A knowledge graph guided coal combustion process characterization method

Abstract

The application discloses a kind of knowledge graph guided coal process characterization method.The application first carries out unified analysis and standardization to the text, image, voice and structured parameter etc. in the coal process scene Multimodal process information, on this basis, the information of different sources is carried out association fusion and structured organization, the key knowledge that can reflect process object, working condition state, running relationship etc. is extracted, and further constructs coal process knowledge graph.The method of the application can form knowledge enhanced state representation based on coal process knowledge graph for current working condition, so as to improve the comprehensive expression ability to complex working condition information, and provide unified characterization basis for subsequent process analysis.

Description

Technical Field

[0001] This invention belongs to the field of intelligent combustion optimization of coal-fired boilers, and particularly relates to a knowledge graph-guided method for characterizing coal-fired processes. Background Technology

[0002] The core issues to be addressed in coal-fired power plant process optimization and control are improving combustion efficiency, reducing pollutant emissions, and controlling operating costs while ensuring stable boiler operation. Achieving these goals involves multiple process stages, including coal blending, coal quality testing, boiler operation, and load regulation. Because these stages are interconnected, coal-fired power plant process optimization is not simply a matter of adjusting a single parameter; it is characterized by multi-stage coupling, multi-objective trade-offs, and dynamic changes.

[0003] In actual production sites, coal-fired processes continuously generate a large amount of multi-source heterogeneous information, which can manifest in various modalities such as text, images or videos, audio, and structured parameters. Typical textual information includes production logs, operation records, and process procedures; image or video information includes equipment images and monitoring screens; and structured parameter information includes coal quality test data, boiler operating parameters, and pollutant emission indicators. This information contains operating conditions, process rules, operational experience, and constraints, forming a crucial foundation for conducting intelligent analysis and characterization of coal-fired processes.

[0004] Currently, most methods related to coal-fired power processes rely on human experience, static rules, or single data-driven models. Traditional experience-based methods typically involve operators analyzing and judging the coal-fired power process status by combining real-time operating parameters, operating records, alarm information, and historical operating experience. Static rule-based methods identify the process status based on preset rules or thresholds. Single data-driven models extract partial features from the current operating conditions for modeling and analysis. Furthermore, with the development of multimodal information processing technology, existing multimodal information processing methods can extract and represent partial information from text, images, speech, and structured data.

[0005] Overall, although existing coal-fired power plant technologies can support operational condition monitoring, parameter analysis, and local feature extraction to some extent, they still have the following shortcomings:

[0006] First, existing methods still rely heavily on human experience, static rules, or single data-driven models, making it difficult to adapt to complex and changing operating conditions. Traditional experience-based methods depend on the professional skills of operators, resulting in limited stability and transferability of the results; while single data-driven models can extract some features based on the current operating conditions, they are usually limited to local conditions, making it difficult to fully utilize process rules and historical experience, and also difficult to adapt to complex operating environments.

[0007] Second, while multimodal process information is widespread, it lacks a unified structured organization method, making it difficult to directly support the analysis and characterization of coal-fired processes. Production logs, operation records, equipment images, monitoring screens, voice recordings, and sensor parameters contain rich process information, but existing methods still fall short in their ability to uniformly analyze and correlate this heterogeneous information. Summary of the Invention

[0008] This invention addresses the problems of dispersed multi-source heterogeneous information, implicit process relationships, and insufficient state representation in coal-fired process scenarios, and proposes a knowledge graph-guided coal-fired process characterization method.

[0009] To achieve the above-mentioned objectives, the present invention specifically adopts the following technical solution:

[0010] In a first aspect, the present invention provides a knowledge graph-guided method for characterizing coal combustion processes, comprising the following steps:

[0011] S1. Analyze, clean and standardize the multimodal process information in the coal-fired process scenario to form a standardized information set. Then extract the operating conditions, process rules, operating experience and constraints related to process analysis from the standardized information set to construct a set of process knowledge units.

[0012] S2. Based on the requirements for characterizing coal-fired processes and the set of process knowledge units, construct the graph pattern of the coal-fired process knowledge graph, and then generate the node set and relation edge set of the coal-fired process knowledge graph. Combine these with the attribute set and constraint set to obtain the coal-fired process knowledge graph.

[0013] S3. Using the current multimodal process information as the retrieval basis, perform relevant subgraph searches in the coal-fired process knowledge graph to determine local knowledge subgraphs; then, compare the current multimodal process information with the local knowledge subgraphs. Figure 1 The input knowledge-enhanced representation model is encoded and fused to obtain a knowledge-enhanced state representation, thereby completing the characterization of the coal-fired process.

[0014] Based on the above scheme, each step can be implemented in the following preferred manner.

[0015] As a preferred embodiment of the first aspect mentioned above, the specific process of S1 is as follows:

[0016] S11. Obtain multimodal process information in the coal-fired process scenario and form a multimodal process information set containing four types of modal process information: text-based process information, image-based process information, voice-based process information, and structured parameter-based process information.

[0017] S12. The various modal process information is parsed, cleaned and standardized to form a standardized information set consisting of multiple standardized information items. Each standardized information item includes at least the set of process objects involved in the standardized information item, standardized information content and time identifier.

[0018] S13. Perform cross-source semantic consistency identification on each standardized information item to determine the same process object, the same operating condition, or the same running event involved in different modal process information, and form a set of associated results composed of multiple associated results;

[0019] S14. Based on the association result set, generate a process knowledge unit set consisting of five types of process knowledge units: process object units, operating condition units, operation event units, rule constraint units, and experience information units. This set is used to structure the expression of process objects, operating conditions, operation events, rule constraints, and experience information in the coal-fired process scenario. Each process knowledge unit... Represented as , Indicates the type of process knowledge unit. This represents the set of process objects involved in a process knowledge unit. This represents the content of the process knowledge unit. A time marker indicating a unit of process knowledge.

