A dynamic knowledge computing and reasoning system and method for digital twinning
By defining knowledge points as six-tuples, establishing a multimodal labeling system and process modeling, and combining reasoning with knowledge logic graphs, the problems of single reasoning mechanism and weak adaptability in digital twin systems are solved, enabling flexible knowledge retrieval and continuous optimization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NORTH CHINA UNIV OF WATER RESOURCES & ELECTRIC POWER
- Filing Date
- 2026-01-05
- Publication Date
- 2026-05-29
AI Technical Summary
Existing digital twin systems have shortcomings such as a single reasoning mechanism, overly tight coupling between knowledge and model, rigid knowledge representation, and weak adaptability, making it difficult to effectively express dynamic and process-oriented business logic and optimize based on real-time data.
It adopts a six-tuple to define knowledge points, establishes a multimodal knowledge tagging system, performs process modeling through an extended modeling language, combines knowledge logic graphs to perform reasoning modes, and utilizes optimization algorithms and machine learning models for online optimization, supporting multiple reasoning modes and real-time data feedback.
It enables flexible knowledge retrieval and association, supports multiple reasoning modes, adapts to different business scenarios, has adaptive capabilities, continuously optimizes knowledge calculation results, and improves decision-making.
Smart Images
Figure CN122114098A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of digital twin technology, specifically a dynamic knowledge computing and reasoning system and method for digital twins. Background Technology
[0002] The intelligence of a digital twin system highly depends on the reasoning and computational capabilities of its knowledge platform, but some limitations have been found in its use: The reasoning mechanism is singular: For example, the patent with announcement number CN118296435A describes a method for constructing a knowledge platform-assisted reasoning application based on water conservancy digital twins. This water conservancy digital twin knowledge platform relies heavily on specific mathematical models (such as water balance equations) and classifiers, and lacks support for diversified and configurable reasoning paths. Knowledge and model are too tightly coupled: For example, a knowledge service method for digital twin models of complex equipment described in patent CN114638160A, although it proposes the principle of loose coupling between knowledge and model, lacks a specific implementation mechanism for how to dynamically associate and organize knowledge. Rigid knowledge representation: Although traditional knowledge graphs (such as simulated knowledge graphs) can integrate multimodal data, they are difficult to effectively express dynamic and process-oriented business logic; Weak adaptability: Most systems are unable to dynamically optimize knowledge computation paths and recommendation results based on real-time data and feedback.
[0003] Based on this, a dynamic knowledge computing and reasoning system and method for digital twins is now provided, which can eliminate the drawbacks of existing technical solutions. Summary of the Invention
[0004] The purpose of this invention is to provide a dynamic knowledge computing and reasoning system and method for digital twins, in order to solve the problems in the background art.
[0005] To achieve the above objectives, the present invention provides the following technical solution: A dynamic knowledge computation and reasoning method for digital twins includes the following steps: Step S1: Define knowledge points as six-tuples and define the knowledge logic graph as a directed graph, where the nodes of the knowledge logic graph are knowledge points and the edges are multi-dimensional relationships between knowledge points. Step S2: Establish a multimodal knowledge tagging system, calculate tag similarity based on knowledge tags to associate knowledge point tags and improve the depth of semantic understanding; Step S3: Use an extended modeling language to perform process modeling and extend the definition of a knowledge computing-oriented meta-model. The meta-model is used to visualize business logic as an executable knowledge computing flowchart. Step S4: Based on knowledge tags and knowledge computation process, execute the reasoning mode on the knowledge logic graph; Step S5: Use optimization algorithms to filter and recommend the reasoning results, and use feedback data from the digital twin platform to optimize the knowledge computation process online.
[0006] Furthermore, the six-tuple in step S1 is represented as: ,in, Indicates knowledge points, As a unique identifier, For a set of attributes and This is used to describe the static characteristics of the knowledge point. A collection of knowledge tags and Used for retrieval and association. For a set of relations and This is used to define the connection between this knowledge point and other knowledge points. For confidence level and This indicates the reliability of the knowledge point. Metadata including source, version, and creation time.
