An intelligent power resource scheduling method, device and equipment

Abstract

The application relates to the technical field of smart grid dispatching, and provides a smart power resource dispatching method, device and equipment, which comprises the following steps: acquiring real-time collection data of a physical power grid, and performing labelization processing on the real-time collection data; constructing a digital twin based on the labelization-processed real-time collection data; synchronously correcting the digital twin, so that the error between the state of the digital twin and the real-time state of the physical power grid is kept within a preset threshold; based on the synchronously corrected digital twin, allocating a computing resource to perform analysis processing, generating and executing a dispatching scheme; the analysis processing comprises power grid state analysis, fault diagnosis, risk early warning or scene simulation. The application solves the problems of single dimension of a digital twin model and insufficient synchronous correction accuracy in the prior art, and realizes high-precision synchronization of the physical power grid and the digital twin and integration of multi-dimensional model capability.

Description

Technical Field

[0001] This invention relates to the field of smart grid dispatching technology, and in particular to a smart power resource dispatching method, apparatus and equipment. Background Technology

[0002] With the continuous expansion of smart grid construction, the power system is exhibiting significant characteristics such as increased structural complexity, explosive growth in data volume, and dynamic operational status. On the one hand, the grid coverage is constantly expanding, with a large number of distributed energy sources and energy storage devices being connected to the grid, making the need for cross-regional and cross-energy type collaborative management increasingly urgent. On the other hand, the amount of real-time data generated by grid monitoring equipment is growing exponentially, coupled with multi-source heterogeneous data such as historical grid load, meteorological changes, and equipment operating status, posing a severe challenge to the data processing and integration capabilities of the dispatch center.

[0003] Existing digital twin models are mostly single-dimensional geometric or physical models, lacking the ability to predict power grid operation behavior and perform multi-objective optimization, resulting in the model failing to fully reflect the power grid's operating status. At the same time, the existing solutions have relatively simple synchronization correction mechanisms between the digital twin model and the physical power grid, making it difficult to achieve high-precision synchronization and affecting the accuracy of dispatching commands.

[0004] Therefore, how to construct an intelligent power resource scheduling method that can achieve high-precision synchronization between the physical power grid and the digital twin, and integrate multi-dimensional model capabilities to ensure scheduling timeliness is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] This invention provides an intelligent power resource scheduling method, device, and equipment, which solves the problems of single-dimensional digital twin models and insufficient synchronization correction accuracy in the prior art, and realizes the integration of high-precision synchronization and multi-dimensional model capabilities between the physical power grid and the digital twin.

[0006] This invention provides an intelligent power resource scheduling method, comprising the following steps: Acquire real-time data from the physical power grid and label the real-time data. The labels include at least node identifier, acquisition time, and topological location. A digital twin is constructed based on real-time collected data after labeling. The digital twin includes a physical model, a power grid behavior prediction model, and a multi-objective dynamic optimization model. The digital twin is synchronously corrected so that the error between the state of the digital twin and the real-time state of the physical power grid is kept within a preset threshold. Based on the synchronized and corrected digital twin, computing resources are allocated for analysis and processing to generate and execute scheduling schemes. The analysis and processing includes power grid status analysis, fault diagnosis, risk warning, or scenario simulation.

[0007] According to a smart power resource scheduling method provided by the present invention, the synchronous correction of the digital twin includes: converting the real-time acquired data and the output data of the digital twin into multi-dimensional electrical feature vectors respectively; calculating a deviation matrix between the electrical feature vector corresponding to the real-time acquired data and the electrical feature vector corresponding to the output data based on the weighted distance of the electrical feature space, wherein the weighted distance of the electrical feature space is differentially weighted using power grid topology weight and operating state weight; calculating a transformation matrix according to the deviation matrix, and correcting the state of the digital twin through the transformation matrix.

[0008] According to the intelligent power resource scheduling method provided by the present invention, the fault diagnosis steps include: acquiring electrical quantity measurement data of each node of the power grid and the topological connection relationship of the power grid; dividing the power grid into electrical islands based on the topological connection relationship of the power grid and calculating the electrical impedance distance between nodes in the same electrical island; performing cluster analysis on the electrical impedance distance to obtain the clustering result; verifying the clustering result and determining the fault location based on the verified clustering result.

[0009] According to the present invention, an intelligent power resource scheduling method is provided, wherein the power grid behavior prediction model is implemented based on a time-series neural network, the input layer of the time-series neural network is embedded with a power grid topology adjacency matrix, the hidden layer of the time-series neural network is provided with a constraint gating unit, which is used to suppress abnormal output when the voltage or frequency exceeds a safety threshold, and the loss function of the time-series neural network includes a data fitting loss term and a power grid power balance penalty term.

[0010] According to the intelligent power resource scheduling method provided by the present invention, the analysis and processing of allocating computing resources includes: dividing the scheduling tasks into different levels according to the complexity and / or priority of the scheduling tasks; allocating different scales of computing resources to the scheduling tasks of different levels, wherein computing resources are allocated preferentially to scheduling tasks with higher priority.

[0011] According to the intelligent power resource scheduling method provided by the present invention, when executing the scheduling scheme, the method further includes: collecting execution status data and feeding back the execution status data to update the digital twin.

[0012] The present invention also provides an intelligent power resource scheduling device, comprising the following modules: The data acquisition module is used to acquire real-time data of the physical power grid and to tag the real-time data. The tags include at least node identifier, acquisition time and topological location. A digital twin construction module is used to construct a digital twin based on real-time collected data after labeling. The digital twin includes a physical model, a power grid behavior prediction model, and a multi-objective dynamic optimization model. The synchronization correction module is used to synchronize and correct the digital twin so that the error between the state of the digital twin and the real-time state of the physical power grid is kept within a preset threshold. The scheduling processing module is used to allocate computing resources for analysis and processing based on the synchronized and corrected digital twin, generate and execute scheduling schemes, and the analysis and processing includes power grid status analysis, fault diagnosis, risk warning or scenario simulation.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the intelligent power resource scheduling method as described above.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the intelligent power resource scheduling method as described above.

[0015] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the intelligent power resource scheduling method as described above.

[0016] This invention provides an intelligent power resource scheduling method, device, and equipment, which has the following beneficial effects: By acquiring real-time data from the physical power grid and performing tagging processing, a structured foundation is laid for subsequent data fusion and model construction, solving the problem of inconsistent formats of multi-source data; on this basis, a digital twin is constructed, including a physical model, a power grid behavior prediction model, and a multi-objective dynamic optimization model. Compared with the single-dimensional geometric or physical models in the prior art, this achieves a comprehensive integration of the power grid's physical characteristics, operational behavior prediction, and multi-objective optimization capabilities, enabling the digital twin to more completely reflect the power grid's operating status; by synchronously correcting the digital twin and... The error between the real-time state of the physical power grid and the actual state of the power grid is controlled within a preset threshold, overcoming the shortcomings of existing synchronization correction mechanisms, such as simplicity and insufficient synchronization accuracy. This ensures that the model state on which the dispatch instructions are based remains highly consistent with the actual operating state of the power grid. Furthermore, based on the synchronized and corrected digital twin, computing resources are allocated for analysis and processing. The computing power can be flexibly allocated according to the needs of different analysis tasks such as power grid state analysis, fault diagnosis, risk warning, or scenario simulation, ensuring the timeliness of dispatch scheme generation. Finally, the dispatch scheme is generated and executed, forming an integrated optimization of the entire process from data acquisition, twin construction, synchronization correction to dispatch execution, effectively improving the accuracy and response efficiency of power dispatch. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this 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 some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the intelligent power resource scheduling method provided by the present invention.

