A power system anomaly governance method and system based on causal inference

Abstract

The application discloses a power system anomaly treatment method and system based on causal inference, and relates to the technical field of power anomaly diagnosis. The method comprises the following steps: obtaining operation state data and power grid topology data of a target power grid area; based on the operation state data, calculating at least one abnormal index value representing power quality, and comparing the abnormal index value with a preset index threshold to determine whether there is an abnormal power quality; if it is determined that there is an abnormality, determining at least one candidate cause leading to the abnormality based on the operation state data and the power grid topology data; for each candidate cause, calculating a comprehensive scheduling cost required for intervention according to a predefined evaluation rule; selecting a cause subset from a preselected set constituted by the candidate causes; and outputting a treatment action suggestion corresponding to the cause subset. The application automatically selects a treatment measure combination with the optimal comprehensive scheduling cost and generates a treatment suggestion, thereby improving treatment efficiency and reducing resource invalid occupation.

Description

Technical Field

[0001] This invention relates to the field of power anomaly diagnosis technology, and in particular to a power system anomaly management method and system based on causal inference. Background Technology

[0002] With the accelerated construction of new power systems and the widespread integration of diverse loads such as distributed power sources and electric vehicles, the operating characteristics of distribution networks are becoming increasingly complex, highlighting power quality issues. Ensuring power supply reliability and power quality has become a core task for power grid companies to improve service levels and optimize operational efficiency.

[0003] Current power quality anomaly management in power systems largely relies on manual experience to determine the causes, which is prone to subjectivity, misjudgments, and omissions. While existing power quality monitoring systems automatically issue alarms for voltage, current, and other indicators by setting thresholds, enabling rapid problem detection, the solutions are often crude and lack automatic root cause analysis. After an anomaly alarm, maintenance personnel still need to combine their personal experience to investigate load characteristics, distributed generation output, equipment status, and other factors to deduce the cause. This process is time-consuming and labor-intensive, failing to meet the needs of efficient management in complex power grids. Summary of the Invention

[0004] To address the problem of low efficiency in governance and diagnosis caused by the crude formulation of existing power quality anomaly mitigation schemes in distribution networks and the inability to automatically analyze root causes, this invention provides a power system anomaly mitigation method and system based on causal inference. This method can automatically generate mitigation action suggestions, shorten the response and execution cycle of anomaly mitigation, and automatically select the optimal combination of mitigation measures with the best overall scheduling cost while ensuring that mitigation effects meet standards. This not only improves governance efficiency but also reduces ineffective resource occupation and enhances the efficiency of scheduling resource utilization. The specific technical solution is as follows: Firstly, this application provides a power system anomaly management method based on causal inference, comprising the following steps: Acquire operational status data and power grid topology data for the target power grid area; Based on the operating status data, at least one abnormal indicator value characterizing power quality is calculated and compared with a preset indicator threshold to determine whether there is a power quality abnormality. If an abnormal power quality is determined, at least one candidate cause for the abnormality is identified based on the operating status data and the power grid topology data. For each of the candidate causes, the comprehensive scheduling cost required to intervene is calculated according to predefined evaluation rules; wherein, the evaluation rules at least integrate the technical complexity level representing the difficulty of the intervention measures, the amount of resources used representing resource consumption, and the time period representing the time consumption. The candidate causes form a pre-selection set, and a subset of causes is automatically selected from the pre-selection set; wherein, the optimization objective of selecting the subset of causes is to minimize the sum of the comprehensive scheduling costs of all candidate causes in the subset of causes, and it is expected that after intervention on the subset of causes, the at least one abnormal indicator value can be restored to within the indicator threshold. Output governance action recommendations corresponding to the subset of causes.

[0005] Preferably, based on the operational status data and power grid topology data, at least one candidate cause leading to the anomaly is determined, specifically including: Based on the operating status data and power grid topology data, a directed graph model is constructed to represent the causal relationships between variables; wherein, when constructing the directed graph model, the model structure is constrained according to the electrical connection relationships represented by the power grid topology data; Based on the directed graph model, at least one candidate cause for power quality anomalies is identified.

[0006] Preferably, before constructing the directed graph model, the operating status data is preprocessed. The preprocessing includes: for time segments with missing or abnormal data, a collaborative repair algorithm based on the power grid topology and time-series correlation is used to complete and correct the data.

[0007] Preferably, the technical complexity level is a parameter used to classify the difficulty of implementing various governance measures based on predefined classification rules; the resource consumption is a parameter used to quantify the human, material, and equipment resources required for intervention based on historical data and a resource weight coefficient library.

[0008] Preferably, the resource weight coefficient library on which the resource occupancy calculation is based is synchronized in real time or periodically with the external operating environment data source to update the resource acquisition difficulty coefficient and the regional human resource occupancy equivalent data, and this update is dynamically mapped to the calculation result of the comprehensive scheduling cost.

[0009] Preferably, in the calculation of resource occupancy, for the same intervention measure, different resource weight coefficients or efficiency coefficients are associated according to different time periods or different urgency levels of its implementation, so as to calculate differentiated comprehensive scheduling costs.

