Digital twin system for power grid

Abstract

The present application relates to the technical field of power digital twin, and discloses a digital twin system for a power grid. The system comprises: a data access module, which accesses multi-source heterogeneous data of a power grid; a modeling module, which is used for receiving the data output by the data access module and establishing digital twin models in multiple service scenes; a fusion module, which is used for performing association and fusion of the digital twin models in the multiple service scenes; a computing module, which is used for carrying out multi-scene collaborative simulation and derivation by using the digital twin models; and an optimization module, which is used for optimizing the digital twin models, wherein the multi-source heterogeneous data comprises real-time monitoring data, historical operation data, meteorological environment data, device model data, and geographic information data. The problem in the prior art of low fusion efficiency of multi-source heterogeneous data of a power grid is solved. A standardized data access interface is constructed, and association analysis, knowledge graph technology, etc. are applied, so that seamless integration of the multi-source heterogeneous data of the power grid is achieved, thereby improving the data fusion efficiency.

Description

A digital twin system for power grids Technical Field

[0001] The present invention relates to the technical field of power digital twin technology, and more specifically, to a digital twin system of a power grid. Background Art

[0002] With the transformation and upgrading of energy systems, smart grid construction is accelerating globally. Grid digitization and informatization have become crucial tools for power companies to achieve intelligent management and refined operations and maintenance. At the same time, digital twins, as a cutting-edge digital technology, have also garnered significant attention within the power industry. Digital twins, leveraging data, models, algorithms, and computing platforms, achieve a virtual miniature of physical assets, mirroring the real and the virtual. They are considered a crucial approach to the digitalization of complex systems.

[0003] The power grid involves multiple scenarios, including substations and transmission and distribution lines, with complex data types. The semantic expressions of models in different scenarios vary, making collaboration difficult. Existing digital twin systems are mostly limited to a single scenario and data type, and cannot support system-level digital twin applications.

[0004] The Chinese patent application, application number CN202211499751.3, published on April 18, 2023, discloses a converter station equipment monitoring and management system and method based on digital twin technology, including: an operation and maintenance status display module, configured to use digital twin technology to integrate the real-time operation data of the converter station with the three-dimensional model of the entire station to establish a three-dimensional scene of the converter station; an equipment information display module, which realizes the presentation of the primary wiring diagram and underground pipeline information within the station; an equipment status analysis module, which monitors and analyzes the changing trends of the set data and the equipment operation risks; an equipment disposal diagnosis module, which performs fault comparative analysis; the Chinese patent application, application number CN202210514806.7, published on July 29, 2022, discloses a digital twin model construction method based on application scenarios. This method innovatively proposes four main digital twin model application scenarios, and on this basis, specifically proposes four digital twin model construction process methods. However, the above technical solutions have at least the following problems: the data fusion mechanism is simple and cannot effectively integrate the multi-source heterogeneous data of the power grid, resulting in the need to further improve the efficiency of multi-source heterogeneous data fusion in the power grid. Summary of the Invention Technical problems to be solved

[0005] In response to the problem of low efficiency in the fusion of multi-source heterogeneous data in power grids in the existing technology, the present invention provides a digital twin system for power grids, which improves the fusion efficiency of multi-source heterogeneous data in power grids by constructing a standardized data access interface and applying association analysis and knowledge graph technology. Technical Solution

[0006] The purpose of the present invention is achieved through the following technical solutions.

[0007] An embodiment of this specification provides a digital twin system for a power grid, including: a data access module, which accesses multi-source heterogeneous data of the power grid, including real-time monitoring data, historical operation data, meteorological environment data, equipment model data and geographic information data; a modeling module, which receives data from the data access module and establishes a digital twin model suitable for multiple business scenarios; a fusion module, which associates and fuses the digital twin models of multiple business scenarios; and a computing module, which uses the associated and fused digital twin models to perform multi-scenario collaborative simulation.

[0008] Furthermore, the data access module includes: a standard interface unit, which defines a standard service interface for data access through OpenAPI and adopts an asynchronous non-blocking input or output model to improve the response speed of the service interface; a format conversion unit, which converts data of different formats through JSON conversion logic; a semantic conversion unit, which converts different semantic data based on the word vector of the improved Word2Vec model, and the improved Word2Vec model is a model that adds semantic constraints to optimize the word vector representation; a latitude conversion unit, which converts data of different latitudes by combining the geodetic coordinate system conversion based on positive calculation and the coordinate measurement based on satellite navigation; a middleware unit, which uses an optimized distributed message queue to access multi-source heterogeneous data, and the optimization is through intelligent routing diversion and dynamic resource scheduling.

[0009] Furthermore, the middleware unit includes: an association sub-unit, which adopts a dependency analysis method based on causal model inference to establish causal dependencies between multi-source heterogeneous data; a routing sub-unit, which performs intelligent routing and diversion based on the causal dependencies between multi-source heterogeneous data to carry out business association between multi-source heterogeneous data; a resource sub-unit, which adjusts the resource queue length by changing the number of queue partitions based on the real-time monitoring results of data traffic, and changes the resource queue width by adjusting the number of threads in the consumer thread pool.

[0010] Furthermore, the modeling module includes: a basic model construction unit, which constructs a basic resource model of the power grid based on the topological structure of the power grid using the power system steady-state analysis method; a digital twin model unit, which uses the basic resource model to construct a digital twin model through visual editing tools according to different business scenarios; a model library unit, which uses a distributed model storage mechanism to store different types of preset model structures for call by the digital twin model unit; a parameter fitting unit, which fits the parameters of the digital twin model based on the federated learning algorithm and monitoring data; a model management unit, which stores each digital twin model in the model library and encapsulates the digital twin model into a service for publication.

[0011] Furthermore, the fusion module includes: an association analysis unit, which uses mutual information theory and association rule algorithm to obtain the association relationship between different multi-source heterogeneous data; a rule fusion unit, which uses time rules, space rules and business rules to perform cross-domain feature mapping and relationship matching based on the acquired association relationship between multi-source heterogeneous data, and generate a matching mapping table between heterogeneous data; a model fusion unit, which uses inference algorithm and sequence pattern matching technology according to the matching mapping table to fuse digital twin models based on multi-source heterogeneous data; a model collaboration unit, which constructs a model association diagram between different businesses based on the association relationship and the fused digital twin model; based on the model association diagram, extracts the interface mapping relationship of the digital twin model, and performs collaborative simulation of digital twin models of different businesses through interface mapping relationship and message passing.

[0012] Furthermore, the computing module includes: a collaborative computing unit, which utilizes the interface mapping relationship of the digital twin model to perform distributed collaborative simulation calculations on the associated and fused digital twin models through parallel computing and distributed communication technology; an output aggregation unit, which uses heterogeneous information serialization and encoding technology to generate output data containing the collaborative simulation calculation results of multiple digital twin models; a simulation analysis unit, which constructs a prediction model based on a deep learning algorithm, and uses the prediction model to analyze the calculation results of the collaborative simulation under different control parameter settings; a simulation control unit, which uses the policy gradient method of the reinforcement learning algorithm to adjust the control parameters of the digital twin model according to the analysis results of the collaborative simulation calculation results.

[0013] Furthermore, the simulation control unit includes: a strategy construction subunit, which uses the policy gradient method in the reinforcement learning algorithm to construct a control strategy model for collaborative simulation of the digital twin model; a parameter determination subunit, which uses the prediction model constructed by the simulation analysis unit to select the top N control parameters with the greatest impact on the collaborative simulation calculation accuracy and recall rate as key control parameters based on the collaborative simulation results under different control parameter settings; a model training subunit, which uses the selected key control parameters to train the control strategy model in the digital twin model collaborative simulation environment to obtain the optimal control strategy; and a model application subunit, which uses the optimal control strategy obtained through training to adjust the parameters of the digital twin model.

