Complex multivariate combination model comprehensive element intelligent extraction and recombination publishing method

Abstract

The invention provides a complex multivariate combination model comprehensive element intelligent extraction and recombination publishing method, and relates to the field of system comprehensive modeling simulation. The complex multivariate combination model comprehensive element intelligent extraction and recombination publishing method is realized through the following technical steps: step 1, preparing and preprocessing multi-source model data, analyzing a *. Mo file of a target system through a Modelica compiler, and generating an AST syntax tree containing component declarations and equation sets; and importing STEP format geometric data of the CAD model, and carrying out grid repair and feature edge extraction by using Paraview. According to the method, a lightweight digital fingerprint algorithm oriented to multi-source heterogeneous data is developed, while encryption strength is kept, feature extraction time of a gigabit-level CAD model is shortened to be within 30 seconds, an interpretable intelligent confirmation framework is constructed, a model element feature map is constructed, a technical barrier is broken through, a model element association rule base based on a semantic network is developed, and a multi-source heterogeneous data encryption algorithm is developed. And automatic binding and intelligent extraction of the geometric parameters and the multi-physical field equation are realized.

Description

Technical Field

[0001] This invention relates to the field of system integrated modeling and simulation, specifically to a method for intelligent extraction, recombination and release of integrated elements in complex multi-element combined models. Background Technology

[0002] A complex multivariate composite model refers to a comprehensive model that uses the Modelica model as its core and integrates geometric models, field models, and other model elements. The intelligent extraction, recombination, and release of comprehensive elements from complex multivariate composite models involves extracting model elements as needed, combining them, and releasing them as digital prototype units in a unified standard format. This supports the packaging and delivery of digital prototypes in digitalization processes.

[0003] Current multi-domain system modeling mainly faces the following technical bottlenecks: The problem of model information silos: Traditional Modelica simulation environments (such as Dymola and OpenModelica) mainly focus on system-level behavioral simulation and lack the ability to explicitly extract implicit geometric features (such as CAD parameters) and physical field properties (such as material constitutive relations) in the model.

[0004] Insufficient platform compatibility: Different engineering software (such as COMSOL, ANSYS, and CAD tools) interpret model data semantically differently. Existing intermediate formats (such as FMI / FMU) focus on interface encapsulation and do not achieve atomic decomposition of model elements.

[0005] Data reorganization is inefficient: manual extraction of model features involves repetitive work (such as manually identifying geometric parameters from code) and is prone to causing data inconsistencies.

[0006] Current multi-domain system modeling faces the following technical challenges: Geometric-physical field coupling is missing: Existing technologies cannot simultaneously extract parametric geometric features (such as the sweep path of rotating machinery) and associated physical constraints (such as bearing contact force equations) in Modelica models.

[0007] Semantic consistency challenge: The mathematical descriptions of model elements in different fields (such as control systems and structural mechanics) have implicit relationships such as unit system and coordinate system, and traditional conversion methods cannot maintain cross-domain semantic coherence.

[0008] Intermediate format limitations: Existing standards (such as STEP AP242) only support static geometric data exchange and do not include topological mapping rules for time-varying field data (such as transient temperature fields), which means that digital twin systems need to develop additional data adaptation layers.

[0009] Therefore, this invention proposes a method for intelligent extraction, recombination and release of comprehensive elements in complex multivariate combination models, thereby effectively solving the aforementioned pain points and problems. Summary of the Invention

[0010] To address the shortcomings of existing technologies, this invention provides a method for intelligent extraction, recombination, and release of comprehensive elements in complex multivariate combined models. By constructing a feature map of model elements, it overcomes technical barriers, develops a semantic web-based rule library for model element association, and achieves automatic binding and intelligent extraction of geometric parameters and multiphysics field equations. It proposes a fractal coding strategy to convert mixed algebraic-differential equation systems into C code, supports intelligent verification, and designs a lightweight encapsulation release protocol based on model-based systems engineering, ensuring that the output files are compatible with both geometric and field data standards. It also supports automatic interface call testing to confirm the correctness of the released digital prototype.