[0019] Figure 2 This is a schematic diagram of the intelligent power dispatching architecture based on digital twins provided by the present invention.

[0020] Figure 3 This is a schematic diagram of the synchronization correction process between the digital twin and the physical power grid provided by the present invention.

[0021] Figure 4 This is a detailed illustration of the specific implementation of "iterative nearest point registration" and "deviation compensation" in the synchronous correction process provided by the present invention.

[0022] Figure 5 This is a flowchart illustrating the principle of the present invention.

[0023] Figure 6 This is a schematic diagram of the intelligent power resource scheduling device provided by the present invention.

[0024] Figure 7 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0026] This invention belongs to the field of smart grid dispatching technology. It specifically designs a smart power dispatching architecture that integrates digital twins, big data, and high-performance computing technologies, and also provides the corresponding implementation method. This architecture and method are applicable to the collaborative dispatching of power systems across regions and different energy types, and are particularly suitable for complex power grid operation scenarios with a high proportion of renewable energy and large fluctuations in electricity load.

[0027] This invention breaks through the technical limitations of traditional power dispatching architecture, builds a dynamic self-optimizing dispatching closed-loop system for the physical power grid based on a digital twin architecture, integrates three core innovative technologies: customized big data fusion, distributed elastic computing power dispatching, and cross-system heterogeneous integration, and designs a five-layer differentiated closed-loop architecture consisting of a physical power grid perception layer, a digital twin modeling layer, a high-performance computing and analysis layer, a collaborative dispatching decision layer, and a physical power grid execution layer.

[0028] Compared with existing technologies, this invention realizes integrated optimization of the entire process of power dispatching, from real-time perception, accurate prediction, intelligent decision-making, efficient execution to state feedback. It can provide core technical support for the intelligent, refined and collaborative dispatching of smart grids, and effectively improve the dispatching response speed and power resource allocation efficiency in complex power grid scenarios.

[0029] The following is combined with Figures 1-7 The embodiments of the present invention are described in detail.

[0030] The intelligent power resource scheduling method provided in this embodiment of the invention is executed by an intelligent power resource scheduling device, which can be configured in a computer. The computer can be a local computer or a cloud computer. The local computer can be a computer, tablet, etc., and no specific limitation is made here.

[0031] Figure 1 This is a flowchart illustrating the intelligent power resource scheduling method provided by the present invention, as shown below. Figure 1 As shown, the method includes the following steps: S110. Acquire real-time data from the physical power grid and label the data. The labels must include at least the node identifier, acquisition time, and topological location.

[0032] S120. Construct a digital twin based on real-time acquired data after tagging. The digital twin includes a physical model, a power grid behavior prediction model, and a multi-objective dynamic optimization model.

[0033] S130. Synchronize and correct the digital twin to keep the error between the state of the digital twin and the real-time state of the physical power grid within a preset threshold.

[0034] S140. Based on the synchronized and corrected digital twin, allocate computing resources for analysis and processing, generate and execute scheduling schemes. Analysis and processing include power grid status analysis, fault diagnosis, risk warning, or scenario simulation.

[0035] Specifically, such as Figure 2The diagram illustrates a smart power dispatch architecture based on digital twins. The physical power grid sensing layer collects core power grid data and externally related data hierarchically through core data acquisition equipment and auxiliary data acquisition equipment. The core data acquisition equipment includes phasor measurement units and high-precision status sensors, acquiring high-precision electrical quantities at a frequency of 10 milliseconds. The auxiliary data acquisition equipment includes data acquisition and monitoring control system equipment and conventional distribution terminals, acquiring conventional monitoring quantities at a frequency of 1 second. All collected data undergoes power grid spatiotemporal tagging, adding three types of tags: node identifier, acquisition timestamp, and topology location code. The node identifier is a unique code across the entire network, formatted as region code-voltage level-equipment number. The acquisition timestamp uses Coordinated Universal Time (UTC) with millisecond-level accuracy. The topology location code is based on node coordinate codes from the power grid map database, including information on the substation, bus, and feeder levels.

[0036] The customized unified big data lake for the digital twin modeling layer performs customized cleaning of tagged data, including deduplication based on the joint primary key of node identifier and collection timestamp, automatic switching of the completion algorithm based on the duration of missing data, and dimension normalization based on the rated operating parameters of the power grid. The processed data is then stored in a hot / cold data tier: real-time collected data is stored in the hot data area using memory and solid-state drives, 72 hours of collected data are retained, and historical data is stored in the cold data area using distributed hard drives. Based on the processed data, a multi-dimensional twin model specific to the power grid is constructed. The physical model replicates the physical and electrical characteristics of power grid lines and equipment using finite element analysis based on the actual power grid topology. The power grid behavior prediction model uses a power grid time-series feature-enhanced neural network, with its input layer embedding the power grid topology adjacency matrix, the hidden layer setting power grid constraint gating units, and the loss function including a power grid power balance penalty term. The multi-objective dynamic optimization model updates constraint parameters in real time when equipment operating conditions are abnormal or system conditions change, automatically adapting to changes in power grid operating conditions.

[0037] The second-level real-time synchronous self-correction engine adopts a combined correction mechanism of iterative nearest-point registration and power grid deviation compensation model. It converts the physical power grid collected data and the digital twin output data into multi-dimensional electrical feature vectors. The multi-dimensional electrical feature vectors include at least voltage amplitude, voltage phase angle, active power, reactive power and equipment status index. The deviation matrix is ​​calculated based on the weighted distance of the electrical feature space. The weighted distance of the electrical feature space is differentiated by the power grid topology weight and the operating status weight. According to the deviation matrix, a power grid-specific transformation matrix containing the offset vector, topology weight scaling factor and operating status weight scaling factor is calculated. The state of the digital twin model is corrected by the transformation matrix, so that the error between the corrected state of the digital twin model and the real-time state of the physical power grid is controlled within a preset threshold range. The synchronous correction cycle is second-level.

[0038] The distributed elastic computing resource pool of the high-performance computing and analysis layer divides scheduling tasks into micro-tasks, medium-tasks, and macro-tasks based on their complexity and priority. Micro-tasks are allocated single-node computing power, medium-tasks are allocated multi-node parallel computing power, and macro-tasks utilize heterogeneous computing power across the entire platform. A preemptive scheduling strategy prioritizes allocating computing resources for urgent scheduling tasks such as fault response. Based on the allocated computing power, the power grid-specific intelligent analysis engine performs analysis and processing: it uses a time-series sliding window algorithm for power grid state analysis with an analysis latency of no more than 2 seconds; it employs a topology-constrained density clustering algorithm for fault diagnosis, using electrical impedance distance as the distance metric, with the neighborhood search range limited to the same electrical island by power grid topology constraints, resulting in a diagnosis latency of no more than 3 seconds; it uses a time-series neural network for prediction and early warning, enabling ultra-short-term and short-term load forecasting; and it performs scenario simulation pre-runs using typical scenario templates with custom parameters, with a simulation response time of no more than 10 seconds. The standardized and unified system integration platform of the collaborative scheduling decision layer integrates multiple power business management systems through a publish-subscribe message model using a heterogeneous system integration protocol, realizes data interoperability and command transmission among multiple systems, generates scheduling schemes based on analysis and processing results, and executes scheduling commands in a hierarchical manner through intelligent execution devices of the physical power grid execution layer.