[0014] Furthermore, the output aggregation unit includes: a data parsing sub-unit, which combines named entity recognition and dependency syntax analysis to extract text data from the collaborative simulation calculation results of different digital twin models, and converts it into structured data to generate first output data; an encoding sub-unit, which uses a compression coding algorithm based on sparse representation to encode the first output data to generate second output data; a relationship extraction sub-unit, which uses knowledge graph technology to extract the association relationship between the second output data; and a relationship fusion sub-unit, which associates the second output data according to the extracted association relationship to generate the final output data.

[0015] Furthermore, the compression coding algorithm based on sparse representation includes: using a dictionary learning method to obtain an overcomplete dictionary of the power grid monitoring data sample set; applying an orthogonal matching pursuit algorithm to the first output data to obtain the sparse representation coefficients of the first output data; retaining the non-zero elements in the sparse representation coefficients to generate the second output data.

[0016] Furthermore, the digital twin system of the power grid also includes an edge computing module, which performs local pre-processing on multi-source heterogeneous data connected to the power grid; the edge computing module includes: an edge node unit, which is arranged in the converter at the power grid end, and collects the sensor data of the power grid through the analog-to-digital conversion module; the edge computing unit performs local pre-processing on the collected sensor data, and the pre-processing adopts a coding technology based on a lossless compression algorithm to remove redundant information in the data, and uses a low-bit first transmission mechanism to increase the data transmission rate; the edge communication unit transmits the pre-processed sensor data to the computing module using the 5G communication network. Beneficial effects

[0017] The beneficial effects that may be brought about by the digital twin system of a power grid provided in the embodiments of this specification include at least:

[0018] (1) Standard interfaces and data formats eliminate the differences between heterogeneous data. Standardized interfaces and data formats enable heterogeneous data to be seamlessly associated and integrated, avoiding repeated data processing steps and improving integration efficiency.

[0019] (2) Utilizing data from different business areas to realize cross-scenario digital twins, it expands the scope of data fusion, avoids data silos, and improves data fusion efficiency;

[0020] (3) Large-scale collaborative simulation requires a large amount of heterogeneous data support. Cloud computing technology improves the integration throughput of heterogeneous data, realizes efficient simulation operations, and improves data fusion efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The specification will be further described in the form of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting, and in these embodiments, the same numbers represent the same structures, wherein:

[0022] FIG1 is an exemplary module diagram of a digital twin system of a power grid according to some embodiments of this specification;

[0023] FIG2 is a schematic diagram of a data access module according to some embodiments of this specification;

[0024] FIG3 is a schematic diagram of a modeling module according to some embodiments of the present specification;

[0025] FIG4 is a schematic diagram of a fusion module according to some embodiments of this specification;

[0026] FIG5 is a schematic diagram of a computing module according to some embodiments of this specification;

[0027] FIG6 is a schematic diagram of an edge computing module according to some embodiments of the present specification.

[0028] Explanation of the numbers in the figure: 100, a digital twin system of a power grid; 110, data access module; 120, modeling module; 130, fusion module; 140, computing module; 150, edge computing module; 111, standard interface unit; 112, format conversion unit; 113, semantic conversion unit; 114, latitude conversion unit; 115, middleware unit; 121, basic model construction unit; 122, digital twin unit; 123, model library unit; 124, parameter fitting unit; 125, model management unit; 131, association analysis unit; 132, rule fusion unit; 133, model fusion unit; 134, model collaboration unit; 141, collaborative computing unit; 142, output aggregation unit; 143, simulation analysis unit; 144, simulation control unit; 151, edge node unit; 152, edge computing unit; 153, edge communication unit. DETAILED DESCRIPTION

[0029] The system provided in the embodiments of this specification is described in detail below with reference to the accompanying drawings.

[0030] The GreenOS power digital twin operating system focuses on three business areas: power grid equipment, dispatching, and marketing. It leverages advanced technologies such as IoT perception, digital-analog fusion, simulation and deduction, cloud-edge collaboration, and virtual-reality iteration. Through the GreenOS power digital twin operating system and the development of digital twin application scenarios, it creates a digital twin power grid solution that features real-time state perception, accurate model fidelity, system self-evolution, intelligent management empowerment, and precise decision-making execution. By tackling key digital twin technologies and iteratively deepening digital twin application scenarios, the company continuously improves its capabilities for enhanced perception, cognition, and decision-making in new power systems, addressing the common needs of new power systems for comprehensive data collection, measurement, and control, real-time simulation and calculation, and intelligent deduction and optimization, thereby promoting the digital and intelligent transformation of power grids.

[0031] Figure 1 is an exemplary module diagram of a digital twin system 100 of a power grid according to some embodiments of this specification. As shown in Figure 1, a digital twin system 100 of a power grid includes: a data access module 110, a modeling module 120, a fusion module 130 and a computing module 140.

[0032] Data access module 110: This module acquires multi-source heterogeneous data from the power grid into the system through the data access unit. This data includes real-time monitoring data, historical operating data, meteorological and environmental data, equipment model data, and geographic information data. This data is transmitted and stored through different interfaces and formats. The standard interface unit 111 in data access module 110 uses OpenAPI to define a standard service interface for data access and adopts an asynchronous non-blocking input or output model to improve the response speed of the service interface. The format conversion unit 112 and semantic conversion unit 113 convert data of different formats and semantics into a unified data format and semantic standard for subsequent processing and analysis.

[0033] Among them, the modeling module 120: the modeling module 120 receives the data provided by the data access module 110, and uses this data to establish a digital twin model suitable for multiple business scenarios. The basic model construction unit 121 uses the power system steady-state analysis method to construct a basic resource model of the power grid based on the topological structure of the power grid. This basic model provides the basic structure and parameter information of the power grid. The digital twin model unit uses the basic resource model to build a digital twin model through a visual editing tool, including modeling of equipment, lines, transformers, etc. The model library unit 123 is responsible for storing and managing different types of preset model structures for the digital twin model unit to call. The parameter fitting unit 124 uses the federated learning algorithm and monitoring data to fit the parameters of the digital twin model so that the model can better adapt to the actual situation.

[0034] Among them, fusion module 130: fusion module 130 is responsible for the association and fusion of digital twin models of multiple business scenarios. The association subunit uses a dependency analysis method based on causal model inference to establish causal dependencies between multi-source heterogeneous data. These causal dependencies reflect the mutual influence relationship between different data. The routing subunit performs intelligent routing diversion based on causal dependencies, and associates different multi-source heterogeneous data for business purposes for subsequent processing and analysis. The resource subunit adjusts the resource queue length by changing the number of queue partitions based on the real-time monitoring results of data traffic, and changes the resource queue width by adjusting the number of threads in the consumer thread pool, thereby optimizing resource allocation and utilization efficiency.

[0035] Among them, the calculation module 140: the calculation module 140 uses the associated and fused digital twin model to perform multi-scenario collaborative simulation calculations. The collaborative calculation unit 141 uses the interface mapping relationship and parallel computing technology of the digital twin model to assign computing tasks to different computing nodes for parallel computing to improve computing efficiency. The output aggregation unit 142 aggregates the calculation results of each computing node to generate output data containing the collaborative simulation calculation results of multiple digital twin models. The simulation analysis unit 143 constructs a prediction model based on the deep learning algorithm, and uses the prediction model to analyze the collaborative simulation calculation results under different control parameter settings to provide prediction and analysis of the power grid operation status. The simulation control unit 144 uses the policy gradient method of the reinforcement learning algorithm to adjust the control parameters of the digital twin model according to the analysis results of the collaborative simulation calculation results, and optimize the operation strategy and decision of the power grid.

[0036] Among them, the edge computing module 150 (preferred embodiment) performs local preprocessing of heterogeneous multi-source data entering the power grid to improve processing efficiency and reduce data transmission requirements. It includes an edge node unit 151, located in a converter at the power grid end, which collects power grid sensor data through an analog-to-digital conversion module. The edge computing unit 152 preprocesses the collected sensor data and removes redundant information using a low-order bit first transmission mechanism and encoding technology based on a lossless compression algorithm. The edge communication unit 153 transmits the preprocessed sensor data to the computing module 140 via the 5G communication network.

