Power grid accident automatic submission method and system based on AI semantic understanding and event chain tracking

Abstract

The invention provides a power grid accident automatic submission method and system based on AI semantic understanding and event chain tracking, and relates to the technical field of intelligent power grid monitoring. The method comprises the following steps: constructing an event chain model through event time alignment and equipment topological association, and deducing a causal relationship between events to generate an electrical law causal graph; identifying a root cause event in combination with the in-degree, out-degree and confidence weight of the node; and automatically judging and generating an accident submission text and an evidence packet according to the accident severity and the confidence threshold. According to the invention, intelligent diagnosis and automatic submission of electrical accidents can be realized, and the analysis accuracy and response efficiency are improved. The problem that in the prior art, cross-system time alignment and automatic power grid accident submission based on causal constraints cannot be achieved under the multi-source heterogeneous data condition is solved.

Description

Technical Field

[0001] This invention relates to the field of smart grid monitoring technology, and in particular to a method and system for automatic reporting of power grid accidents based on AI semantic understanding and event chain tracing. Background Technology

[0002] In current power system monitoring and dispatching, multiple information sources, such as operation monitoring systems, protection and fault recording systems, operation management systems, field communication records, and maintenance planning systems, operate in parallel, generating a large amount of structured and unstructured multi-source data. When an anomaly or accident occurs in the power grid, on-duty personnel need to retrieve records from multiple systems, compare signals, analyze the cause, and manually generate accident reporting materials. The entire process relies on manual experience, resulting in low efficiency and susceptibility to personal judgment, leading to delays and inconsistencies in accident information reporting.

[0003] In recent years, with the development of artificial intelligence and big data technologies, the industry has begun to shift from alarm compression and templated reports based on fixed rules to intelligent analysis based on semantic understanding and causal inference. Some research attempts to unify the event expression forms of different systems using semantic recognition technology and introduce event chain analysis models based on topological logic relationships to achieve intelligent analysis and report generation of content such as accident causes, correctness of actions, and scope of impact. At the same time, to meet compliance and audit requirements, the power grid automatic reporting system is evolving towards data traceability, verifiable evidence, and closed-loop learning.

[0004] However, existing technologies still have significant shortcomings in multi-system time synchronization and causal chain reasoning. Timestamps in different systems drift and are delayed, and the impact of time uncertainty on accident assessment cannot be accurately quantified; event correlation mainly relies on static rules or experience templates, making it difficult to balance physical topology constraints and temporal logical order; automatic reporting functions generally lack triggering strategies based on dynamic judgment of severity and confidence levels, and also lack bidirectional consistency verification between field reports and system judgments, resulting in fragmented evidence chains that are not effectively auditable. These problems collectively lead to insufficient accuracy, timeliness, and traceability in power grid accident reporting. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, the purpose of this invention is to provide an automatic reporting method and system for power grid accidents based on AI semantic understanding and event chain tracing. This invention solves the problem that existing technologies cannot achieve cross-system time alignment and automatic reporting of power grid accidents based on causal constraints under multi-source heterogeneous data conditions.

[0006] To achieve the above objectives, the present invention provides the following solution:

[0007] An automatic reporting method for power grid accidents based on AI semantic understanding and event chain tracing includes:

[0008] Acquire multi-source data from the power grid and perform semantic parsing on the multi-source data to obtain event types, time clues, and equipment entity information;

[0009] Based on the device vocabulary and topological knowledge, the event types, time clues and device entity information are normalized and hierarchically mapped to obtain the initial event model dataset;

[0010] Based on the synchronization anchor points in the multi-source data, the time information from different sources in the initial event model dataset is drifted and aligned, and a confidence interval is established for the aligned time information to obtain a time-constrained event set after cross-system time-scale alignment.

[0011] Based on the time sequence of the time-constrained event set and the topological relationships between electrical devices, the events are sorted and aggregated to obtain an event chain model;

[0012] Based on time and topological constraints, causal inferences are made about the relationships between event chain models to obtain an electrical law causal graph.

[0013] The root cause analysis results data are determined based on the aforementioned electrical law cause-effect diagram;

[0014] The root cause analysis results are automatically triggered to make judgments based on the severity of the accident and the confidence threshold, resulting in the accident report text and the corresponding evidence package.

[0015] An automatic power grid accident reporting system based on AI semantic understanding and event chain tracing includes:

[0016] The multi-source data acquisition module is used to acquire multi-source data from the power grid and perform semantic parsing on the multi-source data to obtain event types, time clues, and equipment entity information;

[0017] The device vocabulary matching and hierarchical mapping module is used to normalize and hierarchically map the event types, time clues and device entity information based on the device vocabulary and topological knowledge to obtain the initial event model dataset.

[0018] The time alignment and confidence modeling module is used to perform drift estimation and alignment processing on time information from different sources in the initial event model dataset based on the synchronization anchor points in the multi-source data, and to establish confidence intervals for the aligned time information to obtain a time-constrained event set after cross-system timescale alignment.

[0019] The event chain construction module is used to sort and aggregate events based on the time sequence of the time-constrained event set and the topological relationship between electrical equipment to obtain the event chain model;

[0020] The causal inference module is used to perform causal inference on the correlation between event chain models based on time constraints and topological constraints, and to obtain an electrical law causal graph.

[0021] The root cause analysis module is used to determine the root cause analysis result data based on the electrical law cause-effect diagram;

[0022] The automatic reporting generation module is used to automatically trigger judgment on the root cause analysis results data based on the severity of the accident and the confidence threshold, and obtain the accident reporting text and the corresponding evidence package.

