Semantic association modeling and adaptive simulation method based on multi-layer correction

Abstract

The present invention relates to the field of virtual simulation and human-computer interaction technology, and in particular to a semantic association modeling and adaptive simulation method based on multi-layer correction, including: collecting feature data from a three-dimensional simulation model and an electrical schematic diagram, generating situational dynamic model data through data cleaning and standardization processing; constructing a semantic association matrix based on the situational dynamic model data, generating geometric layer, logic layer and cross-domain layer models, and generating dynamic state and logic state data through distributed cellular automaton simulation; combining simulation data, optimizing the semantic association matrix and dynamic modeling rules using a causal path inference algorithm and a multi-layer correction mechanism, and realizing state linkage and user interaction feedback between a three-dimensional simulation model and an electrical schematic diagram. The present invention significantly improves the state synchronization accuracy and interaction efficiency of complex systems, and is suitable for the fields of industrial design, simulation verification, and teaching and training.

Description

Technical Field

[0001] The present invention relates to the technical field of virtual simulation and human-computer interaction, and in particular to a semantic association modeling and adaptive simulation method based on multi-layer correction. Background Art

[0002] Semantic association modeling and adaptive simulation technology aims to achieve real-time linkage between three-dimensional simulation models and electrical schematics by integrating multi-dimensional simulation data and logical information. This technology is of great significance to the complex interactive design, simulation verification and training of industrial systems. Its core is to improve the accuracy and dynamic adaptability of interaction by building a semantic association matrix and multi-layer feedback correction based on simulation data, thereby achieving global optimization and efficient operation of the system.

[0003] The existing technology uses a model and graphics synchronization technology based on a single mapping relationship, and realizes the linkage between the 3D model and the electrical schematic diagram through a static ID mapping method. However, this technical method has significant defects:

[0004] Static mapping relationships are difficult to dynamically adapt to the multi-state changes of components, which leads to inconsistent state synchronization in complex scenarios; existing methods only focus on the simple correspondence between geometry and logic, without considering the association of semantic levels, and it is difficult to support deep interaction of complex systems; when abnormal states or erroneous operations occur, there is a lack of effective feedback mechanism to adjust models and graphics, which easily leads to system response delays or errors; the system fails to effectively combine user behavior to generate real-time feedback and guidance information, which reduces the operability of the system and the teaching application effect. Summary of the invention

[0005] In view of the many problems existing in the above-mentioned prior art, the present invention provides a semantic association modeling and adaptive simulation method based on multi-layer correction. The present invention deeply links the dynamic state and logical state of the three-dimensional simulation model with the electrical schematic diagram by constructing a semantic association matrix of multi-layer correction, and realizes the adaptive operation of the complex system through dynamic modeling rules and distributed cellular automaton simulation. Combined with causal path inference and multi-layer correction mechanism, the system can quickly respond to and optimize abnormal states, and finally achieve efficient and accurate state synchronization and user interactive feedback.

[0006] A semantic association modeling and adaptive simulation method based on multi-layer correction includes the following steps:

[0007] Collect geometric feature data, kinematic feature data and component status data from the three-dimensional simulation model, collect primitive structure data and circuit logic data from the electrical schematic diagram, pre-process the collected data, and generate context dynamic model data, wherein the context dynamic model data includes semantic node data, context node attribute data and context association data;

[0008] Based on the situational dynamic model data, a semantic association matrix is ​​generated through an iterative semantic aggregation algorithm, a geometric layer model, a logical layer model and a cross-domain layer model are constructed using the semantic association matrix, and the geometric layer model, the logical layer model and the cross-domain layer model are dynamically simulated in combination with a distributed cellular automation to generate simulation data, wherein the simulation data includes dynamic state data of the three-dimensional simulation model and logical state data of the electrical schematic diagram;

[0009] Based on simulation data, a causal path inference algorithm is used to perform feedback optimization on the semantic association matrix and dynamic modeling rules to generate optimized semantic association matrix and dynamic modeling rules. The optimized semantic association matrix and dynamic modeling rules are adjusted through a real-time correction mechanism, a trend correction mechanism and a global correction mechanism to generate synchronization status data. The synchronization status data is used to synchronize the status display of the three-dimensional simulation model and the electrical schematic diagram as well as the dynamic feedback of the user interaction interface.

[0010] Preferably, the data preprocessing comprises the following steps:

[0011] The collected geometric feature data, kinematic feature data and component status data of the three-dimensional simulation model, as well as the graphic element structure data and circuit logic data of the electrical schematic diagram are de-redundant, and the de-redundant processing includes identifying and deleting duplicate data through similarity calculation; aligning the time series data, and the time series alignment processing is completed based on a dynamic time warping algorithm; and standardizing the processed data to generate structured semantic node data in a unified format.

[0012] Preferably, the semantic node data generates the situational dynamic model data through the following steps:

[0013] The structured semantic node data is semantically annotated by combining a rule-based parsing method with a natural language processing algorithm; the geometric attributes, logical attributes and state attributes of each semantic node are defined; context node attribute data is generated based on the geometric adjacency relationship, logical dependency relationship and state transfer relationship of the semantic nodes, and context-related data is constructed through multi-layer relationship analysis.

[0014] Preferably, the semantic association matrix is ​​generated by an iterative semantic aggregation algorithm, and the iterative semantic aggregation algorithm comprises the following steps:

[0015] Based on semantic nodes and The calculation formula of the association weight is as follows:

[0016] , where the association weight Represents a semantic node and The final weight of the semantic association rule set Represents a set of rules used for association calculation; weights of semantic association rules Representation Rules Importance in calculations; rule scoring Representation Rules Semantic Node and The strength of association;

[0017] The dynamic adjustment formula of the association weight is as follows:

[0018] , where the association weight after iteration is Represents a semantic node and In the The association weight after iterations; adjustment coefficient Represents the smoothing coefficient of the dynamic adjustment of the weight; The association weight of the iteration Represents the weight of the semantic node in the previous iteration; attention score Represents a semantic node calculated based on a multi-head attention mechanism and of relevance.

[0019] Preferably, the geometric layer model, logical layer model and cross-domain layer model supported by the semantic association matrix include:

[0020] The geometric layer model describes the geometric position relationship, motion constraints and physical coupling of components in the three-dimensional simulation model; the logical layer model describes the connectivity, signal flow and control logic of the graphics elements in the electrical schematic diagram; the cross-domain layer model describes the control relationship between the three-dimensional simulation model and the electrical schematic diagram components.

[0021] Preferably, the dynamic modeling rules include state transition rules of distributed cellular automata, and the state transition rules are generated by the following optimization process:

[0022] The reward function calculation formula is as follows:

[0023] , where the reward value Represents the state reward of the optimization process; state consistency score Used to measure the degree of match between the cell simulation state and the actual state; simulation efficiency score Used to evaluate the efficiency of simulation calculations; weight coefficient Indicates the importance of the state consistency score;

[0024] The cell state update rules are as follows:

[0025] , where the cell state Representation Cell In the The state at the iteration; cell state Representation Cell In the The state at the iteration; the neighborhood state set Representation and Cell The state set of other directly related cells; state transition function Represents a function that calculates a new state by inputting the current cell state, a set of neighborhood states, and a reward value.

[0026] Preferably, the process of optimizing the state transition rule of the distributed cellular automaton is performed based on a reinforcement learning method, and the optimization process includes the following steps:

[0027] Constructing a reward function, wherein the optimization objectives of the reward function include minimizing simulation error and maximizing simulation efficiency;

[0028] Constructing training samples based on simulation historical data, wherein the training samples are used for training a reinforcement learning model;

[0029] During the reinforcement learning training process, the weight distribution of the cell neighborhood state is dynamically adjusted to optimize the state transition parameters of the neighborhood rule;

[0030] The state transition rules optimized by reinforcement learning are applied to dynamic simulation to update the cell state.

