Multi-terminal collaborative modular power operation and maintenance method and system based on digital twinning

Abstract

The invention discloses a multi-terminal collaborative modular power operation and maintenance method and system based on digital twinning, and the method comprises the following steps: constructing a global power topological graph of a power energy site, mapping power equipment, sensors and connecting lines into graph nodes and graph edges, and binding the static attributes of the equipment with the topological structure; and acquiring real-time operation data of the power equipment, and distributing a unified time label for the multi-source heterogeneous data. According to the method, a unified twinborn model of cross-voltage class, cross-region and cross-equipment type is established in a digital twinborn platform, and operation data of different equipment such as photovoltaic equipment, battery energy storage equipment, an inverter, a protection device, an electric energy meter and temperature control equipment are expressed in a unified data semantic mode; according to the invention, the system can still automatically establish a consistent device portrait under the conditions of a large number of multi-terminal devices and diversified communication protocols, and the multi-terminal cooperative computing capability is significantly improved.

Description

Technical Field

[0001] This invention relates to the field of power operation and maintenance technology, specifically to a multi-terminal collaborative modular power operation and maintenance method and system based on digital twins. Background Technology

[0002] With the development of large-scale power grids, ultra-high-voltage transmission projects, and new power systems, the number of devices in power systems, such as photovoltaic power plants, energy storage systems, substations, distribution rooms, user-side loads, and charging infrastructure, is rapidly increasing. The system structure is becoming increasingly modular, multi-layered, and cross-scenario. In such complex power grids, maintenance personnel not only need to monitor a large number of heterogeneous devices in real time, but also need to perform unified analysis and collaborative processing of operating status across regions, voltage levels, and subsystems.

[0003] Currently, power operation and maintenance systems generally use data acquisition terminals, edge controllers, and monitoring center platforms to manage power equipment, but the following technical challenges still exist: First, the data sources of multiple terminal devices in the power system are severely heterogeneous.

[0004] Inverters, meters, protection devices, energy storage PCS, and temperature control equipment of different brands, with different communication protocols and sampling periods constitute a highly heterogeneous data system. Due to the inconsistency in data granularity, refresh frequency, and data format, it is difficult for the system to achieve unified modeling and unified calculation, which greatly increases the difficulty of cross-terminal collaboration.

[0005] Second, the temporal inconsistency between multiple data sources prevents the system from forming a complete real-time operational profile.

[0006] In multiple subsystems such as grid connection point, grid side, high voltage room, photovoltaic side, and energy storage side, the data sampling period often ranges from seconds to minutes. Data delay and packet loss occur frequently, making it difficult to construct an accurate overall system status and making it impossible to use for scheduling analysis, power flow prediction, and equipment health assessment.

[0007] Third, existing digital twins are mainly limited to device-level or single-site-level modeling, and cannot achieve real-time collaboration across regions and multiple terminals.

[0008] Existing technologies generally construct digital twins using traditional 3D models, finite element models, or physical mechanism models, but a large proportion of these are "static twins," lacking real-time synchronization capabilities. They cannot achieve data complementarity or topological correlation reasoning between multiple terminals, nor can they form a "globally unified twin" at the system level.

[0009] Fourth, insufficient multi-station collaboration and cross-terminal correction capabilities make it difficult to identify anomalies in the original data.

[0010] In complex power grids, when noise, discrepancies, or distortions occur in the data of a certain terminal, existing systems typically only perform threshold judgment or simple filtering. They cannot use the power topology relationship between devices for correction, nor can they infer the true state based on the physical laws of neighboring devices, which leads to a decrease in the accuracy and reliability of system operation and maintenance.

[0011] Fifth, the existing operation and maintenance system lacks self-learning capabilities and cannot automatically optimize itself in response to aging power equipment, load changes, and weather changes.

[0012] After long-term operation, the performance indicators, efficiency curves and response parameters of power equipment will change. Traditional systems rely on manual maintenance or manual parameter adjustment, which makes it impossible for the operation and maintenance system to automatically update models and strategies, and thus makes it difficult to continuously improve operating efficiency.

[0013] Sixth, there is a lack of a high-precision global digital twin model that can support scheduling and can be expanded according to different scenarios.

[0014] For large-scale photovoltaic bases, energy storage power stations, and power distribution systems, most existing systems cannot integrate electrical topology, real-time data, equipment health status, and operation simulation into a unified digital twin model, making it difficult to meet the needs of scheduling, predictive operation and maintenance, and full life cycle management. To address this, a multi-terminal collaborative modular power operation and maintenance method and system based on digital twins is proposed. Summary of the Invention

[0015] The purpose of this invention is to provide a multi-terminal collaborative modular power operation and maintenance method and system based on digital twins. This aims to solve one of the problems existing in the prior art.

[0016] Firstly, to address the aforementioned technical problems, this application adopts a technical solution: a multi-terminal collaborative modular power operation and maintenance method based on digital twins, comprising the following steps: Construct a global power topology map of power energy sites, mapping power equipment, sensors and connecting lines as graph nodes and graph edges, and binding the static attributes of equipment to the topology structure; Acquire real-time operating data of power equipment, assign unified time labels to multi-source heterogeneous data, and align data of different frequencies to the same time slice through resampling technology. Then bind the aligned data as dynamic attributes of nodes and edges. The state propagation calculation of the global power topology graph is performed based on the graph neural network, and the electrical correlation between nodes is analyzed by the graph message passing mechanism to perform global consistency correction of the equipment operating status. The photovoltaic, energy storage and power grid subsystems are mapped as independent subgraphs. Based on the main wiring diagram, the multi-subgraphs are integrated and simulated collaboratively, and the energy flow state across subsystems is coupled and corrected. The system monitors the difference between the actual value of the equipment operation data and the predicted value of the digital twin model in real time. When the difference exceeds the preset threshold, it triggers an operation and maintenance warning and dynamically updates the edge weight parameters of the topology graph based on long-term operation data. The corrected digital twin data is output to a multi-terminal operation and maintenance platform, providing equipment status monitoring, fault location, and collaborative scheduling decisions based on a consistent digital twin graph.

[0017] In one possible implementation, the construction of a global power topology map of power energy stations specifically includes: Based on the main wiring diagram and site layout diagram of the power station, transformers, inverters and switchgear equipment are defined as diagram nodes, and cables, busbars and protection circuits are defined as diagram edges; The equipment type, rated voltage, rated current, and installation location information are stored as static attributes in the corresponding graph nodes, forming a digital infrastructure that reflects the physical connection relationships.

[0018] In one possible implementation, the method of assigning a unified time label to multi-source heterogeneous data and aligning data of different frequencies to the same time slice through resampling technology specifically includes: Real-time data from power equipment is uploaded via edge acquisition devices, and the data is managed hierarchically according to the sampling frequency. By setting a unified time base, for data with inconsistent sampling frequencies, resampling or interpolation algorithms are used to map them into the same time window, ensuring that the dynamic attributes of all devices are kept in time synchronization in the global power topology map.

[0019] In one possible implementation, the state propagation calculation of the global power topology graph based on a graph neural network specifically includes: By utilizing the message passing mechanism of graph neural networks, abnormal state features are propagated along the topology between device nodes; Based on the physical constraints of current, voltage, and power attributes between upstream and downstream nodes, the output status values ​​of the nodes are automatically adjusted to eliminate local measurement errors and maintain the conservation and consistency of power flow data across the entire network.

[0020] In one possible implementation, mapping the different subsystems of photovoltaics, energy storage, and the power grid into independent subgraphs specifically includes: The photovoltaic power generation system, energy storage system and distribution network side are treated as independent subgraph structures. By establishing connection edges between subgraphs through power transmission paths, device nodes in different subgraphs can communicate and interact across subgraphs, realizing cross-system state sharing and boundary condition constraints.

