Voltage stability risk dynamic assessment method and system based on data driving

Abstract

The invention provides a voltage stability risk dynamic assessment method and system based on data driving, and belongs to the technical field of voltage stability risk assessment. Constructing a power grid state vector based on the preprocessed power grid multi-source operation data; learning a causal dependency relationship among parameters in the power grid state vector by using a causal discovery algorithm, and generating a dynamic causal graph representing an internal causal structure of current power grid operation; performing risk traceability analysis based on a dynamic causal graph, and performing causal effect deduction on regulation and control strategies associated with the key nodes by using a causal intervention model; and generating a voltage stability risk dynamic assessment report according to a risk traceability analysis result and a causal effect deduction result. According to the invention, by constructing the dynamic causal graph, real-time and accurate traceability and quantitative evaluation of the voltage stability risk are realized, and causal effect deduction can be carried out on the regulation and control strategy, so that the intervention effect is predicted in advance.

Description

Technical Field

[0001] This invention relates to the field of voltage stability risk assessment technology, specifically to a data-driven dynamic assessment method and system for voltage stability risk. Background Technology

[0002] Voltage stability is the core foundation for the safe and stable operation of a power system. With the increasing proportion of renewable energy, the growing complexity of load characteristics, and the increasing variability of power grid operation modes, voltage stability issues are becoming increasingly prominent, and their assessment and prevention are extremely challenging.

[0003] Traditional voltage stability assessment methods primarily rely on physical model-based simulation calculations, such as continuous power flow methods and eigenvalue analysis. While these methods have a rigorous theoretical foundation, they have significant limitations in practical applications. Their assessment results heavily depend on accurate power grid models and parameters, and inaccurate equipment parameters and lagging model updates are unavoidable in real-world systems, leading to biased results. Secondly, simulation-based methods are computationally intensive and time-consuming, making it difficult to meet the real-time requirements of modern large-scale power grids for online and dynamic risk sensing. These methods typically only provide a margin of the power system's current distance from stability limits, representing a static and lagging assessment that cannot dynamically track risk evolution trends, hindering dispatchers from providing accurate early warnings and interventions. Summary of the Invention

[0004] The purpose of this invention is to provide a data-driven dynamic assessment method and system for voltage stability risks. This system can automatically identify the root causes of voltage instability risks from real-time data by constructing a dynamic causal graph, and quantitatively deduce the effects of different control measures, thereby providing accurate and interpretable decision support for power grid dispatchers.

[0005] To achieve the above objectives, this invention provides a data-driven dynamic assessment method for voltage stability risk, comprising: collecting multi-source operation data of the power grid and preprocessing the multi-source operation data of the power grid; constructing a power grid state vector based on the preprocessed multi-source operation data of the power grid; based on the power grid state vector, using a causal discovery algorithm to learn the causal dependencies between parameters in the power grid state vector, generating a dynamic causal graph characterizing the inherent causal structure of the current power grid operation; determining a voltage stability risk index according to the power grid state vector; when the voltage stability risk index exceeds a preset risk index threshold, performing risk tracing analysis based on the dynamic causal graph, calculating the contribution of each parameter in the power grid state vector to voltage stability risk, and locating nodes whose contribution exceeds a preset contribution threshold as key nodes; based on the dynamic causal graph, using a causal intervention model to perform causal effect deduction on the control strategies associated with the key nodes; and generating a voltage stability risk dynamic assessment report based on the results of the risk tracing analysis and the results of the causal effect deduction to complete the dynamic assessment of voltage stability risk.

[0006] Optionally, constructing the power grid state vector based on the preprocessed multi-source power grid operation data includes: extracting node electrical measurement features from the preprocessed multi-source power grid operation data, the node electrical measurement features including node voltage amplitude, node voltage phase angle, generator active power output, generator reactive power output, load active power, and load reactive power; extracting line power flow features from the preprocessed multi-source power grid operation data, the line power flow features including line active power, line reactive power, and line current amplitude; and concatenating the node electrical measurement features and the line power flow features in a preset order to obtain the power grid state vector.