[0011] To achieve the above objectives, the present invention provides the following technical solution: A method for intelligent extraction, recombination, and release of comprehensive elements in complex multivariate combination models, the method being implemented through the following technical steps: Step 1: Multi-source model data preparation and preprocessing. The *.mo file of the target system is parsed using the Modelica compiler to generate an AST syntax tree containing component declarations and equations; STEP format geometric data of the CAD model is imported, and mesh repair and feature edge extraction are performed using Paraview; Discretized matrix data of the field model is collected, and time series alignment and spatial coordinate normalization are performed. Step 2: Intelligent extraction of comprehensive elements, initiating a hybrid neural network hierarchical processing architecture; Geometric channel: Triangular patch data is input into the improved ResNet-50 network, and topological features such as assembly interference and fit tolerance are extracted through 3D convolution kernel (11×11×11); Syntax Channel: Traverse Modelica syntax tree nodes to identify variable scope, equation coupling relationships, and interface definitions; Field Model Channel: Deploy a graph attention network to analyze field strength distribution and boundary effects based on node gradient propagation; Step 3: Construct the element knowledge graph, creating four types of nodes in the graph database: Physical entity nodes (geometric feature extraction results), mathematical model nodes (Modelica analysis results), physical field nodes (multi-field model analysis results), and constraint rule nodes (domain knowledge base). Step 4: Simulation code generation and verification, calling DRL-Coder to perform layered code generation: The upper-level strategy network planning component topology generates the Modelica code framework; Mid-level strategy network population connect statement and parameter type declaration; The underlying policy network optimizes numerical parameters; Step 5: Dynamic model reorganization and optimization. A multi-objective optimization algorithm is adopted, with performance indicators (simulation speed), accuracy indicators (error with measured data), and resource consumption (memory usage) as optimization objectives. The reorganization scheme is iterated in the digital twin sandbox. The design variables are the element association weights and solver configuration parameters, and the constraints are mandatory clauses in the engineering rule base (such as a safety factor ≥ 2.5). Step 6: Standardized release of digital prototypes. Perform release preprocessing on the validated recombinant model: Generate a digital fingerprint, which includes the model hash value, version number, and contributor signature; Automatically package into PMU standard format and construct PMI interface description XML file; Generate a C language interface wrapper layer; Step 7: Smart Verification and Deployment. Perform the following verification process in the target deployment environment: 1: Geometric consistency confirmation; 2: Interface compatibility testing; 3: Real-time verification; 4. Resource integrity check; 5: Traceability Verification 6: User confirmation.

[0012] Furthermore, in the multi-source model data preparation and preprocessing steps, the preprocessed data is stored in a multimodal database in HDF5 format, and the physical dimensions and engineering constraint labels of each element are labeled.

[0013] Furthermore, in the comprehensive element intelligent extraction step, the feature fusion stage performs tensor normalization and injects domain knowledge rules to output an element list with confidence rating.

[0014] Furthermore, in the element knowledge graph construction step, edge relationship weights are calculated to establish an association network containing semantics such as "spatial assembly", "energy transfer", and "data dependency".

[0015] Furthermore, in the simulation code generation and verification step, the generated code triggers triple verification in real time: syntax verification calls the Modelica compiler, physical verification performs dimensional balance calculations, and functional verification uses an automated verification framework to perform test case coverage analysis.

[0016] Furthermore, in the model dynamic reorganization optimization step, a Pareto front solution set is generated in each iteration, and the optimal reorganization scheme is selected by the Top-Order Ideal Solution Ranking (TOPSIS) decision method.

[0017] This invention provides a method for intelligent extraction, recombination, and publishing of comprehensive elements in complex multivariate combination models. It has the following beneficial effects: 1. This invention provides a method for intelligent extraction, recombination and release of comprehensive elements in complex multivariate combined models. It develops a lightweight digital fingerprint algorithm for multi-source heterogeneous data, which reduces the feature extraction time of gigabit-level CAD models to within 30 seconds while maintaining encryption strength. It constructs an interpretable intelligent confirmation framework. When boundary condition violations are detected, the system automatically traces back to the associated Modelica equation or geometric assembly constraints and generates visual correction suggestions. It introduces a version management mechanism. Each released prototype generates a smart contract containing a timestamp, verification digest and contributor signatures, realizing traceable auditing of modification history.

[0018] 2. This invention provides a method for intelligent extraction, recombination and release of comprehensive elements in complex multivariate combined models. By constructing a feature map of model elements, it overcomes technical barriers, develops a semantic web-based model element association rule library, realizes automatic binding and intelligent extraction of geometric parameters and multiphysics field equations, proposes a fractal coding strategy to convert the mixed algebraic-differential equation system into C code, supports intelligent verification, designs a lightweight encapsulation release protocol based on model-based systems engineering, makes the output file compatible with both geometric and field data standards, supports automatic interface call testing, and confirms the correctness of the released digital prototype. Attached Figure Description

