Dynamic electric power question answering system optimization method and system based on knowledge graph

Abstract

The invention provides a dynamic electric power question-answering system optimization method and system based on a knowledge graph, and the method comprises the steps: collecting heterogeneous operation data of multiple positions of electric power equipment in real time, including regulation document rule constraint items, sensor time sequence state signals and historical maintenance event association items; and mapping the rule constraint item and the event association item to a unified semantic space, eliminating semantic cross interference, and constructing a topological structure. And then dynamically coupling the topological structure with a real-time sequence signal to form a power equipment state evolution graph. Based on the atlas and current state verification parameters uploaded by the distributed sensors, basic association strength values among the nodes are dynamically calibrated, and boundary condition constraints are applied through rule constraint terms. And finally, matching the event association item with the constrained association strength value to generate an abnormal reasoning path optimization strategy for the question and answer request. According to the invention, the accuracy and real-time performance of power equipment abnormity diagnosis are improved.

Description

Technical Field

[0001] The present application relates to the field of knowledge graph technology, and in particular to a method and system for optimizing a dynamic power question-answering system based on a knowledge graph. Background Art

[0002] As smart grid construction progresses, power systems must process multi-source, heterogeneous data on equipment operating status in real time, including sensor monitoring data, operational and maintenance procedures, and historical fault records. Traditional manual question-and-answer models struggle to rapidly locate and reason about equipment anomalies. Automated systems that integrate multimodal data, dynamically update knowledge graphs, and support intelligent question-and-answering are urgently needed to accurately diagnose faults and optimize response plans.

[0003] Current solutions utilize a power question-and-answer system based on a static knowledge graph. This system pre-builds a graph model containing device attributes and relationships, then matches graph nodes with real-time sensor data for question-and-answer reasoning. This system utilizes a graph database to store device topology relationships. When sensor data exceeds a threshold, a pre-defined rule chain is triggered to identify anomalies.

[0004] Static knowledge graphs are difficult to adapt to the dynamic changes in equipment operating status. The simple threshold matching method between sensor data and graph nodes ignores the spatiotemporal correlation characteristics between multimodal data, resulting in deviations between abnormal reasoning results and actual fault propagation paths. The handling strategies output by the question-answering system often lack foresight for complex chain failures. Summary of the Invention

[0005] The present application provides a method and system for optimizing a dynamic power question-answering system based on a knowledge graph, which is used to solve the problems of low accuracy and poor real-time performance in the prior art of abnormal diagnosis of power equipment.

[0006] In a first aspect, the present application provides a method for optimizing a dynamic power question-answering system based on a knowledge graph, comprising:

[0007] Real-time collection of heterogeneous operating data at different locations of power equipment, including rule constraints in procedure documents, time-series status signals continuously generated by sensors, and event-related items in historical maintenance records;

[0008] Mapping the rule constraint items and the event association items into a unified semantic space, eliminating cross-interference of repetitive semantic units in the semantic space, and constructing a topological structure;

[0009] Dynamically coupling the topological structure with the time series state signal in a cross-modal manner to generate a state evolution graph of the power equipment;

[0010] Based on the state evolution graph of the power equipment and the current device state verification parameters uploaded by each node in the distributed sensor network deployed on the power equipment, the basic association strength values ​​between the nodes in the topological structure are dynamically calibrated, and boundary condition constraints are imposed on the calibrated association strength values ​​through the rule constraint items;

[0011] The event association items are matched with the association strength values ​​constrained by boundary conditions to generate an abnormal reasoning path optimization strategy for the question-answering system request.

[0012] Optionally, the current device status verification parameter includes a spatial location identifier and a real-time monitoring value of a distributed sensor node;

[0013] The method dynamically calibrates the basic association strength values ​​between nodes in the topology structure based on the power equipment state evolution graph and the current equipment state verification parameters uploaded by each node in the distributed sensor network deployed on the power equipment, and imposes boundary condition constraints on the calibrated association strength values ​​through the rule constraint items, including:

[0014] Calculating the deviation between the real-time monitoring value and the historical state transition path of the corresponding entity node according to the matching result between the spatial location identifier and the corresponding entity node in the power equipment state evolution map;

[0015] When the deviation result exceeds the preset allowable fluctuation range of the correlation strength, the state transition path network in the topological structure is traced back to the deviation starting node, and the basic correlation strength values ​​between the nodes on the backtracking path are calibrated;

[0016] The operation critical value defined in the rule constraint item is used as a boundary condition constraint. When the calibrated association strength value is greater than the operation critical value, the operation critical value is used as the association strength value constrained by the boundary condition.

[0017] Optionally, tracing back along the state transition path network in the topological structure to the deviation starting node and calibrating the basic association strength values ​​between nodes on the tracing back path includes:

[0018] Detecting a difference direction between the deviation result and a preset correlation strength allowable fluctuation range, and determining a forward or reverse backtracking mode along the state transition path network according to the difference direction;

[0019] According to the backtracking mode, starting from the current entity node, backtracking along the direction of the connection line in the state transfer path network to the deviation starting node, recording the time stamp interval length and the accumulated value of the state transfer amount of the current entity node in the backtracking path;

[0020] Calculating a target association strength attenuation coefficient between every two adjacent entity nodes in the backtracking path according to the timestamp interval length and the accumulated value of the state transfer amount;

[0021] On the backtracking path, in the direction from the deviation starting node to the current entity node, the target association strength attenuation coefficient is multiplied by the basic association strength value between the corresponding entity nodes to generate a calibrated association strength value.

[0022] Optionally, calculating the target association strength attenuation coefficient between every two adjacent entity nodes in the backtracking path according to the timestamp interval length and the accumulated state transfer amount includes:

[0023] Comparing the timestamp interval length with a preset reference time window to generate a time span impact factor;

[0024] Comparing the state transfer amount cumulative value with a preset reference transfer amount to generate a transfer amount impact factor;

[0025] generating an initial correlation strength attenuation coefficient according to the inverse relationship between the time span influencing factor and the transfer amount influencing factor;

[0026] When the value of the initial association strength attenuation coefficient exceeds the preset attenuation allowable range, the upper and lower limits of the attenuation allowable range are used as mandatory constraint boundaries, and the initial association strength attenuation coefficient is subjected to boundary truncation processing to generate a target association strength attenuation coefficient.

[0027] Optionally, matching the event association item with the association strength value constrained by boundary conditions to generate an abnormal reasoning path optimization strategy for the question-answering system request includes:

[0028] Extracting a characteristic parameter set of historical abnormal events from the event association item, wherein the characteristic parameter set includes an event triggering node identifier, a correlation strength mutation threshold, and an impact path length;

[0029] Searching for a key entity node that matches the event triggering node identifier in the power equipment state evolution graph;

[0030] When the association strength value between the key entity node and the adjacent nodes, which is constrained by boundary conditions, reaches the association strength mutation threshold, the key entity node is marked as an abnormal trigger source;

[0031] Taking the abnormal trigger source as the starting point, tracing back the dependency path along the direction of attenuation of the correlation strength in the state evolution graph of the power equipment;

[0032] All nodes in the traceability dependency path are sorted according to the spatial attenuation pattern to generate an ordered path chain from the exception trigger source to the set threshold node; based on the ordered path chain and the question-answering system request, an exception reasoning path optimization strategy is generated.

[0033] Optionally, mapping the rule constraint items and the event association items into a unified semantic space, eliminating cross-interference of repetitive semantic units in the semantic space, and constructing a topological structure includes:

[0034] Converting the text description unit in the rule constraint item and the historical event identification unit in the event association item into a semantic vector unit;

[0035] Identifying repetitive semantic unit groups in the semantic vector that are used to describe the same power equipment entity or the same operating behavior, and semantically superimposing all repetitive semantic units in each repetitive semantic unit group to generate a semantic vector expression unit;

[0036] Determining the subordinate relationships between the semantic vector expression units, and establishing a unidirectional dependency link between the power equipment entities in a unified semantic space based on the subordinate relationships;

[0037] A topology structure is formed by hierarchical expansion of the unidirectional dependent links.

