A power grid digital twin modeling method and system based on dynamic assembly and update

Abstract

This invention discloses a method and system for power grid digital twin modeling based on dynamic assembly and updating. The method includes: completing device-level virtual-physical mapping of power grid equipment based on a preset standard data structure, assembling it into a cluster-level model and performing boundary behavior fusion; achieving system integration by constructing electrical quantity coupling relationships to obtain a first system-level digital twin model; quantifying the error uncertainty between the model and the physical entities of the power grid through information entropy, and using dual-path differentiated updates combined with hierarchical resource scheduling to obtain a second system-level digital twin model. This invention also constructs a closed loop of "perception-assembly-simulation-update," enabling the self-evolution and continuous optimization of the digital twin model. This invention achieves consistency in multi-level model assembly and dynamic self-updating of the model, significantly improving the construction efficiency, simulation accuracy, real-time performance, and engineering applicability of the digital twin model, providing core support for intelligent operation and maintenance and precise decision-making in the power grid.

Description

Technical Field

[0001] This invention belongs to the field of power systems, and particularly relates to a method and system for digital twin modeling of power grids based on dynamic assembly and updating. Background Technology

[0002] As a core technology for digital transformation, power grid digital twins achieve panoramic perception of power grid status, in-depth analysis of operational patterns, and intelligent support for business decision-making by constructing a precise mapping of all elements of the physical power grid in the digital space. Essentially, it uses data to simulate the real operating state of the power grid in the digital space, supporting real-time simulation, predictive modeling, and closed-loop optimization. Digital twin power grids can facilitate renewable energy generation and load forecasting, achieve local balancing of distributed energy resources, and effectively address the stability challenges brought about by high-proportion renewable energy integration. Simultaneously, through global data connectivity and business reshaping, digital twin technology can significantly improve the efficiency of power grid status perception, analysis and decision-making, and resource allocation, providing key technical support for building a new power system and achieving "dual-carbon" goals.

[0003] The accuracy and real-time performance of digital twin modeling directly determine its application effectiveness. An ideal digital twin model needs to achieve bidirectional dynamic mapping between physical devices and virtual objects, meaning that the model's state is updated in real time through sensor data, and model feedback is used to optimize the operation of the physical system. In the operation and maintenance of power transmission and transformation equipment, the model needs to synchronously reflect the coupling characteristics of multiple physical fields such as current, temperature, and mechanical deformation in order to accurately predict the health status of the equipment. The modeling process needs to integrate mechanistic rules and data-driven methods to build a unified model system covering the entire "source-grid-load-storage" process, supporting full lifecycle management from the equipment level to the system level. If the model deviates significantly from the physical entity or is updated lagging behind schedule, it will directly lead to inaccurate simulations, ineffective decisions, and even trigger cascading failures.

[0004] Despite the enormous potential of digital twin technology for power grids, existing modeling methods still have significant limitations:

[0005] 1. Traditional modeling relies on full data synchronization, with non-critical parameters consuming significant computational resources, resulting in second-level delays in updating critical equipment models and making it difficult to capture transient fault characteristics. Model updates depend on global optimization through knowledge fine-tuning, making it difficult to balance update accuracy and computational load.

[0006] 2. The heterogeneous sources and inconsistent standards of power grid data create "data silos," and the lack of complete unification of cross-disciplinary data models leads to timing misalignment during multi-physics coupling modeling. Furthermore, the assembly of heterogeneous equipment models lacks structured support, and the update mechanism does not quantify the uncertainty of errors between the model and the entity. Relying solely on fixed triggering conditions can easily lead to update lags or resource waste, making it difficult to achieve accurate self-evolution of the model.

[0007] 3. Traditional mechanistic models are difficult to describe the behavior of distributed resources that are nonlinear and highly random, while pure data-driven models lack physical constraints and are prone to violating fundamental laws such as the law of conservation of energy. Summary of the Invention

[0008] Objective of this invention: The objective of this invention is to provide a power grid digital twin modeling method based on dynamic assembly and updating. This method enables multi-level model closed-loop optimization, data-mechanism hybrid driving, and intelligent resource scheduling, significantly improving the construction efficiency, simulation accuracy, real-time performance, and engineering applicability of the digital twin model, thus providing core support for intelligent operation and maintenance and precise decision-making in the power grid. Another objective of this invention is to provide a power grid digital twin modeling system based on dynamic assembly and updating.

[0009] Technical solution: According to a first aspect of the present invention, the power grid digital twin modeling method based on dynamic assembly and updating includes:

[0010] S1. Based on the preset standard data structure, complete the virtual-physical mapping of power grid equipment at the equipment level, assemble the equipment-level model into a cluster-level model and perform boundary behavior fusion, realize system integration by constructing electrical quantity coupling relationship, and obtain the first system-level digital twin model;

[0011] S2. Based on information entropy, quantify the current error uncertainty between the first system-level digital twin model and the power grid physical entity, and trigger a model update request based on the entropy threshold; obtain the equipment importance of each power grid device based on future operation risk assessment; combine the information entropy change magnitude and the equipment importance, and update the first system-level digital twin model using a dual-path differentiated update method, and simultaneously perform hierarchical scheduling and dynamic resource allocation based on power grid topology partitioning to obtain the second system-level digital twin model.