[0039] This embodiment achieves structured unification of multi-source heterogeneous data by acquiring and tagging real-time data from the physical power grid, laying a data foundation for subsequent model construction. By constructing a digital twin comprising a physical model, a power grid behavior prediction model, and a multi-objective dynamic optimization model, it integrates the capabilities of replicating power grid physical characteristics, predicting operational behavior, and optimizing multiple objectives, enabling the digital twin to comprehensively reflect the power grid's operational status. Through synchronous correction of the digital twin and controlling errors within preset thresholds, high-precision synchronization between the digital twin model and the physical power grid is achieved, ensuring the accuracy of the model state upon which dispatch instructions are based. By allocating computing resources for analysis and processing based on the synchronously corrected digital twin, and flexibly adjusting computing resources according to the analysis task type while prioritizing urgent tasks, the timeliness of dispatch scheme generation and the efficiency of computing resource utilization are improved. Ultimately, this forms an integrated optimization process from data acquisition, twin construction, synchronous correction to dispatch execution, effectively improving the accuracy and response efficiency of power dispatch.

[0040] According to the present invention, an intelligent power resource scheduling method is provided, in which the power grid behavior prediction model is implemented based on a temporal neural network. The input layer of the temporal neural network is embedded with the power grid topology adjacency matrix, and the hidden layer of the temporal neural network is provided with a constraint gating unit to suppress abnormal output when the voltage or frequency exceeds the safety threshold. The loss function of the temporal neural network includes a data fitting loss term and a power grid power balance penalty term.

[0041] Specifically, the power grid behavior prediction model is implemented using a power grid time-series feature-enhanced neural network. The input layer of this time-series neural network embeds a power grid topology adjacency matrix, whose elements represent the electrical connections between power grid nodes, binding the neuron connection weights to the physical connections of the power grid. The hidden layer of this time-series neural network incorporates a power grid constraint gating unit, which performs gating calculations based on whether the voltage or frequency exceeds a safety threshold. When the voltage or frequency exceeds the safety threshold, abnormal outputs are automatically suppressed. The loss function of this time-series neural network includes a data fitting loss term and a power grid power balance penalty term. The power grid power balance penalty term is used to constrain the prediction results to satisfy Kirchhoff's laws, and its calculation formula is based on the sum of the squares of the differences between the node input power and the node output power. ,in, Input power to the node. Output power to the node.

[0042] This embodiment enhances the model's ability to perceive the power grid topology by embedding the grid topology adjacency matrix into the input layer of a temporal neural network, thus binding the neuron connection weights to the physical connections of the power grid. By setting a grid constraint gating unit in the hidden layer, abnormal outputs are automatically suppressed when voltage or frequency exceeds a safety threshold, enhancing the safety of the model's output and preventing the generation of predictions that violate grid safety constraints. Furthermore, by introducing a grid power balance penalty term into the loss function, the prediction results are constrained to satisfy Kirchhoff's laws, solving the problem that general neural networks generate physically infeasible predictions by only pursuing data fit. This allows the power grid behavior prediction model to satisfy the physical constraints of the power grid while ensuring prediction accuracy.

[0043] According to the present invention, an intelligent power resource scheduling method is used to synchronously correct a digital twin, including: converting real-time acquired data and output data of the digital twin into multi-dimensional electrical feature vectors respectively; calculating a deviation matrix between the electrical feature vectors corresponding to the real-time acquired data and the electrical feature vectors corresponding to the output data based on the weighted distance of the electrical feature space, wherein the weighted distance of the electrical feature space is differentially weighted using power grid topology weights and operating state weights; calculating a transformation matrix based on the deviation matrix, and correcting the state of the digital twin using the transformation matrix.

[0044] Specifically, a second-level real-time synchronous self-correcting engine acquires real-time data collected by the physical power grid sensing layer and output data from the digital twin modeling layer. The real-time data and the digital twin output data are converted into multi-dimensional electrical feature vectors, including voltage amplitude, voltage phase angle, active power, reactive power, and equipment status index. A deviation matrix is ​​calculated between the electrical feature vectors corresponding to the real-time data and those corresponding to the output data, based on a weighted distance in the electrical feature space. This weighted distance uses differentiated weighting based on power grid topology weights and operational status weights. The power grid topology weights are determined according to the electrical connection hierarchy between nodes, with higher weights for nodes within the same substation than for nodes across substations. The operational status weights are dynamically adjusted based on equipment health index and load factor. A power grid-specific transformation matrix is ​​calculated based on the deviation matrix. This transformation matrix includes an offset vector, a topology weight scaling factor, and an operational status weight scaling factor. The state of the digital twin is corrected using this transformation matrix, ensuring that the error between the corrected digital twin state and the real-time state of the physical power grid is controlled within a preset threshold range.

[0045] This embodiment converts real-time acquired data and digital twin output data into multi-dimensional electrical feature vectors, using electrical quantities instead of spatial geometric coordinates as the registration benchmark. This solves the problem that the general iterative nearest point algorithm, which uses three-dimensional spatial coordinates, cannot reflect the real electrical coupling relationship of the power grid. By using power grid topology weights and operating state weights to calculate the deviation matrix of the electrical feature spatial weighted distance, the registration process can be adaptively adjusted according to the power grid topology and equipment operating state, avoiding the registration distortion caused by uniform correction in general algorithms. By calculating the transformation matrix based on the deviation matrix and correcting the state of the digital twin, high-precision synchronization between the digital twin model and the physical power grid is achieved, ensuring that dispatch instructions are generated based on the real operating state of the power grid.

[0046] According to the present invention, an intelligent power resource scheduling method allocates computing resources for analysis and processing, including: classifying scheduling tasks into different levels based on their complexity and / or priority; allocating different scales of computing resources to scheduling tasks of different levels, wherein computing resources are allocated preferentially to scheduling tasks with higher priority.

[0047] Specifically, the distributed elastic computing resource pool of the high-performance computing and analysis layer integrates multi-core computing units and adopts a hybrid heterogeneous computing architecture. Scheduling tasks are divided into different levels based on their complexity, including: routine status monitoring as micro-tasks, load forecasting as medium-level tasks, and fault response and cross-domain collaboration as macro-tasks. Simultaneously, task scheduling is based on priority, with emergency scheduling tasks such as fault response having higher priority than regular tasks. Different scales of computing resources are allocated to different levels of scheduling tasks: single-node computing power is allocated to micro-tasks, multi-node parallel computing power is allocated to medium-level tasks, and heterogeneous computing power across the entire platform is utilized for macro-tasks. A preemptive scheduling strategy is employed, prioritizing the allocation of computing resources to higher-priority scheduling tasks when emergency tasks arrive, ensuring computing power supply in emergency scenarios.

[0048] This embodiment divides scheduling tasks into different levels based on their complexity and priority, and allocates different amounts of computing resources to tasks of different levels, thereby achieving fine-grained allocation of computing resources and avoiding idle or insufficient computing resources. By prioritizing the allocation of computing resources to high-priority scheduling tasks and adopting a preemptive scheduling strategy, it can ensure priority supply of computing power in emergency scenarios such as fault response, improve the response timeliness of emergency scheduling tasks, and solve the problems of fixed computing power allocation and low utilization efficiency in the prior art.

[0049] According to the intelligent power resource scheduling method provided by the present invention, the fault diagnosis steps include: acquiring electrical quantity measurement data of each node of the power grid and the topological connection relationship of the power grid; dividing the power grid into electrical islands based on the topological connection relationship of the power grid and calculating the electrical impedance distance between nodes in the same electrical island; performing cluster analysis on the electrical impedance distance to obtain the clustering results; verifying the clustering results and determining the fault location based on the verified clustering results.