[0020] As a preferred embodiment of the first aspect mentioned above, in S13, the cross-source semantic consistency identification specifically involves: for each normalized information item in the normalized information set, if different normalized information items contain the same process object, then the association type of these normalized information items is defined as the association of the same process object; when different normalized information items involve the same or related process objects and the time identifiers are in the same or adjacent time ranges, if the normalized information content of different normalized information items describes the same or similar operating conditions, then the association type of these normalized information items is defined as the association of the same operating condition; if the normalized information content of different normalized information items describes the same or similar operating events, then the association type of these normalized information items is defined as the association of the same operating event; finally, the association result is constituted by the associated normalized information items, the unified semantic content corresponding to the association of normalized information items, and the association type, wherein the unified semantic content includes process object, operating condition, and operating event.

[0021] As a preferred embodiment of the first aspect mentioned above, in S14, if the association type of the association result is the same process object association, then the associated process object is taken as a process object unit; if the association type is the same operating condition state association, then the unified semantic content of the association result is taken as the operating condition state unit content, and the process object and time identifier involved in the operating condition state are determined to generate the operating condition state unit; if the association type is the same running event association, then the unified semantic content of the association result is taken as the running event unit content, and the process object and time identifier involved in the running event are determined to generate the running event unit; through semantic parsing, the process rules, constraints and operating experience contained in the standardized information set are identified, and the process rules and constraints are organized according to the preset knowledge unit structure to generate rule constraint units, and the operating experience is used to generate experience information units.

[0022] As a preferred embodiment of the first aspect mentioned above, in S2, when constructing the graph pattern, firstly, based on the requirements for representing the coal combustion process, the scope of objects to be expressed by the coal combustion process knowledge graph and the set of mapping rules are determined; then, based on the content and purpose of various process knowledge units in S1, the set of node types, the set of relation types, the set of attribute types, and the set of constraint types are determined, thereby forming the graph pattern; wherein, the set of mapping rules includes four types of mapping rules, namely node generation rules, relation edge generation rules, attribute generation rules, and constraint generation rules.

[0023] As a preferred embodiment of the first aspect above, the node generation rule is specifically as follows: process object class nodes are generated from the process object names in the process object unit, operating condition class nodes are generated from the knowledge unit content in the operating condition unit, and operating event class nodes are generated from the knowledge unit content in the operating event unit.

[0024] As a preferred embodiment of the first aspect above, the relation edge generation rule is specifically as follows: if If it is a working condition unit, then search for the node set that matches the condition. For the corresponding process object class nodes, if two found process object class nodes are related, a relationship edge is generated between the two corresponding process object class nodes; if It is a working condition unit and has been... If a working condition status node is generated, then the node set is searched for a match. The corresponding process object class node is then used to generate a relationship edge between this process object class node and the operating condition class node; if For running event units and already by If a node representing a runtime event is generated, then the node set is searched for a match. The corresponding process object class node is used to generate a relationship edge between the process object class node and the running event class node; if the process knowledge unit content of the experience information unit... If a relationship exists between a certain operating condition state and a running event, then a relationship edge is generated between the corresponding operating condition state class node and the running event class node.

[0025] As a preferred embodiment of the first aspect above, the attribute generation rule is specifically as follows: If If it is a working condition unit, then... As a state property, attached to The corresponding process object class node; if To run the event unit, then... As an event property attached to The corresponding process object class node; when the process knowledge unit contains a time identifier. At that time, then As a time attribute, it is attached to the corresponding node or relation edge.

[0026] As a preferred embodiment of the first aspect above, the constraint generation rule is specifically: if When it is a rule-constrained unit, Match with process object type nodes, operating condition type nodes, and running event type nodes, and... As a constraint, it is attached to the corresponding matching node.

[0027] As a preferred embodiment of the first aspect mentioned above, in S3, the newly added multimodal process information at the current time compared to the previous time is obtained as the input set. The node set, relation edge set, attribute set, and constraint set in the coal combustion process knowledge graph at the current time are used as graph elements. If there is a graph element in the coal combustion process knowledge graph at the current time that corresponds to the input set, the corresponding graph element is updated; if there is no corresponding graph element, the corresponding graph element is added according to the graph pattern and mapping rules to achieve incremental updating of the coal combustion process knowledge graph, resulting in the updated coal combustion process knowledge graph and redetermining the local knowledge subgraph. Then, the multimodal process information at the current time is compared with the redetermined local knowledge subgraph. Figure 1 The input knowledge-enhanced representation model is encoded and fused to obtain an updated knowledge-enhanced state representation.

[0028] In a second aspect, the present invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, can implement the knowledge graph-guided coal combustion process characterization method as described in any of the solutions of the first aspect above.

[0029] Thirdly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the knowledge graph-guided coal combustion process characterization method as described in any of the solutions of the first aspect above.

[0030] Fourthly, the present invention provides a computer electronic device, which includes a memory and a processor;

[0031] The memory is used to store computer programs;

[0032] The processor is configured to, when executing the computer program, implement the knowledge graph-guided coal combustion process characterization method as described in any of the first aspects above.

[0033] Compared with existing technologies, the knowledge graph-guided coal combustion process characterization method proposed in this invention has at least the following beneficial effects:

[0034] First, this invention explicitly unifies the representation of multimodal process information, improving the consistency of information organization. By parsing, cleaning, merging, and unifying the representation of multimodal information such as text, images, speech, and structured parameters in coal-fired process scenarios, this invention organizes information scattered from different sources into unified process knowledge units, thereby reducing the fragmentation of expression between heterogeneous multi-source information and improving the consistency and completeness of process information organization.

[0035] Second, this invention constructs a coal-fired power process knowledge graph to enhance the ability to express the relationships between process elements. This invention integrates and structures process objects, operating conditions, operational events, rule constraints, and experiential information to construct a coal-fired power process knowledge graph, enabling the explicit expression of process elements and their relationships, thereby enhancing the ability to express process relationships, state evolution, and constraints.

[0036] Third, this invention generates knowledge-enhanced state representations for the current operating conditions, improving the representation quality under complex operating conditions. Based on a coal-fired power process knowledge graph, this invention performs graph mapping and knowledge enhancement on the current operating conditions, generating knowledge-enhanced state representations for process analysis. These representations not only include original observation information but also integrate process relationships, constraint information, and semantic knowledge, thereby improving the state representation capability and interpretability under complex operating conditions.

