A power distribution network fault processing method and system

Abstract

The application discloses a kind of power distribution network fault processing method and system, applied to power distribution network fault processing technical field, method includes: based on the judgment result of real-time operation data of power distribution network, the abnormal data of abnormal power distribution equipment is determined;Abnormal data of abnormal power distribution equipment and structured knowledge data are carried out rule inference, and rule link data is obtained;Based on rule link data, abnormal data is carried out semantic inference, and initial fault processing data is obtained;When initial fault processing data and rule link data match, based on initial fault processing data, the fault processing of abnormal power distribution equipment is realized;When initial fault processing data and rule link data do not match, based on rule link data, initial fault processing data is optimized, to realize the fault processing of abnormal power distribution equipment based on optimized fault processing data.The application improves the accuracy of power distribution network fault inference.

Description

Technical Field

[0001] This invention relates to the field of power distribution network fault handling technology, and in particular to a power distribution network fault handling method and system. Background Technology

[0002] In distribution network fault handling scenarios, existing methods typically employ a single inference architecture to handle faults in abnormal distribution equipment. For example, inference methods based on expert rule bases pre-encode power safety regulations, equipment operating specifications, and protection setting rules into condition-action rules. When processing real-time operational data of the distribution network, rule matching is used to determine the equipment fault type and output a handling plan. However, existing distribution network fault handling methods suffer from low rule matching efficiency due to the proliferation of rules when facing complex fault scenarios; for fault handling scenarios without pre-defined rules, fault inference may also fail. This results in inaccurate fault inference in existing distribution network fault handling methods. Summary of the Invention

[0003] This invention provides a method and system for handling power distribution network faults, in order to solve the technical problem of inaccurate fault reasoning in existing power distribution network fault handling methods.

[0004] To address the aforementioned technical problems, embodiments of the present invention provide a method for handling power distribution network faults, including: Based on the judgment results of real-time operation data of the power distribution network, the abnormal data of abnormal power distribution equipment is determined; Rule-based reasoning is performed on the abnormal data and structured knowledge data of the abnormal power distribution equipment to obtain the rule link data of the abnormal power distribution equipment; Based on the rule link data, semantic reasoning is performed on the abnormal data to obtain the initial fault handling data of the abnormal power distribution equipment. When the initial fault handling data is matched with the rule link data, the fault handling of the abnormal power distribution equipment is realized based on the initial fault handling data. When the initial fault handling data does not match the rule link data, the initial fault handling data is optimized based on the rule link data so as to realize the fault handling of the abnormal power distribution equipment based on the optimized fault handling data.

[0005] As one preferred embodiment, the judgment result based on the real-time operation data of the power distribution network determines the abnormal data of the abnormal power distribution equipment, including: The data source is based on the real-time operation data of the distribution network, and the judgment rule data of the distribution network is determined. When the real-time operating data is determined to be abnormal based on the judgment rule data, the abnormal power distribution equipment of the power distribution network is determined based on the data source; Based on the real-time operating data of the abnormal power distribution equipment, the abnormal data of the abnormal power distribution equipment is determined.

[0006] As one preferred embodiment, the step of performing rule-based reasoning on the abnormal data and structured knowledge data of the abnormal power distribution equipment to obtain the rule-based link data of the abnormal power distribution equipment includes: Pattern matching is performed on the abnormal data to obtain the fault mode of the abnormal power distribution equipment; Based on the fault mode, determine the fault topology range and adjustment constraint tuples of the abnormal power distribution equipment; Based on the adjustment action elements in the adjustment constraint tuple, the parameter adjustment range of the abnormal power distribution equipment is determined; The fault mode, the fault topology range, the adjustment constraint tuple, and the parameter adjustment range are combined to obtain the rule link data of the abnormal power distribution equipment.

[0007] As one preferred embodiment, the step of performing semantic reasoning on the abnormal data based on the rule link data to obtain the initial fault handling data for the abnormal power distribution equipment includes: Input the rule link data and the anomaly data into the large language model; The abnormal data is semantically reasoned using the large language model to obtain the structured reasoning text and processing action vectors output by the large language model. The structured reasoning text and the processing action vector are combined to obtain the initial fault handling data of the abnormal power distribution equipment.

[0008] As one preferred embodiment, when the initial fault handling data does not match the rule link data, optimizing the initial fault handling data based on the rule link data to achieve fault handling of the abnormal power distribution equipment based on the optimized fault handling data includes: The processing action vector is validated based on the rule link data to obtain the validation result; When the verification result indicates that the processing action vector does not match the rule link data, the structured reasoning text and the processing action vector are adjusted based on the verification result to optimize the initial fault processing data. Based on the optimization results of the fault handling data, the fault handling of the abnormal power distribution equipment is realized.

[0009] Another embodiment of the present invention provides a power distribution network fault handling system, comprising: The abnormal data determination module is used to determine abnormal data of abnormal power distribution equipment based on the judgment results of real-time operation data of the power distribution network. The rule link data determination module is used to perform rule reasoning on the abnormal data and structured knowledge data of the abnormal power distribution equipment to obtain the rule link data of the abnormal power distribution equipment. The initial fault handling data determination module is used to perform semantic reasoning on the abnormal data based on the rule link data to obtain the initial fault handling data of the abnormal power distribution equipment. An initial fault processing data verification module is used to implement fault processing of the abnormal power distribution equipment based on the initial fault processing data when the initial fault processing data matches the rule link data. The power distribution network fault handling module is used to optimize the initial fault handling data based on the rule link data when the initial fault handling data does not match the rule link data, so as to realize the fault handling of the abnormal power distribution equipment based on the optimized fault handling data.