[0050] Specifically, the fault diagnosis module acquires electrical quantity measurement data and grid topology connections for each node in the power grid. The electrical quantity measurement data includes at least voltage, current, and power data. Based on the grid topology connections, electrical islands are defined, grouping nodes with electrical connections into the same electrical island. The electrical impedance distance between any two nodes within the same electrical island is calculated. This distance is determined based on the equivalent impedance and current change between nodes, and the formula is: electrical impedance distance equals the modulus of the equivalent impedance multiplied by the current change. A grid topology-constrained density clustering algorithm is used to cluster the electrical impedance distance. This algorithm's neighborhood search range is constrained by the grid topology, searching only within the same electrical island. The clustering results are verified through grid protection action logic, excluding clusters that do not conform to protection coordination relationships. Based on the verified clustering results, the fault location is determined, and fault diagnosis information is output.

[0051] This embodiment divides electrical islands based on the power grid topology and limits the neighborhood search range of cluster analysis to the same electrical island, avoiding misclassification of electrically isolated fault points into the same cluster and improving the accuracy of fault location. By using electrical impedance distance instead of Euclidean distance as the distance metric for the clustering algorithm, the distance calculation can reflect the electrical coupling relationship between nodes, avoiding misjudgments caused by physical proximity but weak electrical coupling. By performing power grid protection action logic verification on the clustering results, false alarms that do not conform to the protection coordination relationship are eliminated, further improving the reliability of fault diagnosis.

[0052] According to the intelligent power resource scheduling method provided by the present invention, when executing a scheduling scheme, execution status data is collected and the execution status data is fed back to update the digital twin.

[0053] Specifically, the physical power grid execution layer is equipped with an encrypted real-time status feedback unit. After the intelligent execution device completes the scheduling command, the encrypted real-time status feedback unit collects the device's execution status data in real time. The execution status data includes the completion status of the action, the device's operating parameters after execution, and the device's health status. Execution result tags are attached to the collected execution status data, including three categories: success, failure, and partial execution. The execution status data with attached execution result tags is fed back to the digital twin modeling layer via wireless and fiber optic links through the encrypted feedback unit protocol. The feedback data is encrypted using a symmetric encryption algorithm and includes a message authentication code. Based on the received execution result tags and feedback parameters, the digital twin modeling layer triggers updates and deviation corrections to the corresponding node's digital twin model, completing the closed loop of the scheduling process.

[0054] This embodiment achieves the quantification of the status of scheduling command execution results by collecting execution status data and attaching execution result labels to the data, providing accurate feedback basis for updating the digital twin model. By feeding back the execution status data to the digital twin modeling layer and triggering the model update and deviation correction of the corresponding nodes, a complete dynamic self-optimizing closed loop is formed from physical power grid perception to scheduling execution and then to twin update. This enables the digital twin model to reflect the real-time changes of the physical power grid in a timely manner and improves the accuracy of subsequent scheduling decisions.

[0055] The present invention will be further illustrated below through complete embodiments.

[0056] The five-layer architecture designed in this invention is interconnected and forms a dynamic closed loop. The hardware composition, software algorithm, implementation mechanism, and interaction protocol of each layer have been comprehensively improved and clearly defined, and all adopt implementation methods that are significantly different from existing technologies, as detailed below: 1. Physical power grid sensing layer (customized data acquisition and tagged transmission) This layer forms the foundation for customized data acquisition within the architecture. It overcomes the limitations of existing simple device integration models by employing multi-source device hierarchical access, heterogeneous data tagging, and redundant wireless-fiber dual-link transmission. Specifically, it comprises three parts: hardware integration, data access, and transmission mechanisms. Hardware Integration: Based on the requirements for data acquisition accuracy and real-time performance, SCADA equipment, PMU phasor measurement units, power distribution terminals, and equipment status sensors are divided into core data acquisition devices and auxiliary data acquisition devices. Among them, the PMU phasor measurement units and high-precision status sensors are the core data acquisition devices, with an acquisition frequency of 10 milliseconds; SCADA equipment and conventional power distribution terminals are the auxiliary data acquisition devices, with an acquisition frequency of 1 second, realizing fine-grained configuration of acquisition resources.

[0057] Data Access: This involves accessing core power grid data and externally correlated data. Core data includes voltage, current, power, and equipment operating status. Externally correlated data includes real-time / forecast weather data, regional load data, electricity market trading signals, and power generation data from renewable energy plants. All collected data undergoes power grid spatiotemporal tagging, adding tags for the data acquisition node, acquisition time, and power grid topology location. This resolves the issue of inconsistent formats among multi-source data and lays a structured foundation for subsequent data fusion in the digital twin modeling layer. The tag structure is as follows: Tag structure = { Node Identifier: A unique code across the entire network (format: region code-voltage level-device number). Data collection timestamp: UTC standard time (accurate to milliseconds), Topology location coding: Node coordinate coding based on the power grid map database (including information on the substation, busbar, and feeder levels). } Transmission Mechanism: Core data is preferentially transmitted via a dedicated fiber optic link, with a transmission latency of no more than 5 milliseconds. The wireless link serves as a hot-standby fault-tolerant supplement, automatically switching to the wireless link within milliseconds in the event of a fiber optic link failure. Auxiliary data is transmitted via a wireless network, with a transmission latency of no more than 1 second. The transmission layer uniformly adopts a lightweight encryption and compression protocol. The encryption module uses the SM4 national standard symmetric encryption algorithm (128-bit key length) to encrypt the transmitted data. In cross-border collaborative scheduling scenarios, it can be replaced with the AES-256 algorithm. The compression module uses the LZ4 lossless compression algorithm to compress the encrypted data stream in real time, reducing transmission bandwidth usage by more than 40% while ensuring data integrity.

[0058] Bandwidth compression ratio calculation formula: In the formula, Bandwidth compression rate (%) The bandwidth of the data stream before compression (Mbps). For the compressed data stream bandwidth (Mbps), η ≥ 40% is required.

[0059] (II) Digital Twin Modeling Layer Based on the data collected by the physical sensing layer, this layer constructs a virtual twin that is 1:1 highly accurate and supports second-level deviation self-correction. Its core comprises three innovative modules: a customized unified big data lake, a power grid-specific multi-dimensional twin model, and a second-level real-time synchronization self-correction engine. The specific implementation method is as follows: Customized Unified Big Data Lake: Customized cleaning and standardization algorithms are used for tagged power grid data to achieve rapid deduplication, completion, and normalization of time-series and heterogeneous power grid data. A hierarchical storage mechanism for hot and cold data is designed to address the low efficiency of existing big data lakes in processing power grid data. Specific steps are as follows: Deduplication: Based on spatiotemporal standards, deduplication uses the node identifier + collection timestamp as a combined primary key to detect and remove duplicate reports caused by network jitter. The false positive rate for deduplication does not exceed 1%. False positive rate calculation formula: In the formula, The deduplication false positive rate (%) To avoid misjudging the number of data entries, This represents the total number of duplicate data entries to be deduplicated.

[0060] Missing value completion: Two adaptive completion algorithms are designed for time-series power grid data. Under normal operating conditions and when the data missing duration does not exceed 500 milliseconds, a linear interpolation completion algorithm based on data from previous and subsequent time points is used. Under conditions of prolonged data missing, such as equipment outages, a statistical mean completion algorithm based on historical data of the same type is used. The two algorithms automatically switch according to the duration of data missing. Linear interpolation completion formula: In the formula, The result is the completion of missing values ​​at time t. They are respectively The collected values ​​at adjacent valid time points before and after the given time point. This is the timestamp for the corresponding moment. The formula for completing the statistical mean is: In the formula, for Time-lapse missing value completion result, This is the first time under similar working conditions in the same period in history. The collected values ​​of each sample, To determine the effective sample size, the following is required: .