[0021] In one possible implementation, the coupling correction of the energy flow state across subsystems specifically includes: Through a cross-subgraph collaborative simulation mechanism, the data coupling relationship between the photovoltaic subsystem, the energy storage subsystem, and the power distribution system is calculated in real time. The system jointly adjusts the equipment status of each subsystem at the same time, eliminates the data silo effect caused by independent calculation of factor systems, and ensures that the collaborative operation logic between multiple systems conforms to physical laws.

[0022] In one possible implementation, the difference between the actual value of the real-time monitoring device's operating data and the predicted value of the digital twin model specifically includes: Construct a difference analysis model to calculate in real time the difference between the predicted state of each device node and the actual value uploaded by the sensor; When the difference exceeds the set threshold range, an early warning signal is automatically triggered, and a graph neural network is used to backtrack and analyze the source of the difference, identify abnormal data sources, and automatically perform data cleaning or correction.

[0023] In one possible implementation, the dynamic updating of the edge weight parameters of the topology graph based on long-term operational data specifically includes: Collect historical data on equipment operation and real-time difference analysis results; The graph neural network is trained online using an incremental learning method. The device model parameters are adjusted according to the aging of the device or changes in the environment, and the node weights and edge weights in the topology graph are dynamically updated to maintain the fidelity of the digital twin model.

[0024] In one possible implementation, the step of outputting the corrected digital twin data to a multi-terminal operation and maintenance platform specifically includes: Generate a visualized digital twin containing data on equipment health status, power flow, real-time load, and predictive maintenance. It supports multiple user terminals to access device information at different levels according to their permissions, and transmits decision data to the dispatching system in real time to support the overall dispatching and emergency response of large-scale power grids.

[0025] Secondly, to solve the aforementioned technical problems, another technical solution adopted in this application is: a multi-terminal collaborative modular power operation and maintenance system based on digital twins, comprising: The topology building module is configured to create a global power topology graph of power energy sites and map the static attributes of devices to the graph structure; The data synchronization module is configured to collect real-time data and perform time stamp alignment and resampling to ensure the synchronization of multi-source data within the same time slice; The intelligent correction module is configured to perform node state propagation and global consistency correction based on graph neural networks, and to perform cross-subgraph co-simulation. The difference optimization module is configured to monitor the difference between the predicted value and the true value, trigger an early warning, and dynamically update the model parameters based on a self-learning mechanism. The operation and maintenance interaction module is configured to output a consistent digital twin graph to a multi-terminal platform, providing status monitoring, fault location, and collaborative scheduling functions.

[0026] Thirdly, to solve the above-mentioned technical problems, another technical solution adopted in this application is: a multi-terminal collaborative modular power operation and maintenance method based on digital twins, the method comprising the following steps: Step 1: Deploy data computing units at the edge layer and cloud layer respectively, establish a unified topology meta-model library, and configure a unified timestamp protocol and clock synchronization strategy for all network data sources; Step 2: Construct a multi-layer topology graph containing electrical, functional, and business dimensions based on the aforementioned topology meta-model library, and map the real-time collected power data into the dynamic attributes of the graph nodes; Step 3: Construct and train a time-series topology graph neural network that incorporates the physical constraints of the power grid. Run a lightweight model at the edge for local state estimation and a high-precision model in the cloud for global state fusion. Step 4: Perform real-time synchronization between the cloud twin and the real network data, calculate the residual between the cloud twin state and the real network collected data to detect abnormal differences, and use the attention weights of the graph neural network to locate the source of the difference; Step 5: Classify the detected differences, perform online adaptive correction according to the difference type, and dynamically correct the parameters of the digital twin model through parameter inversion and online learning mechanisms; Step 6: Map the analysis results output by the graph neural network into modular operation and maintenance action units, and distribute them to multiple terminals through the operation and maintenance bus for collaborative execution and work order closure.

[0027] In one possible implementation, constructing the multi-layer topology graph includes: Static attributes of nodes and edges are extracted from the meta-model library to construct electrical topology layer, functional topology layer and business topology layer; A partitioned modeling strategy is adopted, forming subgraph groups by geographical or power grid segmentation, and establishing convergence edges between different subgraphs. Cross-regional message passing and global consistency checks are performed through these convergence edges.

[0028] In one possible implementation, the construction and training of the time-series topological graph neural network incorporating power grid physical constraints includes: The message passing unit is designed to handle both node timing characteristics and topological space characteristics, and incorporates soft coding of physical constraints on power flow direction and device capacity boundaries into the message function; By using structural pruning and hybrid precision quantization techniques, a lightweight edge model deployed on edge nodes and a high-precision cloud model deployed in the cloud are generated, forming a binary model system.

[0029] In one possible implementation, the synchronization and difference detection of real-time twin and real-network data includes: Edge nodes send summaries of their local state estimates to the cloud, and the cloud graph neural network receives the summaries of each subgraph and performs global message fusion. Based on the graph-level residual monitoring mechanism, the cloud twin status is compared with the real network aggregated data. When an over-threshold alarm is triggered, the graph information backtracking mechanism is activated to locate the source of the difference in the device, line or data link.

[0030] In one possible implementation, classifying the detected differences and performing online adaptive correction includes: A graph-level classifier is used to identify differences as data problem type, equipment behavior deviation type, or external disturbance type. For data discrepancies, trigger edge retransmission or interpolation to fill in the gaps; For deviations in equipment behavior, perform parameter inversion to estimate the effective parameter offset of the equipment, and issue parameter correction suggestions; For external perturbation-type differences, the short-term predicted distribution of the twin model is updated in a graph-level mode.

[0031] In one possible implementation, the step of dynamically correcting the parameters of the digital twin model through parameter inversion and online learning mechanisms further includes: Based on the difference correction results, an incremental learning mechanism is used to correct the weights of the graph neural network model. After batch updates are completed in the cloud, the data is periodically distributed to the edge model. If the device's performance parameters deviate from their normal range for an extended period, the node's parameters will be updated in real time in the digital twin model, and a maintenance work order containing the topology context will be automatically generated.

[0032] In one possible implementation, mapping the analysis results output by the graph neural network into modular operation and maintenance action units includes: Define standardized operation and maintenance action units, which include automatic alarm, priority dispatch, temporary isolation and power reallocation; Set action rollback thresholds and manual confirmation procedures for high-voltage or ultra-high-voltage nodes, and register all twin scheduling actions and model correction actions on an immutable audit chain.

[0033] Fourthly, to solve the above-mentioned technical problems, another technical solution adopted in this application is: an electronic device, including a processor, a memory and a communication interface, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the steps of the above-mentioned multi-terminal collaborative modular power operation and maintenance method based on digital twins.

[0034] Fifth aspect: To solve the above-mentioned technical problems, another technical solution adopted in this application is: a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the multi-terminal collaborative modular power operation and maintenance method based on digital twin as described above.

