Power grid operation risk prediction and collaborative governance system based on digital twinning

Abstract

The invention discloses a power grid operation risk prediction and collaborative treatment system based on digital twinning. The system comprises a basic support layer, a service management and control layer, an intelligent analysis layer and a collaborative treatment layer. Wherein the basic support layer constructs a power grid equipment operation state real-time mapping model through a digital twinning technology; the business management and control layer realizes cooperation of operation monitoring and maintenance management; the intelligent analysis layer analyzes the data based on an improved Bayesian network to realize power grid situation assessment and risk prediction; and the collaborative governance layer executes hierarchical response according to an analysis result. The technical problems of information island, inaccurate risk prediction, untimely response and the like in traditional power grid operation management are solved, collaborative linkage of accurate sensing of the power grid operation state and intelligent risk prediction and disposal is realized, the intelligent and automatic level of power grid operation management is remarkably improved, and the power grid operation management efficiency is improved. And a powerful guarantee is provided for safe and stable operation of a power grid.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent operation and maintenance of power systems, and in particular to a power grid operation risk prediction and collaborative management system based on digital twins. Background Art

[0002] With the in-depth promotion of smart grid construction and the large-scale grid connection of new energy, the grid structure is becoming increasingly complex and the operation mode is becoming more diverse, which puts higher requirements on grid dispatching, operation and control. Although modern power grids have a complete automation and information foundation, there are still the following problems in intelligent analysis and collaborative governance:

[0003] 1. Insufficient data value mining: Although the existing SCADA system has high-frequency sampling capabilities and a wide coverage, it still needs to be strengthened in terms of multi-source data fusion analysis and equipment operation mechanism modeling. In particular, the lack of deep mining and knowledge extraction of massive operating data makes it difficult to support more accurate status assessment and decision analysis.

[0004] 2. The depth of business collaboration is not enough. Although the power grid dispatching technical support system has achieved data integration, the depth of collaboration at the business level is still insufficient. The linkage mechanism of operation monitoring, maintenance management, fault handling and other links needs to be improved, making it difficult to give full play to the value brought by data sharing.

[0005] 3. Limited risk prediction capabilities. Existing early warning systems mainly focus on anomaly detection of single devices or local systems, and do not adequately consider the correlation between devices and fault propagation characteristics. The lack of intelligent analysis models based on network topology and historical experience makes it difficult to accurately predict and proactively prevent risks.

[0006] 4. The emergency response mechanism needs to be optimized. Although a relatively complete emergency plan system has been established, there is still room for improvement in rapid assessment and intelligent decision-making support.

[0007] 5. The overall coordination level is not high. Although the existing management and control system covers all business links, it lacks a systematic collaborative governance mechanism. The linkage effect between various links has not been fully utilized, making it difficult to achieve active management of the entire life cycle of the power grid.

[0008] Although the industry has developed a variety of intelligent power grid operation and maintenance management systems, there are still the following technical challenges in deep application:

[0009] 1. Existing digital twin applications lack unified modeling of equipment topology, maintenance operations, and scheduling operations, which affects the deep mining of data value;

[0010] 2. The existing risk prediction model does not adequately consider the correlation between devices and the fault propagation mechanism, especially in the use of Bayesian networks for multi-dimensional analysis. There is still room for improvement;

[0011] 3. The coordination mechanism between operation monitoring and maintenance management is imperfect, making it difficult to achieve seamless connection and closed-loop management of risk warning and emergency response.

[0012] Therefore, it is necessary to develop a power grid operation and control system that integrates digital twin modeling, Bayesian network analysis, and collaborative governance mechanism to improve the intelligence level and security assurance capabilities of power grid operation. Summary of the invention

[0013] The problem to be solved by the present invention is how to provide a power grid operation control system based on digital twin technology, with intelligent risk prediction capabilities and supporting global collaborative governance. In order to solve the above technical problems, the technical solutions adopted by the present invention are as follows:

[0014] A power grid operation risk prediction and collaborative governance system based on digital twins, characterized by comprising a basic support layer, a business management and control layer, an intelligent analysis layer and a collaborative governance layer;

[0015] The basic support layer includes a data acquisition submodule, a data processing submodule and a data modeling submodule, which collect and store power grid operation data, and after data cleaning and standardization, construct a power grid digital twin model including equipment topology, maintenance operation procedures and dispatching operation procedures. The digital twin model reflects the operating status of the physical equipment of the power grid in real time.

