Multilevel information extraction and intelligent reasoning fusion optimization method and system based on hydropower industry knowledge graph

Abstract

The invention discloses a multilevel information extraction and intelligent reasoning fusion optimization method and system based on a hydropower industry knowledge graph. The method comprises the following steps: acquiring original knowledge graph data; segmenting the original knowledge graph data into a first graph region and a second graph region according to a data magnitude threshold judgment result; according to the extraction command, generating a linkage guide label from the first atlas area, and inputting the linkage guide label into the second atlas area for information fusion operation to obtain a reasoning fusion result; matching and capturing the inference fusion result and elements in the original knowledge graph to obtain an extraction verification result; the problems of low efficiency, poor flexibility, weak cross-level reasoning ability and lack of quality self-verification in the prior art are solved, and the method has the characteristics of being capable of intelligently dividing map structures, dynamically guiding cross-regional information fusion and having result self-verification ability.

Description

Technical Field

[0001] This invention belongs to the field of information extraction and optimization, and in particular relates to a multi-level information extraction and intelligent reasoning fusion optimization method and system based on a knowledge graph of the hydropower industry. Background Technology

[0002] With the deepening of digitalization in the hydropower industry, knowledge graphs, as a core technology for organizing and managing massive amounts of domain knowledge, are playing an increasingly important role in equipment management, fault diagnosis, and operation optimization. However, existing knowledge graph-based information extraction and optimization methods still face significant technical bottlenecks when dealing with the complex, multi-layered, and dynamically changing data structures of the hydropower industry, mainly in the following aspects: The first type of method is extraction techniques based on global rule matching. This method typically relies on pre-set fixed rules or templates to traverse and match the entire knowledge graph to extract target information. Its drawback is that hydropower industry knowledge graphs are usually enormous, encompassing multiple levels from macro-basin planning to micro-equipment parameters; global traversal leads to huge computational overhead and slow response times. More importantly, fixed rules cannot adapt to the dynamic evolution of the graph structure. For example, when new equipment types are added, operating procedures are changed, or the fault mode library is expanded, the original rules are easily rendered invalid, resulting in incomplete or erroneous extraction results, making it difficult to meet the high real-time requirements of business scenarios.

[0003] The second type of method employs a strategy of independently extracting data after partitioning static subgraphs. While this approach reduces the data size per processing run through pre-segmentation, it severs the semantic relationships between subgraphs, leading to information silos. For example, when analyzing the causes of a turbine failure, its associated maintenance records, historical operating data, and upstream and downstream equipment information may be distributed across different static subgraphs. Existing methods struggle to achieve cross-subgraph correlation reasoning, thus failing to form a complete fault chain analysis. The extracted information lacks depth and contextual support, limiting the effectiveness of intelligent decision-making.

[0004] The third type of method attempts to introduce simple graph neural networks (GNNs) for feature learning, but their model structures are usually uniform and fixed, making it impossible to adaptively adjust to different extraction tasks or the local structural characteristics of the graph. When faced with highly heterogeneous nodes and relationship types in hydropower knowledge graphs, such as spatiotemporal data, equipment attributes, and text reports, fixed models struggle to accurately capture the intrinsic connections between different levels and modalities of information, resulting in insufficient feature representation and limited reasoning capabilities. Furthermore, existing methods generally lack closed-loop verification mechanisms for the validity of extraction results, failing to guarantee the accuracy and reliability of information extraction.

[0005] In summary, existing technologies for applying knowledge graphs in the hydropower industry generally suffer from low efficiency, poor flexibility, weak cross-level reasoning capabilities, and a lack of quality self-verification. Therefore, it is necessary to propose a multi-level information extraction and intelligent reasoning fusion optimization method and system based on hydropower industry knowledge graphs to address these issues. Summary of the Invention

[0006] The technical problem to be solved by this invention is to provide a multi-level information extraction and intelligent reasoning fusion optimization method and system based on the knowledge graph of the hydropower industry. It aims to solve the problems of low efficiency, poor flexibility, weak cross-level reasoning ability and lack of quality self-verification of existing technologies. Thus, it has the characteristics of intelligently dividing the graph structure, dynamically guiding cross-regional information fusion, and having the ability to self-verify the results.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A multi-level information extraction and intelligent reasoning fusion optimization method based on hydropower industry knowledge graph includes the following steps: Obtain the raw knowledge graph data; The original knowledge graph data is divided into the first graph region and the second graph region based on the data volume threshold. Based on the extraction command, a linkage guidance tag is generated from the first map region, and the linkage guidance tag is input into the second map region for information fusion operation to obtain the reasoning fusion result; The reasoning fusion results are matched and extracted with the elements in the original knowledge graph to obtain the extraction and verification results.