[0031] Preferably, the causal path inference algorithm comprises the following steps:

[0032] Construct a causal path diagram based on simulation data to analyze the causal relationship between component state changes and logic changes during the simulation process; identify abnormal state nodes and their triggering conditions in the causal path diagram, and locate the cause of the abnormality through path backtracking; adjust the semantic association matrix and dynamic modeling rules to eliminate abnormal states and optimize the simulation path.

[0033] Preferably, the real-time correction mechanism is used to correct slight deviations between simulation data and actual status; the trend correction mechanism adjusts the semantic association matrix and dynamic modeling rules in advance based on the time series prediction model; and the global correction mechanism optimizes the global structure of the semantic association matrix through consistency constraints.

[0034] Preferably, the synchronization status data is generated in the following manner:

[0035] Combined with the dynamic state and logical state in the simulation data, the display state of the three-dimensional simulation model and the electrical schematic diagram is updated in real time; the user operation behavior is recorded through the user interaction interface, and dynamic feedback information is generated based on the intention prediction model; the dynamic feedback information includes real-time highlighting of related components and providing operation suggestions.

[0036] Compared with the prior art, the advantages and beneficial effects of the present invention are:

[0037] The present invention is based on semantic association matrix and dynamic modeling rules, supports the construction and real-time update of geometric layer, logical layer and cross-domain layer models, and overcomes the limitations of static mapping;

[0038] The present invention adopts real-time correction, trend correction and global correction mechanisms, which can quickly recover and optimize the simulation path when state deviation occurs, thereby improving the robustness of the system;

[0039] The present invention constructs a causal path diagram based on simulation data, locates abnormal states and adjusts dynamic modeling rules to achieve deep optimization of simulation logic;

[0040] The present invention combines user operation records and intention prediction models to provide real-time feedback information and operation guidance, which significantly improves the user-friendliness of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a schematic diagram of the process of the present invention;

[0042] Figure 2 A schematic diagram of constructing a semantic association matrix in the present invention;

[0043] Figure 3 It is a flow chart of the multi-layer correction mechanism in the present invention;

[0044] Figure 4 This is a flowchart for optimizing the causal path inference in the present invention. DETAILED DESCRIPTION

[0045] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure.

[0046] like Figure 1 As shown, a semantic association modeling and adaptive simulation method based on multi-layer correction includes the following steps:

[0047] Collect geometric feature data, kinematic feature data and component status data from the three-dimensional simulation model, collect primitive structure data and circuit logic data from the electrical schematic diagram, pre-process the collected data, and generate context dynamic model data, wherein the context dynamic model data includes semantic node data, context node attribute data and context association data;

[0048] In the present invention, data collection and situational dynamic model generation are the basic steps to achieve multi-layer correction semantic association modeling and adaptive simulation. Its core goal is to transform the heterogeneous data in the three-dimensional simulation model and the electrical schematic diagram into situational dynamic model data that supports semantic association analysis and dynamic simulation through unified formatting and semantic annotation, thereby realizing the deep integration of multimodal information. Specifically, the geometric feature data, kinematic feature data and component state data collected from the three-dimensional simulation model represent the structural information, motion behavior information and real-time physical state information of the simulation model, respectively, wherein the geometric feature data mainly describes the topological structure, size and relative position relationship of the components of the model, the kinematic feature data includes the speed, acceleration and trajectory of the components, and the component state data records the physical properties (such as temperature, pressure) and logical properties (such as enabled state, execution state). The primitive structure data and circuit logic data collected from the electrical schematic diagram provide a functional description of the electrical components. The former defines the physical connection relationship and topological distribution of the electrical components, and the latter includes signal flow, switch logic and control dependency. By collecting the above data, it is possible to achieve comprehensive coverage of the information in the three-dimensional simulation model and the electrical schematic diagram, and provide a data basis for subsequent association analysis.

[0049] After completing the data collection, the collected data is preprocessed, including de-redundancy, time series alignment and standardization. First, de-redundancy processing removes duplicate or irrelevant data by calculating the similarity between data. After de-redundancy, data items with similarity below the threshold are retained. Time series alignment is based on the dynamic time warping algorithm (Dynamic Time Warping, DTW), which is used to synchronize data collected at different times. After time series alignment processing, it is ensured that all collected data have a consistent time reference. Then the data is standardized, and data of different units and dimensions are converted into a unified range and format for subsequent analysis.

[0050] After preprocessing, the situational dynamic model data is generated through semantic annotation technology. The situational dynamic model data includes semantic node data, situational node attribute data and situational association data. The semantic node data is generated by giving semantic tags to the preprocessed data. The attribute data of each semantic node consists of its geometric attributes (such as size, position), logical attributes (such as signal direction) and state attributes (such as real-time temperature). The situational association data is generated by analyzing the geometric relationship (such as contact relationship), logical dependency relationship (such as control and controlled relationship) and state transfer relationship (such as data flow) between nodes. Taking the association between components in the three-dimensional simulation model and components in the electrical schematic as an example, by defining the motion trajectory of the three-dimensional component and the transmission path of the control signal of the electrical component, the association relationship between the two can be generated and stored in the situational association data, thus laying the foundation for subsequent semantic association modeling. Through the above steps, the situational dynamic model data describes the multi-level and multi-modal association information between the three-dimensional simulation model and the electrical schematic in a standardized form, which can efficiently support the subsequent semantic aggregation, simulation analysis and feedback optimization process.

[0051] The present invention provides a solid foundation for the deep fusion and association modeling of multimodal information by integrating heterogeneous data and generating situational dynamic model data. Specifically, the generation of situational dynamic models significantly reduces the complexity of multimodal information in the analysis process, unifies the semantic expression of geometric, logical and state data, and provides accurate and comprehensive basic data for association matrix generation and dynamic simulation optimization in subsequent steps. In addition, by introducing standardized processing and semantic annotation technology, the present invention effectively improves the accuracy and adaptability of heterogeneous data fusion, ensures the efficiency and accuracy in the association modeling and simulation process, and thus provides an innovative solution to the problem of multimodal interaction complexity.

[0052] Preferably, the data preprocessing comprises the following steps:

[0053] The collected geometric feature data, kinematic feature data and component status data of the three-dimensional simulation model, as well as the graphic element structure data and circuit logic data of the electrical schematic diagram are de-redundant, and the de-redundant processing includes identifying and deleting duplicate data through similarity calculation; aligning the time series data, and the time series alignment processing is completed based on a dynamic time warping algorithm; and standardizing the processed data to generate structured semantic node data in a unified format.

[0054] Data preprocessing is a key step to ensure that heterogeneous data can achieve semantic association modeling and dynamic simulation in the subsequent stage. Specifically, this process includes redundancy removal, time series alignment, and standardization, and ultimately generates structured semantic node data in a unified format, laying the foundation for the construction of situational dynamic model data.

[0055] First, after data collection, redundancy removal is performed. Its goal is to improve data quality and reduce the complexity of subsequent modeling by identifying and removing duplicate or redundant data. In redundancy removal, similarity calculation is used to evaluate the similarity between data items and determine whether there is duplication in the data. The specific calculation formula is as follows:

[0056] ,in, Represents a data item and The similarity of and Data items and In the The weight on the dimension, Indicates the dimension of the data. If the similarity is greater than the preset threshold , then determine and If there are duplicate data, one of them will be removed. For example, in the geometric feature data of a 3D simulation model, for the vertex coordinates of two components, if their spatial Euclidean distance is less than a threshold, they can be considered as duplicates to be deleted.

[0057] Secondly, the alignment of time series data is the core step in preprocessing to ensure that dynamic feature data has temporal consistency. In the present invention, the dynamic time warping algorithm (Dynamic Time Warping, DTW) is used to achieve sequence alignment. DTW can process data with different sampling rates or time offsets, and achieve alignment by minimizing the cumulative matching distance. Its optimization formula is as follows:

[0058] ,in, Representation sequence and The alignment cost is and Respectively represent the sequence and The and data points, It is a distance function between data points (such as Euclidean distance). Through this algorithm, the kinematic feature data of the 3D simulation model is aligned with the time-dependent signal data in the electrical schematic diagram, so that the two are synchronized in the time dimension. For example, there is a timing misalignment between the motion trajectory of a component in the simulation model and the electrical control signal. After DTW alignment, it can ensure that these data accurately reflect their dynamic association in subsequent analysis.