[0007] Furthermore, the directed graph in step S1 is represented as follows: ,in, For knowledge logic diagrams, For a set of nodes and Each node corresponds to a knowledge point. , Let be the set of edges and This indicates that there is a relationship between knowledge points. For edge type mapping functions, This is a predefined set of relation types, including causal, temporal, spatial, logical dependency, and composition.
[0008] Furthermore, the formula for calculating tag similarity in step S2 is expressed as follows: ; in, For two tags and The similarity of tags, where tags are represented as: , For the label name, For tag type, As weight and , To calculate semantic similarity based on word vectors, To calculate path similarity in a labeling network, To calculate similarity based on contextual information about the co-occurrence of tags, , , The weighting coefficients are satisfied. .
[0009] Furthermore, in step S3, the knowledge computation process is defined as a function: ; in, For knowledge points, As a set of knowledge tags, As a knowledge computing process, For a set of input data, This is a set of results output after performing the calculation.
[0010] Furthermore, the knowledge computing process supports at least one of three reasoning modes: forward reasoning, backward reasoning, and analogical reasoning.
[0011] Furthermore, the forward reasoning specifically includes: given initial conditions, activating and propagating information in the knowledge logic graph along a specified path of the knowledge computation process, represented as finding a set of nodes that satisfy the conditions: ; The initial condition is the input label set. , Knowledge Logic Graph One of the nodes, Used to determine in the knowledge computing process Is it possible to reach the node from the input under the constraints? Used to determine the degree of matching between the node label and the input label; The reverse reasoning specifically includes: given a target state, the knowledge computation process is traversed in reverse to find all possible preconditions leading to the target, represented in the knowledge logic graph. and knowledge computing process The problem involves a path backtracking task that satisfies constraints within a search space. The target state consists of the target label set and the target node, and the preconditions are the target label set and the path. The analogical reasoning specifically includes: when faced with a new scenario, in the knowledge logic graph The system identifies known cases that have the highest structural analogy to the current problem scenario and recommends them after adaptively adjusting their solution processes. These known cases are knowledge logic graphs. The analogy score is obtained by combining the label similarity calculation formula with the subgraph matching algorithm. The analogy score calculation formula is as follows: ; in, Known knowledge points With target knowledge points The analogy value, For two tags and Tag similarity, For tag sets.
[0012] Furthermore, the knowledge computation process generates a candidate result set during the reasoning process, and uses an optimization algorithm to select the optimal subset from the candidate result set. The optimization algorithm is a multi-objective optimization function, expressed as: ; in, For the candidate result set The optimal subset in Indicates in the candidate result set All subsets Find the subset with the largest function value. For query context, Used to measure the relevance of results to the query. Used to measure the non-redundancy of results This is used to ensure that the result set covers different aspects. This indicates the cost incurred during the optimization process. This is the relevance weighting coefficient. This is the novelty weighting coefficient. For diversity weighting coefficients, These are cost weighting coefficients, all of which are non-negative and satisfy the following conditions: .
[0013] A dynamic knowledge computation and reasoning system for digital twins, applied to a dynamic knowledge computation and reasoning method for digital twins, includes: The knowledge representation module is used to construct and maintain a knowledge logic graph containing knowledge points and multiple relationships. The tag management module is used to manage the multimodal knowledge tag system and calculate tag similarity. The process modeling module uses a unified modeling language to define executable knowledge computation processes. The optimization and recommendation module is used to load models to perform multimodal inference, multi-objective optimization recommendation, and online optimization based on feedback data. The interface module is used for data interaction with the digital twin platform.
[0014] Furthermore, the optimization and recommendation module incorporates a machine learning model, employing a graph neural network to learn the low-dimensional vector representation of nodes, specifically as follows: ; in, To add self-connected adjacency matrices, For degree matrix, For the first The node feature matrix of the layer, For trainable weight matrix, This is the activation function.