[0012] Optionally, the preset standard data structure adopts a standard data structure table, which includes a geometric model index of power grid equipment, a static attribute dictionary, a set of behavioral rules, and a data flow interface.

[0013] Optionally, the device-level virtual-real mapping includes the virtual-real mapping of the device's physical structure, state, and control behavior.

[0014] Optionally, the boundary behavior fusion algorithm includes: based on the power grid business topology relationship, by comparing the predicted outputs of adjacent device-level models or arbitrary-level cluster-level models at the topology connection point, and using a sliding mode control law or consensus algorithm to coordinate the model state.

[0015] Optionally, the electrical quantity coupling relationship is constructed using differential-algebraic equations, as shown in the following formula:

[0016]

[0017] in, For time, Let the system state variable vector be... For the differential term of the state variable, The system input vector;

[0018] The differential-algebraic equations contain algebraic constraints formed by the nodal power balance equations; the nodal power balance equations include:

[0019] Active power balance equation:

[0020]

[0021] Reactive power balance equation:

[0022]

[0023] in, , The active and reactive power injected into the nodes, , The active and reactive power of the node load. , For nodes , voltage amplitude, For nodes , voltage phase angle difference, , Let be the real and imaginary parts of the network admittance matrix.

[0024] Optionally, the differential algebraic equation is solved using the Newton-Raphson method.

[0025] Optionally, the importance of the equipment is dynamically determined through forward simulation and decision analysis, including: injecting typical operating scenarios or extreme disturbance conditions into the first system-level digital twin model to perform forward simulation to predict the future state of the power grid; assessing the system risk level based on the simulation results, and assigning high equipment importance to power grid equipment at high risk levels to guide the dual-path differentiated update and generate pre-control strategies to support scheduling decisions.

[0026] Optionally, the forward simulation includes, but is not limited to, calculating the dynamic trajectories of key state variables of the power grid such as node voltage, line power and frequency in future time periods, and predicting the transient stability and energy flow distribution trend of the power grid system.

[0027] Optionally, the preset system risks include, but are not limited to, equipment overload, voltage exceeding limits, and current exceeding limits, which can be extracted from the dynamic trajectory.

[0028] Furthermore, based on the simulation results, a probabilistic security assessment algorithm is used to quantify the system risk level of the first system-level digital twin model in various scenarios, thereby automatically identifying potential weaknesses and security risks.

[0029] Optionally, the probabilistic security assessment algorithm includes:

[0030] By employing the Monte Carlo method combined with importance sampling techniques, a large number of random samples are taken from the uncertainties of renewable energy output and load to generate a massive dataset of power grid operating conditions containing uncertainties. Under certain operating conditions, historical statistical data is used to predict the probability of various preset system risks occurring under those conditions. ;

[0031] Define risk indicators ,in For the first The probability of a pre-defined system risk occurring. Costs of load shedding or equipment damage resulting from this risk;

[0032] Based on the relevant risk indicators and preset thresholds The system automatically classifies risk levels and identifies risk areas and key contributing factors.

[0033] Optionally, the generation of the pre-regulation strategy includes: combining the risk assessment results, generating a pre-regulation strategy set through a multi-objective optimization algorithm, and determining the optimal strategy based on sensitivity analysis.

[0034] Optionally, the forward simulation is performed using a numerical simulation method.

[0035] Preferably, the numerical simulation method includes:

[0036] The numerical solution method for differential-algebraic equations is adopted, combined with the implicit trapezoidal integral method to ensure the numerical stability of the simulation. Its discretization form is as follows:

[0037]

[0038] in, The integration step size is... , The first and The state vector of the step-by-step digital twin model, The function described by the system's differential equations. The state vector of the digital twin model represents the equivalent virtual topology of the entire power grid and its dynamic energy flow process, constructed by the above-mentioned fusion rules of node power balance and boundary behavior.

[0039] Preferably, the simulation uses parallel computing technology to decompose the task into multi-core processors or computing clusters, thereby achieving a linear increase in simulation speed.

[0040] Optionally, updating the system-level digital twin model using a dual-path differentiated update method includes:

[0041] When the information entropy change is lower than the severity threshold and the corresponding device importance is identified as a non-core node, a lightweight parameter correction is adopted, and the parameters of the first system-level digital twin model are quickly adjusted through an online learning algorithm.

[0042] When the information entropy change exceeds the severity threshold or the corresponding equipment is identified as a core critical equipment, the first system-level digital twin model is reconstructed using a physical-guided neural network, and the physical laws of the power grid are embedded as constraints into the neural network loss function for training.

[0043] Optionally, the reconstructing of the first system-level digital twin model using a physically guided neural network includes:

[0044]

[0045] in, The mean square error between the predicted values ​​of the first-level digital twin model and the measured data is denoted as . For physical constraints, This is a hyperparameter used to balance the contributions of data-driven terms and physical law terms.

[0046] Optionally, the hierarchical scheduling includes:

[0047] The update regions are divided based on the power grid topology, and priority is assigned to the model update tasks in different regions. Among them, core equipment with high equipment importance is updated synchronously at high frequency, while edge equipment with low equipment importance is updated asynchronously at low priority.

[0048] The dynamic resource allocation is used to balance computing load and real-time update requirements.