[0007] Optionally, the step of learning the causal dependencies between parameters in the power grid state vector using a causal discovery algorithm includes: performing a stationarity test on the power grid state vector in historical time series form, and performing cointegration processing on power grid state vectors determined to be non-stationary after the test; using a constraint-based causal discovery algorithm, gradually eliminating redundant edges in the causal network through conditional independence tests to construct an initial causal skeleton graph; the conditional independence test uses the partial correlation coefficient test method or the kernel function-based conditional independence test method; based on the initial causal skeleton graph, using collision detection rules to determine the causal direction of some edges, and using the directed propagation rule to deduce the causal direction of edges applicable to the rule; for edges whose direction cannot be determined by the collision detection rule and the directed propagation rule, they are retained as undirected connections to construct a partially directed acyclic graph; using a score-based greedy equivalence search algorithm to optimize the structure of the partially directed acyclic graph, evaluating the fitting degree between different directed graph structures and the power grid state vector through a preset scoring function, determining the optimal causal direction of each undetermined direction edge in the partially directed acyclic graph, and obtaining the causal dependencies between power grid state parameters.

[0008] Optionally, generating a dynamic causal graph representing the inherent causal structure of the current power grid operation includes: configuring a time-window-based causal graph update mechanism, using a fixed-length sliding window or an adaptive variable-length window to extract a power grid state vector sequence within a specific time length before the current moment; recalculating the weight coefficients of each edge in the dynamic causal graph based on the newly extracted power grid state vector sequence in each update cycle; setting a causal structure stability monitoring mechanism, triggering a complete re-execution of the causal discovery algorithm when the number of edges or the change in the weight coefficients of edges in the dynamic causal graph exceeds a preset threshold within multiple consecutive time windows; and encoding the updated causal dependencies into a dynamic causal graph in the form of a directed acyclic graph.

[0009] Optionally, determining the voltage stability risk index based on the power grid state vector includes: inputting the power grid state vector into a pre-trained voltage stability assessment model, wherein the voltage stability assessment model outputs a numerical index characterizing the degree of voltage stability, and using this numerical index as the voltage stability risk index; the voltage stability assessment model is constructed based on a neural network or gradient boosting decision tree and is trained using historical power grid operation data.

[0010] Optionally, the step of performing risk tracing analysis based on the dynamic causal graph and calculating the contribution of each parameter in the power grid state vector to voltage stability risk includes: selecting at least one state variable related to voltage stability as the target variable for tracing analysis from the power grid state vector; identifying all directed causal paths in the dynamic causal graph that terminate at the node of the target variable; calculating the contribution of each parameter in the power grid state vector to the target variable through the identified causal path; associating the contribution of each parameter to the target variable with the contribution of each parameter to the voltage stability risk index, and locating key nodes based on this contribution.

[0011] Optionally, the step of performing risk source analysis based on the dynamic causal graph and calculating the contribution of each parameter in the power grid state vector to voltage stability risk further includes: calculating the path influence intensity of each identified directed causal path; for each parameter in the power grid state vector, arithmetically summing the path influence intensities of all directed causal paths to which it belongs to obtain the total contribution of the parameter; and normalizing the total contribution of all parameters to obtain the contribution of each parameter to voltage stability risk.

[0012] Optionally, the step of using a causal intervention model to extrapolate the causal effects of control strategies associated with the key node includes: selecting a set of control strategies associated with the key node, wherein each control strategy is defined as a specific intervention on one or more control variables in the power grid; constructing a structured causal intervention model based on the causal structure defined by the dynamic causal graph; and performing intervention simulation for each control strategy in the structured causal intervention model to extrapolate the evolution path of the power grid state vector and the voltage stability risk index after the implementation of the control strategy.

[0013] Optionally, the deduction of the evolution path of the power grid state vector and the voltage stability risk index after the implementation of the control strategy includes: for each control strategy, performing an intervention in the structured causal intervention model, calculating the difference between the voltage stability risk index after the intervention and the current voltage stability risk index, and quantifying the difference as the causal effect value of the control strategy.

[0014] On the other hand, the present invention provides a data-driven dynamic voltage stability risk assessment system for implementing a data-driven dynamic voltage stability risk assessment method. The system includes a control module, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program to implement the data-driven dynamic voltage stability risk assessment method.

[0015] The aforementioned technical solution, by constructing a dynamic causal graph, enables real-time, accurate source tracing and quantitative assessment of voltage stability risks. It not only dynamically identifies key nodes leading to risks but also extrapolates the causal effects of control strategies, thereby predicting intervention outcomes in advance. Ultimately, it generates a decision report that combines risk location and strategy optimization, significantly improving the accuracy, response speed, and intelligence level of grid security early warning and dispatching decisions, providing reliable technical support for proactive grid defense under high-proportion renewable energy integration.

[0016] Other features and advantages of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the following detailed description to explain the invention, but do not constitute a limitation thereof. In the drawings:

[0018] Figure 1 This is a flowchart of a data-driven dynamic assessment process for voltage stability risks.