[0008] Preferably, the original knowledge graph data is divided into a first knowledge graph region and a second knowledge graph region based on the data volume threshold, including: Criteria for determining the magnitude of data generated based on application scenario information; By comparing the data volume in the original knowledge graph with the aforementioned volume judgment criteria, a graph segmentation strategy is obtained. The original knowledge graph data is segmented according to the graph segmentation strategy to obtain the first graph region and the second graph region. The first graph region includes all data root index elements.

[0009] Preferably, a linkage guidance tag is generated from the first map region according to the extraction command, and the linkage guidance tag is input into the second map region for information fusion operation to obtain the inference fusion result, including: The extraction command generated by the client is collected, and the extraction command includes the content to be extracted and the index to be extracted; Generate linkage guidance tags from the first map region according to the extraction command; Generate inference fusion results using inference fusion formulas: ; Where R is the inference fusion result, n is the number of linkage guidance tags, and L is the data element of the second graph region.

[0010] Preferably, the reasoning fusion results are matched and extracted with elements in the original knowledge graph to obtain the extraction and verification results, including: The reasoning fusion results are retrieved from the root index pool of the original knowledge graph, and matching, crawling and comparison operations are performed to obtain the extraction verification results; the extraction verification results include qualified extraction and unqualified extraction.

[0011] Preferably, a multi-level information extraction and intelligent reasoning fusion optimization system based on a hydropower industry knowledge graph, used in the aforementioned hydropower feature extraction and optimization method based on cross-modal heterogeneous data fusion, includes: The acquisition module is used to acquire raw knowledge graph data; The segmentation module is used to segment the original knowledge graph data into a first graph region and a second graph region based on the data volume threshold. The input module is used to generate linkage guidance tags from the first map region according to the extraction command, and input the linkage guidance tags into the second map region for information fusion operation to obtain the inference fusion result. The crawling module is used to match and crawl the reasoning fusion results with the elements in the original knowledge graph to obtain the extraction and verification results.

[0012] Preferably, the segmentation module includes: The generation unit is used to generate a scale judgment standard based on application scenario information; The comparison unit is used to compare the data volume and the volume judgment criteria in the original knowledge graph to obtain the graph segmentation strategy. The segmentation unit is used to segment the original knowledge graph data according to the graph segmentation strategy to obtain the first graph region and the second graph region. The first graph region includes all data root index elements.

[0013] Preferably, the input module includes: The acquisition unit is used to acquire the extraction command generated by the client, the extraction command including the extraction content and the extraction index; An extraction unit is used to generate the linkage guidance tag from the first map region according to an extraction command; The fusion unit is used to generate inference fusion results through inference fusion formulas: ; Where R is the inference fusion result, n is the number of linkage guidance tags, and L is the data element of the second graph region.

[0014] Preferably, the crawling module includes: The processing unit is used to retrieve the reasoning fusion results from the root index pool in the original knowledge graph, perform matching, crawling and comparison operations, and obtain the extraction verification results, which include qualified extraction and unqualified extraction.

[0015] Preferably, a non-volatile storage medium includes a stored program, wherein the program, when running, controls the device where the non-volatile storage medium is located to execute the multi-level information extraction and intelligent reasoning fusion optimization method based on hydropower industry knowledge graph.

[0016] Preferably, an electronic device includes a processor and a memory; the memory stores computer-readable instructions, and the processor is used to execute the computer-readable instructions, wherein the computer-readable instructions execute the multi-level information extraction and intelligent reasoning fusion optimization method based on hydropower industry knowledge graph.

[0017] The beneficial effects of this invention are as follows: 1. This invention dynamically segments the knowledge graph into a first and second graph region based on a data volume threshold, retaining the root index in the first region. This achieves efficient and controllable hierarchical information processing, solving the inefficiency and resource waste problems caused by global traversal in existing technologies. It generates a volume judgment standard based on real-time application scenarios and intelligently segments the original graph into a lightweight first region containing the core index and a second region containing detailed data. This design allows the system to focus on the core first region first, rather than processing the entire complex graph each time it performs extraction tasks, significantly reducing the initial computational load. This divide-and-conquer strategy significantly improves the system's processing speed and responsiveness when dealing with large-scale, multi-layered hydropower knowledge graphs, laying an efficient data foundation for subsequent accurate reasoning.