[0049] Optionally, the dynamic resource allocation is shown in the following formula:

[0050]

[0051] in, To update the priority score, Let x be the information entropy, and let x be the time series of the target device's operating error during the evaluation period; This is the equipment importance coefficient. The ratio of available computing resources; This is a weighting coefficient, which is dynamically adjusted according to the actual operating strategy.

[0052] Optionally, the method further includes:

[0053] The deviations between the state prediction data output by the second system-level digital twin model and the real-time multi-source operating data of the power grid equipment, as well as the response time and data under specific operating conditions, are identified. Based on the deviations, the accuracy and real-time performance of the second system-level digital twin model are evaluated. When the deviations exceed the limits, an offline deep optimization process is initiated. The offline deep optimization includes: identifying the root causes of problems through error source analysis, and using Bayesian optimization or low-level deep reconstruction based on physical guided neural networks for targeted correction; at the same time, the verification and optimization experiences of each time are stored in a knowledge base, and a reinforcement learning training strategy network is used to optimize the decision parameters of the dual-path differentiated update and hierarchical scheduling.

[0054] Optionally, the deviation between the state prediction data output by the second system-level digital twin model and the real-time collected multi-source operation data of power grid equipment is quantified by indicators such as mean square error, and used to evaluate the accuracy and real-time performance of the model.

[0055] Optionally, the step of identifying the root cause of the problem through error source analysis and performing targeted corrections using Bayesian optimization or deep reconstruction based on physical-guided neural networks includes:

[0056] For parameter drift caused by equipment aging or environmental changes, Bayesian optimization algorithm is used to make targeted corrections to key hyperparameters such as static impedance and control gain in the second system-level digital twin model.

[0057] For structural deviations caused by unmodeled dynamics (such as complex nonlinear control of new energy sources), a physical-guided neural network is used to use the differential algebraic equations of the power grid as loss function constraints to perform a deep reconstruction and correction of the virtual-real mapping structure inside the model.

[0058] According to a second aspect of the present invention, the power grid digital twin modeling system based on dynamic assembly and updating includes: a data sensing module, a model assembly module, an update scheduling module, and a verification feedback module that interact sequentially in a closed loop.

[0059] The data sensing module is used to collect the operating data of the physical entities of the power grid, perform standardized processing, and then transmit the data to the model assembly module.

[0060] The model assembly module is used to complete the equipment-level virtual-real mapping of power grid equipment based on the standardized operating data and combined with the preset standard data structure, assemble the equipment-level model into a cluster-level model and execute the boundary behavior fusion algorithm, construct the electrical quantity coupling relationship through differential algebraic equations to realize system integration, and output the first system-level digital twin model.

[0061] The update scheduling module is used to quantify the current error uncertainty between the first system-level digital twin model and the power grid physical entity based on information entropy, and trigger a model update request based on the entropy threshold; obtain the equipment importance of each power grid device based on future operation risk assessment; update the first system-level digital twin model using a dual-path differentiated update method in combination with the information entropy change magnitude and the equipment importance, and simultaneously perform hierarchical scheduling and dynamic resource allocation based on power grid topology partitioning, and output the second system-level digital twin model;

[0062] The verification feedback module is used to verify the model accuracy of the second system-level digital twin model and feed the verification results back to the model assembly module and the update scheduling module to realize the self-evolution optimization of the system-level digital twin model.

[0063] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0064] 1. This invention breaks down the barriers of data silos and splicing distortion in traditional heterogeneous models by dynamically assembling and fusionling device-level models. It realizes structured, pluggable, and dynamic assembly from device to system level, which greatly reduces the construction time and computing cost of panoramic digital twin models and ensures the numerical consistency of multiphysics simulation.

[0065] 2. This invention employs a dual-path differentiated update method to update the system-level digital twin model, abandoning the traditional fixed-threshold, full-scale blind update mode. It couples uncertainty quantification with dynamic resource scheduling, performing lightweight corrections on non-core or low-deviation nodes and deep reconstruction on high-deviation core nodes. While ensuring high-fidelity global mapping, it significantly reduces model update latency, meeting the stringent real-time requirements of power grid transient control.

[0066] 3. This invention provides a dual-path differential update method that includes physical-guided neural network reconstruction, overcoming the shortcomings of purely data-driven models that are prone to non-physical phenomena such as violations of energy conservation under complex and unknown disturbances in the power grid. By embedding mechanistic constraints such as Kirchhoff's laws of the power grid into the loss function, the advantages of data-driven and physical mechanisms are complemented, significantly improving the prediction generalization ability and robustness of the digital twin model under extreme conditions. Attached Figure Description

[0067] Figure 1This is a flowchart of the method of the present invention;

[0068] Figure 2 The structured dynamic assembly flowchart provided for this invention;

[0069] Figure 3 A flowchart illustrating the dynamic update process of the digital twin model provided by this invention;

[0070] Figure 4 This is a flowchart illustrating the method of the present invention in one embodiment;

[0071] Figure 5 The flowchart illustrates the self-evolution and continuous optimization process of the digital twin model provided by this invention. Detailed Implementation

[0072] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0073] like Figure 1 As shown, this invention provides a method for digital twin modeling of power grids based on dynamic assembly and updating, comprising:

[0074] S1. Based on the preset standard data structure, complete the virtual-physical mapping of power grid equipment at the equipment level, assemble the equipment-level model into a cluster-level model and perform boundary behavior fusion, realize system integration by constructing electrical quantity coupling relationship, and obtain the first system-level digital twin model;

[0075] S2. Based on information entropy, quantify the current error uncertainty between the first system-level digital twin model and the power grid physical entity, and trigger a model update request based on the entropy threshold; obtain the equipment importance of each power grid device based on future operation risk assessment; combine the information entropy change magnitude and the equipment importance, and update the first system-level digital twin model using a dual-path differentiated update method, and simultaneously perform hierarchical scheduling and dynamic resource allocation based on power grid topology partitioning to obtain the second system-level digital twin model.