[0035] The present invention has the following beneficial effects: 1. This invention establishes a unified twin model across voltage levels, regions, and equipment types in a digital twin platform, expressing the operating data of different devices such as photovoltaics, battery energy storage, inverters, protection devices, electricity meters, and temperature control equipment in a unified data semantic manner. This invention enables the system to automatically establish a consistent device profile even when there are a large number of multi-terminal devices and diverse communication protocols, significantly improving the multi-terminal collaborative computing capabilities. 2. This invention utilizes the power topology in a digital twin to enable the system to cross-correct data from a single terminal based on the electrical relationships between devices; 3. This invention adopts a modular twin construction method, which allows different functional units (such as grid connection point, feeder unit, photovoltaic array unit and energy storage unit) to be created, combined and upgraded independently; 4. By constructing a cross-terminal time-series fusion mechanism, this invention can automatically handle problems such as inconsistent sampling periods, asynchronous delays, and missing data, and achieve dynamic alignment of data from different sources in the digital twin model; 5. This invention utilizes a parameter self-learning mechanism within the twin to automatically fine-tune the internal model of the twin based on factors such as equipment aging, temperature changes, and load fluctuations. 6. This invention enables various devices such as energy storage, inverters, switching equipment, and power quality devices to share status information and achieve collaborative scheduling through the cross-terminal message propagation mechanism within the digital twin system. 7. The twin of the present invention can automatically generate future operating trends based on the current state, including equipment health trends, power change trends, voltage deviation trends, etc. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 This is a flowchart illustrating the multi-terminal collaborative modular power operation and maintenance method based on digital twins according to the present invention. Figure 2 This is a flowchart illustrating Embodiment 2 of the present invention; Figure 3 This is a block diagram of a multi-terminal collaborative modular power operation and maintenance system based on digital twins according to Embodiment 3 of the present invention; Figure 4 This is a schematic diagram of the structure of the electronic device of the present invention. Detailed Implementation

[0038] 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.

[0039] Figure 1 This is a flowchart illustrating a multi-terminal collaborative modular power operation and maintenance method based on digital twins, according to an embodiment of the present invention. It should be noted that if substantially the same result is achieved, the method of this application is not necessarily identical. Figure 1 The sequence of processes shown is limited.

[0040] Example 1: like Figure 1 The multi-terminal collaborative modular power operation and maintenance method based on digital twins, as shown, includes the following steps: S10. Construct a global power topology map of power energy stations, map power equipment, sensors and connecting lines as graph nodes and graph edges, and bind the static attributes of equipment to the topology structure; In this embodiment, specifically, the construction of the global power topology map of the power energy station includes: Based on the main wiring diagram and site layout diagram of the power station, transformers, inverters and switchgear equipment are defined as diagram nodes, and cables, busbars and protection circuits are defined as diagram edges; The equipment type, rated voltage, rated current, and installation location information are stored as static attributes in the corresponding graph nodes, forming a digital infrastructure that reflects the physical connection relationships.

[0041] S20. Obtain real-time operating data of power equipment, assign unified time labels to multi-source heterogeneous data, and align data of different frequencies to the same time slice through resampling technology, and bind the aligned data as dynamic attributes of nodes and edges. In this embodiment, specifically, the process of assigning a unified time tag to multi-source heterogeneous data and aligning data of different frequencies to the same time slice using resampling technology includes: Real-time data from power equipment is uploaded via edge acquisition devices, and the data is managed hierarchically according to the sampling frequency. By setting a unified time base, for data with inconsistent sampling frequencies, resampling or interpolation algorithms are used to map them into the same time window, ensuring that the dynamic attributes of all devices are kept in time synchronization in the global power topology map.

[0042] S30. Perform state propagation calculation on the global power topology graph based on graph neural network, analyze the electrical association between nodes using graph message passing mechanism, and perform global consistency correction on the equipment operating status. In this embodiment, specifically, the state propagation calculation of the global power topology graph based on the graph neural network includes: By utilizing the message passing mechanism of graph neural networks, abnormal state features are propagated along the topology between device nodes; Based on the physical constraints of current, voltage, and power attributes between upstream and downstream nodes, the output status values ​​of the nodes are automatically adjusted to eliminate local measurement errors and maintain the conservation and consistency of power flow data across the entire network.

[0043] S40. Map different subsystems of photovoltaic, energy storage and power grid into independent subgraphs, perform fusion and collaborative simulation of multiple subgraphs based on the main wiring diagram, and perform coupling correction on the energy flow state across subsystems. In this embodiment, specifically, mapping different subsystems of photovoltaics, energy storage, and the power grid into independent subgraphs includes: The photovoltaic power generation system, energy storage system and distribution network side are treated as independent subgraph structures. By establishing connection edges between subgraphs through power transmission paths, device nodes in different subgraphs can communicate and interact across subgraphs, realizing cross-system state sharing and boundary condition constraints. The coupling correction of the energy flow state across subsystems specifically includes: Through a cross-subgraph collaborative simulation mechanism, the data coupling relationship between the photovoltaic subsystem, the energy storage subsystem, and the power distribution system is calculated in real time. The system jointly adjusts the equipment status of each subsystem at the same time, eliminates the data silo effect caused by independent calculation of factor systems, and ensures that the collaborative operation logic between multiple systems conforms to physical laws.

[0044] S50: Real-time monitoring of the difference between the actual value of equipment operation data and the predicted value of the digital twin model. When the difference exceeds the preset threshold, an operation and maintenance warning is triggered, and the edge weight parameters of the topology graph are dynamically updated based on long-term operation data. In this embodiment, specifically, the difference between the actual value of the real-time monitoring device's operating data and the predicted value of the digital twin model includes: Construct a difference analysis model to calculate in real time the difference between the predicted state of each device node and the actual value uploaded by the sensor; When the difference exceeds the set threshold range, an early warning signal is automatically triggered, and a graph neural network is used to backtrack and analyze the source of the difference, identify abnormal data sources, and automatically perform data cleaning or correction. The edge weight parameters of the topology graph that are dynamically updated based on long-term operating data specifically include: collecting historical data of equipment operation and real-time difference analysis results; The graph neural network is trained online using an incremental learning method. The device model parameters are adjusted according to the aging of the device or changes in the environment, and the node weights and edge weights in the topology graph are dynamically updated to maintain the fidelity of the digital twin model.

[0045] S60. Output the corrected digital twin data to the multi-terminal operation and maintenance platform, and provide equipment status monitoring, fault location and collaborative scheduling decision-making based on the consistent digital twin graph.

[0046] In this embodiment, specifically, the step of outputting the corrected digital twin data to the multi-terminal operation and maintenance platform includes: Generate a visualized digital twin containing data on equipment health status, power flow, real-time load, and predictive maintenance. It supports multiple user terminals to access device information at different levels according to their permissions, and transmits decision data to the dispatching system in real time to support the overall dispatching and emergency response of large-scale power grids.

[0047] In this embodiment, specifically, all original data of the target power station are collected, including the main wiring diagram, equipment list, equipment model and rated parameters, site layout diagram, existing data acquisition point list and communication channel description, historical operation data samples, and equipment maintenance records. This data is then structured and organized into three lists: an equipment list, a connection list, and a measurement point list. Each record in these lists uses standardized terminology to describe the equipment type, location, number, rated parameters, and corresponding measurement items.

[0048] Based on the compiled list, define the representation rules for two types of graphical elements. One type is equipment nodes, used to represent various physical devices such as transformers, switchgear, inverters, combiner boxes, energy storage units, distribution cabinets, meters, and monitoring sensors. The other type is connection edges, used to represent the connections between devices, such as cables, busbars, circuit breakers, and relay protection circuits. Define a set of mandatory static attributes for each type of equipment node, such as equipment type, rated voltage, rated current, rated capacity, installation location, serial number, and manufacturer information. Define a set of mandatory static attributes for each type of connection edge, such as connection type, conductor specifications, rated current carrying capacity, and connection start and end device identifiers. All static attributes use uniform field names and units and are filled in the list.

[0049] Based on the existing data acquisition system on site, the dynamic attributes corresponding to each measurement point are clearly defined, such as voltage, current, active power, reactive power, temperature, switch status, and battery charging / discharging status. A dynamic attribute mapping table is established for each device node and connection edge, specifying which real-time data should be collected by the device or connection edge, the collection frequency of this data, the communication protocol of the data source, and the compensation strategy for missing data. Priority and reliability indicators are set for each type of dynamic attribute for subsequent data alignment and anomaly detection.

[0050] The equipment set with cohesive function within the power plant is divided into several subsystems.

[0051] Examples include photovoltaic power generation systems, energy storage subsystems, power distribution systems, and critical load subsystems.