[0019] Fig. 1 This is a schematic diagram of the hybrid neural network element hierarchical recognition architecture of the present invention; Fig. 2 The workflow of the deep reinforcement learning code generator of this invention; Fig. 3 This is a typical workflow for digital prototype release and intelligent verification of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1: like Figs. 1-3 As shown, this invention provides a method for intelligent extraction, recombination, and release of comprehensive elements in complex multivariate combination models. The main purpose of this method is to provide three technologies: Intelligent extraction technology for comprehensive elements: The hybrid neural network feature hierarchical recognition architecture is a multimodal feature fusion technology for complex multivariate composite models. This architecture constructs parallel three-level feature extraction channels to process heterogeneous data sources such as Modelica syntax models, geometric models, and physical field models, achieving accurate identification and association of cross-domain model elements. Its core consists of three main modules: a residual network (ResNet-50), a bidirectional long short-term memory network (BiLSTM), and a graph neural network (GNN). The geometry processing channel uses a deep residual network to extract the topological features of the CAD model and identifies assembly relationships by analyzing the spatial distribution patterns of point cloud data. The syntax parsing channel uses a bidirectional LSTM to process the Modelica abstract syntax tree (AST), capturing the temporal dependencies of variable declarations and equation systems. The field model processing channel deploys a graph neural network to reconstruct the physical field propagation path based on the discrete matrix of partial differential equations, identifying key field features such as temperature gradients and stress concentrations. This architecture innovatively introduces a tensor normalization layer to achieve cross-modal feature alignment, mapping the 2048-dimensional vector of geometric features, the 256-dimensional temporal encoding of syntactic features, and the 512-dimensional topological representation of the field model to a unified mathematical space. In the processing layer, semantic enhancement is achieved by embedding domain knowledge graphs, for example, transforming tolerance and fit rules in mechanical engineering into constraints in the feature space, effectively improving the physical rationality of parameter associations. In the feature fusion stage, a self-attention mechanism is used to dynamically adjust the contribution of each modality. The weight allocation formula integrates feature similarity measurement and domain expert experience, enabling the system to adapt to the needs of different application scenarios—for example, automatically strengthening mechanical connection features in automotive powertrain modeling, while prioritizing thermal field distribution patterns in electronic device heat dissipation simulation.

[0022] Example 2: like Figs. 1-3 As shown, this invention provides a method for intelligent extraction, recombination, and release of comprehensive elements in complex multivariate combination models. The main purpose of this method is to provide three technologies: Simulation code generation and intelligent verification technology: Deep Reinforcement Learning Code Generator (DRL-Coder) is an innovative code generation technology for modeling complex physical systems. This technology constructs a reinforcement learning framework with physical rule guidance capabilities to achieve intelligent generation and optimization of Modelica language code. The core architecture consists of a three-layer policy network: the upper-layer policy network parses the abstract syntax tree path based on the Transformer model and dynamically plans the component topology; the middle-layer policy network uses a bidirectional LSTM (an improved recurrent neural network structure) to capture the semantic context of the code and decide on the type of decision parameters; and the lower-layer policy network performs specific numerical optimization through a distributed reinforcement learning algorithm.

[0023] DRL-Coder innovatively embeds physical rule verification into a reinforcement learning reward mechanism, designing a three-dimensional reward function that includes syntactic correctness, semantic matching degree, and physical rationality. During training, the system verifies the dynamic behavior of the generated code in real time through a virtual simulation environment. If the code violates physical laws such as energy conservation and dimensional consistency, a negative reward-guided strategy network is triggered to self-correct.

[0024] To adapt to complex multidisciplinary scenarios, the deep reinforcement learning code generator employs a course-based learning strategy to progressively increase model complexity. Initially, it generates only basic models of single physical fields (such as thermodynamics), gradually transitioning to modeling multi-field coupled systems involving electro-thermal-mechanical systems. During training, a hybrid online-offline learning approach is implemented: the offline phase utilizes a historical codebase to pre-train the strategy network, while the online phase uses a digital twin sandbox for dynamic environment interaction. The technological advantages are prominent in three aspects: first, by hard-coding physical rules into the attention mechanism, it ensures that the generated code strictly adheres to domain knowledge constraints; second, it develops a syntax-guided action space for the Modelica language, mapping code generation actions to syntax tree node operations to avoid illegal code generation; and third, it constructs a value evaluation network with temporal memory capabilities, accurately predicting long-term code quality and guiding the generation process towards high-performance code. Compared to traditional methods, this technology achieves orders-of-magnitude improvements in key indicators such as multi-field coupling support, generation efficiency, and code reliability, providing a new generation of intelligent code generation solutions for the development of digital prototypes for complex systems.