[0010] As one preferred embodiment, the abnormal data determination module includes: The judgment rule data determination unit is used to determine the judgment rule data of the distribution network based on the data source of real-time operation data of the distribution network; An abnormal power distribution equipment determination unit is used to determine the abnormal power distribution equipment of the power distribution network based on the data source when the real-time operating data is determined to be abnormal based on the judgment rule data. An abnormal data determination unit is used to determine the abnormal data of the abnormal power distribution equipment based on the real-time operating data of the abnormal power distribution equipment.

[0011] As one preferred embodiment, the rule link data determination module includes: A fault mode determination unit is used to perform pattern matching on the abnormal data to obtain the fault mode of the abnormal power distribution equipment. The fault mode analysis unit is used to determine the fault topology range and adjustment constraint tuple of the abnormal power distribution equipment according to the fault mode. The parameter adjustment range determination unit is used to determine the parameter adjustment range of the abnormal power distribution equipment based on the adjustment action elements in the adjustment constraint tuple. The rule link data determination unit is used to combine the fault mode, the fault topology range, the adjustment constraint tuple, and the parameter adjustment range to obtain the rule link data of the abnormal power distribution equipment.

[0012] As one preferred embodiment, the initial fault handling data determination module includes: The model input unit is used to input the rule link data and the abnormal data into the large language model; The semantic reasoning unit is used to perform semantic reasoning on the abnormal data through the large language model to obtain the structured reasoning text and processing action vector output by the large language model. The initial fault handling data determination unit is used to combine the structured reasoning text and the processing action vector to obtain the initial fault handling data of the abnormal power distribution equipment.

[0013] As one preferred embodiment, the power distribution network fault handling module includes: The action vector verification unit is used to verify the action vector based on the rule link data and obtain the verification result. An initial fault handling data optimization unit is used to adjust the structured inference text and the processing action vector based on the verification result when the verification result shows that the processing action vector does not match the rule link data, so as to optimize the initial fault handling data. The power distribution network fault handling unit is used to handle the faults of the abnormal power distribution equipment based on the optimization results of the fault handling data.

[0014] This application achieves the determinism and security of rule-based reasoning, organically integrating it with the generalization and semantic understanding capabilities of semantic reasoning, through technical logic involving anomaly data identification, rule-based reasoning, semantic reasoning, compliance matching judgment, and optimization processing. The rule-based reasoning chain outputs compliance handling boundaries at the front end of the process, establishing an insurmountable safety barrier for semantic reasoning. Within this safety barrier, the semantic reasoning chain leverages the semantic understanding and generation capabilities of a large language model. The compliance matching judgment mechanism ensures that only handling solutions that fully meet safety constraints are output. The optimization processing mechanism corrects content when non-compliant situations arise. This approach overcomes both the high maintenance costs of pure rule-based reasoning when facing new and complex faults, and the semantic illusions and safety rule violations inherent in pure semantic reasoning. It achieves a fusion of safety compliance and intelligent generalization, improving the accuracy of fault reasoning in distribution networks. Attached Figure Description

[0015] Figure 1 This is one of the flowcharts of the power distribution network fault handling method provided by the present invention; Figure 2 This is the second flowchart of the power distribution network fault handling method provided by the present invention; Figure 3 This is a schematic diagram of the structure of the power distribution network fault handling system provided by the present invention.

[0016] Figure label: Among them, 301 is the abnormal data determination module; 302 is the rule link data determination module; 303 is the initial fault handling data determination module; 304 is the initial fault handling data verification module; and 305 is the distribution network fault handling module. Detailed Implementation

[0017] 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 embodiments of the present invention, and not all embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0018] In the description of this application, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0019] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal communication between two components. The terms "vertical," "horizontal," "left," "right," "upper," "lower," and similar expressions used herein are for illustrative purposes only and do not indicate or imply that the device or component referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as limiting the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0020] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0021] See Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the power distribution network fault handling method provided by the present invention, as shown below. Figure 1 As shown, this embodiment includes steps 100 to 500, and the specific steps are as follows: Step 100: Based on the judgment results of the real-time operation data of the power distribution network, determine the abnormal data of abnormal power distribution equipment; In this embodiment, the real-time operating data of the distribution network refers to the operating parameters and status information of the distribution network collected in real time through the Supervisory Control and Data Acquisition (SCADA) system, the Power Management System (PMS) system, and the distribution IoT platform. Real-time operating data includes, but is not limited to, telemetry data (e.g., three-phase voltage, current, active power, and reactive power), remote signaling data (e.g., switch position signals and protection action signals), alarm data, equipment status data, maintenance records, and environmental monitoring data. The judgment result in this embodiment is a binary result indicating whether there is an operational anomaly in the distribution network, or a quantitative result including the degree of anomaly.

[0022] As one implementation method, firstly, a data interface is established with the SCADA system, PMS system, and distribution IoT platform through a message bus and real-time subscription mechanism, and multiple heterogeneous real-time operation data streams from various sources are received using one or more protocols. The received multi-source data is uniformly mapped into standard event objects, which include event identifier, event timestamp, data source identifier, device identifier, event type, observation vector, quality code, electrical location, and additional metadata fields. Online detection of the standard event objects is performed based on preset anomaly criteria. The above anomaly criteria can be a single threshold criterion, such as a voltage over-limit criterion: when the observed voltage value is greater than the upper voltage limit or less than the lower voltage limit, it is judged as an anomaly; it can also be a composite criterion, such as a short-circuit trip criterion: when the three-phase fault current characteristic value is greater than the instantaneous trip setting value, the protection action flag is true, and the switch trip flag is true, all three conditions must be met simultaneously to determine a short-circuit trip anomaly; it can also be a trend criterion, such as a distribution transformer oil temperature rising trend anomaly criterion: when the derivative of oil temperature with respect to time is greater than a first threshold and the average temperature over the past three hours is greater than a second threshold, it is judged as an anomaly. When an anomaly trigger condition is detected, the power distribution equipment experiencing the anomaly is automatically identified. Since real-time alarm data typically represents equipment identity in non-standardized forms such as point table addresses, interval numbers, channel numbers, or remote signaling codes, an entity resolution function is used. This function performs joint resolution based on the point table mapping table, equipment coding rules, topology adjacency table, and geographical ledger, mapping the original alarm identifier to a unified equipment entity. In cases of one-to-many ambiguity, the maximum a posteriori probability identification method is used, combined with historical matching records, to determine the most probable equipment entity. After equipment location is completed, anomaly data is determined based on the equipment's real-time operating data. This anomaly data includes the observations that triggered the anomaly criterion, as well as relevant timestamps and data source information.