[0019] Figure 2 This is a flowchart for identifying key nodes in the power grid. Detailed Implementation

[0020] The following is in conjunction with the appendix Figure 1 -Appendix Figure 2 The specific implementation methods of the embodiments of the present invention will be described in detail below. It should be understood that the specific implementation methods described herein are only for illustrating and explaining the embodiments of the present invention, and are not intended to limit the embodiments of the present invention.

[0021] It should be noted that the acquisition, transmission, storage, use, and processing of data in the technical solution of this application all comply with the relevant provisions of national laws and regulations. In the embodiments of this application, certain existing industry solutions such as software, components, and models may be mentioned. These should be considered exemplary, intended only to illustrate the feasibility of implementing the technical solution of this application, and do not imply that the applicant has already used or necessarily used such solutions.

[0022] In the process of realizing this invention, the inventors of this application discovered that the existing technology using simulation methods involves large amounts of computation and is time-consuming, making it difficult to meet the real-time requirements of modern large power grids for online and dynamic risk perception.

[0023] Example 1

[0024] Reference Figures 1-2 This is the first embodiment of the present invention, which provides a data-driven dynamic assessment method for voltage stability risk, including:

[0025] S100: Collect multi-source operation data of the power grid, preprocess the multi-source operation data of the power grid, and construct the power grid state vector based on the preprocessed multi-source operation data of the power grid.

[0026] In a preferred embodiment of this application, multi-source power grid operation data is collected. The multi-source power grid operation data includes real-time power grid operation data and historical operation data. Real-time operation data includes, but is not limited to, node voltage, line power, generator output and load power from the data acquisition and monitoring system, as well as voltage phasors, current phasors and system frequency from the synchronous phasor measurement device. Historical operation data includes, but is not limited to, time-series data from the same source as the real-time operation data within a specific historical period.

[0027] Furthermore, the collected multi-source operational data is preprocessed, including data cleaning, outlier removal, multi-source timescale alignment, and data normalization.

[0028] Furthermore, constructing a power grid state vector based on preprocessed multi-source power grid operation data includes: extracting node electrical measurement features from the preprocessed multi-source power grid operation data according to the rule mapping relationship between equipment identifiers and measurement points defined in the power grid model file. These node electrical measurement features include node voltage amplitude, node voltage phase angle, generator active power output, generator reactive power output, load active power, and load reactive power. Based on the rule mapping relationship, extracting line power flow features from the preprocessed multi-source power grid operation data. These line power flow features include line active power, line reactive power, and line current amplitude. The node electrical measurement features and line power flow features are then concatenated in a preset order. First, a fixed sorting rule is defined for all nodes and lines based on the power grid topology. Then, according to this sorting rule, all parameters in the node electrical measurement features are arranged sequentially according to node order, forming the first part of the state vector. Next, all parameters in the line power flow features are arranged sequentially according to line order, following the first part, forming the second part of the state vector. Finally, the first part and the second part are connected sequentially to form a power grid state vector with a unified structure and fixed dimensions, thus obtaining the power grid state vector.

[0029] The above scheme constructs a power grid state vector with fixed dimensions and a unified structure, which can comprehensively and structurally represent the operating state of the power grid at any time, providing an accurate and consistent data foundation for all subsequent analyses.

[0030] S200: Based on the power grid state vector, the causal discovery algorithm is used to learn the causal dependencies between parameters in the power grid state vector and generate a dynamic causal graph that represents the internal causal structure of the current power grid operation.

[0031] In the embodiments of this application, a causal discovery algorithm is used to learn the causal dependencies between parameters in the power grid state vector. This includes: performing a stationarity test on the power grid state vector in historical time series form, and cointegrating the power grid state vectors determined to be non-stationary after the test; employing a constraint-based causal discovery algorithm to gradually eliminate redundant edges in the causal network through conditional independence tests to construct an initial causal skeleton graph; the conditional independence test uses either the partial correlation coefficient test or a kernel function-based conditional independence test; based on the initial causal skeleton graph, collision detection rules are used to determine the causal directions of some edges, and causal directions are derived for edges applicable to the directed propagation rule; edges whose directions cannot be determined by collision detection and directed propagation rules are retained as undirected connections, thus constructing a partially directed acyclic graph; a score-based greedy equivalence search algorithm is used to optimize the structure of the partially directed acyclic graph, and a preset scoring function is used to evaluate the fit between different directed graph structures and the power grid state vector, determining the optimal causal direction of each edge in the partially directed acyclic graph to be determined, thereby obtaining the causal dependencies between the power grid state parameters.