[0052] Following the node and edge rules defined in step two, a local topology diagram is created within each subsystem, marking the device connections and measurement point distribution within the subsystem. Then, the subsystems are aggregated through key connection points in the main wiring diagram to form the overall topology diagram of the power plant. The overall topology diagram should simultaneously retain the hierarchical relationships of the subsystems and the power flow direction information between devices to facilitate subsequent propagation calculations and correlation analysis.

[0053] To ensure the comparability of node states at each time point in the subsequent graph, time synchronization and data preprocessing rules were established between edge devices and the centralized acquisition system. These rules include: a unified method for selecting a time reference, prioritizing high-precision field time sources and calibrating other devices; mapping data to a unified time slice using given resampling rules for data with different sampling frequencies; filling in data loss or delays using edge buffer playback or interpolation based on the most recent value; and implementing unified anomaly filtering rules for monitoring data, eliminating obvious acquisition errors and marking anomalies as low-confidence for subsequent correction reference.

[0054] The overall topology graph and subgraphs are stored in a searchable structure in a graph database or graphical data structure. The graph database records the static and dynamic attribute mapping tables for each node and edge, and provides query, update, and subscription interfaces. The interface specifications should include how to obtain all node attributes of the graph at a certain moment, how to subscribe to real-time changes of a node or path, how to submit device additions or topology change information, and how to record versioned topology snapshots to achieve traceable management of the twin model.

[0055] After completing the graph modeling, a joint verification should be organized between the field and the dispatching end. Each item in the graph should be checked to ensure that the equipment and connections are completely consistent with the actual objects on site, that the dynamic attribute mapping corresponds to the acquisition system, and that the actual performance of the time synchronization strategy on site is verified. After verification, the graph model should be put into production as the basic data for the digital twin of the power station for subsequent use in graph learning, anomaly detection, or operation and maintenance decision-making.

[0056] For example, suppose there are the following real objects and structures at a grid-connected photovoltaic power station site: several photovoltaic module arrays, several combiner boxes, a large inverter, a low-voltage distribution box, a transformer, and several monitoring instruments. The following is a specific example of how to visualize them.

[0057] First, on-site data was collected. Engineers compiled the main wiring diagram, equipment nameplates, inverter operation manuals, and data acquisition unit communication configurations into an equipment list and a measurement point list. The equipment list recorded static information such as "Photovoltaic array A, capacity several kilowatts, installation location on the west side of the site," "Combiner box one, number of branches sixteen, connection cable number thirty," "Inverter one, model [model number missing], maximum grid-connected power eighty kilowatts," and "Transformer one, capacity five hundred kVA." The measurement point list recorded real-time measurements for each device, such as the current in each branch of the photovoltaic array, the output current of the combiner box, the DC input voltage of the inverter, the transformer oil temperature, and the low-voltage side voltage.

[0058] Next, define the nodes and edges of the graph. Define photovoltaic array A as node 1, combiner box 1 as node 2, inverter 1 as node 3, low-voltage distribution box as node 4, and transformer as node 5. Define the cable between photovoltaic array A and combiner box 1 as edge A, the cable between combiner box 1 and inverter 1 as edge B, the busbar between inverter 1 and low-voltage distribution box as edge C, and the connection between low-voltage distribution box and transformer as edge D. Fill in the static attributes for each node, such as the rated grid-connected power, model, and installation location of node 3; fill in the static attributes for each edge, such as the rated conductor model and maximum current carrying capacity of edge B.

[0059] Then, bind the dynamic attributes. Configure the field acquisition system to associate the current measurement points of each branch of combiner box 1 and the output power measurement points of inverter 1 with the corresponding node or edge dynamic attributes. For example, associate the current value of branch 3 of combiner box 1 with the current attribute of branch 3 of node 2, associate the output active power of inverter 1 with the output power attribute of node 3, and associate the bus current of edge C with the line current attribute of edge C. Set the sampling frequency of these data: the branch current of the combiner box is collected every ten seconds, the total power of the inverter is uploaded every minute, and the transformer oil temperature is reported every five minutes.

[0060] Define time synchronization and resampling strategies. Because the sampling frequency of each measurement point is different, a unified time slice is defined as one complete time slice per minute. For each minute time slice, the acquisition system will fill the time slice with the average value of the branch current sampled over ten seconds within that minute or the last value according to priority. If a sample is missing within that minute, the edge device will replay the most recent valid value from the local cache or use data from adjacent times for interpolation to fill in the missing value, and mark the value as low confidence on the data.

[0061] Construct subgraphs and integrate them into the overall topology graph. The photovoltaic array, combiner box, and inverter form the photovoltaic subgraph, and the low-voltage distribution box and transformer form the distribution subgraph. These subgraphs are connected via transformers and busbar interfaces to form the overall topology graph of the power station. This overall topology graph is imported into the graph database and a consistency check is performed. The system automatically verifies the completeness of the static attributes of all nodes and edges and checks whether all dynamic attributes are bound to the corresponding measurement points.

[0062] Finally, on-site verification is conducted. On-site engineers and maintenance personnel check the equipment item by item to confirm the correct correspondence between nodes and physical objects, ensuring that the collected data flows correctly to the dynamic attributes of the corresponding nodes and edges. A joint debugging test is then performed at the dispatch center to observe whether the time synchronization and resampling strategies maintain the integrity of each minute's time slice during on-site operation. After verification, this graph model is used as the basis for the power plant's digital twin for subsequent use.

[0063] In this embodiment, real-time data from different devices, different sampling frequencies, and different communication channels are uniformly aligned to the same point in time, so that the digital twin of the entire power station remains consistent on the same time slice.

[0064] The internal clocks of equipment within the power station, such as inverters, combiner boxes, energy storage controllers, meters, metering cabinets, and transformer protection devices, differ. Therefore, the system first selects a unified time base for the entire station, such as the clock of the station control host. Then, by issuing time synchronization commands, the time of each device is aligned within the same second range. If some older equipment does not support automatic time synchronization, the gateway will re-mark the data according to the current accurate time when the data enters the gateway.

[0065] Based on the site's operational characteristics, a unified time slice is defined for the digital twin, for example, each minute is considered a complete time slice. All real-time data must find corresponding values ​​within this minute to compose the graph state for that minute. For high-frequency sampled data, such as inverter DC-side power being sampled every five seconds, the system will take the last value or the average of multiple values ​​within that minute as the data for that minute. For low-frequency data, such as transformer oil temperature being updated every five minutes, if no updated data is found within a certain minute, the most recent data is retained and marked as low confidence.

[0066] Data from different devices passes through gateways, meters, and branch monitoring units, and each channel may experience delays, packet loss, or inconsistent data formats. The system automatically fills in delayed data based on its arrival time; for lost data, it fills it with the most recent value or automatically replays it using the data acquisition unit's cache; for data with different formats, the gateway standardizes it before writing it into the time slice of this step.

[0067] During this one-minute time slice, the system extracts voltage, current, temperature, power, and any real-time status information of all devices from various data sources and writes it into the corresponding device nodes or line connection edges. This ultimately forms a complete digital twin graph for that minute, preparing it for subsequent graph calculations.

[0068] For example, in a photovoltaic power station, the inverter uploads DC voltage every ten seconds, the combiner box uploads branch current every thirty seconds, the meter uploads active power data every minute, and the transformer oil temperature uploads every five minutes.

[0069] When the system reaches 10:05 AM: Between 10:05:00 and 10:59, the inverter uploaded DC voltage data six times, and the system took the last DC voltage as the value for that minute.

[0070] The combiner box recorded the branch current twice, and the system takes the average of these two readings.

[0071] The meter uploads active power exactly once, and the system uses it directly.

[0072] Since the transformer oil temperature is only uploaded every five minutes, there was no new value at 10:05. The system retains the oil temperature at 10:04 and marks it as low reliability.