[0037] Fourth, this invention introduces a local mapping and incremental update mechanism to improve the timeliness and adaptability of the representation. When new process information or execution feedback arrives, this invention incrementally updates the knowledge graph and the knowledge-enhanced state representation, enabling the knowledge-enhanced state representation to continuously iterate and optimize with changes in the field conditions, avoiding repeated reconstruction of historical information, thereby improving the response speed and adaptability of the knowledge-enhanced state representation to changes in operating conditions.

[0038] Fifth, this invention provides a unified interface for subsequent process analysis and solution support, improving the method's transferability. This invention unifies multimodal information, knowledge graphs, and knowledge-enhanced state representation into a single process, providing a unified input interface for subsequent process analysis, state understanding, constraint matching, and solution support, thereby enhancing the method's ability to adapt to different operating conditions and data sources. Attached Figure Description

[0039] Figure 1 This is a schematic flowchart of the method of the present invention;

[0040] Figure 2 This is a flowchart of the multimodal process information parsing and unified characterization in the method of this invention;

[0041] Figure 3 This is a flowchart of the local knowledge subgraph extraction process in the method of this invention;

[0042] Figure 4 This is a flowchart of the knowledge organization and state representation in the method of the present invention;

[0043] Figure 5 This is a flowchart supporting the process scheme analysis in the method of the present invention;

[0044] Figure 6 This is a schematic diagram of the characterization update and iterative optimization process in the method of the present invention. Detailed Implementation

[0045] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below. Technical features in the various embodiments of the present invention can be combined accordingly without mutual conflict.

[0046] In the description of this invention, it should be understood that the terms "first" and "second" are used only for descriptive purposes and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first" and "second" may explicitly or implicitly include at least one of those features.

[0047] To facilitate understanding of the technical solution of this invention, the key terms involved in this invention are defined and explained as follows:

[0048] 1) Multimodal process information: refers to the original process information in the coal combustion process scenario, which consists of multiple sources such as text, images, voice and structured parameters, and is used to characterize the operating status, operation records, equipment status and environmental characteristics in the coal combustion process.

[0049] 2) Standardized information set: refers to the information set formed after parsing, cleaning and standardizing the multimodal process information, which is used to unify the terminology, time reference and recording format of different modal information.

[0050] 3) Association result set: refers to the result set formed after identifying the correspondence between different information items in the normalized information set and associating them across modes. It is used to represent the same process object, the same working condition, or the same running event involved in different modal information.

[0051] 4) Process knowledge unit: refers to a structured knowledge object formed by merging, extracting and organizing process objects, operating conditions, running events, rule constraints and experience information based on a standardized information set and a set of associated results. It is used for subsequent knowledge graph construction.

[0052] 5) Graph pattern: refers to the structural specification used to limit the organization of coal-fired process knowledge graph, which includes at least the set of node types, the set of relation types, the set of attribute types, the set of constraint types and their organization rules.

[0053] 6) Coal-fired process knowledge graph: refers to a graph structure knowledge representation built based on process knowledge units, used to uniformly express process objects, operating conditions, running events, relation edges, attribute information and constraint information in coal-fired processes.

[0054] 7) Current operating condition input: refers to the multimodal process information acquired at a certain moment, used to characterize the real-time input content under the current coal-fired operating conditions, and as the basic input for generating knowledge-enhanced state representations.

[0055] 8) Local knowledge subgraph: refers to the local graph structure related to the current working condition retrieved from the coal-fired process knowledge graph based on the current working condition input, used to provide the process knowledge context corresponding to the current working condition.

[0056] 9) Graph mapping: refers to the process of confirming the correspondence between the current working condition input and the graph elements in the local knowledge subgraph, used to determine the corresponding position of objects, states, events or constraints in the current working condition in the knowledge graph.

[0057] 10) Mapping result: refers to the correspondence result obtained after the graph mapping, which is used to represent the mapping relationship between various types of information in the current working condition input and knowledge graph nodes, relation edges or constraint edges.

[0058] 11) Knowledge-enhanced state representation: refers to the unified representation result generated based on the current operating condition input and local knowledge subgraph, which is used to simultaneously encode the original operating condition information, process relationship and constraint information to support subsequent process analysis.

[0059] 12) Candidate process scheme set: refers to the set of multiple alternative process adjustment schemes formed for the current operating conditions, which are used for consistency analysis, constraint matching analysis and adaptability analysis.

[0060] 13) Analysis result set: refers to the result set obtained after analyzing the candidate process scheme set, which is used to characterize the degree of matching between each candidate scheme and the current operating conditions and its applicability.

[0061] 14) Input set: refers to the multimodal process information, equipment status changes, execution feedback or adjustment records newly generated during the process, which are used to update the coal-fired process knowledge graph and knowledge-enhanced status representation.

[0062] like Figure 1 As shown, in a preferred embodiment of the present invention, the knowledge graph-guided coal combustion process characterization method includes the following steps S1 to S3. The specific implementation process of each step will be described in detail below.

[0063] S1. The multimodal process information in the coal-fired process scenario is analyzed, cleaned and standardized to form a standardized information set. Then, the operating conditions, process rules, operating experience and constraints related to the process analysis are extracted from the standardized information set to construct a set of process knowledge units, which provides basic data for subsequent knowledge organization and representation generation.

[0064] It should be noted that, in this invention, as Figure 2 As shown, the specific process of step S1 is as follows:

[0065] S11. Obtain multimodal process information in the coal-fired process scenario and form a multimodal process information set. Multimodal process information includes four types of modal process information, namely text-based process information. Image-based process information Voice-based process information Structured parameter-based process information .

[0066] In this embodiment S11, the multimodal process information set can be used as the original multimodal process input for subsequent analysis and characterization. Among them, text-based process information includes production logs, operation records, and process procedures; image-based process information includes equipment images, monitoring screens, and furnace flame images; voice-based process information includes on-site voice recordings and dispatch voice recordings; and structured parameter-based process information includes coal quality parameters, coal feed rate, and sensor parameters.

[0067] S12. The various modal process information is parsed, cleaned and standardized to form a standardized information set consisting of multiple standardized information items. Each standardized information item includes at least the set of process objects involved in the standardized information item, standardized information content and time identifier.