[0010] Preferably, a subset of reasons is automatically selected from the pre-selected set, specifically including: Construct a combinatorial optimization problem, where decision variables represent whether to intervene in each candidate cause, and the objective function is set to minimize the total integrated scheduling cost of the intervened candidate causes; The constraints of the combinatorial optimization problem include: First constraint: The probability that the abnormal indicators return to normal after intervention is not less than a preset probability threshold, and the prediction of the preset probability threshold is based on the directed graph model; Second constraint: Total intervention resource consumption shall not exceed the acceptable resource consumption limit; The third constraint is that the implementation order of each intervention measure satisfies the causal temporal relationship derived from the directed graph model. Solve the combinatorial optimization problem to obtain the subset of causes.

[0011] Preferably, the combinatorial optimization problem is solved using a graph search algorithm, which, in the heuristic evaluation of the search path, simultaneously considers the cumulative scheduling cost of the selected causes and the prediction of the effectiveness of intervention on the unselected causes.

[0012] Preferably, the proposed governance action includes a list of specific maintenance, repair, or infrastructure measures for each cause in the subset of causes, as well as a resource consumption estimate based on the level of technical complexity and resource consumption.

[0013] Preferably, the output form of the governance action recommendations includes structured task instructions, in which each specific measure is automatically associated with best practice documents, standard operating procedure videos or key risk point prompts in its historical execution case library, and sent together to the execution terminal; The structured task instructions are directly parsed by the downstream execution terminal to create corresponding inspection, maintenance or construction work orders.

[0014] Secondly, this application also provides a power system anomaly management system based on causal inference, comprising: The data acquisition unit is used to acquire the operating status data and power grid topology data of the target power grid area; The first analysis unit is used to calculate at least one abnormal indicator value representing power quality based on the operating status data, and compare it with a preset indicator threshold to determine whether there is a power quality abnormality. The second analysis unit is used to determine at least one candidate cause of the abnormality based on the operating status data and the power grid topology data if it is determined that there is an abnormality in power quality. The computing unit is used to calculate the comprehensive scheduling cost required to intervene in each of the candidate causes, according to a predefined evaluation rule; wherein the evaluation rule at least integrates the technical complexity level representing the difficulty of the intervention measure, the resource consumption representing the resource consumption, and the time period representing the time consumption. A filtering unit is configured to construct a pre-selection set based on the candidate causes, and automatically select a subset of causes from the pre-selection set; wherein the optimization objective of selecting the subset of causes is to minimize the sum of the comprehensive scheduling costs of all candidate causes in the subset of causes, and it is expected that after intervention on the subset of causes, the at least one abnormal indicator value can be restored to within the indicator threshold. The output unit is used to output governance action recommendations corresponding to the subset of causes.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention provides a power system anomaly management method based on causal inference. It integrates technical complexity, resource consumption, and time cycle to construct a comprehensive scheduling cost evaluation system, providing a unified standard for scheme optimization. With the dual optimization objectives of minimizing scheduling costs and restoring anomaly indicators, it automatically selects a subset of causes, balancing the scheduling cost of governance with the anomaly elimination effect, avoiding excessive or ineffective intervention, and effectively reducing the ineffective use of power grid resources. It automatically generates governance action suggestions, replacing subjective manual decision-making, shortening the anomaly governance response and execution cycle, and improving power grid operation and maintenance efficiency. Simultaneously, based on causal inference, it accurately locates candidate causes of anomalies, establishing a logical connection between anomalies and their causes from the root, improving the accuracy of cause identification, and reducing the probability of misjudgment and omission. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0017] Figure 1 A flowchart of a power system anomaly management method based on causal inference provided in an embodiment of the present invention.

[0018] Figure 2 This is a schematic diagram of a power system anomaly management system based on causal inference, provided for an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] It should be understood that, when used in this specification, the terms “comprising” and “including” indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0021] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0022] It should also be further understood that the term "and / or" as used in this specification refers to any combination of one or more of the associated listed items, as well as all possible combinations, and includes such combinations.

[0023] Please refer to the following examples. Figure 1 and Figure 2 .

[0024] This application provides a power system anomaly management method based on causal inference, comprising the following steps: Step S1: Obtain the operating status data and power grid topology data of the target power grid area; Two types of core data are acquired synchronously from multiple data sources in the target power grid area. Operational status data originates from data acquisition and monitoring systems, advanced measurement systems, etc., collecting electrical measurement data in real time, including voltage, current, power, and power factor, and updating at minute or second-level frequencies. Power grid topology data comes from the power grid graphical management system and equipment asset management system, providing the connection relationships of equipment such as transformers, lines, switches, and distributed power sources, as well as key parameters such as line impedance and transformer capacity.

[0025] For example, by acquiring real-time operational status data node voltages from the target power grid area. Branch current Active power reactive power And so on, as well as power grid topology data, to map It means that, among them For a set of nodes, For the set of branches.