[0059] Finally, the data is standardized to convert data of different units and dimensions into a unified numerical range for subsequent semantic association calculations. The standardization formula is as follows:

[0060] ,in, represents the normalized value, Represents the original data value, represents the mean of the data, Represents the standard deviation of the data. Through standardization, the geometric feature data (such as length and area) of the 3D simulation model and the logical state data (such as signal level) of the electrical schematic diagram are unified into dimensionless numerical form, thereby avoiding the influence of different data dimensions on the semantic calculation results.

[0061] Through the above steps, structured semantic node data in a unified format are generated. These data store the core features of each component and its dynamic properties in the form of nodes, and provide a clear basic data structure for subsequent semantic annotation and association modeling. For example, for a certain 3D simulation component, its structured semantic node data may include geometric properties (such as position and size), logical properties (such as control dependencies), and state properties (such as real-time temperature). These node data can be directly mapped to the situational dynamic model to support the subsequent construction of the semantic association matrix and multi-layer dynamic modeling.

[0062] The present invention significantly improves data quality and reduces redundant computing overhead through efficient redundancy removal processing; time series alignment ensures the temporal consistency between dynamic data, so that the dynamic association between the three-dimensional simulation model and the electrical schematic diagram can be accurately modeled; standardization processing realizes the unified dimensional expression of data in the present invention, providing a reliable data foundation for semantic association modeling. This preprocessing method provides an efficient and accurate implementation path for the fusion and analysis of heterogeneous data, and is a key step to ensure the success of semantic association modeling and adaptive simulation.

[0063] Preferably, the semantic node data generates the situational dynamic model data through the following steps:

[0064] The structured semantic node data is semantically annotated by combining a rule-based parsing method with a natural language processing algorithm; the geometric attributes, logical attributes and state attributes of each semantic node are defined; context node attribute data is generated based on the geometric adjacency relationship, logical dependency relationship and state transfer relationship of the semantic nodes, and context-related data is constructed through multi-layer relationship analysis.

[0065] In the present invention, if Figure 2As shown in the figure, the generation of semantic node data is a key step in building situational dynamic model data. Its main goal is to transform the original structured semantic node data into a multidimensional semantic model that can reflect geometric, logical and state associations through semantic annotation and relationship modeling, thereby providing support for the subsequent generation of semantic association matrices and dynamic simulation. Specifically, this process is divided into three stages: semantic annotation, node attribute definition, and association data construction.

[0066] First, semantic annotation of structured semantic node data is the process of giving clear semantic meaning to each data item, so as to realize semantic-based association analysis in subsequent modeling. The present invention uses a rule-based parsing method combined with a natural language processing algorithm to complete the annotation. Among them, the rule-based parsing method parses the semantic nodes through a predefined rule base, for example, the rule base defines "electrical component type rules" (such as "resistance", "capacitor") and "three-dimensional component type rules" (such as "bracket", "connector"), which are used to quickly classify data items; the natural language processing algorithm is used to process unstructured data or ambiguous semantic mapping, such as matching the description of the data item with the semantics in the rule base through semantic similarity calculation. For example, when the description of an electrical component is "1kΩ resistor", it is directly mapped to the "resistance" type based on rule parsing, and the natural language processing algorithm further analyzes its attribute information and classifies it into specific categories. After semantic annotation, each node is given an independent semantic label, so that it can be distinguished and applied in subsequent association calculations.

[0067] Next, defining the geometric attributes, logical attributes, and state attributes of each semantic node is the core step in generating context node attribute data. Geometric attributes are used to describe the physical characteristics of nodes in the three-dimensional simulation model, mainly including position, size, and topology; logical attributes are used to describe the control dependency and signal flow of nodes in the electrical schematic, such as the logical relationship between the power controller and the actuator; state attributes reflect the dynamic characteristics of the node, including real-time status (such as temperature, movement speed) and timing behavior. Taking the three-dimensional component "bracket" as an example, its geometric attributes can be described as "size is (200mm, 50mm, 20mm), position is (100, 200, 300)", logical attribute is "connected to electrical component X", and state attribute is "real-time temperature 40°C". The above attributes are stored in the context node attribute data through the direct association relationship of the node.

[0068] Based on the above attribute definitions, context-related data is further generated through the geometric adjacency, logical dependency, and state transfer relationships of semantic nodes. The geometric adjacency relationship is mainly constructed based on the positional relationship of nodes in three-dimensional space. For example, the geometric adjacency relationship between two nodes can be calculated by the spatial distance function, and its formula is as follows:

[0069] ,in, Representation Node and The spatial distance and is the three-dimensional coordinate of the node. The logical dependency is realized by analyzing the signal flow and control link in the electrical schematic. For example, if the signal flow of component A points to component B, then it is defined that there is a control dependency between A and B. The state transfer relationship is realized by monitoring the state changes between nodes. For example, if the temperature change of node C directly causes the state change of node D, then it is defined that there is a state transfer association between C and D.

[0070] After generating the situation node attribute data and situation association data, the situation dynamic model data is constructed through multi-layer relationship analysis. Multi-layer relationship analysis integrates the association information of geometric, logical and state dimensions, and realizes the fusion of multimodal information through a unified relationship representation. For example, for the multi-layer association between the component "bracket" and the electrical component "controller", its geometric adjacency relationship describes the spatial position relationship between the bracket and the controller, the logical dependency relationship reflects the controller's action control of the bracket, and the state transfer relationship describes the feedback effect of the bracket state change on the controller signal. By integrating the above relationships, the situation dynamic model data represents the association structure of multimodal information in the form of a high-dimensional semantic graph.

[0071] The present invention realizes the clarification and standardization of data semantics through semantic annotation, significantly improving the accuracy of association modeling; through attribute definition and association construction, a unified input format is provided for the subsequent generation of semantic association matrix; through multi-layer relationship analysis, a deep fusion of geometric, logical and state information is realized, laying the foundation for the accuracy and adaptability of complex dynamic simulation. Overall, this method effectively solves the problem of heterogeneous data fusion and significantly improves the robustness and efficiency of semantic association modeling.

[0072] Based on the situational dynamic model data, a semantic association matrix is ​​generated through an iterative semantic aggregation algorithm, a geometric layer model, a logical layer model and a cross-domain layer model are constructed using the semantic association matrix, and the geometric layer model, the logical layer model and the cross-domain layer model are dynamically simulated in combination with a distributed cellular automation to generate simulation data, wherein the simulation data includes dynamic state data of the three-dimensional simulation model and logical state data of the electrical schematic diagram;

[0073] In the present invention, generating a semantic association matrix through situational dynamic model data is the core step of constructing geometric layer models, logical layer models and cross-domain layer models, which provide a multi-level structural foundation for dynamic simulation. The semantic association matrix is ​​generated by an iterative semantic aggregation algorithm, and the above model is dynamically simulated in combination with a distributed cellular automaton, and finally simulation data is generated. The simulation data includes the dynamic state data of the three-dimensional simulation model and the logical state data of the electrical schematic diagram, which can reflect the dynamic behavior and logical interaction relationship of the system as a whole.

[0074] Based on the contextual dynamic model data, a semantic association matrix is ​​generated through an iterative semantic aggregation algorithm. The contextual dynamic model data includes semantic node data, contextual node attribute data and contextual association data, which respectively describe the basic semantics, attribute characteristics and mutual association information of the nodes. In the present invention, the semantic association matrix is ​​a quantitative expression of these multidimensional data to represent the semantic association strength between nodes. The iterative semantic aggregation algorithm adopts a multi-head attention mechanism and a rule optimization strategy.