[0037] In summary, the digital twin system of this power grid, through the combination of data access, modeling, fusion, and computing module 140, achieves the fusion of multi-source heterogeneous data and multi-scenario collaborative simulation computing in the power grid, which can improve the operational efficiency and reliability of the power grid. The addition of edge computing module 150 further optimizes the efficiency of data processing and transmission.

[0038] Figure 2 is a schematic diagram of the data access module 110 shown in some embodiments of this specification. As shown in Figure 2, the data access module 110 includes: a standard interface unit 111, a format conversion unit 112, a semantic conversion unit 113, a latitude conversion unit 114 and a middleware unit 115.

[0039] Standard interface unit 111 defines a standard service interface for data access through OpenAPI, enabling all data sources to communicate with data access module 110 using a standardized method. Adopting an asynchronous, non-blocking input or output model improves the service interface's response speed and ensures efficient data transmission.

[0040] Specifically, OpenAPI is an open standard interface specification that defines a set of rules and conventions for describing the communication methods between software applications or services. It provides a unified method that enables different data sources or applications to interact with the data access module 110 in a consistent manner. OpenAPI can use common data exchange formats such as JSON or XML to define the transmission format of data. By using OpenAPI, the standard interface unit 111 can expose the functions of the data access module 110 to external data sources and ensure that the data sources communicate through a unified standardized method. The advantage of this is that each data source can use the same request and response structure, regardless of its internal implementation details, to maintain interoperability with the data access module 110.

[0041] Specifically, the asynchronous non-blocking input or output model is a communication model that aims to improve the response speed of the service interface and the efficiency of data transmission. In this model, the input and output of data are non-blocking, that is, when data is transmitted, the sender and the receiver can continue to perform other operations without waiting for the other party to complete. Asynchronous means that the sender and the receiver can perform other operations at the same time without having to wait for the other party's response. Non-blocking means that the input and output process of data will not block the execution of the application. This model is suitable for high-concurrency and fast-response scenarios and can improve the throughput and performance of the system. By adopting the asynchronous non-blocking input or output model, the standard interface unit 111 can process multiple requests in an efficient manner and maintain concurrent communication with various data sources. This can improve the response speed of the service interface while ensuring efficient data transmission, providing an optimized mechanism for the overall performance of the system and data exchange efficiency.

[0042] Among them, the format conversion unit 112: This unit is responsible for processing the conversion between data of different formats. By using the conversion logic of JSON, data of different formats are converted into a unified JSON format for subsequent processing and analysis. This can eliminate the differences between data of different formats, making it easier to process the data in a unified manner. Among them, the semantic conversion unit 113: This unit converts between different semantic data based on the word vectors of the improved Word2Vec model. By mapping different semantic data into word vector representations, calculations and comparisons can be performed in the vector space, thereby achieving conversion and fusion between different semantic data. This can improve the semantic consistency and comparability of the data, allowing the data to be better associated and analyzed. Among them, the latitude conversion unit 114: This unit is responsible for processing the conversion between data of different latitudes. By combining the geodetic coordinate system conversion based on positive calculation and the coordinate measurement based on satellite navigation, the coordinate conversion between data of different latitudes can be achieved. This allows data of different latitudes to be compared and fused under the same coordinate system, improving the consistency and visualization of the data.

[0043] Middleware unit 115 utilizes an optimized distributed message queue to access multi-source heterogeneous data. Through intelligent routing and dynamic resource scheduling, data traffic can be distributed to different processing nodes, and resources can be rationally allocated to improve data access efficiency and processing capabilities. This reduces system load and improves data processing efficiency and response speed. Specifically, middleware unit 115 includes an association subunit, a routing subunit, and a resource subunit. The association subunit utilizes a dependency analysis method based on causal model inference to establish causal dependencies between multi-source heterogeneous data. By analyzing the causal relationships between data, the dependencies and impact relationships between data can be determined. This helps identify data relevance, enabling better data fusion and business integration. The routing subunit performs intelligent routing and diversion based on the causal dependencies between multi-source heterogeneous data, enabling business integration between multi-source heterogeneous data. By understanding the dependencies between data, the data transmission path and sequence can be determined, allowing data to be distributed to the appropriate processing nodes. This avoids unnecessary data transmission and processing, improving data processing efficiency and response speed. Resource Subunit: Based on real-time data traffic monitoring, this subunit adjusts resource queue lengths by changing the number of queue partitions and resource queue widths by adjusting the number of threads in the consumer thread pool. This allows for dynamic resource allocation and utilization based on real-time data traffic. By properly adjusting resource queue lengths and thread pool sizes, resource waste and bottlenecks can be avoided, improving data processing efficiency and system performance.

[0044] These middleware units 115 collaborate to implement data correlation analysis, intelligent routing, and resource adjustment. The correlation subunit establishes causal dependencies between data and provides information about their relevance. The routing subunit intelligently routes data based on these relationships, ensuring data flows to the target processing nodes. The resource subunit adjusts resources based on real-time data flow monitoring to optimize data processing capabilities and system performance. Through this collaborative approach, the middleware units 115 effectively improve the efficiency of integrating heterogeneous data from multiple sources in the power grid.

[0045] In this embodiment, the data access module 110 uses a RESTful API-style interface design, defining standard interfaces for accessing real-time data, historical data, and device data. It uses the HTTP / 2 protocol to support asynchronous, non-blocking, and concurrent processing of requests from multiple data sources. An OpenAPI specification file is defined, allowing different systems to access data using the same interface standard. A conversion tool is implemented to support multiple data formats, such as CSV, XML, and Protocol Buffers, to JSON. Libraries such as Jackson and Gson are used to parse complex formats, and asynchronous, non-blocking I / O is used to convert data of various formats to JSON. A format validation tool is also developed to ensure the standardization of the converted JSON data. Based on the Word2Vec model, word embeddings are first trained on a large-scale corpus of data from the power industry. A power system-specific dictionary is then added, and weights are refined to construct a domain-specific semantic space. Semantic relationships within the word embeddings are used to map and transform different terms, outputting standardized vocabulary. A combined call to the Java Geospatial Toolkit (GeoTools) and the GPS Toolkit (JGPS) is implemented. GeoTools performs forward calculations and conversions, while JGPS obtains real-time satellite positioning data. Dynamically select a conversion scheme based on the different coordinate systems of the data source to output a unified latitude and longitude representation. A distributed messaging system based on Kafka is constructed, with topic partitions divided into internal power grid enterprise regions, dispatch instructions, and equipment monitoring topics. Data generation rates are monitored, and the number of partitions is dynamically adjusted to balance the load. Data associations are identified based on message keywords, and related data is routed to the same consumer group. By controlling consumer threads, server load is balanced to achieve resource scheduling. A causal model inference algorithm based on Bayesian networks is used to construct a directed acyclic graph representing the dependencies between variables. Variable entities are identified based on multidimensional text data from the power grid, and their conditional dependency probabilities are learned. This allows the causal relationships between different data sources to be derived, creating a dependency graph for heterogeneous data. Based on this constructed heterogeneous data dependency graph, a graph search algorithm is used to find the shortest dependency paths between data sources. Based on the distance between dependencies, related heterogeneous data is routed to the same Kafka topic partition, achieving business-related data aggregation. The Kafka Consumer consumer group management module monitors message accumulation in each partition in real time. When accumulation reaches a threshold, the number of partitions for this type of data is dynamically increased, thereby expanding the queue length. It also monitors the processing rate of consumer threads, dynamically adjusts the thread pool size, and controls queue width. This allows the association subunit to analyze the dependencies between heterogeneous data, the routing subunit to distribute data based on these dependencies, and the resource subunit to continuously monitor and optimize resources. These three elements work together to improve the efficiency of heterogeneous data fusion and association in the power grid.

