Semantic enhancement and dynamic completion method and system for power market data graph

Abstract

The invention provides a semantic enhancement and dynamic completion method and system for a power market data graph. The method comprises the following steps: acquiring target entity information and external data of a power market data graph by responding to a data query instruction; converting external data into structured semantic tags, then traversing the topological structure, establishing a mapping relationship between the tags and entity attributes, and combining query instruction influence logic to calculate topological integrity indexes and mark missing entities or correlate to form a missing tag set; establishing attribute mapping based on a target entity and external data, generating a semantic vector fusing cross-domain features, and projecting the semantic vector to a topological space according to a missing mark set to form cross-domain feature topological nodes and connections; embedding a missing position to reconstruct a topological structure, and checking logic connectivity; the dynamic conduction chain is analyzed, and a feedback interface containing the semantic relation and the completion structure is generated to respond to query. According to the method, conduction logic closed-loop verification of the missing elements in the power market data graph is realized.

Description

Technical Field

[0001] This application relates to the technical field of power market informatization, and particularly to a method and system for semantic enhancement and dynamic completion of a power market data graph. Background Art

[0002] For power market trading decisions, a dynamic knowledge graph needs to be established to integrate multi-source heterogeneous data. This graph must achieve semantic fusion of unstructured external data and structured operation data, dynamically repair missing associations between entities caused by market changes, and ensure that the topological structure after completion conforms to the conduction rules unique to the power market.

[0003] Currently, some use static graph completion techniques based on ontology mapping to standardize entities through a predefined power trading ontology library, match external data to ontology fields using a rule engine, supplement static association edges based on historical co-occurrence frequencies when detecting data missing, and finally execute predefined integrity verification rules to verify the completeness of the basic data.

[0004] However, this technology has an essential limitation: it relies on predefined ontology rules and static historical statistics, cannot analyze implicit relationships in cross-domain data, ignores the constraints of the power market conduction mechanism during the completion process, resulting in topological logic conflicts, and the response interface only displays isolated completed entities without being able to restore the dynamic conduction path, making it difficult to support decision traceability. Summary of the Invention

[0005] This application provides a method and system for semantic enhancement and dynamic completion of a power market data graph to solve the problem in the prior art that a closed-loop conduction path between cross-domain data cannot be dynamically established.

[0006] In a first aspect, this application provides a method for semantic enhancement and dynamic completion of a power market data graph, including: In response to a data query instruction sent by a user, collect the target entity information of structured operation and unstructured external data in a pre-established power market data graph; Convert the external data into structured semantic tags, traverse the topological structure of the target entity information, establish a mapping relationship between the structured semantic tags and the target entity attributes, combine the influence logic in the data query instruction, calculate the integrity index of the topological structure based on the mapping relationship, and mark missing entities or missing associations related to the data query instruction to generate a missing mark set; Based on the target entity information and the external data, establish an attribute mapping between the external data attributes and the entity attributes, generate a semantic vector integrating cross-domain features by analyzing the implicit relationship between the external data and the target entity information in the attribute mapping, project the semantic vector into the topological space according to the missing mark set, and generate topological nodes and topological connections inheriting cross-domain features. Embed the topological nodes and topological connections into the missing marker positions of the power market data graph, reconstruct the topological structure, and verify the logical connectivity of the topological structure after completion based on the relationship conduction rules; Based on the verified topological structure, analyze the dynamic conduction chain formed by the target entity information through the completed path, and generate a feedback interface containing semantic relationships and a completed structure to respond to the data query instruction.

[0007] Optionally, it further includes: For the target entity information, call the ontology layer in the pre-constructed power market data graph to identify the domain categories to which the entities belong and the domain relationship categories between the entities; In the data layer of the power market data graph, extract the neighbor node features and multi-hop path features associated with the identified entities and relationships as entity context features; Fuse the domain categories to which the entities belong, the domain relationship categories between the entities, and the entity context features to generate domain-enhanced semantic vectors; Inject the domain-enhanced semantic vectors into the semantic vectors that fuse cross-domain features, and output enhanced semantic vectors.

[0008] Optionally, in the topological space of the power market data graph, based on the structural similarity and attribute similarity between entities, locate the cross-domain entities that have an associated mapping with the target entity information, and output the cross-domain entities and their associated relationships; Taking the cross-domain entities and the target entities as anchor points, dynamically construct an inference sub-network containing multi-hop neighbors in the topological space, and perform logical rule reasoning along the multi-hop path in the dynamic inference sub-network, and output the inference result; Fuse the target entity information, the cross-domain entities and their associated relationships, and the inference result to generate a joint semantic representation; Decompose the joint semantic representation into a sub-vector set according to the entity hierarchy, and hierarchically project the sub-vector set onto the topological space based on the voltage level and service type to generate the completed nodes and associated edges corresponding to the missing marker set; Verify the business logic compliance of the completed nodes and associated edges through a rule engine, and perform cross-level backtracking correction on the nodes that fail the verification until the logical connectivity requirements of the topological structure are met.

[0009] Optionally, during the establishment of the attribute mapping, call the pre-constructed power domain term library to identify the professional terms in the external data and target entity information, perform semantic disambiguation, and output standardized entity attributes; Using the standardized entity attributes, when traversing the target entity information and semantic tags, detect whether they hit predefined atomic metrics, derived metrics, or composite metrics. If a hit occurs, according to the predefined metric calculation rules, output the reconstructed node connection data completion path and calculation logic description; Based on the node connection data completion path and calculation logic description, dynamically construct a sub-network containing relevant entities, attributes, and calculation rules in the topological space, and execute the metric definition rules within the sub-network to output the calculated predefined metric values; Use the predefined metric values as cross-domain features to inject into the semantic vectors of the corresponding entities or relationships, and based on the semantic vectors containing the metric values, call the rule-based diagnostic templates associated with the metrics to perform anomaly analysis and interpretation.

[0010] Optionally, based on the target entity information and the external data, establish an attribute mapping between the external data attributes and entity attributes. By analyzing the implicit relationship between the external data and the target entity information in the attribute mapping, generate semantic vectors that integrate cross-domain features, and project the semantic vectors into the topological space according to the missing marker set to generate topological nodes and topological connections that inherit cross-domain features, including: Based on the target entity information and the external data, construct a bidirectional mapping dictionary between the external data attributes and entity attributes, and record the direct mapping rules, mathematical derivation formulas, and unit conversion coefficients between the attributes; Through a deep semantic association model, analyze the implicit multi-order derivation relationships between the attributes in the bidirectional mapping dictionary, and identify cross-domain attribute combinations with common change characteristics or causal constraints; Generate semantic vectors for each cross-domain attribute combination, where the semantic vectors integrate the encoded values, relationship weight coefficients, and time sequence markers of the associated attributes; According to the topological coordinates located by the missing marker set, create blank topological nodes in the power market data graph, and project the semantic vectors onto the corresponding blank topological nodes to form topological nodes carrying cross-domain features; According to the relationship weight coefficients in the semantic vectors, establish weighted topological connections that inherit cross-domain features between the topological nodes carrying cross-domain features. The attributes and connection relationships of the topological nodes and topological connections need to satisfy the topological constraint rules of the power market graph.

[0011] Optionally, through a deep semantic association model, analyze the implicit multi-order derivation relationships between the attributes in the bidirectional mapping dictionary, and identify cross-domain attribute combinations with common change characteristics or causal constraints, including: Receive a set of mapping entries from the bidirectional mapping dictionary through the deep semantic association model. The set of mapping entries records the corresponding rules and relationship identifiers between external data attributes and target entity attributes, and construct a derivation network with the set of mapping entries; Perform a traversal operation on the derivation network, identify the consistency of the time dimension fluctuation characteristics between the nodes in the derivation network, record the co-varying attribute group, and detect the conduction paths that meet the causal constraint rules in the derivation network and record the response sequence; Calculate the fluctuation consistency quantization value of the co-varying attribute group, and measure the trigger delay time of the response sequence; when the fluctuation consistency quantization value exceeds the set threshold and the trigger delay time is within the preset interval, output the cross-domain attribute combination with the relationship identifier.

[0012] Optionally, based on the verified topological structure, parse the dynamic conduction chain formed by the target entity information through the completion path, and generate a feedback interface including semantic relationships and completion structures to respond to the data query instruction, including: Based on the verified topological structure, extract the dynamic conduction chain formed by the target entity information through the completion path, and parse the relationship weight value and the time sequence marking trajectory of each conduction path in the dynamic conduction chain; Map the relationship weight value to a preset semantic intensity level, convert the time sequence marking trajectory into a time axis, generate a node relationship chain with a conduction direction identifier, and mark the semantic intensity level and the time axis on the node relationship chain; Convert the conduction paths that fail the verification into warning identifiers, integrate the node relationship chain and the warning identifiers, and generate a visual feedback interface to respond to the data query instruction.

[0013] Optionally, convert the external data into structured semantic tags, traverse the topological structure of the target entity information, establish the mapping relationship between the structured semantic tags and the target entity attributes, combine the influence logic in the data query instruction, and calculate the integrity index of the topological structure based on the mapping relationship, and mark the missing entities or missing associations related to the data query instruction to generate a missing marker set, including: Perform semantic annotation processing on the unstructured external data, extract the attribute category and numerical description, generate structured semantic tags, start from the target entity and traverse its associated topological structure, and identify all the conduction paths between entities and the attribute associations on the paths; Establish the mapping relationship between the structured semantic tags and the target entity attributes, combine the influence logic in the data query instruction, calculate the integrity index of each conduction path in the mapping relationship, and synchronously detect the association breakpoints that violate the influence logic in the conduction path; When the integrity index is lower than the preset threshold or there is an associated breakpoint, mark the corresponding entity as a missing entity or the association as a missing association, aggregate the topological coordinates of all the missing entities and missing associations, and generate a missing marker set.