[0075] The generated semantic association matrix is ​​used to construct the geometric layer model, the logical layer model and the cross-domain layer model. The geometric layer model describes the geometric position relationship, physical coupling relationship and motion constraint relationship between the components of the 3D simulation model by parsing the geometric-related weight items in the association matrix; the logical layer model describes the signal flow and control link in the electrical schematic diagram by parsing the logical-related weight items; the cross-domain layer model combines the geometric layer and the logical layer to describe the control dependency and state feedback between the 3D simulation model and the electrical schematic diagram. For example, in the geometric layer model, the distance and relative position between "component A" and "component B" are quantified as geometric weights, in the logical layer model, the signal transmission from "controller X" to "actuator Y" is modeled as logical weights, and in the cross-domain layer model, the coupling of the signal trigger relationship and motion changes between "controller X" and "component A" is quantified as cross-domain weights.

[0076] After constructing the above models, the geometric layer model, logical layer model and cross-domain layer model are dynamically simulated in combination with distributed cellular automata. Distributed cellular automata map each semantic node to a cell and simulate the dynamic behavior between nodes according to neighborhood rules.

[0077] The neighborhood rules are jointly determined by the geometric layer model, the logical layer model and the cross-domain layer model. For example, the displacement of "component A" will trigger the change of the logical signal of "controller X" and further act on the motion state of "component B".

[0078] The simulation data generated by dynamic simulation includes the dynamic state data of the 3D simulation model and the logical state data of the electrical schematic. The dynamic state data describes the motion trajectory, speed, and physical state changes of the components of the 3D simulation model; the logical state data describes the timing changes and control logic behaviors of the signals in the electrical schematic. For example, in a set of dynamic simulations, when controller X sends a signal to trigger the movement of component A, the simulation data records the propagation path of the control signal, the response time, and the real-time position changes of component A.

[0079] The present invention realizes the deep fusion and interactive modeling of multimodal data through the generation and dynamic simulation of semantic association matrix, and can accurately reflect the dynamic association and logical dependency between the three-dimensional simulation model and the electrical schematic. Compared with the traditional separate modeling method, the present invention significantly improves the dynamic adaptability and association expression ability of the system, and provides strong support for the behavior analysis and fault diagnosis of complex systems. For example, in the simulation of aviation equipment, the present invention can efficiently capture the triggering relationship of the controller to the control surface action, and simultaneously simulate the real-time motion behavior and feedback signal changes of the control surface, thereby providing comprehensive data support for control system optimization.

[0080] Preferably, the semantic association matrix is ​​generated by an iterative semantic aggregation algorithm, and the iterative semantic aggregation algorithm comprises the following steps:

[0081] Based on semantic nodes and The calculation formula of the association weight is as follows:

[0082] , where the association weight Represents a semantic node and The final weight of the semantic association rule set Represents a set of rules used for association calculation; weights of semantic association rules Representation Rules Importance in calculations; rule scoring Representation Rules Semantic Node and The strength of association;

[0083] The dynamic adjustment formula of the association weight is as follows:

[0084] , where the association weight after iteration is Represents a semantic node and In the The association weight after iterations; adjustment coefficient Represents the smoothing coefficient of the dynamic adjustment of the weight; The association weight of the iteration Represents the weight of the semantic node in the previous iteration; attention score Represents a semantic node calculated based on a multi-head attention mechanism and of relevance.

[0085] In the present invention, the generation of the semantic association matrix is ​​a key step in achieving deep fusion of multimodal data and semantic modeling. Through the iterative semantic aggregation algorithm, the semantic association matrix can quantitatively represent the association strength between semantic nodes and dynamically adjust the weights to reflect the changes in multidimensional semantic relationships. The present invention generates an accurate and dynamically adaptive semantic association matrix by introducing rule weight calculation and dynamic attention mechanism optimization, which provides strong support for subsequent association modeling and dynamic simulation.

[0086] First, the semantic association matrix is ​​initialized based on the context dynamic model data, and the initial association weight between semantic nodes iii and jjj is calculated by the following formula:

[0087]

[0088] The specific score is calculated based on the content of the rule. For example, for a geometric adjacency rule, the score can be the inverse of the spatial distance between nodes:

[0089] ,in, For Node and The Euclidean distance between .

[0090] Secondly, based on the initial weights, the semantic association matrix is ​​optimized through a dynamic adjustment mechanism. The dynamic adjustment adopts the weight update formula based on the multi-head attention mechanism:

[0091]

[0092] is the smoothing coefficient, which is used to control the influence ratio of the current weight and attention score in the update, and its value range is usually ; For the The association weight at the iteration; Score the node relevance calculated based on the multi-head attention mechanism.

[0093] The implementation of the multi-head attention mechanism in this invention is based on the following formula:

[0094] ,in, , and Respectively represent nodes and nodes query, key, and value vectors, Represents the dimension of the vector, used to normalize the relevance score; The function maps the relevance score to the probability distribution range to emphasize highly relevant nodes. Through the attention mechanism, the semantic relevance matrix can adaptively capture the highly relevant features between nodes and dynamically optimize the association relationship.

[0095] The core of the iterative semantic aggregation algorithm is to optimize the semantic association matrix through multiple iterations so that it gradually converges on local and global semantic associations. For example, in multiple iterations, geometric adjacency relationships with higher initial rule weights may be dynamically adjusted to lower weights to highlight the importance of logical dependencies. This dynamic optimization mechanism ensures that the semantic association matrix can accurately reflect the multi-dimensional association characteristics between semantic nodes.

[0096] In the practical application of the present invention, the semantic association matrix generation process significantly improves the fusion efficiency and semantic expression ability of multimodal data. For example, in the simulation of aviation control systems, semantic nodes can represent aircraft components (such as wings, controllers, etc.). Through iterative optimization, the semantic association matrix can accurately capture the logical dependency between the wing and the controller and the feedback effect of the wing motion state on the control signal, providing an accurate semantic association description for subsequent dynamic simulation.

[0097] The semantic association matrix generated by the iterative semantic aggregation algorithm has the following advantages: first, the dynamic optimization mechanism can capture the semantic changes of multimodal data in real time, improving the adaptability of association modeling; second, the combination of rules and attention mechanisms ensures the accuracy of matrix weights and significantly improves the robustness of association description; finally, the generated semantic association matrix provides a strong foundation for the construction of geometric layer, logical layer and cross-domain layer models, and can support the efficient implementation of complex dynamic simulation. Through this step, the entire simulation system achieves deep fusion at the semantic level, laying a solid foundation for the dynamic analysis and optimization of complex systems.

[0098] Preferably, the geometric layer model, logical layer model and cross-domain layer model supported by the semantic association matrix include:

[0099] The geometric layer model describes the geometric position relationship, motion constraints and physical coupling of components in the three-dimensional simulation model; the logical layer model describes the connectivity, signal flow and control logic of the graphics elements in the electrical schematic diagram; the cross-domain layer model describes the control relationship between the three-dimensional simulation model and the electrical schematic diagram components.

[0100] In the present invention, the geometric layer model, logical layer model and cross-domain layer model supported by the semantic association matrix are the key structures for realizing the deep fusion and simulation of multimodal data. They model the system from the dimensions of geometric relationship, logical relationship and cross-domain interaction respectively. These three models use the weight information provided by the semantic association matrix to construct accurate and efficient multi-level association descriptions, thus providing a unified modeling framework for the dynamic simulation of complex systems.

[0101] First, the geometric layer model describes the geometric position relationship, motion constraints and physical coupling of components in the 3D simulation model based on the weights related to the geometric features in the semantic association matrix. This model is mainly calculated through the geometric adjacency relationship and motion characteristics of the components, such as the distance, angle, contact relationship and constraint conditions between components. The geometric position relationship of the components can be quantified by the following formula:

[0102] ,in For Node and The spatial distance and is the three-dimensional coordinate of the node. The physical coupling relationship is described by the contact force and friction force between nodes. For example, the spring coupling between components can be calculated by Hooke's law:

[0103] ,in, is the spring force, is the elastic coefficient, The displacement change between components. Through the geometric layer model, geometric properties such as "the contact relationship between the bracket and the bearing" or "the spatial constraint between component A and component B" can be constructed.