[0018] 2. This invention achieves deep cross-regional association and intelligent reasoning by generating linkage guidance tags from a first region and using them to guide information fusion in a second region. This solves the problems of fragmented information extraction and lack of context caused by data segmentation in existing technologies. A highly summarized "linkage guidance tag" is extracted from the core first region using an extraction command. This tag carries key semantic directions and related information. Subsequently, this tag is injected into the second region, and through a specific reasoning fusion formula, it activates and integrates related deep and detailed data. This process simulates the thought process of human experts who develop associative reasoning from key clues, effectively connecting relevant information scattered across different regions and levels of the graph to form a complete information chain. This results in more in-depth and logical reasoning fusion results, significantly improving the comprehensiveness and intelligence of information extraction.

[0019] 3. This invention achieves closed-loop self-verification of the extraction result's accuracy by matching and verifying the reasoning fusion result with the root index pool of the original knowledge graph. This solves the technical problem of uncontrollable extraction quality caused by the lack of an effective verification mechanism in existing technologies. The reasoning fusion result generated in this invention is not directly output but needs to be returned to the "root index pool" of the original graph for consistency matching and comparison. This step is equivalent to setting a "quality checkpoint" for the information extraction process. If the result can effectively match the authoritative core elements in the root index pool, it is verified as "extraction qualified," ensuring the reliability of the result; if it cannot match, it is judged as "unqualified," triggering an alarm or reprocessing. This embedded verification mechanism constructs a closed-loop quality control system, fundamentally guaranteeing the accuracy and credibility of the final information extraction result and enhancing the robustness of the entire system in practical applications. Attached Figure Description

[0020] Figure 1 This is a flowchart of the method in an embodiment of the present invention; Figure 2 This is a system structure block diagram in an embodiment of the present invention; Figure 3 This is a block diagram of a terminal device for executing the method according to the present invention in an embodiment of the present invention; Figure 4 It is a storage unit used in embodiments of the present invention to hold or carry program code that implements the method according to the present invention. Detailed Implementation

[0021] Example 1: like Figure 1 The image shows an embodiment of a multi-level information extraction and intelligent reasoning fusion optimization method based on a hydropower industry knowledge graph, comprising the following steps: Step S102: Obtain the original knowledge graph data.

[0022] Step S104: The original knowledge graph data is divided into a first graph region and a second graph region based on the data volume threshold judgment result.

[0023] Step S106: Generate a linkage guidance tag from the first map region according to the extraction command, and input the linkage guidance tag into the second map region for information fusion operation to obtain the inference fusion result.

[0024] Step S108: Match and capture the reasoning fusion result with the elements in the original knowledge graph to obtain the extraction verification result.

[0025] Optionally, dividing the original knowledge graph data into a first graph region and a second graph region based on the data volume threshold includes: generating a volume judgment standard based on application scenario information; comparing the data volume in the original knowledge graph with the volume judgment standard to obtain a graph segmentation strategy; and performing a segmentation operation on the original knowledge graph data according to the graph segmentation strategy to obtain a first graph region and a second graph region, wherein the first graph region includes all data root index elements.

[0026] Optionally, the step of generating linkage guidance tags from the first map region according to the extraction command and inputting the linkage guidance tags into the second map region for information fusion operation to obtain the inference fusion result includes: collecting the extraction command generated by the client, wherein the extraction command includes: extracting content and extracting index; generating the linkage guidance tags from the first map region according to the extraction command; and using the inference fusion formula. ; Generate the inference fusion result, where R is the inference fusion result, n is the number of linkage guidance tags, and L is the data element of the second graph region.

[0027] Optionally, the step of matching and capturing the reasoning fusion result with the elements in the original knowledge graph to obtain the extraction verification result includes: retrieving the reasoning fusion result from the root index pool of the original knowledge graph, performing matching and comparison operations to obtain the extraction verification result, wherein the extraction verification result includes: qualified extraction and unqualified extraction.

[0028] Example 2: Figure 2 This is a structural block diagram of a multi-level information extraction and intelligent reasoning fusion optimization system based on a knowledge graph of the hydropower industry according to an embodiment of the present invention, such as... Figure 2 As shown, the system includes: Module 20 is used to acquire raw knowledge graph data.

[0029] The segmentation module 22 is used to segment the original knowledge graph data into a first graph region and a second graph region based on the data volume threshold.

[0030] The input module 24 is used to generate linkage guidance tags from the first map region according to the extraction command, and input the linkage guidance tags into the second map region for information fusion operation to obtain the inference fusion result.

[0031] The crawling module 26 is used to match and crawl the reasoning fusion result with the elements in the original knowledge graph to obtain the extraction verification result.

[0032] Optionally, the segmentation module includes: a generation unit for generating a magnitude judgment standard based on application scenario information; a comparison unit for comparing the data magnitude in the original knowledge graph with the magnitude judgment standard to obtain a graph segmentation strategy; and a segmentation unit for performing segmentation operations on the original knowledge graph data according to the graph segmentation strategy to obtain a first graph region and a second graph region, wherein the first graph region includes all data root index elements.