[0038] Optionally, the performing cross-modal dynamic coupling of the topological structure and the time sequence state signal to generate a state evolution graph of the power device includes:

[0039] Extracting a set of power equipment state parameters at each acquisition time point in the time series state signal;

[0040] Binding the power equipment state parameter set to the corresponding power equipment entity node in the topology structure;

[0041] At continuous collection time points, calculating the state transfer amount between adjacent nodes along the connection line direction of the topological structure according to the device state parameter set;

[0042] When the state transfer amount exceeds a preset transfer threshold, a state transfer mark from the current acquisition time point to the next acquisition time point is added in the topology structure;

[0043] By aggregating state transition marks of all collected time points, a state transition path network with timestamps is formed in the topological structure and basic correlation strength values ​​between adjacent nodes are generated;

[0044] A power equipment state evolution graph is constructed based on the state transition path network and the basic association strength value.

[0045] In a second aspect, the present application provides a dynamic power question-answering system optimization system based on a knowledge graph, comprising:

[0046] An acquisition module is used to collect heterogeneous operating data at different locations of power equipment in real time. The heterogeneous operating data includes rule constraints in procedure documents, time-series status signals continuously generated by sensors, and event-related items in historical maintenance records;

[0047] A construction module, configured to map the rule constraint items and the event association items into a unified semantic space, eliminate cross-interference of repetitive semantic units in the semantic space, and construct a topological structure;

[0048] A generation module, configured to dynamically couple the topological structure with the time sequence state signal in a cross-modal manner to generate a state evolution graph of the power equipment;

[0049] A calibration module is configured to dynamically calibrate the basic correlation strength values ​​between nodes in the power equipment state evolution graph based on the current device state verification parameters uploaded by each node in the distributed sensor network deployed on the power equipment, and impose boundary condition constraints on the calibrated correlation strength values ​​through the rule constraint items;

[0050] The matching module is used to match the event association items with the association strength values ​​constrained by boundary conditions, and generate an abnormal reasoning path optimization strategy for the question-answering system request.

[0051] In a third aspect, the present application provides a computing device comprising a processor and a memory, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute a dynamic power question-and-answer system optimization method based on a knowledge graph as described in any one of the first aspects.

[0052] In a fourth aspect, the present application provides a computer storage medium having computer program instructions stored thereon, which, when executed by a processor, implements a dynamic power question-and-answer system optimization method based on a knowledge graph as described in any one of the first aspects.

[0053] In the present application, a dynamic power question-and-answer system optimization method based on a knowledge graph is provided, which includes: real-time collection of heterogeneous operating data at different locations of power equipment, the heterogeneous operating data including rule constraint items of procedure documents, time series status signals continuously generated by sensors, and event association items in historical maintenance records; mapping the rule constraint items and the event association items into a unified semantic space, eliminating cross-interference of repetitive semantic units in the semantic space, and constructing a topological structure; dynamically coupling the topological structure with the time series status signal across modalities to generate a power equipment state evolution graph; dynamically calibrating the basic association strength values ​​between nodes in the topological structure based on the power equipment state evolution graph and the current equipment state verification parameters uploaded by each node in the distributed sensor network deployed on the power equipment, and applying boundary condition constraints to the calibrated association strength values ​​through the rule constraint items; matching the event association items with the association strength values ​​constrained by the boundary conditions to generate an abnormal reasoning path optimization strategy for the question-and-answer system request.

[0054] The technical solution provided by this application has the following beneficial effects:

[0055] This application fully obtains the multi-dimensional characteristics of power equipment operation, providing a comprehensive data foundation for knowledge graph construction; eliminates semantic conflicts between multi-source data, and establishes accurate equipment entity relationship expressions; realizes the deep integration of static topology structure and dynamic timing signals, and forms equipment status representation with spatiotemporal evolution characteristics; adaptively adjusts graph node relationships based on real-time sensor data to ensure the timeliness of status assessment; converts procedural requirements into graph parameter constraints to ensure that system output complies with power safety regulations; combines historical event patterns with real-time status matching to output a disposal plan that complies with actual fault propagation laws.

[0056] Furthermore, this application also achieves precise matching of monitoring data and graph nodes through the spatial location identification of distributed sensor nodes, calculates the deviation between real-time values ​​and historical states, and back-calibrates the association strength along the state transition path when it exceeds the allowable range. The operation critical value defined by the regulations is used as a mandatory constraint boundary to ensure that the calibrated association strength value meets safety specifications.

[0057] In addition, the upgrade of power equipment status assessment from static threshold judgment to dynamic path tracing has been achieved. Through the dual mechanisms of spatial positioning matching and time series backtracking calibration, the abnormal diagnosis results can reflect real-time status changes and comply with procedural constraints, thereby improving the accuracy of complex fault location and the compliance of disposal plans.

[0058] These and other aspects of the present application will become more readily apparent from the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0060] Figure 1 A flowchart of a method for optimizing a dynamic power question-answering system based on a knowledge graph provided in an embodiment of the present application;

[0061] Figure 2 A schematic diagram of the structure of a dynamic power question-answering system optimization system based on a knowledge graph provided in an embodiment of the present application;

[0062] Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0063] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0064] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to being different types.

[0065] Current intelligent question-answering technologies for power systems, based on static knowledge graphs, have limitations: their pre-defined graph node relationships cannot adapt to the dynamic evolution of equipment states, and the simple threshold matching mechanism between sensor data and graphs breaks the spatiotemporal correlations between multimodal data, leading to systematic deviations between anomaly inference results and actual fault propagation paths. This shortcoming stems from the fact that static modeling methods ignore the time-varying characteristics of power equipment operating states and the laws of chain reactions. As a result, the response strategies output by question-answering systems often lack the ability to foresee complex fault evolution.

[0066] In response to the above problems, this application proposes a dynamic power question-answering system optimization method based on knowledge graphs, which realizes the deep fusion of multimodal data by constructing a device state graph with spatiotemporal evolution characteristics. The method first maps the procedural documents, sensor signals and historical events to a unified semantic space to construct a topological structure, and then generates an evolutionary graph reflecting the device state migration path through dynamic coupling of time series signals, and calibrates the node association strength based on real-time sensor data, and finally generates an abnormal reasoning path optimization strategy based on historical event patterns. This solution breaks through the rigid modeling limitations of static graphs, accurately captures the fault propagation path through a dynamic association strength calibration mechanism, and uses boundary condition constraints to ensure the compliance of the disposal plan, effectively solving the problems of abnormal reasoning deviation and insufficient strategy foresight in the existing technology.

[0067] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0068] Figure 1 A flow chart of a method for optimizing a dynamic power question-answering system based on a knowledge graph is provided in an embodiment of the present application, such as Figure 1 As shown, the method includes:

[0069] Step 101: collecting heterogeneous operation data at different locations of power equipment in real time, wherein the heterogeneous operation data includes rule constraint items of a regulation document, time sequence status signals continuously generated by sensors, and event association items in historical maintenance records.

[0070] In step 101, heterogeneous operational data refers to a collection of operational data from different sources and formats within the power system. The rule constraint items in the procedural document represent equipment operating specifications extracted from standard documents in the power industry, such as the clause "Main transformer oil temperature must not exceed 70°C" in the "Substation Operation and Maintenance Regulations." Sensor time series status signals represent continuous monitoring data collected at regular intervals by sensors on power equipment, such as temperature, current, and voltage. The event association items in historical maintenance records represent triples of equipment failure symptoms, handling measures, and consequences recorded in past maintenance work orders.

[0071] In this embodiment of the application, the system uses a sensor network deployed on power equipment such as transformers and circuit breakers to collect real-time monitoring data such as temperature and current at preset sampling intervals. It also extracts the latest version of equipment operating procedures from the power company's document management system and retrieves historical fault records from the maintenance work order database over the past several years. These three types of data are formatted and standardized before being stored in different partitions of a distributed data warehouse. Sensor data is accompanied by equipment location codes and timestamps, procedure data is annotated with applicable equipment types, and maintenance record data is supplemented with fault level labels, providing structured input for subsequent processing.

[0072] For example, the oil temperature sensor on the main transformer of a 220kV substation collects temperature data every five minutes. The system also extracts the clause "Normal top oil temperature must not exceed 85°C" from the "Operation and Maintenance Regulations for Oil-Immersed Transformers" from the knowledge base and correlates it with the fault event "Abnormal oil temperature causing light gas activation" from the station's maintenance records over the past three years. These three types of data are labeled as time-series signal data for device number B01, rule clause R01, and event record E01, and are transmitted to the processing module via the data bus.