[0104] Secondly, the logic layer model describes the connectivity, signal flow, and control logic of the elements in the electrical schematic based on the weights related to the logical features in the semantic association matrix. The connectivity between electrical components is defined by the logical rules in the semantic association matrix, and the signal flow and control logic are constructed based on the logical dependencies between the elements. For example, for two elements and , the logical weight of the signal transmission path can be expressed as:

[0105] ,in, Indicates the signal from arrive The propagation delay of The logic layer model also supports complex control logic modeling. For example, the relationship between a controller and an actuator can be described by Boolean logic. The high level of the signal triggers the actuator The expression of these logical relationships provides support for the functional simulation of the system.

[0106] Finally, the cross-domain layer model is constructed through the cross-domain weights in the semantic association matrix to describe the control relationship between the 3D simulation model and the electrical schematic components. The core of this model is to fuse the data of the geometric layer model and the logic layer model to capture the dynamic interaction between the geometric state and the logic signal. For example, when the controller signal triggers the movement of the component, the cross-domain layer model describes the relationship between the signal propagation and the change of the component position through real-time association weight updates. The weight calculation formula of the cross-domain layer is as follows:

[0107] ,in, is the cross-domain association weight, is the geometry layer weight, is the logic layer weight, is a fusion parameter used to adjust the importance of geometric and logical information. For example, when the controller "When the signal triggers the movement of "component A", the geometric layer model describes the movement path and position change of the component, the logical layer model describes the signal trigger path, and the cross-domain layer model unifies the two parts of information to achieve overall dynamic association.

[0108] Through the construction of the above three models, the present invention realizes multi-level modeling and efficient fusion of multimodal data. For example, in an aviation equipment simulation, the geometric layer model can describe the motion coupling between the wing and the aileron, the logic layer model can describe the transmission logic of the aileron control signal, and the cross-domain layer model can describe the interactive process of how the control signal drives the aileron movement and feeds back to the controller. The overall modeling framework can dynamically capture the complex interactive relationship between multimodal data, providing accurate and scalable modeling support for dynamic simulation and fault diagnosis.

[0109] Through the construction of geometric layer models, logical layer models and cross-domain layer models, the system realizes complete modeling from structure to logic to multi-domain interaction, greatly improving the fusion depth and simulation accuracy of multimodal data. This multi-level modeling method can effectively support the behavior analysis, control optimization and dynamic fault detection of complex systems, and provides an efficient and innovative solution for application scenarios in a wide range of fields (such as aviation simulation, intelligent manufacturing, robot control, etc.).

[0110] Based on simulation data, a causal path inference algorithm is used to perform feedback optimization on the semantic association matrix and dynamic modeling rules to generate optimized semantic association matrix and dynamic modeling rules. The optimized semantic association matrix and dynamic modeling rules are adjusted through a real-time correction mechanism, a trend correction mechanism and a global correction mechanism to generate synchronization status data. The synchronization status data is used to synchronize the status display of the three-dimensional simulation model and the electrical schematic diagram as well as the dynamic feedback of the user interaction interface.

[0111] In the present invention, the feedback optimization and correction mechanism of simulation data is the core step to ensure the synchronous display and dynamic interaction of the three-dimensional simulation model and the electrical schematic diagram. The semantic association matrix and dynamic modeling rules are optimized by the causal path inference algorithm, and multi-level adjustments are made in combination with the real-time correction mechanism, trend correction mechanism and global correction mechanism to generate synchronous state data and realize accurate modeling and feedback of the dynamic behavior of the system.

[0112] First, the semantic association matrix and dynamic modeling rules are optimized through feedback through the causal path inference algorithm. The core of causal path inference is to identify the causal relationship of node state changes in simulation data and update the association matrix and modeling rules based on the causal inference model. The specific process includes three parts: causal path construction, anomaly identification and rule adjustment.

[0113] The optimized semantic association matrix and dynamic modeling rules are further adjusted through real-time correction mechanism, trend correction mechanism and global correction mechanism to generate synchronized state data. Based on the real-time comparison between simulation data and actual state, the dynamic deviation of the simulation model is corrected. The core is to adjust the state parameters through the error minimization model. The semantic association matrix and dynamic modeling rules are trend-adjusted using the time series prediction model. The prediction model fits the future state change trend based on historical data, such as predicting state changes through the autoregressive integrated moving average model (ARIMA). The global structure of the semantic association matrix is ​​optimized through consistency constraints to ensure the integrity and robustness of the matrix in multi-level associations. The goal of global correction is to minimize the overall error through iterative optimization algorithms (such as gradient descent).

[0114] The corrected semantic association matrix and dynamic modeling rules generate synchronized state data, which is used to synchronize the state display of the 3D simulation model and the electrical schematic diagram as well as the dynamic feedback of the user interaction interface. The synchronized state data includes the real-time motion state of the 3D simulation model and the logical state of the electrical schematic diagram, and is presented in a dynamic and visual form. For example, in a user-operated simulation system, after the user adjusts the control signal through the interface, the system instantly updates the controller state and synchronously displays the component motion trajectory and signal flow, ensuring that the user can intuitively perceive the system response and interact.

[0115] By combining causal path inference with a multi-layer correction mechanism, the accuracy, adaptability, and dynamic interaction capabilities of the simulation system are significantly improved. Causal path inference achieves a deep correlation between simulation data and actual status, providing a basis for rule optimization; the multi-level adjustment mechanism of real-time correction, trend correction, and global correction ensures the efficient adaptability of the model in a dynamic environment. Overall, this step can effectively support the dynamic simulation and user interaction of complex systems, and provide accurate and real-time synchronous feedback for a wide range of application scenarios (such as aviation equipment simulation, intelligent manufacturing, etc.).

[0116] Preferably, the dynamic modeling rules include state transition rules of distributed cellular automata, and the state transition rules are generated by the following optimization process:

[0117] The reward function calculation formula is as follows:

[0118] , where the reward value Represents the state reward of the optimization process; state consistency score Used to measure the degree of match between the cell simulation state and the actual state; simulation efficiency score Used to evaluate the efficiency of simulation calculations; weight coefficient Indicates the importance of the state consistency score;

[0119] The cell state update rules are as follows:

[0120] , where the cell state Representation Cell In the The state at the iteration; cell state Representation Cell In the The state at the iteration; the neighborhood state set Representation and Cell The state set of other directly related cells; state transition function Represents a function that calculates a new state by inputting the current cell state, a set of neighborhood states, and a reward value.

[0121] In the present invention, the state transition rules of the distributed cellular automaton in the dynamic modeling rules are implemented through reward-driven and state feedback mechanisms, aiming to dynamically adjust the cellular state to optimize the simulation accuracy and efficiency. The optimization process of the rule combines the reward function calculation with the state update strategy, and realizes the global optimization and local adaptation of the cellular state through multiple iterations, and finally generates efficient and dynamically consistent modeling rules, providing core support for the high-precision implementation of multimodal simulation.

[0122] The reward function is used to evaluate the quality of the cell simulation state and guide the optimization of the cell state through the reward value. The calculation formula is: .

[0123] Status consistency score : Measure the current state of the cell With target status The difference is calculated as:

[0124] ,in, For cells The target state, is the maximum value range of the state, used for normalization.

[0125] Simulation efficiency score : Measures the computational overhead of the current simulation, and the calculation formula is: ,in, is the computation time of the simulation process. By incorporating the efficiency score into the reward function, it ensures that state optimization not only focuses on the accuracy of the results, but also takes into account the efficiency of computing resources.

[0126] The state transition of the cellular automaton is realized by the following formula:

[0127]

[0128] The specific calculation is as follows:

[0129] ,in, is the learning rate, which is used to control the step size of state adjustment; is the neighborhood weight, indicating the cell For cells The intensity of the impact; is the number of neighborhood cells. The state transition rule combines neighborhood influence and reward drive to achieve dynamic adjustment of cell states.