[0073] After all the data is integrated, the system generates a complete digital twin status for "10:05".

[0074] In this embodiment, specifically, the actual topology of the power station is used to enable "automatic verification" between devices and identify erroneous data.

[0075] Establish association verification rules by utilizing the upstream and downstream relationships between devices in the topology diagram.

[0076] For example, the sum of the currents in the photovoltaic strings should equal the output current of the combiner box; the output current of the combiner box should match the DC input current of the inverter; the AC output of the inverter and the current on the low-voltage switchgear side should be consistent within a reasonable error range; the power ratio of the low-voltage side to the high-voltage side of the transformer should conform to the transformer's turns ratio.

[0077] The system sets reasonable ranges for the relationships between each type of device.

[0078] For example, if a combiner box has sixteen branch currents, and the sum of the sixteen branch currents deviates from the total output current of the combiner box by more than five percent, it indicates that the current of a certain branch may be abnormal; if the inverter output power is zero but its DC input voltage is normal, it is judged that the inverter may be in standby or fault state; if the power on the high-voltage side of the transformer is zero, but the load on the low-voltage side shows obvious power, it is judged that the data is inconsistent and needs to be corrected.

[0079] When the system detects that a piece of data violates the topological relationship, it will automatically mark it as an outlier and automatically generate a reasonable alternative value based on the upstream and downstream data.

[0080] For example, if the current in a branch of a combiner box suddenly jumps from eight amperes to zero amperes, while the currents in other branches and the total current remain unchanged, the system will automatically replace the current in that branch with a value similar to that of a branch of the same type and record it as an automatic correction value.

[0081] After all verifications and automatic repairs are completed, the data of each device node and connection edge will be entered into the digital twin graph in the form of "corrected", providing a highly reliable foundation for subsequent predictive analysis and status judgment.

[0082] For example, consider a combiner box with sixteen branch lines: At 11:02 AM, the current in all 15 branch circuits was between 8 and 9 amps, but one of them suddenly dropped to zero amps. However, the total output current of the combiner box did not change significantly compared to the previous minute.

[0083] The system determines based on topological relationships: The sum of the currents in the fifteen branch circuits is approximately 130 amperes. The total output current of the combiner box is 131 amperes; If a branch with a value of zero is actually zero, then the total output should decrease; Therefore, the system automatically corrects the current of this branch to be close to the average value of other branches, namely 8.5 amperes, and marks it as automatically corrected data.

[0084] In this embodiment, specifically, based on the main wiring diagram within the station, a cross-system association table is established to define the output and input relationships of the four major subsystems.

[0085] For example, the output of a photovoltaic inverter is connected to a low-voltage distribution cabinet, which is then connected to a transformer; the charging and discharging current of energy storage simultaneously affects the power balance on the distribution cabinet side; and the electricity consumption of important loads directly affects the transformer load.

[0086] Make a reasonable judgment on the power flow between different systems.

[0087] For example, when photovoltaic power generation increases, the inverter output power should increase, and a corresponding power increase should be seen in the distribution cabinet; if the energy storage is charging, a corresponding load increase should be seen on the distribution cabinet side; if the energy storage is discharging, the power of the distribution cabinet should decrease.

[0088] Correct inconsistent data across systems.

[0089] For example, if the photovoltaic inverter displays a 50 kW output, but the distribution cabinet only displays a 40 kW load, the system will automatically determine that there may be a deviation at a certain measurement point and check whether the line current between the inverter and the distribution cabinet is normal. If the line current is normal, the power on the inverter side will be corrected to the same value as that on the distribution cabinet.

[0090] Write the cross-system calibration results into the digital twin graph.

[0091] Once all cross-system rationality checks are completed, the digital twin of the entire site will reach a highly consistent state.

[0092] For example, a photovoltaic inverter may show a 60-kilowatt output at 1 p.m., but the power output on the distribution cabinet side is only 55 kilowatts.

[0093] The system determined that the energy storage device was charging at a power of 5 kilowatts, therefore the power distribution cabinet side should be 5 kilowatts less than the photovoltaic side, which is consistent with the measurement results. Therefore, the photovoltaic side measurement was determined to be correct.

[0094] If the energy storage is not charging, it indicates that there is a deviation at a certain measuring point on the photovoltaic side or the distribution cabinet side, which needs to be corrected.

[0095] In this embodiment, specifically, a deviation record table is established for each device. The record table includes the actual measured value, the corrected graph value, the device health status score, the historical deviation trend, and other contents.

[0096] Set deviation thresholds. For example, if the deviation of the inverter output power exceeds 3%, it is considered a minor abnormality, and if it exceeds 10%, it is considered a serious abnormality. If the transformer temperature deviation exceeds 3 degrees, it indicates that the measurement point is suspected to be abnormal.

[0097] Each time a time slice is generated, the system compares the values ​​in the graph with the actual values. For example, if the inverter displays 50 kilowatts, but the graph infers 55 kilowatts, the deviation is 5 kilowatts, and the system judges it as a slight anomaly.

[0098] Record deviation trends and create equipment profiles. If a certain measurement point has a large deviation multiple times in a row, the system will determine that the measurement point may be aging or faulty.

[0099] For example, an inverter has had a deviation of about 5 kilowatts per day over the past seven days, and the deviation is gradually increasing.

[0100] Based on the deviation trend, the system determines that the measurement error may be gradually increasing due to the aging of the inverter-side current transformer, or the output may be unstable due to excessive internal temperature of the inverter. The system automatically marks the inverter as a "weak reliability device" and prompts the maintenance personnel.

[0101] In this embodiment, specifically, all deviation changes are recorded, including deviation magnitude, deviation direction, deviation duration, and changes in equipment ambient temperature.

[0102] The system analyzes the patterns of deviations. For example, if a certain inverter is more prone to deviations when the temperature rises at noon every day, the system learns this pattern and adjusts the corresponding weights accordingly.

[0103] The parameters of devices in the digital twin diagram are automatically corrected. For example, as the rated conversion efficiency of an inverter decreases with aging, the system automatically reduces its efficiency in the twin model, making future predictions more accurate.

[0104] If a device malfunctions for an extended period, its inspection priority will be increased. For example, if the charge / discharge curve of an energy storage device does not match the actual value for more than three days, the system will list that device as a key focus.

[0105] For example, a transformer may experience frequent increases in oil temperature deviation during the summer, but remain completely normal during the winter.

[0106] After automatic learning, the system concluded that: in summer, the ambient temperature is high and the heat dissipation efficiency of the equipment decreases. The oil temperature sensor is prone to positive deviation in high-temperature environments. Therefore, in future inferences, the system will automatically consider seasonal factors and correct the reliability of the sensor.

[0107] In this embodiment, the graph data after correction, inference and learning is summarized to form a "real status graph" of the whole station. This graph is used for real-time monitoring, including equipment health, power flow and fault alarms, and is used for operation and maintenance management.

[0108] For example, it can help determine which inverters may age prematurely, which branches need cleaning, and which energy storage units are experiencing performance degradation.

[0109] This graph can be provided to the scheduling system to help identify potential overload risks, peak shaving and valley filling needs, and the need to activate or deactivate energy storage.

[0110] It supports future predictions, such as predicting the output curve of an inverter one hour in the future, and predicting changes in the available capacity of energy storage.

[0111] For example, at 3 PM, the digital twin shows: An inverter is expected to show a downward trend in the evening; an energy storage device will reach its charging limit in twenty minutes; the current fluctuation range of a certain branch is greater than usual today; therefore, the operation and maintenance system reminds maintenance personnel to check the branch in advance.