[0014] Optionally, embed the topological nodes and topological connections into the missing marker positions of the power market data graph, reconstruct the topological structure, and verify the logical connectivity of the topological structure after completion based on the relationship conduction rules, including: Locate the entity missing positions and association missing positions in the power market data graph according to the missing marker set, embed the topological nodes into the entity missing positions and write the attribute coding values, embed the topological connections into the association missing positions and configure the weight coefficients, and reconstruct the topological structure; Traverse the reconstructed topological structure based on the relationship conduction rules, and verify the conduction direction, weight decay gradient, and time sequence continuity of the newly added conduction paths; If the conduction direction conforms to the preset constraints, the weight decay gradient is within the allowable range, and the time sequence markers are continuous, it is determined that the conduction path is logically connected, and record the conduction paths that fail the logical connectivity verification and the failure rules.

[0015] In a second aspect, the present application provides a semantic enhancement and dynamic completion system for a power market data graph, including: An acquisition module, in response to a data query instruction sent by a user, acquires the target entity information running in a structured manner and unstructured external data in a pre-established power market data graph; A mapping module, converts the external data into structured semantic tags, traverses the topological structure of the target entity information, establishes a mapping relationship between the structured semantic tags and the target entity attributes, combines the influence logic in the data query instruction, calculates the integrity index of the topological structure based on the mapping relationship, and marks the missing entities or missing associations related to the data query instruction to generate a missing marker set; A generation module, based on the target entity information and the external data, establishes an attribute mapping between the external data attributes and the entity attributes, generates a semantic vector integrating cross-domain features by analyzing the implicit relationship between the external data and the target entity information in the attribute mapping, projects the semantic vector into the topological space according to the missing marker set, and generates topological nodes and topological connections inheriting cross-domain features; A reconstruction module, embeds the topological nodes and topological connections into the missing marker positions of the power market data graph, reconstructs the topological structure, and verifies the logical connectivity of the topological structure after completion based on the relationship conduction rules; The verification module, based on the verified topological structure, parses the dynamic conduction chain formed by the target entity information through the complemented path, and generates a feedback interface containing semantic relationships and complemented structures to respond to the data query instruction.

[0016] This application collects target entities and external data in response to data query instructions, establishes the mapping relationship between structured semantic tags and entity attributes to achieve cross-domain data semantic connection; combines the impact on the integrity of the logical calculation topology and marks the missing set, breaks through the limitations of static detection to accurately locate the conduction fault; constructs attribute mapping to analyze implicit relationships and generates semantic vectors integrating cross-domain features, projects to form topological elements inheriting external characteristics; embeds the missing positions, reconstructs the topological structure, and then verifies the logical connectivity based on the relationship conduction rules to ensure that the complemented path conforms to the power market constraints; finally, parses the dynamic conduction chain to generate visual feedback, forming a technical closed-loop from cross-domain data fusion, rule constraint complementation to interpretable output of the conduction path.

[0017] Furthermore, by calling the pre-constructed ontology layer to identify the domain categories and their relationship types of power market entities, extracting the associated neighbor nodes and multi-hop path features from the data layer to form entity context features, fusing the domain category information and context features to generate domain-enhanced semantic vectors, and injecting them into the cross-domain semantic vectors to achieve dual feature enhancement; this mechanism eliminates potential conflicts between external data and power market rules through domain knowledge constraints, such as correcting the deviation between meteorological load prediction and the actual regulation capacity of units, and at the same time optimizing the transmission capacity constraints of cross-regional topological connections, ensuring that the generated topological nodes and connections not only inherit cross-domain data features but also conform to power operation rules, and finally completing domain adaptability enhancement at the semantic vector layer, providing dual verification guarantees of domain rules and external features for topological projection.

[0018] These aspects or other aspects of this application will be more clearly understood in the following description of the embodiments. Description of the Drawings

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of this application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0020] Figure 1 Shows the flowchart of a method for semantic enhancement and dynamic complementation of a power market data graph provided by this application; Figure 2 Shows the scenario diagram of a method for semantic enhancement and dynamic complementation of a power market data graph provided by this application; Figure 3The structure diagram of a semantic enhancement and dynamic completion system for an electricity market data graph provided by this application is shown. Detailed implementation manners

[0021] In order to enable those skilled in the art to better understand the solution of this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this application.

[0022] In some processes described in the specification, claims and the above-mentioned drawings of this application, there are multiple operations that appear in a specific order. However, it should be clearly understood that these operations may not be executed in the order in which they appear in this article or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. 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 such as "first" and "second" in this article are used to distinguish different messages, devices, modules, etc., do not represent the sequence, and do not limit that "first" and "second" are of different types.

[0023] Researchers have found that existing electricity market data analysis methods have serious deficiencies: relying on predefined rules and static data, it is difficult to integrate multi-source heterogeneous information; when performing data completion, the conduction mechanism unique to the market is ignored, resulting in frequent logical conflicts in the reconstructed topological structure; the completion results are presented in isolation, unable to clearly display the dynamic conduction path of information, and it is difficult to provide traceability support for decision-making. Therefore, there is an urgent need for a dynamic completion method that can deeply integrate data, ensure logical self-consistency, and visualize the conduction chain.

[0024] In response to the above problems, the present invention proposes a semantic enhancement and dynamic completion method for an electricity market data graph. The core lies in using attribute mapping to generate semantic vectors that fuse cross-domain features, and realizing logically self-consistent dynamic completion and conduction chain analysis under the constraint of conduction rules. Specifically, the method responds to a query instruction, collects target entity information and external data; structures the external data and analyzes the topological integrity, and marks the missing points; establishes attribute mapping and analyzes the implicit relationship, generates a fused semantic vector, and intelligently projects it to the topological missing position; applies the relationship conduction rule to strictly verify the logical connectivity of the new structure; finally, analyzes and displays the dynamic conduction chain formed by the completed entities and their semantic relationships. This method effectively overcomes the limitations of predefined rules, realizes deep integration of cross-domain data; avoids topological logic conflicts caused by ignoring the conduction mechanism; and accurately restores the dynamic conduction path, significantly improving the traceability support ability for market decision-making.

[0025] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application.

[0026] Figure 1 The flowchart of a semantic enhancement and dynamic completion method for a power market data graph provided by an embodiment of the present application is as follows Figure 1 As shown, the method includes: 101. In response to a data query instruction sent by a user, collect the target entity information operating in a structured manner and unstructured external data in a pre-established power market data graph; In the above step, the data query instruction refers to a command sent by the user to request power market data. The power market data graph refers to a graph structure that describes power market entities and their relationships in advance. The entity information operating in a structured manner refers to data entities stored in a structured format and capable of being operated. The target entity information refers to the specific data entity content specified by the user instruction. The unstructured external data refers to data from external sources that does not follow a fixed format, and the collected external data can be specifically obtained according to the data required by the user for query.

[0027] In the embodiments of the present application, first, in response to a data query instruction sent by the user, receive and parse the instruction to determine the target entity information and unstructured external data requirements to be collected; second, based on the parsing result, collect the target entity information operating in a structured manner from the pre-established power market data graph and extract the required data; then, further obtain the corresponding unstructured external data according to the specific content of the user query instruction. For example, when the user asks about the correlation between power load and meteorology, external data is obtained from the meteorological bureau data source, or when the user asks about the impact scope of the electricity price policy, relevant external policy documents are obtained; finally, complete the entire collection process to ensure that the target entity information and external data are successfully integrated and obtained.

[0028] In practical applications, in a certain power market operation platform, the user inputs a data query instruction about recent regional power supply and demand through the interface, and the system immediately responds to the instruction, automatically extracts the structured operation information of 7 target entities from the pre-established power market data graph, including the maximum output value of the power plant and the node electricity price data. At the same time, 3 unstructured market analysis reports and policy documents are collected from external websites using a crawler tool; the entire collection process is completed within 15 milliseconds, successfully integrating key data, greatly enhancing the information analysis ability and decision-making support ability, and helping users efficiently identify potential market risks.

[0029] After receiving the user's data query instruction, the system can obtain structured target entity information from the pre-established power market data graph, and at the same time collect unstructured external data, effectively integrating the two data types to achieve a comprehensive query response, thereby improving the efficiency of data mining and the depth of results. This process ensures that the query results incorporate the internal structured operation targets and rich unstructured supplementary information, providing accurate and detailed power market data analysis support for users.

[0030] 102. Convert the external data into structured semantic tags, traverse the topological structure of the target entity information, establish the mapping relationship between the structured semantic tags and the target entity attributes, combine the influence logic in the data query instruction, calculate the integrity index of the topological structure based on the mapping relationship, and mark the missing entities or missing associations related to the data query instruction to generate a missing tag set; Optionally, step 102 may specifically include the following steps: 1021. Perform semantic annotation processing on the unstructured external data, extract the attribute categories and numerical descriptions, generate structured semantic tags, traverse the associated topological structure starting from the target entity, and identify all the conduction paths between entities and the attribute associations on the paths; 1022. Establish the mapping relationship between the structured semantic tags and the target entity attributes, combine the influence logic in the data query instruction, calculate the integrity index of each conduction path in the mapping relationship, and synchronously detect the associated breakpoint that violates the influence logic in the conduction path; 1023. When the integrity index is lower than the preset threshold, or there is such an associated breakpoint, mark the corresponding entity as a missing entity or the association as a missing association, summarize the topological coordinates of all the missing entities and missing associations, and generate a missing tag set.