[0033] Optionally, the input module includes: a collection unit for collecting the extraction command generated by the client, wherein the extraction command includes: extraction content and extraction index; an extraction unit for generating the linkage guidance tag from the first map region according to the extraction command; and a fusion unit for inferring and fusing formulas. ; Generate the inference fusion result, where R is the inference fusion result, n is the number of linkage guidance tags, and L is the data element of the second graph region.

[0034] Optionally, the crawling module includes a processing unit, used to retrieve the reasoning fusion result from the root index pool of the original knowledge graph, perform matching crawling and comparison operations, and obtain the extraction verification result, wherein the extraction verification result includes: extraction qualified, extraction unqualified.

[0035] The above embodiments solve the technical problem that the information extraction and optimization process in the prior art is merely based on the rule extraction of all elements and data of the knowledge graph, and the extraction results are optimized and corrected in a fixed way, which cannot efficiently extract and optimize on the basis of multi-level knowledge graphs or variable graph structures.

[0036] According to another aspect of the present invention, a non-volatile storage medium is also provided, the non-volatile storage medium including a stored program, wherein the program, when running, controls the device where the non-volatile storage medium is located to execute a multi-level information extraction and intelligent reasoning fusion optimization method based on a hydropower industry knowledge graph.

[0037] Specifically, the above method includes: acquiring original knowledge graph data; dividing the original knowledge graph data into a first graph region and a second graph region based on a data volume threshold judgment result; generating linkage guidance tags from the first graph region according to an extraction command, and inputting the linkage guidance tags into the second graph region for information fusion operation to obtain a reasoning fusion result; matching and capturing the reasoning fusion result with elements in the original knowledge graph to obtain an extraction verification result. Optionally, dividing the original knowledge graph data into a first graph region and a second graph region based on a data volume threshold judgment result includes: generating a volume judgment standard based on application scenario information; comparing the data volume in the original knowledge graph with the volume judgment standard to obtain a graph segmentation strategy; performing a graph segmentation operation on the original knowledge graph data according to the graph segmentation strategy to obtain a first graph region and a second graph region, wherein the first graph region includes all data root index elements. Optionally, the step of generating linkage guidance tags from the first map region according to the extraction command and inputting the linkage guidance tags into the second map region for information fusion operation to obtain the inference fusion result includes: collecting the extraction command generated by the client, wherein the extraction command includes: extracting content and extracting index; generating the linkage guidance tags from the first map region according to the extraction command; and using the inference fusion formula. ; Generate a reasoning fusion result, where R is the reasoning fusion result, n is the number of linked guiding tags, and L is the data element of the second knowledge graph region. Optionally, matching and extracting the reasoning fusion result with elements in the original knowledge graph to obtain an extraction verification result includes: retrieving the reasoning fusion result from the root index pool of the original knowledge graph, performing matching, extraction, and comparison operations to obtain the extraction verification result, wherein the extraction verification result includes: successful extraction and unsuccessful extraction.

[0038] According to another aspect of the present invention, an electronic device is also provided, comprising a processor and a memory; the memory stores computer-readable instructions, and the processor is configured to execute the computer-readable instructions, wherein the computer-readable instructions, when executed, perform a multi-level information extraction and intelligent reasoning fusion optimization method based on a hydropower industry knowledge graph.

[0039] Specifically, the above method includes: acquiring original knowledge graph data; dividing the original knowledge graph data into a first graph region and a second graph region based on a data volume threshold judgment result; generating linkage guidance tags from the first graph region according to an extraction command, and inputting the linkage guidance tags into the second graph region for information fusion operation to obtain a reasoning fusion result; matching and capturing the reasoning fusion result with elements in the original knowledge graph to obtain an extraction verification result. Optionally, dividing the original knowledge graph data into a first graph region and a second graph region based on a data volume threshold judgment result includes: generating a volume judgment standard based on application scenario information; comparing the data volume in the original knowledge graph with the volume judgment standard to obtain a graph segmentation strategy; performing a graph segmentation operation on the original knowledge graph data according to the graph segmentation strategy to obtain a first graph region and a second graph region, wherein the first graph region includes all data root index elements. Optionally, the step of generating linkage guidance tags from the first map region according to the extraction command and inputting the linkage guidance tags into the second map region for information fusion operation to obtain the inference fusion result includes: collecting the extraction command generated by the client, wherein the extraction command includes: extracting content and extracting index; generating the linkage guidance tags from the first map region according to the extraction command; and using the inference fusion formula. ; Generate the inference fusion result, where R is the inference fusion result, n is the number of linkage guidance tags, and L is the data element of the second graph region.