[0076] Example 1

[0077] Step S1 of the present invention realizes the structured dynamic assembly of twin models from the device level to the system level.

[0078] In this embodiment, as Figure 2 As shown, step S1 specifically includes:

[0079] S11. Construct a device-level model.

[0080] The preset standard data structure adopts a standard data structure table. Based on the standard data structure table, a unified geometric model, attribute information table, and behavioral logic file are established for power grid equipment such as circuit breakers, transformers, capacitors, and feeder switches. Through field binding, the virtual-physical mapping of the equipment's physical structure, state, and control behavior is completed, forming an equipment-level digital twin model.

[0081] Furthermore, the standard data structure table includes:

[0082] The field definitions of the standard data structure table include mapping relationships with the device geometric model, static attribute parameters, dynamic behavior rules, and real-time measurement data, so as to achieve structural unification and association binding of multi-source heterogeneous data;

[0083] The standard data structure table is defined using an object-oriented paradigm, creating a unique data template for each type of power grid equipment. The template includes:

[0084] The geometric model index field is used to associate and load the corresponding 3D mesh or point cloud model file;

[0085] A static attribute dictionary stores the device's nameplate parameters, physical constants, and topology connection point information in key-value pairs.

[0086] The behavior rule method set encapsulates differential equations or state machine logic that describe the dynamic characteristics of a device;

[0087] The real-time data stream interface defines the subscription and mapping rules with physical sensor data sources, ensuring that real-time measurement data can be updated to the corresponding state variables of the model at a preset frequency.

[0088] S12. Assemble the device-level model into a cluster-level model.

[0089] Using predefined topology-assembleable templates, multiple device-level models are automatically combined into cluster models such as feeder-level, distribution room-level, or subsystem-level models based on the business topology of the power grid; and a boundary behavior fusion algorithm is introduced to verify and ensure the consistency of the model's state and behavior logic at the boundary.

[0090] Furthermore, the boundary behavior fusion algorithm includes:

[0091] Based on the power grid business topology, by comparing the predicted outputs of adjacent device-level models or arbitrary-level cluster-level models at topology connection points, a sliding mode control law or consensus algorithm is used for state coordination to ensure the overall behavioral consistency of the cluster model. When the sliding mode control law is used, it specifically includes:

[0092] Define adjacent device models at connection points The power or state variable deviation is a sliding surface function:

[0093]

[0094] in, and respectively equipment and Output vector at boundary points. Design control law:

[0095]

[0096] in, For the control gain matrix, The symbolic function is used. The energy function is constructed using Lyapunov's second method. and ensure its derivative This causes the system state to be attracted to the sliding surface within a finite time. This enables fast and consistent convergence of the boundary states.

[0097] This algorithm effectively suppresses boundary power oscillations caused by model simplification or parameter mismatch, ensuring the numerical stability of cluster model simulation.

[0098] S13, System Integration, namely, system-level model fusion.

[0099] Specifically, this includes: at the system level, constructing coupling relationship equations between electrical quantities such as node voltage, current, power, and switching states based on differential algebraic equations, realizing the continuous expression of energy flow and information flow at the boundary of multiple sub-models, and completing the integration of system-level digital twin models.

[0100] Furthermore, the equations for constructing the coupling relationships between electrical quantities such as node voltage, current, power, and switching states based on differential-algebraic equations include:

[0101] The formula for the coupling relationship of differential-algebraic equations is:

[0102]

[0103] in, For time, Let the system state variable vector be... For the differential term of the state variable, This represents the system input vector, such as generator active power output and load power. The equation system is specifically composed of nodal power balance equations:

[0104] (1) Active power balance equation:

[0105]

[0106] (2) Reactive power balance equation:

[0107]

[0108] in, , The active and reactive power injected into the nodes, , The active and reactive power of the node load. , For nodes , voltage amplitude, For nodes , voltage phase angle difference, , Let represent the real and imaginary parts of the network admittance matrix. The Newton-Raphson method is used to solve this system of differential-algebraic equations, achieving a physically accurate simulation of energy flow and information dynamics in the system-level model, accurately reflecting the transient and steady-state operating characteristics of the power grid.

[0109] Example 2

[0110] like Figure 3 As shown, in this embodiment, step S2 specifically includes:

[0111] S21. Quantify the error uncertainty between the first system-level digital twin model and the power grid physical entity based on information entropy.