[0046] In summary, through the coordinated and collaborative work of various units, the data access module 110 efficiently handles the access and integration of multi-source heterogeneous power grid data. The standard interface unit 111 provides a standardized interface and efficient communication methods. The format conversion unit 112 and semantic conversion unit 113 ensure the consistency of data with different formats and semantics. The dimension conversion unit 114 enables unified processing of data at different dimensions. The middleware unit 115 optimizes data access and distribution. This collaborative work enables better data integration and analysis.

[0047] Figure 3 is a schematic diagram of the modeling module 120 shown in some embodiments of this specification. As shown in Figure 3, the modeling module 120 includes: a basic model construction unit 121, a digital twin model unit, a model library unit 123, a parameter fitting unit 124 and a model management unit 125.

[0048] The basic model construction unit 121 is primarily responsible for abstracting various power equipment within the power grid into mathematical models based on the power grid topology and establishing the connections between them. These equipment include generators, transformers, and transmission lines. Using power system steady-state analysis methods, the basic model construction unit 121 considers the electrical characteristics, operating status, and interconnections of the power equipment to construct a basic resource model for the power grid. Before constructing the basic resource model, data collection and processing are required. This data includes the physical parameters, operating status, and connectivity of the power equipment. Data collection can be performed through field measurements, monitoring systems, or other reliable data sources. The collected data undergoes quality checks and preprocessing to ensure accuracy and usability. The basic model construction unit 121 abstracts the power equipment into mathematical models. Different types of equipment require different modeling methods. For example, an equivalent circuit model or a power-voltage characteristic curve model can be used to describe its electrical characteristics. Transformers can be modeled using a parametric model or an impedance matrix model. Transmission lines can be described using parameters such as resistance, reactance, and conductance. The appropriate modeling method should be selected based on the characteristics of the equipment and actual needs. In a power grid, various power devices have complex connections. The basic model building unit 121 analyzes the topology of the power grid to establish the connection relationships between power devices. These connection relationships can be represented using a node admittance matrix, a branch impedance matrix, or other topological analysis methods. By establishing these connection relationships, it is possible to describe the current, voltage, and power transmission relationships between various devices in the power grid.

[0049] The digital twin model unit utilizes the basic resource model and, based on different business scenarios, uses visual editing tools to construct digital twin models. A digital twin model is a virtual simulation of the actual operation of a power grid. The model can be designed and customized according to different business needs. The digital twin model can be used to understand the dynamic operating status of the power grid and predict future operations. The model library unit 123 utilizes a distributed model storage mechanism to store different types of preset model structures for use by the digital twin model unit. The model library contains a variety of predefined model structures, allowing users to select the appropriate model structure as needed. This improves model reusability and development efficiency. The parameter fitting unit 124 fits the parameters of the digital twin model based on the federated learning algorithm and monitoring data. Using monitoring data and the federated learning algorithm, the parameters of the digital twin model can be optimized and adjusted to better align the model with actual conditions, improving its accuracy and predictive capabilities. The model management unit 125 is responsible for storing each digital twin model in the model library and packaging the digital twin models into services for publishing. The model management unit 125 manages the digital twin model in a unified manner, including storage, version management, and publishing. By encapsulating the digital twin model into a service, it can be easily called and used by other systems or applications.

[0050] In this embodiment, the modeling module 120 collects grid topology data and performs steady-state analysis using MATPOWER software to obtain node impedance parameters, line parameters, and other information. This information is used to construct a node relationship diagram for the basic grid model, representing the grid's connectivity structure, transmission network, and other aspects. A browser-based visual model editor has been developed, providing drag-and-drop interface components for editing digital twin models of grid transmission, substations, and power consumption. The editor can quickly bind component input and output parameters and generate model execution code. Users can flexibly build models for different business scenarios. Based on the Neo4j graph database, various preset grid model templates are stored. Templates contain model structure information, parameter configurations, and other information. Templates have version management capabilities and support iterative upgrades. Digital twin model units can query and call templates in the model library for rapid modeling. A CNN deep learning model for grid device parameter detection is trained using a federated learning algorithm to protect device data privacy. Detection data is fed into the model to iteratively update parameters, continuously optimizing the model's detection accuracy and the digital twin model's fit. The constructed digital twin model is registered in the model library for unified ID management and access control. The model is encapsulated as a microservice using the Spring Cloud framework and published via a RESTful interface. Other modules can then easily call the services of related models for simulation calculations.

[0051] In summary, the various units of modeling module 120 collaborate to implement functions such as building a basic model, customizing and optimizing the digital twin model, and storing and managing the model. The basic model construction unit 121 provides accurate foundational data for the digital twin model; the digital twin model unit constructs the digital twin model based on the basic model; the model library unit 123 provides preset model structures for use by the digital twin model; the parameter fitting unit 124 optimizes model parameters using a federated learning algorithm and monitoring data; and the model management unit 125 centrally manages and publishes the digital twin model. Through this collaborative work, modeling module 120 effectively improves the efficiency of integrating heterogeneous data from multiple sources in the power grid.

[0052] FIG4 is a schematic diagram of a fusion module 130 according to some embodiments of this specification. As shown in FIG4 , the fusion module 130 includes: an association analysis unit 131 , a rule fusion unit 132 , a model fusion unit 133 and a model collaboration unit 134 .

[0053] The association analysis unit 131 uses mutual information theory and association rule algorithms to analyze the associations between different multi-source heterogeneous data. By analyzing the correlation and dependency between statistical data, it can discover hidden associations between data, providing a basis for subsequent fusion.

[0054] Among them, the rule fusion unit 132: This unit uses temporal rules, spatial rules, and business rules to perform cross-domain feature mapping and relationship matching based on the associations between the acquired multi-source heterogeneous data. By formulating unified rules, it can match and map features from different data sources, generating a matching mapping table between heterogeneous data.

[0055] Specifically, temporal rules. Power grid data often contains temporal attributes, such as equipment monitoring time and alarm time. Temporal rules refer to the association of data based on temporal characteristics when matching heterogeneous data sources. For example, voltage data and alarm data monitored at the same time can be associated, reflecting the monitoring of the same event by different systems. Spatial rules. Power grid data also contains location / spatial attributes, such as device numbers and installation locations. Spatial rules refer to matching based on spatial characteristics. For example, load data and line data in the same area can be associated, reflecting the modeling of the same power grid area by different systems. Business rules refer to data association based on power grid business knowledge. For example, upstream and downstream current and voltage data can be associated using power grid topology connectivity. Alternatively, corresponding relationships between heterogeneous data sources can be mapped based on the business flow logic between devices. By considering the three dimensions of time, space, and business, semantic associations between heterogeneous data can be more comprehensively determined, guiding effective data integration and facilitating the construction of digital twin systems.

[0056] Specifically, cross-domain feature mapping refers to mapping or converting data from different sources and their corresponding features in the rule fusion unit 132 so that they can be effectively compared, associated, or analyzed during the fusion process. When multiple different data sources or data domains are involved, each data source may have a different data structure and feature representation. The goal of cross-domain feature mapping is to map these different features into a unified feature space for further fusion and processing.

[0057] Specifically, inference algorithms can represent digital twin models of different business systems in the power grid as knowledge graphs. Rule-based inference algorithms (such as RDFS and OWL) are then applied to infer relationships between models and merge model attributes. For example, the "voltage" attribute in the substation model and the line model can be determined through inference to be the same concept, and the attribute characteristics of the two models can then be unified. Sequential pattern matching technology: The operation of digital twin models in the power grid generates various time series data. Sequential pattern matching algorithms can be used to analyze the behavioral sequences of different models and discover similar patterns between them. For example, usage patterns can be found in data sequences from line simulations and actual operations. This can be used to guide the fusion of models for similar business scenarios, resulting in more accurate models. In summary, inference algorithms can infer model relationships, while sequential pattern matching can determine model behavioral relationships. The combination of these two can better integrate heterogeneous models and improve the performance of digital twin systems.