[0068] In this embodiment S12, for text-based process information, this embodiment performs tokenization, entity normalization, and entity recognition to extract key information related to process objects, operating states, and abnormal events; for image-based process information, this embodiment performs equipment target recognition, instrument reading recognition, flame status recognition, and image semantic description generation to extract equipment status and environmental features from the image; for voice-based process information, this embodiment performs speech transcription, noise filtering, and keyword recognition to obtain the operation instructions, status descriptions, and experience information involved in the speech; for structured parameter-based process information, this embodiment performs missing value processing, outlier detection, time alignment, unit unification, and statistical feature extraction to form parameter features that can be directly used in process analysis.

[0069] Through this step, various types of process information from different sources are transformed into a standardized information set with unified terminology and a unified time reference. ,in, Indicates the number of normalized information items. (The rest of the text appears to be a list of terms or tags, possibly related to Standardized information items For example, it includes at least a set of process objects. Standardized information content and time markers This is used to represent a unified record format of various process information after it has been organized, and serves as the input for subsequent cross-modal semantic alignment; the standardized information content includes operating status, operating events, process rules, operating experience, or constraints.

[0070] For example, for text logs " "Current of No. 1 coal mill is too high" (voice transcription) The message "Current of coal mill No. 1 is a bit high" and the structured parameter "M1_Current exceeds the set limit" indicate that the current of the No. 1 coal mill is too high. After processing in step S12, standardized information items can be generated respectively:

[0071]

[0072]

[0073]

[0074] S13. Perform cross-source semantic consistency identification on each standardized information item to determine the same process object, the same operating condition, or the same running event involved in different modal process information, and form a set of associated results composed of multiple associated results.

[0075] In this embodiment S13, cross-source semantic consistency identification specifically involves: for each normalized information item in the normalized information set, if different normalized information items contain the same process object, then the association type of these normalized information items is defined as the association of the same process object; when different normalized information items involve the same or related process objects and the time identifiers are in the same or adjacent time ranges, if the normalized information content of different normalized information items describes the same or similar operating conditions, then the association type of these normalized information items is defined as the association of the same operating condition; if the normalized information content of different normalized information items describes the same or similar running events, then the association type of these normalized information items is defined as the association of the same running event; finally, the association result is constituted by the associated normalized information items, the unified semantic content corresponding to the association of normalized information items, and the association type, wherein the unified semantic content includes process object, operating condition, and running event.

[0076] The unified semantic content refers to the common part of the two standardized information items, that is, the completely identical part, which can be a process object, operating condition, or running event.

[0077] Specifically, this embodiment performs consistency checks on each standardized information item, including process object consistency check, operating condition consistency check, and running event consistency check:

[0078] (1) For the consistency judgment of process objects, this embodiment identifies whether the same process objects are contained in different standardized information items based on the set of process objects in the standardized information items;

[0079] (2) Regarding the consistency judgment of working conditions, in this embodiment, when different standardized information items involve the same or related process objects and the time identifiers are in the same or adjacent time ranges, the semantic consistency of the standardized information content is used to determine whether different standardized information items describe the same or similar working conditions.

[0080] (3) For the consistency judgment of the running event, based on the set of process objects, standardized information content and time identifier in the standardized information item, it is similar to identify whether different standardized information items describe the same running event.

[0081] Finally, in this embodiment, the standardized information items that meet the above consistency judgment conditions are respectively grouped into association results of the same process object, association results of the same operating condition, and association results of the same operating event, forming a set of association results. ;in, Indicates the number of associated results. (The last part is a repetition of the previous one and can be left as is.) One related result For example, it can be represented as , Indicates the first The set of normalized information items contained in each associated result; This represents the unified semantic content corresponding to the association of standardized information items; The association type is indicated, which includes at least the association of the same process object, the association of the same operating condition, and the association of the same running event.

[0082] For example, based on the normalized information item obtained in step S12 , and Since all three processes include "Coal Mill No. 1" in their process object sets, and their standardized information content describes the "abnormal high current" status, and , and Within the same time frame, therefore , and It can form associations between the same process object and operating conditions, such as:

[0083]

[0084]

[0085] S14. Based on the association result set, generate a process knowledge unit set consisting of five types of process knowledge units: process object unit, operating condition state unit, running event unit, rule constraint unit, and experience information unit. It is used to structure and represent process objects, operating conditions, operational events, rule constraints, and experience information in coal-fired process scenarios. This indicates the number of process knowledge units.

[0086] Furthermore, in this embodiment S14, if the association type of the association result is the same process object association, then the associated process object is taken as a process object unit; if the association type is the same operating condition state association, then the unified semantic content of the association result is taken as the operating condition state unit content, and the process object and time identifier involved in the operating condition state are determined to generate the operating condition state unit; if the association type is the same running event association, then the unified semantic content of the association result is taken as the running event unit content, and the process object and time identifier involved in the running event are determined to generate the running event unit; through semantic parsing, the process rules, constraints and operating experience contained in the standardized information set are identified, and the process rules and constraints are organized according to the preset knowledge unit structure to generate rule constraint units, and the operating experience is used to generate experience information units.

[0087] Each process knowledge unit can be represented as: ,in, Indicates the type of process knowledge unit. This represents the set of process objects involved in a process knowledge unit. This represents the content of the process knowledge unit. A time marker indicating a unit of process knowledge.

[0088] For example, different types of process knowledge units can be represented as:

[0089]

[0090]

[0091] This results in a unified, associative, and searchable set of process knowledge units, which serves as the foundational data for subsequent construction of coal-fired process knowledge graphs and generation of knowledge-enhanced representations.

[0092] S2. Based on the requirements for representing coal-fired processes and the set of process knowledge units, construct the schema of the coal-fired process knowledge graph, and then generate the node set and relation edge set of the coal-fired process knowledge graph. These are then merged with the attribute set and constraint set to obtain the coal-fired process knowledge graph, which serves as the basis for subsequent knowledge-enhanced state representation generation.

[0093] It should be noted that in S2 of this invention, the graph pattern is used to define the organization of the set of process knowledge units in the coal combustion process knowledge graph, and to specify the mapping rules from different types of process knowledge units to nodes, relationships, attributes, and constraints. When constructing the graph pattern, the scope of objects to be expressed by the coal combustion process knowledge graph and the set of mapping rules are first determined according to the requirements for representing the coal combustion process. Then, based on the content and purpose of each process knowledge unit in S1, determine the set of node types. Set of relation types Attribute type collection and constraint type set Thus forming a map pattern .