[0061] Normalization Processing: For electrical quantity data with different dimensions such as voltage, current, and power, a component-dimensional normalization algorithm based on the rated operating parameters of the power grid is adopted to normalize various electrical quantities to the [0,1] interval, replacing the general maximum-minimum normalization method and avoiding excessive compression of normal operation data due to extreme values ​​such as transient overvoltages. Component-dimensional normalization formula: In the formula, The values ​​are normalized. These are the actual collected values ​​of electrical quantities. The rated operating parameters of the power grid corresponding to this electrical quantity Hot and cold tiered storage: Real-time collected tagged data is stored in the hot data area (using memory + SSD storage, retaining 72 hours of collected data), while historical data is stored in the cold data area (using distributed hard disk storage) to ensure data retrieval efficiency.

[0062] A multi-dimensional twin model specifically designed for power grid operation: A three-layer model is custom-built to address the characteristics of power grid operation, achieving high-precision replication of the physical power grid, accurate load forecasting, and multi-objective dynamic optimization. Specifically: High-precision physical model: Based on the actual topology of the power grid, the physical and electrical characteristics of the power grid lines and equipment are replicated using the finite element analysis algorithm to achieve a high-precision mapping with the physical power grid.

[0063] Power Grid Behavior Prediction Model: The behavior prediction model employs a power grid time-series feature-enhanced neural network, with improvements including: The input layer embeds a power grid topology adjacency matrix A, binding the neuron connection weights to the physical connections within the power grid. The adjacency matrix elements... This indicates that node i and node j are electrically connected. Indicates no connection; The hidden layer is equipped with a power grid constraint gating unit G, which automatically suppresses abnormal output when the voltage / frequency exceeds a safety threshold. Where σ is the sigmoid activation function, This is the gated weight matrix; The loss function adds a power balance penalty term to the grid. To ensure that the prediction results satisfy the constraints of Kirchhoff's laws, , .

[0064] Multi-objective dynamic optimization model: Combining multiple objectives such as electricity market, grid security, and resource utilization efficiency, an optimization model with nonlinear constraints and adaptive adjustment is designed to adapt to different grid dispatching scenarios. When equipment operating conditions are abnormal (such as transformer temperature rise exceeding limits) or system conditions change (such as after an N-1 fault), the model updates the corresponding constraint parameters in real time without manual intervention to reset them, achieving automatic adaptation to changes in grid operating conditions. In the formula, To comprehensively optimize the target value, For electricity purchase, sale and operating costs, For power grid losses, To improve the efficiency of new energy consumption; Let be the weighting coefficient, satisfying It can be dynamically adjusted according to the power grid operation scenario.

[0065] Dispatchers can modify dispatch instructions within preset constraints (such as power grid safety limits and power balance requirements). After modification, the system performs real-time constraint verification through a multi-objective optimization model. If the verification fails, it will automatically prompt the violation and refuse to issue instructions.

[0066] A second-level real-time synchronization self-correction engine: Employing a combined correction mechanism of ICP registration and a power grid deviation compensation model, it achieves second-level synchronization between the digital twin model and the physical power grid. Addressing the coupling characteristics of the power grid's multi-dimensional electrical features, it replaces the general point cloud Euclidean distance with a weighted distance in the electrical feature space. Feature vectors are constructed using core power grid features such as voltage amplitude, phase angle, active / reactive power, and equipment status as dimensions. The power grid-specific transformation matrix is ​​calculated by minimizing the weighted distance in the electrical feature space. (Including offset vector) Topology weight scaling factor Running status weight scaling factor This achieves accurate registration between data collected from the physical power grid and output data from the digital twin model. Iterative objective function. , The optimal transformation matrix (customized for power grid scenarios). (R is the eigenvector rotation matrix, adapted to the angular characteristics of the power grid topology). The output of the digital twin model One electrical feature vector (dimension: , For voltage amplitude, For voltage phase angle, For active power, For reactive power, (Equipment status index). Data collected for physical power grids and The first match One electrical feature vector; iterative convergence condition: when the weighted distance error changes between two adjacent iterations... Stop iterating when the time comes.

[0067] Figure 3This diagram illustrates the synchronization correction process between the digital twin and the physical power grid. Multi-dimensional electrical feature vectors are extracted from real-time data collected from the physical power grid, and corresponding feature vectors are output from the digital twin, providing a unified data foundation for registration calculations. An iterative nearest-point registration algorithm is used to calculate the deviation between the physical power grid feature vectors and the twin's feature vectors. Power grid topology constraints are introduced during the calculation process, meaning that the registration weight of adjacent nodes is higher than that of distant nodes, ensuring that the registration results reflect electrical coupling relationships. Based on the calculated deviation matrix, a transformation matrix containing offset vectors, topology weight scaling factors, and operating state weight scaling factors is generated through a power grid deviation compensation model to correct the digital twin's state. A consistency check is performed. If the error between the corrected twin state and the real-time state of the physical power grid does not exceed a preset threshold, the correction is complete and the twin state is updated. If the error exceeds the threshold, a deep correction process is triggered, calling the offline model retraining module to retrain the digital twin model to ensure that synchronization accuracy continues to meet the target.

[0068] Figure 4 The specific implementation of "iterative nearest point registration" and "deviation compensation" in the synchronous correction process is explained in detail. First, the multi-dimensional electrical state vectors of the power grid nodes (including voltage amplitude, voltage phase angle, active power, reactive power, frequency, etc.) are mapped to a "state point cloud" in an n-dimensional feature space. The physical power grid data constitutes the "physical state point cloud" as the target set, and the digital twin output data constitutes the "twin state point cloud" as the source set. The Euclidean distance between the source set and the target set is minimized through an iterative algorithm, and a transformation matrix containing the offset vector and scaling factor is calculated to achieve initial registration and adjustment of the twin state. In each iteration, physical constraints of the power grid (including power flow equations and voltage upper and lower limits) are introduced to project the transformation matrix into the feasible region, ensuring that the correction result satisfies the physical laws of the power grid and avoiding the generation of physically infeasible state parameters. After iterative convergence, for any remaining residual deviations, a weighted compensation is allocated based on node electrical distance and topological connectivity using a power grid deviation compensation model. This means that nodes with tight electrical coupling are assigned higher compensation weights, while nodes with weak electrical coupling are assigned lower compensation weights, thereby achieving high-precision synchronization between the twin state and the physical power grid state.

[0069] (III) High-performance computing and analysis layer (automatically allocates computing power according to tasks) This layer breaks through the centralized computing power allocation and single-algorithm analysis model by using a distributed elastic computing power resource pool and a power grid-specific intelligent analysis engine. Specifically, it includes: Distributed Elastic Computing Resource Pool: Deploys a distributed elastic computing high-performance computing platform integrating multi-core computing units, employing a hybrid heterogeneous computing architecture. A dynamic computing power scheduling mechanism based on task complexity is designed, categorizing tasks into micro-tasks (routine status monitoring), medium-tasks (load forecasting), and macro-tasks (fault emergency response, cross-domain collaboration) according to their complexity. Differentiated computing resources are allocated to tasks of different levels: micro-tasks are allocated single-node computing power, medium-tasks are allocated multi-node parallel computing power, and macro-tasks utilize heterogeneous computing power across the entire platform. Simultaneously, a task priority scheduling strategy is adopted, prioritizing urgent scheduling tasks such as fault emergency response over regular tasks, achieving efficient utilization of computing resources and addressing the problems of fixed computing power allocation and low utilization efficiency in existing technologies.