[0026] Step S2: Based on the operating status data, calculate at least one abnormal indicator value that characterizes power quality, and compare it with a preset indicator threshold to determine whether there is a power quality abnormality. The system performs real-time calculations on the acquired operational status data to generate a series of power quality indicators, including voltage-related indicators such as voltage deviation rate, voltage fluctuation and flicker, and voltage sag / boost; waveform distortion indicators such as total harmonic distortion rate; and frequency deviation indicators such as frequency deviation. For example, the most crucial indicator is the voltage deviation rate. The voltage deviation rate is calculated by comparing the real-time voltage value of each monitoring point with a pre-set or standard-specified rated voltage value to determine the percentage deviation. Furthermore, the system does not trigger an alarm based on a single instance of exceeding a limit; instead, it makes a comprehensive judgment based on preset judgment rules, such as if the voltage deviation exceeds a threshold ±7% for more than 80% of consecutive preset time intervals. This identifies persistent power quality anomalies requiring remediation and records the specific location and numerical value of the anomaly indicator.

[0027] By calculating key power quality indicators, such as voltage deviation rate : in, This is the rated voltage. The voltage is compared with a pre-set or standard-specified threshold. If the voltage exceeds the threshold for multiple consecutive sampling periods, it is determined to be a persistent voltage anomaly.

[0028] Step S3: If it is determined that there is an abnormal power quality, then based on the operating status data and the power grid topology data, determine at least one candidate cause of the abnormality. Based on the data integrated in step S1, causal discovery algorithms, such as the PC algorithm and the NOTEARS algorithm, are used to automatically analyze the potential causal relationship network among numerous variables, including voltage, current, load, and distributed power generation output. Power grid topology data is injected into the learning process as physical rules. Finally, the system generates a causal directed graph model under power grid topology constraints. The model shows which factors lead to anomalous indicators. Tracing back along the causal graph to the anomalous nodes, the system identifies several fundamental cause nodes with the largest contributions and lists them as candidate causes.

[0029] Specifically, based on the operational status data and power grid topology data, at least one candidate cause leading to the anomaly is identified, including: Based on the operating status data and power grid topology data, a directed graph model is constructed to represent the causal relationships between variables; wherein, when constructing the directed graph model, the model structure is constrained according to the electrical connection relationships represented by the power grid topology data; Based on the directed graph model, at least one candidate cause for power quality anomalies is identified.

[0030] Before constructing the directed graph model, the operating status data is preprocessed. The preprocessing includes: for time segments with missing or abnormal data, a collaborative repair algorithm based on the power grid topology and time-series correlation is used to complete and correct the data.

[0031] Assuming nodes Data missing, utilize its topological neighbor nodes Data repair: in, As weight, and branch impedance It is inversely proportional, reflecting the strength of electrical coupling.

[0032] Assume the target power grid area includes Each monitoring node, such as the distribution transformer, feeder head, and distributed power grid connection point, is monitored within a time window. Internal data collection yielded: in, The dimension number includes voltage amplitude. Current amplitude Active power reactive power Power factor Harmonic distortion rate ; The power grid topology is represented as a diagram. ,in For a set of nodes, Let the edge set be defined. Define the adjacency matrix. : Simultaneously define the electrical distance weight matrix ,in , This is the branch impedance.

[0033] Calculate the voltage deviation rate at each node: in, The rated voltage. Define the anomaly identification vector. ,in: In the formula, This is the voltage deviation threshold. This represents the threshold for the percentage of time spent on the task.

[0034] The NOTEARS algorithm (a non-combinatorial optimization structure learning algorithm based on the trace exponent and augmented Lagrange method) is adopted, incorporating topological constraints. The optimization objective is: in, It is a weighted adjacency matrix. Representing variables For variables The strength of the causal effect; It is the Frobenius norm; It is an L1 regularization term, which promotes sparsity; For topological penalty terms, For element-wise multiplication, this penalty applies to the causal boundaries between non-adjacent nodes in the topology. and This is the penalty coefficient.

[0035] In this embodiment, The value range is 0.01~0.1. The value range is 0.1 to 1.0, and can be adjusted according to the power grid scale and data noise level. For example, for distribution networks with fewer than 100 nodes, take... =0.05、 =0.5.

[0036] To ensure that the resulting graph is a directed acyclic graph (DAG), we introduce the differentiability constraint proposed by NOTEARS: in, For matrix exponents.

[0037] Solve using the augmented Lagrange method: Iterative updates via gradient descent Simultaneously, adaptive updates and The process continues until the adjacency matrix of the causal graph is obtained. In this embodiment, The initial value is 1.0. The initial value is 10.0; the adaptive update rule is: if ,but , ;otherwise , The convergence condition is: and .

[0038] For each abnormal node (Right now ), in the diagram Tracing back to its ancestral causes and effects: Calculate each ancestor node For abnormal nodes The overall causal effect is specifically achieved through the sum of the effects of all pathways: in, From arrive The set of all directed paths.

[0039] Select the one with the largest contribution One node as a candidate reason: Among them, a contribution threshold can be set. ,reserve The node.

[0040] For each abnormal node Generate contribution decomposition: The final output is a list of candidate causes for each anomalous node. Each cause includes the cause node, node type, causal contribution, and description of the main influencing path.