[0094] In this embodiment, the node type set includes process object type nodes, operating condition type nodes, and running event type nodes; the relationship type set includes the relationship between process objects and operating conditions, the relationship between process objects and running events, the relationship between operating conditions and running events, and the relationship between process objects; the attribute type set includes operating condition attributes, running event attributes, and time attributes; and the constraint type set is determined according to the constraint conditions in the rule constraint unit.

[0095] In this embodiment, the mapping rule set includes four types of mapping rules: node generation rules, relation edge generation rules, attribute generation rules, and constraint generation rules. These mapping rules refer to the methods used to map different types of process knowledge units to nodes, relation edges, attributes, or constraints in the knowledge graph, based on knowledge in the coal combustion process domain, the characteristics of the coal combustion process, and the field structure of the process knowledge unit. Specifically, the node generation rule determines whether a corresponding process knowledge unit can generate a node and the type of node generated, based on the process knowledge unit type, the set of process objects involved in the process knowledge unit, the content of the process knowledge unit, and the time identifier of the process knowledge unit. The relation edge generation rule determines the starting node, ending node, and relation type of the relation edge based on the process knowledge unit type, the set of process objects involved in the process knowledge unit, the content of the process knowledge unit, and the set of generated nodes. The attribute generation rule determines the attribute content, attribute type, and its associated object based on the process knowledge unit type, the content of the process knowledge unit, and the time identifier. The constraint generation rule determines the constraint content, constraint type, and its associated object based on the set of process objects involved in the constraint unit, the content of the process knowledge unit, and the time identifier.

[0096] The node generation rules are as follows: process object class nodes are generated from the process object names in the process object unit, operating condition class nodes are generated from the knowledge unit content in the operating condition unit, and operating event class nodes are generated from the knowledge unit content in the operating event unit.

[0097] The specific rule for generating relation edges is: if If it is a working condition unit, then search for the node set that matches the condition. For the corresponding process object class nodes, if two found process object class nodes are related, a relationship edge is generated between the two corresponding process object class nodes; if It is a working condition unit and has been... If a working condition status node is generated, then the node set is searched for a match. The corresponding process object class node is then used to generate a relationship edge between this process object class node and the operating condition class node; if For running event units and already by If a node representing a runtime event is generated, then the node set is searched for a match. The corresponding process object class node is used to generate a relationship edge between the process object class node and the running event class node; if the process knowledge unit content of the experience information unit... If a relationship exists between a certain operating condition state and a running event, then a relationship edge is generated between the corresponding operating condition state class node and the running event class node.

[0098] The specific attribute generation rule is as follows: If If it is a working condition unit, then... As a state property, attached to The corresponding process object class node; if To run the event unit, then As an event property attached to The corresponding process object class node; when the process knowledge unit contains a time identifier. At that time, then As a time attribute, it is attached to the corresponding node or relation edge.

[0099] The constraint generation rule is specifically as follows: If When it is a rule-constrained unit, Match with process object type nodes, operating condition type nodes, and running event type nodes, and... As a constraint, it is attached to the corresponding matching node. For example, for the rule constraint unit "the current of No. 1 coal mill shall not exceed the set upper limit", "the current shall not exceed the set upper limit" is used as the constraint content, and this constraint content is attached to the process object class node corresponding to "No. 1 coal mill" and the relevant operating condition status class node of "No. 1 coal mill current".

[0100] It should be noted that in S2 of this invention, the node set for generating the coal combustion process knowledge graph is... Specifically, based on the set of node types and node generation rules in the graph pattern, it is determined whether each process knowledge unit in the process knowledge unit set can generate a node; then, according to the node generation rules, the process knowledge units that meet the requirements are generated into nodes, thereby obtaining the node set.

[0101] It should be noted that in S2 of this invention, the set of relation edges for generating the knowledge graph of coal combustion processes is... Specifically, based on the set of relation types and the relation edge generation rules in the graph pattern, it is determined whether each process knowledge unit in the process knowledge unit set can generate a relation edge; for process knowledge units or combinations of process knowledge units that can generate relation edges, the starting node, ending node, and relation type of the relation edge are determined through the relation edge generation rules; the nodes corresponding to the content of the starting node and the ending node are searched in the node set; when both the starting node and the ending node exist, the corresponding relation edge is generated, thus obtaining the relation edge set.

[0102] It should be noted that in S2 of this invention, the attribute set for generating the coal combustion process knowledge graph is... and constraint set Specifically, based on the attribute type set, constraint type set, and mapping rules in the graph pattern, it is determined whether each process knowledge unit in the process knowledge unit set can generate attributes or constraints. For process knowledge units that can generate attributes, the attribute content, attribute type, and attribute attachment object are determined through attribute generation rules, and then attached to the corresponding node or relation edge to form an attribute set, which is used to describe the state, event, and time information of the node or relation edge. For process knowledge units that can generate constraints, the constraint content, constraint type, and constraint attachment object are determined through constraint generation rules, and then attached to the corresponding node or relation edge to form a constraint set, which is used to limit the applicable scope or operating conditions of the node or relation edge.

[0103] S3. Using the current multimodal process information as the retrieval basis, perform relevant subgraph searches in the coal-fired process knowledge graph to determine local knowledge subgraphs; then, compare the current multimodal process information with the local knowledge subgraphs. Figure 1 The input knowledge-enhanced representation model is encoded and fused to obtain a knowledge-enhanced state representation, thereby completing the characterization of the coal-fired process.

[0104] It should be noted that, in S3 of this invention, the set of multimodal process information at the current moment is represented as follows: , , , and These represent the current moment's text-based process information, image-based process information, voice-based process information, and structured parameter-based process information, respectively. Based on this, such as... Figure 3 As shown, knowledge graph subgraph retrieval methods, such as entity linking-based graph retrieval, can be used to select knowledge contexts related to the current operating conditions from the coal combustion process knowledge graph, i.e., local knowledge subgraphs. This provides a knowledge base for subsequent representation generation.