[0023] By establishing an endogenous data linkage mechanism with the underlying production system, the inference engine can subscribe to and perceive changes in the power grid status in real time. It can proactively identify abnormal equipment and acquire abnormal data the instant an anomaly occurs, eliminating the need for manual input of query commands or timed batch processing triggers, thus shortening the time delay from anomaly occurrence to data acquisition. Simultaneously, through multi-source data standardization and entity parsing mechanisms, heterogeneous data from different systems can be uniformly identified and processed, improving the accuracy and reliability of anomaly data identification.

[0024] Step 200: Perform rule reasoning on the abnormal data and structured knowledge data of the abnormal power distribution equipment to obtain the rule link data of the abnormal power distribution equipment; In this embodiment, structured knowledge data refers to pre-stored, structured knowledge in the power distribution field in a static knowledge base, including but not limited to equipment ledger information, power grid topology connections, protection setting parameters, operating mode data, historical defect records, maintenance records, operation ticket templates, power safety regulations constraints, and related environmental information. Rule-based reasoning in this embodiment refers to the process of deterministic logical deduction of input data based on pre-encoded expert rules, with the reasoning results being interpretable and tamper-proof. Rule-based link data in this embodiment refers to the structured data set output by the rule-based reasoning link, used to constrain subsequent semantic reasoning, including fault assessment results and compliance handling boundaries.

[0025] As one implementation method, after identifying the abnormal power distribution equipment and abnormal data, relevant structured knowledge data is extracted from a static knowledge base centered on the abnormal power distribution equipment. Specifically, a knowledge graph is constructed using a hybrid storage method of graph database and relational database. The graph database is used to represent equipment nodes, electrical connection relationships, protection logic, regional hierarchy, and historical event associations, while the relational database is used to store ledger details, setting tables, maintenance work orders, and procedure clauses. Given an abnormal power distribution equipment entity, a k-hop neighborhood subgraph is extracted centered on it. Preferably, for fault emergency handling scenarios, k=3 is selected, including upstream power supply points, downstream branches, tie switches, protection configurations, adjacent transformer areas, and the most recent operation record.

[0026] To avoid the inference efficiency degradation caused by an excessively large knowledge subgraph, candidate nodes are scored based on importance. The scoring criteria include topological proximity, temporal relevance, equipment risk level, and correlation with historical similar cases. Nodes with the highest scores are selected to form a lightweight structured knowledge context. Anomaly data and the loaded structured knowledge data are input into the rule-based inference engine, which performs deterministic inference based on expert thought chains. These expert thought chains include state recognition, topology tracing, setting verification, and safety boundary solving. In the state recognition stage, the rule engine performs pattern matching between real-time observed vectors and the expected features of each fault mode in the rule base, calculates the matching score, and selects the fault mode with the highest score exceeding a threshold as the current fault mode. In the topology tracing stage, for tripping faults, the rule engine starts from the protection switch and follows the possible propagation path of the fault current. Based on the switch states of each edge in the current topology graph, it determines the reachable set and locates the fault candidate segment by comparing the boundary division of the pressurized and unpressurized areas. In the setting verification stage, the rule engine compares the observed fault current with the instantaneous overcurrent setting and overcurrent setting of the protection device, and verifies whether the protection action time is within the expected time window to determine the rationality of the protection action. In the safety boundary solution stage, the rule engine, based on power safety regulations and operation and maintenance specifications, solves for the compliance handling boundary that cannot be broken in subsequent operations. This compliance handling boundary is represented by a set of constraint inequalities. After the rule reasoning is completed, the rule link data is output, which includes fault type, fault segment, compliance handling boundary, and set of mandatory safety requirements.

[0027] By employing a rule-based reasoning chain to perform deterministic fault assessment and security boundary determination at the very beginning of the reasoning process, the interpretability and determinism of expert rules are leveraged to ensure that fault diagnosis results and security boundaries are verifiable and tamper-proof. Simultaneously, the compliance handling boundaries output by rule-based reasoning define an inviolable range for subsequent semantic reasoning, fundamentally guaranteeing the security and compliance of the entire reasoning process.

[0028] Step 300: Perform semantic reasoning on the abnormal data based on the rule link data to obtain the initial fault handling data of the abnormal power distribution equipment; In this embodiment, semantic reasoning refers to the process of semantic understanding and text generation of input data using a large language model. It is capable of understanding unstructured natural language information and generating text output with contextual coherence. The initial fault handling data in this embodiment refers to fault handling-related data generated for the first time by the semantic reasoning model under the constraints of rule-based link data, without compliance verification. This includes text explaining the fault cause and suggested handling steps.

