulation model, corresponding regional sub-models are built for each divided region. After data acquisition and analysis, the results are directly reflected in the constructed model, allowing for feedback on the specific situation within the region.

Description

Technical Field

[0001] This invention relates to the field of power system technology, specifically to a line loss anomaly analysis and operation monitoring system based on big data. Background Technology

[0002] Currently, line loss, also known as line power loss or energy loss, refers to the loss of effective electrical energy due to physical and technical reasons during the transmission, transformation, and distribution of electrical energy. Chinese patent CN114580819B discloses "A line loss anomaly analysis and operation monitoring system based on big data, including an indicator statistics module, an anomaly classification module, an anomaly analysis module, and an analysis result module. The indicator statistics module is used to obtain line loss statistics data from the metering system for standard verification, obtaining a line loss anomaly list and a high loss list. The anomaly classification module is used to classify the anomaly causes in the line loss anomaly list data according to the line loss rate. The anomaly analysis module is used to specifically analyze and judge the anomaly causes. The analysis result module is used to display and summarize the specific causes of the anomalies. This invention realizes intelligent line loss anomaly analysis in transformer substations, minimizing the workload and errors of manual analysis, and forming a transformation from manual analysis to intelligent analysis." Existing technologies only address the following issues: manual judgment requires verifying the accuracy of information such as marketing and metering system records, electricity statistics, and data collection to analyze the causes of transformer area anomalies, which is a cumbersome and time-consuming process; analyzing line loss anomalies requires strong comprehensive business skills and data sensitivity, but the professional competence of township power supply station staff is relatively low, making it difficult for them to effectively manage line loss by comprehensively utilizing their own experience; there is no unified analysis and monitoring platform for details and data on transformer areas with line loss anomalies, and the back-end big data of various information systems has not been fully explored and utilized, resulting in "information silos." However, in the process of handling line losses, most of the verification and processing are still done manually, which cannot quickly and accurately locate the problem. This results in low efficiency in troubleshooting when anomalies occur. In addition, the routine operation of the power grid relies on limited measurement point data and human experience judgment, which makes a lot of data invisible and unable to accurately grasp the power flow distribution, loss composition and real-time equipment efficiency. Furthermore, problems cannot be handled and optimized in the first place, resulting in a lag in problem handling. Summary of the Invention

[0003] The purpose of this invention is to provide a big data-based line loss anomaly analysis and operation monitoring system to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a big data-based line loss anomaly analysis and operation monitoring system, comprising a region division unit, a data acquisition and analysis unit, a virtual model building unit, and a platform sharing and management unit. The region division unit is electrically connected to the data acquisition and analysis unit and the virtual model building unit. The data acquisition and analysis unit is electrically connected to the virtual model building unit. The virtual model building unit is electrically connected to the platform sharing and management unit. The area division unit divides the area under management and monitoring, and performs identification processing after the division is completed; The data acquisition and analysis unit collects data within the divided regions and performs multi-faceted analysis on the data after the data acquisition is completed. The virtual model building unit builds a power grid simulation model based on the collected data and the divided areas, provides real-time feedback on the collected data, and formulates a dynamic simulation and prediction scheme for line loss after the model is built. The platform sharing and management unit establishes an internal account management scheme for the power grid, requiring users to log in with an account to access the power grid simulation model, and formulates a sharing scheme to achieve data sharing.

[0005] Preferably, after the area division unit completes the area division, it divides and identifies the areas according to the method of industrial load area, residential load area, commercial load area and mixed load area, and sets corresponding abnormal thresholds within the divided areas. After the area division is completed, data layer fusion and analysis layer fusion are performed. Data layer fusion: each partition is assigned "physical ID + logical label + management label", and real-time data collection is automatically associated with multi-dimensional labels to support cross-dimensional analysis. Analysis layer fusion: when judging line loss anomalies in real time, physical topology consistency, logical threshold compliance and management responsibility matching are checked at the same time to avoid false alarms.