[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention combines knowledge logic graphs with business process modeling to uniformly represent structured and procedural knowledge, solving the problem that traditional knowledge graphs cannot effectively handle procedural knowledge. It utilizes a multimodal knowledge tagging system to achieve flexible and semantically rich knowledge retrieval and association. Through configurable knowledge computation processes, it supports multiple reasoning modes such as forward, backward, and analogy to adapt to different business scenarios. It introduces online learning and optimization mechanisms so that the knowledge computation results can be continuously improved with the accumulation of digital twin data. It uses machine learning models to continuously optimize the effect of knowledge computation, making the system scalable. Attached Figure Description
[0016] Figure 1 This is a schematic diagram illustrating the steps of the dynamic knowledge calculation and reasoning method of the present invention.
[0017] Figure 2 This is a schematic diagram of the dynamic knowledge computing and reasoning system of the present invention.
[0018] Figure 3 This is a schematic diagram of the knowledge point structure of the flood control knowledge case of water conservancy projects in this invention.
[0019] Figure label annotations: Knowledge representation module 10, Tag management module 20, Process modeling module 30, Optimization and recommendation module 40, Interface module 50. Detailed Implementation
[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments.
[0021] Example 1 In this embodiment, as Figure 1 As shown, a dynamic knowledge computation and reasoning method for digital twins specifically includes the following steps: Step S1: Define knowledge points as six-tuples and the knowledge logic graph as a directed graph. The nodes of the knowledge logic graph are knowledge points, and the edges are the multi-dimensional relationships between knowledge points. The knowledge logic graph construction supports two methods: manual modeling and automatic extraction. Manual modeling involves dragging and dropping nodes and edges using a visualization tool and filling in attributes, labels, and relationships. Automatic extraction is based on business system logs, document libraries, and database table structures. It uses entity recognition and relationship extraction algorithms (such as BERT + relationship classification model) to generate an initial graph, which is then manually verified and stored in the database. Step S2: Establish a multimodal knowledge tagging system, calculate tag similarity based on knowledge tags to associate knowledge point tags and improve the depth of semantic understanding; Step S3: Use the Extended Modeling Language (UML) for process modeling, and extend the definition of the knowledge computing metamodel. The metamodel is used to visualize business logic as an executable knowledge computing flowchart. Metamodel elements include «KnowledgeFlow», «KnowledgeNode», «InferenceRule», «Condition», etc. The above metamodel elements extend the Unified Modeling Language (UML) to make it possible to visualize business logic as a computable flowchart composed of start nodes, end nodes, task nodes, decision nodes, merge nodes, and control flow edges. «KnowledgeFlow» represents a complete knowledge computing task flow. «InferenceNode» represents a reasoning step and can be bound to a specific reasoning algorithm (such as rule reasoning, graph query). «DecisionNode» can make branch judgments based on intermediate results. «DataFlow» represents the transfer of data between nodes. The data can be the original input, intermediate knowledge nodes, or the final result. Step S4: Based on knowledge tags and knowledge computation process, execute the reasoning mode on the knowledge logic graph; Step S5: Optimize the reasoning results using an optimization algorithm, and optimize the knowledge computation process online using feedback data from the digital twin platform. The feedback data from the digital twin platform includes, but is not limited to: sensor measured data such as temperature, pressure, flow rate, and displacement; simulation result data such as simulation prediction curves, convergence status, and error indicators; user operation logs such as decision adoption records, manual correction instructions, and evaluation feedback; and system execution indicators such as response time, resource utilization, and fault alarm information.
[0022] In this embodiment, the six-tuple in step S1 is represented as: ,in, Indicates knowledge points, As a unique identifier, For a set of attributes and This is used to describe the static characteristics of the knowledge point. A collection of knowledge tags and Used for retrieval and association. For a set of relations and This is used to define the connection between this knowledge point and other knowledge points. For confidence level and This indicates the reliability of the knowledge point. Metadata including source, version, and creation time; In this embodiment, the directed graph in step S1 is represented as: ,in, For knowledge logic diagrams, For a set of nodes and Each node corresponds to a knowledge point. , Let be the set of edges and This indicates that there is a relationship between knowledge points. For edge type mapping functions, It is a predefined set of relation types, including causal, temporal, spatial, logical dependency, and composition.