[0031] In the above steps, the structured semantic tag refers to the structured identifier that converts unstructured external data into an identifier containing attribute categories and numerical descriptions through semantic annotation. The topological structure of the target entity information refers to the network structure that describes its associated entities and conduction relationships starting from the target entity. The conduction path refers to the attribute conduction link between entities in the topological structure. The attribute association refers to the dependency relationship between entity attributes on the conduction path. The mapping relationship refers to the corresponding connection between the structured semantic tag and the target entity attribute. The influence logic refers to the attribute change conduction rule defined in the data query instruction. The integrity index refers to the quantitative evaluation value of the attribute coverage rate of a single conduction path in the mapping relationship. The association breakpoint refers to the breakpoint of the conduction path that violates the influence logic. The missing entity refers to the entity whose integrity does not meet the standard or is located at the breakpoint. The missing association refers to the association edge in the conduction path that does not meet the influence logic. The topological coordinate refers to the position identifier of the missing entity or association in the topological structure. The missing tag set refers to the set that summarizes all topological coordinates.

[0032] In the embodiment of the present application, first, in step 1021, through semantic annotation processing of unstructured external data, the entity recognition algorithm is used to extract attribute categories and numerical descriptions to generate structured semantic tags; secondly, starting from the target entity, the breadth-first traversal algorithm is used to scan the topological structure to identify all conduction paths between entities and record the attribute associations on the paths; finally, the generated structured semantic tags and the topological data set containing the attribute associations of the conduction paths are output as the input for the subsequent steps.

[0033] Then, through step 1022, a mapping relationship table between the structured semantic tag and the target entity attribute is established for the topological data set; secondly, each conduction path is traversed, and the attribute matching algorithm is used in combination with the influence logic rule in the data query instruction to calculate the attribute coverage rate of the path in the mapping relationship as the integrity index. In the calculation of the attribute coverage rate, first, based on the attribute matching algorithm, each node attribute in the conduction path is traversed, and the mapping relationship table is used to check whether there is a match for the attribute, then the number of matching attributes is counted, and it is divided by the total number of attributes in the mapping table to obtain a ratio value. The formula is: attribute coverage rate = (number of matching attributes / total number of attributes) × 100%); then, the association breakpoints that violate the influence logic in the conduction path are synchronously detected, for example, by checking for variable missing or conflicts through a logic rule engine; finally, the conduction path analysis set with the integrity index and association breakpoints is output.

[0034] Finally, when the integrity index of a certain path in the conduction path analysis set is lower than the preset threshold or there are association breakpoints, the marking algorithm is used to mark the corresponding uncovered entity as a missing entity and the broken association edge as a missing association; secondly, the topological coordinates of all missing entities and missing associations in the topological structure are aggregated, for example, the position is recorded through the coordinate identification algorithm; finally, all topological coordinates are summarized to generate a missing tag set.

[0035] In practical applications, in a certain power market analysis system, after the system obtains the query instruction for regional power balance data submitted by the user, it first performs semantic parsing on the 3 unstructured policy reports collected, and extracts three types of structured semantic tags: "environmental protection production limit ratio of coal-fired power plants", "capacity of cross-provincial power transmission channels", and "adjustment of new energy subsidy policies". Subsequently, starting from the 4 target power plants involved in the instruction, it traverses the connected power transmission network topology structure and identifies 12 conduction paths with attributes such as power plant output value, line transmission limit, and node load. The system automatically establishes the mapping relationship between semantic tags and target entity attributes. For example, it maps the "environmental protection production limit ratio" to the available output upper limit attribute of coal-fired power plants, and based on the logic of "policy changes affecting power supply reliability" in the user instruction, uses the formula to calculate the integrity index of each conduction path one by one. The formula is: Integrity index of the conduction path = Σ(Existence flag of the mapped attribute × Weight coefficient) / Total number of required attributes of the path. When it is found that the mapping integrity of a path containing a cross-provincial power transmission line is only 75% due to the lack of grid structure adjustment data, which is lower than the preset threshold of 85%, and there is an interruption in the association between the environmental protection production limit policy and the unit maintenance plan in another path, the system immediately marks 2 coal-fired power plants as missing entities and 3 cross-regional tie lines as missing associations, and finally generates a missing mark set containing 5 topological coordinates to accurately locate the data gap affecting the power supply reliability assessment.

[0036] In the overall solution of step 102 above, by converting unstructured external data into structured semantic tags, establishing a mapping relationship based on the target entity topology structure, and combining the impact logic to calculate the integrity index of the conduction path, the data missing problem can be systematically detected and marked. After generating semantic tags by extracting attribute features through semantic annotation in the data conversion stage, it traverses the topology structure starting from the target entity to identify all conduction paths and their attribute associations. In the mapping analysis stage, it verifies the association continuity of each path based on the impact logic and calculates the integrity index. Finally, it marks the missing topological coordinates with integrity lower than the threshold or with association interruption. The whole process realizes the in-depth fusion analysis of unstructured data and structured graphs. Through the quantitative evaluation of the conduction path integrity and the location of break points, it accurately outputs the missing entities and association mark sets with topological coordinates.

[0037] 103. Based on the target entity information and the external data, establish an attribute mapping between the external data attributes and the entity attributes. By analyzing the implicit relationship between the external data and the target entity information in the attribute mapping, generate a semantic vector integrating cross-domain features. According to the missing mark set, project the semantic vector into the topological space to generate topological nodes and topological connections inheriting cross-domain features; Optionally, step 103 may specifically include the following steps: 1031. Based on the target entity information and the external data, construct a bidirectional mapping dictionary between external data attributes and entity attributes, and record the direct mapping rules, mathematical derivation formulas, and unit conversion coefficients between the attributes. 1032. Through a deep semantic association model, analyze the implicit multi-order derivation relationships between the attributes in the bidirectional mapping dictionary, and identify cross-domain attribute combinations with common change characteristics or causal constraints. Among them, step 1032 may specifically include the following process: Through the deep semantic association model, receive a set of mapping entries from the bidirectional mapping dictionary. The set of mapping entries records the corresponding rules and relationship identifiers between external data attributes and target entity attributes, and construct a derivation network with the set of mapping entries; perform a traversal operation on the derivation network, identify the consistency of the time dimension fluctuation characteristics between the nodes in the derivation network, and record the co-changing attribute groups. Detect the conduction paths that meet the causal constraint rules in the derivation network and record the response sequences; calculate the fluctuation consistency quantization value of the co-changing attribute groups, and measure the trigger delay time of the response sequences; when the fluctuation consistency quantization value exceeds the set threshold and the trigger delay time is within the preset interval, output the cross-domain attribute combination with the relationship identifier.

[0038] 1033. Generate semantic vectors for each of the cross-domain attribute combinations. The semantic vectors fuse the encoded values, relationship weight coefficients, and time sequence markers of the associated attributes. 1034. According to the topological coordinates located by the missing marker set, create blank topological nodes in the power market data graph, and project the semantic vectors onto the corresponding blank topological nodes to form topological nodes carrying cross-domain features. 1035. According to the relationship weight coefficients in the semantic vectors, establish weighted topological connections inheriting cross-domain features between the topological nodes carrying cross-domain features. The attributes and connection relationships of the topological nodes and topological connections need to meet the topological constraint rules of the power market graph.

[0039] In the above steps, attribute mapping refers to a set of association rules between external data attributes and target entity attributes. A two-way mapping dictionary refers to a structured dictionary that records the mathematical derivation formulas of direct mapping rules between attributes and unit conversion coefficients. An implicit multi-level derivation relationship refers to a multi-level mathematical or logical conduction association between attributes that is not explicitly defined. A cross-domain attribute combination refers to a group of attributes in different data domains with common variation characteristics or causal constraints. A derivation network refers to a network of derivation relationships between attributes constructed from mapping entries. A co-varying attribute group refers to a set of attributes with consistent fluctuation characteristics in the time dimension. A response sequence refers to a sequence of attributes of a conduction path that satisfies causal constraint rules. A fluctuation consistency quantization value refers to a numerical measure of the temporal fluctuation similarity of a co-varying attribute group. A trigger delay time refers to the response time interval of a causal constraint conduction path. A semantic vector refers to a feature vector that fuses the relationship weight coefficients of attribute coding values and temporal markers. A blank topological node refers to an unassigned node created in the power market data graph according to a missing marker set. A topological connection inheriting cross-domain features refers to a node connection link established according to the relationship weight coefficients of semantic vectors. Topological constraint rules refer to the logical restrictions on node and connection relationships predefined in the power market data graph.

[0040] In the embodiment of the present application, first, attribute pairs are extracted based on the target entity information and external data through step 1031. Secondly, a key-value pair matching algorithm is used to construct a two-way mapping dictionary to record the mathematical derivation formulas of direct mapping rules between attributes and unit conversion coefficients. Then, a set of mapping entries is formed through a structured storage technology. For example, a mapping relationship is established between the temperature attribute of meteorological data and the load attribute of the distribution network, and the conversion formula from temperature to load is recorded. Then, the set of mapping entries is output as the input for the subsequent steps. Finally, the construction process of the two-way mapping dictionary is completed.

[0041] Secondly, an operation process is executed based on the set of mapping entries through step 1032: First, a derivation network is constructed through a graph structure conversion algorithm. The source attribute and target attribute of each mapping entry are converted into network nodes with metadata, and the metadata includes value type, range, and unit. For example, when assigning the node identifier N15 to the meteorological temperature, the metadata value type is defined as float32, the range is defined as , the unit is defined as °C, and when assigning the node N22 to the power load, the metadata value type is defined as float32, the range is defined as , the unit is defined as MW; a directed edge with a mathematical relationship expression is established between the nodes, and the relationship expression such as N15 pointing to N22 stores the conversion formula: , where the letter load represents the load value and the letter temp represents the temperature value in the formula. Furthermore, the dynamic time warping algorithm is used to traverse the derivation network to identify the consistency of the fluctuation characteristics in the time dimension between the nodes. The specific processing process includes extracting the time series data corresponding to the nodes, such as the temperature series T equal to the series value and the load sequence L is equal to the sequence value , 290.0; perform linear interpolation on the missing values, and use the formula to complete the filling at the missing position k. The formula letter represents the value before the position k, and the letter represents the value after the position k, and the letter represents the value after interpolation; the normalization process uses the formula , where the letter x in the formula represents the original data value, and the letter represents the sequence mean, and the letter represents the sequence standard deviation, and the letter represents the value after normalization; construct a dynamic time warping matrix to calculate the path distance, and the matrix elements are recursively calculated using the formula: , where the letter represents the th data point of the temperature sequence, and the letter represents the jth data point of the load sequence, and the letter represents the minimum cumulative distance from the start point of the sequence to the position , and the letter min represents the minimum value operation; output the fluctuation - consistency quantization value , where the letter d in the formula represents the cumulative distance calculated by dynamic time warping. When d is equal to 0.087, Sim is calculated to be 0.92; finally, record the attribute groups with fluctuation consistency, such as the combination of N15 and N22.