[0112] In this embodiment, specifically: I. This invention establishes a unified twin model across voltage levels, regions, and equipment types within a digital twin platform, expressing operational data from various devices such as photovoltaic systems, battery storage, inverters, protection devices, electricity meters, and temperature control equipment using a unified data semantic approach. Compared to traditional decentralized management methods based on brand or specification, this invention enables the system to automatically establish consistent device profiles even with a large number of multi-terminal devices and diverse communication protocols, significantly improving multi-terminal collaborative computing capabilities.

[0113] Second, this invention utilizes the power topology structure within a digital twin, enabling the system to cross-correct data from a single terminal based on the electrical relationships between devices (such as branches, voltage levels, and power flow). When a terminal experiences discrepancies, jitter, delays, or anomalies, this invention can automatically combine measurements from neighboring devices to infer the true operating state, thereby recovering data closer to the actual operating condition. Compared to traditional methods that rely solely on single-terminal threshold judgments, this invention significantly improves data accuracy and the precision of operational and maintenance assessments.

[0114] Third, this invention employs a modular twin construction method, enabling different functional units (such as grid connection points, feeder units, photovoltaic array units, and energy storage units) to be independently created, combined, and upgraded. When power plants are expanded, new energy sources are added, or equipment is replaced, only the sub-twin modules need to be added, removed, or reconstructed, without rewriting the entire system. Compared to traditional static models, this invention has significant advantages in terms of scalability and maintainability.

[0115] Fourth, by constructing a cross-terminal time-series fusion mechanism, this invention can automatically handle problems such as inconsistent sampling periods, asynchronous delays, and missing data, achieving dynamic alignment of data from different sources in the digital twin model. This mechanism enables the system to obtain a complete, continuous, and unified global power operation curve for the first time, which is beneficial for forming more accurate power flow analysis and operation trend judgment.

[0116] Fifth, this invention utilizes a parameter self-learning mechanism within the twin to automatically fine-tune the internal model of the twin based on factors such as equipment aging, temperature changes, and load fluctuations. Traditional methods require manual recalibration or adjustment of model parameters, while this invention automatically updates and optimizes, significantly reducing manual maintenance costs and improving equipment operational accuracy and stability.

[0117] VI. This invention utilizes a cross-terminal message propagation mechanism within the digital twin system to enable various devices, such as energy storage, inverters, switching equipment, and power quality devices, to share status information and achieve coordinated scheduling. For example, it can realize functions such as balanced power generation from multiple inverters, coordinated charging and discharging of energy storage systems, and optimized load-side matching, thereby improving the overall energy utilization efficiency and stability of the system.

[0118] VII. The twin of this invention can automatically generate future operating trends based on the current state, including equipment health trends, power change trends, voltage deviation trends, etc. Compared with traditional methods that rely on manual experience or historical reports, this invention can achieve real-time prediction and risk warning, which is conducive to taking maintenance measures in advance and reducing the risk of power outages.

[0119] 8. By continuously recording the equipment's operating status and learning through digital twins, this invention can construct a complete digital life archive for equipment from installation, commissioning, operation to decommissioning. This archive can provide a basis for equipment operation and maintenance, maintenance planning, fault tracing, and replacement decisions, making operation and maintenance more scientific, precise, and quantifiable.

[0120] Example 2: like Figure 2 As shown, to solve the above-mentioned technical problems, based on Embodiment 1, another technical solution adopted in this application is: a multi-terminal collaborative modular power operation and maintenance method based on digital twins, the method including the following steps: Step 1: Deploy data computing units at the edge layer and cloud layer respectively, establish a unified topology meta-model library, and configure a unified timestamp protocol and clock synchronization strategy for all network data sources; Step 2: Construct a multi-layer topology graph containing electrical, functional, and business dimensions based on the aforementioned topology meta-model library, and map the real-time collected power data into the dynamic attributes of the graph nodes; Step 3: Construct and train a time-series topology graph neural network that incorporates the physical constraints of the power grid. Run a lightweight model at the edge for local state estimation and a high-precision model in the cloud for global state fusion. Step 4: Perform real-time synchronization between the cloud twin and the real network data, calculate the residual between the cloud twin state and the real network collected data to detect abnormal differences, and use the attention weights of the graph neural network to locate the source of the difference; Step 5: Classify the detected differences, perform online adaptive correction according to the difference type, and dynamically correct the parameters of the digital twin model through parameter inversion and online learning mechanisms; Step 6: Map the analysis results output by the graph neural network into modular operation and maintenance action units, and distribute them to multiple terminals through the operation and maintenance bus for collaborative execution and work order closure.

[0121] In this embodiment, specifically, constructing the multi-layer topology graph includes: Static attributes of nodes and edges are extracted from the meta-model library to construct electrical topology layer, functional topology layer and business topology layer; A partitioned modeling strategy is adopted, forming subgraph groups by geographical or power grid segmentation, and establishing convergence edges between different subgraphs. Cross-regional message passing and global consistency checks are performed through these convergence edges.

[0122] In this embodiment, specifically, the construction and training of the time-series topology graph neural network incorporating power grid physical constraints includes: The message passing unit is designed to handle both node timing characteristics and topological space characteristics, and incorporates soft coding of physical constraints on power flow direction and device capacity boundaries into the message function; By using structural pruning and hybrid precision quantization techniques, a lightweight edge model deployed on edge nodes and a high-precision cloud model deployed in the cloud are generated, forming a binary model system.

[0123] In this embodiment, specifically, the synchronization and difference detection of real-time twin and real-network data includes: Edge nodes send summaries of their local state estimates to the cloud, and the cloud graph neural network receives the summaries of each subgraph and performs global message fusion. Based on the graph-level residual monitoring mechanism, the cloud twin status is compared with the real network aggregated data. When an over-threshold alarm is triggered, the graph information backtracking mechanism is activated to locate the source of the difference in the device, line or data link.

[0124] In this embodiment, specifically, classifying the detected differences and performing online adaptive correction includes: A graph-level classifier is used to identify differences as data problem type, equipment behavior deviation type, or external disturbance type. For data discrepancies, trigger edge retransmission or interpolation to fill in the gaps; For deviations in equipment behavior, perform parameter inversion to estimate the effective parameter offset of the equipment, and issue parameter correction suggestions; For external perturbation-type differences, the short-term predicted distribution of the twin model is updated in a graph-level mode.

[0125] In this embodiment, specifically, the step of dynamically correcting the parameters of the digital twin model through parameter inversion and online learning mechanisms further includes: Based on the difference correction results, an incremental learning mechanism is used to correct the weights of the graph neural network model. After batch updates are completed in the cloud, the data is periodically distributed to the edge model. If the device's performance parameters deviate from their normal range for an extended period, the node's parameters will be updated in real time in the digital twin model, and a maintenance work order containing the topology context will be automatically generated.

[0126] In this embodiment, specifically, mapping the analysis results output by the graph neural network into modular operation and maintenance action units includes: Define standardized operation and maintenance action units, which include automatic alarm, priority dispatch, temporary isolation and power reallocation; Set action rollback thresholds and manual confirmation procedures for high-voltage or ultra-high-voltage nodes, and register all twin scheduling actions and model correction actions on an immutable audit chain.

[0127] In this embodiment, data acquisition and computing units are deployed at the edge and cloud layers respectively: the edge layer is responsible for collecting high-frequency time-series data and event logs from field devices; the cloud layer deploys a network-wide topology twin and a global graph neural network engine.

[0128] Physical devices, sites, lines, power electronic terminals, work order systems, and operation and maintenance terminals are uniformly registered in the system to establish a unified topology meta-model library. This library records the metadata format of the static topology attributes and variable operational attributes of each node and edge.

[0129] Configure a unified timestamp protocol and clock synchronization strategy for each data source, preferably NTP / PTP, and set edge buffering and retransmission strategies to ensure message ordering and reliability.

[0130] For example: The circuit breakers, busbars, and instrument transformers of substation A are entered into the topology library as nodes; the photovoltaic inverters and energy storage units are mapped as nodes with control interfaces as grid-connected terminals, and all devices complete static description registration when they are first connected.