[0016] The business control layer includes an operation monitoring submodule and a maintenance management submodule. A business monitoring system is built based on the digital twin model to realize operation monitoring and maintenance management, and transmit structured monitoring data to the intelligent analysis layer;

[0017] The intelligent analysis layer includes a situation awareness submodule and a risk analysis submodule, which analyzes data based on an improved Bayesian network, outputs power grid situation assessment and risk prediction results, and transmits them to the collaborative governance layer;

[0018] The collaborative governance layer includes an emergency response submodule, which executes emergency response according to the analysis results and issues disposal instructions through the mobile application terminal.

[0019] Furthermore, the basic support layer includes: a data acquisition submodule for collecting real-time operation data of the power grid, specifically including: equipment operation attributes, including voltage level, operating current, active power, reactive power, equipment opening and closing status, equipment operation mode; transmission parameters, including data sampling period, data quality identifier, data timestamp, data source identifier; basic equipment information, including basic equipment attributes and connection relationships; equipment maintenance information; spatial location information;

[0020] The data processing submodule performs denoising, deduplication and standardization on the collected data and establishes a unified data format standard;

[0021] The data modeling submodule constructs a power grid digital twin model based on the processed data. The digital twin model includes an equipment topology relationship model, a maintenance operation procedure model, and a scheduling operation process model.

[0022] Furthermore, the method for constructing the power grid digital twin model by the data modeling submodule includes:

[0023] Equipment topology relationship modeling, abstracting the power equipment in the power grid as nodes, abstracting the electrical connection relationship between the nodes as edges, and constructing a directed graph model of the power grid structure; the nodes of the directed graph model contain basic equipment attributes and operation attributes; the edges of the directed graph model contain connection attributes and transmission attributes;

[0024] Maintenance operation procedure modeling breaks down the maintenance operation process into three levels: operation items, operation steps, and operation points. Each operation item contains specific operation instructions, safety measures, and quality standards. Each operation step sets the execution sequence and completion standard. Each operation point records the key control points and acceptance criteria.

[0025] Scheduling operation process modeling divides the scheduling operation process into three stages: pre-arrangement, execution and supervision. For each stage, a sequence of operation steps is established, and the triggering conditions, execution actions and completion marks of each step are defined. The system automatically records the execution time, operator and operation results of each step.

[0026] Among them, the digital twin model maintains synchronization with the actual power grid status in the following ways: determining the optimal collection cycle based on the importance and operating status of the equipment and dynamically adjusting it; using data caching and version control mechanisms to ensure data consistency; setting up model update exception handling and rollback mechanisms; updating the power grid operation data according to the preset collection cycle; updating the equipment status data according to the preset collection cycle; updating the operating status in response to scheduling operation instructions; and updating the maintenance status in response to maintenance operation information.

[0027] Furthermore, the specific implementation method of running the monitoring submodule in the business management and control layer includes:

[0028] Establish monitoring objects based on the equipment topology relationship model, map equipment nodes to monitoring points, map topological relationships to monitoring associations, and establish parameter threshold ranges for each monitoring point;

[0029] Establish an operation condition library based on the scheduling operation process model, extract the parameter change characteristics of the standard operation steps, establish the corresponding relationship between the operation type and the parameter change mode, and form a standard condition feature library;

[0030] Perform operation monitoring, collect monitoring point data in real time, identify parameter out-of-limit situations, extract related measurement point data, match operating condition characteristics, and generate structured abnormality descriptions;

[0031] Furthermore, the specific implementation method of the maintenance management submodule in the business control layer includes:

[0032] Establish operating standards based on the maintenance operating procedure model, convert operating procedures into standard operating items, define the execution and acceptance rules of operating items, and establish the association between operating items and equipment status;

[0033] Perform maintenance management, track maintenance work execution, record operation process data, associate equipment operating status, and generate standardized maintenance records;

[0034] Data association analysis: establish the time series association between maintenance work and equipment status, analyze parameter changes before and after maintenance, and extract maintenance effect characteristics;

[0035] Furthermore, the intelligent analysis layer specifically includes: constructing observation variables and state variables of the improved Bayesian network, taking the equipment abnormality description and operating condition characteristics output by the operation monitoring submodule as the network observation variable X, taking the operation record and equipment status output by the maintenance management submodule as the network state variable Y, and determining the connection relationship E between network nodes based on the equipment topology relationship matrix T in the digital twin model;

[0036] Furthermore, the specific implementation of improving the Bayesian network includes:

[0037] Based on the network observation variables, state variables and connection relationships, a dynamic Bayesian network structure B(V,E) containing T time segments is constructed, where V={X∪Y} represents the network node set, each node contains state variables at time t and t-1, and E represents the temporal transfer relationship between nodes;