[0032] In a preferred embodiment of this application, the power grid state vector in historical time series form is divided into a training set and a validation set according to the time dimension. The training set accounts for 70%-80% and is used for training the causal discovery model, while the validation set accounts for 20%-30% and is used to verify the rationality of the causal structure. During the division process, care is taken to maintain temporal continuity to avoid distortion of causal relationships caused by data disruption. A dual validation mechanism of Augmented Dickey-Fuller Test (ADF) and Phillips-Perron Test (PP) is used for stationarity testing. The ADF is used to detect whether the sequence has a unit root (the null hypothesis is that a unit root exists, i.e., the sequence is non-stationary). The null hypothesis is rejected when the test statistic is less than a critical value (such as the critical value at the 1%, 5%, or 10% significance level). The Phillips-Perron Test is used to handle sequences with heteroscedasticity, compensating for the limitation of the ADF in its assumptions about residual terms.

[0033] It is important to note that during the verification process, verification should be performed separately for different parameters in the power grid state vector (such as node voltage amplitude, line active power, etc.) to avoid overall misjudgment due to differences in the stability of some parameters.

[0034] For power grid state vector parameters that are determined to be non-stationary by both double tests, a preliminary processing method (first-order or second-order difference) is first used. If the differencing sequence is still non-stationary, cointegration processing is performed. Cointegration processing uses the Johansen cointegration test to identify long-term equilibrium relationships among multiple non-stationary sequences and determine the number of cointegrating vectors. For parameter sets with cointegration relationships, an error correction model (ECM) is constructed to transform the non-stationary sequence into a stationary error correction term, while preserving long-term causal association information between sequences. After processing, the stationarity test is re-performed on the transformed sequence to ensure that the input requirements of the causal discovery algorithm are met.

[0035] The core of the causal discovery algorithm uses the Peter-Clark algorithm (PC algorithm), with the initial adjacency matrix set to fully connected, and the significance level of the conditional independence test. We set the value to 0.05 (a standard adaptation value for power grid data), and the maximum condition set size to min(4, n-2) (where n is the parameter dimension). For each pair of parameters (X, Y) in the fully connected adjacency matrix, we perform conditional independence tests under different condition sets S (S being a subset of parameters other than X and Y). If we use the partial correlation coefficient test, we calculate the partial correlation coefficient r between X and Y given S. If the absolute value of r is less than the corresponding significance level... If the threshold is reached, X and Y are determined to be conditionally independent given S, and the connecting edges between X and Y in the adjacency matrix are deleted. Alternatively, if a kernel-based conditional independence test, such as the KCIT algorithm (kernel-based conditional independence test), is used, for parameter pairs with nonlinear associations, the data is mapped to a high-dimensional feature space using a kernel function (such as a Gaussian kernel or a polynomial kernel). The conditional mutual information of the kernel matrix is ​​calculated. If the conditional mutual information is less than a threshold, the condition is determined to be independent, and the connecting edges are deleted. During the test, the condition set size is incremented from 0 until all possible condition sets have been tested, resulting in an initial causal skeleton graph (only connecting edges between parameters with statistical associations are retained).

[0036] In a preferred embodiment of this application, determining the causal direction of some edges using collision detection rules and deriving the causal direction of edges applicable to the rules based on directional propagation rules includes: traversing all triple structures (X, Y, Z) in the initial causal skeleton graph, where X and Y, Y and Z have connected edges, and X and Z have no connected edges. If there exists another parameter W such that X and Z are conditionally independent given W, but conditionally related given Y, then Y is determined to be a collision point, and the edge is oriented as X→Y←Z. If there are edges in the triple with known causal directions (such as the generator output → node voltage amplitude correlation that can be determined according to the physical laws of the power grid), then the orientation is preferentially based on prior knowledge, and then the direction of other edges is optimized by combining the collision detection rules. For example, if it is known that there is a causal relationship between generator active power output (G) and node voltage amplitude (U) G→U, and if the triple is G, U, L (L is the load active power), and G and L are not directly connected, then the direction L→U or U→L can be further determined through collision detection.

[0037] Based on the already directed edges, execute the directed propagation rules:

[0038] 1. If there exists an edge X→Y and there is an undirected edge between Y and Z, and there is no connecting edge between X and Z, then orient YZ as Y→Z.

[0039] 2. If there exist edges X→Y and Y→Z, then if there is an undirected edge between X and Z, its orientation is X→Z.

[0040] 3. If there exist edges XY and Y→Z, and X and Z are not connected by any edge, then orient XY as X→Y.

[0041] It is important to note that during the propagation process, directed cycles in the graph need to be checked in real time. If a cycle structure is found (such as X→Y→Z→X), the direction of the previous edge should be adjusted back to ensure the graph is acyclic.