[0058] Specifically, the model association graph is a knowledge graph that represents the relationships between digital twin models in different business domains. Nodes in the graph represent model entities, and edges represent the types of associations between models. Constructing such a graph can intuitively reflect the connections between digital twin models across different power grid systems and businesses, facilitating model fusion. Interface mappings refer to the correspondence between input and output variables and event triggers between different models. For example, the "voltage" output of a substation model is mapped to the "voltage" input of a line model. Defining such interface mappings enables interoperability between models, parameter and event transfer, and model collaborative simulation. Model collaborative simulation involves enabling models from different business domains to run in parallel and collaborate, passing parameters and events to each other, and jointly completing multi-scenario simulation tasks based on clear model associations and interface mappings. This is crucial for joint simulation and fault prediction in power grids. Therefore, the model association graph intuitively reflects model relationships, and interface mappings enable model interconnection, ultimately enabling collaborative simulation of different business models. This effectively addresses the issues of linking and integrating heterogeneous data in power grids.

[0059] Among them, the model fusion unit 133: This unit uses the inference algorithm and sequence pattern matching technology according to the matching mapping table to fuse the digital twin model based on multi-source heterogeneous data. By fusing information from different data sources into the digital twin model, the integrity and expression ability of the model are improved, making it closer to the actual situation. Among them, the model collaboration unit 134: This unit constructs a model association diagram between different businesses based on the association relationship and the fused digital twin model. The model association diagram shows the relationship and dependency between different businesses, which can help understand the mutual influence of different businesses. According to the model association diagram, the interface mapping relationship of the digital twin model is extracted, and the collaborative simulation of the digital twin models of different businesses is carried out through the interface mapping relationship and message passing.

[0060] In this embodiment, the fusion module 130 uses the mutual information theory formula to calculate the mutual information between different data sources, and extracts sources with mutual information greater than a threshold as associated sources. The Apriori algorithm is used to learn the association rules between data sources and obtain rules that meet the threshold. Time, space and business metadata are defined, and the Jaccard similarity algorithm is used to match the metadata of heterogeneous data sources. A matching mapping table is generated between heterogeneous sources to represent the corresponding relationship between their business entities. Turtle files of different models are loaded, and RDFS / OWL reasoning is performed using the Jena reasoning framework to merge the attribute features of the same entities between models. A sequential pattern matching algorithm is used to identify event-driven model behavior sequences and merge models with similar behavior patterns. An RDF graph of the associated model is constructed, and SPARQL is used to query the input and output variables of the model interface. An interface mapping table is generated to drive the parameter and event transmission between models for collaborative simulation.

[0061] In summary, the various units of fusion module 130 work together through association analysis, rule fusion, model fusion, and model collaboration to achieve the fusion of multi-source heterogeneous power grid data. Association analysis unit 131 discovers associations between data, rule fusion unit 132 performs feature mapping and relationship matching, model fusion unit 133 integrates data into the digital twin model, and model collaboration unit 134 constructs a model association diagram and implements collaborative simulation of different services. Through this collaborative approach, fusion module 130 effectively improves the efficiency of fusion of multi-source heterogeneous power grid data.

[0062] FIG5 is a schematic diagram of a computing module 140 according to some embodiments of this specification. As shown in FIG5 , the computing module 140 includes: a collaborative computing unit 141 , an output aggregation unit 142 , a simulation analysis unit 143 , and a simulation control unit 144 .

[0063] Specifically, collaborative computing unit 141 utilizes the interface mapping relationships of the digital twin models, parallel computing, and distributed communication technologies to perform distributed collaborative simulation calculations on the associated and fused digital twin models. By allocating computing tasks to multiple computing nodes and utilizing distributed communication for data exchange and collaborative computing, simulation speed can be accelerated and computational efficiency improved.

[0064] In this embodiment, the collaborative computing unit 141 is distributedly deployed based on the Spark cluster, taking advantage of Spark's good parallel computing capabilities and scalability. The interface mapping relationship of different models is represented as an RDD distributed data set and registered as a temporary table for computing task queries. Spark SQL is used to shard the computing tasks, and tasks are assigned to different nodes based on the model dependencies. A timeout retry mechanism is used to ensure task execution. ZeroMQ is used to implement data exchange between computing nodes. Point-to-point low-latency message publishing and subscription are performed based on the interface mapping relationship. Tools such as Ganglia are used to monitor node loads, implement dynamic resource scheduling of the scheduler, and maintain cluster load balancing. Through this specific implementation, Spark's distributed computing engine can be fully utilized to perform parallel collaborative simulation calculations of digital twin models, thereby improving the efficiency of power grid simulation. At the same time, the consistency, scalability, and fault tolerance of the calculations are also guaranteed.

[0065] Among them, the output aggregation unit 142: This unit adopts heterogeneous information serialization and encoding technology to generate output data containing the collaborative simulation calculation results of multiple digital twin models. By aggregating the calculation results of multiple digital twin models and performing appropriate serialization and encoding, unified output data can be generated, which is convenient for subsequent analysis and application. The output aggregation unit 142 includes: a data parsing subunit, an encoding subunit, a relationship extraction subunit and a relationship fusion subunit. Specifically, the data parsing subunit: This unit combines named entity recognition and dependency syntax analysis to extract text data from the collaborative simulation calculation results of different digital twin models, and converts it into structured data to generate the first output data. Named entity recognition is used to identify entity information in the text, such as place, time, etc., while dependency syntax analysis is used to analyze the dependency relationship between different words in a sentence, so as to obtain sentence structure and semantic information. Specifically, the encoding subunit: This unit adopts a compression coding algorithm based on sparse representation to encode the first output data to generate the second output data. Sparse representation is a method of representing data by finding the optimal linear combination to represent the original data, thereby achieving data dimensionality reduction and compression. Sparse representation encoding algorithms can reduce data redundancy and noise, extract key features of the data, and retain important information in the data. Specifically, the relationship extraction subunit: This unit uses knowledge graph technology to extract the association relationships between data from the second output data. Knowledge graph is a method of representing knowledge in a graph structure, where nodes represent entities and edges represent relationships between entities. By performing semantic analysis and association rule mining on the second output data, association relationships between data, such as causal relationships and similarity relationships, can be extracted. Specifically, the relationship fusion subunit: This unit associates the second output data based on the extracted association relationships to generate the final output data. By fusing and integrating the association relationships between different data, a complete connection between the data can be established, thereby obtaining more comprehensive and accurate data results.

[0066] The various subunits of output aggregation unit 142 collaborate through data parsing, encoding, relationship extraction, and relationship fusion to improve the efficiency and technical effectiveness of fusion of multi-source heterogeneous power grid data. The data parsing subunit extracts text data and converts it into a structured form. The encoding subunit performs data compression and feature extraction. The relationship extraction subunit extracts relationships between data, and the relationship fusion subunit integrates these relationships into the final output data. Through this collaborative work, output aggregation unit 142 can effectively improve the efficiency of data fusion.

[0067] Specifically, compression coding algorithms based on sparse representation primarily include technical features such as dictionary learning, orthogonal matching pursuit, and non-zero element extraction. Dictionary learning is the first key step in compression coding algorithms. Its goal is to learn an overcomplete dictionary so that the atoms (basis) in the dictionary can best represent the power grid monitoring data sample set. Such dictionaries typically have a certain degree of redundancy, meaning that the bases in the dictionary are not linearly independent, to better represent the data. Dictionary learning can be performed based on a training sample set through iterative optimization methods to learn the optimal dictionary to represent the characteristics of power grid data.

[0068] Orthogonal matching pursuit is a key step in compression coding algorithms. It is used to calculate the sparse representation coefficients of the first output data on an overcomplete dictionary. The orthogonal matching pursuit algorithm gradually projects the input data onto the dictionary basis and uses orthogonalization conditions to find the optimal representation. It iteratively searches for the representation coefficients, making them as sparse as possible while accurately reconstructing the original data.