[0042] Next, through step 1033, based on cross - domain attribute combinations, such as combination records including meteorological temperature and power load attributes, first use the word embedding technology to extract the associated attributes of each combination to generate attribute coding values. Among them, the word embedding technology uses a pre - trained Word2Vec model to process the attribute names and relationship description texts, and generates numerical vectors with a fixed dimension. For example, generate a coding value of 0.72 for the meteorological temperature attribute and a coding value of 1.15 for the power load attribute, forming an attribute coding vector [0.72, 1.15]. Secondly, use the weight calculation algorithm according to the relationship identification field to generate relationship weight coefficients. Among them, the weight calculation algorithm is based on the relationship index data in the combination, such as the fluctuation - consistency Sim value and the delay time, and calculates the cosine similarity score. The formula for the weight coefficient is , where the letter A represents the Sim value vector and the letter B represents the delay time vector. Then add a time series marker to record the time - varying characteristics of the attributes, extract the minimum time interval from the timestamp field of the combined data and calculate the time difference. For example, calculate the time feature equal to 8 minutes for the difference in the sequence change time points of meteorological temperature and power load. Finally, through the feature fusion technology, fuse the attribute coding values, relationship weight coefficients, and time series markers into a semantic vector. Among them, the feature fusion uses the weighted average formula , where the letter V represents the output semantic vector, the letter w represents the weight coefficient, the letter E represents the attribute encoding vector, and the letter represents the time feature value. For example, the encoding vector [0.72, 1.15] formed by fusing the meteorological temperature encoding value 0.72 and the power load encoding value 1.15, combined with the weight 0.92 and the time series feature of 8 minutes, outputs a 32-dimensional semantic vector after weighted average (such as the specific value can be represented as , … to represent the cross-domain semantic features), forming a semantic representation that can be directly used for subsequent knowledge graph construction. Conversely, if the attribute encoding value is missing or the weight coefficient is lower than the 0.8 threshold, the fusion is skipped and the data is marked as incomplete.

[0043] Then, based on the missing marker set and the semantic vector output in step 1033 through step 1034, first use the graph database indexing algorithm to locate the topological coordinates in the missing marker set. Secondly, create a blank topological node at the corresponding position in the power market data graph through the node creation interface. Then, use the vector projection algorithm to map the semantic vector to the blank topological node. Finally, form a topological node carrying cross-domain features. For example, create a blank node with load meteorological features at the missing position of the distribution network node and complete the vector projection process.

[0044] Finally, access the semantic vector field of the topological node through step 1035, locate the weight coefficient field at index positions 128 to 135 according to the predefined data structure, and directly read the field value through the array indexing method. For example, when processing the meteorological temperature node, extract the value 0.92 stored at index 130 from its associated semantic vector as the weight coefficient. Call the connection line generation algorithm to perform node unique identifier matching to identify the correspondence between the source node ID = N15 and the target node ID = N22. Calculate the node spatial positions based on the topological coordinate system, such as the source node coordinates (120, 80) and the target node coordinates (310, 150)), and use the Bresenham line algorithm to generate a set of pixel points between the two points to create a black solid connection line segment with a width of 1 pixel. The algorithm actually calculates and generates an oriented connection line segment entity containing 210 pixel points. Write the parsed weight coefficient value 0.92 into the weight attribute field of the connection line entity, and at the same time automatically configure the visualization parameters according to the preset rules: the weight value 0.92 belongs to the range of 0.9 to 1.0, and set the connection line to a 5-pixel-wide red solid line format. The topological constraint engine loads the rule "Unidirectional Constraint from Environment Domain → Power Domain" (ID = R007) in the rule library, and verifies the source node metadata. The domain type field value of the meteorological temperature node is "Environment Domain"; the domain type field value of the grid load node is "Power Domain"; the connection direction detection confirms that the arrow points from the environment domain to the power domain, and all three verification results meet the rule requirements. Assign a unique connection ID = CONN_001, record the correspondence between the source node {ID: N15, attribute: "Meteorological Temperature", domain type: "Environment Domain"} and the target node {ID: N22, attribute: "Grid Load", domain type: "Power Domain"}, bind the weight value 0.92, mark the verification status as PASS and associate the rule ID = R007. Finally, render and output a 5-pixel-wide red connection line in the topological space to generate a topological connection entity carrying complete metadata.

[0045] In practical applications, in a certain smart grid modeling system, based on the structured operation information of 4 target power plants and 3 collected external policy reports, the system first constructs a bidirectional mapping dictionary containing 5 entries, which details the direct correspondence rules, mathematical conversion formulas, and unit conversion coefficients between external data attributes and entity attributes. Subsequently, the dictionary data is parsed through a pre-trained deep semantic association model, which automatically constructs a derivation network with 28 nodes, traverses the network to identify the time-fluctuation consistency features between attributes, and simultaneously detects the conduction paths that meet the causal constraints, calculates the fluctuation consistency quantization value, and measures the response delay time. When the quantization value exceeds the set threshold and the delay time is within the preset interval, 4 groups of cross-domain attribute combinations with causal identifiers are automatically output. Then, semantic vectors integrating multi-domain features are generated for each group of attribute combinations, where the encoded values include attribute numerical encodings and relationship strength coefficients, and temporal marker information is additionally attached. Based on 5 topological coordinates located by the previously generated missing marker set, the system creates 5 blank topological nodes in the power market data graph and precisely projects the semantic vectors to the corresponding node positions, forming entity node units carrying policy analysis features. Finally, based on the relationship strength coefficients embedded in the semantic vectors, weighted topological connection links are automatically constructed between the newly created nodes, and strictly following the topological rules such as the node degree constraint and thermal stability threshold of the power market graph, the compliance of all connection relationships is ensured. The entire process realizes the deep integration of unstructured policy data and structured power grid entity features, providing cross-domain feature support for market risk prediction.

[0046] In the overall solution of step 103 above, through constructing a derivation network to perform topological traversal, common change attribute groups with time-dimensional fluctuation consistency between nodes are identified, and the conduction path response sequence that meets the causal constraint rules is detected. At the same time, the fluctuation consistency quantization value is calculated and the trigger delay time is measured. When the quantization value exceeds the set threshold and the delay is within the preset interval, cross-domain attribute combinations with relationship identifiers are generated. Semantic vectors integrating associated attribute encodings, relationship weight coefficients, and temporal markers are dynamically generated for each cross-domain attribute combination. For the topological coordinates located by the missing marker set, blank nodes are created in the power market data graph, and the semantic vectors are projected to the corresponding nodes to form topological node entities carrying cross-domain features. Finally, based on the relationship weight coefficients in the semantic vectors, weighted topological connections are established between the nodes to ensure that all node attributes and connection relationships strictly follow the topological constraint rules of the power market graph, thereby completing the topological reconstruction of the cross-domain features in the power market data graph.

[0047] 104. Embed the topological nodes and topological connections into the missing marker positions of the power market data graph, reconstruct the topological structure, and based on the relationship conduction rules, verify the logical connectivity of the topological structure after completion. Optionally, step 104 may specifically include the following steps: 1041. Locate the entity missing positions and associated missing positions in the power market data graph according to the missing tag set, embed the topological nodes into the entity missing positions and write the attribute coding values, embed the topological connections into the associated missing positions and configure the weight coefficients, and reconstruct the topological structure; 1042. Traverse the reconstructed topological structure based on the relationship conduction rules, and verify the conduction direction, weight attenuation gradient and time sequence continuity of the newly added conduction paths; 1043. If the conduction direction conforms to the preset constraints, the weight attenuation gradient is within the allowable range and the time sequence markers are continuous, then it is determined that the conduction path is logically connected, and record the conduction paths that fail to pass the logical connectivity verification and the failure rules.

[0048] In the above steps, the relationship conduction rule refers to the direction constraints, weight attenuation gradient and time sequence continuity conditions predefined in the power market field. Logical connectivity refers to the effective conduction state after the newly added conduction path meets the relationship conduction rules. The attribute coding value refers to the characteristic coding numerical value embedded by the topological node. The weight coefficient refers to the relationship weight value configured by the topological connection. Reconstructing the topological structure refers to the updated network structure after embedding nodes and connections. The newly added conduction path refers to the conduction link between entities newly generated in the reconstructed topological structure. The conduction direction refers to the flow direction of attribute changes in the conduction path. The weight attenuation gradient refers to the change rate numerical value of the weight coefficient on the conduction path. The time sequence continuity refers to the smooth and continuous state of the time markers in the conduction path. The preset constraints refer to the direction matching requirements set in the relationship conduction rules. The allowable range refers to the tolerance interval of the change value of the weight attenuation gradient. The continuous time sequence marker refers to the conduction state without interruption in the time interval change. The conduction path that fails to pass the logical connectivity verification refers to the conduction link that fails the verification. The failure rule refers to the specific constraint rule identifier that causes the verification to fail.