[0042] In a preferred embodiment of this application, edges whose direction cannot be determined by collision detection rules and directional propagation rules are marked as "edges with directions to be determined," and their statistical indicators such as conditional correlation coefficients and mutual information under different condition sets are recorded to provide a basis for subsequent structural optimization. Simultaneously, based on knowledge of power grid physical topology, undirected edges are initially screened. For example, edges between line power flow parameters and the voltage parameters of the two end nodes can have their potential directions initially determined based on circuit theory, serving as prior information for subsequent optimization.

[0043] The preset scoring function is as follows:

[0044]

[0045] in, The goodness-of-fit score is represented by G, which represents the candidate directed acyclic graph, and D represents the power grid state vector dataset. It is the log-likelihood of the data under graph G. Represents the regularization parameter. Representing graph complexity, E represents the number of directed edges in graph G.

[0046] In a preferred embodiment of this application, a partial directed acyclic graph (DAG) is first used as the initial structure, and the undirected edges are defined as the set of edges to be oriented, constructing a search space containing all possible direction combinations. Next, a score-based search algorithm is used to iteratively optimize the search space, including generating a set of neighborhood graphs for the current structure, calculating the fitting score of each neighborhood graph using a preset scoring function, selecting the neighborhood graph with the higher score to update the current structure, until the iteration termination condition is met. Finally, the directions of each edge in the optimized DAG are determined as the optimal causal directions of the edges to be oriented, thereby obtaining the complete causal dependencies between power grid state parameters.

[0047] In the embodiments of this application, generating a dynamic causal graph representing the inherent causal structure of the current power grid operation includes: configuring a causal graph update mechanism based on a time window, using a fixed-length sliding window or an adaptive variable-time window to extract a power grid state vector sequence within a specific time length before the current moment; recalculating the weight coefficients of each edge in the dynamic causal graph based on the newly extracted power grid state vector sequence in each update cycle; setting a causal structure stability monitoring mechanism, triggering a complete re-execution of the causal discovery algorithm when the number of edges or the change in the weight coefficients of edges in the dynamic causal graph exceeds a preset threshold within multiple consecutive time windows; and encoding the updated causal dependencies into a dynamic causal graph in the form of a directed acyclic graph.

[0048] In a preferred embodiment of this application, to ensure the consistency between the voltage stability risk assessment model and the dynamic causal graph when the operating mode changes, the present invention sets up a model consistency guarantee mechanism. When the change in the causal structure or edge weights in the dynamic causal graph over time exceeds a preset threshold, the online update or periodic incremental training of the voltage stability assessment model is automatically triggered to ensure that the discrimination criteria of the risk assessment model match the current causal structure, thereby guaranteeing the consistency between risk source tracing analysis and risk index based on the dynamic causal graph.

[0049] The aforementioned scheme utilizes causal discovery algorithms to automatically learn the causal dependencies between power grid state parameters from data, thus surpassing traditional correlation analysis and revealing the intrinsic mechanisms and transmission paths of risks. The generated dynamic causal graph not only possesses strong interpretability and conforms to the physical laws of the power grid, but also continuously updates through a sliding window mechanism, adapting to changes in power grid operation. This ensures that the causal structure of the power grid upon which the risk assessment model is based remains up-to-date and accurate, providing a scientific basis for precise risk tracing.

[0050] S300: Determine the voltage stability risk index based on the power grid state vector. When the voltage stability risk index exceeds the preset risk index threshold, perform risk source analysis based on the dynamic cause-effect graph, calculate the contribution of each parameter in the power grid state vector to the voltage stability risk, and locate the nodes whose contribution exceeds the preset contribution threshold as key nodes.

[0051] In the embodiments of this application, determining the voltage stability risk index based on the power grid state vector includes: inputting the power grid state vector into a pre-trained voltage stability assessment model, the voltage stability assessment model outputting a numerical index characterizing the degree of voltage stability, and using this numerical index as the voltage stability risk index; the voltage stability assessment model is constructed based on a neural network or gradient boosting decision tree and trained using historical power grid operation data.

[0052] In a preferred embodiment of this application, the voltage stability assessment model constructs a training sample set by collecting historical power grid operation data. The input features of the sample set are historical power grid state vectors containing node voltage, line power, and generator output parameters. Simultaneously, offline simulation calculations (using continuous power flow or optimal power flow methods) generate voltage stability margin labels representing the distance from the operating point to the voltage collapse point for each historical power grid state vector. The state vectors and corresponding labels are then combined to form a complete training sample set. Next, model structure selection and training are performed. A neural network or gradient boosting decision tree is selected as the model structure. The training sample set is divided into a training set and a test set according to time sequence. The training set is used to minimize the error between the model's predicted values ​​and the simulation labels using an optimization algorithm, and the test set is used to evaluate the model's generalization ability. Finally, the trained model is saved as a pre-trained voltage stability assessment model. Preferably, the voltage stability margin label can be any one of the load margin index or local voltage stability index.