[0041] Step S4: For each candidate cause, calculate the comprehensive scheduling cost required to intervene according to the predefined evaluation rules; wherein the evaluation rules at least integrate the technical complexity level representing the difficulty of the intervention measures, the resource consumption representing resource consumption, and the time period representing the time consumption. Specifically, the technical complexity level is a parameter used to classify the difficulty of implementing various governance measures based on predefined classification rules; the resource consumption is a parameter used to quantify the human, material, and equipment resources required for intervention based on historical data and a resource weight coefficient library.

[0042] The classification of technical complexity levels can be achieved using methods such as feature clustering analysis based on historical data, hierarchical analysis, decision tree method, and fuzzy comprehensive evaluation method.

[0043] In practice, a feature clustering analysis method based on historical data is used to classify the levels of technical complexity, providing feature data of historical measures. Similarity is calculated based on objective feature vectors, eliminating subjective biases such as personal preferences and varying scales that may exist in expert scoring. The historical governance measures database contains... Items, each measure extract 3D feature vector ,include: Characteristics of the working environment: ; Skill requirements characteristics: (Skill levels: Beginner = 1, Intermediate = 2, Advanced = 3); Device dependency characteristics: ; Cooperative complexity characteristics: ; Security risk characteristics: ; By using an automatic classification based on improved K-means, continuous features are standardized using Z-score, while binary features retain their original values.

[0044] In the formula, For the first The first governance measure The original values ​​of each feature; For the first The first governance measure The value after standardizing the features; For all The first measure The sample mean of each feature; For all The first measure The sample standard deviation of each feature.

[0045] Introducing feature weight vectors defined by experts The distance is calculated as follows: in, For clusters The center of mass, Reflecting characteristics The degree of influence on complexity is determined by the following feature weight vector w: working environment feature weight 0.3, skill requirement feature weight 0.25, equipment dependence feature weight 0.2, collaborative complexity feature weight 0.15, and safety risk feature weight 0.1.

[0046] Clustering optimization objective: In the formula, is the regularization coefficient, used to control the strength of the second ordinal regularization term. >0, with values ​​ranging from 0.1 to 0.5. A larger value forces adjacent cluster centroids to be more ordered. The number of clusters K defaults to 5, corresponding to complexity levels 1 (very low) to 5 (very high). The second term is a regularization term, which forces the centroids to be arranged in an ordered manner, ensuring the monotonicity of the complexity level.

[0047] Will Each cluster is sorted by the score of the first principal component of its centroid, and mapped to a complexity level. .

[0048] Online update mechanism for new measures At that time, calculate its distance to the centroid of each cluster: In the formula, New measures The corresponding feature vectors undergo the same standardization process as the training data.

[0049] If the minimum distance is greater than the threshold (In this embodiment, If the average distance within all clusters is taken as 1.5 times, the clustering model will be retrained.

[0050] Define resource type set Each type of resource is further subdivided, such as , , .

[0051] For each measure Establish a standard operating procedure sequence Each step resource consumption quota Based on historical work order data, the following was obtained: in, For historical work orders intermediate steps Consume resources Quantity, This represents the number of work orders.

[0052] Basic consumption of resources: Environmental adjustment factors, taking temperature into account ,humidity The impact of environmental factors on efficiency: in, For regression coefficients, For reference only.

[0053] Final consumption: in, This is a skill adjustment factor.

[0054] The resource weight coefficient library is dynamically updated, and the weight composition model is as follows: in, As the benchmark weight; Environmental fluctuation factors are obtained from historical power grid operation databases or meteorological forecasting systems. This is the regional coefficient.

[0055] Based on the above quantitative results, the measures The overall scheduling cost is: in, The technical complexity level of the measure is categorized into levels 1-5, based on manual labeling or automatic classification of implementation difficulty in historical work orders. For resources The amount consumed comes from the Standard Operating Procedures (SOP). For resources The dynamic weighting coefficients are related to the regional human resource utilization equivalent and the implementation period. Related information is provided by the resource weight coefficient library. This refers to the standard operating time. This is the urgency coefficient, used to adjust the weight of time scheduling costs. These are the weighting coefficients for balancing the three scheduling costs.

[0056] Among them, the scheduling cost coefficient of technical complexity This can be obtained through regression analysis of historical project scheduling costs: in, This represents the actual total scheduling cost of project i. To manage the total number of work orders; For resources The actual amount consumed; For resources The actual weighting coefficients; This refers to the actual working time. This represents the actual urgency coefficient. This represents the level of technical complexity.

[0057] Time period coefficient The impact of a power outage per unit time can be estimated: The impact coefficient reflects the difference in the degree of impact of power outages on different user types. The coefficient is 0.1 to 0.3 for residential users, 0.5 to 1.0 for industrial and commercial users, and 1.5 to 2.0 for important users. The comprehensive impact coefficient can be obtained by weighting the user composition of the target power grid area.