[0023] The present invention discloses the following technical effects:

[0024] This invention provides an automatic reporting method and system for power grid accidents based on AI semantic understanding and event chain tracing. The method includes: acquiring multi-source data from the power grid and performing semantic parsing on the multi-source data to obtain event types, time clues, and equipment entity information; normalizing and hierarchically mapping the event types, time clues, and equipment entity information according to an equipment vocabulary and topological knowledge to obtain an initial event model dataset; performing drift estimation and alignment processing on time information from different sources in the initial event model dataset based on synchronization anchors in the multi-source data, and establishing confidence intervals for the aligned time information to obtain a time-constrained event set after cross-system time-scale alignment; sorting and aggregating events according to the time sequence in the time-constrained event set and the topological relationships between electrical equipment to obtain an event chain model; performing causal inference on the relationships between the event chain models based on time constraints and topological constraints to obtain an electrical law causal graph; determining root cause analysis result data based on the electrical law causal graph; and automatically triggering judgment on the root cause analysis result data based on accident severity and confidence thresholds to obtain an accident reporting text and corresponding evidence package. This invention addresses the problems of isolated multi-source data, misaligned time information, and reliance on manual judgment for event correlation in existing power grid accident reporting. It proposes an automated reporting method based on AI semantic understanding and event chain tracing. Through semantic parsing, topology hierarchical mapping, and time alignment mechanisms, it achieves unified representation and synchronization constraints of data from multiple systems including monitoring, protection, management, and communication. Combining an event chain model and an electrical law causal graph, it can automatically identify the accident evolution path and root cause, eliminating reliance on manual experience. Furthermore, by triggering severity and confidence thresholds, it achieves automated reporting of accident information and evidence packages. This solution effectively improves the accuracy and real-time performance of accident analysis and reporting, significantly reduces the workload of manual intervention, and enhances the intelligence level of power grid operation monitoring. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 The flowchart illustrates an automatic reporting method for power grid accidents based on AI semantic understanding and event chain tracing, as provided in this embodiment of the invention. Detailed Implementation

[0027] 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. 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.

[0028] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0029] like Figure 1 As shown, this invention provides an automatic reporting method for power grid accidents based on AI semantic understanding and event chain tracing, including:

[0030] Step 100: Obtain multi-source data from the power grid and perform semantic parsing on the multi-source data to obtain event types, time clues, and equipment entity information;

[0031] Step 200: Based on the device vocabulary and topological knowledge, normalize and hierarchically map the event types, time clues and device entity information to obtain the initial event model dataset;

[0032] Step 300: Based on the synchronization anchor points in the multi-source data, perform drift estimation and alignment processing on the time information from different sources in the initial event model dataset, and establish confidence intervals for the aligned time information to obtain a time-constrained event set after cross-system time-scale alignment;

[0033] Step 400: Based on the time sequence of the time-constrained event set and the topological relationships between electrical devices, sort and aggregate the events to obtain the event chain model;

[0034] Step 500: Based on time constraints and topological constraints, perform causal inference on the correlation between event chain models to obtain the electrical law causal graph;

[0035] Step 600: Determine the root cause analysis results data based on the electrical law cause-effect diagram;

[0036] Step 700: Automatically trigger judgment on the root cause analysis results data based on the severity of the accident and the confidence threshold to obtain the accident report text and the corresponding evidence package.

[0037] Furthermore, the specific implementation process of step 100 is as follows:

[0038] This embodiment first acquires multi-source data from the operation monitoring system, protection waveform recording device, operation management platform, and field communication channels. The acquired data undergoes format recognition, time stamp extraction, and abnormal noise filtering to eliminate communication errors and invalid signal information, and the timestamp field is simultaneously added to form a unified data format. For records where the original signal lacks complete identification, this embodiment uses the source port number, device number, and communication channel identifier for traceability matching, ensuring consistency in physical and temporal attributes of the multi-source data, forming a parsable pre-processed data set.

[0039] This embodiment performs word segmentation, part-of-speech tagging, and named entity recognition on text data in the preprocessed multi-source dataset. It then uses a semantic template library to identify verbs, equipment nouns, and time expressions related to the accident event. Simultaneously, it extracts alarm tags, timestamps, and measurement sequence features from signal data, and utilizes signal mutation detection and interval trend analysis to extract key features representing electrical disturbances. The text semantic results and signal features are aligned and fused along the time dimension to generate a unified data representation that simultaneously includes semantic pragmatic features and dynamic signal features, providing consistent input for subsequent event element extraction.

[0040] This embodiment, based on a unified data representation, invokes a semantic rule base and combines contextual features for semantic classification, identifying semantic elements such as event type candidates, timeline fragments, and device entity designations. Then, it performs homonymy normalization and number mapping on device entity designations, ensuring that the same device has a unique identifier when appearing in different data sources. After structured processing, an event semantic element set is obtained. Finally, based on semantic association rules between elements, event types, timelines, and device entity information are extracted, achieving unified extraction and structured output of multi-source heterogeneous power grid data at the semantic level.

[0041] Furthermore, the specific implementation process of step 200 is as follows:

[0042] This embodiment first invokes a preset equipment vocabulary and power grid topology knowledge base to perform vocabulary matching and node localization processing on the equipment entities involved in the event semantic element set. During vocabulary matching, this embodiment identifies standard equipment entries corresponding to the equipment entities in the semantic elements by comparing equipment standard naming conventions, abbreviation libraries, and operation numbering rules. When non-standard or colloquial expressions appear in the text, synonym merging is performed based on word vector similarity and equipment function tags to improve recognition accuracy. In the node localization stage, based on the equipment node numbers and connection path information in the topology knowledge base, the matched equipment entities are located to specific node positions in the primary or secondary network of the power grid, forming an event element set containing node coordinates and equipment hierarchical attributes.