[0053] In the embodiments of this application, risk tracing analysis is performed based on a dynamic causal graph, calculating the contribution of each parameter in the power grid state vector to voltage stability risk. This includes: selecting at least one state variable strongly correlated with voltage stability as the target variable for tracing analysis from the power grid state vector, wherein the selected target variable and the voltage stability risk index are mapped through predefined association rules; identifying all directed causal paths in the dynamic causal graph that terminate at the node of the target variable; calculating the contribution of each parameter in the power grid state vector to the target variable through the identified causal path; and, based on the association rules, associating the contribution of each parameter to the target variable as the contribution of each parameter to the voltage stability risk index, and locating key nodes based on this contribution. For example, based on the contribution of each parameter to the risk index obtained after mapping, the power grid nodes corresponding to parameters whose contribution exceeds a preset threshold (contribution greater than 0.1 or ranked in the top 5) are located as key nodes causing voltage stability risk.

[0054] Preferably, for each identified directed causal path, its path influence strength is calculated by multiplying the standardized regression coefficients of each causal edge along the path. For each parameter in the power grid state vector, the path influence strengths of all directed causal paths to which it belongs are arithmetically summed to obtain the total contribution of that parameter. The total contribution of all parameters is then normalized to obtain the contribution of each parameter to voltage stability risk. The association rule is as follows: the target variable is a key input feature for constructing the voltage stability assessment model, and its strong causal association with the voltage stability risk index is confirmed through causal discovery or statistical analysis (such as calculating the Pearson correlation coefficient or the magnitude of the causal effect based on the causal graph). For example, its causal relationship strength or correlation coefficient exceeds a preset threshold (such as 0.7). This rule ensures that the source tracing analysis results for the target variable can effectively represent the source tracing of the overall risk index.

[0055] The above-described scheme utilizes pre-trained models to replace complex simulations, enabling online, rapid, and quantitative assessment of voltage stability risks and timely early warning. Furthermore, based on dynamic cause-effect graphs for risk tracing, it can accurately calculate the contribution of each node's parameters to the overall risk, thereby precisely identifying the key source nodes causing the risk. This provides dispatchers with clear objectives for implementing targeted control measures, significantly improving dispatching efficiency.

[0056] S400: Based on dynamic causal graphs, it uses causal intervention models to extrapolate the causal effects of regulatory strategies associated with key nodes.

[0057] In the embodiments of this application, a causal intervention model is used to extrapolate the causal effects of control strategies associated with critical nodes, including: selecting a set of control strategies associated with critical nodes, wherein each control strategy is defined as a specific intervention on one or more control variables in the power grid; constructing a structured causal intervention model based on the causal structure defined by the dynamic causal graph; and performing intervention simulation for each control strategy in the structured causal intervention model to extrapolate the evolution path of the power grid state vector and the voltage stability risk index after the implementation of the control strategy.

[0058] In the embodiments of this application, the deduction of the evolution path of the power grid state vector and the voltage stability risk index after the implementation of the control strategy includes: for each control strategy, performing intervention in a structured causal intervention model, calculating the difference between the voltage stability risk index after intervention and the current voltage stability risk index, and quantifying the difference as the causal effect value of the control strategy.

[0059] In a preferred embodiment of this application, based on the power grid topology and the causal paths revealed by the dynamic causal graph, control strategies that can directly or indirectly affect the electrical state of key nodes are selected from a pre-set control strategy library, forming a set of control strategies associated with key nodes. The control strategy library is pre-established based on power grid dispatching procedures and expert experience, and each strategy includes at least the control object, control type, and control value range. Based on the dynamic causal graph, a structural equation corresponding to its parent node is established for each node in the causal graph, and the parameters of each structural equation are estimated using historical power grid operation data to form a structured causal intervention model. In the causal intervention model, for each selected control strategy, the values ​​of the control variable nodes corresponding to the strategy are forcibly set to the pre-set intervention values. The posterior distribution of other nodes is solved recursively through the structural equations to obtain the system state evolution process under the intervention. Based on the posterior distribution of each node after intervention, the evolution path of the power grid state vector within a pre-set time range is deduced using time-series recursion or Monte Carlo sampling methods, and the voltage stability risk index of each time segment is calculated using a voltage stability assessment model. For each control strategy, the difference between the voltage stability risk index at steady state or at a specified time after intervention and the current risk index is calculated, and this difference is used as the causal effect value of the control strategy. A negative causal effect indicates a decrease in risk, a positive value indicates an increase in risk, and the absolute value represents the intensity of the effect.