[0006] Preferably, the data acquisition and analysis unit collects data within the divided areas, builds a database within each area, and identifies the corresponding databases as industrial, residential, commercial, and mixed databases. The collected data is centrally collected and stored, and timestamped. The data includes power grid operation data, equipment asset data, environmental meteorological data, marketing business data, and external data. Power grid operation data includes real-time data such as voltage, current, power, and electricity consumption from smart meters, transformers, DTU / FTU / TTU, and SCADA systems. Equipment asset data includes line parameters, transformer ledgers, switchgear information, and GIS geospatial data. Environmental meteorological data includes environmental factors affecting line losses such as temperature, humidity, wind, rain, and snow. Marketing business data includes user profiles, electricity usage categories, electricity price standards, and meter reading records. External data includes power grid topology, load forecasting data, and power grid planning data.

[0007] Preferably, the data acquisition method includes real-time acquisition, batch acquisition, edge computing preprocessing, and manual acquisition. Real-time acquisition: data is acquired at the second / minute level through communication technologies such as 4G / 5G, power line carrier, and fiber optics. Batch acquisition: historical data and non-real-time data are synchronized at regular intervals. Edge computing preprocessing: data is cleaned, format converted, and preliminarily analyzed at the data source to reduce the pressure on the cloud. Manual acquisition: in special areas or when anomalies occur in the above acquisition process, professional personnel are used to collect and verify the data.

[0008] Preferably, the virtual model building unit first builds a central processing model using virtual technology and digital twin technology. Then, within the central model, sub-models are built according to the different regions after division. The sub-models include industrial models, residential models, commercial models, and hybrid models. Anomaly analysis and processing schemes, prediction schemes, learning schemes, and segmented control schemes are built within different sub-models. A control scheme is built within the central processing model. The control scheme is associated with the anomaly analysis and processing scheme, prediction optimization scheme, and learning scheme. The control scheme includes online expert processing and software processing. The software processing uses big data processing technology to store, calculate, and stream the data generated during the real-time generation of the anomaly analysis and processing scheme, prediction optimization scheme, and learning scheme. The online expert processing is implemented by having experts remotely process the data online in real-time through a rotation system. A central database is built to centrally collect and process the generated data. The stored data is labeled and classified. The segmented control scheme performs segmented control within each sub-model, identifies the controlled area, and adds a label to the data when it is generated.

[0009] Preferably, the anomaly analysis and handling scheme first classifies the anomalies into three categories: Category A, Category B, and Category C. Category A anomalies are the most severe, Category B anomalies are severe, and Category C anomalies are minor. Then, a first, second, and third analysis and handling scheme are formulated. The first analysis and handling scheme analyzes and handles Category A anomalies, employing online expert collaboration, graph computing, machine learning / artificial intelligence, edge computing, and collaboration with experienced experts working on the front lines. The second analysis and handling scheme analyzes and handles Category B anomalies, employing online expert collaboration, graph computing, machine learning / artificial intelligence, and edge computing. The third analysis and handling scheme handles Category C anomalies, employing graph computing, machine learning / artificial intelligence, and edge computing.

[0010] Preferably, the prediction scheme performs predictions on the established model during implementation and includes the following steps; (1) Establish a fast mapping relationship between “state-line loss” using a digital twin model. ① Scenario generation: In the digital twin model, a large number of typical operating scenarios are generated in batches by changing the load distribution, generator output, topology and reactive power compensation configuration. ② Perform power flow calculation on each scenario to obtain the accurate theoretical line loss value under the scenario. ③ Build a proxy model: Use the above massive “input scenario parameters-output line loss value” data pair to train a lightweight machine learning model. (2) Key boundary condition prediction includes ① Load prediction: using historical load, weather, holidays, and event data, predicting the short-term and medium-term load curves for each prediction node / bus. This is the most influential factor. ② New energy power generation prediction: predicting the output curves for photovoltaic, wind power, and other access points. ③ Topology and operation mode planning: obtaining future planned maintenance, switching operations, and network reconfiguration information from OMS and operation plans. ④ Obtaining future temperature, humidity, wind speed, and sunlight, which affect load and equipment parameters, as well as understanding future transaction plans and dispatch instructions. (3) Perform fusion prediction. Input the prediction results of (2) into the proxy model trained in (1) for calculation. ① Determine the spatiotemporal range of the prediction: For example, predict "the line loss rate of the city's power grid every 15 minutes tomorrow". ② Assemble future scenarios: For each prediction time point, assemble a scenario vector according to the predicted load distribution, new energy output and planned topology at that time. ③ Proxy model calculation: Input the scenario vector into the proxy model and output the theoretical technical line loss value (ΔP_tech) at that time. ④ Superimpose the management line loss baseline: Technical line loss is the "calculable" part. For "management line loss", due to its randomness and concealment, it is difficult to predict through simulation. Therefore, the method of baseline value + dynamic adjustment is used for prediction processing. ⑤ Synthesize the final prediction: Total predicted line loss = technical line loss predicted by the proxy model + management line loss prediction value.