[0023] In this embodiment, the formula for calculating tag similarity in step S2 is expressed as follows: ; in, For two tags and The similarity of tags, where tags are represented as: , For the label name, For tag types, an example is: , As weight and , To calculate semantic similarity based on word vectors, To calculate path similarity in a labeling network, To calculate similarity based on contextual information about the co-occurrence of tags, , , The weighting coefficients are satisfied. It can be adjusted according to specific application scenarios.
[0024] In this embodiment, in step S3, the knowledge computation process is defined as a function: ; in, For knowledge points, As a set of knowledge tags, As a knowledge computing process, For a set of input data, This is a set of results output after performing the calculation.
[0025] In this embodiment, the knowledge computing process supports at least one of three reasoning modes: forward reasoning, backward reasoning, and analogical reasoning.
[0026] Forward reasoning specifically includes: given initial conditions, activating and propagating information in the knowledge logic graph along a prescribed path of knowledge computation, represented as finding a set of nodes that satisfy the conditions. ; The initial condition is the input label set. , Knowledge Logic Graph One of the nodes, Used to determine in the knowledge computing process Is it possible to reach the node from the input under the constraints? Used to determine the degree of matching between the node label and the input label; Reverse reasoning specifically includes: given a target state, the knowledge computation process is traversed in reverse to find all possible preconditions leading to the target, represented in the knowledge logic graph. and knowledge computing process The problem involves a path backtracking task that satisfies constraints within a search space. The target state consists of the target label set and the target node, and the preconditions are the target label set and the path. Analogical reasoning specifically includes: when faced with a new scenario, in a knowledge logic diagram The system identifies known cases that have the highest structural analogy to the current problem scenario (described as a set of feature labels), and recommends solutions based on adaptively adjusted flow patterns. These known cases are knowledge logic graphs. In the subgraph, known cases are stored in the knowledge logic graph in the form of a subgraph structure. In the process, each subgraph is labeled as a case node and associated with a case tag set and a solution process identifier. During retrieval, a subgraph matching algorithm searches for subgraphs in the knowledge logic graph with a structural similarity higher than a preset threshold, and then sorts and returns them based on tag similarity. The preset threshold can be adjusted according to actual needs. The analogy score is obtained by combining the tag similarity calculation formula with the subgraph matching algorithm. The analogy score calculation formula is as follows: ; in, Known knowledge points With target knowledge points The analogy value, For two tags and Tag similarity, For tag sets.
[0027] In this embodiment, the knowledge computation process generates a candidate result set during the reasoning process, and an optimization algorithm is used to select the optimal subset from the candidate result set. The optimization algorithm is a multi-objective optimization function, expressed as: ; in, For the candidate result set The optimal subset in Indicates in the candidate result set All subsets Find the subset with the largest function value. For query context, Used to measure the relevance of results to the query. Used to measure the non-redundancy of results This is used to ensure that the result set covers different aspects. This indicates the cost incurred during the optimization process. This is the correlation weighting coefficient, with a reference range of 0.4 to 0.7. This is the highest priority, ensuring that the output is highly relevant to the problem. For example, in flood control scheduling, the drainage efficiency of the plan must be guaranteed first. This is the novelty weighting coefficient, with a reference range of 0.1 to 0.3. This is a medium priority approach, encouraging the system to explore uncommon but potentially effective solutions from historical data to avoid getting stuck in local optima. It is valuable for dealing with new situations (such as extreme weather). This is the diversity weighting coefficient, with a reference range of 0.1 to 0.3. Prioritize this approach to ensure that recommendations cover different aspects (such as different regions and different types of measures), providing users with a comparable and selectable set of options. This is the cost weighting coefficient, with a reference range of 0.1 to 0.4. Variable priority represents a penalty for resource consumption, risk, or negative impact. In resource-sensitive (e.g., power shortages) or risk-averse (e.g., ensuring safety) scenarios, the initial weight is set higher, and all weight coefficients are non-negative and satisfy the following conditions: They are usually normalized to facilitate comparison and interpretation across different tasks.