[0043] This embodiment performs unified normalization on different naming or numbering formats of the same device in the event element set. During normalization, based on standardized naming rules, multiple identifiers of the same physical device appearing in different systems are uniformly mapped, prioritizing the retention of the main network code or standard operating number, and constructing a device entity mapping table. The device nodes in the mapping table are deduplicated and their consistency is verified to ensure the consistency of topological relationships across the entire network. Then, based on the normalization results and the structural hierarchy of primary and secondary equipment in the power grid, the subordinate relationships, connection relationships, and signal transmission paths between devices are analyzed. Based on this, an electrical hierarchy structure model is established to reflect the hierarchical and interconnection attributes of devices within the system.

[0044] This embodiment, based on the electrical hierarchy model, binds event types, timelines, and device entity information to corresponding device nodes, thereby forming a structured event record set after hierarchical mapping. For each event in this record set, this embodiment generates a unique index number and establishes a number management table to track the correspondence between events at different levels and times. Finally, by verifying and synchronizing the association indexes between event nodes, an initial event model dataset containing standardized device information, time associations, and hierarchical relationships is constructed, providing a standardized data foundation for subsequent event timescale alignment and causal inference.

[0045] Specifically, this embodiment calculates the electrical hierarchy model by modeling the topology and device attributes of power grid equipment nodes. Specifically, this embodiment first extracts all equipment nodes and their connections from the power grid topology knowledge base, and obtains the shortest path information between nodes using a graph structure algorithm to characterize the transmission depth between devices at the physical connection level. For each pair of equipment nodes, this embodiment determines its electrical connectivity coefficient based on the electrical connection attributes in the topology knowledge base, reflecting the accessibility of energy or control signal transmission between nodes. Subsequently, this embodiment extracts the category code values ​​for each device; for example, main transformers are coded as one type of device, circuit breakers as another, and communication and protection devices correspond to different category numbers. These category code values ​​reflect the functional hierarchy differences of the devices in the electrical system.

[0046] In constructing the model, this embodiment multiplies the electrical connectivity coefficients of all node pairs by the corresponding node category codes to obtain the direct association strength between each node pair. Then, it multiplies this by a topology attenuation coefficient based on the path distance between nodes to describe the diminishing influence effect between devices as the topology distance increases. By summing the weights of all node pairs, a comprehensive matrix representing the entire electrical network hierarchy is obtained, thus reflecting the comprehensive association degree of different levels and types of devices in the topology. This matrix, as the expression of the electrical hierarchy model, provides a quantitative basis for the hierarchical mapping between events and devices.

[0047] The electrical connectivity coefficients are derived from the connection table in the topology knowledge base and can be values ​​between zero and one, representing the strength of electrical interactions. Equipment category codes are generated based on the equipment standard type table and are discrete integer numbers used to map the hierarchical attributes of different equipment categories. The path distance is the shortest path length between equipment nodes, obtained through a distance calculation module based on a graph search algorithm. The average hierarchical scale constant is taken as the average path length of the topology network and is used to control the scale of the attenuation term, ensuring consistent attenuation across networks of different sizes. For example, in a four-layer network including substation buses, main transformers, feeders, and protection devices, nodes can be assigned electrical connectivity and category codes, their topological distances and average hierarchical constants can be calculated, and a weighted sum can be obtained to obtain the overall structural representation of the electrical interactions between each layer. This calculation process requires no manual adjustment of weight parameters; all values ​​are derived from the power grid topology and equipment standard information database, ensuring the model's repeatability and accuracy.

[0048] More specifically, this embodiment instantiates and illustrates the parameters of the electrical hierarchy model in a substation network containing several main equipment nodes to ensure sufficient disclosure and ease of understanding of its calculation process. Assume the network consists of 5 equipment nodes: main transformer T1, busbar B1, circuit breaker Q1, line L1, and protection device P1. The total number of equipment nodes N in the power grid is then 5.

[0049] In this network, the electrical connectivity coefficients between nodes can be determined based on the topology knowledge base. The coefficient is higher when two devices are directly connected, and decreases according to the path hierarchy when they are indirectly connected. For example, the main transformer T1 is directly connected to bus B1, with a connectivity coefficient R of 1.0; bus B1 is directly connected to circuit breaker Q1, with a coefficient of 0.9; the main transformer T1 and line L1 are indirectly connected via the bus and circuit breaker, with a connectivity coefficient of 0.7; and the protection device P1 is connected to the circuit breaker via a signal path, with an electrical connectivity coefficient of 0.5.

[0050] Equipment category codes are assigned integer values ​​based on the functional level of the equipment, used to distinguish the hierarchical attributes of different types of equipment in the model. For example, the main transformer T1, as a primary equipment main component, is assigned a category code of 3; busbar B1, as a central aggregation node, is assigned a value of 2; circuit breaker Q1, as a control device, is assigned a value of 1; line L1, as a transmission channel, is assigned a value of 2; and protection device P1, as a secondary detection device, is assigned a value of 1. The shortest path distance between nodes is represented by the result calculated by a graph search algorithm. The distance between directly connected devices is 1, and the distance is 2 if an intermediate node is required. For example, the path distance d between T1 and B1 is 1, the path distance between T1 and Q1 is 2, and the path distance between T1 and L1 is 3. The average hierarchical scale constant L of the network structure is calculated based on the average path length of the entire network, and is taken as 2 in this example.

[0051] When calculating the exponential decay relationship between T1 and L1, the negative exponent of the exponential term 'e' in the formula is equal to the ratio of the path distance to the hierarchical constant, i.e., 3 divided by 2 equals 1.5, corresponding to an exponential value of approximately 0.22, indicating that the electrical influence between the two is attenuated to about 22%. Meanwhile, the exponential value corresponding to the path distance 1 between T1 and B1 is approximately 0.61, indicating a closer electrical relationship between directly connected devices. Therefore, by using the specific values ​​of the connectivity coefficients of each node, category codes, path distances, and the average hierarchical scale constant, a specific hierarchical correlation matrix can be calculated and summarized, comprehensively reflecting the strength and attenuation characteristics of the electrical effects between devices at different levels, thus achieving a quantitative expression of the electrical hierarchical structure model.