[0058] Urgency coefficient The hierarchical quantization can be defined as: Step S5: Construct a pre-selection set based on the candidate causes, and automatically select a subset of causes from the pre-selection set; wherein, the optimization objective of selecting the subset of causes is to minimize the sum of the comprehensive scheduling costs of all candidate causes in the subset of causes, and it is expected that after intervening in the subset of causes, the at least one abnormal indicator value can be restored to within the indicator threshold. Specifically, a subset of reasons is automatically selected from the pre-selected set, including: Construct a combinatorial optimization problem, where decision variables represent whether to intervene in each candidate cause, and the objective function is set to minimize the total integrated scheduling cost of the intervened candidate causes; The constraints of the combinatorial optimization problem include: First constraint: The probability that the abnormal indicators return to normal after intervention is not less than a preset probability threshold, and the prediction of the preset probability threshold is based on the directed graph model.

[0059] Second constraint: Total intervention resource consumption shall not exceed the acceptable resource consumption limit; The third constraint is that the implementation order of each intervention measure satisfies the causal temporal relationship derived from the directed graph model. Solve the combinatorial optimization problem to obtain the subset of causes.

[0060] The combinatorial optimization problem is solved using a graph search algorithm, which, in the heuristic evaluation of the search path, simultaneously considers the cumulative scheduling cost of the selected causes and the prediction of the effectiveness of interventions on the unselected causes.

[0061] In practice, the combinatorial optimization problem is constructed as a 0-1 integer programming problem. Let the set of candidate causes determined in step S3 be... For each reason In step S4, one or more governance measures have been matched to form a set. Each measure The overall scheduling cost has been calculated. Expected contribution to the recovery of abnormal indicators .

[0062] Among them, the expected contribution Based on the overall causal effect of this measure on the anomaly node in the causal graph. The calculation formula is as follows: in, This represents the maximum total causal effect of all candidate causes.

[0063] Define binary decision variables: The objective function is to minimize the total scheduling cost of the intervention cause: The constraints include: (1) Probabilistic constraints (based on directed graphical model prediction): recovery probability after intervention The requirement is not lower than the preset threshold. : In the formula, The expected contribution.

[0064] Using a linear approximation: (2) Resource usage limit constraint (total intervention resource usage shall not exceed the limit) ): (3) Temporal Constraints (Deriving the Order of Causal Relationships from a Directed Graph Model): Let the causal order set be... Then the constraint is: (4) Logical constraints (at most one measure can be selected for each cause): The graph search algorithm is used to solve the problem, and the state is defined as follows: ,in, Selected reason index set; For each selected reason Specified Measures Index ; To accumulate scheduling costs ; Current recovery probability Initial state The target state is satisfied. and The state.

[0065] Transition from state to state Extend to new state Conditions: New reasons for choosing And all its precursor reasons (satisfy All of these have been included. The middle. For new reasons. Selection measures New status parameters updated: State ,definition ,in This represents the estimated minimum remaining scheduling cost.

[0066] Probability deficit calculation: like ,but .

[0067] Let the set of reasons for not selecting be... Excluding the possibility that the timing constraints are not met, i.e., not all predecessors are present. middle.

[0068] For each remaining cause Calculate its optimal unit scheduling cost contribution rate: Will according to Arrange in descending order to obtain the sequence .

[0069] Assuming partial selection of measures is possible, the contribution rate is accumulated from high to low until a certain condition is met. Let the cumulative contribution be... Estimate scheduling cost Process them in sequence: Regarding the cause Choose the measure with the highest contribution rate ; like Select all: , ; Otherwise, select a portion: take the proportion. , ,termination; final This heuristic is admissible.

[0070] This relaxed solution is a lower bound of the actual minimum remaining scheduling cost, therefore the heuristic function is admissible, and the optimal solution is found.

[0071] Step S6: Output governance action recommendations corresponding to the subset of causes.

[0072] Specifically, the proposed governance actions include a list of specific maintenance, repair, or infrastructure measures for each cause in the subset of causes, as well as a resource consumption estimate based on the level of technical complexity and resource consumption.

[0073] After the system completes the optimization and selection of the cause subset, it enters the automatic report generation stage. First, the system calls upon the measures knowledge base, which stores standard templates for various power grid operation, maintenance, and infrastructure measures in a structured manner. For each cause in the cause subset, the knowledge base matches one or more standard remedial measures based on the cause type and attributes. The system automatically fills in detailed information for each matched measure, forming a measure list. For example, if a transformer in a distribution area is overloaded, the knowledge base matches one or more measures based on the overload ratio and equipment model, such as adjusting the three-phase load balance, adding parallel capacitors, or increasing transformer capacity.

[0074] Secondly, the system performs resource consumption estimation. For each measure in the list, the system retrieves the technical complexity level and resource consumption details calculated in step S4. This data is then combined and formatted according to a predefined report template to generate a clear resource consumption estimation table. Finally, the system integrates the list of cause-and-effect measures with the resource consumption estimation table, and includes an overview of the overall expected governance goals and scheduling costs, forming a complete governance action recommendation report.

[0075] The resource utilization estimates in the report are directly linked to the dynamic scheduling cost model and serve as the basis for subsequent resource utilization plans, material procurement plans, and human resource allocation. This changes the extensive model of rough estimation and facilitates precise resource utilization control and optimal resource allocation, avoiding resource waste or shortages.