[0073] Step 102: Map the rule constraint items and the event association items into a unified semantic space, eliminate cross-interference of repetitive semantic units in the semantic space, and construct a topological structure.

[0074] In step 102, the unified semantic space represents the conversion of data from different sources into a mathematical space with feature vectors of the same dimension. Semantic unit cross-interference indicates semantic ambiguity or conflict in the descriptions of the same device attributes across different data sources. The topology represents a graph structure describing the relationships between power device entities, with nodes representing device entities and edges representing functional dependencies between devices.

[0075] In an embodiment of the present application, the text of the procedural clauses and historical event descriptions are converted into feature vectors using natural language processing technology. The procedural clauses are decomposed into vector representations of three dimensions: equipment type, operating action, and restriction conditions. Historical events are converted into vector representations of three dimensions: equipment number, fault type, and treatment measures. By calculating the cosine similarity between vectors, repeated vector groups describing the same equipment attributes are identified, and a weighted average is taken for each repeated vector group to generate a representative vector. Directed edges are established based on the subordinate relationships between the representative vectors, such as the "transformer" vector pointing to the "cooler" vector and annotated with the "control" relationship type, ultimately forming a topological structure with a weighted directed graph structure.

[0076] For example, rule clause R01 is converted into the vector [Transformer, Oil Temperature Monitoring, ≤85°C], and event record E01 is converted into the vector [B01, Abnormal Oil Temperature, Minor Gas Reset]. Calculation shows that the similarity between the two in the device type dimension exceeds a threshold, and they are merged to generate a new vector [B01 Transformer, Abnormal Oil Temperature Monitoring, ≤85°C or Minor Gas Reset]. This vector establishes a "Protection Action" relationship edge with the station circuit breaker vector, forming a topological structure that includes device nodes such as main transformers, circuit breakers, and coolers, and their relationships.

[0077] Step 103: Perform cross-modal dynamic coupling on the topological structure and the time sequence state signal to generate a state evolution graph of the power equipment.

[0078] In step 103, cross-modal dynamic coupling represents the fusion of the changing characteristics of the time series signal with the static relationship of the topological structure. The power equipment state evolution graph represents a dynamic graph structure formed by superimposing the time dimension on the basic topology, and the edge weights change over time.

[0079] In an embodiment of the present application, sensor time series data is matched to corresponding nodes in the topology structure by device number, and the characteristic values ​​of each time point are extracted to form a state snapshot. The difference between the state values ​​of adjacent time points is calculated, and the state change amount is transmitted along the edge direction of the topology structure. For example, the increase in the main transformer oil temperature transmits the temperature adjustment demand to the associated cooler node. The edge weight is adjusted according to the size of the transfer amount. When the transfer amount exceeds the threshold at multiple consecutive time points, a timestamp mark is added to the corresponding edge. Ultimately, an evolutionary graph containing the initial topological relationship and the time-varying state transition path is formed.

[0080] For example, if the oil temperature of main transformer B01 rises from 65°C to 78°C between 8:00 AM and 8:30 AM, the temperature rise difference is calculated every five minutes and a regulation signal is transmitted along the topological edge to the cooler node. If the temperature rise difference exceeds the 0.5°C per minute threshold at 8:25 AM, an "8:25 emergency cooling" event is marked on the edge from B01 to the cooler, forming an evolutionary subgraph that reflects the temperature anomaly propagation path.

[0081] Step 104: Based on the state evolution graph of the power equipment and the current equipment state verification parameters uploaded by each node in the distributed sensor network deployed on the power equipment, the basic association strength values ​​between the nodes in the topology structure are dynamically calibrated, and boundary condition constraints are imposed on the calibrated association strength values ​​through the rule constraints.

[0082] In step 104, the current device status verification parameter represents the latest monitoring data with location identification uploaded by the sensor. The basic association strength value represents the initial setting value of the edge weight in the topological structure. The graph node represents the power equipment entity and its state attributes (such as transformers, circuit breakers, etc.), and the network node refers to the physical sensor hardware unit. The two are associated through the mapping relationship between sensor data and graph entities. The boundary condition constraint represents the value range limit set by the procedural clauses on the association strength. The rule constraint item is the explicit rule extracted from the procedural document (such as "the temperature must be shut down when it exceeds the threshold"), and the boundary condition constraint is the dynamic restriction on the association strength calculation by applying these rules (such as forcibly cutting off the graph connection of the over-temperature node). The former is the source of the rule, and the latter is the form of rule execution.

[0083] In an embodiment of the present application, monitoring data with device location encoding uploaded by the sensor in real time is received, and the corresponding device node is located in the evolution map. The deviation between the current monitoring value and the mean value of the node's historical state is calculated. When the deviation exceeds the allowable fluctuation range, the deviation source node is searched in reverse along the time series of the state transition mark. Starting from the source node, the association strength value of each edge on the path is recalculated forward in time, and a recursive formula is used to make the strength value decay with the transmission distance. The calculated value is compared with the limit value specified in the regulations, and the excess part is truncated to generate the final association strength that meets the safety specifications.

[0084] For example, at 8:30 AM, an alarm indicating an oil temperature of 82°C was received for B01. The deviation from the historical mean of 70°C was calculated. Tracing back along the evolution graph, the cooler performance degradation event at 8:25 AM was identified as the source. The correlation strength from the cooler node to B01 was recalculated. Based on the upper limit of 85°C in regulation R01, the final correlation strength was limited to a safe range.

[0085] Step 105: Match the event association item with the association strength value constrained by the boundary condition, and generate an abnormal reasoning path optimization strategy for the question-answering system request.

[0086] In step 105 , the abnormal reasoning path optimization strategy represents a sequence of treatment suggestions generated for a specific abnormal scenario.

[0087] In an embodiment of the present application, the fault propagation pattern in the historical event is matched with the correlation strength distribution of the current evolution graph to find a highly matching historical pattern as a reference. Based on the matching results, the current abnormal source node is identified, and the scope of influence is traced along the direction of correlation strength attenuation. The nodes on the path are sorted by strength value, and the equipment corresponding to the strength mutation point is processed first. Combined with the operating requirements of each device in the regulations, an optimization strategy including inspection sequence and disposal measures is generated.

[0088] For example, the E01 event pattern matches the current B01 oil temperature anomaly, identifying the cooler as the source. Tracing back along the graph reveals a sudden change in the strength of the associated circuit breaker node. This generates a response strategy: "First check the cooler fan power supply, then test the circuit breaker protection settings." This strategy is then validated against regulatory requirements and output.

[0089] This method constructs a dynamic evolution graph by fusing multi-source data, enabling spatiotemporal correlation analysis of power equipment status, upgrading anomaly diagnosis from single-point judgment to system-level path tracing. A correlation strength mechanism calibrated with real-time data accurately reflects fault propagation patterns. Combined with procedural constraints, the resulting response strategy improves the accuracy and safety of operation and maintenance responses, providing reliable decision support for power question-and-answer systems.

[0090] In order to solve the accuracy problem of dynamic evaluation of power equipment status, in some embodiments, step 104: the current equipment status verification parameter includes the spatial location identifier and real-time monitoring value of the distributed sensor node.

[0091] The method dynamically calibrates the basic association strength values ​​between nodes in the topology structure based on the power equipment state evolution graph and the current equipment state verification parameters uploaded by each node in the distributed sensor network deployed on the power equipment, and imposes boundary condition constraints on the calibrated association strength values ​​through the rule constraint items, including:

[0092] Step 201: Based on the matching result between the spatial location identifier and the corresponding entity node in the power equipment state evolution graph, the deviation between the real-time monitoring value and the historical state transition path of the corresponding entity node is calculated.

[0093] In step 201, the spatial location identifier is the physical installation location code of the distributed sensor node on the power equipment, which is used to establish a spatial mapping relationship with the map node. The matching result refers to the precise comparison of the spatial location identifier of the distributed sensor node (such as the equipment number ST01) with the spatial code of the entity node in the power equipment state evolution map to determine the mapping relationship of the specific equipment node (such as the main transformer B01 node) corresponding to the sensor monitoring data. This result ensures that the real-time monitoring value can be accurately associated with the target analysis object in the map. The real-time monitoring value is the latest equipment operating parameter collected by the sensor. The historical state transition path is formed by the accumulation of state transition marks, that is, the path set generated by aggregating the equipment state parameter change trajectories at all time points. Deviation calculation is a quantitative process of comparing the degree of deviation between the current monitoring value and the historical state trend.