[0112] The information entropy of the multidimensional error sequence between the output of the digital twin model and the actual operating state of the physical entity is calculated in real time. The formula for the information entropy is as follows:

[0113]

[0114] in, For error sequences, For the first Error-like states, This represents the probability of this state occurring. When the information entropy value... When the threshold is exceeded, the uncertainty of the model is determined to increase significantly, and a model update request is automatically triggered.

[0115] S22. Combining the information entropy change magnitude and device importance, the system-level digital twin model is updated using a dual-path differentiated update method, specifically including:

[0116] When the information entropy change is below the severity threshold and the corresponding device is identified as a non-core node, a lightweight parameter correction is used, and the model parameters are quickly adjusted through an online learning algorithm. If the information entropy change exceeds the severity threshold or the corresponding device is identified as a core critical device, a physical-guided neural network reconstruction is triggered, and the physical laws of the power grid are embedded as constraints into the neural network loss function for training.

[0117] The physical-guided neural network reconstruction includes:

[0118]

[0119] in, The mean square error between the predicted values ​​of the first-level digital twin model and the measured data is denoted as . For physical constraints, This is a hyperparameter used to balance the contributions of data-driven terms and physical law terms.

[0120] S23. Perform regional hierarchical update scheduling of the model.

[0121] The update regions are divided based on the power grid topology, and priority is assigned to the model update tasks in different regions. Core equipment uses high-frequency synchronous updates, while edge equipment uses asynchronous low-priority updates. A dynamic resource allocation algorithm is used to balance computational load and real-time update requirements.

[0122] Furthermore, the method of balancing computational load and real-time update requirements through dynamic resource allocation algorithms includes:

[0123] Calculate the updated priority score using the dynamic resource allocation formula:

[0124]

[0125] in, To update priority scores; Here, x represents the information entropy, and x is the time series of the target device's operating error during the evaluation period, used to characterize the current error uncertainty of the model. The equipment importance coefficient is generated by dynamic mapping of the system risk level assessed in the forward simulation and is used to characterize the physical vulnerability of the equipment under future operating conditions. The ratio of available computing resources; These are weighting coefficients, dynamically adjusted based on the actual operating strategy. This mechanism ensures that when the model currently has significant errors (high...),... Furthermore, this node is highly susceptible to physical overruns in the future (high risk). When a system performs a deep refactoring, it prioritizes allocating computing resources to the system.

[0126] Example 3

[0127] like Figure 4 As shown, the power grid digital twin modeling method based on dynamic assembly and updating further includes: S3, completing verification feedback and closed-loop optimization, thereby closing with the previous steps S1 and S2 to form a closed loop of "perception-assembly-simulation-update" to realize the self-evolution and continuous optimization of the digital twin model.

[0128] like Figure 5 As shown, in this embodiment, the function of step S1 specifically includes:

[0129] Multi-source sensing and data fusion: Sensors deployed at various nodes of the power grid collect multi-source data such as electrical quantities, non-electrical quantities and environmental parameters in real time. Data cleaning, time alignment and spatial alignment are performed at the edge to form a unified and standardized data stream, providing real-time and consistent input for the model.

[0130] Dynamic model assembly and integration: Based on standard data structure tables and topology-assembleable templates, preprocessed data drives the updating of device-level and cluster-level models, and the coupling mechanism of differential-algebraic equations is used to complete the fusion of boundary states of multiple sub-models at the system level, ensuring the consistency and integrity of the entire system model in terms of structure, state and behavioral logic.

[0131] In this embodiment, the function of step S2 specifically includes:

[0132] Simulation and decision analysis: Typical operating scenarios or extreme disturbance conditions are injected into the first system-level digital twin model to perform forward simulation and predict the future state of the power grid; the system risk level is assessed based on the simulation results and pre-control strategies are generated to support scheduling decisions.

[0133] Dynamic update triggering and strategy execution: Based on the deviation between simulation results and actual operating data, and combined with the information entropy quantification model uncertainty, an update strategy of lightweight parameter correction or physical guidance neural network reconstruction is automatically triggered; computing resources are allocated based on equipment importance and regional priority to achieve differentiated update scheduling.

[0134] In this embodiment, the function of step S3 specifically includes:

[0135] Verification feedback and closed-loop optimization: The updated model output is compared with actual power grid operation data, and indicators such as mean square error and mean absolute error are calculated to verify the model accuracy. The verification results are fed back into the model assembly and update module through a feedback mechanism to form closed-loop control and continuously optimize the model's accuracy, stability and adaptability.

[0136] In one implementation, the dynamic assembly and integration of the model includes:

[0137] Based on standard data structure tables, a unified geometric model, attribute information table, and behavioral logic file are established for power grid equipment such as circuit breakers, transformers, capacitors, and feeder switches. Through field binding, the virtual and real mapping of the equipment's physical structure, status fields, and control behaviors is realized, forming a device-level digital twin model.

[0138] Using predefined topology-assembleable templates, multiple device-level models are automatically combined into a cluster model based on the electrical connection relationship of the power grid and business logic. In this process, a boundary behavior fusion algorithm is introduced to verify and ensure the consistency of electrical quantities and behavioral logic of the model at the boundary.