[0052] Furthermore, the specific implementation process of step 300 is as follows:

[0053] This embodiment first extracts time attribute information from different sources from the initial event model dataset, and then groups and organizes it according to the type of data source to form a source-specific time attribute set. The data sources include operation monitoring systems, protection waveform recording devices, operation management platforms, and field communication terminals, etc., and each type of data source records its own timestamp during sampling or reporting. This embodiment standardizes the timestamp format, unifying all times expressed in different time systems into Coordinated Universal Time (UTC). To ensure the comparability of times from different data sources at the semantic event level, this embodiment records the original sampling period, trigger identifier, and trigger channel number during the extraction process to support subsequent time alignment calculations.

[0054] This embodiment then identifies synchronization anchor points for the source time attribute set, extracting trigger signals with common event identifiers or simultaneous responses from each data source as time alignment references. Synchronization anchor points may include relay protection action signals, circuit breaker trip commands, and message time identifiers from communication channels. Based on event name matching rules and signal waveform similarity calculation results, this embodiment automatically identifies corresponding trigger points between different data sources and establishes a synchronization anchor point matching relationship table, thereby forming a time anchor point set. The anchor point set not only contains matching event numbers but also records the trigger times of each source and their time differences, providing raw supporting data for cross-source drift estimation.

[0055] This embodiment further calculates the drift amount and drift trend of each data source relative to the reference time source based on the time anchor point set, forming a cross-source time drift parameter set, and corrects the time attributes of the initial event model dataset accordingly. The reference time source can be GPS synchronization data or a master control source with certain stability, using its time as a benchmark to perform linear drift compensation or moving average adjustment on the times of other data sources. After correction, this embodiment calculates the confidence interval of each event time based on the variance of historical drift errors to quantify the time alignment accuracy. The confidence interval indicates the error range that the actual occurrence time of the event may fall into. This embodiment stores the confidence level and the adjusted time together to generate a time-calibrated event set with time confidence information. Finally, a unified index and formatting management is performed on the time field and confidence field in the time-calibrated event set to obtain a time-constrained event set after cross-system time-scale alignment, providing accurate time-series input for subsequent event chain model construction and causal inference.

[0056] Furthermore, the specific implementation process of step 400 is as follows:

[0057] This embodiment constructs an event chain model by comprehensively analyzing events in a time-constrained event set according to their chronological order and device topology associations. Specifically, this embodiment first performs preliminary sorting of events based on the timestamp information of each event in the time-constrained event set to ensure the accuracy of the event arrangement order on the timeline. Subsequently, based on the device topology relationships established in the electrical hierarchy model, it determines whether there are direct or indirect physical relationships between the corresponding devices of the events, and generates a topology connection index between events based on the connection paths and action links of the device nodes. This index is used to determine whether two events may have a causal triggering relationship. When the corresponding devices are in the same loop or have control logic associations, the index value is higher; if the event association path is long or they are in different electrical areas, the index value is correspondingly lower.

[0058] In constructing the event chain model, this embodiment further calculates the temporal sequence relationship and time difference between events. The temporal sequence symbol is used to identify the sequential relationship between events; a positive symbol is assigned when an event is earlier than another event, and a negative symbol otherwise. The time difference is the timestamp difference between two events, which can be in milliseconds or seconds, and is used to measure the actual time interval between events. Combining the above topology connectivity indicators and temporal sequence symbols, this embodiment establishes corresponding association weights for any two events. Since the triggering effect between events usually gradually weakens as the time interval increases, this embodiment sets a time decay constant to control the magnitude of this decrease. The value of the time decay constant can be set according to the event type and system response characteristics. For example, in events involving transformer protection or tripping actions, this constant can be selected as 0.5 seconds; in communication events with large signal delays, it can be taken as 1 second, to reflect the difference in the duration of event impact under different behavioral mechanisms.

[0059] In the specific calculation, this embodiment traverses all event pairs in the time-constrained event set, multiplies the topological connectivity index and time sequence symbol of each event pair by an exponential term reflecting the time decay effect, and then sums them to obtain the overall correlation strength matrix of the event chain model. This matrix reflects the comprehensive influence between events in the entire event sequence and can characterize the laws of event propagation from front to back, causal triggering, and influence transmission. The topological connectivity index is derived from the node connectivity matrix in the electrical hierarchy, the time sequence symbol is obtained through timestamp comparison, the time difference is directly calculated from the event's time attribute, and the time decay constant is preset as a model control parameter in the system operation configuration. Through this calculation process, an event chain model representing cross-time and cross-device events is finally formed to support subsequent event causal analysis and risk propagation inference.

[0060] Specifically, to facilitate understanding of the construction principle of the event chain model in this embodiment, a set of specific examples are used to illustrate each parameter in the model, and corresponding numerical values ​​are given to ensure full disclosure and clear calculation process. Assume that during the operation of a substation in a certain area, a total of four typical events are recorded in the time-constrained event set: Event E1 (circuit breaker Q1 trips), Event E2 (line L1 protection starts), Event E3 (main transformer T1 current exceeds limit alarm), and Event E4 (communication equipment P1 sends protection action signal). The total number of events, M, is 4.

[0061] Based on the electrical hierarchy model, the topological connections between the devices corresponding to these events can be determined. Main transformer T1 is directly connected to circuit breaker Q1, with a connection index S of 1.0; line L1 is indirectly connected to circuit breaker Q1 via a busbar, with an index of 0.8; communication device P1 is indirectly connected to the protection device via a signal link, with an index of 0.6; the connection between main transformer T1 and communication device P1 is relatively weak, with an index of 0.3. This index reflects the strength of the electrical and signalal connection between the devices to which the event belongs, and is a real number ranging from 0 to 1.