[0094] In an embodiment of the present application, the system receives real-time data with position coding uploaded by the sensor, searches for device nodes with matching position coding in the evolution map, extracts the state values ​​of the node for several past sampling cycles to form a historical trajectory curve, calculates the vertical distance between the real-time data point and the trajectory curve as the deviation, and converts the deviation into a standardized score for subsequent judgment.

[0095] Step 202: When the deviation result exceeds the preset association strength allowable fluctuation range, trace back to the deviation starting node along the state transition path network in the topological structure, and calibrate the basic association strength values ​​between the nodes on the backtracking path.

[0096] In step 202, the allowed fluctuation range of the association strength is a reasonable range of state changes preset based on the device type. The state transition path network is a dynamic subgraph in the topology structure that records the propagation direction of device state changes. It is generated by coupling the time-series state signal with the topology structure. Its edges represent the transmission direction of the state change (such as the transmission of the oil temperature anomaly from the cooler node to the transformer node). The connecting lines between nodes are accompanied by timestamps and transfer quantity parameters for tracing the abnormal propagation path. Backtracking refers to tracing back in time along the state transition mark. The deviation starting node is the device node where the state anomaly first occurs. The backtracking path specifically refers to the dependency link that traces back along the direction of the association strength attenuation in the power equipment state evolution graph, starting from the node with the most recent abnormal data reported by the distributed sensor. The current entity node path specifically refers to the dynamic tracing path in the direction of the association strength attenuation in the state transition path network, which is essentially different from the static unidirectional dependency link when constructing the topology structure. The basic association strength value reflects the inherent relationship strength of state transfer between devices.

[0097] In an embodiment of the present application, when the standardized deviation score exceeds the fluctuation range threshold, starting from the current node, the node where the abnormal feature first appeared is searched reversely along the timestamp of the state transition mark as the starting point, and the association strength between each pair of nodes on the path is recalculated forward in time from the starting point. A recursive decay algorithm is used to make the strength value decrease as the transmission distance increases, forming a new association strength distribution.

[0098] Step 203: Using the operation critical value defined in the rule constraint item as a boundary condition constraint, and when the calibrated association strength value is greater than the operation critical value, using the operation critical value as the association strength value constrained by the boundary condition.

[0099] In step 203, the operation threshold is the safety limit of the equipment parameter specified in the specification document. The boundary condition constraint is a mandatory rule that limits the association strength to a safe range.

[0100] In an embodiment of the present application, the recalculated association strength value is compared with the parameter limit of the corresponding equipment in the regulations. For strength values ​​that exceed the limit, they are forcibly adjusted to the limit value to ensure that the final output association strength reflects both the actual state changes and complies with safety specifications.

[0101] Here's a specific example:

[0102] At 08:35, the oil temperature sensor ST01 of the main transformer B01 in a 220kV substation uploaded its spatial location code ST01 and a real-time temperature value of 86°C. The system matched the location code to node B01 in the power equipment state evolution graph. The system compared the 86°C value with the historical state transition path formed by the oil temperature data from node B01 over the past 24 hours. The system calculated the deviation between the current temperature and the historical average to be 16°C. This value is calculated by subtracting the historical 24-hour average temperature of 70°C from the current temperature. Since the preset correlation strength fluctuation range is ±10°C, the system determined that the deviation exceeded the threshold. Tracing back along the state transition path marked in the graph, the system found an anomaly flagged "30% decrease in heat dissipation efficiency" at cooler node C01 at 08:25, identifying it as the starting point of the deviation. Starting from node C01, the correlation strength value to node B01 was recalculated along the state transfer direction. Using a recursive formula, the new correlation strength value was 1.8, calculated by multiplying the base correlation strength of 1.2 by the temperature rise influence coefficient of 1.5. In accordance with the maximum allowable correlation strength of 1.5 corresponding to the upper limit of oil temperature of 85°C specified in regulation clause R01, the calibrated 1.8 is forcibly adjusted to 1.5 to ensure compliance with safety regulations.

[0103] In the embodiment of the present application, dynamic association between sensor data and graph nodes is achieved through precise matching of spatial positions, deviation detection based on historical trajectories is used to improve the sensitivity of anomaly recognition, and strength calibration combined with procedural constraints is used to ensure that the evaluation results are safe and reliable, forming a closed-loop dynamic state evaluation mechanism, which effectively improves the accuracy and safety of abnormal diagnosis of power equipment.

[0104] To further improve the accuracy of backtracking abnormal states of power equipment, in some embodiments, step 202: backtracking along the state transition path network in the topology structure to the deviation starting node and calibrating the basic correlation strength values ​​between the nodes on the backtracking path includes:

[0105] Step 301: Detect the direction of the difference between the deviation result and the preset correlation strength allowable fluctuation range, and determine the forward or reverse backtracking mode along the state transition path network according to the direction of the difference.

[0106] In step 301, the difference direction refers to the direction of the comparison result between the current monitoring value and the preset fluctuation range. A positive difference indicates that the value exceeds the upper limit, and a negative difference indicates that the value falls below the lower limit. For example, when the power question-and-answer system monitors that the oil temperature sensor timing status signal of a substation main transformer shows that the current oil temperature value is 75 degrees Celsius, it first queries the preset correlation strength of the main transformer node in the power equipment state evolution map, and the allowable fluctuation range is 65 to 70 degrees Celsius. The system calculates that the deviation between the current value 75 and the upper limit of the allowable range 70 is +5 degrees Celsius. Since the calculated deviation result is a positive value and exceeds the upper limit threshold of the allowable range by 2 degrees Celsius, which is derived from the safety margin coefficient 1.2 defined in the power equipment regulation document multiplied by the sensor measurement error of 1.67 degrees Celsius and rounded, the system determines that the direction of the difference is a positive difference. Based on this, it is determined to use the reverse tracing mode to trace the source of the fault. The definition standard for a positive difference is that the measured value exceeds the upper limit of the allowable range and the deviation is greater than the threshold of 2 degrees Celsius. The reverse difference is that the measured value is lower than the lower limit of the allowable range and the absolute value of the deviation is greater than 1.5 degrees Celsius. The standard value is derived from the optimal interval of fault warning effectiveness statistically calculated in historical maintenance records. The backtracking mode is divided into forward backtracking and reverse backtracking. Forward backtracking searches for the source along the direction of state transfer, while reverse backtracking searches for the impact range in the opposite direction of state transfer.

[0107] In an embodiment of the present application, the system first determines whether the deviation is positive or negative deviation from the allowable range. If it is a positive deviation, the reverse tracing mode is selected to trace back to the source of the fault. If it is a negative deviation, the forward tracing mode is selected to track the affected device to ensure that the tracing direction is consistent with the actual fault propagation direction.

[0108] Step 302: Starting from the current entity node, according to the backtracking mode, backtrack along the direction of the connection line in the state transfer path network to the deviation starting node, and record the timestamp interval length and the accumulated value of the state transfer amount of the current entity node in the backtracking path.

[0109] In step 302, the timestamp interval length is the time difference between adjacent state transition marks. The state transfer amount cumulative value is the accumulated value of the state change amount of each node along the backtracking path.

[0110] In an embodiment of the present application, starting from the current abnormal node, the predecessor node is gradually searched along the time sequence of the state transition mark according to the selected backtracking mode, and the time interval and the cumulative value of the state change between each node are recorded to form a complete backtracking path and its spatiotemporal characteristic parameters.

[0111] Step 303: Calculate the target association strength attenuation coefficient between every two adjacent entity nodes in the backtracking path according to the timestamp interval length and the accumulated value of the state transfer amount.

[0112] In step 303, the target association strength attenuation coefficient is a weight adjustment parameter calculated according to the spatiotemporal characteristics, reflecting the attenuation degree of the state change on the path.

[0113] In an embodiment of the present application, the timestamp interval length and the cumulative value of the state transfer amount are input into the attenuation model. The longer the time interval and the smaller the transfer amount, the larger the attenuation coefficient. The dynamic attenuation coefficient between each pair of adjacent nodes on the backtracking path is calculated.