[0139] At the system level, coupling relationship equations between electrical quantities such as node voltage, current, power, and switching states are constructed based on differential algebraic equations, realizing the continuous expression of energy flow and information flow at the boundary of multiple sub-models; by associating the virtual model set and stacking it block by block and layer by layer, a digital twin full-space model is formed, completing the integration of the system-level digital twin model;

[0140] The assembled digital twin model is verified for structural completeness, field consistency, and behavioral correctness; the numerical stability and dynamic response accuracy of the model are verified through simulation to ensure that the model is feasible for simulation.

[0141] In one implementation, the simulation and decision analysis includes:

[0142] Based on the first system-level digital twin model, typical operating scenarios or extreme disturbance conditions are injected to construct a set of multiple simulation scenarios including steady state, transient state and fault state, providing diverse input conditions for the simulation.

[0143] The first system-level digital twin model is simulated and extrapolated using numerical simulation methods to calculate the dynamic trajectory of key state variables of the power grid, such as node voltage, line power, and frequency, in the future time period. This is used to predict the transient stability of the system and the energy flow distribution trend. Preset system risks such as equipment overload, voltage over-limit, and current over-limit are extracted from the dynamic trajectory.

[0144] Based on the simulation results, a probabilistic security assessment algorithm is used to quantify the system risk level of the first system-level digital twin model in various scenarios, automatically identify potential weak links and security risks, and assign high equipment importance to power grid equipment in high-risk areas to guide the execution of the dual-path differentiated update.

[0145] Based on the risk assessment results, a set of pre-regulation strategies is generated using a multi-objective optimization algorithm, and the optimal strategy is determined based on sensitivity analysis.

[0146] In one implementation, the forward simulation and deduction of the first system-level digital twin model using numerical simulation methods includes:

[0147] The numerical solution method for differential-algebraic equations is adopted, combined with the implicit trapezoidal integral method to ensure the numerical stability of the simulation. Its discretization form is as follows:

[0148]

[0149] in The integration step size is... , The first and The state vector of the step-by-step digital twin model, The function described by the system's differential equations. The state vector of the digital twin model represents the equivalent virtual topology of the entire power grid and its dynamic energy flow process, constructed by the above-mentioned fusion rules of node power balance and boundary behavior.

[0150] Furthermore, parallel computing technology is used to accelerate the simulation of large-scale power grids by decomposing tasks to multi-core processors or computing clusters, thereby achieving a linear increase in simulation speed.

[0151] In one implementation, the probabilistic security assessment algorithm includes:

[0152] By employing the Monte Carlo method combined with importance sampling techniques, a large number of random samples are taken from the uncertainties of renewable energy output and load to generate a massive power grid operating state boundary that includes uncertainties. Under a given operating state, historical statistical data is used to predict the probability of various preset system risk level assessment indicators occurring under that operating condition. ;

[0153] Define risk indicators ,in For the first The probability of occurrence of the risk level assessment indicators for this type of system. Cost of load shedding or equipment damage resulting from the failure;

[0154] The system automatically triggers warning levels based on risk indicator thresholds and highlights risk areas and key causative factors through a visual interface.

[0155] In one implementation, the verification feedback and closed-loop optimization include:

[0156] Multi-dimensional model validation index calculation: The output of the second system-level digital twin model is compared with the actual power grid operation data to obtain the deviation between the two. The deviation is obtained by calculating accuracy indicators including root mean square error and mean absolute percentage error, which are used to evaluate the performance of the second system-level digital twin model in terms of static parameter accuracy, dynamic parameter accuracy, and parameter correlation accuracy. At the same time, by comparing the response time and data synchronization error of the second system-level digital twin model and the actual power grid under specific operating conditions, the real-time performance and synchronization index of the model are verified.

[0157] Dynamic optimization based on verification results: If the deviation exceeds the preset over-limit threshold, the offline deep optimization process is initiated. This process first identifies the root causes of problems that cannot be eliminated by online scheduling updates through error source analysis. For parameter drift caused by equipment aging or environmental changes, Bayesian optimization algorithm is used to specifically correct key hyperparameters such as static impedance and control gain in the model. For structural deviations caused by unmodeled dynamics (such as complex internal nonlinear control of new energy sources) that are difficult to fit by online dual-path updates, a slow loop evolution mechanism is triggered. The physical-guided neural network uses the power grid differential algebra equation as a loss function constraint to perform a deep reconstruction and correction of the mapping structure inside the model.

[0158] Closed-loop feedback and knowledge accumulation: Each verification result, optimization operation, and its performance improvement effect are stored as experience samples in the model knowledge base, and a policy network is trained using reinforcement learning algorithm, enabling the system to automatically formulate trigger thresholds and weight strategies for subsequent online scheduling and dual-path differentiated updates based on historical optimization experience; thus forming a complete closed loop of "simulation deduction - multi-dimensional verification - dynamic optimization - experience feedback", realizing the autonomous and continuous evolution of the digital twin model throughout its entire life cycle.

[0159] Example 4

[0160] This invention provides a power grid digital twin modeling system based on dynamic assembly and updating, comprising a data sensing module, a model assembly module, an update scheduling module, and a verification feedback module that interact sequentially in a closed loop.

[0161] The data sensing module is used to collect operational data of the physical entities of the power grid, perform standardized processing, and then transmit the data to the model assembly module.