[0062] The time sequence symbol is used to identify the chronological order of events. If event m occurs before event n, the symbol is +1; if event m occurs after event n, the symbol is -1. For example, in this instance, circuit breaker Q1 trips (E1) before line L1 protection starts (E2), so the time sequence symbol T is +1; while the communication device P1 uploads the action signal (E4) after the protection start event (E2), so T is -1.

[0063] The time difference measures the time interval between events in milliseconds. For example, the time difference Δt between E1 and E2 is 200 milliseconds, between E2 and E3 is 500 milliseconds, and between E3 and E4 is 800 milliseconds. The time decay constant τ represents the rate attenuation of the triggering effect between events over time, and its value is set according to the typical response characteristics of the electrical system. In this example, τ is taken as 400 milliseconds. Based on this setting, the numerical calculation result of the second exponential term can reflect the degree of decay of the event's effect. For example, when the time difference is 200 milliseconds, the exponential term is approximately 0.61, indicating that the effect decays by about 39%; when the time difference is 800 milliseconds, the exponential term drops to 0.14, indicating that the triggering effect strength remains at only about 14%.

[0064] By combining the above parameters, the association strength matrix of the event chain model can be obtained. The results show that events that are temporally adjacent and topologically close (such as E1 and E2) have high association strength values ​​(approximately 0.49 to 0.55), while events that are spaced further apart and have weaker connections (such as E1 and E4) have association strength values ​​below 0.05. This example clearly demonstrates the practical meaning, numerical source, and role of each parameter in the event chain model in the model calculation, enabling this embodiment to be accurately reproduced and implemented by those skilled in the art.

[0065] Furthermore, the specific implementation process of step 500 is as follows:

[0066] This embodiment first extracts the temporal sequence information of event nodes and the attributes of their respective device topology nodes from the event chain model, forming an event chain input set containing both temporal and topological constraints. In this process, based on the aforementioned time-constrained event set aligned with time scales, this embodiment obtains the precise occurrence time and confidence interval of each event, and simultaneously extracts the topology node number, connection path, and hierarchical position of the device to which the event belongs from the electrical hierarchy model. By sorting the events by time and matching them with their device topology information, a comprehensive input set that reflects both the temporal sequence of events and describes the physical relationships between devices is constructed, ensuring that subsequent causal inference has dual constraints.

[0067] This embodiment then identifies the pre- and post-event triggering relationships and logical connectivity relationships between events based on time constraints and topological constraints, respectively. Time constraint identification is achieved by comparing the chronological order of events and the degree of overlap in their confidence intervals. When the chronological order of two events is clear and the interval is within the allowable range of the system response characteristics, a possible triggering relationship is determined. Topological constraint identification determines the logical adjacency of events in space based on the connection paths of device nodes, hierarchical dependencies, and the direction of control signals. This embodiment establishes a set of event association pairs, each pair containing its pre- and post-event temporal relationship and topological connection status, thereby uniformly expressing the constraints of temporal causality and device connectivity, providing accurate input for subsequent causal relationship modeling.

[0068] This embodiment further calculates the causal correlation strength and directionality index for each event pair based on the set of event association pairs, and corrects them using time confidence intervals and topological path constraints to obtain a set of effective causal relationships. The causal correlation strength reflects the probability of triggering between events and can be calculated based on time difference, topological connection weights, and historical event co-occurrence frequency. The directionality index is used to identify the direction of causal propagation; when an upstream device event triggers a downstream device response, the direction is indicated as positive, otherwise as negative. This embodiment uses the corrected effective causal relationships as directed edges and event nodes as vertices to construct an electrical law causal graph. This causal graph can intuitively present the causal transmission paths of different electrical events in both time and topological dimensions, providing verifiable knowledge-based model support for power grid operation anomaly analysis, protection action tracing, and automated diagnosis.

[0069] Specifically, this embodiment first extracts the time interval, event type label, and corresponding device node information for each event pair from the event association pair set, constructing an event association feature set for causal analysis. The time interval is calculated from the timestamps of the two events, reflecting the chronological order and relative time interval between events; the event type label is encoded according to the event category (e.g., alarm, protection action, operation command); the device node information is taken from the electrical hierarchy model, including device node number, hierarchy attribute, and connection path information. This embodiment, through the combined description of the above three types of elements, gives each event pair sufficient feature expression dimensions, providing a structured data foundation for subsequent calculation of causal strength and direction.

[0070] This embodiment then calculates the probability of event sequence based on the trigger time difference, sampling period, and timeline arrangement. Under conditions of short time intervals and consistent sampling periods, the probability of event sequence is higher, indicating a stronger likelihood of the event being triggered. When the time confidence intervals of events partially overlap, this embodiment corrects the probability value using a confidence weighting factor to reduce the risk of misjudgment. The corrected index forms the sequence measurement result, used to measure the credibility of the causal order of two events in the time dimension. This embodiment also calls the electrical hierarchy model to calculate the topological influence index of upstream events on downstream events. The topological influence is derived from the connection type between devices, energy flow, or signal transmission path. When upstream devices directly control or power downstream devices in the topology, this index takes a high value to reflect the dominant role at the topological level, thus obtaining the topological influence result.

[0071] Based on this, this embodiment normalizes and synthesizes the sequential measurement results and topological influence results according to a preset weight model to form a set of causal relationships between events. The weight model is determined by empirical parameters or training results, typically setting the weight of the time factor to 0.6 and the weight of the topological factor to 0.4 to balance the contribution of the two constraints to the causal judgment. After normalization and synthesis, each pair of events obtains a causal relationship strength value between 0 and 1, with higher values ​​indicating a stronger causal relationship. This embodiment further determines the causal direction of events based on the temporal relationship and a preset causal strength threshold. When the relationship strength is higher than 0.5 and the upstream event is earlier than the downstream event, a positive causal direction is marked; if the strength is lower than the threshold or the time order is reversed, it is marked as a reverse or invalid relationship. This yields a set of candidate causal relationships containing causal direction labels, providing fundamental data support for constructing the final electrical law causal graph.