[0060] The above-described scheme transforms dispatching decisions from passive response to proactive intervention. By simulating the effects of different control strategies in a model built on causal graphs, the causal effects of these strategies on voltage stability risks can be quantified and extrapolated. This helps dispatchers select the optimal solution from multiple alternatives, thereby improving the scientific rigor and safety of decision-making.

[0061] S500: Based on the results of risk source analysis and causal effect deduction, generate a voltage stability risk dynamic assessment report to complete the voltage stability risk dynamic assessment.

[0062] In a preferred embodiment of this application, firstly, the key nodes and their contribution ranking obtained from risk tracing analysis, the causal effect values ​​of each control strategy obtained from causal effect deduction, the current voltage stability risk index, and the recent risk index sequence obtained from historical database retrieval are integrated to form a complete assessment dataset. Then, based on this integrated assessment dataset, an automated document generation method is used to generate a structured risk assessment report. This report covers an overview of the risk situation (including the current risk level, index value, and development trend), the results of risk tracing analysis (including a list of key nodes and main propagation paths), and the results of control strategy deduction (including a list of recommended strategies and their expected effects and priorities). Based on this, targeted decision-making suggestions are automatically generated by combining the risk level and core data, recommending differentiated control measures and associating the key nodes with the highest contribution with the most effective control strategies. Finally, the report is formatted according to a standard template, supplemented with basic information such as timestamps, and simultaneously generated visual charts that can be directly displayed on a human-machine interface, as well as machine-readable structured data, completing the dynamic assessment of voltage stability risk.

[0063] The present invention also provides a data-driven voltage stability risk dynamic assessment system for implementing a data-driven voltage stability risk dynamic assessment method. The system includes a control module, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program to implement the data-driven voltage stability risk dynamic assessment method.

[0064] This invention provides a storage medium storing a program that, when executed by a processor, implements a data-driven dynamic assessment method for voltage stability risks.

[0065] This invention provides a processor for running a program, wherein the program executes the data-driven dynamic assessment method for voltage stability risk.

[0066] This invention provides a device including a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements a data-driven dynamic assessment method for voltage stability risks. The device described herein can be a server, PC, PAD, mobile phone, etc.

[0067] This application also provides a computer program product that, when executed on a data processing device, is suitable for performing a data-driven dynamic assessment method for voltage stability risks.

[0068] Those skilled in the art will understand that embodiments of this application can provide methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0069] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0070] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0071] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0072] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0073] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0074] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0075] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0076] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A data-driven based dynamic voltage stability risk assessment method, characterized in that, The method comprises the following steps: Collecting power grid multi-source operation data, preprocessing the power grid multi-source operation data, and constructing a power grid state vector based on the preprocessed power grid multi-source operation data; Based on the power grid state vector, the causal dependence relationship between each parameter in the power grid state vector is learned by using a causal discovery algorithm, and a dynamic causal graph representing the internal causal structure of the current power grid operation is generated; According to the power grid state vector, the voltage stability risk index is determined, when the voltage stability risk index exceeds the preset risk index threshold, the risk traceability analysis is carried out based on the dynamic causal graph, the contribution degree of each parameter in the power grid state vector to the voltage stability risk is calculated, and the node whose contribution degree exceeds the preset contribution degree threshold is located as the key node; Based on the dynamic causal graph, the causal effect of the control strategy associated with the key node is deduced by using the causal intervention model; According to the results of the risk traceability analysis and the results of the causal effect deduction, a voltage stability risk dynamic evaluation report is generated to complete the voltage stability risk dynamic evaluation.

2. The data-driven based voltage stability risk dynamic assessment method according to claim 1, wherein, The method comprises the following steps: Extracting node electrical quantity measurement features from the preprocessed power grid multi-source operation data, the node electrical quantity measurement features including node voltage amplitude, node voltage phase angle, generator active power, generator reactive power, load active power and load reactive power; Extracting line flow characteristics from the preprocessed power grid multi-source operation data, the line flow characteristics including line active power, line reactive power and line current amplitude; The node electrical quantity measurement features and the line flow characteristics are spliced in a preset order to obtain a power grid state vector.