[0105] It should be noted that in S3 of this invention, the knowledge enhancement representation model can be implemented using architectures such as Transformer or GNN, and may include processing methods such as feature encoding, graph structure encoding, concatenation fusion, or attention fusion, etc., which are not limited in this invention. Figure 4 As shown, through the model encoding and fusion steps, this invention unifies the original operating condition information and the prior knowledge of the graph into the same representation space, forming a state representation that takes into account both real-time operating condition information and prior knowledge of the graph. This enhances the comprehensive expression capability of complex operating conditions such as operating condition characteristics, process relationships and constraint information, thereby providing a foundation for subsequent representation updates and analysis.

[0106] The above knowledge-enhanced state representation It can be further written as:

[0107]

[0108] in, Represents the characteristics of the process object. This indicates the characterization of the operating condition. This indicates the characterization of process relationships. This represents the process constraints. These knowledge-enhanced state representations can serve as a unified input for subsequent process analysis, used to perform consistency analysis, constraint matching analysis, or adaptability analysis on candidate process schemes, providing representational support for the formulation, adjustment, and optimization of candidate process schemes. Specifically, such as... Figure 5 As shown, this embodiment first obtains candidate process solutions generated by process control, process adjustment, or customization strategies. The candidate process solutions are then used as input for subsequent analysis. These candidate solutions can come from historical operational experience, a strategy library, or a preset strategy set, and may include the operation object, operation conditions, operation sequence, operation content, parameter settings, and application scenarios. This embodiment then evaluates the conformity of each candidate process solution with the current operating condition in terms of state representation, process constraints, and operational adaptation; that is, it performs consistency analysis, constraint matching analysis, or adaptability analysis on the candidate process solutions. Consistency analysis can be implemented using existing rule matching methods or representation similarity calculation methods to determine whether the operation object, operation content, or parameter settings of the candidate process solution correspond to the current process object, operating condition state representation, and process relationship representation. Constraint matching analysis can be implemented using existing graph constraint verification methods to determine whether the parameter settings, operation conditions, or application scenarios of the candidate process solution satisfy the current process constraint representation and the operating boundaries, rule conditions, or security restrictions corresponding to the local knowledge subgraph. Adaptability analysis can be implemented using existing precondition validation methods, feasibility analysis methods, or rule matching methods. It is used to determine whether the operating conditions, parameter settings, or application scenarios of candidate process solutions are compatible with the current operating state representation, and whether the candidate process solution has the conditions for execution under the current operating conditions. After the above multi-dimensional analysis, this embodiment summarizes the consistency evaluation results, constraint matching results, and adaptability evaluation results of each candidate process solution to obtain the final set of analysis results. This set serves as the analytical basis for process solution formulation, adjustment, and optimization, providing analytical information for each candidate process solution, but not directly determining the final execution result. Through this step, process solution analysis support based on knowledge-enhanced state representation is achieved, providing an output foundation for subsequent solution adjustment and optimization.

[0109] It should also be noted that in S3 of this invention, as Figure 6 As shown, the multimodal process information added at the current time compared to the previous time is obtained as the input set. The knowledge graph of coal-fired technology at the current moment The set of nodes, edges, attributes, and constraints in the graph are used as graph elements. If a graph element corresponding to the input set exists in the current coal combustion process knowledge graph, the corresponding graph element is updated; otherwise, a corresponding graph element is added according to the graph pattern and mapping rules, realizing incremental updates of the coal combustion process knowledge graph and obtaining the updated coal combustion process knowledge graph. Then, the local knowledge subgraph is redefined, and the multimodal process information at the current moment is combined with the redefined local knowledge subgraph. Figure 1 The input knowledge-enhanced representation model is encoded and fused to obtain the updated knowledge-enhanced state representation. .

[0110] Through the above process, this invention only corrects local nodes, relation edges, attribute sets, and constraint sets affected by the newly added information, without reconstructing the entire knowledge graph, thus maintaining the continuity of historical knowledge and reducing update overhead. The finally updated knowledge-enhanced state representation will serve as the basic input for the next round of process analysis, enabling subsequent representation generation to continuously absorb new process information and execution feedback, achieving iterative optimization of the representation results. Therefore, this invention forms a closed-loop update mechanism for coal-fired process analysis, allowing the knowledge graph and knowledge-enhanced state representation to continuously evolve with the process, thereby improving their accuracy and applicability for subsequent process analysis and solution support.

[0111] To better demonstrate the specific implementation and technical effects of the present invention, the knowledge graph-guided coal combustion process characterization method shown in steps S1 to S3 of the above preferred implementation is applied to two specific examples.

[0112] Example 1

[0113] This embodiment is an experiment on anomaly identification and future state ranking based on boiler time-series datasets. It aims to demonstrate the application effect of knowledge graph-guided coal-fired process characterization method on boiler operation data. The specific implementation process of the knowledge graph-guided coal-fired process characterization method used is as described above and will not be repeated here.

[0114] This embodiment uses a publicly available time-series dataset of coal-fired boilers as the verification object. This dataset was collected from a coal-fired boiler in a chemical plant in Zhejiang Province, with a sampling interval of 5 seconds. It contains 86,400 samples and 30 operating variables, including pressure, temperature, flow rate, air volume, flue gas temperature, and steam temperature. On this dataset, this embodiment sets two tasks: first, an anomaly identification task, using whether the future boiler outlet steam temperature exceeds the normal range [530, 545] as a binary classification label to verify the model's ability to identify abnormal operating conditions; second, a future state ranking task, discretizing the boiler outlet steam temperature after 6 future time steps into 8 candidate states and requiring the model to rank these candidate states to verify the model's ability to represent future operating conditions.

[0115] The evaluation metrics used in this embodiment are as follows: For the anomaly identification task, Accuracy, Precision, Recall, F1, AUC, and AP are used as evaluation metrics to comprehensively reflect the accuracy, precision, recall, and overall discriminative ability of the classification results; For the future state ranking task, Hits@1 and MRR are used as evaluation metrics, where Hits@1 is used to measure the proportion of the true state ranked first, and MRR is used to measure the average reciprocal ranking of the true state.

[0116] This embodiment selects Majority, LR, RF, GBDT, Temporal Transformer, and PatchTransformer as comparison methods. The experimental results for each method on the anomaly recognition task are shown in Table 1, and the experimental results for the future state ranking task are shown in Table 2. Here, Majority represents consistently predicting the class or state with the highest frequency in the training set, serving as the most basic comparison baseline; LR, RF, and GBDT are traditional machine learning baselines; TemporalTransformer is the standard Transformer temporal modeling method based on time windows; and PatchTransformer is a Transformer temporal modeling method based on time segmentation.