[0029] As one implementation method, rule-based data (including fault types, fault segments, compliance handling boundaries, and mandatory security requirements) are combined with anomaly data and structured knowledge data to construct a prompting context for the semantic reasoning model. The prompting context can be organized using a preset template format and includes at least the following slots: event summary (to summarize the basic situation of the current anomaly event), device and topology context (to provide structured information about the anomaly device and its associated devices), deterministic fault conclusion (to clarify the fault type and fault segment determined by rule-based reasoning), inviolable security boundaries (to inform the model of the specific content of compliance handling boundaries in natural language), mandatory warnings (to emphasize the safety procedures that must be followed), and output format constraints (to specify the structure and format of the model's output text).

[0030] The constructed contextual clues are input into a large language model, which then generates an output sequence step by step within an autoregressive decoding framework. Unlike conventional large language model applications, the model in this embodiment is not generated entirely freely, but conditionally within the compliance boundaries defined by the rule-based data. The model output includes at least two parts: first, an explanation of the fault cause, describing the cause and mechanism of the anomaly in natural language; and second, a detailed breakdown of the handling steps, expanding the large-granularity handling strategy given by rule inference into a sequence of specific, executable operational steps. In addition, the model can also output operational precautions and auxiliary suggestions. After the model completes generation, it outputs initial fault handling data.

[0031] By introducing a large language model for semantic reasoning, the model's semantic understanding and context learning capabilities are fully utilized to generate readable explanations of fault causes and detailed handling steps, thus compensating for the shortcomings of pure rule-based reasoning in semantic understanding and generalization to complex scenarios. Furthermore, since semantic reasoning is performed within the compliance boundaries defined by rule-based data chains, the security of the model-generated content is guaranteed.

[0032] Step 400: When the initial fault handling data matches the rule link data, the fault handling of the abnormal power distribution equipment is implemented based on the initial fault handling data; In this embodiment, matching the initial fault handling data with the rule-based data means that the key elements in the initial fault handling data, such as the handling actions, operation objects, and operation sequences, all satisfy all the constraints defined by the compliant handling boundaries in the rule-based data. Fault handling in this embodiment refers to outputting the verified fault handling data to maintenance personnel or downstream execution systems to guide or execute actual power distribution fault handling operations.

[0033] As one implementation method, compliance verification is performed on the initial fault handling data. For the handling text, a step-level verification approach is adopted. Specifically, the handling steps in the initial fault handling data are divided into independent action units, each containing three elements: operation type, operation object, and preconditions. Each action unit is mapped into a structured action vector using a semantic parser. Then, each action vector is compared with the compliance handling boundary in the rule link data to determine whether the action vector satisfies all the constraints defined by the compliance handling boundary. These constraints include, but are not limited to: prohibiting closing the circuit with the ground wire connected (if the operation type of an action is closing and the operation object is currently in a grounded state, it is considered a violation); requiring voltage verification before grounding (if there is a grounding action but no voltage verification action was performed beforehand, it is considered a violation); the transferred load does not exceed the tie line capacity (if the transferred load exceeds the tie line capacity limit, it is considered a violation); and isolation operations must target fault candidate sections (if the object of the isolation operation is not within the fault candidate section set, it is considered a violation).

[0034] If all action units in the initial fault handling data satisfy all the above constraints, then the initial fault handling data is determined to match the rule link data. In this case, the initial fault handling data is directly output as the final fault handling result. The output method can be to display it to maintenance personnel in structured text form on the human-machine interface, or to issue the handling instructions to the distribution automation system for execution through a standardized interface.

[0035] By performing compliance checks on the semantic reasoning results at the output stage, it is ensured that only handling solutions that fully comply with security procedures can be output and executed, thereby effectively preventing non-compliant handling suggestions from entering the actual operation stage. Simultaneously, through step-level fine-grained verification, specific violating action units can be located, providing precise basis for subsequent optimization processing.

[0036] Step 500: When the initial fault handling data does not match the rule link data, optimize the initial fault handling data based on the rule link data, so as to realize the fault handling of the abnormal power distribution equipment based on the optimized fault handling data.

[0037] In this embodiment, the mismatch between the initial fault handling data and the rule chain data means that at least one action unit in the initial fault handling data does not meet the constraints defined by the compliance handling boundary. Optimizing the initial fault handling data in this embodiment refers to the process of correcting the mismatched initial fault handling data to make it meet the compliance handling boundary.

[0038] As one implementation method, an optimization process is initiated when non-compliant content is detected in the initial fault handling data. The optimization methods include any one or more combinations of the following: The first method is output rejection, i.e., directly deleting the non-compliant action unit without any replacement or supplementation. This method is suitable for scenarios where the non-compliant content cannot be corrected through simple replacement. The second method is nearest-neighbor compliance replacement, i.e., retrieving the compliant fragment from a pre-generated compliance template library that is closest in semantic intent to the current non-compliant action unit and fully satisfies the compliance constraints, and replacing the non-compliant content with this compliant fragment. The aforementioned compliance template library stores standard operating statements that have been pre-reviewed by experts and comply with power safety regulations, along with their corresponding structured action vectors. Nearest-neighbor replacement maps the non-compliant action unit and the compliance template to semantic vectors using a semantic encoding function, calculates the similarity between the vectors, and selects the compliance template with the highest similarity as the replacement content. The third method is constraint-based regeneration, i.e., adding the reason for the violation (including which specific safety constraint was violated) as additional information to the prompt context, guiding the large language model to regenerate the content of this step when the reason for the violation is clearly known. Regardless of the optimization method used, each optimized action unit must undergo compliance verification again to ensure that all action units meet all constraints of the compliance handling boundary before the optimized fault handling data can be output as the final result. Detailed logs of the verification and optimization process are also recorded, including what content was blocked, the specific reasons for the violation that triggered the blocking, and the optimization method used. This log information can be used for subsequent manual review and analysis.