[0040] Optionally, the step of matching and capturing the reasoning fusion result with the elements in the original knowledge graph to obtain the extraction verification result includes: retrieving the reasoning fusion result from the root index pool of the original knowledge graph, performing matching and comparison operations to obtain the extraction verification result, wherein the extraction verification result includes: qualified extraction and unqualified extraction.

[0041] Figure 3 This is a schematic diagram of the hardware structure of a terminal device provided in an embodiment of this application.

[0042] like Figure 3 As shown, the terminal device may include an input device 30, a processor 31, an output device 32, a memory 33, and at least one communication bus 34. The communication bus 34 is used to realize communication connections between components. The memory 33 may include high-speed RAM memory, and may also include non-volatile memory (NVM), such as at least one disk storage device. The memory 33 may store various programs for performing various processing functions and implementing the method steps of this embodiment.

[0043] Optionally, the processor 31 may be implemented as a central processing unit (CPU), application-specific integrated circuit (ASIC), digital signal processor (DSP), digital signal processing device (DSPD), programmable logic device (PLD), field-programmable gate array (FPGA), controller, microcontroller, microprocessor or other electronic components. The processor 31 is coupled to the input device 30 and output device 32 via wired or wireless connection.

[0044] Optionally, the input device 30 may include various input devices, such as a user interface, a device interface, a programmable software interface, a camera, and a sensor. Optionally, the device interface may be a wired interface for data transmission between devices, or a hardware interface (e.g., USB interface, serial port) for data transmission between devices. Optionally, the user interface may be a user-facing control button, a voice input device for receiving voice input, or a touch-sensing device for receiving user touch input (e.g., a touchscreen, touchpad). Optionally, the programmable software interface may be an entry point for users to edit or modify programs, such as a chip's input pin interface or input interface. Optionally, the transceiver may be a radio frequency transceiver chip with communication functions, a baseband processing chip, and a transceiver antenna. Audio input devices such as microphones can receive voice data. Output device 32 may include displays, speakers, and other output devices.

[0045] In this embodiment, the processor of the terminal device includes functions for executing the modules of the data processing device in each device. The specific functions and technical effects can be referred to in the above embodiments, and will not be repeated here.

[0046] Figure 4 This is a schematic diagram of the hardware structure of a terminal device provided in another embodiment of this application. Figure 4 Yes Figure 3 A specific implementation example in the implementation process. For example... Figure 4 As shown, the terminal device in this embodiment includes a processor 41 and a memory 42.

[0047] The processor 41 executes the computer program code stored in the memory 42 to implement the method in the above embodiments.

[0048] Memory 42 is configured to store various types of data to support operation on the terminal device. Examples of this data include instructions for any application or method operating on the terminal device, such as messages, pictures, videos, etc. Memory 42 may include random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device.

[0049] Optionally, the processor 41 is located in the processing component 40. The terminal device may also include: a communication component 43, a power supply component 44, a multimedia component 45, an audio component 46, an input / output interface 47, and / or a sensor component 48. The specific components included in the terminal device are determined according to actual needs, and this embodiment does not limit this.

[0050] Processing component 40 typically controls the overall operation of the terminal device. Processing component 40 may include one or more processors 41 to execute instructions to complete all or part of the steps of the above-described method. Furthermore, processing component 40 may include one or more modules to facilitate interaction between processing component 40 and other components. For example, processing component 40 may include a multimedia module to facilitate interaction between multimedia component 45 and processing component 40.

[0051] Power supply component 44 provides power to various components of the terminal device. Power supply component 44 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the terminal device.

[0052] Multimedia component 45 includes a display screen that provides an output interface between a terminal device and a user. In some embodiments, the display screen may include a liquid crystal display (LCD) and a touch panel (TP). If the display screen includes a touch panel, the display screen may be implemented as a touchscreen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors may sense not only the boundaries of the touch or swipe action but also the duration and pressure associated with the touch or swipe operation.

[0053] Audio component 46 is configured to output and / or input audio signals. For example, audio component 46 includes a microphone (MIC) configured to receive external audio signals when the terminal device is in an operating mode, such as a voice recognition mode. The received audio signals may be further stored in memory 42 or transmitted via communication component 43. In some embodiments, audio component 46 also includes a speaker for outputting audio signals.

[0054] Input / output interface 47 provides an interface between processing component 40 and peripheral interface modules, such as click wheels, buttons, etc. These buttons may include, but are not limited to, volume buttons, start buttons, and lock buttons.

[0055] Sensor assembly 48 includes one or more sensors for providing status assessments of various aspects of the terminal device. For example, sensor assembly 48 can detect the on / off state of the terminal device, the relative positioning of components, and the presence or absence of user contact with the terminal device. Sensor assembly 48 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact, including detecting the distance between the user and the terminal device. In some embodiments, sensor assembly 48 may also include a camera, etc.