[0049] In the embodiment of this application, first, locate the entity missing positions and associated missing positions according to the missing tag set in step 1041, and use the Cypher query language of the graph database to retrieve the topological coordinates to be complemented in the power market data graph; secondly, call the graph node creation API to embed the topological nodes into the entity missing positions and write the attribute coding values, for example, create a meteorological feature node at the missing coordinate of the distribution network node and write the coding value [0.72, 1.15]; then, embed the topological connections into the associated missing positions and configure the weight coefficients through the edge creation interface, for example, establish a connection edge with a weight of 0.92 between the meteorological node and the load node; finally, reconstruct the topological structure to generate a network structure including the newly added nodes and connections, and output the updated power market data graph.

[0050] Secondly, traverse the reconstructed topological structure based on the relationship conduction rule through step 1042. First, use the breadth-first traversal algorithm to scan all newly added conduction paths. Secondly, calculate the dot product of the direction vectors to verify whether the conduction direction conforms to the preset constraints, such as verifying the one-way rule of "meteorological domain → power domain". Then calculate the weight decay gradient difference value δ = ∣(w1 - w2) / d∣ (w is the weight, d is the path length), and verify whether it is within the preset allowable range of 0.05 - 0.2. Finally, use the time series dynamic programming algorithm to detect the time series continuity to ensure that the time mark interval conforms to the continuous threshold of Δt ≤ 1 hour, such as checking the 15-minute sampling interval continuity of meteorological data and load data.

[0051] Finally, through step 1043, if the conduction direction conforms to the preset constraints, the weight decay gradient is within the allowable range, and the time series marks are continuous, then mark the conduction path as logically connected; if the dot product value of the direction vector < 0.8, then record the failure of the direction constraint, if δ > 0.2, then record the failure of the gradient exceeding the limit, if the time series fault > 1 hour, then record the failure of the time series interruption; finally, output a verification report containing the ID of the failed path and the corresponding failure rule, such as recording that the weight decay failure of path P35 is caused by δ = 0.28 exceeding the threshold.

[0052] In practical applications, in the digital twin platform of a regional power grid, the system performs the following operations: According to the missing marker set containing the missing coordinates of 2 coal power plants and 3 transmission lines, accurately locate 5 defective positions in the power market data graph. Embed the topological node unit carrying cross-domain features into the missing positions of the 2 coal power plant entities, and write the 0.75 output encoding value mapped by the environmental protection production restriction policy; at the same time, embed the topological connection link with weight coefficients into the missing positions associated with the 3 transmission lines, and configure weight coefficients of 0.62, 0.79, and 0.85 respectively to complete the reconstruction of the topological structure. Then start the full-topology traversal based on the power supply and demand conduction rule: check whether the conduction direction of the newly added cross-provincial transmission path conforms to the constraint rule of "power plant → load center", verify whether the decay gradient of the weight along the transmission distance is controlled within the allowable threshold of ±0.1 per 100 kilometers, and check the timestamp continuity of the new energy output data and the actual load curve. When it is found that the directional check of a reconstructed tie line fails due to the lack of associated regional load data, and there is a 4-hour time fault in the output fluctuation mark of another line, the system automatically records the 2 failed paths and the corresponding violations of the "power flow unidirectionality rule" and "time series synchronization rule" to ensure that the reconstructed graph has strict interpretability and business logic consistency.

[0053] In the overall solution of step 104 above, by embedding topological nodes carrying cross-domain features into entity missing positions and writing attribute coding values, and at the same time embedding weighted topological connections into associated missing positions and configuring weight coefficients, the integrity reconstruction of the topological structure is achieved. Subsequently, based on the relationship conduction rules, the reconstructed topological structure is traversed, and three-dimensional logical verification is performed on the newly added conduction paths: verifying whether the conduction direction conforms to the preset constraint rules, detecting whether the weight decay gradient is within the allowable range, and checking whether there are continuous breaks in the timing marks. Only when the conduction direction is compliant, the weight decay gradient is controllable, and the timing marks are continuous, is it determined that the conduction path is logically connected. For conduction paths that fail the verification, the system will accurately record their path coordinates and the specific failure rules violated, form a logical connectivity verification report, and finally complete the closed-loop complementation and verification of the data graph on the premise of ensuring that the newly added topological elements meet the constraints of the power market graph ontology.

[0054] 105. Based on the verified topological structure, parse the dynamic conduction chain formed by the target entity information through the complementation path, and generate a feedback interface including semantic relationships and a complemented structure to respond to the data query instruction.

[0055] Optionally, step 105 may specifically include the following steps: 1051. Based on the verified topological structure, extract the dynamic conduction chain formed by the target entity information through the complementation path, and parse the relationship weight value and the timing mark trajectory of each conduction path in the dynamic conduction chain; 1052. Map the relationship weight value to a preset semantic intensity level, convert the timing mark trajectory into a time axis, generate a node relationship chain with a conduction direction identifier, and mark the semantic intensity level and the time axis on the node relationship chain; 1053. Convert the conduction paths that fail the verification into warning identifiers, integrate the node relationship chain and the warning identifiers, and generate a visual feedback interface to respond to the data query instruction.

[0056] In the above steps, the verified topological structure refers to the topological structure after the logical connectivity verification output in step 104. The target entity information refers to the entity data specified by the user data query instruction. The complemented path refers to the newly added conduction link formed through topological reconstruction and verification in steps 103 to 104. The dynamic conduction chain refers to the conduction sequence of attribute changes formed on the complemented path. The semantic relationship refers to the meaning association between nodes in the conduction chain. The complemented structure refers to the graph structure that includes newly added elements after update. The feedback interface refers to the visual display interface that responds to user queries. The relationship weight value refers to the relationship weight value configured in the topological connection. The time sequence marking trajectory refers to the time marking sequence in the semantic vector. The preset semantic intensity level refers to the predefined high, medium, and low-level mapping table. The time axis refers to the visual time axis for the conversion of time sequence markings. The conduction direction identifier refers to the direction arrow marking in the node relationship chain. The node relationship chain refers to the visual representation element of the dynamic conduction chain. The warning identifier refers to the warning icon for the conduction path that fails to pass the verification. The visual feedback interface refers to the visual display interface that integrates all elements. The conduction path that fails to pass the verification refers to the conduction path for which the logical connectivity verification fails in step 104.

[0057] In the embodiment of the present application, first, perform the following operation process based on the verified topological structure through step 1051: First, adopt the depth-first traversal algorithm to extract the dynamic conduction chain formed by the target entity information through the complemented path starting from the selected node. For example, select the generator node identifier G1 and start traversing and accessing adjacent nodes, including the transformer node T1 and the load node L1 to form a complete path where the generator G1 points to the transformer T1 and the transformer T1 points to the load L1; when it is detected that the intermediate nodes of the path are all complemented links, generate a conduction chain identifier. For example, assign the chain ID as DC001, corresponding to the complete path list . Second, use the parsing algorithm to scan each conduction path in the dynamic conduction chain one by one, and extract the numerical value of the relationship weight attribute in the path text field through the regular expression matching algorithm. For example, match the weight value 0.85 for the text description field of the conduction path and match the weight value 0.93 for the conduction path . Then, extract the time marking trajectory in the time stamp vector corresponding to the conduction chain through the time series extraction technique. Based on the time point sequence values, such as the specific values of the time stamp sequence being 10:00, 10:03, and 10:08, calculate the interval difference between adjacent time points using the formula , where the letter represents the time point serial number starting from 1, the letter represents the value of the th time point, the letter represents the value of the previous time point, and the letter represents the A time interval difference value is used to generate a time marker sequence The unit is hours. Finally, a structured dynamic conduction chain dataset is output

[0058] Secondly, the relationship weight value is input into the mapping algorithm through step 1052, and is matched with the preset semantic intensity level table to be converted into high, medium, and low semantic intensity levels; at the same time, the time series processing technology is used to convert the time series marker trajectory into a time axis visualization component; then, a node relationship chain with a conduction direction identifier is constructed through a visualization generation tool, and the direction is identified by an arrow; finally, the semantic intensity level and the time axis are marked on the node relationship chain to generate a complete conduction path display element

[0059] Finally, the unpassed conduction path is input into the conversion logic module through step 1053 and converted into a warning identifier such as a red warning icon; the node relationship chain and the warning identifier are fused through the interface integration module; then, the visualization engine is called to generate a feedback interface including semantic relationships and a complemented structure; finally, the feedback interface responds to the data query instruction, for example, when the user queries the association between power load and meteorology, a complete conduction chain is displayed and the warning mark of the interrupted path is highlighted

[0060] In practical applications, in a regional energy decision-making platform, the system analyzes 3 newly added cross-provincial dynamic conduction chains from 4 target power plant entities based on the verified reconstructed graph spectrum. First, the core path features are extracted: identify the 0.82 weight value of a certain coal power transmission channel and the continuous 72-hour output curve, and simultaneously capture the 0.67 weight value of the new energy consumption chain and the fluctuation trajectory including 5 time markers. Subsequently, semantic conversion is performed: the weight value is mapped to a three-level intensity identifier (weak / medium / strong), and the time series trajectory is converted into a three-segment time axis marked with "morning peak", "noon valley", and "evening peak". When generating a conduction chain with an arrow flow direction on the visualization interface, a dark red "strong association" identifier is marked on the coal power channel and the time axis is superimposed, while the consumption chain with a 4-hour data fault is converted into a flashing lightning warning icon. Finally, 3 dynamic conduction chains and 2 failure warning channels are integrated to generate an interactive feedback interface including a color flow topology map, a time series scale axis, and a floating warning box, dynamically responding to the power balance assessment instruction initially submitted by the user

[0061] In the overall solution of step 105 above, a query response closed-loop is achieved by parsing the dynamic conduction chain formed by the complementation path between target entities. Specifically, when executing, first extract the dynamic conduction chain generated in the verified topological structure, and accurately analyze the relationship weight values and timing marking trajectories of each conduction path; then map the relationship weight values to the preset semantic intensity levels, convert the timing marking trajectories into visual time axes, construct a node relationship chain with conduction direction identifiers, and mark the semantic intensity levels and time axis coordinates on the chain. For the conduction paths that fail in logical verification, convert them into eye-catching warning identifiers (such as flashing marks or highlighted borders). Finally, integrate the complete node relationship chain and warning identifiers into a unified visual panel to generate an interactive feedback interface containing three-level information.