[0039] By establishing a multi-layered mechanism to optimize non-compliant content, the system corrects non-compliant outputs into compliant handling suggestions without losing the rich information generated by semantic reasoning. This ensures the security of the final output while maximizing the value of the large language model in explaining fault causes and refining handling steps. Simultaneously, recording verification and optimization logs provides a data foundation for continuous system improvement.

[0040] This application achieves the determinism and security of rule-based reasoning, organically integrating it with the generalization and semantic understanding capabilities of semantic reasoning, through technical logic involving anomaly data identification, rule-based reasoning, semantic reasoning, compliance matching judgment, and optimization processing. The rule-based reasoning chain outputs compliance handling boundaries at the front end of the process, establishing an insurmountable safety barrier for semantic reasoning. Within this safety barrier, the semantic reasoning chain leverages the semantic understanding and generation capabilities of a large language model. The compliance matching judgment mechanism ensures that only handling solutions that fully meet safety constraints are output. The optimization processing mechanism corrects content when non-compliant situations arise. This approach overcomes both the high maintenance costs of pure rule-based reasoning when facing new and complex faults, and the semantic illusions and safety rule violations inherent in pure semantic reasoning. It achieves a fusion of safety compliance and intelligent generalization, improving the accuracy of fault reasoning in distribution networks.

[0041] In another embodiment of the power distribution network fault handling method provided by the present invention, step 100 specifically includes: Step 110: Based on the data source of real-time operation data of the distribution network, determine the judgment rule data of the distribution network; Step 120: When it is determined that the real-time operating data is abnormal based on the judgment rule data, the abnormal power distribution equipment of the power distribution network is determined based on the data source; Step 130: Determine the abnormal data of the abnormal power distribution equipment based on the real-time operating data of the abnormal power distribution equipment.

[0042] In this embodiment, the judgment rule data of the distribution network refers to a set of pre-defined rules used to determine whether the operating status of the distribution network is abnormal. Each judgment rule includes a judgment function, a trigger event category, and a trigger priority mapping. The data source refers to the source device that generates real-time operating data, including SCADA systems, PMS systems, distribution IoT platforms, etc.

[0043] As one implementation method, an intrinsic data interface is established with multiple data sources. Through a message bus and real-time subscription mechanism, it simultaneously receives telemetry and telecontrol data from the SCADA system, equipment ledgers and maintenance data from the PMS system, and sensor data from the power distribution IoT platform. For each data source, corresponding judgment rules are pre-configured. The judgment rules can differ between different data sources. For example, for the SCADA system, the judgment rules may include short-circuit tripping criteria (based on three-phase current, protection action signals, and switch position signals) and voltage over-limit criteria (based on voltage observations); for the power distribution IoT platform, the judgment rules may include oil temperature rising trend criteria (based on the derivative of temperature with respect to time and historical average temperature).

[0044] The system receives data streams from various data sources in real time and maps the received data to standard event objects. For each standard event object, it calls the judgment function in the judgment rule data corresponding to its data source for online detection. When an abnormal trigger condition is detected, the real-time running data is determined to be abnormal. At this time, the data source identifier in the standard event object is used to determine which data source the abnormal data comes from, and the abnormal power distribution equipment is determined based on the equipment identifier. In the process of determining abnormal power distribution equipment, since different data sources may identify the same equipment in different ways, an entity resolution function can be used to uniformly identify equipment across data sources. The entity resolution function performs joint resolution based on the point table mapping table, equipment coding rules, topology adjacency table, and geographic ledger. When there is one-to-many ambiguity, the maximum a posteriori probability identification method is used, combined with historical matching records, to determine the most likely equipment entity. After determining the abnormal power distribution equipment, the observations related to triggering the abnormality and the corresponding timestamps, quality codes, and other information are extracted from the real-time running data of the equipment as abnormal data.

[0045] This embodiment achieves real-time access and unified processing of multi-source heterogeneous data by establishing endogenous data interfaces with multiple data sources and a unified event standardization mechanism, thereby improving the comprehensiveness and timeliness of anomaly detection. Simultaneously, by configuring different judgment rule data for different data sources and employing the most appropriate anomaly criteria for different types of data characteristics, the accuracy of anomaly detection is improved. Furthermore, unified device identification across data sources is achieved through entity parsing functions, avoiding misidentification problems caused by inconsistent device identifiers between data sources.

[0046] See Figure 2 , Figure 2 This is a flowchart illustrating another embodiment of the power distribution network fault handling method provided by the present invention, as shown below. Figure 2 As shown, this embodiment includes steps 210 to 240, and the specific steps are as follows: Step 210: Perform pattern matching on the abnormal data to obtain the fault mode of the abnormal power distribution equipment; Step 220: Based on the fault mode, determine the fault topology range and adjustment constraint tuples of the abnormal power distribution equipment; Step 230: Determine the parameter adjustment range of the abnormal power distribution equipment based on the adjustment action elements in the adjustment constraint tuple; Step 240: Combine the fault mode, the fault topology range, the adjustment constraint tuple, and the parameter adjustment range to obtain the rule link data of the abnormal power distribution equipment.

[0047] In this embodiment, a fault mode refers to a standardized description of various distribution fault types predefined in the rule base. Each fault mode corresponds to a set of desired feature vectors and corresponding handling logic. The fault topology range refers to the electrical area related to the fault, determined based on the current power grid topology, including upstream power sources, downstream load branches, and related tie switches that may be affected by the fault. An adjustment constraint tuple refers to a set of structured rules that constrain fault handling operations. Each adjustment constraint tuple includes an operation type element, an operation object element, and a precondition element, representing a complete operation constraint. The parameter adjustment range refers to the upper and lower limits of allowed values ​​for relevant electrical parameters (e.g., transferred load and current setting) when performing operations such as load transfer and setting adjustment.