[0038] A node conditional probability table (CPT) is constructed based on the equipment-associated fault records, and the CPT parameters are iteratively optimized using the EM algorithm. The EM algorithm calculates the conditional probability of the unobserved state through the E step, and maximizes the expected likelihood estimation probability parameters of the complete data through the M step. The optimal parameters are obtained through iterative optimization. The optimization process of the EM algorithm is as follows:

[0039] Step E, estimate the conditional probability distribution of unobserved data based on the current parameter θ:

[0040] P(Y|D,CPT)

[0041] Wherein D is the observation data, i.e., the equipment abnormality description and operating condition characteristics output by the operation monitoring submodule, and Y is the state variable, i.e., the operation record and equipment status output by the maintenance management submodule;

[0042] In the M step, the expected likelihood function of the complete data is maximized to obtain new parameter estimates:

[0043] CPT * =argmaxP(D|CPT)

[0044] Among them, CPT * is the optimal conditional probability estimate, P(D|CPT) is the likelihood probability of observing data D under the current conditional probability;

[0045] The probability transfer matrix is ​​modified by the device importance weight, and the formula is:

[0046] P ′ (Y|X)=W×P(Y|X)

[0047] Among them, P ′ (Y|X) is the corrected conditional probability, P(Y|X) is the original conditional probability, X is the observed variable, i.e., the device abnormality description and operating condition characteristics output by the operation monitoring submodule, W is the device importance weight matrix, which is determined based on the voltage level in the basic attributes of the device, the connection relationship in the topological relationship matrix, and the operation attributes; for the variable X, the probability distribution of the network state variable is calculated using the MAP criterion, and the formula is:

[0048] P(Y|X)=argmaxP(X|Y)P(Y)

[0049] Among them, P(X|Y) is the likelihood probability (the dependence of the observed variable on the state variable), and P(Y) is the prior probability of the state variable;

[0050] The forward algorithm is used to calculate the risk probability at time t+k, and the formula is:

[0051] P(Y t+k |X 1:t )=∑P(Y t+k |Y t )P(Y t |X 1:t )

[0052] Among them, Y t+k is the state at time t+k, X 1:t is the observation sequence from time 1 to time t, P(Y t+k |Y t ) is the state transition probability, P(Y t |X 1:t ) is the state posterior probability at time t;

[0053] The risk propagation path is calculated based on the conditional risk propagation probability matrix, and the formula is:

[0054] P=P(Y|X)×R

[0055] Wherein, R is the risk association matrix, which describes the association relationship between the devices based on the device topology relationship matrix T;

[0056] Furthermore, the specific implementation of the collaborative governance layer includes: establishing a hierarchical response mechanism based on the risk probability distribution and propagation path output by the intelligent analysis layer, mapping the risk probability P(Y|X) to four-level response levels, determining the key equipment within the affected range according to the risk propagation probability matrix P, and pushing the risk level and key equipment information to the mobile application terminal of relevant personnel;

[0057] Furthermore, the system also includes a mobile application terminal:

[0058] The mobile application terminal interacts with the system in real time through the 4G / 5G network, receives the risk level and key equipment information output by the intelligent analysis layer, displays the disposal instructions issued by the collaborative governance layer, and transmits the on-site disposal situation to the system in real time, including equipment status changes and disposal progress, to cooperate with the collaborative governance layer to form a closed-loop management.

[0059] The beneficial effects of the present invention are as follows:

[0060] 1. By building a digital twin model that integrates equipment topology, maintenance procedures, and dispatching operation processes, unified modeling and real-time mapping of power grid operation status are achieved, providing a comprehensive data foundation for risk prediction;

[0061] 2. Based on the improved Bayesian network and equipment importance weight correction mechanism, combined with the conditional risk propagation probability matrix, the accuracy and reliability of risk prediction are improved, and a breakthrough is achieved from single equipment failure warning to global risk prediction;

[0062] 3. Adopt a hierarchical response mechanism based on risk level and impact scope, and realize real-time feedback of on-site disposal through mobile terminals, forming a complete closed-loop management from risk warning to on-site disposal; BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 This is the architecture diagram of the power grid operation risk prediction and collaborative governance system based on digital twins.

[0064] Figure 2 Schematic diagram of the power grid digital twin model. DETAILED DESCRIPTION

[0065] The present invention provides a power grid operation risk prediction and collaborative governance system based on digital twins, which includes a basic support layer, a business management layer, an intelligent analysis layer, and a collaborative governance layer, and a closed-loop management mechanism for data circulation and business collaboration is formed between each layer. The technical solution of the present invention is described in detail below in conjunction with specific embodiments.