[0114] Step 304: On the backtracking path, in the direction from the deviation starting node to the current entity node, the target association strength attenuation coefficient is multiplied by the basic association strength value between the corresponding entity nodes to generate a calibrated association strength value.

[0115] In step 304, the calibrated association strength value is the dynamic relationship strength after adjustment of the spatiotemporal characteristics.

[0116] In an embodiment of the present application, starting from the deviation starting node, the basic correlation strength between each pair of nodes is multiplied by the corresponding attenuation coefficient along the backtracking path to generate a calibration value reflecting the actual state propagation strength, thereby completing the strength update of the entire path.

[0117] Here's a specific example:

[0118] At 08:35, the oil temperature sensor ST01 of the main transformer B01 in a 220kV substation uploaded a real-time temperature value of 86°C. The system calculated that this value deviated by 16°C from the historical 24-hour average temperature of 70°C, exceeding the preset tolerance of ±10°C. Because the deviation was positive, the system determined that a reverse backtracking mode was used to trace the fault source. Starting from node B01, tracing back along the connection lines in the state transition path network revealed an abnormality at 08:25, indicating a 30% drop in heat dissipation efficiency at cooler node C01. The time interval between the timestamps recorded from B01 to C01 was 10 minutes, and the cumulative state transfer value was 16°C. Based on the inverse relationship between the time interval and the transfer value, the attenuation coefficient was calculated using the formula α = 1 / (1 + Δt × ΔQ), where Δt is the time interval of 10 minutes and ΔQ is the transfer value of 16°C. The calculated attenuation coefficient for the target association strength between B01 and C01 is 0.8. Starting from the deviation starting node C01, the base correlation strength value 1.2 from C01 to B01 is multiplied by the attenuation coefficient 0.8 to generate a calibrated correlation strength value of 0.96.

[0119] In the embodiment of the present application, the difference direction is determined to ensure that the backtracking path conforms to the fault propagation law, the attenuation coefficient calculation based on the time and space characteristics accurately reflects the attenuation characteristics of the state change, and the dynamically calibrated correlation strength value more realistically presents the state influence relationship between devices, providing an accurate quantitative basis for abnormal diagnosis.

[0120] To further improve the accuracy of calculating the association strength attenuation coefficient, in some embodiments, step 303: calculating the target association strength attenuation coefficient between each two adjacent entity nodes in the backtracking path based on the timestamp interval length and the accumulated state transfer amount includes:

[0121] Step 401: Compare the timestamp interval length with a preset reference time window to generate a time span impact factor.

[0122] In step 401, the reference time window is a standard state transfer time preset according to the device type. The time span impact factor reflects the degree of deviation between the actual transfer time and the standard time.

[0123] In an embodiment of the present application, the system divides the actual recorded time interval between adjacent nodes on the backtracking path by the standard time window for state transmission of this type of device to generate a time span impact factor that characterizes time efficiency. The longer the time interval, the smaller the factor value.

[0124] Step 402: Compare the state transfer amount cumulative value with a preset reference transfer amount to generate a transfer amount impact factor.

[0125] In step 402, the reference transfer amount is the standard state change amount when the equipment is operating normally. The transfer amount influencing factor reflects the proportional relationship between the actual transfer amount and the standard amount.

[0126] In the embodiment of the present application, the actual cumulative transfer amount is divided by the standard transfer amount of the device in the reference state to generate a transfer amount impact factor that represents the amplitude of the state change. The larger the transfer amount, the larger the factor value.

[0127] Step 403: generating an initial correlation strength attenuation coefficient according to the inverse relationship between the time span influence factor and the transfer amount influence factor.

[0128] In step 403, the inverse relationship refers to the mechanism by which the time span influencing factor and the transfer volume influencing factor act inversely on the attenuation coefficient. That is, longer time intervals (smaller factor values) or smaller transfer volumes (smaller factor values) result in larger attenuation coefficients. This relationship stems from the physical laws of power equipment fault propagation—state changes naturally decay over time and with path loss during the transmission process. This relationship is derived by analyzing the quantitative relationship between time and transfer volume on the final impact in historical fault events, ensuring that the calculation model conforms to actual physical processes. The initial correlation strength attenuation coefficient is a preliminary adjustment parameter generated based on spatiotemporal characteristics.

[0129] In an embodiment of the present application, the time span influence factor is divided by the transfer amount influence factor to obtain the initial attenuation coefficient, and the time factor accounts for no less than 60%, ensuring the dominant position of the time factor in the attenuation calculation.

[0130] Step 404: When the value of the initial association strength attenuation coefficient exceeds the preset attenuation allowable range, the upper and lower limits of the attenuation allowable range are used as mandatory constraint boundaries, and the initial association strength attenuation coefficient is subjected to boundary truncation processing to generate a target association strength attenuation coefficient.

[0131] In step 404, the attenuation allowable range is a coefficient boundary preset according to the safety operation requirements of the equipment. Boundary truncation processing is a process of forcibly adjusting coefficients that exceed the range to the boundary value.

[0132] In the embodiment of the present application, when the initial attenuation coefficient is less than the lower limit, it is adjusted to the lower limit value; when it is greater than the upper limit, it is adjusted to the upper limit value, ensuring that the final coefficient reflects the actual state and complies with safety constraints.

[0133] Here's a specific example:

[0134] In a state transfer analysis between the main transformer B01 and cooler C01 nodes in a 220kV substation, the system determined that the timestamp interval between the two nodes was 10 minutes. The baseline time window for this device type was preset to 8 minutes. By dividing the actual interval of 10 minutes by the baseline of 8 minutes, a time span impact factor of 1.25 was obtained. The cumulative state transfer quantity was also recorded as 16°C. The baseline transfer quantity for this device was preset to 12°C. By dividing the actual transfer quantity of 16°C by the baseline of 12°C, a transfer quantity impact factor of 1.33 was obtained. According to the inverse proportional relationship formula α = time span impact factor / transfer quantity impact factor × 0.6 + 0.4 × time span impact factor, where the time factor is weighted at 60%, the initial correlation strength attenuation coefficient was calculated to be 1.25 / 1.33 × 0.6 + 0.4 × 1.25 = 1.06. The system preset allowable attenuation range is 0.8 to 1.2. Since 1.06 is within the allowable range, the final target correlation strength attenuation coefficient remains unchanged at 1.06.

[0135] In the embodiment of the present application, the dual consideration of time and space characteristics ensures that the attenuation coefficient accurately reflects the state transfer characteristics, and the boundary constraint mechanism ensures the rationality of the calculation results, so that the calibrated correlation strength is consistent with the actual operating status of the equipment and meets the safety specifications.

[0136] To further improve the accuracy of the abnormal reasoning path, in some embodiments, step 105: matching the event association items with the association strength values ​​constrained by boundary conditions to generate an abnormal reasoning path optimization strategy for the question-answering system request includes:

[0137] Step 501: extracting a feature parameter set of a historical abnormal event from the event association item, wherein the feature parameter set includes an event triggering node identifier, a correlation strength mutation threshold, and an impact path length.

[0138] In step 501, historical abnormal events are recorded fault events extracted from historical maintenance records. Through structured processing, they form an event template library containing fault nodes, strength mutation characteristics, and impact range. The feature parameter set is a collection of typical fault characteristics extracted from historical events. The event trigger node identifier is the device code where the fault first occurred. The association strength mutation threshold is the critical change in node relationship strength when a fault occurs. The impact path length is the maximum range of devices affected by the fault propagation.

[0139] In an embodiment of the present application, the system filters historical fault cases that match the current equipment type from the maintenance record library, extracts the fault first-occurring equipment number, correlation strength jump amplitude and number of affected equipment in each case, and forms a standardized feature template set.

[0140] Step 502: Search the power equipment state evolution graph for a key entity node that matches the event triggering node identifier.

[0141] In step 502 , the key entity nodes are candidate device nodes in the evolution graph that match the historical fault characteristics.

[0142] In an embodiment of the present application, the device number in the historical fault feature is accurately matched with the node identifier in the evolution graph to locate candidate device nodes that may reproduce the same type of fault, while considering the similarity of device type and topological location.

[0143] Step 503: When the association strength value between the key entity node and the adjacent nodes constrained by the boundary conditions reaches the association strength mutation threshold, the key entity node is marked as an abnormal trigger source.

[0144] In step 503, the abnormal trigger source is the fault starting node determined by double verification, and the verification is achieved by comparing the current correlation strength value with the historical mutation threshold.