[0070] A dedicated intelligent analysis engine for the power grid: Four customized functional modules, each employing algorithm combinations different from existing technologies, enable real-time analysis of power grid status, accurate fault diagnosis, early risk warning, and scenario simulation. Specifically: Real-time analysis module: Based on power grid tagged data, it adopts a time-series sliding window analysis algorithm to analyze power grid load trends, reserve capacity distribution, and equipment operation trends in seconds, with an analysis delay of no more than 2 seconds.

[0071] Fault Diagnosis Module: The fault diagnosis module adopts the power grid topology-constrained DBSCAN algorithm, and its improvements include: Distance measurement uses electrical impedance distance instead of Euclidean distance. , in For nodes arrive The equivalent impedance, This refers to the change in current. The neighborhood search range is constrained by the power grid topology, and neighborhood points are searched only within the same electrical island; The clustering results are verified by the power grid protection action logic to eliminate false alarms that do not conform to the protection coordination relationship.

[0072] By embedding power grid topology constraints and electrical impedance distance into the DBSCAN algorithm, spatial density clustering is achieved. In a power grid, physically adjacent devices may belong to different electrical islands (such as the two sides of an open tie switch). Without topology constraints, the algorithm may misclassify electrically isolated fault points into the same cluster, leading to a significant decrease in fault location accuracy.

[0073] Prediction and early warning module: Based on time-series neural network, the prediction module combines weather forecasts and the output characteristics of new energy sources to carry out ultra-short-term load forecasts of 1-15 minutes and short-term load forecasts of 15 minutes to 2 hours. At the same time, it provides early warnings for grid risks such as voltage anomalies and islanding operation. The early warning accuracy rate is not less than 95%, and the early warning time is not less than 10 minutes.

[0074] What-if scenario simulation module: Based on a high-precision digital twin model, it presets more than 20 typical power grid scenarios such as sudden changes in new energy output, equipment failure, and load surge. It supports dispatchers to customize scenario parameters, realize rapid scenario simulation and dispatch scheme pre-playing, and has a short simulation response time and strong adaptability.

[0075] (iv) Coordinated scheduling decision-making layer (generating scheduling schemes) This layer is the core of the architecture's business applications. Based on a standardized, unified system integration platform and a scenario-based scheduling and decision-making module that supports human-machine collaboration, it breaks down data silos between systems, moving beyond simple multi-system data exchange models. Specifically, it includes: Standardized and unified system integration platform: Based on a proprietary protocol for heterogeneous power grid system integration, this platform integrates multiple power business management systems such as EMS, DMS, and WAMS, establishing a unified data bus, shared database, and access control center. This enables standardized data interoperability and bidirectional command transmission across multiple systems, with data interoperability latency not exceeding 50 milliseconds and command transmission success rate no less than 99.9%. The access control center sets tiered operation permissions according to dispatcher positions, ensuring the security of dispatch command operations and resolving the existing problems of lack of standardized protocols and low collaboration efficiency in cross-system integration.

[0076] Master Node: Power Grid Dispatch Center (sole), responsible for: Assign subscription IDs and manage access permissions for slave nodes; Issue scheduling instructions (function code 02) and receive responses from slave nodes (function code 03); Prioritize the acquisition of computing power / bandwidth from slave nodes.

[0077] From the nodes: each business subsystem (SCADA / EMS / WAMS, etc.), responsible for: Send a subscription request (function code 01) to the master node, carrying the slave node ID and business type; Publish real-time data according to the master node command (function code 02), and send heartbeats periodically (function code 04). The master-slave relationship is achieved through triple binding of master node ID, slave node ID, and subscription ID, which prevents unauthorized interactions.

[0078] The communication process is as follows: The master device assigns a subscription number; the slave device initiates a subscription; the master device replies and establishes a connection; the slave device periodically uploads data; the master device issues commands; the slave device replies with the execution result. A scenario-based scheduling decision-making module supporting log-based collaboration: Three dedicated decision-making modules are designed for routine scheduling, emergency scheduling, and cross-domain collaboration. Each module is configured with customized scheduling algorithms and rolling optimization strategies to adapt to different power grid scheduling scenarios. Specifically: Routine scheduling module: For daily load allocation, a multi-objective genetic algorithm is used to optimize the allocation of power resources with an optimization cycle of 5 minutes, achieving high-frequency rolling optimization.

[0079] Emergency dispatch module: For scenarios such as emergency response to faults and sudden changes in the output of new energy sources, a fast greedy algorithm and a pre-simulation scheme matching method are used to generate dispatch schemes in seconds, with a generation time of no more than 5 seconds.

[0080] Cross-domain collaborative module: For cross-regional power mutual assistance and cross-energy collaborative scheduling, a distributed collaborative optimization algorithm is adopted to realize dynamic power allocation of multi-regional power grids, and improve the efficiency of power resource optimization and allocation by more than 30%.

[0081] Based on digital twin models, the system enables panoramic display of power grid topology, real-time monitoring of equipment status, precise fault location, and simulation of dispatching schemes. Dispatch instructions generated from digital twin models support manual intervention and correction. Dispatchers can manually modify the automatically generated dispatching schemes and perform system constraint verification based on the content of the digital twin model, the analysis results of the analysis engine, and the actual scenario, ensuring the accuracy and flexibility of dispatching decisions.

[0082] (v) Physical power grid execution layer This layer is responsible for the high-precision execution of dispatch commands and real-time feedback of grid status. Through hierarchical response from intelligent execution devices and encrypted real-time status feedback, it breaks through the simple command execution and status reporting model. Specifically, it includes: Intelligent execution equipment: Integrating intelligent circuit breakers, reactive power compensation devices, new energy inverters, energy storage converters, and other intelligent devices, it is divided into primary and secondary execution equipment based on the real-time requirements of dispatch commands. Specifically, fault isolation and emergency power regulation utilize primary execution equipment (intelligent circuit breakers, emergency power regulation devices), with a command response time of no more than 100 milliseconds; daily load distribution and reactive power compensation utilize secondary execution equipment (reactive power compensation devices, energy storage converters, etc.), with a command response time of no more than 500 milliseconds. All intelligent devices have built-in standardized command parsing modules, which can directly parse standardized dispatch commands issued by the collaborative dispatch decision-making level, solving the problems of inconsistent command parsing and delayed response of existing equipment.

[0083] Encrypted Real-Time Status Feedback Unit: All intelligent execution devices are equipped with a status acquisition module and a secure encrypted transmission module. After executing scheduling instructions, the device collects the execution status in real time (action completion status, post-execution device operating parameters, and device health status). Three types of execution result tags—success, failure, and partial execution—are attached to the data. This data is fed back to the digital twin modeling layer via a wireless + fiber optic link through an encrypted feedback unit protocol, with an end-to-end feedback delay of no more than 1 second. The feedback data is encrypted using the SM4 national cryptographic algorithm and accompanied by a MAC authentication code to ensure data security and validity. Based on the execution result tags and feedback parameters, the digital twin modeling layer triggers updates and deviation corrections for the corresponding node's digital twin model, completing the closed-loop process of the entire scheduling process.