[0066] Example 1: Power grid operation risk prediction and collaborative governance system based on digital twins

[0067] Reference Figure 1 The system of this embodiment adopts a layered architecture design, and realizes the prediction and management of power grid operation risks through the coordinated cooperation of the basic support layer, business management layer, intelligent analysis layer and collaborative governance layer. The specific implementation process of the system is as follows:

[0068] The basic support layer collects real-time grid operation data through the data acquisition submodule. This submodule adopts a distributed data acquisition architecture and deploys data acquisition units in each substation and key nodes. The data acquisition unit adopts a redundant design. Each unit contains two sets of acquisition systems, the main and the backup. When the main system fails, it automatically switches to the backup system to ensure the reliability of data acquisition. The collected data includes the following categories:

[0069] Equipment operation attribute data, including: voltage level (35kV, 110kV, 220kV, 500kV, etc.), operating current (accurate to 0.1A), active power (accurate to 0.1kW), reactive power (accurate to 0.1kVar), equipment opening and closing status (open, closed, fault), equipment operation mode (normal, maintenance, standby), etc. The sampling period of these parameters is configured differently according to the importance of the equipment. The sampling period of important equipment such as the main transformer is 100ms, and that of general equipment is 1s.

[0070] Transmission parameter data, including: data sampling period (configurable, range 100ms-60s), data quality identification (normal, suspicious, error), data timestamp (accurate to milliseconds), data source identification (equipment code, acquisition unit number) and other information. The system monitors data quality in real time, and automatically triggers the data review mechanism when the data quality identification is found to be "suspicious" or "error".

[0071] Equipment basic information data, including: equipment model, manufacturer, commissioning time, rated parameters, maintenance cycle and other basic attributes, as well as primary system connection relationship and secondary system connection relationship between equipment, etc. These data are synchronously obtained from the equipment asset management system, and full synchronization is performed every 24 hours, during which the incremental synchronization method is used to update the changed data.

[0072] Equipment maintenance information data, including historical maintenance records, fault records, technical transformation records, etc. The system establishes an equipment maintenance archive to record detailed information on each maintenance activity, including maintenance time, maintenance content, maintenance personnel, maintenance effect evaluation, etc.

[0073] Spatial location information data uses the Beidou positioning system to obtain the precise geographic coordinates of the device. The location accuracy of outdoor equipment is better than 10 meters, and the location coordinates of indoor equipment are determined through pre-calibration.

[0074] The collected data is processed by the data processing submodule. This submodule adopts a three-level data processing mechanism: the first level is real-time processing, which performs noise filtering on the collected data, uses the median filter algorithm to eliminate mutation interference, and uses the mean filter algorithm to smooth the data curve for analog quantities; the second level is quasi-real-time processing, including data deduplication, data completion and outlier processing. The system establishes a data change feature model through historical data analysis for the identification and processing of outliers; the third level is offline processing, including data cleaning, data standardization and data quality assessment, and performs in-depth processing of historical data through batch processing.

[0075] Standardization processing uses a unified data format standard, including: the timestamp is uniformly UTC time, and the time zone information is recorded; analog data is uniformly converted to the international system of units; state data uses standard state coding; equipment identification uses a unified coding rule. The processed data is stored in a distributed time series database according to the time series, which is convenient for subsequent rapid retrieval and analysis.

[0076] Reference Figure 2 The data modeling submodule builds a digital twin model of the power grid based on the processed standard data. The model adopts a hierarchical modeling method, including three levels: device layer, system layer and business layer. At the device layer, a digital model of independent equipment is built, including the geometric model, physical model and behavior model of the equipment; at the system layer, an association model between devices is built to describe the physical connection, logical relationship and information interaction between devices; at the business layer, a business process model is built to describe the rules and processes of business activities such as maintenance, operation and scheduling.

[0077] In the process of modeling the topological relationship of equipment, a graph database is used to store the connection relationship between devices. The power equipment in the power grid is abstracted as nodes, and each node contains basic attributes such as equipment ID, equipment type, voltage level, and operating attributes such as operating current and active power. The electrical connection relationship between nodes is abstracted as edges, and the attributes of the edges include connection type (busbar connection, T-connection, etc.), wire model, line length and other connection attributes, as well as transmission attributes such as flow direction and impedance parameters. The system supports dynamic update of topological relationships, and automatically updates the topological model when the equipment status changes (such as switch opening and closing, line shutdown, etc.).