[0145] In an embodiment of the present application, it is checked whether the calibrated association strength value between the candidate node and the adjacent nodes reaches the mutation threshold of historical similar faults. When multiple consecutive adjacent relationships meet the threshold condition, the node is determined to be an abnormal trigger source.

[0146] Step 504: starting from the abnormal trigger source, tracing back the dependency path in the direction of attenuation of the correlation strength in the power equipment state evolution graph.

[0147] In step 504, the direction of association strength attenuation reflects the direction of the fault impact spread. "Tracing back to the deviation starting node" is used to calibrate the association strength, and "reverse tracing the dependent path" is used to locate the source of the anomaly. The operation objects are the same but the purposes are different. Both are technical actions of tracing back along the path. The association between the tracing dependent path and the backtracking path: the tracing dependent path specifically refers to the process of reversely searching for the fault propagation path from the anomaly trigger source along the direction of association strength attenuation, while the backtracking path is the path of reversely searching for the deviation source from the current anomaly node. The two technologies are essentially the same but have different application stages. The tracing dependent path is used for final anomaly location, and the backtracking path is used for data calibration in the intermediate process. In addition, the generation of the tracing dependent path directly depends on the association strength attenuation rule established in the backtracking path stage.

[0148] In an embodiment of the present application, starting from the trigger source node, the affected device nodes are gradually searched along the direction of decreasing association strength, and each node passed through and its strength change are recorded to form a complete fault propagation path.

[0149] Step 505: Sort all nodes in the traceability dependency path according to the spatial attenuation pattern to generate an ordered path chain from the exception trigger source to the set threshold node, and generate an exception reasoning path optimization strategy based on the ordered path chain and the question-answering system request.

[0150] In step 505, a spatial attenuation pattern describes the decreasing pattern of correlation strength values ​​in abnormal events as they propagate outward from the triggering node. This pattern is derived from statistical analysis of historical maintenance records of fault events. Specifically, with the fault source as the center, the correlation strength values ​​decrease according to the inverse square of the physical distance between devices. The influencing factors of the power network topology are superimposed to form the final spatial attenuation coefficient curve. An ordered path chain is a sequence of critical devices ranked by fault impact. Requests to the question-and-answer system refer to abnormality diagnosis requirements input by power operation and maintenance personnel through the human-computer interface. These requests come from two sources: real-time alarm requests automatically triggered by the system, generated when monitoring data exceeds a threshold; and proactive query requests, such as those submitted by operators regarding abnormal conditions of specific equipment. Requests typically include the target device identifier, a description of the abnormal phenomenon, and the desired diagnostic depth parameters. The system matches these structured request parameters with the abnormality inference path to generate an optimization strategy that meets actual requirements.

[0151] In an embodiment of the present application, the traced path nodes are sorted from large to small according to the association strength value, and equipment with obvious strength mutation is processed first. Combined with the processing resource constraints in the question and answer request, the optimal inspection and maintenance sequence is generated.

[0152] Here's a specific example:

[0153] A 220kV substation system detected that the oil temperature at the main transformer node B01 reached 88°C, exceeding the 85°C limit specified in regulation R01, triggering an automated Q&A request. The system extracted characteristic parameters of event record E01 from the historical event database and determined that cooler node C01 was a typical trigger source, with a correlation strength mutation threshold of 1.8 and an impact path length of 3. The correlation strength between the current node C01 and node B01, as matched in the evolution graph, was 2.1. This was calculated by dividing the current oil temperature deviation of 18°C ​​by the baseline temperature difference of 10°C, which yields 1.8. Adding a cooling efficiency reduction factor of 0.3 yields 2.1. Because this value exceeded the historical threshold of 1.8, C01 was identified as the abnormal trigger source. Tracing the path along the direction of strength decay revealed that the strength value from C01 to B01 decreased from 2.1 to 1.5. Continuing to trace the path to the cooling water pump node P01, the strength value reached 1.2, forming a complete path from P01 to C01 to B01. Sorting by intensity values ​​generates an ordered path chain, C01-B01-P01. Based on the Q&A request's requirement to prioritize the primary fault point, the inspection sequence is optimized to: first address the C01 cooling system, then review the B01 oil temperature protection, and finally check the P01 water supply line. This strategy, after being verified to be consistent with the requirement in regulation clause R01 that "abnormal oil temperature should prioritize checking the cooling system," is then output as the final optimization strategy to the O&M terminal.

[0154] In the embodiment of the present application, accurate positioning of the abnormal source is ensured by dual matching of historical fault characteristics and real-time status, and path tracing based on the intensity attenuation law truly reflects the fault propagation process. The generated optimization strategy not only conforms to the actual change characteristics of the equipment status, but also meets the requirements of the operation and maintenance regulations, thereby improving the decision-making reliability of the power question and answer system.

[0155] To further improve the accuracy of power equipment knowledge representation, in some embodiments, step 102: mapping the rule constraint items and the event association items into a unified semantic space, eliminating cross-interference of repetitive semantic units in the semantic space, and constructing a topological structure, includes:

[0156] Step 601: Convert the text description unit in the rule constraint item and the historical event identification unit in the event association item into a semantic vector unit.

[0157] In step 601, the text description unit in the rule constraint item is derived from the semantic parsing of the power system specification document. By extracting the clauses defining equipment operating specifications from the document, each clause is converted into a structured text fragment containing the three elements of the operator, action type, and constraint condition, forming a text description unit. The historical event identification unit in the event association item is derived from the structured processing of maintenance records. By identifying the key fields of the record, including equipment number, fault type, and treatment measure, each maintenance event is encoded as a triplet identifier of "equipment number-fault code-treatment code", forming a historical event identification unit. The semantic vector unit is a feature vector representation in a multidimensional space.

[0158] In an embodiment of the present application, the system decomposes the procedural clauses into triple segments according to the grammatical structure of subject-predicate-object, encodes historical events into feature tags according to the structure of equipment-fault-measures, and converts the two types of data into vector representations of the same dimension through a semantic model to ensure that data from different sources can be compared.

[0159] Step 602: Identify repetitive semantic unit groups in the semantic vector that are used to describe the same power equipment entity or the same operation behavior, perform semantic superposition on all repetitive semantic units in each group of repetitive semantic units, and generate a semantic vector expression unit.

[0160] In step 602, the repetitive semantic unit group is a multi-source data set describing the same device attribute or operation behavior. Semantic superposition is a weighted fusion process of similar vectors. The semantic vector expression unit is the representative vector after fusion.

[0161] In an embodiment of the present application, the cosine similarity between all semantic vectors is calculated, and the vector groups whose similarity exceeds a threshold are weighted averaged. The weights are determined based on the authority and timeliness of the data source to generate a unified vector expression representing the semantic group.

[0162] Step 603: Determine the subordinate relationships between the semantic vector expression units, and establish a unidirectional dependency link between the power equipment entities in a unified semantic space based on the subordinate relationships.

[0163] In step 603, the subordinate relationship is a logical relationship between device entities such as control and protection. A unidirectional dependency link is an edge with an arrow in a directed graph that represents a master-slave relationship.

[0164] In an embodiment of the present application, the correlation of the semantic vector expression unit in dimensions such as device type and functional attributes is analyzed, and a directed connection is established from the dominant device vector to the controlled device vector according to the device hierarchy relationship defined by the regulations. The connection types include control, protection, linkage, etc.

[0165] Step 604: forming a topology structure through hierarchical expansion of the unidirectional dependent links.

[0166] In step 604 , hierarchical expansion is a process of establishing a complete device relationship network through recursive traversal.

[0167] In an embodiment of the present application, starting from the core device node, the network is gradually expanded along the established unidirectional dependency link, adding secondary devices and association relationships until all device nodes are incorporated into the network to form a complete device topology.