[0076] Specifically, the output form of the governance action recommendations includes structured task instructions. In these structured task instructions, each specific measure is automatically associated with best practice documents, standard operating procedure videos, or key risk point prompts from its historical execution case library, and these are sent together to the execution terminal. The structured task instructions are directly parsed by downstream execution terminals to create corresponding inspection, maintenance, or construction work orders. The structured task instructions are in JSON format and include fields such as task ID, anomaly location, measure type, work steps, resource list, execution time limit, and associated knowledge ID. They interface with the downstream work order system via a RESTful API, conforming to the "Power System Operation and Maintenance Work Order Data Exchange Specification" DL / T 2035-2019.

[0077] In practice, the system transforms the governance action recommendation report generated in the previous stage into machine-readable structured task instructions. The work order information interface recorded at a specific moment in the power system includes the work order title, execution period, and related case information, defining multiple task nodes. Secondly, when generating each task node, the system automatically activates a knowledge association engine. This engine uses the characteristics of the current measure as keywords to search the historical execution case library in real time. The case library stores complete files of similar past operations, including successful best practice documents, recorded standard operation videos, and summarized key risk points. The ID or link of the most relevant knowledge document is embedded into the related knowledge field of the current task node. Finally, a complete structured task instruction data package is encapsulated. The work order content includes governance measures, related historical cases, automatically associated best practice documents, and the creation of inspection and maintenance work orders, and is automatically pushed to the downstream work order management system or the mobile inspection terminal of on-site personnel. After receiving the instruction, the downstream system can directly parse the data package and automatically create the corresponding electronic work order, and push the associated knowledge documents to the work order executor.

[0078] Traditional work order creation relies on manual input, which is prone to errors and time-consuming. This preferred implementation solution automates the workflow from intelligent analysis to work order generation. On-site personnel receive structured instructions via mobile terminals, with clear and concise work steps and resource lists, avoiding misunderstandings that may arise from telephone communication or paper work orders, and improving the accuracy of first-time operations.

[0079] Transforming optimal decision-making outcomes into governance action recommendations that directly guide on-site work results in a structured, actionable report. This report clearly lists each selected root cause and its corresponding specific governance measures and technical aspects; it details the personnel, materials, equipment, and estimated scheduling costs required for each measure, forming a complete resource allocation list and schedule. Furthermore, it can generate machine-readable structured task instructions from these recommendations, automatically pushing them to the work order management system or mobile inspection terminal to directly create work orders. The instructions can even link to historical videos of similar cases and risk warnings, empowering on-site personnel to operate safely and efficiently.

[0080] Specifically, in a preferred embodiment of this application, after outputting the governance action suggestion, feedback on the actual execution result of the suggestion is received; based on the feedback data, the calculation rules for the comprehensive scheduling cost or the prediction model for the abnormal governance effect conditions are updated.

[0081] This invention provides a power system anomaly management method based on causal inference. It integrates technical complexity, resource consumption, and time cycle to construct a comprehensive scheduling cost evaluation system, providing a unified standard for scheme optimization. With the dual optimization objectives of minimizing scheduling costs and restoring anomaly indicators, it automatically selects a subset of causes, balancing the scheduling cost of governance with the anomaly elimination effect, avoiding excessive or ineffective intervention, and effectively reducing the ineffective use of power grid resources. It automatically generates governance action suggestions, replacing subjective manual decision-making, shortening the anomaly governance response and execution cycle, and improving power grid operation and maintenance efficiency. Simultaneously, based on causal inference, it accurately locates candidate causes of anomalies, establishing a logical connection between anomalies and their causes from the root, improving the accuracy of cause identification, and reducing the probability of misjudgment and omission.

[0082] After the governance measures are implemented, collect actual post-governance operational data and actual scheduling costs. and actual time spent Using feedback data, Bayesian updates or online learning methods are employed to dynamically adjust resource weight coefficients. Or efficiency coefficient.

[0083] in, It is a forgetting factor.

[0084] This governance case will be used as a new sample for incrementally updating the causal graph. Alternatively, an effect prediction model can be used to achieve system self-optimization.

[0085] Specifically, in a preferred embodiment of this application, the resource weight coefficient library on which the resource occupancy calculation is based is synchronized in real time or periodically with the external operating environment data source to update the resource acquisition difficulty coefficient and the regional human resource occupancy equivalent data, and this update is dynamically mapped to the calculation result of the comprehensive scheduling cost.

[0086] By establishing a standardized data interface layer, information is retrieved from external data sources in real time or periodically via APIs, transforming unstructured data into structured time-series data. In practice: For each resource Its environmental reference weight The weighted synthesis yielded the following: in, For data source The weights are dynamically adjusted based on their historical accuracy and update frequency. The confidence level of each weight is also calculated. This reflects the consistency and timeliness of the data.

[0087] When the weight change exceeds the threshold An update is triggered when the confidence level is higher than the threshold. To prevent drastic fluctuations in weights, an exponential smoothing method is used to update the weights in the resource weight coefficient library. : in, This is a smoothing factor, and a larger value can be taken when the confidence level is high, such as 0.7.