[0162] In one implementation, the data sensing module is configured to: collect multi-source data such as electrical quantities, non-electrical quantities, and environmental parameters in real time through sensors deployed at various nodes of the power grid, and perform data cleaning, time alignment, and spatial alignment processing at the edge side to form a unified standardized data stream, providing real-time and consistent input for subsequent model construction and updates.

[0163] The model assembly module is used to complete the equipment-level virtual-real mapping of power grid equipment based on the standardized operating data and combined with a preset standard data structure, assemble the equipment-level model into a cluster-level model and execute the boundary behavior fusion algorithm, construct electrical quantity coupling relationships through differential algebraic equations to achieve system integration, and output the first system-level digital twin model.

[0164] In one implementation, the model assembly module is configured to perform structured assembly of digital twin models from the device level to the system level. Based on standard data structure tables, this module establishes geometric models, attribute information tables, and behavioral logic files for power grid equipment such as circuit breakers and transformers, achieving device-level virtual-physical mapping through field binding. Using predefined topology-assembleable templates, it automatically combines device-level models into cluster models such as feeder-level and substation-level models according to power grid business topology relationships, and calls boundary behavior fusion algorithms to ensure consistency between the internal states and behavioral logic of the clusters. At the system level, it constructs electrical quantity coupling relationships based on differential-algebraic equations, realizing the continuous expression of energy flow and information flow in multiple sub-models, and completing the integration of system-level digital twin models.

[0165] The update scheduling module is used to quantify the current error uncertainty between the first system-level digital twin model and the power grid physical entity based on information entropy, and trigger a model update request based on the entropy threshold; obtain the equipment importance of each power grid device based on future operational risk assessment; update the first system-level digital twin model using a dual-path differentiated update method by combining the information entropy change magnitude and the equipment importance, and simultaneously perform hierarchical scheduling and dynamic resource allocation based on power grid topology partitioning, and output the second system-level digital twin model.

[0166] In one implementation, the update scheduling module includes an intelligent update decision and execution submodule and a simulation and decision support submodule.

[0167] The intelligent update decision and execution submodule is used to realize dual-path dynamic updates of the model. This module calculates the error information entropy between the model output and the physical entity state in real time, and compares it with a preset threshold to automatically trigger update requests; based on the magnitude of information entropy change and equipment importance, it dynamically selects and executes update strategies such as lightweight parameter correction or physical-guided neural network reconstruction; based on the power grid topology, it divides update regions and schedules computing resources for update tasks in different priority regions according to a dynamic resource allocation algorithm to achieve load balancing and real-time guarantee of the update process.

[0168] The simulation and decision support submodule is used to perform power grid state prediction and risk analysis using the first system-level digital twin model. This module injects typical or extreme operating conditions into the model, performs forward simulation, and predicts the dynamic trajectory of key state variables. Based on the simulation results, it uses a probabilistic safety assessment algorithm to quantify system risks, automatically identifies safety hazards, and generates and recommends pre-control strategies through a multi-objective optimization algorithm to support scheduling decisions.

[0169] The verification feedback module is used to verify the model accuracy of the second system-level digital twin model and feed the verification results back to the model assembly module and the update scheduling module to realize the self-evolution optimization of the system-level digital twin model.

[0170] In one implementation, the verification feedback module is configured to enable the self-evolution and continuous optimization of the model. This module compares the model output with actual running data, calculates multi-dimensional verification metrics to evaluate model accuracy and real-time performance; when deviations exceed limits, it initiates an offline deep optimization process, identifies the root cause of the problem through error source analysis, and performs targeted corrections using Bayesian optimization or deep reconstruction based on a physics-guided neural network; simultaneously, it stores the verification and optimization experiences from previous iterations in a knowledge base and uses reinforcement learning to train a strategy network to optimize the decision parameters of the dual-path differentiated update and hierarchical scheduling, forming a complete closed loop of "simulation-verification-optimization-feedback" to drive the model's autonomous and continuous evolution.

Claims

1. A method for digital twin modeling of power grids based on dynamic assembly and updating, characterized in that, The method includes: S1. Based on the preset standard data structure, complete the virtual-physical mapping of power grid equipment at the equipment level, assemble the equipment-level model into a cluster-level model and perform boundary behavior fusion, realize system integration by constructing electrical quantity coupling relationship, and obtain the first system-level digital twin model; S2. Based on information entropy, quantify the current error uncertainty between the first system-level digital twin model and the power grid physical entity, and trigger a model update request based on the entropy threshold; obtain the equipment importance of each power grid device based on future operation risk assessment; combine the information entropy change magnitude and the equipment importance, and update the first system-level digital twin model using a dual-path differentiated update method, and simultaneously perform hierarchical scheduling and dynamic resource allocation based on power grid topology partitioning to obtain the second system-level digital twin model.

2. The power grid digital twin modeling method based on dynamic assembly and updating according to claim 1, characterized in that, The preset standard data structure adopts a standard data structure table, which includes a geometric model index of power grid equipment, a static attribute dictionary, a set of behavioral rules, and a data flow interface.

3. The power grid digital twin modeling method based on dynamic assembly and updating according to claim 1, characterized in that, The boundary behavior fusion algorithm includes: based on the power grid business topology relationship, by comparing the predicted outputs of adjacent device-level models or arbitrary-level cluster-level models at the topology connection point, and using sliding mode control law or consensus algorithm to coordinate the model state.