[0011] Preferably, the learning scheme trains and optimizes the built model, and establishes a pass / fail rating level before training. The rating levels are Level 1, Level 2, and Level 3. The pass / fail standard for Level 1 is 60%, with the highest intensity; the pass / fail standard for Level 2 is 80%, with relatively strong intensity; and the pass / fail standard for Level 3 is 100%, with moderate intensity. The learning scheme also includes a first training analysis scheme, a second training analysis scheme, and a third training analysis scheme. Level 1 corresponds to the first training analysis scheme, which has the highest intensity during training and the most comprehensive analysis process. Level 2 corresponds to the second training analysis scheme, which has relatively strong intensity during training and moderate comprehensiveness in its analysis process. Level 3 corresponds to the third training analysis scheme, which has moderate intensity during training and moderate comprehensiveness in its analysis process. During the training process, each level is trained, and only after the training at each level intensity is qualified can the model be determined and implemented.

[0012] Preferably, the platform sharing and management unit supports data interaction with government departments and power users, and centralized management is achieved through account registration. When government departments or power users use the model, they need to register an account in advance, and can log in directly to use it. It also includes a problem feedback and data upload scheme, establishing a hierarchical supervision system comprising a basic response monitoring layer, a middle collaborative monitoring layer, and a top collaborative monitoring layer. The basic response monitoring layer consists of several sub-monitoring layers. This layer performs preliminary screening of the data and provides feedback on simple issues. Issues collected by the basic response monitoring layer are then fed back to the middle collaborative monitoring layer, which processes the received issues. The system performs detailed analysis and processing of issues, and handles slightly more complex problems. Sensitive and complex issues are transmitted to the top collaborative monitoring layer for review and analysis before processing. Feedback time is set for the basic response monitoring layer and the middle collaborative monitoring layer. If the data collected by the basic response monitoring layer cannot be processed within the specified time, it is directly fed back to the middle collaborative monitoring layer. If the middle collaborative monitoring layer fails to resolve the issue within the specified time, it is directly fed back to the top collaborative monitoring layer. The top collaborative monitoring layer adopts a 24-hour shift system. If any issues transmitted to the top collaborative monitoring layer are not processed, adjustments are made to the basic response monitoring layer and the middle collaborative monitoring layer.

[0013] Compared with the prior art, the beneficial effects of the present invention are: This invention, in the process of analyzing and monitoring line loss anomalies, pre-divides the power grid area and builds a virtual simulation model based on the divided areas. Within each divided area, a corresponding sub-model is built. After data collection and analysis, the results are directly reflected in the constructed model. This model provides feedback on the specific situation within each area. Through zoned management, anomalies can be handled accurately and efficiently. Regular automatic and proactive model training and optimization ensure the model remains in optimal condition. Furthermore, a shared platform facilitates data sharing and user-managed monitoring, enabling timely problem resolution and guaranteeing high efficiency in monitoring and processing. Attached Figure Description

[0014] Figure 1 This is a system block diagram provided for an embodiment of the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Please see Figure 1 The present invention provides a technical solution: a line loss anomaly analysis and operation monitoring system based on big data, comprising a region division unit, a data acquisition and analysis unit, a virtual model building unit, and a platform sharing and management unit. The region division unit is electrically connected to the data acquisition and analysis unit and the virtual model building unit. The data acquisition and analysis unit is electrically connected to the virtual model building unit. The virtual model building unit is electrically connected to the platform sharing and management unit. The area division unit divides the area under management and monitoring, and performs identification processing after the division is completed; The data acquisition and analysis unit collects data within the divided regions and performs multi-faceted analysis on the data after the data acquisition is completed. The virtual model building unit builds a power grid simulation model based on the collected data and the divided areas, provides real-time feedback on the collected data, and formulates a dynamic simulation and prediction scheme for line loss after the model is built. The platform sharing and management unit establishes an internal account management scheme for the power grid, requiring users to log in with an account to access the power grid simulation model, and formulates a sharing scheme to achieve data sharing.