[0062] For steps 101 to 105 above, the following is a specific embodiment: For example, when a user queries the trading volume data of each month in the electricity trading market in 2024 (as shown in Figure 2 the bar chart), the data for June is missing in the original graph. The present application dynamically executes the following process: First, in response to the user's query instruction for "trading volume of each month in 2024", the system collects structured operation data from the electricity market data graph (obtain the trading volume from January to May and from July to December, such as 8 million kWh in May and 9 million kWh in July), and at the same time obtains external unstructured information (such as the news report that "the continuous high temperature in June 2024 led to a 20% year-on-year increase in electricity load"). Through semantic analysis, the external data is converted into structured labels {time: 2024-06, event: high temperature, impact: 20% increase in electricity load}. When traversing the monthly conduction chain (January → February →... → December), it is found that the June node is missing and the adjacent paths are interrupted. Combining the logical relationship in the electricity market rules that "an increase in electricity load usually causes an increase in trading volume", calculate the topological integrity as 91.6% (lower than the 95% threshold), and finally mark the missing entity (trading volume in June) and the missing association (the conduction relationship between May and June, and between June and July).

[0063] Secondly, establish an attribute association rule according to the cross-domain fusion mechanism: historical data shows that for every 10% increase in electricity load, the trading volume on average increases by 8% (i.e., the impact coefficient is 0.8). Through in-depth analysis, identify the strong statistical association between the high temperature event and the trading volume (historical correlation coefficient 0.85), and generate a semantic vector with integrated features {high temperature intensity: 0.9, load increase amplitude: 0.2, impact coefficient: 0.8, time: 2024-06}. Create a June node at the topological missing position, calculate the trading volume complement value based on the May data (8 million kWh) and the load increase amplitude: 800×(1 + 0.2×0.8) = 9.28 million kWh, and establish the conduction relationships of May → June (weight 0.8) and June → July (weight 0.7).

[0064] Next, embed the newly generated June node and conduction relationship into the original graph to form a complete time sequence chain: January →... → May → June → July →... → December. Verify the rationality of the newly added path: the conduction direction conforms to the time sequence (May → June → July), the weight change is within the allowable range (0.8 → 0.7, decay rate 12.5% < 15% threshold), and the month marks are continuous without breaks, confirming the topological logic completeness.

[0065] Finally, generate a visual feedback interface as shown in Figure 2 : Bar chart: Display the complete 12-month data, where the June column (9.28 million kWh) is marked with a special color to complete the value.

[0066] Conduction chain identification: Mark "Dynamic Completion Based on High Temperature Events" and the influence coefficient 0.8 (not shown in the figure) on the path of May → June → July.

[0067] Data interpretation: Hover prompt "June data is calculated by a 20% increase in electricity load and historical association rules, and the conduction intensity with adjacent months > 0.7" (not shown in the figure).

[0068] Through cross-domain attribute fusion and topological dynamic completion, this application converts unstructured external data (high temperature events) into structured transaction volume data, inherits cross-domain semantic relationships in the reconstructed conduction chain (such as the highlighted path in the attached drawing), and finally outputs a visual result with complete logic and including the explanation of the completion process.

[0069] Optionally, the method further includes: for the target entity information, invoking the ontology layer in the pre-constructed power market data graph to identify the domain category to which the entity belongs and the domain relationship category between entities; in the data layer of the power market data graph, extracting neighbor node features and multi-hop path features associated with the identified entities and relationships as entity context features; fusing the domain category to which the entity belongs, the domain relationship category between entities, and the entity context features to generate a domain-enhanced semantic vector; injecting the domain-enhanced semantic vector into the semantic vector that fuses cross-domain features to output an enhanced semantic vector.

[0070] In the above steps, the ontology layer refers to a structured description system of predefined domain entity categories and relationship categories in the power market data graph. The domain category refers to the classification of the power market professional fields to which the entities belong. The domain relationship category refers to the identification of the relationship types that conform to the power market rules between entities. The domain entity type refers to the classification result of the entity domain identified according to the ontology layer. The neighbor node feature refers to the set of node attributes directly connected to the target entity. The multi-hop path feature refers to the sequence of node attributes reachable by the target entity through multiple associated paths. The entity context feature refers to the composite attribute set that integrates the neighbor node feature and the multi-hop path feature. The domain-enhanced semantic vector refers to the enhanced feature vector injected with domain knowledge. The enhanced semantic vector refers to the optimized feature expression that integrates the domain-enhanced semantic vector and the original semantic vector.

[0071] In the embodiment of the present application, first, the ontology layer of the power market data graph is called to identify the domain category to which the target entity belongs and the domain relationship category between entities, and the domain entity type and relationship type are output; secondly, based on the identified domain entity type and relationship type, the neighbor node features are extracted through a graph neural network in the data layer, and the multi-hop path traversal algorithm is synchronously used to extract the path features of the associated nodes within three hops, and the two are fused to generate the entity context feature; then, the domain entity type relationship type and the entity context feature are input into the feature fusion model to generate the domain-enhanced semantic vector; finally, the domain-enhanced semantic vector is injected into the semantic vector of the original fused cross-domain feature, and the enhanced semantic vector is output through the vector splicing layer. For example, the entity type of the generator set and the transmission relationship type are injected into the meteorological-load semantic vector to form an optimized vector containing the domain weight coefficient.

[0072] In practical applications, in a certain power grid intelligent analysis system, the system calls the ontology layer of the pre-constructed power market data graph, and identifies that 3 out of 4 target power plants belong to the "coal power field" and 1 belongs to the "new energy field", and parses 6 types of domain relationships such as "fuel supply relationship" and "power transmission relationship". Subsequently, feature extraction is performed in the data layer: for a certain coal power plant entity, the load data of 4 neighbor nodes such as its coal supplier and power transmission and transformation facilities are extracted; the three-hop path feature of "coal power - power transmission - industrial user" is extracted along the ontology path, and the price fluctuation features of 7 nodes on the path are captured. Then, the domain entity type identifier and the entity context feature are fused to generate a domain-enhanced semantic vector containing the domain code 0.88 and the supply stability score. Finally, this vector is injected into the cross-domain semantic vector generated by the external policy feature, so that the original output power coding value of 0.75 is enhanced to 0.82, and the reconstructed topological nodes show the coupling effect of fuel cost fluctuation and environmental protection policy in the feedback interface, making the market risk assessment result more in line with the actual business scenario.

[0073] This application accurately identifies the domain category to which the target entity belongs and the domain relationship category between entities by invoking the ontology layer of the power market data graph, and clearly outputs professional entity types such as power generators and electricity sellers, as well as relationship types such as transmission contracts and dispatching instructions. Based on this domain knowledge, multi-hop neighbor node features and conduction path features of the target entity are deeply extracted at the data layer, and entity context features including topological constraint relationships are dynamically output. Subsequently, by integrating domain entity type identifiers, relationship type tags, and entity context features, domain-enhanced semantic vectors carrying market rule logic are generated. Finally, the domain-enhanced semantic vectors are injected into the original cross-domain feature semantic vectors to output enhanced semantic vectors that carry both external data relevance and power market professional rules. This process ensures that the semantic feature representation in the subsequent topological projection step has both the breadth of cross-domain data association and meets the professional depth requirements of the power market domain, significantly improving the adaptation accuracy of the feature representation to power trading scenarios.

[0074] Optionally, the method further includes: in the topological space of the power market data graph, based on the structural similarity and attribute similarity between entities, locating cross-domain entities associated with the target entity information, and outputting the cross-domain entities and their association relationships; using the cross-domain entities and the target entity as anchor points, dynamically constructing an inference sub-network including multi-hop neighbors in the topological space, and performing logical rule inference along the multi-hop path in the dynamic inference sub-network, and outputting an inference result; fusing the target entity information, the cross-domain entities and their association relationships, and the inference result to generate a joint semantic representation, and complementing the inference result to the target entity information to output the complemented target entity information.

[0075] In the above steps, structural similarity refers to the connection structure matching degree of entities in the topological space of the power market data graph. Attribute similarity refers to the cosine similarity calculation result of numerical or categorical features between entities. Cross-domain entities refer to external entities that are identified through association mapping and have cross-domain associations with the target entity information. Association relationship refers to the attribute conduction or logical dependence connection between cross-domain entities and the target entity. Anchor points refer to the target entity and cross-domain entities that serve as the starting points of reasoning. The inference sub-network refers to a local network composed of multi-hop neighbor nodes expanded with the anchor points as the center. Multi-hop path refers to a conduction path that spans two or more associated edges in the inference sub-network. Logical rule inference refers to the calculation process of executing the preset rules of the power market (such as the supply-demand balance formula) along the multi-hop path. The inference result refers to the attribute derivation value or status judgment output by the logical rule inference. The joint semantic representation refers to a feature vector that fuses the target entity information, the cross-domain entity association relationship, and the inference result. The complemented target entity information refers to the enhanced entity data injected with the inference result.

[0076] In the embodiments of the present application, first, in the topological space of the power market data graph, the domain categories of the target entity information are parsed through the ontology layer, and the cosine similarity algorithm is used to calculate the structural similarity and attribute similarity between entities, locate the cross-domain entities with associated mappings and output their association relationships; secondly, taking the target entity and the cross-domain entity as anchor points, the random walk algorithm with a decay factor is used to dynamically construct an inference sub-network including three-hop neighbors; then, along the multi-hop path in the inference sub-network, the rule engine is called to execute the preset logical rule inference, for example, applying the supply-demand balance formula to the "power plant, transmission grid, user" path to deduce the load gap value and output the inference result; finally, the target entity information, the cross-domain entity association relationship and the inference result are input into the feature fusion model to generate a joint semantic representation and supplement it to the target entity information, and output the target entity information including the inference result, such as the load prediction data with the meteorological impact coefficient complemented.