4. The power grid digital twin modeling method based on dynamic assembly and updating according to claim 1, characterized in that, The electrical quantity coupling relationship is constructed using differential algebraic equations, as shown in the following formula: in, For time, Let the system state variable vector be... For the differential term of the state variable, The system input vector; The differential-algebraic equations contain algebraic constraints formed by the nodal power balance equations; the nodal power balance equations include: Active power balance equation: Reactive power balance equation: in, , The active and reactive power injected into the nodes, , The active and reactive power of the node load. , For nodes , voltage amplitude, For nodes , voltage phase angle difference, , Let be the real and imaginary parts of the network admittance matrix.

5. The power grid digital twin modeling method based on dynamic assembly and updating according to claim 1, characterized in that, The importance of the equipment is dynamically determined through forward simulation and decision analysis, including: injecting typical operating scenarios or extreme disturbance conditions into the first system-level digital twin model to perform forward simulation to predict the future state of the power grid; assessing the system risk level based on the simulation results, and assigning high equipment importance to power grid equipment at high risk levels to guide the dual-path differentiated update and generate pre-control strategies.

6. The power grid digital twin modeling method based on dynamic assembly and updating according to claim 5, characterized in that, The system risk level under various scenarios is quantified using a probabilistic security assessment algorithm, thereby automatically identifying potential weaknesses and security vulnerabilities; wherein, the probabilistic security assessment algorithm includes: By employing the Monte Carlo method combined with importance sampling techniques, a large number of random samples are taken from the uncertainties of renewable energy output and load to generate a massive dataset of power grid operating conditions containing uncertainties. Under certain operating conditions, historical statistical data is used to predict the probability of various preset system risks occurring under those conditions. ; Define risk indicators ,in For the first The probability of a pre-defined system risk occurring. Costs of load shedding or equipment damage resulting from this risk; Based on the aforementioned risk indicators The system automatically classifies risk levels based on thresholds and indicates risk areas and key causative factors.

7. The power grid digital twin modeling method based on dynamic assembly and updating according to claim 1, characterized in that, The method of updating the first system-level digital twin model using a dual-path differentiated update approach includes: When the information entropy change is lower than the severity threshold and the corresponding device importance is identified as a non-core node, a lightweight parameter correction is adopted, and the parameters of the first system-level digital twin model are quickly adjusted through an online learning algorithm. When the information entropy change exceeds the severity threshold or the corresponding equipment is identified as a core critical equipment, the first system-level digital twin model is reconstructed using a physical-guided neural network, and the physical laws of the power grid are embedded as constraints into the neural network loss function for training.

8. The power grid digital twin modeling method based on dynamic assembly and updating according to claim 1, characterized in that, The hierarchical scheduling includes: The update regions are divided based on the power grid topology, and priority is assigned to the model update tasks in different regions. Among them, core equipment with high equipment importance is updated synchronously at high frequency, while edge equipment with low equipment importance is updated asynchronously at low priority.

9. The power grid digital twin modeling method based on dynamic assembly and updating according to claim 1, characterized in that, The method also includes: The deviations between the state prediction data output by the second system-level digital twin model and the real-time collected multi-source operation data of power grid equipment and the response time and data under specific operating conditions are identified. The accuracy and real-time performance of the second system-level digital twin model are evaluated based on the deviations. When the deviations exceed the limits, the offline deep optimization process is initiated. The offline deep optimization includes: for parameter drift caused by equipment aging or environmental changes, a Bayesian optimization algorithm is used to specifically correct the key hyperparameters in the second system-level digital twin model; for structural deviations caused by unmodeled dynamics, a physical-guided neural network is used to use the power grid differential algebraic equation as a loss function constraint to perform a low-level deep reconstruction and correction of the virtual-real mapping structure inside the model; at the same time, the verification and optimization experience of each time is stored in the knowledge base, and a reinforcement learning training strategy network is used to optimize the decision parameters of the dual-path differentiated update and hierarchical scheduling.

10. A power grid digital twin modeling system based on dynamic assembly and updating, characterized in that, The system includes: a data sensing module, a model assembly module, an update scheduling module, and a verification feedback module that interact sequentially in a closed loop; The data sensing module is used to collect the operating data of the physical entities of the power grid, perform standardized processing, and then transmit the data to the model assembly module. The model assembly module is used to complete the equipment-level virtual-real mapping of power grid equipment based on the standardized operating data and combined with the preset standard data structure, assemble the equipment-level model into a cluster-level model and execute the boundary behavior fusion algorithm, construct the electrical quantity coupling relationship through differential algebraic equations to realize system integration, and output the first system-level digital twin model. The update scheduling module is used to quantify the current error uncertainty between the first system-level digital twin model and the power grid physical entity based on information entropy, and trigger a model update request based on the entropy threshold; obtain the equipment importance of each power grid device based on future operation risk assessment; update the first system-level digital twin model using a dual-path differentiated update method in combination with the information entropy change magnitude and the equipment importance, and simultaneously perform hierarchical scheduling and dynamic resource allocation based on power grid topology partitioning, and output the second system-level digital twin model; The verification feedback module is used to verify the model accuracy of the second system-level digital twin model and feed the verification results back to the model assembly module and the update scheduling module to realize the self-evolution optimization of the system-level digital twin model.

Images