[0072] Furthermore, the specific implementation process of step 600 is as follows:

[0073] This embodiment first reads the attribute information of event nodes and causal edges from the electrical cause-effect graph to construct a causal node feature set. The attribute information includes the node's in-degree, out-degree, event type code, time confidence weight, and topological level position. The in-degree represents the number of times an event is triggered by other events, and the out-degree represents the number of times the event triggers other events; nodes with lower in-degree and higher out-degree typically have stronger triggering characteristics. The event type code is graded according to the nature of the event; for example, a primary equipment failure event is assigned a value of 3, a protection device action event is assigned a value of 2, and an alarm or communication upload event is assigned a value of 1. The time confidence weight is calculated from the time-constrained event set and is used to measure the reliability of the event's timing. By extracting the above multi-dimensional attribute information, this embodiment forms a causal node feature set that includes event topological relationships, time reliability, and logical impact, providing a data foundation for subsequent root cause identification.

[0074] This embodiment then traverses and identifies event chains along the causal direction based on the causal node feature set. By tracing the event propagation path step by step along the directed edges starting from the starting node of the causal graph, it detects terminal event groups that are no longer triggered by other events in terms of temporal sequence and topological propagation direction. Such event groups typically represent potential root cause triggering sources. If an event node has an in-degree of 0 or is the earliest occurring in time and its device is located in the upstream region of the power supply or control link, it is marked as a candidate root cause event. To avoid false causal interference, this embodiment combines time confidence weights and topological path constraints during the traversal process to automatically exclude events with temporal overlap or indirect connections, ensuring that the identified candidate root cause event set retains only physically feasible triggering source nodes.

[0075] This embodiment further calculates the impact and overall confidence score of each candidate event based on the candidate root cause event set. The impact is calculated based on the candidate event's hierarchical position in the electrical topology, the number of related events, and the depth of causal propagation. For example, equipment nodes located at the main transformer or busbar have a wider control range, so their impact weight is set between 0.8 and 1.0; the impact of end load nodes can be set between 0.3 and 0.5. The overall confidence score is generated by weighting time confidence and topology impact. In this embodiment, the time confidence weight is set to 0.6, and the topology causal weight is set to 0.4. After calculation, candidate events with scores higher than 0.7 are selected as the deterministic root cause event set. Finally, based on the deterministic root cause event set, combined with its related event chain, temporal relationship, and topology node location information, this embodiment outputs root cause analysis results data. This result includes the root cause event identifier, impact range, trigger path, and time confidence description, which can clearly reflect the cause, propagation link, and dominant factors of abnormal events in the electrical system, thereby achieving verifiable root cause location and diagnostic results output.

[0076] Furthermore, the specific implementation process of step 700 is as follows:

[0077] This embodiment first parses the root cause analysis results data, extracts the accident type, severity level, and confidence index, and constructs an accident analysis input set. The accident type field includes type codes such as equipment failure, protection failure, communication anomaly, and erroneous operation commands; the severity level is divided into four levels according to the degree of impact on the operation of electrical equipment, namely general (1), relatively serious (2), serious (3), and extremely serious (4); the confidence index is a confidence score generated in the root cause analysis stage, with a value range of 0 to 1. By centrally parsing these three types of elements, this embodiment generates a structured accident analysis input set, which serves as the core basis for subsequent automatic triggering judgment. At the same time, in order to make the accident judgment logic traceable, this embodiment records the event timestamp, equipment node name, and associated causal chain number in the input set for subsequent evidence association and retrieval.

[0078] This embodiment then compares the accident type and impact range with a preset severity threshold table based on the accident analysis input set to determine whether the accident reaches the reporting level. The severity threshold table is configured by the operating standard library before system deployment. For example, the severity threshold is 3 for main transformer fault events with voltage levels of 220 kV and above, and for events where two or more switching devices operate simultaneously; the threshold is 2 for events where line tripping affects more than 20% of the power supply area; and a threshold of 1 can be set for general signal delays and communication interruptions. If the severity level in the accident analysis input set is greater than or equal to the threshold value, it is marked as a valid preliminary accident judgment result. To avoid misjudgment, this embodiment compares the confidence index of the root cause event with a preset confidence threshold. For example, when the confidence level is higher than 0.7, the data is deemed reliable and included in the set of valid reporting events; if the confidence level is lower than the threshold, the reporting logic is not triggered temporarily.

[0079] Based on this, this embodiment determines whether the automatic triggering conditions are met according to the valid set of reported events. If met, the accident reporting template generation module is invoked, and the corresponding evidence path is locked based on the event node index and analysis records, forming a trigger status identifier and an evidence package index set. The evidence package index set includes the event record file path, action waveform capture position, equipment operating status snapshot, and the identifier number corresponding to the cause-effect graph. Subsequently, this embodiment binds the trigger status identifier and evidence index set with the accident analysis input set information, and automatically generates the accident reporting text according to the preset reporting template. The reporting text contains fields such as accident type, severity, occurrence time, root cause description, and evidence file path, and outputs an accident report in a unified format that can be directly used for scheduling reporting and archiving. Thus, the accident reporting text and the corresponding evidence package can be automatically generated and traceably collected after the root cause analysis phase, ensuring information consistency and timeliness.