[0117] Table 1. Results of the anomaly detection task

[0118]

[0119] Table 2. Results of the Future State Ranking Task

[0120]

[0121] The results above demonstrate that the method of this invention achieves superior results in both anomaly identification and future state ranking tasks, indicating that the knowledge graph-guided boiler operating condition representation process can effectively improve the unified representation capability and future state discrimination capability of boiler time-series operating condition information. This is because this invention does not simply model the original time-series features, but rather combines the processes described in S1-S3 above to uniformly parse the operating condition data, organize it into a knowledge graph, map local subgraphs, and generate knowledge-enhanced state representations. Furthermore, it introduces constraint information and an incremental update mechanism, thereby enabling the model to simultaneously possess the comprehensive expression capability of original operating condition features, process relationship information, and rule constraint information, ultimately improving the overall performance of anomaly identification and state ranking.

[0122] Example 2

[0123] This embodiment is a scheme analysis support experiment for knowledge-enhanced state representation, aiming to demonstrate the application effect of the knowledge graph-guided coal combustion process characterization method in the unified characterization and cross-modal alignment of multi-source heterogeneous information of coal and heat. The specific implementation process of the knowledge graph-guided coal combustion process characterization method used is as described above and will not be repeated here.

[0124] This embodiment uses a publicly available near-infrared multimodal dataset of coal and rock as the verification object. This dataset contains four types of data: sample image information (Photos), sample description documents (Documentation), reflectance spectral data (Spectra), and sample analysis results (Analysis). Among them, sample image information (Photos) is used to characterize the appearance characteristics of coal samples, sample description documents (Documentation) are used to characterize the source and category of samples, reflectance spectral data (Spectra) is used to characterize the spectral response characteristics of coal samples, and sample analysis results (Analysis) are used to characterize the analytical properties of coal samples.

[0125] In this embodiment, reflectance spectral data is used as the query modality, and sample description documents, sample image information, and sample analysis results are used as the target modality candidate library to construct a cross-modal retrieval and alignment task. That is, based on the spectral information, the corresponding documents, images, and analysis results are retrieved, and the true corresponding items are required to be ranked as high as possible in the candidate sorting.

[0126] The evaluation metrics used in this example are R@1, R@5, R@10, and MRR. R@1 measures the proportion of true correspondences ranked first, R@5 and R@10 measure the proportions of true correspondences ranked in the top 5 and top 10, respectively, and MRR measures the average reciprocal ranking of true correspondences. MRR can be used to assess the accuracy and ranking of cross-modal alignment results.

[0127] This example selects Raw-Cosine, CCA, and PLS as comparison methods. The document retrieval results based on spectral data for each method are shown in Table 3, the image retrieval results based on spectral data for each method are shown in Table 4, and the retrieval results based on spectral analysis results for each method are shown in Table 5. Raw-Cosine is a naive retrieval method based on the cosine similarity of the original features; CCA is a classic canonical correlation analysis method; and PLS is a partial least squares projection method.

[0128] Table 3. Spectrum-based document retrieval results

[0129]

[0130] Table 4. Image retrieval results based on spectral data

[0131]

[0132] Table 5. Retrieval Results Based on Spectral Analysis

[0133]

[0134] The results above demonstrate that the method of this invention effectively improves the modeling ability of the correspondence between spectra and documents, images, and analysis results in cross-modal retrieval and alignment experiments on near-infrared multimodal datasets of coal and rock. Compared with traditional alignment methods such as Raw-Cosine, CCA, and PLS, the method of this invention shows significant advantages in both R@1 and MRR metrics, indicating that the knowledge graph-guided representation method described in this invention can more effectively integrate multi-source heterogeneous information and enhance cross-modal semantic alignment capabilities.

[0135] It is understood that the knowledge graph-guided coal combustion process characterization method described in S1-S3 above can essentially be implemented by a computer program. Therefore, based on the same inventive concept, another preferred embodiment of the present invention also provides a computer program product corresponding to the knowledge graph-guided coal combustion process characterization method provided in the above embodiments, which includes a computer program / instructions. When the computer program / instructions are executed by a processor, they can implement the knowledge graph-guided coal combustion process characterization method as described in the above embodiments.

[0136] Similarly, based on the same inventive concept, another preferred embodiment of the present invention also provides a computer electronic device corresponding to the knowledge graph-guided coal combustion process characterization method provided in the above embodiment, which includes a memory and a processor;

[0137] The memory is used to store computer programs;

[0138] The processor is configured to implement a knowledge graph-guided coal combustion process characterization method according to the above embodiments when executing the computer program.

[0139] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0140] Therefore, based on the same inventive concept, another preferred embodiment of the present invention also provides a computer-readable storage medium corresponding to the knowledge graph-guided coal combustion process characterization method provided in the above embodiments. The storage medium stores a computer program, which, when executed by a processor, can implement the knowledge graph-guided coal combustion process characterization method in the above embodiments.

[0141] Specifically, in the computer-readable storage medium of the above three embodiments, the stored computer program is executed by a processor, which can perform the aforementioned steps S1 to S3.

[0142] It is understood that the aforementioned storage media may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Furthermore, the storage media may also be various media capable of storing program code, such as USB flash drives, external hard drives, magnetic disks, or optical discs.

[0143] It is understood that the processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0144] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, all technical solutions obtained through equivalent substitution or transformation fall within the protection scope of the present invention.

Claims

1. A knowledge graph-guided method for characterizing coal combustion processes, characterized in that, Includes the following steps: S1. Analyze, clean and standardize the multimodal process information in the coal-fired process scenario to form a standardized information set. Then extract the operating conditions, process rules, operating experience and constraints related to process analysis from the standardized information set to construct a set of process knowledge units. S2. Based on the requirements for characterizing coal-fired processes and the set of process knowledge units, construct the graph pattern of the coal-fired process knowledge graph, and then generate the node set and relation edge set of the coal-fired process knowledge graph. Combine these with the attribute set and constraint set to obtain the coal-fired process knowledge graph. S3. Using the current multimodal process information as the retrieval basis, perform relevant subgraph retrieval in the coal combustion process knowledge graph to determine the local knowledge subgraph; The multimodal process information at the current moment is then input into the knowledge-enhanced representation model along with the local knowledge subgraph for encoding and fusion to obtain a knowledge-enhanced state representation, thereby completing the coal-fired process representation.