[0028] Example 2 The difference from Example 1 is that, as Figure 2 As shown, the present invention also provides a dynamic knowledge computation and reasoning system for digital twins, applied to a dynamic knowledge computation and reasoning method for digital twins, comprising: Knowledge representation module 10 is used to construct and maintain a knowledge logic graph containing knowledge points and multiple relationships; The tag management module 20 is used to manage the multimodal knowledge tag system and calculate tag similarity. Process modeling module 30 uses a unified modeling language to define executable knowledge computing processes; The optimization and recommendation module 40 is used to load the model to perform multi-modal reasoning, multi-objective optimization recommendation, and online optimization based on feedback data. The online optimization process adopts a strategy optimization mechanism based on reinforcement learning. Based on the execution effect data fed back by the digital twin platform, the node weights and reasoning path priorities in the knowledge computing process are dynamically adjusted. The update frequency supports real-time updates, batch updates, or timed updates, which can be configured according to business scenarios. Interface module 50 is used to interact with the digital twin platform. The system establishes a continuously running feedback learning loop. When the recommendation scheme is implemented in the digital twin environment, its real effect data (such as goal achievement and efficiency indicators) is collected as supervision signals. Using these signals, the system can update the confidence and relation weight of knowledge points, adjust the weight in the label similarity calculation, fine-tune the parameters of the machine learning model, and even trigger the addition, deletion and modification of the knowledge logic graph structure (knowledge update).
[0029] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the above-described dynamic knowledge calculation and reasoning method; In this embodiment, the optimization and recommendation module has a built-in machine learning model (graph neural network model), which uses graph neural networks (GNNs) to learn the low-dimensional vector representation (embedding) of nodes (knowledge points), specifically as follows: ; in, To add self-connected adjacency matrices, Its degree matrix, For the first The node feature matrix of the layer, For trainable weight matrix, As an activation function, this operation can delve deeper into the complex structural information contained in the knowledge logic graph, and the learned node embeddings can be used to calculate more accurate node similarity (auxiliary). It can be used as a function, or directly as a feature input to the ranking stage of a multi-objective optimization recommendation model (which implements ranking logic through a multi-objective optimization function, receives node embedding features generated by a graph neural network as input, and performs result ranking and recommendation), or in analogical reasoning, it can be used to calculate the structural similarity between subgraphs, replacing or enhancing label-based similarity calculation.
[0030] Specifically, based on the flood control knowledge logic of this water conservancy project, the operational process of constructing a knowledge logic graph in this dynamic knowledge computation and reasoning method for digital twins is demonstrated: Scenario: City XX needs to build a flood control decision support system based on the knowledge connection between rainfall, hydrology, and flood control. The flood control knowledge of water conservancy projects can be directly regarded as the connection between the two knowledge points of rainfall and flood control.