[0056] Communication component 43 is configured to facilitate wired or wireless communication between the terminal device and other devices. The terminal device can access wireless networks based on communication standards, such as WiFi, 2G, or 3G, or combinations thereof. In one embodiment, the terminal device may include a SIM card slot for inserting a SIM card, enabling the terminal device to log in to a GPRS network and establish communication with a server via the Internet.

[0057] As can be seen from the above, in Figure 4 The communication component 43, audio component 46, input / output interface 47, and sensor component 48 involved in the embodiment can all be used as... Figure 3 The implementation method of the input device in the embodiment.

[0058] Example 3: This embodiment provides a specific implementation process for step S104 in Embodiment 1 as follows: Phase 1: Judging the scale of data generated based on application scenario information: 1. Input application scenario information: including scenario type, response time limit requirements, and query depth and breadth.

[0059] 2. Generation scale judgment criteria: Based on the above scenario information, the system dynamically generates a set of segmentation criteria, mainly including: Core subgraph size threshold: Define the maximum number of nodes and relationships allowed in the first graph region; for example, for real-time alarm scenarios, this threshold would be set very small, such as less than 1,000 nodes, to ensure lightning-fast retrieval speed.

[0060] Relationship Hop Threshold: Define the shortest relationship path length to be retained within the first region starting from the "extraction index" entry point. For example, for in-depth root cause analysis, all relationships within 3 hops may need to be retained; while for real-time alerts, only the direct relationships within 1 hop are retained.

[0061] Entity Criticality Weight: A predefined or learned weight table that identifies which entities are "root index elements"; typically includes: Critical facility entities: such as dams, power stations, main transformers, and core turbines.

[0062] Core type entities: Definition nodes for all fault types and all maintenance operation types.

[0063] High-level relationship nodes: Relationships representing top-level semantics such as "belong to", "located in", "has a fault", etc.

[0064] The output of this stage is: A structured magnitude judgment criterion: Criteria = { max_size: 1000, max_hops: 2, critical_entities: ["Dam_A", "Turbine_Type", "hasFault",... ]}.

[0065] Stage Two: Compare the magnitude of the original graph data with the standard to obtain the graph segmentation strategy: 1. Magnitude evaluation: The system analyzes the original knowledge graph, calculates its total number of nodes, relationships, and identifies all root index elements that meet the "entity criticality weight".

[0066] 2. Strategy formulation: If the scale of the original graph < max_size, there is no need to segment, and only lightweight optimization is performed; however, in large-scale hydropower graphs, this situation is rare.

[0067] If the scale of the original graph is much larger than max_size. At this time, the system formulates the following graph segmentation strategy: (1) Set the goal: Construct a highly interconnected core subgraph, i.e., the first graph region, that contains all root index elements and meets the max_size and max_hops constraints.

[0068] (2) Adopt an expansionary pruning algorithm driven by root index elements.

[0069] Step 1: Take all root index elements as seed nodes and add them to the first region.

[0070] Step 2: Starting from these seed nodes, expand outwards along all relational directions, with the expansion depth not exceeding max_hops.

[0071] Step 3: During the expansion process, if it is found that the size of the first region is about to exceed max_size, pruning is performed based on indicators such as node degree and centrality, prioritizing the retention of nodes connected to more root index elements to ensure the connectivity and coreness of the network.

[0072] The output at this stage is: An executable splitting strategy: Strategy = { method: "root_expansion_pruning", seeds: [critical_entities...], parameters: {max_hops: 2, target_size: 1000}}。

[0073] Phase 3: Perform the segmentation operation to generate the first and second map regions: 1. Perform segmentation: The system performs segmentation on the original knowledge graph according to the above strategy.

[0074] 2. First map region: The content contains all root index elements, as well as the high-level relationship network between them within a limited number of hops.

[0075] Its features include lightweight, high speed, and global view.

[0076] Its role in subsequent processes: As a generator of "linkage guidance tags", because all core entities and relationships are here, the entry point for extracting commands can be located very quickly, and based on this highly generalized network, the reasoning path can be intelligently planned to generate guidance tags.

[0077] 3. Second spectral region: The content includes all remaining data in the original map except for the first region.

[0078] Its features include massive, detailed, and deep storage, containing all detailed attributes, historical time-series data, specific text reports, long-path relationships, etc.

[0079] Its role in subsequent processes: As a data pool for "information fusion operations", it receives "linkage guidance tags" sent from the first region and performs in-depth mining and calculation within a specific, local scope according to the instructions of the tags.