[0131] In this embodiment, specifically, a multi-layer topology diagram is constructed from the meta-model library and real-time data, including but not limited to: electrical topology layer (physical devices and connections), functional topology layer (control and protection logic), and business topology layer (operation and maintenance units and team relationships).

[0132] To reduce computational complexity, a partitioned modeling approach is adopted: subgraphs are formed by geographical or power grid segmentation, and convergence edges are established between subgraphs for cross-regional message passing and global consistency checks.

[0133] Each node and edge carries multiple types of attributes: static attributes (equipment type and rating), short time series (current, voltage, temperature and vibration, etc.), event logs (alarms and work orders), and operation and maintenance context (task assignment and personnel location).

[0134] For example, a sub-graph is created for the 750kV transmission corridor, and another sub-graph is created for the substation and the photovoltaic-storage cluster. The two sub-graphs are connected by a convergence edge through the access site to conduct cross-layer impact analysis.

[0135] In this embodiment, specifically, the message passing unit of the temporal topology graph neural network simultaneously processes the temporal characteristics of node states and the spatial characteristics of the topology; soft coding of power grid physical constraints (such as power flow direction constraints and equipment capacity boundaries) is incorporated into the message function, so that the graph neural network retains both learning ability and physical interpretability.

[0136] The graph neural network was trained using historical operational data, historical fault records, and work order logs for several sub-tasks: state reconstruction, short-term load prediction, fault propagation probability estimation, and difference prediction. During training, a hierarchical loss function was used to constrain both local node accuracy and global energy consumption / stability metrics.

[0137] To meet the needs of edge deployment, the trained graph neural network is structurally pruned and mixed precision quantization is performed to generate a binary model system consisting of a lightweight edge model and a high-precision cloud model.

[0138] For example, a model can be trained using historical load and event logs from the past two years to predict the possible voltage drop propagation path within the next 30 minutes after a circuit breaker alarm occurs.

[0139] In this embodiment, specifically, edge nodes send time-series data to the local edge graph neural network. The edge graph neural network first performs local state estimation and sends the summary (compressed representation) to the cloud graph neural network. The cloud graph neural network receives the summaries of each subgraph and performs global message fusion to output the overall network twin state estimate.

[0140] The cloud-based twin status is compared online with the aggregated data collected from the real network, and graph-level residual monitoring (node ​​residual, path residual, and group residual) is used to automatically label the spatial distribution and time series of the differences.

[0141] When an over-threshold alarm is triggered, a graph information backtracking mechanism is activated to quickly locate possible sources of discrepancies (devices, lines, or data links) using the attention weights of the graph neural network.

[0142] For example, when an edge graph neural network detects a rise in temperature and an abnormal current at a site, it immediately makes an estimate locally and informs the cloud. The cloud then combines data from surrounding sites to determine whether this is a decrease in the efficiency of a local inverter or a loss in upstream power transmission.

[0143] In this embodiment, the detected differences are specifically classified into data problem type (timing misalignment and packet loss), equipment behavior deviation type (actual performance degradation) and external disturbance type (climate or sudden load), and a graph-level classifier is used to automatically identify the difference type.

[0144] Trigger edge retransmission, timestamp backtracking calibration, and interpolation filling mechanisms; perform time-series reordering of critical paths to restore consistency between twins and the real network.

[0145] The parameter inversion module is executed in the cloud. It estimates the effective parameter offset of the equipment through reverse engineering of local subgraphs and sends parameter correction suggestions to the edge or field maintenance system, generating work orders when necessary.

[0146] Trigger temporary policy adjustments (such as temporarily reducing some load, adjusting grid connection strategy, and switching microgrid operation mode), and update the short-term forecast distribution of the twin model in graph-level mode.

[0147] Based on the difference correction results, a small-batch online learning mechanism (reinforcement learning or incremental learning) is used to correct the weights of the graph neural network model. Batch updates are first completed in the cloud and then periodically distributed to the edge model.

[0148] For example, if it is found that the discharge capacity of a group of energy storage units decreases throughout the day, it can be inferred through inversion that the battery health has declined. The system will automatically adjust its available capacity, generate a maintenance work order, and update the node parameters in real time in the twin.

[0149] In this embodiment, the graph neural network output (anomaly location, propagation path, risk level, and recommended strategy) is specifically transformed into modular operation and maintenance action units, including: automatic alarm, priority dispatch, temporary isolation, local scheduling, and power reallocation. Each action unit defines a standard input / output interface and fallback logic.

[0150] Action execution is distributed to relevant terminals (on-site maintenance personnel, remote control system, SCADA interface and microgrid controller) via the operation and maintenance bus, and execution feedback is recorded as a basis for further correction of the twin.

[0151] When a device malfunction or performance degradation is identified, the system automatically generates a work order with a topology context. The work order includes the scope of the malfunction, recommended operating steps, priority, and required spare parts information.

[0152] For example, if the system determines that overheating of a line joint will lead to a risk of load transfer to an adjacent substation, it will automatically trigger a temporary load redistribution and issue a high-priority work order to the corresponding maintenance team.

[0153] In this embodiment, key information (alarms, topology changes, and work orders) is synchronized using primary and backup channels, and the edge must retain local operating policies to ensure safe operation during temporary disconnections.

[0154] All twin scheduling actions and model correction actions are registered on a blockchain or an immutable audit chain (optional implementation) to support incident backtracking.

[0155] For high-voltage / ultra-high-voltage nodes, stricter action rollback thresholds and manual confirmation procedures should be set to prevent automated decision-making from causing a wider range of impacts.

[0156] For example, when the system recommends disconnecting a main outgoing line, if the line is a 750kV main line, the system must simultaneously meet multiple safety conditions and execute the operation after confirmation by two operators.

[0157] In this embodiment, specific factors include twin-to-real network residuals, fault prediction accuracy, root cause location time, average work order closed-loop time, and the impact on PUE / system stability.

[0158] Regularly conduct A / B comparison trials: Maintain two models (experimental group and control group) in the cloud to verify the benefits and stability of online learning. If online updates cause a drop in KPIs, automatically roll back to the historical stable model and trigger manual review.

[0159] For example, using average root cause localization time as the main KPI, the goal is to reduce it to half the current system's time.

[0160] Example 3: like Figure 3 As shown, to solve the above-mentioned technical problems, based on Embodiment 1, another technical solution adopted in this application is: a multi-terminal collaborative modular power operation and maintenance system based on digital twins, comprising: The topology building module is configured to create a global power topology graph of power energy sites and map the static attributes of devices to the graph structure; The data synchronization module is configured to collect real-time data and perform time stamp alignment and resampling to ensure the synchronization of multi-source data within the same time slice; The intelligent correction module is configured to perform node state propagation and global consistency correction based on graph neural networks, and to perform cross-subgraph co-simulation. The difference optimization module is configured to monitor the difference between the predicted value and the true value, trigger an early warning, and dynamically update the model parameters based on a self-learning mechanism. The operation and maintenance interaction module is configured to output a consistent digital twin graph to a multi-terminal platform, providing status monitoring, fault location, and collaborative scheduling functions.

[0161] For other details regarding the implementation technical solutions of each module in the above embodiment system, please refer to the description in the multi-terminal collaborative modular power operation and maintenance method based on digital twin in the above embodiment one, which will not be repeated here.

[0162] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For system-type embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0163] Example 4: like Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present disclosure. It illustrates a structural schematic diagram suitable for implementing the electronic device in the embodiment of the present disclosure. Figure 4 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0164] like Figure 4 As shown, the electronic device includes a processor, a memory, and a communication interface. The memory stores a computer program, and when the processor executes the computer program, it implements the multi-terminal collaborative modular power operation and maintenance method based on digital twins as described in the embodiments of this disclosure. The electronic device can exchange data with other devices or systems through the communication interface to achieve real-time updates and sharing of drug information.