[0031] Step 1: Obtain the rainfall information for XX City in the next 24 hours through the meteorological department's rainfall forecast. Break down the rainfall information into two key points: rainfall intensity and rainfall range. The rainfall intensity is defined as "the cumulative rainfall in 12 hours will reach 80mm (heavy rainstorm level)", and the rainfall range is defined as "covering the main urban area of XX City and the upstream river basin (approximately 500km²)". Step 2: Combine rainfall information with hydrological stations and geographical conditions to trigger the linkage of three types of knowledge points: Input basic conditions such as geographical region and underlying surface. The geographical region is "the river in XX city flows through the main urban area, with mountains upstream and plains downstream". The underlying surface is "the vegetation coverage rate of the upstream mountains is 60%, and the proportion of hardened road surface in the downstream urban area is 85% (affecting the runoff velocity)". Hydrological process calculations for runoff generation and confluence: runoff generation is "the upstream mountainous area has a runoff velocity of approximately 20 mm / h due to vegetation interception, and the downstream urban area has a runoff velocity of approximately 50 mm / h", and confluence is "10 hours after rainfall, the upstream water will flow into the main urban river section"; Hydrological station monitoring: "Real-time collection of water level and flow data from hydrological stations in the middle reaches of rivers"; Step 3: Based on hydrological data, trigger the execution of two knowledge points related to water conservancy and flood control: Water level and flow rate process: "The hydrological station has monitored that the current water level has reached 90% of the warning level, and it is expected to exceed the warning level in 2 hours." Warning rule: "Triggering a Level II flood warning, corresponding to rainfall of 80mm and water level exceeding the warning level." According to the warning rule, the dispatch and emergency rescue linkage is triggered: "Open the upstream reservoir floodgate and control the downstream flow to not exceed 500m³ / s" and "Organize personnel to reinforce the weak sections of the downstream river embankment with sandbags."
[0032] The above process logically connects the knowledge of "rainfall forecasting (broken down into rainfall intensity / rainfall range) → hydrological stations (combined with geography / runoff / confluence) → flood control (water level and flow processes / early warning rules)," achieving a structured implementation from "meteorological data" to "flood control actions," such as... Figure 3 As shown, each box corresponds to a specific task node in actual execution.
[0033] In summary, this invention discloses a dynamic knowledge computation and reasoning method for digital twins. This method constructs a hybrid knowledge representation structure that integrates knowledge logic graphs and business process modeling, utilizes a multimodal knowledge tagging system to achieve flexible knowledge association, and supports multiple reasoning modes through a configurable knowledge computation process. The system employs a pluggable engine to execute reasoning tasks and continuously optimizes computation results through online learning based on feedback data from the digital twin platform. This effectively addresses the shortcomings of existing digital twin knowledge platforms in expressing complex knowledge, handling business processes, and adaptive optimization, thereby improving decision-making capabilities.
[0034] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A dynamic knowledge computation and reasoning method for digital twins, characterized in that, Specifically, the following steps are included: Step S1: Define knowledge points as six-tuples and define the knowledge logic graph as a directed graph, where the nodes of the knowledge logic graph are knowledge points and the edges are multi-dimensional relationships between knowledge points. Step S2: Establish a multimodal knowledge tagging system, calculate tag similarity based on knowledge tags to associate knowledge point tags and improve the depth of semantic understanding; Step S3: Use an extended modeling language to perform process modeling and extend the definition of a knowledge computing-oriented meta-model. The meta-model is used to visualize business logic as an executable knowledge computing flowchart. Step S4: Based on knowledge tags and knowledge computation process, execute the reasoning mode on the knowledge logic graph; Step S5: Use optimization algorithms to filter and recommend the reasoning results, and use feedback data from the digital twin platform to optimize the knowledge computation process online.
2. The dynamic knowledge computation and reasoning method for digital twins according to claim 1, characterized in that, The six-tuple in step S1 is represented as follows: ,in, Indicates knowledge points, As a unique identifier, For a set of attributes and This is used to describe the static characteristics of the knowledge point. A collection of knowledge tags and Used for retrieval and association. For a set of relations and This is used to define the connection between this knowledge point and other knowledge points. For confidence level and This indicates the reliability of the knowledge point. Metadata including source, version, and creation time.
3. The dynamic knowledge computation and reasoning method for digital twins according to claim 2, characterized in that, The directed graph in step S1 is represented as follows: ,in, For knowledge logic diagrams, For a set of nodes and Each node corresponds to a knowledge point. , Let be the set of edges and This indicates that there is a relationship between knowledge points. For edge type mapping functions, This is a predefined set of relation types, including causal, temporal, spatial, logical dependency, and composition.