[0080] This implementation also provides an automatic power grid accident reporting system based on AI semantic understanding and event chain tracing, including:

[0081] The multi-source data acquisition module is used to acquire multi-source data from the power grid and perform semantic parsing on the multi-source data to obtain event types, time clues, and equipment entity information;

[0082] The device vocabulary matching and hierarchical mapping module is used to normalize and hierarchically map the event types, time clues and device entity information based on the device vocabulary and topological knowledge to obtain the initial event model dataset.

[0083] The time alignment and confidence modeling module is used to perform drift estimation and alignment processing on time information from different sources in the initial event model dataset based on the synchronization anchor points in the multi-source data, and to establish confidence intervals for the aligned time information to obtain a time-constrained event set after cross-system timescale alignment.

[0084] The event chain construction module is used to sort and aggregate events based on the time sequence of the time-constrained event set and the topological relationship between electrical equipment to obtain the event chain model;

[0085] The causal inference module is used to perform causal inference on the correlation between event chain models based on time constraints and topological constraints, and to obtain an electrical law causal graph.

[0086] The root cause analysis module is used to determine the root cause analysis result data based on the electrical law cause-effect diagram;

[0087] The automatic reporting generation module is used to automatically trigger judgment on the root cause analysis results data based on the severity of the accident and the confidence threshold, and obtain the accident reporting text and the corresponding evidence package.

[0088] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0089] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for automatic reporting of power grid accidents based on AI semantic understanding and event chain tracing, characterized in that, include: Acquire multi-source data from the power grid and perform semantic parsing on the multi-source data to obtain event types, time clues, and equipment entity information; Based on the device vocabulary and topological knowledge, the event types, time clues and device entity information are normalized and hierarchically mapped to obtain the initial event model dataset; Based on the synchronization anchor points in the multi-source data, the time information from different sources in the initial event model dataset is drifted and aligned, and a confidence interval is established for the aligned time information to obtain a time-constrained event set after cross-system time-scale alignment. Based on the time sequence of the time-constrained event set and the topological relationships between electrical devices, the events are sorted and aggregated to obtain an event chain model; Based on time and topological constraints, causal inferences are made about the relationships between event chain models to obtain an electrical law causal graph. The root cause analysis results data are determined based on the aforementioned electrical law cause-effect diagram; The root cause analysis results are automatically triggered to make judgments based on the severity of the accident and the confidence threshold, resulting in the accident report text and the corresponding evidence package.

2. The automatic reporting method for power grid accidents based on AI semantic understanding and event chain tracing as described in claim 1, characterized in that, The process of acquiring multi-source data from the power grid and performing semantic parsing on the multi-source data to obtain event types, time clues, and equipment entity information includes: Acquire multi-source data from operation monitoring, protection waveform recording, operation management, and field communication channels, and perform format recognition, noise filtering, and time tag extraction processing on the multi-source data to obtain a pre-processed multi-source data set; The text data in the preprocessed multi-source dataset is segmented, part-of-speech tagging is performed, and named entity recognition is performed. Alarm tags, timestamps, and measurement sequence features of signal data in the multi-source dataset are extracted to obtain a unified data representation with semantic and signal features. Semantic classification is performed on the unified data representation based on a semantic rule base and contextual features to identify semantic elements containing event type candidates, timeline fragments, and device entity references; Based on semantic elements, the device entity references are normalized and numbered, resulting in a structured set of event semantic elements. The event type, timeline, and device entity information are obtained from the event semantic element set.

3. The automatic reporting method for power grid accidents based on AI semantic understanding and event chain tracing as described in claim 1, characterized in that, The process of normalizing and hierarchically mapping the event types, timelines, and device entity information based on the device vocabulary and topological knowledge yields an initial event model dataset, including: By calling the preset equipment vocabulary and power grid topology knowledge base, the equipment entities involved in the event semantic element set are matched with words and located by nodes to obtain the event element set. Perform unified normalization processing on different naming or numbering formats of the same device in the event element set to obtain a standardized device entity mapping table; Based on the standardized equipment entity mapping table, and combined with the primary and secondary system structure of the power grid, the subordinate relationships, connection relationships and action paths between equipment are determined, and an electrical hierarchy structure model is obtained. Based on the electrical hierarchy model, the event type, timeline, and device entity information are bound to the associated device nodes to obtain a structured event record set after hierarchical mapping; The event relationships in the structured event record set are uniquely indexed and numbered to obtain the initial event model dataset; The expression for the electrical hierarchy model is: ； in, This is the comprehensive representation matrix of the electrical hierarchy model; This represents the total number of equipment nodes in the power grid. For nodes With nodes The electrical connectivity coefficient between them and They are nodes and The device category code value; For nodes With nodes The shortest path distance in the topology; The average hierarchical scale constant of the network structure; The first exponential term describes the attenuation relationship of electrical influence between topology levels.

4. The automatic reporting method for power grid accidents based on AI semantic understanding and event chain tracing as described in claim 1, characterized in that, The method involves performing drift estimation and alignment processing on time information from different sources in the initial event model dataset based on synchronization anchors in multi-source data, and establishing confidence intervals for the aligned time information to obtain a time-constrained event set aligned across system timescales, including: Extract the time attribute information corresponding to each data source from the initial event model dataset, and group and organize the time attribute information according to the data source type to obtain the source-specific time attribute set; Analyze the common event identifiers and trigger signals of the source time attribute sets to identify the synchronization anchors that can be aligned between different sources, and obtain a set of time anchors containing synchronization anchor matching relationships; Based on the set of time anchor points, the time drift and drift trend of each data source relative to the reference time source are calculated to obtain the cross-source time drift parameter set. The time attributes in the initial event model dataset are corrected using a cross-source time drift parameter set, and all event times are adjusted and aligned based on the synchronization anchor point to obtain a time-scaled event model dataset. Establish confidence intervals for the time attributes of each event in the event model dataset to obtain a time-calibrated event set with confidence information; The event times and confidence levels in the time calibration event set are uniformly formatted and indexed to obtain a time constraint event set after cross-system time-scale alignment.