[0080] Example 4: This embodiment provides a specific implementation process for step S106 in Embodiment 1 as follows: In step S106, a lightweight first map region is used as the command center to generate precise navigation instructions, i.e., linkage guidance tags. Then, targeted fishing and information integration are carried out in the vast ocean of data in the second map region to finally arrive at a single, reliable reasoning conclusion.

[0081] Its specific execution process can be broken down into the following stages: Phase 1: Parsing the extraction command and locating the initial target: 1. Input: The extraction command issued by the client.

[0082] 2. Content Parsing: The system receives and parses the extraction command, extracting two core elements: Content extracted: Defines the questions that need to be answered, such as: Analyze the potential causes of abnormal vibration in the XX turbine.

[0083] Extracting the index: provides entry points for starting a search in the knowledge graph, such as "Device ID: WT-001" and "Abnormal phenomenon: Vibration amplitude exceeds the standard".

[0084] 3. Initial location: The system uses the extracted index to perform a fast search in the first graph area; the first area stores the core indexes and key relationships of all entities, just like a company's organizational chart; in this example, the system quickly locates entity WT-001 and its fault relationship, namely hasFault.

[0085] Phase Two: Generating Linkage Guidance Tags 1. Generation process: Based on the initial nodes and relationships found in the first region, the system does not directly return data, but generates one or more linkage guide tags.

[0086] 2. The essence of tags: Linked guidance tags are not simple keywords, but a structured query and reasoning plan that includes: Core Entities / Relationships: Key nodes and relationship types extracted from the first region. For example: [Core Entity: WT-001, Relationship Type: hasFault, causedBy].

[0087] Reasoning path assumption: Based on the pre-defined exploration direction of domain knowledge, for example, for abnormal vibration, the pre-defined path may include: mechanical cause, electrical cause, hydraulic cause.

[0088] Feature extraction command: Specify the data type to be of interest in the second area; for example: extract vibration spectrum data from the past week, or find entries related to "bearing" in historical maintenance records.

[0089] Fusion strategy identifier: Indicates which algorithm or formula will be used for information fusion; for example, it points to the inference fusion formula.

[0090] The output of this stage is a set of structured, machine-readable linked guide labels, for example: LeadTag={core_entity:"WT-001",paths:["mechanical","electrical"],features:["vibration_spectrum","maintenance_log"],fusion_strategy:"formula_1"}; Phase 3: Perform information fusion operations in the second map region: The system injects the aforementioned guidance tags into the second map region: 1. Path Exploration and Data Activation: Based on the inference path assumptions in the guidance labels, the system traverses along the specified relationship path in the second region, for example: Following the mechanical cause path: WT-001 -> hasComponent -> Bearings -> hasStatus -> "Wear"; Along the electrical cause path: WT-001 -> hasComponent -> Generator -> linkedTo -> ExcitationSystem -> hasParameter -> "Unstable excitation current".

[0091] At the same time, based on the "feature extraction instruction", detailed attribute data of relevant entities on the path are collected, such as the numerical sequence of vibration sensors and the text description of maintenance work orders.

[0092] 2. Information Fusion Computation: The system integrates the collected multi-path, multi-type information according to the "fusion strategy" specified by the guidance labels. Here, the physical meaning of the formula can be explained: ; in, This is the feature vector of the nth data element extracted from the second region. For example, the bearing wear state is a vector, and the vibration spectrum after Fourier transform is also a vector. To calculate the L2 norm of the vector, which represents the strength or significance of the evidence, the norm of a violently anomalous vibration signal will be much larger than that of a normal signal.

[0093] The purpose of the double summation operation is to aggregate the strength of all evidence across all paths.

[0094] The inner layer summation summarizes the strength of all evidence under a single path, such as bearing wear and shaft misalignment, to obtain the total confidence level of that path.

[0095] The outer summation sums the total confidence scores of all candidate paths, including mechanical and electrical paths, to obtain a global inference fusion result R.

[0096] 4. Result Generation: The final R is a quantified value representing the overall explanatory strength of the "vibration anomaly" problem based on the current spectrogram data. The system can also retain intermediate results, such as the confidence scores of each path, to form richer conclusions.

[0097] Phase 4: Generating the final inference fusion result: The inference fusion result is a structured output, for example: Fusion_Result = { primary_cause: "mechanical cause", confidence: 0.85, supporting_evidences: ["Bearings_Wear", "Shaft_Misalignment"], alternative_causes: [{"electrical cause", 0.15}]}.