4. The dynamic knowledge computation and reasoning method for digital twins according to claim 1, characterized in that, The formula for calculating tag similarity in step S2 is expressed as follows: ； in, For two tags and The similarity of tags, where tags are represented as: , For the label name, For tag type, As weight and , To calculate semantic similarity based on word vectors, To calculate path similarity in a labeling network, To calculate similarity based on contextual information about the co-occurrence of tags, , , The weighting coefficients are satisfied. .
5. The dynamic knowledge computation and reasoning method for digital twins according to claim 1, characterized in that, In step S3, the knowledge computation process is defined as a function: ； in, For knowledge points, As a set of knowledge tags, As a knowledge computing process, For a set of input data, This is a set of results output after performing the calculation.
6. A dynamic knowledge computing and reasoning system and method for digital twins according to claim 4, characterized in that, The knowledge computing process supports at least one of three reasoning modes: forward reasoning, backward reasoning, and analogical reasoning.
7. A dynamic knowledge computation and reasoning method for digital twins according to claim 6, characterized in that, The forward reasoning specifically includes: given initial conditions, activating and propagating information in the knowledge logic graph along a defined path of the knowledge computation process, represented as finding a set of nodes that satisfy the conditions: ； The initial condition is the input label set. , Knowledge Logic Graph A node on, Used to determine in the knowledge computing process Is it possible to reach the node from the input under the constraints? Used to determine the degree of matching between the node label and the input label; The reverse reasoning specifically includes: given a target state, the knowledge computation process is traversed in reverse to find all possible preconditions leading to the target, represented in the knowledge logic graph. and knowledge computing process The problem involves a path backtracking task that satisfies constraints within a search space. The target state consists of the target label set and the target node, and the preconditions are the target label set and the path. The analogical reasoning specifically includes: when faced with a new scenario, in the knowledge logic graph The system identifies known cases that have the highest structural analogy to the current problem scenario and recommends solutions based on adaptive adjustments to their process flow. These known cases are knowledge logic graphs. The analogy score is obtained by combining the label similarity calculation formula with the subgraph matching algorithm. The analogy score calculation formula is as follows: ； in, Known knowledge points With target knowledge points The analogy value, For two tags and Tag similarity, For tag sets.
8. The dynamic knowledge computation and reasoning method for digital twins according to claim 1, characterized in that, The knowledge computation process generates a candidate result set during reasoning, and uses an optimization algorithm to select the optimal subset from the candidate result set. The optimization algorithm is a multi-objective optimization function, expressed as: ； in, For the candidate result set The optimal subset in Indicates in the candidate result set All subsets Find the subset with the largest function value. For query context, Used to measure the relevance of results to the query. Used to measure the non-redundancy of results This is used to ensure that the result set covers different aspects. This indicates the cost incurred during the optimization process. This is the relevance weighting coefficient. This is the novelty weighting coefficient. For diversity weighting coefficients, These are cost weighting coefficients, all of which are non-negative and satisfy the following conditions: .
9. A dynamic knowledge computation and reasoning system for digital twins, applied to the dynamic knowledge computation and reasoning method for digital twins as described in any one of claims 1-8, characterized in that, include: The knowledge representation module is used to construct and maintain a knowledge logic graph containing knowledge points and multiple relationships. The tag management module is used to manage the multimodal knowledge tag system and calculate tag similarity. The process modeling module uses a unified modeling language to define executable knowledge computation processes. The optimization and recommendation module is used to load models to perform multimodal inference, multi-objective optimization recommendation, and online optimization based on feedback data. The interface module is used for data interaction with the digital twin platform.
10. A dynamic knowledge computing and reasoning system for digital twins according to claim 9, characterized in that, The optimization and recommendation module incorporates a machine learning model, employing a graph neural network to learn the low-dimensional vector representation of nodes, specifically as follows: ； in, To add self-connected adjacency matrices, For degree matrix, For the first The node feature matrix of the layer, For trainable weight matrix, This is the activation function.