[0077] In actual applications, in a certain power dispatching decision-making system, the system performs the following operations through the topological analysis module of the power market data graph: Based on the entity structure similarity and attribute similarity in the topological space, accurately locate 3 cross-domain coal transportation enterprises with fuel supply associations with the target coal power plant and output the "coal price conduction" association relationship. Taking these 3 coal enterprises and 4 target power plants as anchor points, automatically construct an inference sub-network covering six-hop neighbors, including 9 types of entities such as generator sets, coal transportation railway nodes and port storage and transportation facilities. Execute the preset inventory warning rule in this network: when it is detected that the railway transport capacity saturation of a certain coal enterprise > 90% and the port inventory < 200,000 tons, trigger the "fuel supply risk" inference conclusion. Finally, fuse the 0.82 output value of the target power plant, the coal enterprise supply association parameter and 3 inference conclusions, including 2 warning conclusions and 1 safety conclusion, generate a joint semantic representation including the fuel risk level identifier, and dynamically supplement the risk level to the operation indicators of the target power plant - for example, a new field of "fuel risk level III" is added to the attribute table of a certain power plant, which improves the accuracy of the power generation plan prediction by 0.12 and significantly optimizes the robustness of the day-ahead market clearing decision.

[0078] Based on multi-dimensional similarity matching calculation, this application accurately locates cross-domain entities with potential associated mappings and outputs a mapping set covering cross-domain entity identifiers and their association relationships. Using the target entity and cross-domain entities as topological anchor points, it dynamically extracts and constructs an inference sub-network containing three-hop neighbor nodes, which completely retains the original conduction path and attribute constraint relationships between entities. In the constructed dynamic inference sub-network, along the multi-hop conduction path, it loads the power market trading rule library and operation constraint library to execute the logic inference engine, and layer by layer derives inference results that comply with the power domain specifications. This process strictly ensures the logical compliance and timeliness consistency of the derivative relationships. Finally, it fuses the native data of the target entity, the cross-domain entity association relationships, and the inference results to generate a joint semantic representation and automatically complete it into the target entity information, forming an enhanced entity information set that simultaneously contains original features and cross-domain derivative features, thereby significantly improving the entity information completeness and decision-making support value in complex association scenarios in the power market.

[0079] Optionally, the method further includes: during the establishment of the attribute mapping, calling a pre-constructed power domain term library to identify professional terms in the external data and target entity information, performing semantic disambiguation, and outputting standardized entity attributes; using the standardized entity attributes, when traversing the target entity information and semantic tags, detecting whether they hit predefined atomic indicators, derivative indicators, or composite indicators. If they hit, according to the predefined indicator calculation rules, output the reconstructed node connection data completion path and calculation logic description; based on the node connection data completion path and calculation logic description, dynamically construct a sub-network containing relevant entities, attributes, and calculation rules in the topological space, and execute the indicator definition rules within the sub-network to output the calculated predefined indicator values; inject the predefined indicator values as cross-domain features into the semantic vectors of the corresponding entities or relationships, and based on the semantic vectors containing the indicator values, call the rule-based diagnosis template associated with the indicator to perform anomaly analysis and interpretation.

[0080] In the above steps, the attribute mapping process refers to the operation process of mapping external data to the target entity attributes. The pre-built power domain term library refers to a pre-established database containing power professional vocabulary. Professional terms refer to the specific domain vocabulary in external data and target entity information. Semantic disambiguation refers to the process of eliminating the polysemy of terms to determine the unique meaning. Standardized entity attributes refer to the attribute data in a unified format after disambiguation. Atomic indicators refer to the most basic independent calculation indicators. Derived indicators refer to the secondary indicators derived from atomic indicators. Composite indicators refer to the complex indicators composed of multiple indicators. The predefined indicator calculation rule refers to the formula or logic for calculating the indicators. The reconstructed node connection data completion path refers to the supplementary path description generated according to the indicator calculation. The calculation logic description refers to the step description of indicator calculation. A sub-network refers to a local network dynamically constructed in the topological space, including relevant entity attributes and calculation rules. The indicator definition rule refers to the instruction set for executing indicator calculation. The predefined indicator value refers to the numerical result of the calculated indicator. Cross-domain feature injection refers to the process of adding the indicator value to the semantic vector. The rule-based diagnosis template refers to the predefined diagnosis rule framework. Abnormal analysis and interpretation refer to the operation of performing diagnosis and generating analysis results.

[0081] In the embodiment of the present application, first, during the establishment of the attribute mapping process, the term recognition interface of the pre-built power domain term library is called, and the professional terms in the external data and target entity information are recognized using a context-based natural language processing algorithm, and semantic disambiguation is performed to output standardized entity attributes; secondly, using the standardized entity attributes, when traversing the target entity information and semantic tags, a depth-first traversal algorithm is adopted to detect whether the predefined atomic indicators, derived indicators or composite indicators are hit. If hit, according to the predefined indicator calculation rule, the rule engine is called to output the reconstructed node connection data completion path and the calculation logic description; then, based on the reconstructed calculation logic description, a sub-network including relevant entity attributes and calculation rules is dynamically constructed in the topological space, and the rule engine is executed using the indicator definition rule within the sub-network to calculate and output the predefined indicator value; then, the predefined indicator value is injected as a cross-domain feature into the semantic vector of the corresponding entity or relationship; finally, based on the semantic vector containing the indicator value, the template matching function of the rule-based diagnosis template associated with the indicator is called to perform abnormal analysis and interpretation and output the diagnosis result.

[0082] In practical applications, in a certain smart grid analysis platform, the system calls a pre-built power domain terminology library to perform terminology standardization processing on external data related to coal supply policies and target power plant entity information. For example, "maximum unit output" is disambiguated as "rated capacity", and "coal calorific value fluctuation" is unified as "fuel calorific value deviation", and 6 standardized entity attributes are output. When traversing the topology structure of the target power plant, it is detected that the "environmental protection production restriction execution rate" hits the predefined atomic index library, and the "inter-regional power transmission margin" matches the derivative index rule, automatically triggering the calculation logic: According to the atomic index calculation rule, reconstruct the node connection path of a certain coal power plant, and complete the calculation logic of "production restriction execution rate = output / approved capacity × 100%". Based on the derivative index formula, a sub-network including transmission lines and load nodes is constructed in the topological space to dynamically calculate "margin = (thermal stability limit of the line - actual power flow) / thermal stability limit". The system automatically executes the rule calculation and outputs the margin value of a certain cross-provincial channel as 62.3% and the production restriction execution rates of 3 power plants. After injecting these index values as cross-domain features into the semantic vector, the preset diagnosis template is called: When it is detected that the execution rate of a certain power plant is 77% lower than the 80% threshold and the associated channel margin < 70%, activate the "section congestion risk" alarm logic, and generate a highlighted interpretation of the congestion cause on the feedback interface: "#1 unit has insufficient execution of environmental protection production restrictions, resulting in a decrease in the regional power transmission capacity", realizing automatic attribution of business anomalies.

[0083] This application identifies professional terms in external data and target entity information by calling a pre-built power domain terminology library, performs standardized disambiguation processing, and outputs standardized entity attributes. Using the standardized entity attributes to traverse the topology structure and semantic tags of the target entity, it is detected whether the defined thresholds of atomic indexes, derivative indexes, or composite indexes are hit. If hit, the node connection path is automatically reconstructed according to the predefined index calculation rule to complete the data breakpoint, and a calculation rule chain with mathematical logic description is output synchronously. Subsequently, a special sub-network including relevant entity attributes and calculation factors is dynamically constructed in the topological space based on the calculation rule chain, and the index definition rule is enforced in this sub-network to output the calculation result. Finally, the calculated index values are injected into the semantic vectors of the corresponding entities or relationships as cross-domain features, triggering the associated rule-based diagnosis template to perform anomaly analysis. This process generates an interpretation report with confidence by mapping the results to the standard early warning event library, realizing an automated index diagnosis closed-loop in the power market data fusion scenario.

[0084] Figure 3 The following is a schematic structural diagram of a semantic enhancement and dynamic completion system for a power market data graph provided by an embodiment of this application, as Figure 3 shown, this system includes: A collection module 31, in response to a data query instruction sent by a user, collects target entity information structured in a pre-established power market data graph and unstructured external data; The mapping module 32 converts the external data into structured semantic tags, traverses the topological structure of the target entity information, establishes the mapping relationship between the structured semantic tags and the target entity attributes, combines the influence logic in the data query instruction, calculates the integrity index of the topological structure based on the mapping relationship, and marks the missing entities or missing associations related to the data query instruction to generate a missing tag set; The generation module 33 establishes an attribute mapping between the external data attributes and the entity attributes based on the target entity information and the external data, generates a semantic vector integrating cross-domain features by analyzing the implicit relationship between the external data and the target entity information in the attribute mapping, projects the semantic vector into the topological space according to the missing tag set, and generates topological nodes and topological connections inheriting cross-domain features; The reconstruction module 34 embeds the topological nodes and topological connections into the missing tag positions of the power market data graph, reconstructs the topological structure, and verifies the logical connectivity of the topological structure after completion based on the relationship conduction rule; The verification module 35 analyzes the dynamic conduction chain formed by the target entity information through the completion path based on the verified topological structure, and generates a feedback interface including semantic relationships and a completion structure to respond to the data query instruction.