[0069] Non-zero element extraction is the last step of the compression coding algorithm. After obtaining the sparse representation coefficients of the first output data, the non-zero elements therein are retained and used as the second output data. This step is essentially a method of data sparsification and feature selection. By retaining important sparse representation coefficients, the redundancy of the data is reduced and the key features of the data are extracted. These technical features work together to improve the fusion efficiency and technical effects of multi-source heterogeneous data in the power grid. Dictionary learning provides an optimized over-complete dictionary that can better represent the characteristics of power grid data. The orthogonal matching pursuit algorithm uses this dictionary to calculate the sparse representation coefficients, thereby capturing important information of the data. Finally, the non-zero element extraction stage generates the second output data by retaining the non-zero elements in the sparse representation coefficients, reducing the redundancy of the data and extracting key features.

[0070] By combining these technical features, the sparse representation-based compression coding algorithm can effectively improve the fusion efficiency of multi-source heterogeneous power grid data. It can reduce the data dimension, extract key features, and retain important information.

[0071] In this embodiment, the output aggregation unit 142 uses Stanford CoreNLP to perform dependency parsing and named entity recognition, extracting structured information such as device entities and status descriptions from the simulation text results, and converting them into JSON formatted data. The power grid monitoring dataset is trained using Scikit-learn's dictionary learning algorithm to obtain an overcomplete dictionary. The parsed JSON result data is sparsely encoded using the orthogonal matching pursuit algorithm in the SPAMS package to obtain sparse representation coefficients. A power grid knowledge graph is constructed, and the LINE model is used to learn the embedding vector representation of the knowledge graph. The relationships between device entities are measured based on the distance in the embedding vector space, and an association table is output. Based on the extracted entity association relationships, the encoded data is associated and fused. The combined simulation output results of different power grid business systems are output. Lasso regression is used for model compression to extract feature factors that have a significant impact on the output results, thereby improving model parsing efficiency. By integrating technologies such as entity extraction, sparse coding, and knowledge graphs, the heterogeneous output aggregation problem of power grid digital twins can be effectively solved, enabling the association and fusion of multi-source data results.

[0072] Among them, the simulation control unit 144 uses the policy gradient method of the reinforcement learning algorithm to adjust the control parameters of the digital twin model according to the analysis results of the collaborative simulation calculation results. This unit builds a prediction model based on the deep learning algorithm, and uses the prediction model to analyze the calculation results of the collaborative simulation under different control parameter settings. By performing deep learning analysis on the simulation calculation results, the associations and patterns between the data can be revealed, thereby providing more detailed and accurate simulation analysis results. The simulation control unit 144 includes: a policy construction subunit, a parameter determination subunit, a model training subunit and a model application subunit. Specifically, the policy construction subunit: This subunit uses the policy gradient method in the reinforcement learning algorithm to construct a control policy model based on the digital twin model collaborative simulation environment. Through interaction with the simulation environment, the policy construction subunit gradually optimizes the control strategy to maximize the expected reward signal. This subunit generates a control strategy based on the current state and optional actions, providing a basis for subsequent model training and application.

[0073] Specifically, the parameter determination subunit analyzes the co-simulation results under different control parameter settings based on the prediction model constructed by the simulation analysis unit 143. It evaluates the impact of each control parameter on the co-simulation calculation accuracy and recall rate and selects the top N control parameters with the greatest impact on the results as key control parameters. The purpose of this subunit is to determine the key adjustment parameters of the control strategy for subsequent model training and optimization.

[0074] More specifically, the evaluation results are sorted according to the degree of influence, and the top N control parameters with the greatest impact on the collaborative simulation calculation accuracy and recall rate are selected as key control parameters. Professional data analysis and sorting tools, such as the pandas library and numpy library in Python, can be used to implement this step. In this embodiment, first, the collaborative simulation is performed under different parameter settings through the collaborative simulation analysis unit 143, and the results of the calculation accuracy and recall rate are recorded. Then, the simulation results are statistically analyzed, such as calculating the correlation coefficient between each parameter and the simulation results, and sorting them according to the degree of correlation, and selecting the top two key control parameters (N=2). Therefore, based on the evaluation results and sorting, we determine the response time of the energy storage system and the charge and discharge power of the energy storage system as key control parameters. A reinforcement learning algorithm is used to optimize the settings of the response time and charge and discharge power to achieve more efficient energy management and grid stability.

[0075] Specifically, the model training subunit uses parameters to determine the key control parameters selected by the subunit and performs training within the digital twin model co-simulation environment. Using reinforcement learning, it gradually optimizes the control strategy model through interaction with the environment and feedback signals. The training subunit's goal is to obtain the optimal control strategy to maximize the performance and effectiveness of the co-simulation.

[0076] Specifically, the model application subunit applies the optimal control strategy learned through training to the digital twin model. It adjusts the parameters of the digital twin model based on the optimal control strategy to achieve coordinated control and optimization of multi-source heterogeneous data in the power grid. The model application subunit's role is to apply the trained strategy to real-world scenarios, improving and optimizing power grid performance.

[0077] The various subunits in the simulation control unit 144 cooperate with each other to achieve collaborative control and optimization of multi-source heterogeneous data of the power grid through digital twin models, reinforcement learning and simulation technologies. The strategy construction subunit provides the initial control strategy model; the parameter determination subunit selects key control parameters; the model training subunit optimizes the control strategy by interacting with the environment; and the model application subunit applies the optimized strategy to the digital twin model. This collaborative working method can maximize the use of information from multi-source heterogeneous data of the power grid.

[0078] In this embodiment, the simulation control unit 144 uses Keras to build an LSTM deep learning model as a prediction model for simulation calculation results. Historical simulation data is used as model input, and control parameters are used as features. The model is trained to predict calculation indicators. A reinforcement learning agent is constructed in TensorFlow. The state space, action space, and reward function of the simulation environment are defined. The agent is trained using the policy gradient method to obtain the optimal policy for the control parameters. The regression coefficients are analyzed to select the control parameters that have the greatest impact on the prediction results. The intermediate layer outputs of the analysis model are correlated to obtain the control variables that significantly affect the system's collaborative performance. An online control model is constructed to receive real-time simulation output and evaluate the current control strategy. The reinforcement learning model is then used online to optimize the strategy to achieve closed-loop control and continuously improve the coordination between the simulation and the actual system. A federated learning approach is used to integrate the control models of different business systems, prevent data silos, and achieve coordinated control of the power grid digital twin. By applying technologies such as deep learning prediction models, reinforcement learning strategy optimization, and closed-loop control, the control parameters of the power grid digital twin can be continuously optimized, effectively improving the accuracy of the simulation and its coordination with the actual system.

[0079] In summary, the various units of computing module 140 work together through collaborative computing, output aggregation, simulation analysis, and simulation control, achieving improved efficiency and technical effectiveness in the fusion of multi-source heterogeneous data in the power grid. Collaborative computing unit 141 uses distributed computing and communication technologies to accelerate the calculation process, output aggregation unit 142 unifies the calculation results of multiple models, simulation analysis unit 143 uses deep learning algorithms to reveal the correlation patterns between data, and simulation control unit 144 uses reinforcement learning algorithms to adjust model parameters to optimize performance. Through this mutual cooperation and collaborative computing, computing module 140 effectively improves the efficiency of the fusion of multi-source heterogeneous data in the power grid.

[0080] FIG6 is a schematic diagram of an edge computing module 150 according to some embodiments of the present specification. As shown in FIG6 , the digital twin system of the power grid of the present application further includes an edge computing module 150 for pre-processing multi-source heterogeneous data connected to the power grid; the edge computing module 150 includes: an edge node unit 151, an edge computing unit 152, and an edge communication unit 153. Among them, the edge node unit 151: This unit is set in the converter at the power grid end, and adopts an analog-to-digital conversion module to collect the sensor data of the power grid in real time. The edge node unit 151 is responsible for converting the data from the analog signal to the digital signal for subsequent processing and transmission.