[0084] The implementation method of this invention is based on the above-mentioned five-layer differentiated architecture, forming a complete implementation scheme with clear steps, specific mechanisms, and significant differences from existing technologies. The scheme flowchart is as follows. Figure 5 As shown, it specifically includes the following five steps: S1: Customized Data Acquisition and Tagging Transmission. The physical power grid sensing layer collects core power grid data and externally related data hierarchically through core and auxiliary equipment. Core data acquisition equipment collects high-precision electrical quantities at a frequency of 10 milliseconds, while auxiliary data acquisition equipment collects routine monitoring quantities at a frequency of 1 second. All collected data undergoes power grid spatiotemporal tagging processing, adding three types of tags: node identifier, acquisition timestamp, and topology location code (acquisition timestamp accuracy down to the millisecond level). The tagged data is encrypted and compressed using the SM4 algorithm and then transmitted to the digital twin modeling layer via redundant wireless + fiber optic dual links. The core data transmission latency does not exceed 5 milliseconds, and the auxiliary data transmission latency does not exceed 1 second.

[0085] S2: High-precision twin construction and second-level self-correcting synchronization. The customized unified big data lake of the digital twin modeling layer performs customized cleaning (duplicate removal, completion, and normalization) and cold / hot tiered storage (72-hour hot data area + historical cold data area) on tagged data; the power grid-specific multi-dimensional twin model completes a 1:1 high-precision replica of the physical power grid based on the processed data (including a high-precision physical model, behavior prediction model, and multi-objective dynamic optimization model); the real-time synchronization self-correcting engine transforms the physical power grid collected data and the twin output data into multi-dimensional electrical feature vectors through ICP registration and power grid deviation compensation model, calculates the deviation matrix, and performs differentiated weighted correction based on the power grid topology and operating status, achieving second-level high-precision synchronization between the digital twin model and the physical power grid, with a corrected state error rate of no more than 1%.

[0086] S3: Distributed Elastic Computing Power Allocation and Grid-Specific Analysis. The distributed computing power resource pool of the high-performance computing and analysis layer allocates differentiated heterogeneous computing power based on the complexity level (micro, medium, and macro) and priority (urgent and routine) of the scheduling tasks. When an urgent scheduling task is triggered, a preemptive scheduling strategy is used to prioritize computing power supply, with a preemption response time not exceeding 500 milliseconds. The grid-specific intelligent analysis engine, based on the allocated computing power, completes one or more of the following analysis tasks through customized algorithms: Real-time status analysis: Employing a time-series sliding window algorithm, the analysis delay does not exceed 2 seconds; Fault diagnosis: Employing a serial workflow of DBSCAN and topology inference, the diagnostic delay is no more than 3 seconds, and the accuracy rate is no less than 98%. Prediction and early warning: The prediction module based on the time-series neural network can achieve ultra-short-term load prediction of 1-15 minutes and short-term load prediction of 15 minutes to 2 hours, with an early warning accuracy of no less than 95%; Scenario simulation: Using a typical What-If scenario template with custom parameters, the simulated response time is no more than 10 seconds. All analysis results are tagged with power grid spatiotemporal labels and synchronized to the collaborative scheduling decision-making layer.

[0087] S4: Standardized Cross-System Collaboration and Scenario-Based Scheduling Decisions. The standardized unified system integration platform of the collaborative scheduling decision layer integrates data from multiple systems such as EMS, DMS, and WAMS through a publish-subscribe messaging model using a heterogeneous system integration protocol, achieving standardized data interoperability with a data interoperability latency of no more than 50 milliseconds. Based on the real-time status of the digital twin model and the analysis results of the analysis engine, the three scenario-based scheduling decision modules generate customized scheduling schemes respectively: Conventional scheduling module: Employs a multi-objective genetic algorithm with a rolling optimization cycle of 5 minutes and a solution generation time of ≤5 minutes; Emergency dispatch module: It adopts a pre-rehearsed solution library for matching (direct call when similarity ≥ 0.85, matching time ≤ 1 second and fast greedy algorithm correction, solution generation time ≤ 5 seconds); Cross-domain collaboration module: Adopts a distributed collaborative optimization algorithm and multi-region iterative negotiation, improving resource allocation efficiency by more than 30%.

[0088] After the scheduling scheme is fully displayed and the scheme is pre-drilled on the human-machine collaborative visual interactive terminal, the dispatcher can perform one-click issuance or manual correction. After manual correction, the system performs constraint verification in real time. After the verification is passed, the standardized scheduling instruction message is sent to the physical power grid execution layer through the heterogeneous system integration protocol.

[0089] S5: Hierarchical precise execution and encrypted real-time status feedback. The intelligent execution devices in the physical power grid execution layer parse the dispatch command messages through the built-in heterogeneous system integration protocol command parsing module, and respond in a hierarchical manner according to level one and level two: the response time of level one execution devices (intelligent circuit breakers, emergency power regulation devices) does not exceed 100 milliseconds, and the response time of level two execution devices (reactive power compensation devices, energy storage converters, etc.) does not exceed 500 milliseconds.

[0090] After the equipment executes the command, the encrypted real-time status feedback unit collects the equipment's execution status data, adds execution result tags (success, failure, partial execution) to the data, encrypts it using the SM4 national cryptographic algorithm and adds a MAC authentication code, and then feeds it back to the digital twin modeling layer via a wireless-fiber hybrid link, with an end-to-end feedback delay of no more than 1 second. Based on the execution result tags and feedback parameters, the digital twin modeling layer triggers updates and deviation corrections for the corresponding node's digital twin model, forming a complete dynamic self-optimizing closed loop from physical power grid sensing → digital twin synchronization → intelligent analysis → scheduling decision → command execution → status feedback to twin update.

[0091] The beneficial effects of this invention are as follows: 1. A synchronization correction mechanism based on power grid feature vectorization ICP registration and power grid deviation compensation model is proposed. The mechanism replaces the three-dimensional point cloud coordinates of the general ICP algorithm with multi-dimensional electrical feature vectors, replaces the Euclidean distance with the weighted distance of electrical feature space, and replaces the uniform correction of the general ICP algorithm with the differential weighted correction of topology perception. This achieves high-precision synchronization between the digital twin model and the physical power grid, so as to address the adaptation problem of feature registration distortion and lack of topology perception of the general algorithm in the power grid scenario, and ensures that the dispatching instructions are generated based on the actual operating state of the power grid.

[0092] 2. Build a distributed elastic computing resource pool, design a dynamic computing power scheduling mechanism based on three-level classification of task complexity and priority preemptive scheduling, and combine various heterogeneous computing and parallel processing technologies such as CPU and GPU to compress the generation time of scheduling schemes in emergency scenarios and improve computing power utilization efficiency to deal with emergencies such as sudden changes in new energy output, load surges, and equipment failures.

[0093] 3. A dedicated protocol for integrating heterogeneous power grid systems was developed, and a standardized unified system integration platform was built. The platform uses a publish-subscribe messaging model to achieve standardized two-way data exchange and highly reliable command transmission between multiple power business management systems, breaking down information silos across regions and energy sources, reducing data exchange latency across systems, and improving the efficiency of power resource optimization and allocation.

[0094] 4. Customized algorithms and models are developed based on the characteristics of power grid operation, including a component dimension normalization algorithm, a three-layer twin model, and a scenario-based three-module scheduling decision algorithm. The three-layer twin model integrates physical characteristics, behavioral prediction, and multi-objective optimization to adapt to complex power grid operation scenarios. Effective power grid fault diagnosis and prediction / early warning are provided to adapt to complex power grid scenarios with a high proportion of renewable energy and drastic load fluctuations.