[0078] In the process of modeling maintenance operation procedures, a hierarchical model structure is adopted. At the operation item level, each operation item contains detailed operation instructions (text instructions, operation videos, precautions, etc.), safety measures (personal protection requirements, on-site safety measures, etc.) and quality standards (key parameter requirements, acceptance criteria, etc.). At the operation step level, define the execution sequence between steps (serial, parallel, conditional execution, etc.), and set the completion standard for each step (parameter requirements, status confirmation, etc.). At the operation point level, record the specific requirements (operation methods, precautions, etc.) and acceptance criteria (qualified standards, acceptance methods, etc.) for each key control point.

[0079] In the process of modeling the dispatching operation process, the state machine model is used to describe the operation process. In the pre-arrangement stage, the system automatically generates an operation ticket, which contains detailed operation steps and safety measures; in the execution stage, the system tracks the operation execution in real time, records the execution time, operator and operation results of each step; in the supervision stage, the system monitors the operation process in real time and promptly alarms when abnormal conditions are found. Each operation step defines clear trigger conditions (time conditions, state conditions, etc.), execution actions (opening and closing operations, parameter adjustment, etc.) and completion marks (state confirmation, parameter verification, etc.).

[0080] To ensure that the digital twin model is synchronized with the actual power grid status, the system adopts a multi-level synchronization mechanism:

[0081] 1. Collection cycle optimization: The system dynamically adjusts the collection cycle based on the importance of the equipment (comprehensively evaluated by factors such as equipment type, voltage level, access capacity, etc.) and operating status (normal, abnormal, alarm, etc.). The collection cycle of important equipment in normal state is 1s, which is automatically shortened to 100ms when an abnormality occurs; the collection cycle of general equipment is 5s, which is shortened to 1s when an abnormality occurs.

[0082] 2. Data consistency guarantee: A two-layer cache mechanism is adopted. The first layer is the memory cache, which is used to store the real-time data of the last 5 minutes; the second layer is the persistent cache, which is used to store the historical data of the last 24 hours. The system adopts a version control mechanism to assign a unique version number to each data update to ensure data consistency.

[0083] 3. Exception handling mechanism: Set up the model update exception handling process, including data anomaly detection (based on statistical features and machine learning algorithms), anomaly classification (divided into three levels: warning, serious, and fatal), exception handling (combination of automatic processing and manual intervention), and exception recording (detailed recording of abnormal situations and handling processes). When a fatal exception occurs, the model rollback mechanism is automatically triggered to restore to the most recent stable version.

[0084] The business control layer conducts operation monitoring and maintenance management based on the digital twin model. The operation monitoring submodule first establishes a monitoring object model and maps the device nodes to monitoring measurement points. Each measurement point contains attributes such as measurement point ID, measurement point type, collection cycle, and alarm threshold. The system supports multiple types of threshold settings: fixed threshold, dynamic threshold (automatically adjusted according to the statistical characteristics of historical data), associated threshold (considering the status of related measurement points), etc.

[0085] When establishing the operating condition library, the system performs analysis based on the above-mentioned dispatching operation process model. By analyzing the operation data of the three stages of pre-arrangement, execution and supervision, the system extracts the parameter change characteristics of the standard operation steps. For example, for the transformer voltage regulation operation, key parameter characteristics including voltage change rate, load current change, voltage value before and after voltage regulation are extracted; for the line switching operation, characteristic parameters such as switch action sequence, bus voltage change, and power flow transfer are extracted. The system establishes a correspondence between these characteristics and the corresponding operation types to form a standard operating condition feature library. Based on the established standard operating condition feature library, the system conducts real-time monitoring. First, the monitoring measurement point data is collected in real time, including the voltage, current, power and other parameters related to the above-mentioned operations. When it is found that the parameters exceed the preset threshold range, the system identifies the parameter out-of-limit situation. Subsequently, the system automatically extracts the relevant measurement point data that has a topological association relationship with the out-of-limit parameter. By matching the parameter change characteristics collected in real time with the standard operating condition feature library, it is determined whether there is an abnormal operating condition at present. Finally, the system generates a structured abnormal description based on the matching results.

[0086] The maintenance management submodule establishes a standard operation library based on the maintenance operation procedure model. Each standard operation item contains detailed execution specifications, such as operation content, operation tools, and operation environment requirements. The system establishes a mapping relationship between operation items and equipment status, and automatically evaluates the necessity and priority of maintenance through status monitoring data. During the maintenance process, the system records the operation execution status in real time, including operation time, operator, operation content, detection data, etc., and evaluates the operation quality through image recognition and sensor data analysis.