[0130] In practical applications, the state transition rules of distributed cellular automata can be used for multimodal simulation of complex systems. For example, in an aviation control system simulation, cells can represent various components of an aircraft (such as wings, engines, control surfaces, etc.), and their states reflect the motion parameters (such as displacement, speed) and physical states (such as temperature, pressure) of the components. Through reward function calculation, the state of each cell is dynamically adjusted to keep it consistent with the control signal and environmental conditions. For example, when wing cell A receives a signal from control cell B, the reward function drives its state optimization to match its motion with the actual control logic while considering the simulation efficiency.

[0131] By combining the reward function with the state transition rule, the distributed cellular automation can achieve the following effects: The state consistency score ensures a high match between the simulation state and the target state, significantly improving the accuracy of the simulation. The simulation efficiency score incorporates the calculation time into the optimization target, effectively reducing the calculation overhead and making the simulation process more efficient. The state transition rule combines neighborhood influence and reward drive to ensure that the cellular state can dynamically adapt to complex multimodal interaction requirements. This rule is suitable for simulation modeling of a variety of complex systems, such as production line scheduling simulation in intelligent manufacturing, multi-agent behavior simulation in robot control, etc.

[0132] Preferably, the process of optimizing the state transition rule of the distributed cellular automaton is performed based on a reinforcement learning method, and the optimization process includes the following steps:

[0133] Constructing a reward function, wherein the optimization objectives of the reward function include minimizing simulation error and maximizing simulation efficiency;

[0134] Constructing training samples based on simulation historical data, wherein the training samples are used for training a reinforcement learning model;

[0135] During the reinforcement learning training process, the weight distribution of the cell neighborhood state is dynamically adjusted to optimize the state transition parameters of the neighborhood rule;

[0136] The state transition rules optimized by reinforcement learning are applied to dynamic simulation to update the cell state.

[0137] In the present invention, optimizing the state transition rules of distributed cellular automata is a key link to improve the accuracy and efficiency of dynamic simulation. By optimizing the state transition rules through reinforcement learning methods, the state weight distribution of the cellular neighborhood can be dynamically adjusted, and the automatic parameter tuning of the neighborhood rules can be realized by combining the reward function and the simulation history data. The present invention constructs a reinforcement learning framework, makes full use of historical simulation data, optimizes the neighborhood interaction rules and state update parameters, thereby realizing global optimization and dynamic adaptation of the cellular state.

[0138] The reward function is the core of the optimization objective in the reinforcement learning framework. The design focus is to balance the minimization of simulation error and the maximization of simulation efficiency. The expression of the reward function is:

[0139] ,in, is the reward value; is the error of the current simulation, defined as the deviation between the simulation state and the actual state, is the maximum allowable error; is the simulation efficiency, defined as the number of computational steps per unit time, is the ideal simulation efficiency; is the weight coefficient, which is used to balance the importance of error and efficiency. The value of can be used to balance simulation accuracy and efficiency. For example, in a high-precision scenario, a higher The value prioritizes the simulation error.

[0140] The training of reinforcement learning models relies on high-quality training samples, which are constructed from simulation history data, including cell states, neighborhood states, state transition parameters, and deviations from actual states. The construction process of training samples is as follows:

[0141] State space definition: the state of each cell and the neighborhood state set Construct a state space to record the current state of the cell, the neighborhood interaction state and its associated weights.

[0142] Action space definition: The action space of reinforcement learning includes adjusting the neighborhood weight distribution and state transition parameters ,in Representation Cell and neighborhood cells The weight between.

[0143] Reward feedback record: reward value calculated based on the reward function As an optimization objective, it is used to guide model learning.

[0144] After the construction is completed, the training samples are input into the reinforcement learning model and trained using deep reinforcement learning (such as deep Q network, DQN). The optimization goal of DQN is to minimize the time difference error of the Q value:

[0145] ,in, is the loss function, For the current state and actions The Q value, are network parameters, are the target network parameters.

[0146] During the reinforcement learning training process, the cell neighborhood weights and state transition parameters are dynamically adjusted according to the model output. The adjustment rules are as follows:

[0147] Neighborhood weight update:

[0148] ,in, is the updated neighborhood weight, is the learning rate, is the reward value. The weight update strengthens the high reward value path and suppresses the low reward value path.

[0149] State transfer parameter update:

[0150] ,in, is the next state transfer parameter, is the learning rate, and the state transition parameters are optimized by gradient descent.

[0151] The state transition rules optimized by reinforcement learning are applied in dynamic simulation to achieve dynamic adjustment of cell states. The optimized rules directly affect the cell state update function:

[0152]

[0153] By dynamically updating states and neighborhood weights, the simulation model can be made adaptive to complex dynamic behaviors.

[0154] Through the state transition rules optimized by reinforcement learning, the dynamic simulation system can significantly improve accuracy, efficiency and dynamic adaptability. In specific implementation: The reinforcement learning framework optimizes the neighborhood weights and state transition parameters based on historical data to ensure that the simulation error is minimized. For example, in the simulation of multi-component collaborative motion, the optimized rules can reflect the coupling relationship between components with high accuracy. By weighing the relationship between error and efficiency through the reward function, reinforcement learning automatically optimizes the simulation performance and significantly reduces the computational overhead without sacrificing accuracy. Dynamically adjust the neighborhood weight distribution and state transition parameters so that the simulation model can adapt to changes in complex scenarios in real time. For example, in aviation simulation, the optimized rules can quickly adapt to dynamic adjustments under airflow disturbances.

[0155] Preferably, Figure 4 As shown, the causal path inference algorithm includes the following steps:

[0156] Construct a causal path diagram based on simulation data to analyze the causal relationship between component state changes and logic changes during the simulation process; identify abnormal state nodes and their triggering conditions in the causal path diagram, and locate the cause of the abnormality through path backtracking; adjust the semantic association matrix and dynamic modeling rules to eliminate abnormal states and optimize the simulation path.

[0157] In the present invention, the causal path inference algorithm is the core tool for analyzing the causal relationship between component state changes and logic changes during the simulation process. The algorithm ensures the rationality of the simulation path and the accuracy of the dynamic simulation by constructing a causal path graph, identifying abnormal state nodes and their triggering conditions, and adjusting the semantic association matrix and dynamic modeling rules, thereby optimizing the overall performance of the system.

[0158] The causal path diagram is used to represent the causal relationship between component states and logic changes during the simulation process. A directed graph is constructed by calculating the causal strength between states. The construction of the causal path diagram is based on the following formula:

[0159] ,in, Representation Components For components the strength of the causal influence; For components and The state covariance of For components By calculating the causal strength, we can quantify the causal relationship between states and construct a directed graph, where nodes represent component states and edges represent causal paths.

[0160] In the causal path diagram, abnormal state nodes and their triggering conditions can be identified through path backtracking analysis. The determination of abnormal state nodes is based on the degree of deviation of causal strength:

[0161] ,in, is the difference between the current causal strength and the baseline causal strength; When , it is determined to be an abnormal path. For example, when the temperature of component A deviates significantly from the normal range and the control signal to component B is abnormal, the abnormal temperature of A can be located as the triggering cause of the abnormal signal through path backtracking.

[0162] According to the analysis results of abnormal state nodes, the semantic association matrix and dynamic modeling rules are adjusted to eliminate abnormal states and optimize the simulation path. The adjustment process includes:

[0163] Semantic association matrix adjustment: recalculate association weights on abnormal paths , the adjustment formula is as follows:

[0164] ,in, To adjust the step size, the influence of abnormal causality is weakened by reducing the weight of abnormal paths.

[0165] Dynamic modeling rule adjustment: Update the parameters in the state transition rule, such as adjusting the reward function of the state update to make it more focused on state consistency:

[0166] ,in, is the state consistency score after anomaly correction.

[0167] After the optimization is completed, the effect of the adjustment is verified by re-simulation to ensure that the abnormal state is eliminated and the simulation path is reasonable. For example, after adjusting the semantic association matrix and dynamic modeling rules, the signal transmission path of components A and B becomes more reasonable, and its state changes can reflect the actual logic.