[0168] Here's a specific example:

[0169] In a 220kV substation system, the processing module receives rule clause R01, "The main transformer oil temperature must not exceed 85°C," and event record E01, "Abnormal main transformer oil temperature in B01 caused a minor gas operation." First, R01 is decomposed into three dimensions: equipment type "transformer," monitoring parameter "oil temperature," and constraint condition "≤85°C." These dimensions are converted into semantic vectors [0.8, 0.9, 0.7]. The weights of each dimension are calculated using the TF-IDF algorithm, with the formula w = term frequency × inverse document frequency. E01 is also converted into the vector [0.9, 0.8, 0.6], and the equipment number B01 is mapped to 0.9 using a hashing algorithm. The cosine similarity of the two vectors in the equipment type dimension is 0.95, exceeding the preset threshold of 0.85, indicating that they describe the same equipment attributes. The two sets of vectors are weighted and fused, with a weight of 0.6 for the rule and 0.4 for the event. This generates a new vector [0.84, 0.86, 0.66], representing the abnormal oil temperature monitoring signature for the B01 main transformer. Based on the "oil temperature abnormality triggers protection" clause in the regulations, a "protection action" edge is established between this vector and the station circuit breaker vector [0.7, 0.5, 0.8]. The edge weight is set to 0.75, which is calculated by multiplying the protection action urgency coefficient of 0.5 and the device association coefficient of 1.5. Continuing with the "cooler controls oil temperature" rule, the cooler node [0.6, 0.7, 0.5] is added and a "control" edge is established with a weight of 0.9. This ultimately forms a topological structure consisting of the B01 main transformer, circuit breaker, and cooler nodes and their relationships, providing a foundational framework for subsequent state evolution analysis.

[0170] In an embodiment of the present application, knowledge representation ambiguity is eliminated through semantic fusion of multi-source data, and a precise topological network is constructed based on the equipment relationship defined by the regulations, providing a structurally complete and semantically clear equipment relationship foundation for subsequent state evolution analysis, thereby improving the accuracy and usability of power knowledge representation.

[0171] In order to further improve the dynamic characterization capability of the state evolution of the power device, in some embodiments, step 103: dynamically coupling the topology structure with the time series state signal across modalities to generate a state evolution graph of the power device includes:

[0172] Step 701: extracting a set of power equipment status parameters at each acquisition time point in the time series status signal.

[0173] In step 701, the power equipment state parameter set is multi-dimensional operating data collected by the sensor at a single sampling moment, including key parameters such as temperature, current, and voltage.

[0174] In an embodiment of the present application, the system obtains device operation data from each sensor node according to a preset sampling period, performs time alignment and format standardization on the multi-source monitoring data at each time point, and forms a set of device status snapshots with timestamps.

[0175] Step 702: Bind the power equipment state parameter set to the corresponding power equipment entity node in the topology structure.

[0176] In step 702, binding refers to the process of establishing a mapping relationship between real-time monitoring data and topological nodes.

[0177] In the embodiment of the present application, each state parameter set is associated with the corresponding device entity node in the topological structure through device location coding and type matching, as the instantaneous state attribute of the node.

[0178] Step 703: At continuous collection time points, the state transfer amount between adjacent nodes is calculated according to the device state parameter set along the connection line direction of the topological structure.

[0179] In step 703, continuity is determined when the sensor generates time series data points at a fixed sampling interval (e.g., 1 minute). When there is no missing data between two time points and the interval is equal to the preset sampling period, the data is considered continuous. For example, if the sampling period is 5 minutes, then "08:00, 08:05, 08:10" are continuous time points, while "08:00, 08:07" are discontinuous. Adjacent refers only to nodes in the topology that are directly connected by a connecting line and is not related to the physical location of the sensor. For example, if there is a connecting line from "transformer to circuit breaker" in the topology, it is considered adjacent. The state transfer quantity is a quantified value of the state change between adjacent device nodes, reflecting the propagation intensity of the fault or anomaly.

[0180] In an embodiment of the present application, the difference value of the state parameters of adjacent nodes at the current time point is calculated, the difference value is transmitted along the topological connection line, and the transmission amount is weighted according to the connection type.

[0181] Step 704: When the state transfer amount exceeds a preset transfer threshold, a state transfer mark from the current collection time point to the next collection time point is added to the topology structure.

[0182] In step 704, the transfer threshold is a preset state change sensitivity based on the device type. The acquisition time point represents the moment when the sensor actually records data (e.g., 08:00). The current time point represents the acquisition time point currently being processed (e.g., 08:00). The next time point represents the acquisition time point immediately following the current time point (e.g., 08:05). The state transition marker is a spatiotemporal label that records the characteristics of anomaly propagation.

[0183] In an embodiment of the present application, when the transmission amount exceeds the device tolerance threshold, a triplet mark including a timestamp, a transmission direction and a value is added to the corresponding connection line.

[0184] Step 705: By aggregating the state transition marks of all collected time points, a state transition path network with timestamps is formed in the topology structure and basic association strength values ​​between adjacent nodes are generated.

[0185] In the embodiment of the present application, the transfer marks of all time points are aggregated to form a propagation network with a time dimension, and the average value of the transfer amount on each connection line is counted as the basic association strength.

[0186] Step 706: constructing a power equipment state evolution graph according to the state transition path network and the basic association strength value.

[0187] In an embodiment of the present application, the basic topological relationship is superimposed on the timing transition network to form a multi-layer graph structure in which nodes represent device entities and edges contain static relationships and dynamic propagation characteristics.

[0188] Here's a specific example:

[0189] In a 220 kV substation system, oil temperature monitoring data for the main transformer B01 node is collected at 5-minute intervals between 08:00 and 08:30. At 08:00, a temperature of 65°C is collected and bound to the B01 node in the topology. At 08:05, a temperature of 67°C is collected and a temperature rise of 2°C is calculated. This temperature rise is then transferred along the control edge from B01 to cooler C01. The transfer amount is calculated as ΔT × Kc, where ΔT is the temperature difference of 2°C and Kc is the cooling system regulation coefficient, set to 0.8. The resulting transfer amount is 1.6, which is below the preset threshold of 3. At 08:25, a temperature of 78°C is collected. The calculated 5-minute temperature rise difference is 7°C, and the transfer amount is 7 × 0.8 = 5.6, exceeding the threshold of 3. A state transition marker is added to the B01 to C01 edge with a timestamp of 08:25, a transfer direction of B01 to C01, and a transfer amount of 5.6. At 08:30, a temperature of 82°C is collected and a transfer amount of 6.4, also exceeding the threshold, and a corresponding marker is added. The system aggregates the transfer markers at all time points and calculates the average transfer amount for the edge from B01 to C01 as 5.6 + 6.4 / 2 = 6.0, which serves as the base correlation strength between the two nodes. In the resulting state evolution graph, the B01 node contains time series temperature data, and the edge from B01 to C01 includes the control relationship type and an average transfer strength of 6.0, fully documenting the dynamic propagation of the oil temperature anomaly from the main transformer to the cooling system.

[0190] In the embodiment of the present application, through the dynamic coupling of time series data and topological structure, a complete characterization of the power equipment status from static relationship to dynamic evolution is achieved, enabling the system to intuitively present the abnormal propagation path and key influencing links, providing a visual analysis basis for accurate diagnosis and disposal decisions.

[0191] Figure 2 A schematic diagram of the structure of a dynamic power question-answering system optimization system based on a knowledge graph provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, the system includes:

[0192] The acquisition module 21 is used to collect heterogeneous operation data at different locations of the power equipment in real time. The heterogeneous operation data includes rule constraints in the procedure document, time sequence status signals continuously generated by sensors, and event association items in historical maintenance records.

[0193] The construction module 22 is configured to map the rule constraint items and the event association items into a unified semantic space, eliminate cross-interference of repetitive semantic units in the semantic space, and construct a topological structure.

[0194] The generating module 23 is used to dynamically couple the topological structure with the time sequence state signal in a cross-modal manner to generate a state evolution graph of the power equipment.

[0195] The calibration module 24 is used to dynamically calibrate the basic correlation strength values ​​between nodes in the power equipment state evolution map based on the current equipment state verification parameters uploaded by each node in the distributed sensor network deployed on the power equipment, and impose boundary condition constraints on the calibrated correlation strength values ​​through the rule constraint items.

[0196] The matching module 25 is used to match the event association items with the association strength values ​​constrained by boundary conditions, and generate an abnormal reasoning path optimization strategy for the question-answering system request.

[0197] Figure 2 The dynamic power question answering system optimization system based on knowledge graph can be executed Figure 1 The implementation principle and technical effects of the knowledge graph-based dynamic power question-answering system optimization method described in the illustrated embodiment will not be elaborated on here. The specific manner in which each module and unit performs operations in the knowledge graph-based dynamic power question-answering system optimization system in the above embodiment has been described in detail in the embodiment of the method and will not be elaborated on here.