[0087] Through data association analysis, the system establishes a time-series association model between maintenance operations and equipment status. The model includes three dimensions: time dimension (before maintenance, during maintenance, and after maintenance), status dimension (operating parameters, alarm information, and fault records), and operation dimension (operation content, operation quality, and operation effect). The system uses this model to analyze maintenance effects, evaluate maintenance quality, and continuously optimize maintenance strategies.

[0088] The intelligent analysis layer conducts analysis based on the improved Bayesian network. When building the network structure, the system first determines the observation variable set X and the state variable set Y. The observation variables include: equipment operating parameters (voltage, current, etc.), equipment status indicators (temperature, vibration, etc.), alarm information (type, level, etc.), operation records (type, time, etc.), etc. The state variables include: equipment health status, fault type, risk level, maintenance suggestions, etc.

[0089] The implementation process of the improved Bayesian network adopts a distributed computing framework to improve the computing efficiency of large-scale networks. The system first constructs a dynamic Bayesian network structure B(V, E) containing T time segments (generally T=24, representing 24 hours). When constructing the node conditional probability table (CPT), the system first constructs an initial node conditional probability table based on the device associated fault record. Specifically, the system analyzes the fault propagation chain in the device associated fault record to determine the conditional probability relationship between each node. For example, when device A fails, the probability of failure of the adjacent device B is analyzed based on the associated fault record, thereby constructing the conditional probability table item from node A to node B. For each node pair in the network, the system constructs the corresponding conditional probability table item by analyzing its historical associated fault records. After completing the initial construction of the CPT, the system uses the EM algorithm to perform iterative parameter optimization. The specific implementation of the EM algorithm includes:

[0090] 1. Step E: Calculate the conditional probability P(Y|D, CPT) of the unobserved state Y. The system uses variational inference to approximate the posterior distribution and improve computational efficiency.

[0091] 2.M step: maximize the expected likelihood function of the complete data. The system uses the gradient ascent method to optimize parameters, and the learning rate adopts an adaptive adjustment strategy.

[0092] 3. Convergence judgment: Stop the iteration when the parameter change is less than the preset threshold (such as 0.001) or reaches the maximum number of iterations (such as 100 times).

[0093] The system modifies the probability transfer matrix by the device importance weights. The calculation of the weight matrix W takes into account the following factors:

[0094] 1. Voltage level weight: 500kV is 1.0, 220kV is 0.8, 110kV is 0.6, and 35kV is 0.4.

[0095] 2. Topological importance weight: calculated based on the degree centrality of the device in the network.

[0096] 3. Operation status weight: calculated based on factors such as equipment operation time and load rate.

[0097] When making risk predictions, the system uses a forward algorithm to calculate the risk probability at time t+k. The specific calculation formula is:

[0098] P(Y t+k |X1: t )=∑P(Y t+k |Y t )P(Y t |X1: t )

[0099] Among them, Y t+k is the state at time t+k, X 1:t is the observation sequence from time 1 to time t, P(Y t+k |Y t ) is the state transition probability, P(Y t |X 1:t ) is the posterior probability of the state at time t. Through this calculation process, the system can predict the risk probability at future moments based on the currently observed device state sequence.

[0100] For the calculation of risk propagation path, the system performs analysis based on the conditional risk propagation probability matrix, and the calculation formula is:

[0101] P=P(Y|X)×R

[0102] Among them, R is a risk association matrix, which describes the association relationship between devices based on the device topology relationship matrix T. Through this calculation process, the system can determine the risk propagation path between devices.

[0103] The collaborative governance layer establishes a four-level response mechanism based on the risk analysis results P(Y|X):

[0104] 1. Level 1 response (red): risk probability > 0.8, large impact range, immediate response required

[0105] 2. Level 2 response (orange): 0.6 < risk probability ≤ 0.8, medium impact, requires timely disposal

[0106] 3. Level 3 response (yellow): 0.4<risk probability≤0.6, impact range

[0107] 4. Level 4 response: risk probability ≤ 0.4

[0108] The system determines the key equipment within the impact range based on the risk propagation probability matrix P. For each response level, the system generates corresponding disposal instructions, including risk description, key equipment list and disposal suggestions.

[0109] The mobile application terminal of this embodiment interacts with the system in real time via the 4G / 5G network. In a preferred embodiment, the basic functions of the terminal include:

[0110] 1. Receive risk level information output by the intelligent analysis layer

[0111] 2. Display the disposal instructions issued by the collaborative governance layer

[0112] 3. Report on-site handling status, including: equipment status change information and handling progress.

[0113] The mobile application terminal transmits the collected on-site information back to the system in real time, and cooperates with the collaborative governance layer to form a closed-loop management. The system records the information of the entire disposal process to achieve full tracking and management of the risk disposal process.