[0095] 5. The hardware components, software algorithms, mathematical models, interaction protocols, and scheduling strategies of the five-layer architecture are comprehensively and clearly defined, with each layer adopting an implementation method that differs from existing technologies. Simultaneously, a human-machine collaborative visual interactive terminal supports dispatchers in manually intervening and correcting the automatically generated scheduling schemes, balancing the efficiency of automated scheduling with human experience and providing technical support for the intelligent and refined scheduling of smart grids.

[0096] The intelligent power resource scheduling device provided by the present invention is described below. The intelligent power resource scheduling device described below can be referred to in correspondence with the intelligent power resource scheduling method described above.

[0097] like Figure 6 The image shows an intelligent power resource scheduling device provided by the present invention, comprising: The data acquisition module 610 is used to acquire real-time data of the physical power grid and to tag the real-time data. The tags include at least node identifier, acquisition time and topology location. The digital twin construction module 620 is used to construct a digital twin based on real-time collected data after labeling. The digital twin includes a physical model, a power grid behavior prediction model, and a multi-objective dynamic optimization model. The synchronization correction module 630 is used to synchronize and correct the digital twin, so that the error between the state of the digital twin and the real-time state of the physical power grid is kept within a preset threshold. The scheduling processing module 640 is used to allocate computing resources for analysis and processing based on the synchronized and corrected digital twin, generate and execute scheduling schemes, and the analysis and processing includes power grid status analysis, fault diagnosis, risk warning or scenario simulation.

[0098] Specifically, the functions of each module in the intelligent power resource scheduling device provided in this embodiment of the invention correspond one-to-one with the operation flow of each step in the above method-like embodiments, and the achieved effects are also the same. For details, please refer to the above embodiments, and this will not be repeated in this embodiment of the invention.

[0099] Figure 7 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 7As shown, the electronic device may include: a processor 710, a communication interface 720, a memory 730, and a communication bus 740. The processor 710, communication interface 720, and memory 730 communicate with each other via the communication bus 740. The processor 710 can call logical instructions in the memory 730 to execute an intelligent power resource scheduling method. This method includes: acquiring real-time data from the physical power grid and tagging the real-time data, with tags including at least node identifiers, acquisition time, and topological location; constructing a digital twin based on the tagged real-time data, the digital twin including a physical model, a power grid behavior prediction model, and a multi-objective dynamic optimization model; synchronously correcting the digital twin to keep the error between the state of the digital twin and the real-time state of the physical power grid within a preset threshold; and allocating computing resources for analysis and processing based on the synchronously corrected digital twin to generate and execute a scheduling scheme, the analysis and processing including power grid state analysis, fault diagnosis, risk warning, or scenario simulation.

[0100] Furthermore, the logical instructions in the aforementioned memory 730 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0101] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the intelligent power resource scheduling method provided by the above methods. The method includes: acquiring real-time data of the physical power grid and tagging the real-time data, wherein the tags include at least node identifiers, acquisition time, and topological location; constructing a digital twin based on the tagged real-time data, wherein the digital twin includes a physical model, a power grid behavior prediction model, and a multi-objective dynamic optimization model; synchronously correcting the digital twin to keep the error between the state of the digital twin and the real-time state of the physical power grid within a preset threshold; and allocating computing resources for analysis and processing based on the synchronously corrected digital twin to generate and execute a scheduling scheme, wherein the analysis and processing includes power grid state analysis, fault diagnosis, risk warning, or scenario simulation.

[0102] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the intelligent power resource scheduling method provided by the above methods. The method includes: acquiring real-time data of the physical power grid and tagging the real-time data, wherein the tags include at least node identifiers, acquisition time, and topological location; constructing a digital twin based on the tagged real-time data, wherein the digital twin includes a physical model, a power grid behavior prediction model, and a multi-objective dynamic optimization model; synchronously correcting the digital twin to keep the error between the state of the digital twin and the real-time state of the physical power grid within a preset threshold; and allocating computing resources for analysis and processing based on the synchronously corrected digital twin to generate and execute a scheduling scheme, wherein the analysis and processing includes power grid state analysis, fault diagnosis, risk warning, or scenario simulation.

[0103] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0104] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0105] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent power resource scheduling, characterized in that, include: Acquire real-time data from the physical power grid and label the real-time data. The labels include at least node identifier, acquisition time, and topological location. A digital twin is constructed based on real-time collected data after labeling. The digital twin includes a physical model, a power grid behavior prediction model, and a multi-objective dynamic optimization model. The digital twin is synchronously corrected so that the error between the state of the digital twin and the real-time state of the physical power grid is kept within a preset threshold. Based on the synchronized and corrected digital twin, computing resources are allocated for analysis and processing to generate and execute scheduling schemes. The analysis and processing includes power grid status analysis, fault diagnosis, risk warning, or scenario simulation.

2. The intelligent power resource scheduling method according to claim 1, characterized in that, The synchronous correction of the digital twin includes: The real-time acquired data and the output data of the digital twin are respectively converted into multi-dimensional electrical feature vectors; The deviation matrix between the electrical feature vector corresponding to the real-time acquired data and the electrical feature vector corresponding to the output data is calculated based on the weighted distance of the electrical feature space. The weighted distance of the electrical feature space is weighted differently using power grid topology weight and operating state weight. The transformation matrix is ​​calculated based on the deviation matrix, and the state of the digital twin is corrected using the transformation matrix.

3. The intelligent power resource scheduling method according to claim 1, characterized in that, The fault diagnosis steps include: Acquire electrical quantity measurement data and power grid topology connections at each node of the power grid; Based on the power grid topology connection relationship, electrical islands are divided, and the electrical impedance distance between nodes within the same electrical island is calculated; Cluster analysis was performed on the electrical impedance distance to obtain the clustering results; The clustering results are verified, and the fault location is determined based on the verified clustering results.

4. The intelligent power resource scheduling method according to claim 1, characterized in that, The power grid behavior prediction model is implemented based on a temporal neural network. The input layer of the temporal neural network is embedded with a power grid topology adjacency matrix. The hidden layer of the temporal neural network is equipped with a constraint gating unit to suppress abnormal output when the voltage or frequency exceeds a safety threshold. The loss function of the temporal neural network includes a data fitting loss term and a power grid power balance penalty term.

5. The intelligent power resource scheduling method according to claim 1, characterized in that, The analysis and processing of allocated computing resources includes: The scheduled tasks are divided into different levels based on their complexity and / or priority. Different levels of computing resources are allocated to scheduling tasks of different priorities, with higher priority scheduling tasks receiving priority in computing resource allocation.

6. The intelligent power resource scheduling method according to claim 1, characterized in that, When executing the scheduling scheme, the method further includes: Collect execution status data and feed it back to update the digital twin.

7. An intelligent power resource dispatching device, characterized in that, include: The data acquisition module is used to acquire real-time data of the physical power grid and to tag the real-time data. The tags include at least node identifier, acquisition time and topological location. A digital twin construction module is used to construct a digital twin based on real-time collected data after labeling. The digital twin includes a physical model, a power grid behavior prediction model, and a multi-objective dynamic optimization model. The synchronization correction module is used to synchronize and correct the digital twin so that the error between the state of the digital twin and the real-time state of the physical power grid is kept within a preset threshold. The scheduling processing module is used to allocate computing resources for analysis and processing based on the synchronized and corrected digital twin, generate and execute scheduling schemes, and the analysis and processing includes power grid status analysis, fault diagnosis, risk warning or scenario simulation.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the intelligent power resource scheduling method as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the intelligent power resource scheduling method as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the intelligent power resource scheduling method as described in any one of claims 1 to 6.

Images