3. The data-driven based voltage stability risk dynamic assessment method of claim 1, wherein, The method comprises the following steps: Performing stationarity test on the historical time series form of the power grid state vector, and performing cointegration processing on the power grid state vector determined as non-stationary after the test; Using a constraint-based causal discovery algorithm, the redundant edges in the causal network are gradually removed through conditional independence test to construct an initial causal skeleton graph; The conditional independence test uses partial correlation coefficient test method or kernel function based conditional independence test method; Based on the initial causal skeleton graph, the causal direction of some edges is determined by using collision detection rules, and the causal direction of the edges applicable to the rules is derived based on directional propagation rules. For the edges whose direction cannot be determined by the collision detection rules and the directional propagation rules, they are kept as undirected connections to construct a partial directed acyclic graph; Using a score-based greedy equivalence search algorithm, the structure of the partial directed acyclic graph is optimized, the fitting degree of different directed graph structures and the power grid state vector is evaluated by a preset scoring function, the optimal causal direction of each edge in the partial directed acyclic graph is determined, and the causal dependence relationship between the power grid state parameters is obtained.

4. The data-driven based voltage stability risk dynamic assessment method of claim 1, wherein, The method comprises the following steps: A time window-based causal graph updating mechanism is configured, which adopts a fixed-length sliding window or an adaptive variable-length time window to intercept a sequence of power grid state vectors within a specific time length before the current time; In each update cycle, the weight coefficients of each edge in the dynamic causal graph are recalculated based on the newly intercepted sequence of power grid state vectors; A causal structure stability monitoring mechanism is set up, which triggers a complete re-execution of the causal discovery algorithm when it is monitored that the number of edges or the change amplitude of the weight coefficients of the edges in the dynamic causal graph in consecutive multiple time windows exceeds a preset threshold; The updated causal dependency relationship is encoded into a dynamic causal graph in the form of a directed acyclic graph.

5. The data-driven based dynamic voltage stability risk assessment method of claim 1, wherein, The determination of the voltage stability risk index according to the power grid state vector comprises: The power grid state vector is input into a pre-trained voltage stability evaluation model, and the voltage stability evaluation model outputs a numerical indicator representing the degree of voltage stability, which is taken as the voltage stability risk index; The voltage stability evaluation model is constructed based on a neural network or a gradient boosting decision tree and is obtained by training historical power grid operation data.

6. The data-driven based dynamic voltage stability risk assessment method of claim 1, wherein, The risk tracing analysis based on the dynamic causal graph to calculate the contribution degree of each parameter in the power grid state vector to the voltage stability risk comprises: At least one state variable related to voltage stability is selected from the power grid state vector as a target variable for the tracing analysis; All directed causal paths in the dynamic causal graph with the node of the target variable as the terminal point are identified; The contribution degree of each parameter in the power grid state vector to the target variable through the identified causal paths is calculated; The contribution degree of each parameter to the target variable is associated with the contribution degree of each parameter to the voltage stability risk index, and the key node is located according to the contribution degree.

7. The data-driven based voltage stability risk dynamic assessment method of claim 6, wherein, The risk tracing analysis based on the dynamic causal graph to calculate the contribution degree of each parameter in the power grid state vector to the voltage stability risk further comprises: For each identified directed causal path, the path impact strength thereof is calculated; For each parameter in the power grid state vector, the path impact strengths of all directed causal paths to which the parameter belongs are arithmetically added to obtain the total contribution degree of the parameter; The total contribution degrees of all parameters are normalized to obtain the contribution degree of each parameter to the voltage stability risk.

8. The data-driven based dynamic voltage stability risk assessment method of claim 1, wherein, The causal effect deduction of the causal intervention model on the regulation strategy associated with the key node comprises: A group of regulation strategies associated with the key node are selected, wherein each regulation strategy is defined as a specific intervention on one or more control variables in the power grid; A structured causal intervention model is constructed based on the causal structure defined by the dynamic causal graph; In the structured causal intervention model, intervention simulation is performed on each regulation strategy to deduce the evolution path of the power grid state vector and the voltage stability risk index after the execution of the regulation strategy.

9. The data-driven based voltage stability risk dynamic assessment method of claim 8, wherein, The evolution path of the power grid state vector and the voltage stability risk index after the control strategy is executed, comprising: for each control strategy, executing intervention in the structured causal intervention model, calculating the difference between the voltage stability risk index after intervention and the current voltage stability risk index, and quantifying the difference as the causal effect value of the control strategy.

10. A data-driven based voltage stability risk dynamic assessment system, characterized in that, The system comprises a control module, the control module comprises a memory, a processor and a computer program stored on the memory and executable on the processor, and the processor executes the computer program to realize the data-driven voltage stability risk dynamic evaluation method according to any one of claims 1-9.

Images