[0048] As one implementation method, pattern matching is performed between the observed value vectors in the abnormal data and the expected features of each fault mode in the rule base. Specifically, for each fault mode in the rule base, a weighted matching score is calculated: the actual values ​​of each dimension in the observed value vector are compared with the expected features of the fault mode. For each dimension, an equality judgment, an interval inclusion judgment, or a distance decay function is used as the matching function to obtain the matching result for that dimension. This result is then multiplied by the corresponding weight coefficient and accumulated to obtain the matching score for that fault mode. The fault mode with the highest matching score that exceeds a preset threshold is selected as the identification result of the current fault mode.

[0049] After determining the fault mode, the fault topology range is determined based on the current distribution network topology diagram. The topology diagram represents the electrical connections of the distribution network using equipment nodes and electrical connection edges, with each edge carrying switch status information. Starting from the protection-operated switch node, the system traverses along the closed electrical connection edges to determine the reachable path from the power supply side to the fault point and the downstream undervoltage area. By comparing the boundaries of the voltage-bearing and voltage-free areas, the fault candidate segment is located as the set of edges at the boundary, thus determining the fault topology range. Simultaneously, based on the fault mode and fault topology range, the corresponding adjustment constraint multi-group is retrieved from the rule base. Each adjustment constraint multi-group includes the operation type (e.g., closing, disconnecting, and switching), the operation object (e.g., specific switchgear or line), and preconditions (e.g., "confirm switch is in the open position" and "confirm fault section is isolated"). Furthermore, based on the operation type and operation object in the adjustment constraint multi-group, the parameter adjustment range is determined. For example, for load transfer operations, the upper limit of tie line capacity is obtained from structured knowledge data, and the allowable range of transferred load is determined to be greater than or equal to zero and less than or equal to the upper limit of tie line capacity. For protection setting adjustment operations, the allowable setting range of the protection device is obtained from structured knowledge data. Finally, the identified fault modes, the determined fault topology range, the retrieved adjustment constraint tuples, and the determined parameter adjustment range are structurally combined to form rule-based link data.

[0050] This embodiment achieves automatic identification of fault types through pattern matching, precise location of fault segments through topology tracing, and comprehensive constraints on subsequent operations through adjusting constraint tuples and parameter adjustment ranges. It provides structured, machine-readable rule constraint data for subsequent semantic reasoning, ensuring the integrity and accuracy of rule reasoning output.

[0051] In another embodiment of the power distribution network fault handling method provided by the present invention, step 300 specifically includes: Step 310: Input the rule link data and the abnormal data into the large language model; Step 320: Perform semantic reasoning on the abnormal data using the large language model to obtain the structured reasoning text and processing action vector output by the large language model; Step 330: Combine the structured reasoning text and the processing action vector to obtain the initial fault handling data of the abnormal power distribution equipment.

[0052] In this embodiment, the large language model refers to a language model based on the Transformer architecture and pre-trained on a large-scale text corpus, which possesses deep semantic understanding of natural language text and the ability to generate context-sensitive text. In this embodiment, the preferred large language model is an industry-specific large language model fine-tuned with power industry corpus to improve the accuracy of understanding terminology and scenarios in the power distribution field. The structured reasoning text refers to natural language text generated by the large language model according to a preset format, containing descriptions of fault cause analysis and handling steps, with a clear structure and distinct levels. The processing action vector refers to the vector representation obtained after parsing the action steps in the structured reasoning text into structured data; each processing action vector includes at least an operation type dimension, an operation object dimension, and a precondition dimension.

[0053] As one implementation method, after obtaining the rule link data, the rule link data, anomaly data, and previously loaded structured knowledge data are organized according to a preset prompt word template to construct a complete prompt context. The prompt word template includes at least the following slots: event summary (used to summarize the basic situation of the current anomaly event), device and topology context (used to provide structured information about the anomaly device and its associated devices), deterministic fault conclusion (used to clarify the fault type and fault segment determined by rule reasoning), inviolable security boundary (used to inform the model of the specific content of the compliance handling boundary in natural language), mandatory warning (used to emphasize the safety procedures that must be followed), and output format constraints (used to specify the structure and format of the model's output text).

[0054] The constructed contextual clues are input into the large language model to initiate the semantic reasoning process. Within an autoregressive decoding framework, the large language model progressively generates an output sequence based on the contextual clues. During generation, the model not only outputs natural language text but also embeds structured action markers into the text according to output format constraints. The complete text output by the model is parsed to extract the structured reasoning text (i.e., the main content describing the cause of the fault and the handling steps) and the processing action vectors (i.e., parsing the action steps in the text into structured operation type-operation object-precondition triples). The structured reasoning text and the processing action vectors together constitute the initial fault handling data.

[0055] This embodiment uses rule-based link data as the core input to the prompt context, enabling the large language model to fully consider the deterministic conclusions and security constraints of rule-based reasoning during the generation process, thereby reducing the risk of generating illegal content at the source. Simultaneously, through preset output format constraints, the model's output includes both readable text for operations and maintenance personnel and structured action vectors for machines, providing a foundation for subsequent compliance verification. Employing a large language model fine-tuned with power industry corpus further improves the model's accuracy in understanding power distribution terminology and scenarios.

[0056] In another embodiment of the power distribution network fault handling method provided by the present invention, step 500 specifically includes: Step 510: Verify the processing action vector based on the rule link data to obtain the verification result; Step 520: When the verification result indicates that the processing action vector does not match the rule link data, adjust the structured reasoning text and the processing action vector based on the verification result to optimize the initial fault processing data. Step 530: Based on the optimization results of the fault handling data, implement the fault handling of the abnormal power distribution equipment.

[0057] In this embodiment, the verification result is a binarized result used to indicate whether the processing action vector matches the rule link data, or a quantized result that includes the degree of mismatch and the specific location of the mismatch.