[0165] The processor in the aforementioned electronic device serves as its core, responsible for executing the computer program stored in the memory to implement various functions of the paperless conference terminal's intelligent interaction method. The processor can employ a high-performance multi-core CPU or a dedicated chip to meet the demands of complex calculations and real-time processing. The memory stores the operating system, applications, data, and computer programs. In this embodiment, the memory stores the computer program implementing the paperless conference terminal's intelligent interaction method. The memory can be RAM, ROM, Flash memory, or other types of non-volatile memory. The communication interface connects the electronic device to other devices or networks, enabling data transmission and exchange. In this embodiment, the communication interface supports multiple communication protocols and interface standards, such as Wi-Fi, Bluetooth, USB, and Ethernet, to meet communication needs in different scenarios.

[0166] For a detailed description of this embodiment, please refer to the corresponding descriptions in the foregoing embodiments, which will not be repeated here.

[0167] Example 5: According to embodiments of the present disclosure, a computer-readable storage medium stores a computer program, which, when executed by a processor, implements the functions of the aforementioned digital twin-based multi-terminal collaborative modular power operation and maintenance method of the present disclosure.

[0168] The aforementioned computer-readable storage media include, but are not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or portable hard drive), media with built-in rewritable non-volatile memory (e.g., memory card), and media with built-in ROM (e.g., ROM cartridge).

[0169] For a detailed description of this embodiment, please refer to the corresponding descriptions in the foregoing embodiments, which will not be repeated here.

[0170] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-terminal collaborative modular power operation and maintenance method based on digital twins, characterized in that, Includes the following steps: Construct a global power topology map of power energy sites, mapping power equipment, sensors and connecting lines as graph nodes and graph edges, and binding the static attributes of equipment to the topology structure; Acquire real-time operating data of power equipment, assign unified time labels to multi-source heterogeneous data, and align data of different frequencies to the same time slice through resampling technology. Then bind the aligned data as dynamic attributes of nodes and edges. The state propagation calculation of the global power topology graph is performed based on the graph neural network, and the electrical correlation between nodes is analyzed by the graph message passing mechanism to perform global consistency correction of the equipment operating status. The photovoltaic, energy storage and power grid subsystems are mapped as independent subgraphs. Based on the main wiring diagram, the multiple subgraphs are integrated and simulated collaboratively, and the energy flow state across subsystems is coupled and corrected. The system monitors the difference between the actual value of the equipment operation data and the predicted value of the digital twin model in real time. When the difference exceeds the preset threshold, it triggers an operation and maintenance warning and dynamically updates the edge weight parameters of the topology graph based on long-term operation data. The corrected digital twin data is output to a multi-terminal operation and maintenance platform, which provides equipment status monitoring, fault location and collaborative scheduling decisions based on a consistent digital twin graph.

2. The multi-terminal collaborative modular power operation and maintenance method based on digital twins according to claim 1, characterized in that, The construction of the global power topology map for power energy stations specifically includes: Based on the main wiring diagram and site layout diagram of the power station, transformers, inverters and switchgear equipment are defined as diagram nodes, and cables, busbars and protection circuits are defined as diagram edges; The equipment type, rated voltage, rated current, and installation location information are stored as static attributes in the corresponding graph nodes, forming a digital infrastructure that reflects the physical connection relationships.

3. The multi-terminal collaborative modular power operation and maintenance method based on digital twins according to claim 1, characterized in that, The process of assigning unified time tags to multi-source heterogeneous data and aligning data of different frequencies to the same time slice using resampling technology specifically includes: Real-time data from power equipment is uploaded via edge acquisition devices, and the data is managed hierarchically according to the sampling frequency. By setting a unified time base, for data with inconsistent sampling frequencies, resampling or interpolation algorithms are used to map them into the same time window, ensuring that the dynamic attributes of all devices are kept in time synchronization in the global power topology map.

4. The multi-terminal collaborative modular power operation and maintenance method based on digital twins according to claim 1, characterized in that, The state propagation calculation of the global power topology graph based on graph neural networks specifically includes: By utilizing the message passing mechanism of graph neural networks, abnormal state features are propagated along the topology between device nodes; Based on the physical constraints of current, voltage, and power attributes between upstream and downstream nodes, the output status values ​​of the nodes are automatically adjusted to eliminate local measurement errors and maintain the conservation and consistency of power flow data across the entire network.

5. The multi-terminal collaborative modular power operation and maintenance method based on digital twins according to claim 1, characterized in that, The mapping of different subsystems of photovoltaics, energy storage, and the power grid into independent subgraphs specifically includes: The photovoltaic power generation system, energy storage system and distribution network side are treated as independent sub-graph structures. By establishing connection edges between subgraphs through power transmission paths, device nodes in different subgraphs can communicate and interact across subgraphs, realizing cross-system state sharing and boundary condition constraints.

6. The multi-terminal collaborative modular power operation and maintenance method based on digital twins according to claim 1, characterized in that, The coupling correction of the energy flow state across subsystems specifically includes: Through a cross-subgraph collaborative simulation mechanism, the data coupling relationship between the photovoltaic subsystem, the energy storage subsystem, and the power distribution system is calculated in real time. The system jointly adjusts the equipment status of each subsystem at the same time, eliminates the data silo effect caused by independent calculation of factor systems, and ensures that the collaborative operation logic between multiple systems conforms to physical laws.

7. The multi-terminal collaborative modular power operation and maintenance method based on digital twins according to claim 1, characterized in that, The difference between the actual value of the real-time monitoring equipment's operating data and the predicted value of the digital twin model specifically includes: Construct a difference analysis model to calculate in real time the difference between the predicted state of each device node and the actual value uploaded by the sensor; When the difference exceeds the set threshold range, an early warning signal is automatically triggered, and a graph neural network is used to backtrack and analyze the source of the difference, identify abnormal data sources, and automatically perform data cleaning or correction.

8. The multi-terminal collaborative modular power operation and maintenance method based on digital twins according to claim 7, characterized in that, The edge weight parameters of the topology graph that are dynamically updated based on long-term operational data specifically include: Collect historical data on equipment operation and real-time difference analysis results; The graph neural network is trained online using an incremental learning method. The device model parameters are adjusted according to the aging of the device or changes in the environment, and the node weights and edge weights in the topology graph are dynamically updated to maintain the fidelity of the digital twin model.

9. The multi-terminal collaborative modular power operation and maintenance method based on digital twins according to claim 1, characterized in that, The step of outputting the corrected digital twin data to the multi-terminal operation and maintenance platform specifically includes: Generate a visual digital twin containing data on equipment health status, power flow, real-time load, and predictive maintenance. It supports multiple user terminals to access equipment information at different levels according to their permissions, and transmits decision data to the dispatching system in real time to support the overall dispatching and emergency response of large-scale power grids.

10. A multi-terminal collaborative modular power operation and maintenance system based on digital twins, applied to the multi-terminal collaborative modular power operation and maintenance method based on digital twins as described in any one of claims 1-9, characterized in that, include: The topology building module is configured to create a global power topology graph of power energy sites and map the static attributes of devices to the graph structure; The data synchronization module is configured to collect real-time data and perform time stamp alignment and resampling to ensure the synchronization of multi-source data within the same time slice; The intelligent correction module is configured to perform node state propagation and global consistency correction based on graph neural networks, and to perform cross-subgraph co-simulation. The difference optimization module is configured to monitor the difference between the predicted value and the true value, trigger an early warning, and dynamically update the model parameters based on a self-learning mechanism. The operation and maintenance interaction module is configured to output a consistent digital twin graph to a multi-terminal platform, providing status monitoring, fault location, and collaborative scheduling functions.

Images