[0081] Edge computing unit 152 receives sensor data collected by edge node unit 151 and performs local preprocessing on the data. This preprocessing utilizes coding techniques based on lossless compression algorithms to remove redundant information from the data and uses a low-order bit-first transmission mechanism to increase data transmission rates. The goal of preprocessing is to reduce data size and transmission bandwidth usage while preserving important information for subsequent transmission and analysis.

[0082] Among them, edge communication unit 153: The pre-processed sensor data is transmitted to the computing module 140 via the 5G communication network through the edge communication unit 153. The edge communication unit 153 is responsible for transmitting the pre-processed data to the computing module 140, using the 5G communication network to provide high-speed, low-latency data transmission capabilities to ensure the timeliness and reliability of the data.

[0083] Specifically, the preprocessing technology of edge computing unit 152 can remove redundant information from the data, reduce data size, and retain key information to improve data transmission efficiency and bandwidth utilization. Using a low-order-first transmission mechanism, edge computing unit 152 prioritizes the transmission of important parts of the data, thereby accelerating data transmission rates, reducing transmission delays, and ensuring the timeliness and real-time nature of the data. Edge communication unit 153 utilizes the 5G communication network for data transmission, providing high-speed, low-latency transmission capabilities to meet the demand for real-time transmission of multi-source heterogeneous data in the power grid.

[0084] In this embodiment, an MCU chip such as STM32 is integrated into the smart meter or sensor at the power grid end, and a built-in analog-to-digital conversion chip is used to collect analog signals such as voltage and current. Parameters such as sampling frequency and accuracy are set. A lightweight computing program based on Docker is deployed on the edge node. The collected data is compressed and encoded using a deep learning model, such as a compressed sensing network based on an autoencoder. Floating-point low-order first encoding such as IEEE 754 is used to optimize data representation. The meter is connected to the 5G network through a narrowband IoT module such as CAT-M1. The QUIC transmission protocol is used to transmit sensor data using the high bandwidth and low latency of the 5G network. Security mechanisms such as DTLS are also used to prevent data leakage. The parameters of the encoding model and transmission module are adaptively optimized based on the data content and flow rate to reduce computing and communication resource consumption. Through intelligent collection, encoding, and transmission at the edge, the bandwidth usage of sensor data can be effectively reduced, ensuring low-latency transmission of key power grid data and providing real-time status data support for digital twins.

[0085] In summary, the various units in edge computing module 150 work together to form a complete edge computing module 150, which plays a role in processing and transmitting multi-source heterogeneous data in the power grid. Edge node unit 151 is responsible for data acquisition, edge computing unit 152 performs local preprocessing, including lossless compression and low-order bit first transmission, and edge communication unit 153 transmits the processed data to computing module 140 via the 5G communication network.

[0086] As a preferred embodiment of the present application, the digital twin system of the power grid of the present application also includes an optimization module, which is connected to the calculation module 140 to optimize the digital twin model; the optimization module includes: a data acquisition unit to obtain real-time operation data of the power grid; wherein the data acquisition unit collects real-time operation data from sensors, smart meters, etc. in the power grid. The model calibration unit uses statistical analysis and other methods to calculate the deviation between real-time data and model output and calibrate the parameters of the model. The correlation analysis unit 131 applies algorithms such as correlation analysis to analyze the correlation between different models. The model adjustment unit adjusts the model parameters or model structure based on the model calibration results and correlation analysis. The optimization control unit coordinates and controls the steps of the entire optimization process. In summary, the optimization module realizes automatic calibration and optimization iteration of the digital twin model by comparing with real-time operation data, thereby ensuring the accuracy of the model. Correlation analysis guides the joint optimization between different models.

[0087] As a preferred embodiment of the present application, the digital twin system of the power grid of the present application also includes a data security module, which securely encrypts the system's data; the data security module includes: a data encryption unit, connected to the input data interface of the digital twin system, to encrypt the multi-source heterogeneous data input by the system; an access control unit, connected to the data encryption unit, to perform hierarchical and authorized access control on the encrypted data by different users; and a data audit unit, connected to the data encryption unit and the access control unit, to audit the encryption and access operations of the data. The data encryption unit formats and converts the structured and unstructured data input by the system, and encrypts the data using an asymmetric encryption algorithm, such as the RSA algorithm, so that the encrypted data can only be decrypted using a matching key. The access control unit constructs an access control policy based on the data classification level and the user's access rights, authenticates and authorizes the user's access behavior, and supports dynamic management and adjustment of access rights. The data audit unit records the operation logs of data encryption and access control, which contain relevant information such as operation time, data, and user, and regularly generates audit reports for security analysis.

[0088] As a preferred embodiment of the present application, the digital twin system of the power grid of the present application also includes a model lightweight module, the model lightweight module, the model lightweight module reduces the size of the scene model of the power grid in the digital twin system; the model lightweight module includes: a model extraction unit, connected to the modeling module 120120 of the digital twin system, extracting the established scene model of the power grid; a model compression unit, connected to the model extraction unit, using model compression technology to reduce the size of the scene model; a model clipping unit, connected to the model compression unit, clipping the redundant structure in the scene model, the redundant structure is a redundant intermediate feature layer; an optimization model unit, connected to the model clipping unit, outputting the optimized lightweight scene model of the power grid; wherein, the redundant intermediate feature layer refers to a hidden layer whose contribution to the final output result of the model is less than a threshold value through importance analysis and structural adjustment during the model compression process. In summary, the redundant intermediate feature layer can be identified and clipped through importance analysis and structural adjustment, thereby achieving the effect of model compression without having a significant negative impact on the quality of the model output. The model extraction unit calls an established large-scale scenario model from the digital twin system. The model is organized in the form of a computation graph. The model compression unit uses deep learning model compression algorithms, such as network pruning and low-rank decomposition, to remove redundant model parameters and reduce the parameter size. The model pruning unit analyzes the model to determine the contribution of intermediate layer outputs to the final output, prunes intermediate layers with low contributions, and obtains the pruned model. The model optimization unit further optimizes the model structure, streamlines model connections, and outputs an optimized lightweight model. Specifically, model compression: A method that uses algorithms to reduce model complexity. Model pruning: A technique that removes redundant parts of a model. Lightweight model: A model form with smaller size and more efficient computation. In summary, the model lightweight module achieves the conversion of large models to lightweight models through compression and pruning, reducing the storage and computational costs of the digital twin system.