[0058] As one implementation method, after obtaining the initial fault handling data, compliance verification is performed on the processing action vectors. Specifically, each action unit in the processing action vector is compared one by one with the compliance handling boundaries in the rule link data. The compliance handling boundaries are represented in the form of a set of constraint inequalities, each constraint inequality corresponding to a safety procedure requirement. For example, for the constraint "prohibit closing the circuit with the ground wire connected," it is checked whether there are action units in the processing action vector with the operation type "closing" and the grounding status of the operation object being "grounded"; for the constraint "verify voltage before connecting the ground wire," it is checked whether there is a voltage verification action for the same operation object before the ground wire connecting action in the processing action vector. For each action unit, it is verified whether it satisfies all constraint inequalities. If all action units satisfy all constraints, the verification result is a match, and the initial fault handling data is directly output.

[0059] If at least one action unit fails to satisfy at least one constraint, the verification result is a mismatch, and the optimization process is initiated. In the optimization process, the specific location and reason for the mismatch are first determined, i.e., which action unit violated which security constraint. Then, the structured reasoning text and processing action vectors are adjusted based on the reason for the violation. Adjustment methods include any one or more combinations of the following: directly deleting the violating action unit; retrieving a compliant fragment from the compliance template library that is closest in semantic intent to the violating action unit and satisfies all constraints, and replacing it; regenerating the prompt context with the reason for the violation as additional information to guide the large language model to regenerate the content of this step. After adjustment, a complete compliance verification is performed again on the optimized processing action vectors to ensure that all action units satisfy all constraints of the compliance handling boundary. Only the optimized data that passes the complete compliance verification is output as the final fault handling data.

[0060] This embodiment establishes a closed-loop processing mechanism of verification, location, optimization, and re-verification to ensure that any non-compliant initial output can be accurately identified, located, and corrected. Furthermore, the corrected result must undergo re-verification before being output, thus guaranteeing a certain level of compliance in the final output. Simultaneously, by combining various optimization methods, compliance correction is achieved without sacrificing the value of semantic reasoning content, maximizing the preservation of the large language model's contribution to explaining fault causes and refining handling steps.

[0061] The distribution network fault handling system provided by the present invention is described below. The distribution network fault handling system described below can be referred to in correspondence with the distribution network fault handling method described above.

[0062] Please refer to Figure 3 The present invention also provides a power distribution network fault handling system, comprising: The abnormal data determination module 301 is used to determine the abnormal data of abnormal power distribution equipment based on the judgment results of the real-time operation data of the power distribution network. The rule link data determination module 302 is used to perform rule reasoning on the abnormal data and structured knowledge data of the abnormal power distribution equipment to obtain the rule link data of the abnormal power distribution equipment. The initial fault handling data determination module 303 is used to perform semantic reasoning on the abnormal data based on the rule link data to obtain the initial fault handling data of the abnormal power distribution equipment. The initial fault processing data verification module 304 is used to implement fault processing of the abnormal power distribution equipment based on the initial fault processing data when the initial fault processing data matches the rule link data. The power distribution network fault handling module 305 is used to optimize the initial fault handling data based on the rule link data when the initial fault handling data does not match the rule link data, so as to realize the fault handling of the abnormal power distribution equipment based on the optimized fault handling data.

[0063] Optionally, the abnormal data determination module includes: The judgment rule data determination unit is used to determine the judgment rule data of the distribution network based on the data source of real-time operation data of the distribution network; An abnormal power distribution equipment determination unit is used to determine the abnormal power distribution equipment of the power distribution network based on the data source when the real-time operating data is determined to be abnormal based on the judgment rule data. An abnormal data determination unit is used to determine the abnormal data of the abnormal power distribution equipment based on the real-time operating data of the abnormal power distribution equipment.

[0064] Optionally, the rule link data determination module includes: A fault mode determination unit is used to perform pattern matching on the abnormal data to obtain the fault mode of the abnormal power distribution equipment. The fault mode analysis unit is used to determine the fault topology range and adjustment constraint tuple of the abnormal power distribution equipment according to the fault mode. The parameter adjustment range determination unit is used to determine the parameter adjustment range of the abnormal power distribution equipment based on the adjustment action elements in the adjustment constraint tuple. The rule link data determination unit is used to combine the fault mode, the fault topology range, the adjustment constraint tuple, and the parameter adjustment range to obtain the rule link data of the abnormal power distribution equipment.

[0065] Optionally, the initial fault handling data determination module includes: The model input unit is used to input the rule link data and the abnormal data into the large language model; The semantic reasoning unit is used to perform semantic reasoning on the abnormal data through the large language model to obtain the structured reasoning text and processing action vector output by the large language model. The initial fault handling data determination unit is used to combine the structured reasoning text and the processing action vector to obtain the initial fault handling data of the abnormal power distribution equipment.

[0066] Optionally, the power distribution network fault handling module includes: The action vector verification unit is used to verify the action vector based on the rule link data and obtain the verification result. An initial fault handling data optimization unit is used to adjust the structured inference text and the processing action vector based on the verification result when the verification result shows that the processing action vector does not match the rule link data, so as to optimize the initial fault handling data. The power distribution network fault handling unit is used to handle the faults of the abnormal power distribution equipment based on the optimization results of the fault handling data.

[0067] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for handling faults in a power distribution network, characterized in that, include: Based on the judgment results of real-time operation data of the power distribution network, the abnormal data of abnormal power distribution equipment is determined; Rule-based reasoning is performed on the abnormal data and structured knowledge data of the abnormal power distribution equipment to obtain the rule link data of the abnormal power distribution equipment; Based on the rule link data, semantic reasoning is performed on the abnormal data to obtain the initial fault handling data of the abnormal power distribution equipment. When the initial fault handling data is matched with the rule link data, the fault handling of the abnormal power distribution equipment is realized based on the initial fault handling data. When the initial fault handling data does not match the rule link data, the initial fault handling data is optimized based on the rule link data so as to realize the fault handling of the abnormal power distribution equipment based on the optimized fault handling data.