5. The automatic reporting method for power grid accidents based on AI semantic understanding and event chain tracing as described in claim 1, characterized in that, The expression for the event chain model is: ； in, This is the association strength matrix of the event chain model; The number of events in the time-constrained event set; For the event With the event Connectivity metrics in the device topology; For the event With the event Time sequence symbol For the event and The time difference between them; The event chain's time decay constant; The second exponential term describes the decay of the triggering effect of an event as the time interval increases.

6. The method for automatic reporting of power grid accidents based on AI semantic understanding and event chain tracing according to claim 1, characterized in that, The process of inferring causal relationships between event chain models based on time and topological constraints to obtain an electrical law causal graph includes: The temporal order of event nodes and their associated device topology node information in the event chain model are analyzed to obtain the event chain input set containing event node and constraint information. Identify the triggering relationships between the input sets of the event chain based on time constraints; Determine the logical connectivity between adjacent device nodes in the event chain input set based on topological constraints; A set of event association pairs is obtained based on the preceding and following trigger relationships and the logical connectivity relationships; Calculate the causal association strength and directionality index of each pair of events in the event association pair set to obtain a set of candidate causal relationships with causal direction labels; The candidate causal relationship set is corrected using time confidence intervals and topological path constraints to obtain the effective causal relationship set; Based on the set of effective causal relationships, an electrical law causal graph is constructed with event nodes as vertices and effective causal relationships as directed edges.

7. The automatic reporting method for power grid accidents based on AI semantic understanding and event chain tracing according to claim 6, characterized in that, The calculation of the causal association strength and directionality index of each pair of events in the event association pair set yields a candidate causal relationship set with causal direction annotations, including: Based on the event association pair set, the time interval, event type label, and associated device node information of each event pair are read to obtain the event association feature set for causal analysis; The probability of timing is calculated based on the event trigger time difference, sampling period, and timeline order. The timing order index of the event association feature set is then corrected according to the confidence interval weight to obtain the timing measurement result. Based on the event association feature set and electrical hierarchy model, the topological influence index of upstream events on downstream events is calculated to obtain the topological influence result; The sequential measurement results and the topological influence results are normalized and synthesized according to a preset weight model to obtain the event causal association set; The causal direction is determined based on the chronological order and the causal strength threshold. A directional identifier is marked in the event causal association set to obtain a set of candidate causal relationships with causal direction labels.

8. The automatic reporting method for power grid accidents based on AI semantic understanding and event chain tracing according to claim 1, characterized in that, The root cause analysis results data are determined based on the aforementioned electrical law cause-effect diagram, including: Read the event node and causal edge attribute information in the electrical law causal graph to extract the node in-degree, out-degree, event type and time confidence weight, and obtain the causal node feature set; Based on the causal node feature set, the event chain is traversed along the causal direction to identify terminal event groups that are no longer triggered by other events in time and topology, thus obtaining a candidate root cause event set. Based on the candidate root cause event set, the impact degree and overall confidence score of each candidate event are calculated according to the location of the event in the electrical topology, the number of related events, the causal transmission length and the time confidence weight, so as to obtain the root cause event score set with confidence score; Based on the root cause event score set, the highest-scoring events are selected according to the confidence threshold and the impact weight to obtain the deterministic root cause event set; Based on a deterministic set of root cause events, combined with the associated event chain, time information, and topological node information, the root cause analysis results data are obtained.

9. The automatic reporting method for power grid accidents based on AI semantic understanding and event chain tracing according to claim 1, characterized in that, The automatic triggering of judgment based on the root cause analysis results data according to the severity of the accident and the confidence threshold yields the accident reporting text and the corresponding evidence package, including: By analyzing the accident type, severity level, and confidence index in the root cause analysis results data, the accident analysis input set is obtained; Based on the accident analysis input set, the accident type and scope of impact are compared with the severity threshold table to determine whether the reporting level has been reached, and a preliminary accident judgment result is obtained. Based on the preliminary accident determination results, the confidence level of the root cause event is compared with the preset confidence threshold to obtain the set of valid reported events; Based on the set of valid reported events, determine whether the triggering conditions are met; if so, call the accident reporting template generation module and lock the relevant evidence index path to obtain the triggering status identifier and the corresponding evidence package index set. Based on the trigger status identifier and evidence package index set, the accident type, severity, time, root cause description and evidence path information are embedded into a preset text template to obtain the accident reporting text and the corresponding evidence package.

10. An automatic power grid accident reporting system based on AI semantic understanding and event chain tracing, characterized in that, include: The multi-source data acquisition module is used to acquire multi-source data from the power grid and perform semantic parsing on the multi-source data to obtain event type, time clues and equipment entity information; The device vocabulary matching and hierarchical mapping module is used to normalize and hierarchically map the event types, time clues and device entity information based on the device vocabulary and topological knowledge to obtain the initial event model dataset. The time alignment and confidence modeling module is used to perform drift estimation and alignment processing on time information from different sources in the initial event model dataset based on the synchronization anchor points in the multi-source data, and to establish confidence intervals for the aligned time information to obtain a time-constrained event set after cross-system timescale alignment. The event chain construction module is used to sort and aggregate events based on the time sequence of the time-constrained event set and the topological relationship between electrical equipment to obtain the event chain model; The causal inference module is used to perform causal inference on the correlation between event chain models based on time constraints and topological constraints, and to obtain an electrical law causal graph. The root cause analysis module is used to determine the root cause analysis result data based on the electrical law cause-effect diagram; The automatic reporting generation module is used to automatically trigger judgment on the root cause analysis results data based on the severity of the accident and the confidence threshold, and obtain the accident reporting text and the corresponding evidence package.

Images