[0089] As a preferred embodiment of the present application, the digital twin system of the power grid of the present application also includes a three-dimensional visualization module, which performs three-dimensional display of the power grid scene of the digital twin system; the three-dimensional visualization module includes: a data acquisition unit, which is connected to the scene database of the digital twin system to obtain data of the power grid scene; a rendering generation unit, which uses rendering technology to perform three-dimensional rendering of the three-dimensional visualization model; a display control unit, which is connected to the rendering generation unit and controls the three-dimensional visualization of the power grid scene, and the control is perspective control and viewing distance control. Among them, the data acquisition unit obtains the model data of the power grid entity from the scene database of the digital twin system, and the data includes spatial information, connection relationship, attribute characteristics, etc.; the rendering generation unit uses graphics APIs such as OpenGL or DirectX for rendering, supports rendering of lighting, material, shading and other effects, and is compatible with the display of various terminal devices; the display control unit provides interactive viewpoint and viewing distance control, supports interactive operations such as model selection and query, and ensures the smoothness of rendering and display. Specifically, the 3D visualization module is a tool for building 3D scenes and rendering simulation visualization services. It primarily enables the unified upload and management of diverse model equipment, including distribution equipment, transmission equipment, primary and secondary substation equipment, scene buildings, and tools. Users can search for required model files using keyword-matching searches. The module also provides the ability to create groups for categorized model file management, supporting operations such as viewing, editing, copying, and deleting. Double-clicking a card or clicking "View" takes you to the corresponding device model library management page, allowing users to perform detailed operations. It primarily includes three functions: Model Processing: Uploaded 3D models undergo a series of processing, including format conversion and coordinate conversion, to ensure they meet system data requirements. This processing includes, but is not limited to, converting the model to a supported format and adjusting the coordinate system to accommodate the system environment. After these processing steps are completed, subsequent operations are performed based on the imported model's level (e.g., component, device, or power station). If it is a reusable model (such as an equipment model or a component model), it will be stored in the three-dimensional resource model library with its model number as an identifier, and an index will be formed as the input model library for scene editing; for integrated scene models that are not universal (such as substation / converter station models, transmission / distribution network models), they will be stored in the three-dimensional scene model library to facilitate subsequent operations by users. Scene editing: The scene editing function of this system is based on WebGL technology. Users can combine three-dimensional resource models into three-dimensional scene models through simple drag and drop operations on the browser side. At the same time, the connected equipment ledger information is associated and mapped with the model to construct a virtual asset instance. For different asset instances, the system will form a searchable tree structure based on specific data such as the city to which it belongs, the operation and maintenance unit, etc., and calibrate the location more accurately based on geographic information data.In addition to the 3D geometric model of the object and the relative position of the object in the scene, the generated model file also includes object attributes such as the name / ID of the object in the scene, the GIS information of the scene, and the name / ID of the scene. After review, this information will be stored in the 3D scene model library for subsequent scene editing, simulation, and analysis. Service publishing: The system forms a RESTful interface through service encapsulation to provide users with a series of general model query services. Through these services, model files can be obtained in batches, and multi-dimensional conditional query interfaces can be used to retrieve model lists and model status data. In this way, the system forms a general model service that can be opened to upper-level applications, providing rich model resources and data support for other applications. In summary, the 3D visualization module, through explicit 3D models, facilitates users' understanding of complex power grid scenarios and provides an intuitive basis for diagnosis and decision-making.

[0090] The above description is merely an illustration of some preferred embodiments of the present disclosure and the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, but also includes other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the above-mentioned inventive concept.

Claims

1. A digital twin system of a power grid, comprising: Data access module, which accesses multi-source heterogeneous data of the power grid, including real-time monitoring data, historical operation data, meteorological environment data, equipment model data and geographic information data; The modeling module receives data from the data access module and builds a digital twin model suitable for multiple business scenarios; Fusion module, which associates and integrates digital twin models of multiple business scenarios; The computing module uses the associated and fused digital twin models to perform multi-scenario collaborative simulation.

2. The digital twin system of the power grid according to claim 1, characterized in that: The data access module includes: The standard interface unit defines the standard service interface for data access through OpenAPI and adopts an asynchronous non-blocking input or output model to improve the response speed of the service interface; The format conversion unit converts data in different formats through JSON conversion logic; A semantic conversion unit, which converts different semantic data based on word vectors of an improved Word2Vec model, wherein the improved Word2Vec model is a model that adds semantic constraints to optimize word vector representation; The latitude conversion unit converts data of different latitudes by combining the geodetic coordinate system conversion based on positive calculation and the coordinate measurement based on satellite navigation; The middleware unit uses an optimized distributed message queue to access multi-source heterogeneous data, and the optimization is through intelligent routing diversion and dynamic resource scheduling.

3. The digital twin system of the power grid according to claim 2, characterized in that: The middleware unit includes: The association subunit uses a dependency analysis method based on causal model inference to establish causal dependency relationships between multi-source heterogeneous data; The routing sub-unit performs intelligent routing and diversion according to the causal dependency between multi-source heterogeneous data, so as to carry out business association between multi-source heterogeneous data; The resource subunit adjusts the resource queue length by changing the number of queue partitions based on the real-time monitoring results of data traffic, and changes the resource queue width by adjusting the number of threads in the consumer thread pool.

4. The digital twin system of the power grid according to claim 1, characterized in that: The modeling module includes: The basic model building unit builds the basic resource model of the power grid by using the power system steady-state analysis method according to the topological structure of the power grid; The digital twin model unit uses the basic resource model to build a digital twin model through visual editing tools according to different business scenarios; The model library unit uses a distributed model storage mechanism to store different types of preset model structures for the digital twin model unit to call; The parameter fitting unit fits the parameters of the digital twin model according to the federated learning algorithm and monitoring data; The model management unit stores each digital twin model in the model library and encapsulates the digital twin model into a service for publication.

5. The digital twin system of the power grid according to claim 1, characterized in that: The fusion module comprises: The association analysis unit uses mutual information theory and association rule algorithm to obtain the association relationship between different multi-source heterogeneous data; The rule fusion unit uses time rules, space rules and business rules to perform cross-domain feature mapping and relationship matching according to the association relationship between the acquired multi-source heterogeneous data, and generates a matching mapping table between the heterogeneous data; The model fusion unit uses the inference algorithm and sequence pattern matching technology according to the matching mapping table to fuse the digital twin model based on multi-source heterogeneous data; The model collaboration unit constructs a model association diagram between different businesses based on the association relationship and the integrated digital twin model. Based on the model association diagram, it extracts the interface mapping relationship of the digital twin model, and performs collaborative simulation of the digital twin models of different businesses through the interface mapping relationship and message passing.

6. The digital twin system of the power grid according to claim 5, characterized in that: The calculation module comprises: The collaborative computing unit uses the interface mapping relationship of the digital twin model to perform distributed collaborative simulation calculations on the associated and fused digital twin models through parallel computing and distributed communication technology; The output aggregation unit uses heterogeneous information serialization and encoding technology to generate output data containing the collaborative simulation calculation results of multiple digital twin models; The simulation analysis unit builds a prediction model based on the deep learning algorithm and uses the prediction model to analyze the calculation results of the collaborative simulation under different control parameter settings; The simulation control unit uses the policy gradient method of the reinforcement learning algorithm to adjust the control parameters of the digital twin model according to the analysis results of the calculation results of the collaborative simulation.

7. The digital twin system of the power grid according to claim 6, characterized in that: The simulation control unit includes: The strategy construction subunit uses the policy gradient method in the reinforcement learning algorithm to build a control strategy model for collaborative simulation of the digital twin model; The parameter determination subunit uses the prediction model constructed by the simulation analysis unit to select the top N control parameters that have the greatest impact on the collaborative simulation calculation accuracy and recall rate as key control parameters according to the collaborative simulation results under different control parameter settings; The model training subunit uses the selected key control parameters to train the control strategy model in the digital twin model collaborative simulation environment to obtain the optimal control strategy; The model application subunit uses the optimal control strategy obtained through training to adjust the parameters of the digital twin model.

8. The digital twin system of the power grid according to claim 6, characterized in that: The output aggregation unit includes: The data parsing subunit extracts text data from the collaborative simulation calculation results of different digital twin models by combining named entity recognition and dependency syntax analysis, and converts it into structured data to generate first output data; The encoding subunit encodes the first output data by using a compression encoding algorithm based on sparse representation to generate second output data; The relation extraction subunit uses the knowledge graph technology to extract the association relationship between the second output data; The relation fusion subunit associates the second output data according to the extracted association relationship to generate the final output data.

9. The digital twin system of the power grid according to claim 8, characterized in that: Compression coding algorithms based on sparse representation include: The dictionary learning method is used to obtain an over-complete dictionary of the power grid monitoring data sample set; Applying an orthogonal matching pursuit algorithm to the first output data to obtain a sparse representation coefficient of the first output data; The non-zero elements in the sparse representation coefficients are retained to generate second output data.

10. The digital twin system of the power grid according to claim 1, characterized in that: The digital twin system of the power grid also includes an edge computing module for preprocessing multi-source heterogeneous data connected to the power grid locally; The edge computing module includes: The edge node unit is set in the converter at the power grid end and collects the sensor data of the power grid through the analog-to-digital conversion module; The edge computing unit performs local preprocessing on the collected sensor data. The preprocessing uses a coding technology based on a lossless compression algorithm to remove redundant information in the data and uses a low-bit-first transmission mechanism to increase the data transmission rate. The edge communication unit transmits the pre-processed sensor data to the computing module using the 5G communication network.

Images