[0085] Figure 3 The semantic enhancement and dynamic completion system of the power market data graph described above can execute Figure 1 The semantic enhancement and dynamic completion method of the power market data graph described in the embodiments shown, the implementation principle and technical effects will not be elaborated. For the semantic enhancement and dynamic completion system of the power market data graph in the above embodiments, the specific ways for each module and unit to execute operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A semantic enhancement and dynamic completion method for an electricity market data graph, characterized in that Including: In response to a data query instruction sent by a user, collect the target entity information operating in a structured manner and unstructured external data in a pre-established power market data graph; Convert the external data into structured semantic tags, traverse the topological structure of the target entity information, establish a mapping relationship between the structured semantic tags and the target entity attributes, combine the influence logic in the data query instruction, calculate the integrity index of the topological structure based on the mapping relationship, and mark the missing entities or missing associations related to the data query instruction to generate a missing tag set; Based on the target entity information and the external data, establish an attribute mapping between the external data attributes and the entity attributes, generate a semantic vector integrating cross-domain features by analyzing the implicit relationship between the external data and the target entity information in the attribute mapping, and project the semantic vector into the topological space according to the missing tag set to generate topological nodes and topological connections inheriting cross-domain features; Embed the topological nodes and topological connections into the missing tag positions of the power market data graph, reconstruct the topological structure, and verify the logical connectivity of the topological structure after completion based on the relationship conduction rule; Based on the verified topological structure, analyze the dynamic conduction chain formed by the target entity information through the completion path, and generate a feedback interface including semantic relationships and completion structures to respond to the data query instruction.

2. The method according to claim 1, wherein Also including: For the target entity information, call the ontology layer in the pre-constructed power market data graph to identify the domain category to which the entity belongs and the domain relationship category between entities; In the data layer of the power market data graph, extract the neighbor node features and multi-hop path features associated with the identified entities and relationships as entity context features; Fuse the domain category to which the entity belongs, the domain relationship category between entities, and the entity context features to generate a domain-enhanced semantic vector; Inject the domain-enhanced semantic vector into the semantic vector integrating cross-domain features and output the enhanced semantic vector.

3. The method according to claim 1, characterized in that, Also including: In the topological space of the power market data graph, based on the structural similarity and attribute similarity between entities, locate the cross-domain entities with an associated mapping to the target entity information and output the cross-domain entities and their associated relationships; Using the cross-domain entities and the target entities as anchor points, dynamically construct an inference sub-network including multi-hop neighbors in the topological space, and perform logical rule inference along the multi-hop path in the dynamic inference sub-network to output an inference result; Fuse the target entity information, the cross-domain entities and their associated relationships, and the inference result to generate a joint semantic representation; Decompose the joint semantic representation into a sub-vector set according to the entity hierarchy, and hierarchically project the sub-vector set into the topological space based on the voltage level and service type to generate completion nodes and associated edges corresponding to the missing tag set; Verify the business logic compliance of the completion nodes and associated edges through a rule engine, and perform cross-level backtracking correction on the nodes that fail the verification until the logical connectivity requirement of the topological structure is met.

4. The method according to claim 1, wherein Also including: During the process of establishing the attribute mapping, call the pre-built terminology library in the power field to identify the professional terms in the external data and target entity information, perform semantic disambiguation, and output standardized entity attributes; Using the standardized entity attributes, when traversing the target entity information and semantic tags, detect whether they hit predefined atomic indicators, derivative indicators, or composite indicators. If they hit, according to the predefined indicator calculation rules, output the reconstructed node connection data completion path and calculation logic description; Based on the node connection data completion path and calculation logic description, dynamically construct a sub-network containing relevant entities, attributes, and calculation rules in the topological space, and execute the indicator definition rules within the sub-network to output the calculated predefined indicator values; Inject the predefined indicator values as cross-domain features into the semantic vectors of the corresponding entities or relationships, and based on the semantic vectors containing the indicator values, call the rule-based diagnostic template associated with the indicator to perform anomaly analysis and interpretation.

5. The method according to claim 1, characterized in that, Based on the target entity information and the external data, establish an attribute mapping between the external data attributes and entity attributes. By analyzing the implicit relationship between the external data and the target entity information in the attribute mapping, generate a semantic vector integrating cross-domain features, and project the semantic vector to the topological space according to the missing marker set to generate topological nodes and topological connections inheriting cross-domain features, including: Based on the target entity information and the external data, construct a bidirectional mapping dictionary between the external data attributes and entity attributes, and record the direct mapping rules, mathematical derivation formulas, and unit conversion coefficients between the attributes; Through a deep semantic association model, analyze the implicit multi-order derivation relationships between the attributes in the bidirectional mapping dictionary, and identify cross-domain attribute combinations with common change characteristics or causal constraints; Generate semantic vectors for each of the cross-domain attribute combinations, where the semantic vectors integrate the encoded values, relationship weight coefficients, and time sequence markers of the associated attributes; According to the topological coordinates located by the missing marker set, create blank topological nodes in the power market data graph, and project the semantic vectors to the corresponding blank topological nodes to form topological nodes carrying cross-domain features; According to the relationship weight coefficients in the semantic vectors, establish weighted topological connections inheriting cross-domain features between the topological nodes carrying cross-domain features. The attributes and connection relationships of the topological nodes and topological connections need to satisfy the topological constraint rules of the power market graph.

6. The method according to claim 5, characterized in that Through a deep semantic association model, analyze the implicit multi-order derivation relationships between the attributes in the bidirectional mapping dictionary, and identify cross-domain attribute combinations with common change characteristics or causal constraints, including: Through a deep semantic association model, receive the set of mapping entries from the bidirectional mapping dictionary. The set of mapping entries records the corresponding rules and relationship identifiers between the external data attributes and target entity attributes, and construct a derivation network with the set of mapping entries; Perform a traversal operation on the derivation network, identify the consistency of the time dimension fluctuation characteristics between the nodes in the derivation network, and record the common change attribute groups. Detect the conduction paths satisfying the causal constraint rules in the derivation network and record the response sequences; Calculate the fluctuation consistency quantization value of the co-varying attribute group, and measure the trigger delay time of the response sequence; When the fluctuation consistency quantization value exceeds a set threshold and the trigger delay time is within a preset interval, output a cross-domain attribute combination with a relationship identifier.

7. The method according to claim 1, characterized in that Based on the verified topological structure, parse the dynamic conduction chain formed by the target entity information through the completion path, and generate a feedback interface including semantic relationships and completion structures to respond to the data query instruction, including: Based on the verified topological structure, extract the dynamic conduction chain formed by the target entity information through the completion path, and parse the relationship weight value and the time sequence marking trajectory of each conduction path in the dynamic conduction chain; Map the relationship weight value to a preset semantic intensity level, convert the time sequence marking trajectory into a time axis, generate a node relationship chain with a conduction direction identifier, and mark the semantic intensity level and the time axis on the node relationship chain; Convert the conduction paths that fail to pass the verification into warning identifiers, integrate the node relationship chain and the warning identifiers, and generate a visual feedback interface to respond to the data query instruction.

8. The method according to claim 1, wherein Convert the external data into structured semantic tags, traverse the topological structure of the target entity information, establish the mapping relationship between the structured semantic tags and the target entity attributes, combine the influence logic in the data query instruction, calculate the integrity index of the topological structure based on the mapping relationship, and mark the missing entities or missing associations related to the data query instruction to generate a missing tag set, including: Perform semantic annotation processing on the unstructured external data, extract the attribute category and numerical description, generate structured semantic tags, traverse the associated topological structure starting from the target entity, and identify all the conduction paths between entities and the attribute associations on the paths; Establish the mapping relationship between the structured semantic tags and the target entity attributes, combine the influence logic in the data query instruction, calculate the integrity index of each conduction path in the mapping relationship, and synchronously detect the associated breakpoints that violate the influence logic in the conduction path; When the integrity index is lower than the preset threshold, or there are such associated breakpoints, mark the corresponding entity as a missing entity or the association as a missing association, summarize the topological coordinates of all the missing entities and missing associations, and generate a missing tag set.

9. The method according to claim 1, characterized in that Embed the topological nodes and topological connections into the missing tag positions of the power market data graph, reconstruct the topological structure, and based on the relationship conduction rules, verify the logical connectivity of the complemented topological structure, including: Locate the entity missing positions and association missing positions in the power market data graph according to the missing tag set, embed the topological nodes into the entity missing positions and write the attribute coding values, embed the topological connections into the association missing positions and configure the weight coefficients, and reconstruct the topological structure; Traverse the reconstructed topological structure based on the relationship conduction rules, and verify the conduction direction, weight attenuation gradient, and time sequence continuity of the newly added conduction paths; If the conduction direction conforms to the preset constraints, the weight decay gradient is within the allowable range, and the timing markers are continuous, it is determined that the conduction path is logically connected, and the conduction paths that fail to pass the logical connectivity verification and the failure rules are recorded.

10. A semantic enhancement and dynamic completion system for an electricity market data graph, characterized in that, Including: A collection module, configured to collect target entity information operating in a structured manner and unstructured external data in a pre-established power market data graph in response to a data query instruction sent by a user; A mapping module, configured to convert the external data into structured semantic tags, traverse the topological structure of the target entity information, establish a mapping relationship between the structured semantic tags and the target entity attributes, combine the influence logic in the data query instruction, and calculate the integrity index of the topological structure based on the mapping relationship, and mark the missing entities or missing associations related to the data query instruction to generate a missing marker set; A generation module, configured to establish an attribute mapping between the external data attributes and the entity attributes based on the target entity information and the external data, generate a semantic vector integrating cross-domain features by parsing the implicit relationship between the external data and the target entity information in the attribute mapping, project the semantic vector into the topological space according to the missing marker set, and generate topological nodes and topological connections inheriting cross-domain features; A reconstruction module, configured to embed the topological nodes and topological connections into the missing marker positions of the power market data graph, reconstruct the topological structure, and verify the logical connectivity of the topological structure after completion based on the relationship conduction rules; A verification module, configured to parse the dynamic conduction chain formed by the target entity information through the completed path based on the verified topological structure, and generate a feedback interface including semantic relationships and a completed structure to respond to the data query instruction.

Images