[0017] After the area division is completed, the area division unit is divided and identified according to the method of industrial load area, residential load area, commercial load area and mixed load area. Corresponding abnormal thresholds are set within the divided areas. After the area division is completed, data layer fusion and analysis layer fusion are performed. Data layer fusion: each partition is assigned "physical ID + logical label + management label". Real-time data collection automatically associates multi-dimensional labels to support cross-dimensional analysis (such as line loss anomaly analysis of "10kV Chengdong Line (physical ID) - industrial load area (logical label) - Chengdong Power Supply Station (management label)"). Analysis layer fusion: when judging line loss anomalies in real time, physical topology consistency, logical threshold compliance and management responsibility matching are checked at the same time to avoid false alarms. Simultaneously, technical safeguards are implemented after zoning, including real-time grid topology sensing technology, a dynamic zoning update engine, and multi-dimensional tag mapping technology. Real-time grid topology sensing technology, based on distribution automation systems (DAS) and IoT acquisition terminals, enables second-level updates of switch status and device connection relationships, providing fundamental data for zoning adjustments. The dynamic zoning update engine employs a streaming computing framework (such as Flink) to process topology and load data in real time, automatically triggering zoning adjustment logic to ensure synchronization between zoning and grid status. Multi-dimensional tag mapping technology constructs a zoning tag library, supporting flexible combinations and rapid queries of physical, logical, and management tags, enabling multi-dimensional tracing of real-time anomalies.

[0018] The data acquisition and analysis unit collects data within the divided areas, builds a database within each area, and identifies the corresponding databases as industrial, residential, commercial, and mixed databases. The collected data is centrally collected and stored, and timestamped. The data includes power grid operation data, equipment asset data, environmental meteorological data, marketing data, and external data. Power grid operation data includes real-time data such as voltage, current, power, and electricity consumption from smart meters, transformers, DTU / FTU / TTU, and SCADA systems. Equipment asset data includes line parameters, transformer ledgers, switchgear information, and GIS geospatial data. Environmental meteorological data includes environmental factors affecting line losses such as temperature, humidity, wind speed, rain, and snow. Marketing data includes user profiles, electricity usage categories, electricity price standards, and meter reading records. External data includes power grid topology, load forecasting data, and power grid planning data. The data acquisition methods include real-time acquisition, batch acquisition, edge computing preprocessing, and manual acquisition. Real-time acquisition utilizes communication technologies such as 4G / 5G, power line carrier, and fiber optics to achieve second-level / minute-level data acquisition. Batch acquisition synchronizes historical data with non-real-time data at regular intervals (hourly / daily). Edge computing preprocessing performs data cleaning, format conversion, and preliminary analysis at the data source to reduce cloud pressure. Manual acquisition employs professional personnel to verify data in special areas or when anomalies occur in the above acquisition stages, ensuring the accuracy and stability of data acquisition after an anomaly. After data acquisition, a multi-level data verification mechanism is established, and data repair techniques such as interpolation and trend analysis are used to address issues such as missing, abnormal, and inconsistent data. An adaptive mechanism of "general model + personalized parameters" is constructed to support online model learning and updates. This addresses the problem of significant differences in line loss characteristics across different regions and types of power grids. Standardized interfaces (such as IEC 61970 and IEC 61968) are used, and middleware is developed to achieve smooth integration. This addresses the complexities of interfacing with existing power information systems, employing security measures such as encrypted transmission, access control, and operation auditing to comply with power industry information security standards. This mitigates issues such as data leakage, network attacks, and illegal operations, and also establishes a data platform.