[0088] When performing scheduling cost calculation, the latest version of the weights is automatically applied. Substitute into the formula: Record the weight version used in each scheduling cost calculation to ensure the traceability of scheduling cost estimation. Abnormal weight monitoring and manual review: When the daily fluctuation of a resource weight exceeds the safety threshold, the system automatically freezes the weight update and pushes an alarm to the administrator for manual review.

[0089] This preferred embodiment eliminates resource occupancy estimation errors by establishing a synchronization and mapping mechanism between the resource weight coefficient library and external operating environment data. Traditional static weight libraries cannot reflect short-term environmental fluctuations, often leading to significant discrepancies between the actual scheduling cost of a project and the initial estimate, resulting in resource overruns or project stagnation. This preferred embodiment achieves near real-time linkage between weights and environmental conditions, ensuring that the scheduling cost estimation of the governance plan generated at any given time is based on current operating environment data, significantly improving the reliability of resource occupancy plans.

[0090] Specifically, in the calculation of resource occupancy, for the same intervention measure, different resource weight coefficients or efficiency coefficients are associated according to different time periods or different urgency levels of its implementation, so as to calculate differentiated comprehensive scheduling costs.

[0091] Based on the characteristics of the power grid load and operational management regulations, the day is divided into multiple time periods, and each time period... Define time period weighting coefficient Example as follows: Peak hours (e.g., 18:00-22:00): ; Normal hours (e.g., 8:00-18:00): ; Off-peak hours (e.g., 23:00-7:00): ; For materials and equipment, and It is usually 1.0.

[0092] Define the set of urgency levels and for each level Associate a resource efficiency coefficient This reflects the potential decrease in resource utilization efficiency or additional scheduling costs that may result from accelerating progress under tight circumstances. Examples are as follows: Routine (planned work, well-prepared): ; Urgent (requires immediate action, short preparation time): 5; Urgent (immediate handling, such as emergency repair): .

[0093] In basic weight The time period and urgency are superimposed to form a contextualized weight. : Urgency level May affect standard operating time And resource utilization efficiency, therefore the actual computation time is adjusted. ,in This is a time compression factor, leading to a higher demand for human resources and other resources. Synchronous adjustment. Substituting the above parameters into the comprehensive scheduling cost formula: in, Urgency level The corresponding urgency coefficient reflects the value of time.

[0094] When generating a governance plan, the system allows users or automatically selects or sets the implementation period and urgency level based on the nature of the anomaly. The system will then adjust the plan according to different... By combining and calculating multiple versions of scheduling costs in parallel, the system can achieve the desired results.

[0095] This preferred embodiment introduces implementation time and urgency level as refined dimensions for calculating dispatch costs, enabling a quantitative display of the differences in dispatch costs when the same governance measure is implemented at different times or with different levels of urgency. This allows decision-makers to make refined trade-offs between governance effectiveness, implementation speed, and dispatch costs. For example, to restore power more quickly, decision-makers can clearly see how much additional dispatch cost is required, thus making a comprehensive and optimal decision.

[0096] This application also provides a power system anomaly management system based on causal inference, including: The data acquisition unit is used to acquire the operating status data and power grid topology data of the target power grid area; The first analysis unit is used to calculate at least one abnormal indicator value representing power quality based on the operating status data, and compare it with a preset indicator threshold to determine whether there is a power quality abnormality. The second analysis unit is used to determine at least one candidate cause of the abnormality based on the operating status data and the power grid topology data if it is determined that there is an abnormality in power quality. The computing unit is used to calculate the comprehensive scheduling cost required to intervene in each of the candidate causes, according to a predefined evaluation rule; wherein the evaluation rule at least integrates the technical complexity level representing the difficulty of the intervention measure, the resource consumption representing the resource consumption, and the time period representing the time consumption. A filtering unit is configured to construct a pre-selection set based on the candidate causes, and automatically select a subset of causes from the pre-selection set; wherein the optimization objective of selecting the subset of causes is to minimize the sum of the comprehensive scheduling costs of all candidate causes in the subset of causes, and it is expected that after intervention on the subset of causes, the at least one abnormal indicator value can be restored to within the indicator threshold. The output unit is used to output governance action recommendations corresponding to the subset of causes.

[0097] The functional explanation of each unit in this embodiment is the same as that of a power system anomaly management method based on causal inference, and the technical effect is the same, so it will not be repeated here.

[0098] Those skilled in the art will recognize that the units of the various examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of the invention.

[0099] In the embodiments provided by the present invention, it should be understood that the division of units is only a logical functional division. In actual implementation, there may be other division methods, such as multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored.

[0100] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0101] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0102] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the specification of the present invention.