[0198] In one possible design, Figure 2 The embodiment shown is a knowledge graph-based dynamic power question answering system optimization system that can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0199] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0200] The processing component 32 is as follows Figure 1 The embodiment provides a method for optimizing a dynamic power question-answering system based on a knowledge graph.

[0201] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above method.

[0202] The storage component 31 is configured to store various types of data to support operations in the terminal. The storage component can be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as static random-access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk, or optical disk.

[0203] Of course, a computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0204] The input / output interface provides an interface between the processing component and the peripheral interface module, which can be an output device, an input device, etc.

[0205] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0206] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0207] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 The embodiment shown is a method for optimizing a dynamic power question-answering system based on a knowledge graph.

[0208] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0209] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0210] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0211] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A knowledge graph-based dynamic power question-answering system optimization method, characterized in that: include: Real-time collection of heterogeneous operating data at different locations of power equipment, including rule constraints in procedure documents, time-series status signals continuously generated by sensors, and event-related items in historical maintenance records; Mapping the rule constraint items and the event association items into a unified semantic space, eliminating cross-interference of repetitive semantic units in the semantic space, and constructing a topological structure; Dynamically coupling the topological structure with the time series state signal in a cross-modal manner to generate a state evolution graph of the power equipment; Based on the state evolution graph of the power equipment and the current device state verification parameters uploaded by each node in the distributed sensor network deployed on the power equipment, the basic association strength values ​​between the nodes in the topological structure are dynamically calibrated, and boundary condition constraints are imposed on the calibrated association strength values ​​through the rule constraint items; The event association items are matched with the association strength values ​​constrained by boundary conditions to generate an abnormal reasoning path optimization strategy for the question-answering system request.

2. The method according to claim 1, characterized in that The current device status verification parameters include the spatial location identifier and real-time monitoring value of the distributed sensor node; The method dynamically calibrates the basic association strength values ​​between nodes in the topology structure based on the power equipment state evolution graph and the current equipment state verification parameters uploaded by each node in the distributed sensor network deployed on the power equipment, and imposes boundary condition constraints on the calibrated association strength values ​​through the rule constraint items, including: Calculate the deviation between the real-time monitoring value and the historical state transition path of the corresponding entity node according to the matching result between the spatial location identifier and the corresponding entity node in the power equipment state evolution map; When the deviation result exceeds the preset allowable fluctuation range of the correlation strength, the state transition path network in the topological structure is traced back to the deviation starting node, and the basic correlation strength values ​​between the nodes on the backtracking path are calibrated; The operation critical value defined in the rule constraint item is used as a boundary condition constraint. When the calibrated association strength value is greater than the operation critical value, the operation critical value is used as the association strength value constrained by the boundary condition.

3. The method according to claim 2, characterized in that The step of tracing back along the state transition path network in the topological structure to the deviation starting node and calibrating the basic association strength values ​​between the nodes on the tracing back path includes: Detecting a difference direction between the deviation result and a preset correlation strength allowable fluctuation range, and determining a forward or reverse backtracking mode along the state transition path network according to the difference direction; According to the backtracking mode, starting from the current entity node, backtracking along the direction of the connection line in the state transfer path network to the deviation starting node, recording the time stamp interval length and the accumulated value of the state transfer amount of the current entity node in the backtracking path; Calculating a target association strength attenuation coefficient between every two adjacent entity nodes in the backtracking path according to the timestamp interval length and the accumulated value of the state transfer amount; On the backtracking path, in the direction from the deviation starting node to the current entity node, the target association strength attenuation coefficient is multiplied by the basic association strength value between the corresponding entity nodes to generate a calibrated association strength value.

4. The method according to claim 3, characterized in that The calculating, according to the timestamp interval length and the accumulated value of the state transfer amount, a target association strength attenuation coefficient between every two adjacent entity nodes in the backtracking path includes: Comparing the timestamp interval length with a preset reference time window to generate a time span impact factor; Comparing the state transfer amount cumulative value with a preset reference transfer amount to generate a transfer amount impact factor; generating an initial correlation strength attenuation coefficient according to the inverse relationship between the time span influencing factor and the transfer amount influencing factor; When the value of the initial association strength attenuation coefficient exceeds the preset attenuation allowable range, the upper and lower limits of the attenuation allowable range are used as mandatory constraint boundaries, and the initial association strength attenuation coefficient is subjected to boundary truncation processing to generate a target association strength attenuation coefficient.

5. The method according to claim 1, wherein The matching of the event association items with the association strength values ​​constrained by boundary conditions to generate an abnormal reasoning path optimization strategy for the question-answering system request includes: Extracting a characteristic parameter set of historical abnormal events from the event association item, wherein the characteristic parameter set includes an event triggering node identifier, a correlation strength mutation threshold, and an impact path length; Searching for a key entity node that matches the event triggering node identifier in the power equipment state evolution graph; When the association strength value between the key entity node and the adjacent nodes, which is constrained by boundary conditions, reaches the association strength mutation threshold, the key entity node is marked as an abnormal trigger source; Taking the abnormal trigger source as the starting point, tracing back the dependency path along the direction of attenuation of the correlation strength in the state evolution graph of the power equipment; All nodes in the traceability dependency path are sorted according to the spatial attenuation pattern to generate an ordered path chain from the exception trigger source to the set threshold node; based on the ordered path chain and the question-answering system request, an exception reasoning path optimization strategy is generated.

6. The method according to claim 1, wherein Mapping the rule constraint items and the event association items into a unified semantic space, eliminating cross-interference of repetitive semantic units in the semantic space, and constructing a topological structure includes: Converting the text description unit in the rule constraint item and the historical event identification unit in the event association item into a semantic vector unit; Identifying repetitive semantic unit groups in the semantic vector that are used to describe the same power equipment entity or the same operating behavior, and semantically superimposing all repetitive semantic units in each repetitive semantic unit group to generate a semantic vector expression unit; Determining the subordinate relationships between the semantic vector expression units, and establishing a unidirectional dependency link between the power equipment entities in a unified semantic space based on the subordinate relationships; A topology structure is formed by hierarchical expansion of the unidirectional dependent links.

7. The method according to claim 1, characterized in that The cross-modal dynamic coupling of the topological structure and the time sequence state signal to generate a state evolution graph of the power equipment includes: Extracting a set of power equipment state parameters at each acquisition time point in the time series state signal; Binding the power equipment state parameter set to the corresponding power equipment entity node in the topology structure; At continuous collection time points, calculating the state transfer amount between adjacent nodes along the connection line direction of the topological structure according to the device state parameter set; When the state transfer amount exceeds a preset transfer threshold, a state transfer mark from the current acquisition time point to the next acquisition time point is added in the topology structure; By aggregating state transition marks of all collected time points, a state transition path network with timestamps is formed in the topological structure and basic correlation strength values ​​between adjacent nodes are generated; A power equipment state evolution graph is constructed based on the state transition path network and the basic association strength value.

8. A knowledge graph-based dynamic power question-answering system optimization system, characterized in that: include: An acquisition module is used to collect heterogeneous operating data at different locations of power equipment in real time. The heterogeneous operating data includes rule constraints in procedure documents, time-series status signals continuously generated by sensors, and event-related items in historical maintenance records; A construction module, configured to map the rule constraint items and the event association items into a unified semantic space, eliminate cross-interference of repetitive semantic units in the semantic space, and construct a topological structure; A generation module, configured to dynamically couple the topological structure with the time sequence state signal in a cross-modal manner to generate a state evolution graph of the power equipment; A calibration module is configured to dynamically calibrate the basic correlation strength values ​​between nodes in the power equipment state evolution graph based on the current device state verification parameters uploaded by each node in the distributed sensor network deployed on the power equipment, and impose boundary condition constraints on the calibrated correlation strength values ​​through the rule constraint items; The matching module is used to match the event association items with the association strength values ​​constrained by boundary conditions, and generate an abnormal reasoning path optimization strategy for the question-answering system request.

9. A computing device, characterized in that It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a dynamic power question-answering system optimization method based on a knowledge graph as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that A computer program is stored, and when the computer program is executed by a computer, the method for optimizing a dynamic power question-answering system based on a knowledge graph as described in any one of claims 1 to 7 is implemented.

Images