[0019] The virtual model building unit first builds a central processing model using virtual technology and digital twin technology. Then, it builds sub-models within the central model according to the different regions after division. The sub-models include industrial models, residential models, commercial models, and hybrid models. Within each sub-model, anomaly analysis and processing schemes, prediction schemes, learning schemes, and segmented control schemes are built. Within the central processing model, a control scheme is built, which is associated with the anomaly analysis and processing scheme, prediction optimization scheme, and learning scheme. The control scheme includes online expert processing and software processing. The software processing uses big data processing technology to store, calculate, and stream the data generated during the real-time generation of the anomaly analysis and processing scheme, prediction optimization scheme, and learning scheme. The online expert processing is implemented by having experts remotely process the data online in real-time through a rotation system. A central database is built to collect and process the generated data. The stored data is labeled and classified. The segmented control scheme divides each sub-model into segments for control and identifies each controlled area. When data is generated, a label is added to the data. Furthermore, the control and handling plan establishes a cross-regional coordination plan. After the data transmitted from each sub-model to the central model is analyzed, if the situation in a certain regional model becomes too serious, resulting in insufficient resources and processing capacity within the region, then the adjacent resources and manpower will be transferred to the abnormal region for collaborative processing through control measures. The anomaly analysis and handling plan first categorizes anomalies into three types: Category A, Category B, and Category C. Category A anomalies are the most severe, Category B anomalies are severe, and Category C anomalies are minor. Then, three analysis and handling plans are developed. The first plan analyzes and handles Category A anomalies, employing online expert collaboration, graph computing, machine learning / artificial intelligence, edge computing, and collaboration with experienced frontline experts. The second plan analyzes and handles Category B anomalies, also employing online expert collaboration, graph computing, machine learning / artificial intelligence, and edge computing. The third plan handles Category C anomalies, employing graph computing, machine learning / artificial intelligence, and edge computing. The prediction scheme makes predictions based on the established model during implementation, and includes the following steps; (1) Establish a fast mapping relationship between “state-line loss” using a digital twin model. ① Scenario generation: In the digital twin model, by changing the load distribution, generator output, topology (switching state) and reactive power compensation configuration, a large number (e.g., tens of thousands) of typical operating scenarios are generated in batches. ② Perform power flow calculation (AC-OPF or simplified DC power flow) on each scenario to obtain the accurate theoretical line loss value under the scenario. ③ Build a proxy model: Use the above-mentioned large number of “input scenario parameters-output line loss value” data pairs to train a lightweight machine learning model (e.g., gradient boosting regression tree GBRT, neural network NN or Gaussian process regression GPR). (2) Key boundary condition prediction includes ① Load prediction: using historical load, weather, holidays, and event data, predicting the future short-term (hourly) and medium-term (daily / weekly) load curves for each prediction node / bus (or aggregated to key nodes). This is the most influential factor. ② New energy power generation prediction: predicting the output curves for photovoltaic, wind power, and other access points. ③ Topology and operation mode planning: obtaining future planned maintenance, switching operations, and network reconfiguration information from the OMS (Outage Management System) and operation plan. ④ Obtaining future temperature, humidity, wind speed, and sunlight, which affect load and equipment parameters (such as conductor resistance changing with temperature), and understanding future transaction plans and dispatch instructions. (3) Perform fusion prediction. Input the prediction results of (2) into the proxy model trained in (1) for calculation. ① Determine the spatiotemporal range of the prediction: For example, predict "the line loss rate of the city's power grid every 15 minutes tomorrow". ② Assemble future scenarios: For each prediction time point (such as 14:00 tomorrow), assemble a scenario vector according to the predicted load distribution, new energy output and planned topology at that time. ③ Proxy model calculation: Input the scenario vector into the proxy model and output the theoretical technical line loss value (ΔP_tech) at