Claims

1. A power system anomaly management method based on causal inference, characterized in that, Includes the following steps: Acquire operational status data and power grid topology data for the target power grid area; Based on the operating status data, at least one abnormal indicator value characterizing power quality is calculated and compared with a preset indicator threshold to determine whether there is a power quality abnormality. If an abnormal power quality is determined, at least one candidate cause for the abnormality is identified based on the operating status data and the power grid topology data. For each of the candidate causes, the comprehensive scheduling cost required to intervene is calculated according to predefined evaluation rules; wherein, the evaluation rules at least integrate the technical complexity level representing the difficulty of the intervention measures, the amount of resources used representing resource consumption, and the time period representing the time consumption. The candidate causes form a pre-selection set, and a subset of causes is automatically selected from the pre-selection set; wherein, the optimization objective of selecting the subset of causes is to minimize the sum of the comprehensive scheduling costs of all candidate causes in the subset of causes, and it is expected that after intervention on the subset of causes, the at least one abnormal indicator value can be restored to within the indicator threshold. Output governance action recommendations corresponding to the subset of causes.

2. The power system anomaly management method based on causal inference according to claim 1, characterized in that, Based on the operational status data and power grid topology data, at least one candidate cause for the anomaly is identified, specifically including: Based on the operating status data and power grid topology data, a directed graph model is constructed to represent the causal relationships between variables; wherein, when constructing the directed graph model, the model structure is constrained according to the electrical connection relationships represented by the power grid topology data; Based on the directed graph model, at least one candidate cause for power quality anomalies is identified.

3. The power system anomaly management method based on causal inference according to claim 2, characterized in that, Before constructing the directed graph model, the operating status data is preprocessed. The preprocessing includes: for time segments with missing or abnormal data, a collaborative repair algorithm based on the power grid topology and time-series correlation is used to complete and correct the data.

4. The power system anomaly management method based on causal inference according to claim 1, characterized in that, The technical complexity level is a parameter used to classify the difficulty of implementing various governance measures based on predefined classification rules; the resource consumption is a parameter used to quantify the human, material, and equipment resources required for intervention based on historical data and a resource weight coefficient library.

5. The power system anomaly management method based on causal inference according to claim 4, characterized in that, The resource weight coefficient library on which the resource occupancy calculation is based is synchronized in real time or periodically with the external operating environment data source to update the resource acquisition difficulty coefficient and the regional human resource occupancy equivalent data, and this update is dynamically mapped to the calculation result of the comprehensive scheduling cost.

6. The power system anomaly management method based on causal inference according to claim 5, characterized in that, In the calculation of resource occupancy, for the same intervention measure, different resource weight coefficients or efficiency coefficients are associated according to different time periods or different urgency levels of its implementation, so as to calculate differentiated comprehensive scheduling costs.

7. The power system anomaly management method based on causal inference according to claim 2, characterized in that, Automatically selecting a subset of reasons from the pre-selected set, specifically including: Construct a combinatorial optimization problem, where decision variables represent whether to intervene in each candidate cause, and the objective function is set to minimize the total integrated scheduling cost of the intervened candidate causes; The constraints of the combinatorial optimization problem include: First constraint: The probability that the abnormal indicators return to normal after intervention is not less than a preset probability threshold, and the prediction of the preset probability threshold is based on the directed graph model; Second constraint: Total intervention resource consumption shall not exceed the acceptable resource consumption limit; The third constraint is that the implementation order of each intervention measure satisfies the causal temporal relationship derived from the directed graph model. Solve the combinatorial optimization problem to obtain the subset of causes.

8. The power system anomaly management method based on causal inference according to claim 1, characterized in that, The proposed governance action includes a list of specific maintenance, repair, or infrastructure measures for each cause in the subset of causes, as well as a resource consumption estimate based on the level of technical complexity and resource consumption.

9. A power system anomaly management method based on causal inference according to claim 8, characterized in that, The output of the governance action recommendations includes structured task instructions. In these structured task instructions, each specific measure is automatically associated with best practice documents, standard operating procedure videos, or key risk point prompts from its historical execution case library, and these are sent together to the execution terminal. The structured task instructions are directly parsed by the downstream execution terminal to create corresponding inspection, maintenance or construction work orders.

10. A power system anomaly management system based on causal inference, characterized in that, include: The data acquisition unit is used to acquire the operating status data and power grid topology data of the target power grid area; The first analysis unit is used to calculate at least one abnormal indicator value representing power quality based on the operating status data, and compare it with a preset indicator threshold to determine whether there is a power quality abnormality. The second analysis unit is used to determine at least one candidate cause of the abnormality based on the operating status data and the power grid topology data if it is determined that there is an abnormality in power quality. The computing unit is used to calculate the comprehensive scheduling cost required to intervene in each of the candidate causes, according to a predefined evaluation rule; wherein the evaluation rule at least integrates the technical complexity level representing the difficulty of the intervention measure, the resource consumption representing the resource consumption, and the time period representing the time consumption. A filtering unit is configured to construct a pre-selection set based on the candidate causes, and automatically select a subset of causes from the pre-selection set; wherein the optimization objective of selecting the subset of causes is to minimize the sum of the comprehensive scheduling costs of all candidate causes in the subset of causes, and it is expected that after intervention on the subset of causes, the at least one abnormal indicator value can be restored to within the indicator threshold. The output unit is used to output governance action recommendations corresponding to the subset of causes.

Images