[0168] The application of the causal path inference algorithm has significantly improved the simulation accuracy and stability of the system, which is mainly reflected in the following aspects: By quantifying the component status and logical relationship through the causal path diagram, the abnormal nodes and their triggering conditions can be accurately located. For example, when the logical signal of a component is inconsistent with the physical state, the causal path inference can quickly analyze the cause of the abnormality. By adjusting the semantic association matrix and dynamic modeling rules, the abnormal state can be effectively eliminated and the system's adaptability to complex dynamic environments can be improved. For example, in aviation simulations under dynamic airflow interference, the optimized path can maintain the dynamic consistency of the system. After the causal path is optimized, the system can achieve higher simulation accuracy with lower computing overhead, significantly improving overall efficiency.

[0169] Example: Taking the simulation of aircraft control systems as an example, the causal path inference algorithm can be used to analyze the abnormal relationship between the wing state and the control signal. Assume that during the simulation, the wing lift angle (state node A) and the controller's command signal (state node B) are inconsistent. Through the construction of the causal path diagram, it is found that:

[0170] Significantly lower, indicating that the controller signal failed to properly trigger the wing state change.

[0171] Path tracing revealed that the airflow disturbance of the wing (state node C) was the trigger condition for the abnormal signal.

[0172] By adjusting the path weights in the semantic association matrix and ,After re-simulation, the lift and drop angle of the wing can accurately respond to the control ,signal, the abnormal state is eliminated, and the simulation path is restored to ,reasonableness.

[0173] Preferably, Figure 3 As shown, the real-time correction mechanism is used to correct slight deviations between simulation data and actual status; the trend correction mechanism adjusts the semantic association matrix and dynamic modeling rules in advance based on the time series prediction model; and the global correction mechanism optimizes the global structure of the semantic association matrix through consistency constraints.

[0174] In the present invention, the real-time correction mechanism, trend correction mechanism and global correction mechanism together constitute a dynamic correction framework to ensure the high consistency and adaptability of simulation data with the actual state. Each correction mechanism optimizes the semantic association matrix and dynamic modeling rules from different time and space dimensions, respectively solving the problems of real-time error, trend change and global consistency, thereby improving the accuracy, efficiency and robustness of the simulation.

[0175] The real-time correction mechanism aims at the slight deviation between the simulation data and the actual state, and realizes the rapid correction of the simulation state by introducing error compensation. Its core is to dynamically adjust the simulation state through the error feedback control model. The correction formula is:

[0176] ,in, is the state after correction; It is the simulation state; For the actual state; It is the correction weight, which is used to control the strength of error correction.

[0177] Example: In aviation simulation, when the sensor detects that the actual height of the aircraft is different from the simulated height, the real-time correction mechanism adjusts the simulated height parameters to ensure that the two are quickly consistent. The simulation altitude is 1000 meters. 995 meters, corrected weight =0.8, then the simulated height after correction is:

[0178]

[0179] By introducing this error compensation, the real-time correction mechanism can quickly respond to dynamic environmental changes and eliminate the instantaneous deviation between simulation and reality.

[0180] The trend correction mechanism uses the time series prediction model to adjust the semantic association matrix and dynamic modeling rules in advance to ensure that the simulation can adapt to future trend changes. Its core is to predict the time series of simulation data and feed the prediction results back to the modeling rules. The trend correction formula is based on the autoregressive integrated moving average model (ARIMA):

[0181] ,in, is the current state; is the regression coefficient; is a random error.

[0182] The predicted state values ​​are used to adjust the associated weights and modeling rules in advance. For example, in smart manufacturing simulation, if it is predicted that the operating temperature of a certain device will continue to rise in the future, the trend correction mechanism will adjust the priority weight of its cooling device to adapt to the upcoming temperature change.

[0183] Example: In traffic flow simulation, the trend correction mechanism can predict the traffic flow changes during future peak hours based on past traffic flow data and adjust the signal light control rules in advance. For example, if it is predicted that the future traffic flow of a certain road section will reach 500 vehicles / minute, the trend correction mechanism will increase the green light duration of the signal light on the road section to optimize traffic flow.

[0184] The global correction mechanism optimizes the global structure of the semantic association matrix through consistency constraints to ensure the coordination and consistency of the matrix at the local and global levels. Its core is to adjust the association weight by minimizing the error function. The optimization formula is:

[0185] ,in, is the error function; is the optimized weight; is the actual observation weight.

[0186] The global correction mechanism uses the gradient descent method to optimize the error function and update the weight distribution of the semantic association matrix to ensure global consistency. For example, when the weight of a path in the semantic association matrix deviates from the average value of the global structure, the global correction mechanism optimizes it to keep it consistent with the global structure.

[0187] Embodiment: In a robot cluster collaboration simulation, if the task allocation weight of a robot is too high, resulting in an imbalance in global resource allocation, the global correction mechanism will reduce the task priority weight of the robot and reallocate resources to ensure the balance of the overall task allocation.

[0188] Realize instant alignment between the simulation state and the actual state to eliminate instantaneous errors. For example, in a complex dynamic environment, real-time correction can quickly respond to external disturbances and maintain the dynamic consistency of the simulation model. By predicting future trend changes in advance and adjusting modeling rules, the foresight and adaptability of the simulation are improved. For example, in a manufacturing system, the maintenance plan can be adjusted according to the gradual decline in equipment performance. Optimize the global structure of the semantic association matrix to eliminate the impact of local deviations on the overall model. For example, in a multi-agent system, ensure the consistency of individual behavior with the global goal, and improve the overall robustness of the system.

[0189] Take the simulation of an unmanned vehicle as an example:

[0190] Real-time correction mechanism: When the vehicle sensor detects that the current speed deviates from the target speed, the real-time correction mechanism adjusts the simulated speed according to the error to make the simulated state consistent with the actual state. Trend correction mechanism: Based on the historical traffic flow, the trend correction mechanism predicts the future road congestion trend and adjusts the navigation path and the cooperation rules between vehicles in advance. Global correction mechanism: In the multi-vehicle collaborative obstacle avoidance simulation, the global correction mechanism optimizes the obstacle avoidance weights in the semantic association matrix to ensure that the behavior of each vehicle is consistent with its own goals and coordinated with the global traffic flow.

[0191] Preferably, the synchronization status data is generated in the following manner:

[0192] Combined with the dynamic state and logical state in the simulation data, the display state of the three-dimensional simulation model and the electrical schematic diagram is updated in real time; the user operation behavior is recorded through the user interaction interface, and dynamic feedback information is generated based on the intention prediction model; the dynamic feedback information includes real-time highlighting of related components and providing operation suggestions.

[0193] In the present invention, the generation of synchronized state data realizes the real-time linkage update of the 3D simulation model and the electrical schematic diagram and the dynamic feedback of user interaction, which significantly improves the response efficiency of the system and the user experience. This process combines the dynamic state and logical state in the simulation data to ensure the real-time synchronization of the state display, and records the user's operation behavior through the user interaction interface, and uses the intention prediction model to generate dynamic feedback information, including real-time highlighting of related components and providing operation suggestions.

[0194] The combination of dynamic state and logical state establishes the linkage relationship between the two through the semantic association matrix. The dynamic state reflects the geometric displacement, speed and rotation information in the three-dimensional simulation model, while the logical state describes the changes in signal flow and control logic in the electrical schematic. For example, during the simulation process, the dynamic state can be expressed by the formula:

[0195]

[0196] Describe, where is the position of the component in three-dimensional space, For its speed; the logic state is through:

[0197]

[0198] To express, For components The core of the linkage update of the two is to establish a state mapping relationship based on the weight of the semantic association matrix. When a switch signal in the electrical schematic changes, the update of its logical state will directly affect the dynamic state of the 3D simulation model through the weight of the association matrix. For example, when the switch signal becomes "on", the rotation movement of the corresponding component in the model occurs immediately, and the formula is:

[0199] ,in is the linkage weight, is the association matrix weight.