that time instantly. ④ Superimpose the management line loss baseline: Technical line loss is the "calculable" part. For "management line loss" (electricity theft, metering error, etc.), due to its randomness and concealment, it is difficult to predict through simulation. Therefore, the method of benchmark value + dynamic adjustment is used for prediction processing (benchmark value: take the average value or moving average of management line loss in the same period of recent history (such as similar working days). Dynamic adjustment: a wavelet analysis or time series model based on historical data can be introduced to predict the fluctuation trend of management line loss). ⑤ Synthesize the final prediction: Total predicted line loss = technical line loss predicted by the proxy model + management line loss prediction value. The learning program trains and optimizes the built model, and establishes a pass / fail rating system before training. The rating systems are Level 1, Level 2, and Level 3. Level 1 has a pass / fail standard of 60% and is the most intensive; Level 2 has a pass / fail standard of 80% and is relatively intensive; and Level 3 has a pass / fail standard of 100% and is of moderate intensity. The learning program also includes a first training analysis scheme, a second training analysis scheme, and a third training analysis scheme. Level 1 corresponds to the first training analysis scheme, which has the highest training intensity and the most comprehensive analysis process. Level 2 corresponds to the second training analysis scheme, which has relatively strong training intensity and moderate comprehensiveness in its analysis process. Level 3 corresponds to the third training analysis scheme, which has moderate training intensity and moderate comprehensiveness in its analysis process. During training, each level is trained, and only after all training levels at each intensity are deemed satisfactory can the program be considered qualified and implemented. When training at the corresponding level, if the set standard is not met, adjustments need to be made and training resumed until the set standard is reached. If the standard is not met, a comprehensive analysis and processing are required. Adjustments are made based on the analysis results, and the data generated during the analysis is stored. Once the set standard is met after training, training can be stopped and implemented directly. Feedback and periodic adjustment plans are formulated. Feedback and adjustment aim to collect problems encountered by users during use. If too many problems are received in a short period of time, training adjustments are implemented immediately. The learning plan is also adjusted through expert review, and the time for periodic adjustments is set in advance. When the set time is reached, training will start automatically. The platform's sharing and management unit supports data interaction with government departments and power users. Centralized management is achieved through account registration. When government departments or power users use the model, they need to register an account in advance and log in directly upon use. The unit includes a problem feedback and data upload scheme, establishing a tiered monitoring system comprising a basic response monitoring layer, a middle collaborative monitoring layer, and a top collaborative monitoring layer. The basic response monitoring layer consists of several sub-monitoring layers. This layer performs preliminary screening of the data and provides feedback on simple issues. Issues collected by the basic response monitoring layer are then fed back to the middle collaborative monitoring layer, which further processes the received issues. Detailed analysis and processing are conducted, and slightly more complex issues are analyzed and processed. Sensitive and complex issues are transmitted to the top collaborative monitoring layer for review and analysis before processing. Feedback time is set for the basic response monitoring layer and the middle collaborative monitoring layer. If the data collected by the basic response monitoring layer cannot be processed within the specified time, it is directly fed back to the middle collaborative monitoring layer. If the middle collaborative monitoring layer fails to resolve the issue within the specified time, it is directly fed back to the top collaborative monitoring layer. The top collaborative monitoring layer adopts a 24-hour shift system. If data transmitted to the top collaborative monitoring layer is not processed, adjustments are made to the basic response monitoring layer and the middle collaborative monitoring layer. Furthermore, the problem feedback aims to collect feedback on problems encountered by users during the login process. Once the problems are collected, they are marked and then fed back to the virtual model building unit. Additionally, first-hand data generated during offline work is collected and fed back centrally. During the feedback process, the generated data is processed uniformly and marked after processing.