[0114] This embodiment achieves accurate prediction and collaborative management of power grid operation risks through multi-level architecture design and the application of digital twin technology. The system uses an improved Bayesian network for risk analysis and implements closed-loop management of on-site disposal through mobile application terminals. Through actual application verification, the system can effectively improve the risk management and control capabilities of power grid operation, and has significant practical value and promotion significance.

Claims

1. A power grid operation risk prediction and collaborative management system based on digital twins, characterized in that: It includes basic support layer, business control layer, intelligent analysis layer and collaborative governance layer; The basic support layer includes a data acquisition submodule, a data processing submodule and a data modeling submodule, which collect and store power grid operation data, and after data cleaning and standardization, construct a power grid digital twin model including equipment topology, maintenance operation procedures and dispatching operation procedures. The digital twin model reflects the operating status of the physical equipment of the power grid in real time. The business control layer includes an operation monitoring submodule and a maintenance management submodule. A business monitoring system is built based on the digital twin model to realize operation monitoring and maintenance management, and transmit structured monitoring data to the intelligent analysis layer; The intelligent analysis layer includes a situation awareness submodule and a risk analysis submodule, which analyzes data based on an improved Bayesian network, outputs power grid situation assessment and risk prediction results, and transmits them to the collaborative governance layer; the collaborative governance layer includes an emergency response submodule, which executes emergency response according to the analysis results and issues disposal instructions through a mobile application terminal.

2. According to a digital twin-based power grid operation risk prediction and collaborative management system according to claim 1, it is characterized in that: The basic support layer comprises: The data acquisition submodule is used to collect real-time operation data of the power grid, including: equipment operation attributes, including voltage level, operating current, active power, reactive power, equipment opening and closing status, and equipment operation mode; transmission parameters, including data sampling period, data quality identification, data timestamp, and data source identification; basic equipment information, including basic equipment attributes and connection relationships; equipment maintenance information; spatial location information; data processing submodule, denoising, de-duplication, and standardization of the collected data to establish a unified data format standard; The data modeling submodule constructs a power grid digital twin model based on the processed data. The digital twin model includes an equipment topology relationship model, a maintenance operation procedure model, and a scheduling operation process model.

3. According to a digital twin-based power grid operation risk prediction and collaborative management system according to claim 2, it is characterized in that: The method for constructing a power grid digital twin model by the data modeling submodule includes: Equipment topology relationship modeling, abstracting the power equipment in the power grid as nodes, abstracting the electrical connection relationship between the nodes as edges, and constructing a directed graph model of the power grid structure; the nodes of the directed graph model contain basic equipment attributes and operation attributes; the edges of the directed graph model contain connection attributes and transmission attributes; maintenance operation procedure modeling, decomposing the maintenance operation process into three levels: operation items, operation steps and operation points. Each operation item contains specific operation instructions, safety measures and quality standards. Each operation step sets the execution sequence and completion standards. Each operation point records the key control points and acceptance criteria. Scheduling operation process modeling divides the scheduling operation process into three stages: pre-arrangement, execution and supervision. For each stage, a sequence of operation steps is established, and the triggering conditions, execution actions and completion marks of each step are defined. The system automatically records the execution time, operator and operation results of each step. Among them, the digital twin model maintains synchronization with the actual power grid status in the following ways: determining the optimal collection cycle based on the importance and operating status of the equipment and dynamically adjusting it; using data caching and version control mechanisms to ensure data consistency; setting up model update exception handling and rollback mechanisms; updating the power grid operation data according to the preset collection cycle; updating the equipment status data according to the preset collection cycle; updating the operating status in response to scheduling operation instructions; and updating the maintenance status in response to maintenance operation information.

4. According to a digital twin-based power grid operation risk prediction and collaborative management system according to claim 1, it is characterized in that: The specific implementation method of the operation monitoring submodule in the business management and control layer includes: Establish monitoring objects based on the equipment topology relationship model, map equipment nodes to monitoring points, map topological relationships to monitoring associations, and establish parameter threshold ranges for each monitoring point; Establish an operation condition library based on the scheduling operation process model, extract the parameter change characteristics of the standard operation steps, establish the corresponding relationship between the operation type and the parameter change mode, and form a standard condition feature library; Perform operation monitoring, collect monitoring point data in real time, identify parameter out-of-limit situations, extract related measurement point data, match operating condition characteristics, and generate structured exception descriptions.