Claims

1. A multi-level information extraction and intelligent reasoning fusion optimization method based on a water and electricity industry knowledge graph, characterized by, The method comprises the following steps: Obtaining original knowledge graph data; Dividing the original knowledge graph data into a first graph region and a second graph region according to a data magnitude threshold judgment result; Generating a linkage guide label from the first graph region according to an extraction command, and inputting the linkage guide label into the second graph region for information fusion operation to obtain a reasoning fusion result; Matching and capturing the reasoning fusion result with elements in the original knowledge graph to obtain an extraction verification result.

2. The multi-level information extraction and intelligent reasoning fusion optimization method based on the water and electricity industry knowledge graph according to claim 1, characterized in that, The step of dividing the original knowledge graph data into a first graph region and a second graph region according to a data magnitude threshold judgment result comprises: Generating a magnitude judgment standard according to application scenario information; Comparing the data magnitude in the original knowledge graph with the magnitude judgment standard to obtain a graph division strategy; Dividing the original knowledge graph data according to the graph division strategy to obtain the first graph region and the second graph region, and the first graph region comprising all data root index elements.

3. The multi-level information extraction and intelligent reasoning fusion optimization method based on the water and electricity industry knowledge graph according to claim 1, characterized in that, The step of generating a linkage guide label from the first graph region according to an extraction command, and inputting the linkage guide label into the second graph region for information fusion operation to obtain a reasoning fusion result comprises: Collecting the extraction command generated by the client, the extraction command comprising extraction content and extraction index; Generating the linkage guide label from the first graph region according to the extraction command; Generating the reasoning fusion result through a reasoning fusion formula: ； Wherein, R is the reasoning fusion result, n is the number of linkage guide labels, and L is a data element of the second graph region.

4. The multi-level information extraction and intelligent reasoning fusion optimization method based on the water and electricity industry knowledge graph according to claim 1, characterized in that, The step of matching and capturing the reasoning fusion result with elements in the original knowledge graph to obtain an extraction verification result comprises: Searching the reasoning fusion result in a root index pool of the original knowledge graph, performing matching and capturing and comparison operations, and obtaining the extraction verification result; the extraction verification result comprising extraction qualification and extraction disqualification.

5. A multi-level information extraction and intelligent reasoning fusion optimization system based on a hydropower industry knowledge graph, used to execute the hydropower feature extraction and optimization method based on cross-modal heterogeneous data fusion according to any one of claims 1-4, characterized in that, The method comprises the following steps: An acquisition module is configured to acquire original knowledge graph data; A division module is configured to divide the original knowledge graph data into a first graph region and a second graph region according to a data magnitude threshold judgment result; An input module is configured to generate a linkage guide label from the first graph region according to an extraction command, and input the linkage guide label into the second graph region for information fusion operation to obtain a reasoning fusion result; A capture module is configured to match and capture the reasoning fusion result with elements in the original knowledge graph to obtain an extraction verification result.

6. The multi-level information extraction and intelligent reasoning fusion optimization system based on the water and electricity industry knowledge graph according to claim 5, characterized in that, The division module comprises: A generation unit is configured to generate a magnitude judgment standard according to application scenario information; A comparison unit is configured to compare the data magnitude in the original knowledge graph with the magnitude judgment standard to obtain a graph division strategy; A division unit is configured to divide the original knowledge graph data according to the graph division strategy to obtain the first graph region and the second graph region, and the first graph region comprising all data root index elements.

7. The multi-level information extraction and intelligent reasoning fusion optimization system based on the water and electricity industry knowledge graph according to claim 5, characterized in that, The input module comprises: A collection unit is configured to collect the extraction command generated by the client, the extraction command comprising extraction content and extraction index; An extraction unit is configured to generate the linkage guide label from the first graph region according to the extraction command; A fusion unit is configured to generate the reasoning fusion result through a reasoning fusion formula: ； Wherein, R is the reasoning fusion result, n is the number of linkage guide tags, and L is the data element of the second graph area.

8. The multi-level information extraction and intelligent reasoning fusion optimization system based on the water and electricity industry knowledge graph according to claim 5, characterized in that, The grabbing module comprises: The treatment unit is configured to search the reasoning fusion result in the root index pool of the original knowledge graph, perform matching grabbing and comparison operations, and obtain an extraction verification result, which includes extraction qualified and extraction unqualified.

9. A non-volatile storage medium, comprising: The non-volatile storage medium comprises a stored program, wherein the program controls the device in which the non-volatile storage medium is located to execute the multi-level information extraction and intelligent reasoning fusion optimization method based on the water and electricity industry knowledge graph according to any one of claims 1-4 when running.

10. An electronic device, comprising: The device comprises a processor and a memory; the memory stores computer readable instructions, and the processor is configured to run the computer readable instructions, wherein the computer readable instructions perform the multi-level information extraction and intelligent reasoning fusion optimization method based on the water and electricity industry knowledge graph according to any one of claims 1-4 when running.

Images