[0020] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0021] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A line loss anomaly analysis and operation monitoring system based on big data, characterized in that: It includes a region division unit, a data acquisition and analysis unit, a virtual model building unit, and a platform sharing and management unit. The region division unit is electrically connected to the data acquisition and analysis unit and the virtual model building unit. The data acquisition and analysis unit is electrically connected to the virtual model building unit, and the virtual model building unit is electrically connected to the platform sharing and management unit. The area division unit divides the area under management and monitoring, and performs identification processing after the division is completed; The data acquisition and analysis unit collects data within the divided regions and performs multi-faceted analysis on the data after the data acquisition is completed. The virtual model building unit builds a power grid simulation model based on the collected data and the divided areas, provides real-time feedback on the collected data, and formulates a dynamic simulation and prediction scheme for line loss after the model is built. The platform sharing and management unit establishes an internal account management scheme for the power grid, requiring users to log in with an account to access the power grid simulation model, and formulates a sharing scheme to achieve data sharing.

2. The line loss anomaly analysis and operation monitoring system based on big data according to claim 1, characterized in that: After the area division is completed, the area division unit is divided and identified according to the method of industrial load area, residential load area, commercial load area and mixed load area. Corresponding anomaly thresholds are set within the divided areas. After the area division is completed, data layer fusion and analysis layer fusion are performed. Data layer fusion: each partition is assigned "physical ID + logical tag + management tag", and real-time data collection is automatically associated with multi-dimensional tags to support cross-dimensional analysis. Analysis layer fusion: when judging line loss anomalies in real time, physical topology consistency, logical threshold compliance and management responsibility matching are checked at the same time to avoid false alarms.

3. The line loss anomaly analysis and operation monitoring system based on big data according to claim 1, characterized in that: The data acquisition and analysis unit collects data within the divided areas, establishes a database within each area, and identifies the corresponding databases as industrial, residential, commercial, and mixed databases. The collected data is centrally collected and stored, and timestamped. The data includes power grid operation data, equipment asset data, environmental meteorological data, marketing data, and external data. Power grid operation data includes real-time data such as voltage, current, power, and electricity consumption from smart meters, transformers, DTU / FTU / TTU, and SCADA systems. Equipment asset data includes line parameters, transformer ledgers, switchgear information, and GIS geospatial data. Environmental meteorological data includes environmental factors affecting line losses such as temperature, humidity, wind speed, rain, and snow. Marketing data includes user profiles, electricity usage categories, electricity price standards, and meter reading records. External data includes power grid topology, load forecasting data, and power grid planning data.

4. The line loss anomaly analysis and operation monitoring system based on big data according to claim 3, characterized in that: The data acquisition methods include real-time acquisition, batch acquisition, edge computing preprocessing, and manual acquisition. Real-time acquisition utilizes communication technologies such as 4G / 5G, power line carrier, and fiber optics to achieve second-level / minute-level data acquisition. Batch acquisition involves periodically synchronizing historical data with non-real-time data. Edge computing preprocessing performs data cleaning, format conversion, and preliminary analysis at the data source to reduce cloud pressure. Manual acquisition employs professional personnel to perform data collection and verification in special areas or when anomalies occur in the above acquisition stages.

5. The line loss anomaly analysis and operation monitoring system based on big data according to claim 1, characterized in that: The virtual model building unit first constructs a central processing model using virtual technology and digital twin technology. Then, within the central model, sub-models are built according to the different regions after division. These sub-models include industrial models, residential models, commercial models, and hybrid models. Within each sub-model, anomaly analysis and processing schemes, prediction schemes, learning schemes, and segmented control schemes are built. Within the central processing model, a control scheme is built, which is associated with the anomaly analysis and processing scheme, prediction optimization scheme, and learning scheme. The control scheme includes online expert processing and software processing. The software processing uses big data processing technology to store, calculate, and stream the data generated during the real-time generation of the anomaly analysis and processing scheme, prediction optimization scheme, and learning scheme. The online expert processing is implemented by allowing experts to remotely process data online in real-time through a rotation system. A central database is built to centrally collect and process the generated data, and the stored data is labeled and classified. The segmented control scheme divides each sub-model into segments for control, identifies the controlled areas, and adds labels to the data when it is generated.

6. The line loss anomaly analysis and operation monitoring system based on big data according to claim 5, characterized in that: The anomaly analysis and handling plan first categorizes anomalies into three types: Category A, Category B, and Category C. Category A anomalies are the most severe, Category B anomalies are severe, and Category C anomalies are minor. Then, three analysis and handling plans are developed. The first plan analyzes and handles Category A anomalies, employing online expert collaboration, graph computing, machine learning / artificial intelligence, edge computing, and collaboration with experienced experts working on the front lines. The second plan analyzes and handles Category B anomalies, also employing online expert collaboration, graph computing, machine learning / artificial intelligence, and edge computing. The third plan handles Category C anomalies, employing graph computing, machine learning / artificial intelligence, and edge computing.