[0200] The recording of user interaction behaviors is completed through the interface operation behavior capture module, which records mouse clicks, drags, and other behaviors as operation sequences:

[0201]

[0202] Each behavior Indicates that the user is at time Based on these operation sequences, the intention prediction model uses the long short-term memory network (LSTM) to model the operation behavior and predict the user's operation intention. The intention prediction formula is:

[0203] ,in, is the weight matrix, is the hidden state of the operation sequence, The model predicts the user's next operation intention through implicit pattern analysis and uses the results to generate dynamic feedback information.

[0204] Dynamic feedback information includes two parts: real-time highlighting of associated components and operation suggestions. The highlighted associated components are based on the weights of the semantic association matrix. By calculating the association strength of the components, the system dynamically displays the part of the 3D simulation model or electrical schematic diagram associated with the user's current operation component. For example, when the user clicks a control switch in the electrical schematic interface, the system highlights the robotic arm component controlled by the switch in the 3D model, prompting the user of the linkage relationship between the components. Operation suggestions generate guidance information based on the user intention prediction results. For example, when the user tries to operate a function but fails to complete it, the system generates a prompt through semantic analysis, "After clicking this button, observe the state changes of the associated components", which improves user operation efficiency.

[0205] Through the above process, the generation of synchronous state data ensures the state linkage between the three-dimensional simulation model and the electrical schematic diagram and the dynamic response of user interaction. In practical applications, for example, in an industrial robot simulation system, when a user operates the motor switch in the electrical schematic diagram, the three-dimensional simulation model displays the motion state of the motor after starting in real time, highlights the associated mechanical arm components, and provides follow-up operation suggestions. The present invention effectively improves the user's intuitive operation, timely feedback, and overall system interaction efficiency.

[0206] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A semantic association modeling and adaptive simulation method based on multi-layer correction, characterized in that: The following steps are involved: Collect geometric feature data, kinematic feature data and component status data from the three-dimensional simulation model, collect primitive structure data and circuit logic data from the electrical schematic diagram, pre-process the collected data, and generate context dynamic model data, wherein the context dynamic model data includes semantic node data, context node attribute data and context association data; Based on the situational dynamic model data, a semantic association matrix is ​​generated through an iterative semantic aggregation algorithm, a geometric layer model, a logical layer model and a cross-domain layer model are constructed using the semantic association matrix, and the geometric layer model, the logical layer model and the cross-domain layer model are dynamically simulated in combination with a distributed cellular automation to generate simulation data, wherein the simulation data includes dynamic state data of the three-dimensional simulation model and logical state data of the electrical schematic diagram; Based on simulation data, a causal path inference algorithm is used to perform feedback optimization on the semantic association matrix and dynamic modeling rules to generate optimized semantic association matrix and dynamic modeling rules. The optimized semantic association matrix and dynamic modeling rules are adjusted through a real-time correction mechanism, a trend correction mechanism and a global correction mechanism to generate synchronization status data. The synchronization status data is used to synchronize the status display of the three-dimensional simulation model and the electrical schematic diagram as well as the dynamic feedback of the user interaction interface.

2. The semantic association modeling and adaptive simulation method based on multi-layer correction according to claim 1 is characterized in that: The data preprocessing includes the following steps: The collected geometric feature data, kinematic feature data and component status data of the three-dimensional simulation model, as well as the graphic element structure data and circuit logic data of the electrical schematic diagram are de-redundant, and the de-redundant processing includes identifying and deleting duplicate data through similarity calculation; aligning the time series data, and the time series alignment processing is completed based on a dynamic time warping algorithm; and standardizing the processed data to generate structured semantic node data in a unified format.

3. The semantic association modeling and adaptive simulation method based on multi-layer correction according to claim 2 is characterized in that: The semantic node data generates situational dynamic model data through the following steps: The structured semantic node data is semantically annotated by combining a rule-based parsing method with a natural language processing algorithm; the geometric attributes, logical attributes and state attributes of each semantic node are defined; context node attribute data is generated based on the geometric adjacency relationship, logical dependency relationship and state transfer relationship of the semantic nodes, and context-related data is constructed through multi-layer relationship analysis.

4. The semantic association modeling and adaptive simulation method based on multi-layer correction according to claim 1 is characterized in that: The semantic association matrix is ​​generated by an iterative semantic aggregation algorithm, which includes the following steps: Based on semantic nodes and The calculation formula of the association weight is as follows: , where the association weight Represents a semantic node and The final weight of the semantic association rule set Represents a set of rules used for association calculation; weights of semantic association rules Representation Rules Importance in calculations; rule scoring Representation Rules Semantic Node and The strength of association; The dynamic adjustment formula of the association weight is as follows: , where the association weight after iteration is Represents a semantic node and In the The association weight after iterations; adjustment coefficient Indicates the smoothing coefficient of dynamic weight adjustment; The association weight of the iteration Represents the weight of the semantic node in the previous iteration; attention score Represents a semantic node calculated based on a multi-head attention mechanism and of relevance.

5. The semantic association modeling and adaptive simulation method based on multi-layer correction according to claim 4 is characterized in that: The geometric layer model, logical layer model and cross-domain layer model supported by the semantic association matrix include: The geometric layer model describes the geometric position relationship, motion constraints and physical coupling of components in the three-dimensional simulation model; the logical layer model describes the connectivity, signal flow and control logic of the graphics elements in the electrical schematic diagram; the cross-domain layer model describes the control relationship between the three-dimensional simulation model and the electrical schematic diagram components.

6. The semantic association modeling and adaptive simulation method based on multi-layer correction according to claim 1 is characterized in that: The dynamic modeling rules include state transition rules of distributed cellular automata, and the state transition rules are generated by the following optimization process: The reward function calculation formula is as follows: , where the reward value Represents the state reward of the optimization process; state consistency score Used to measure the degree of match between the cell simulation state and the actual state; simulation efficiency score Used to evaluate the efficiency of simulation calculations; weight coefficient Indicates the importance of the state consistency score; The cell state update rules are as follows: , where the cell state Representation Cell In the The state at the iteration; cell state Representation Cell In the The state at the iteration; the neighborhood state set Representation and Cell The state set of other directly related cells; state transition function Represents a function that calculates a new state by inputting the current cell state, a set of neighborhood states, and a reward value.

7. The semantic association modeling and adaptive simulation method based on multi-layer correction according to claim 6 is characterized in that: The process of optimizing the state transition rule of the distributed cellular automaton is performed based on a reinforcement learning method, and the optimization process includes the following steps: Constructing a reward function, wherein the optimization objectives of the reward function include minimizing simulation error and maximizing simulation efficiency; Constructing training samples based on simulation historical data, wherein the training samples are used for training a reinforcement learning model; During the reinforcement learning training process, the weight distribution of the cell neighborhood state is dynamically adjusted to optimize the state transition parameters of the neighborhood rule; The state transition rules optimized by reinforcement learning are applied to dynamic simulation to update the cell state.

8. The semantic association modeling and adaptive simulation method based on multi-layer correction according to claim 1 is characterized in that: The causal path inference algorithm includes the following steps: Construct a causal path diagram based on simulation data to analyze the causal relationship between component state changes and logic changes during the simulation process; identify abnormal state nodes and their triggering conditions in the causal path diagram, and locate the cause of the abnormality through path backtracking; adjust the semantic association matrix and dynamic modeling rules to eliminate abnormal states and optimize the simulation path.

9. The semantic association modeling and adaptive simulation method based on multi-layer correction according to claim 8, characterized in that: The real-time correction mechanism is used to correct slight deviations between simulation data and actual status; the trend correction mechanism adjusts the semantic association matrix and dynamic modeling rules in advance based on the time series prediction model; the global correction mechanism optimizes the global structure of the semantic association matrix through consistency constraints.

10. The semantic association modeling and adaptive simulation method based on multi-layer correction according to claim 1, characterized in that: The synchronization status data is generated in the following manner: Combine the dynamic state and logical state in the simulation data to update the display state of the 3D simulation model and the electrical schematic in real time; record user operation behavior through the user interaction interface and generate dynamic feedback information based on the intention prediction model; The dynamic feedback information includes real-time highlighting of associated components and providing operation suggestions.

Images