2. The power distribution network fault handling method as described in claim 1, characterized in that, The judgment results based on the real-time operation data of the power distribution network determine the abnormal data of abnormal power distribution equipment, including: The data source is based on the real-time operation data of the distribution network, and the judgment rule data of the distribution network is determined. When the real-time operating data is determined to be abnormal based on the judgment rule data, the abnormal power distribution equipment of the power distribution network is determined based on the data source; Based on the real-time operating data of the abnormal power distribution equipment, the abnormal data of the abnormal power distribution equipment is determined.

3. The power distribution network fault handling method as described in claim 1, characterized in that, The rule-based reasoning process performed on the abnormal data and structured knowledge data of the abnormal power distribution equipment to obtain the rule link data of the abnormal power distribution equipment includes: Pattern matching is performed on the abnormal data to obtain the fault mode of the abnormal power distribution equipment; Based on the fault mode, determine the fault topology range and adjustment constraint tuples of the abnormal power distribution equipment; Based on the adjustment action elements in the adjustment constraint tuple, the parameter adjustment range of the abnormal power distribution equipment is determined; The fault mode, the fault topology range, the adjustment constraint tuple, and the parameter adjustment range are combined to obtain the rule link data of the abnormal power distribution equipment.

4. The power distribution network fault handling method as described in claim 1, characterized in that, The step of performing semantic reasoning on the abnormal data based on the rule link data to obtain the initial fault handling data for the abnormal power distribution equipment includes: Input the rule link data and the anomaly data into the large language model; The abnormal data is semantically reasoned using the large language model to obtain the structured reasoning text and processing action vectors output by the large language model. The structured reasoning text and the processing action vector are combined to obtain the initial fault handling data of the abnormal power distribution equipment.

5. The power distribution network fault handling method as described in claim 4, characterized in that, The step of optimizing the initial fault handling data based on the rule link data when the initial fault handling data does not match the rule link data, so as to realize the fault handling of the abnormal power distribution equipment based on the optimized fault handling data, includes: The processing action vector is validated based on the rule link data to obtain the validation result; When the verification result indicates that the processing action vector does not match the rule link data, the structured reasoning text and the processing action vector are adjusted based on the verification result to optimize the initial fault processing data. Based on the optimization results of the fault handling data, the fault handling of the abnormal power distribution equipment is realized.

6. A power distribution network fault handling system, characterized in that, include: The abnormal data determination module is used to determine abnormal data of abnormal power distribution equipment based on the judgment results of real-time operation data of the power distribution network. The rule link data determination module is used to perform rule reasoning on the abnormal data and structured knowledge data of the abnormal power distribution equipment to obtain the rule link data of the abnormal power distribution equipment. The initial fault handling data determination module is used to perform semantic reasoning on the abnormal data based on the rule link data to obtain the initial fault handling data of the abnormal power distribution equipment. An initial fault processing data verification module is used to implement fault processing of the abnormal power distribution equipment based on the initial fault processing data when the initial fault processing data matches the rule link data. The power distribution network fault handling module is used to optimize the initial fault handling data based on the rule link data when the initial fault handling data does not match the rule link data, so as to realize the fault handling of the abnormal power distribution equipment based on the optimized fault handling data.

7. The power distribution network fault handling system as described in claim 6, characterized in that, The abnormal data determination module includes: The judgment rule data determination unit is used to determine the judgment rule data of the distribution network based on the data source of real-time operation data of the distribution network; An abnormal power distribution equipment determination unit is used to determine the abnormal power distribution equipment of the power distribution network based on the data source when the real-time operating data is determined to be abnormal based on the judgment rule data. An abnormal data determination unit is used to determine the abnormal data of the abnormal power distribution equipment based on the real-time operating data of the abnormal power distribution equipment.

8. The power distribution network fault handling system as described in claim 6, characterized in that, The rule link data determination module includes: A fault mode determination unit is used to perform pattern matching on the abnormal data to obtain the fault mode of the abnormal power distribution equipment. The fault mode analysis unit is used to determine the fault topology range and adjustment constraint tuple of the abnormal power distribution equipment according to the fault mode. The parameter adjustment range determination unit is used to determine the parameter adjustment range of the abnormal power distribution equipment based on the adjustment action elements in the adjustment constraint tuple. The rule link data determination unit is used to combine the fault mode, the fault topology range, the adjustment constraint tuple, and the parameter adjustment range to obtain the rule link data of the abnormal power distribution equipment.

9. The power distribution network fault handling system as described in claim 6, characterized in that, The initial fault handling data determination module includes: The model input unit is used to input the rule link data and the abnormal data into the large language model; The semantic reasoning unit is used to perform semantic reasoning on the abnormal data through the large language model to obtain the structured reasoning text and processing action vector output by the large language model. The initial fault handling data determination unit is used to combine the structured reasoning text and the processing action vector to obtain the initial fault handling data of the abnormal power distribution equipment.

10. The power distribution network fault handling system as described in claim 9, characterized in that, The power distribution network fault handling module includes: The action vector verification unit is used to verify the action vector based on the rule link data and obtain the verification result. An initial fault handling data optimization unit is used to adjust the structured inference text and the processing action vector based on the verification result when the verification result shows that the processing action vector does not match the rule link data, so as to optimize the initial fault handling data. The power distribution network fault handling unit is used to handle the faults of the abnormal power distribution equipment based on the optimization results of the fault handling data.

Images