7. The line loss anomaly analysis and operation monitoring system based on big data according to claim 6, characterized in that: The prediction scheme makes predictions based on the established model during implementation, and includes the following steps; (1) Establish a fast mapping relationship between "state-line loss" using a digital twin model. ① Scenario generation: In the digital twin model, a large number of typical operating scenarios are generated in batches by changing the load distribution, generator output, topology and reactive power compensation configuration. ② Perform power flow calculation on each scenario to obtain the accurate theoretical line loss value under the scenario. ③ Build a proxy model: Use the above massive "input scenario parameters-output line loss value" data pairs to train a lightweight machine learning model. (2) Key boundary condition prediction includes ① Load prediction: using historical load, weather, holidays, and event data, predicting the short-term and medium-term load curves for each prediction node / bus. This is the most influential factor. ② New energy power generation prediction: predicting the output curves for photovoltaic, wind power, and other access points. ③ Topology and operation mode planning: obtaining future planned maintenance, switching operations, and network reconfiguration information from OMS and operation plans. ④ Obtaining future temperature, humidity, wind speed, and sunlight, which affect load and equipment parameters, as well as understanding future transaction plans and dispatch instructions. (3) Perform fusion prediction. Input the prediction results of (2) into the proxy model trained in (1) for calculation. ① Determine the spatiotemporal range of the prediction: For example, predict "the line loss rate of the city's power grid every 15 minutes tomorrow". ② Assemble future scenarios: For each prediction time point, assemble a scenario vector according to the predicted load distribution, new energy output and planned topology at that time. ③ Proxy model calculation: Input the scenario vector into the proxy model and output the theoretical technical line loss value (ΔP_tech) at that time. ④ Superimpose the management line loss baseline: Technical line loss is the "calculable" part. For "management line loss", due to its randomness and concealment, it is difficult to predict through simulation. Therefore, the method of baseline value + dynamic adjustment is used for prediction processing. ⑤ Synthesize the final prediction: Total predicted line loss = technical line loss predicted by the proxy model + management line loss prediction value.

8. The line loss anomaly analysis and operation monitoring system based on big data according to claim 1, characterized in that: The learning program trains and optimizes the built model, establishing pass / fail rating levels before training. These levels are Level 1, Level 2, and Level 3. Level 1 has a pass / fail standard of 60% and is the most intensive; Level 2 has a pass / fail standard of 80% and is relatively intensive; and Level 3 has a pass / fail standard of 100% and is of moderate intensity. The program also includes three training analysis schemes: Level 1 corresponds to the first training analysis scheme, which has the highest training intensity and the most comprehensive analysis process. Level 2 corresponds to the second training analysis scheme, which has relatively intensive training and a moderately comprehensive analysis process. Level 3 corresponds to the third training analysis scheme, which has a moderate training intensity and a moderately comprehensive analysis process. Training is conducted at each level, and only after all levels have passed the training can the model be considered qualified and implemented.

9. The line loss anomaly analysis and operation monitoring system based on big data according to claim 8, characterized in that: The platform's sharing and management unit supports data interaction with government departments and power users. Centralized management is achieved through account registration. When government departments or power users use the model, they need to register an account in advance and log in directly upon use. The unit includes a problem feedback and data upload scheme, establishing a tiered monitoring system comprising a basic response monitoring layer, a middle collaborative monitoring layer, and a top collaborative monitoring layer. The basic response monitoring layer consists of several sub-monitoring layers. This layer performs preliminary screening of the data and provides feedback on simple issues. Issues collected by the basic response monitoring layer are then fed back to the middle collaborative monitoring layer, which further processes the received issues. Detailed analysis and processing are conducted, and slightly more complex issues are analyzed and processed. Sensitive and complex issues are transmitted to the top collaborative monitoring layer for review and analysis before processing. Feedback time is set for the basic response monitoring layer and the middle collaborative monitoring layer. If the data collected by the basic response monitoring layer cannot be processed within the specified time, it is directly fed back to the middle collaborative monitoring layer. If the middle collaborative monitoring layer fails to resolve the issue within the specified time, it is directly fed back to the top collaborative monitoring layer. The top collaborative monitoring layer adopts a 24-hour shift system. If any data transmitted to the top collaborative monitoring layer is not processed, adjustments are made to the basic response monitoring layer and the middle collaborative monitoring layer.

Images