2. The knowledge graph-guided coal combustion process characterization method as described in claim 1, characterized in that, The specific process of S1 is as follows: S11. Obtain multimodal process information in the coal-fired process scenario and form a multimodal process information set containing four types of modal process information: text-based process information, image-based process information, voice-based process information, and structured parameter-based process information. S12. The various modal process information is parsed, cleaned and standardized to form a standardized information set consisting of multiple standardized information items. Each standardized information item includes at least the set of process objects involved in the standardized information item, standardized information content and time identifier. S13. Perform cross-source semantic consistency identification on each standardized information item to determine the same process object, the same operating condition, or the same running event involved in different modal process information, and form a set of associated results composed of multiple associated results; S14. Based on the association result set, generate a process knowledge unit set consisting of five types of process knowledge units: process object units, operating condition units, operation event units, rule constraint units, and experience information units. This set is used to structure the expression of process objects, operating conditions, operation events, rule constraints, and experience information in the coal-fired process scenario. Each process knowledge unit... Represented as , Indicates the type of process knowledge unit. This represents the set of process objects involved in a process knowledge unit. This represents the content of the process knowledge unit. A time marker indicating a unit of process knowledge.

3. The knowledge graph-guided coal combustion process characterization method as described in claim 2, characterized in that, In S13, cross-source semantic consistency identification specifically means: for each normalized information item in the normalized information set, if different normalized information items contain the same process object, then the association type of these normalized information items is defined as the association of the same process object; When different standardized information items involve the same or related process objects and their time markers are within the same or adjacent time ranges, if the standardized information content of different standardized information items describes the same or similar operating conditions, then the association type of these standardized information items is defined as the association of the same operating condition. If the standardized information content of different standardized information items describes the same or similar operating events, then the association type of these standardized information items is defined as the association of the same operating event. Finally, the association result is composed of the associated standardized information items, the unified semantic content corresponding to the association of the standardized information items, and the association type. The unified semantic content includes process objects, operating conditions, and operating events.

4. The knowledge graph-guided coal combustion process characterization method as described in claim 3, characterized in that, In S14, if the association type of the association result is the same process object association, then the associated process object is regarded as the process object unit. If the association type is the same working condition association, then the unified semantic content of the association result is used as the content of the working condition unit, and the process object and time identifier involved in the working condition are determined to generate the working condition unit. If the association type is the same running event association, then the unified semantic content of the association result is used as the running event unit content, and the process object and time identifier involved in the running event are determined to generate the running event unit; Through semantic parsing, the process rules, constraints, and operational experience contained in the standardized information set are identified. The process rules and constraints are organized into rule constraint units according to the preset knowledge unit structure, and the operational experience is used to generate experience information units.

5. The knowledge graph-guided coal combustion process characterization method as described in claim 4, characterized in that, In S2, when constructing the graph pattern, firstly, based on the requirements for representing the coal combustion process, the scope of objects to be expressed by the coal combustion process knowledge graph and the set of mapping rules are determined; then, based on the content and purpose of various process knowledge units in S1, the set of node types, the set of relation types, the set of attribute types, and the set of constraint types are determined, thereby forming the graph pattern; wherein, the set of mapping rules includes four types of mapping rules, namely node generation rules, relation edge generation rules, attribute generation rules, and constraint generation rules.

6. The knowledge graph-guided coal combustion process characterization method as described in claim 5, characterized in that, The node generation rules are as follows: process object class nodes are generated from the process object names in the process object unit, operating condition class nodes are generated from the knowledge unit content in the operating condition unit, and operating event class nodes are generated from the knowledge unit content in the operating event unit.

7. The knowledge graph-guided coal combustion process characterization method as described in claim 5, characterized in that, The specific rule for generating relation edges is: if If it is a working condition unit, then search for the node set that matches the condition. For the corresponding process object class nodes, if two found process object class nodes are related, a relationship edge is generated between the two corresponding process object class nodes; if It is a working condition unit and has been... If a working condition status node is generated, then the node set is searched for a match. The corresponding process object class node is then used to generate a relationship edge between this process object class node and the operating condition class node; if For running event units and already by If a node representing a runtime event is generated, then the node set is searched for a match. The corresponding process object class node is generated, and a relationship edge is created between the process object class node and the running event class node; If the process knowledge unit content of the experience information unit If a relationship exists between a certain operating condition state and a running event, then a relationship edge is generated between the corresponding operating condition state class node and the running event class node.

8. The knowledge graph-guided coal combustion process characterization method as described in claim 5, characterized in that, The specific attribute generation rule is as follows: If If it is a working condition unit, then... As a state property, attached to The corresponding process object class node; if To run the event unit, then As an event property attached to The corresponding process object class node; when the process knowledge unit contains a time identifier. At that time, then As a time attribute, it is attached to the corresponding node or relation edge.

9. The knowledge graph-guided coal combustion process characterization method as described in claim 5, characterized in that, The constraint generation rule is specifically as follows: If When it is a rule-constrained unit, Match with process object type nodes, operating condition type nodes, and running event type nodes, and... As a constraint, it is attached to the corresponding matching node.

10. The knowledge graph-guided coal combustion process characterization method as described in claim 1, characterized in that, In S3, the newly added multimodal process information at the current time relative to the previous time is obtained as the input set. The set of nodes, relation edges, attributes, and constraints in the coal combustion process knowledge graph at the current time are used as graph elements. If there are graph elements in the coal combustion process knowledge graph at the current time that correspond to the input set, the corresponding graph elements are updated. If there are no corresponding graph elements, the corresponding graph elements are added according to the graph pattern and mapping rules to realize the incremental update of the coal combustion process knowledge graph. The updated coal combustion process knowledge graph is obtained and the local knowledge subgraph is redefined. Then, the multimodal process information at the current time and the redefined local knowledge subgraph are input into the knowledge enhancement representation model for encoding and fusion to obtain the updated knowledge enhancement state representation.

Images