5. According to a digital twin-based power grid operation risk prediction and collaborative management system according to claim 1, it is characterized in that: The specific implementation method of the maintenance management submodule in the business control layer includes: Establish operating standards based on the maintenance operating procedure model, convert operating procedures into standard operating items, define the execution and acceptance rules of operating items, and establish the association between operating items and equipment status; Perform maintenance management, track maintenance work execution, record operation process data, associate equipment operating status, and generate standardized maintenance records; Data correlation analysis, establish the time series correlation between maintenance work and equipment status, analyze the parameter changes before and after maintenance, and extract the maintenance effect characteristics.

6. According to a digital twin-based power grid operation risk prediction and collaborative management system according to claim 1, it is characterized in that: The intelligent analysis layer specifically includes: Construct the observation variables and state variables of the improved Bayesian network, take the equipment abnormality description and operating condition characteristics output by the operation monitoring submodule as the network observation variable X, take the operation record and equipment status output by the maintenance management submodule as the network state variable Y, and determine the connection relationship E between network nodes based on the equipment topology relationship matrix T in the digital twin model.

7. A digital twin-based power grid operation risk prediction and collaborative management system according to claim 6, characterized in that: The specific implementation of the improved Bayesian network includes: Based on the network observation variables, state variables and connection relationships, a dynamic Bayesian network structure B(V,E) containing T time segments is constructed, where V={X∪Y} represents the network node set, each node contains state variables at time t and t-1, and E represents the temporal transfer relationship between nodes; A node conditional probability table (CPT) is constructed based on the equipment-associated fault records, and the CPT parameters are iteratively optimized using the EM algorithm. The EM algorithm calculates the conditional probability of the unobserved state through the E step, and maximizes the expected likelihood estimation probability parameters of the complete data through the M step. The optimal parameters are obtained through iterative optimization. The optimization process of the EM algorithm is as follows: Step E, estimate the conditional probability distribution of unobserved data based on the current parameter θ: P(Y|D,CPT) Wherein D is the observation data, i.e., the equipment abnormality description and operating condition characteristics output by the operation monitoring submodule, and Y is the state variable, i.e., the operation record and equipment status output by the maintenance management submodule; In the M step, the expected likelihood function of the complete data is maximized to obtain new parameter estimates: CPT * =argmaxP(D|CPT) Among them, CPT * is the optimal conditional probability estimate, P(D|CPT) is the likelihood probability of observing data D under the current conditional probability; The probability transfer matrix is ​​modified by the device importance weight, and the formula is: P′(Y|X)=W×P(Y|X) Among them, P′(Y|X) is the corrected conditional probability, P(Y|X) is the original conditional probability, X is the observed variable, that is, the device abnormality description and operating condition characteristics output by the operation monitoring submodule, W is the device importance weight matrix, which is determined based on the voltage level in the basic attributes of the device, the connection relationship in the topological relationship matrix, and the operation attributes; for the variable X, the probability distribution of the network state variable is calculated using the MAP criterion, and the formula is: P(Y|X)=argmaxP(X|Y)P(Y) Among them, P(X|Y) is the likelihood probability (the dependence of the observed variable on the state variable), and P(Y) is the prior probability of the state variable; The forward algorithm is used to calculate the risk probability at time t+k, and the formula is: P(Y t+k |X 1:t )=∑P(Y t+k |And t )P(Y t |X 1:t ) Among them, Y t+k is the state at time t+k, X 1:t is the observation sequence from time 1 to time t, P(Y t+k |Y t ) is the state transition probability, P(Y t |X 1:t ) is the state posterior probability at time t; The risk propagation path is calculated based on the conditional risk propagation probability matrix, and the formula is: P=P(Y|X)×R Wherein, R is the risk association matrix, which describes the association relationship between the devices based on the device topology relationship matrix T.

8. The digital twin-based power grid operation risk prediction and collaborative management system according to claim 1 is characterized in that: The specific implementation of the emergency response submodule of the collaborative governance layer includes: Based on the risk probability distribution and propagation path output by the intelligent analysis layer, a hierarchical response mechanism is established to map the risk probability P(Y|X) to four levels of response. The key equipment within the affected range is determined according to the risk propagation probability matrix P, and the risk level and key equipment information are pushed to the mobile application terminal of the relevant personnel.

9. The power grid operation risk prediction and collaborative management system based on digital twin according to claim 1 is characterized in that: The system also includes a mobile application terminal: The mobile application terminal interacts with the system in real time through the 4G / 5G network, receives the risk level and key equipment information output by the intelligent analysis layer, displays the disposal instructions issued by the collaborative governance layer, and transmits the on-site disposal situation to the system in real time, including equipment status changes and disposal progress, to cooperate with the collaborative governance layer to form a closed-loop management.

Images