[0025] Example 3: like Figs. 1-3 As shown, this invention provides a method for intelligent extraction, recombination, and release of comprehensive elements in complex multivariate combination models. The main purpose of this method is to provide three technologies. Digital prototype release and intelligent verification technology: Digital prototype release and intelligent verification technology is a reliable lifecycle assurance system for the digital delivery of complex engineering systems. This technology uses a digital fingerprint algorithm at its core, constructing a unique identifier based on quantum hashing to achieve tamper-proof storage of prototype versions, parameter configurations, and associated constraints. During the release phase, the system first performs multi-dimensional intelligent verification on the digital prototype: geometric topology consistency verification employs an improved ICP point cloud matching algorithm, achieving sub-millimeter accuracy detection by establishing a KD-Tree (K-dimensional tree) accelerated structure; boundary condition compliance detection integrates a rule engine to automatically verify engineering constraint indicators such as the maximum working pressure of pressure vessels and the critical speed of rotating components; and a real-time data-driven twin comparison module synchronizes the physical entity's operational data stream through standard protocols (such as OPC-UA), using a dynamic time warping algorithm to verify the similarity between the simulation model and the measured curves, ensuring a steady-state error ≤1.5% and a transient feature capture rate ≥95%.

[0026] Technological innovation is highlighted in three aspects: First, a lightweight digital fingerprint algorithm for multi-source heterogeneous data was developed, which reduces the feature extraction time of gigabit-level CAD models to within 30 seconds while maintaining encryption strength. Second, an interpretable intelligent verification framework was constructed. When a boundary condition violation is detected, the system automatically traces back to the associated Modelica equation or geometric assembly constraint and generates a visual correction suggestion. Third, a version management mechanism was introduced. Each released prototype generates a smart contract containing a timestamp, verification digest, and contributor signature, enabling traceable auditing of modification history.

[0027] The following points should be noted in this article: 1. The accompanying drawings of the embodiments disclosed herein only relate to the structures involved in the embodiments disclosed herein; other structures can be referred to in a general design.

[0028] 2. Where there is no conflict, the embodiments of this disclosure and the features in the embodiments can be combined with each other to obtain new embodiments.

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

Claims

1. A method for intelligent extraction, recombination, and release of comprehensive elements in complex multivariate combination models, characterized by: The method is implemented through the following technical steps: Step 1: Multi-source model data preparation and preprocessing. The *.mo file of the target system is parsed using the Modelica compiler to generate an AST syntax tree containing component declarations and equations; STEP format geometric data of the CAD model is imported, and mesh repair and feature edge extraction are performed using Paraview; Discretized matrix data of the field model is collected, and time series alignment and spatial coordinate normalization are performed. Step 2: Intelligent extraction of comprehensive elements, initiating a hybrid neural network hierarchical processing architecture; Geometric channel: Triangular patch data is input into the improved ResNet-50 network, and topological features such as assembly interference and fit tolerance are extracted through 3D convolution kernel (11×11×11); Syntax Channel: Traverse Modelica syntax tree nodes to identify variable scope, equation coupling relationships, and interface definitions; Field Model Channel: Deploy a graph attention network to analyze field strength distribution and boundary effects based on node gradient propagation; Step 3: Construct the element knowledge graph, creating four types of nodes in the graph database: Physical entity nodes (geometric feature extraction results), mathematical model nodes (Modelica analysis results), physical field nodes (multi-field model analysis results), and constraint rule nodes (domain knowledge base). Step 4: Simulation code generation and verification, calling DRL-Coder to perform layered code generation: The upper-level strategy network planning component topology generates the Modelica code framework; Mid-level strategy network population connect statement and parameter type declaration; The underlying policy network optimizes numerical parameters; Step 5: Dynamic model reorganization and optimization. A multi-objective optimization algorithm is adopted, with performance indicators (simulation speed), accuracy indicators (error with measured data), and resource consumption (memory usage) as optimization objectives. The reorganization scheme is iterated in the digital twin sandbox. The design variables are the element association weights and solver configuration parameters, and the constraints are mandatory clauses in the engineering rule base (such as a safety factor ≥ 2.5). Step 6: Standardized release of digital prototypes. Perform release preprocessing on the validated recombinant model: Generate a digital fingerprint, which includes the model hash value, version number, and contributor signature; Automatically package into PMU standard format and construct PMI interface description XML file; Generate a C language interface wrapper layer; Step 7: Smart Verification and Deployment. Perform the following verification process in the target deployment environment: 1: Geometric consistency confirmation; 2: Interface compatibility testing; 3: Real-time verification; 4. Resource integrity check; 5: Traceability Verification 6: User confirmation.

2. The method for intelligent extraction, recombination, and release of comprehensive elements in a complex multivariate combination model according to claim 1, characterized in that: In the multi-source model data preparation and preprocessing steps, the preprocessed data is stored in a multimodal database in HDF5 format, and the physical dimensions and engineering constraint labels of each element are labeled.

3. The method for intelligent extraction, recombination, and release of comprehensive elements in a complex multivariate combination model according to claim 1, characterized in that: In the comprehensive element intelligent extraction step, the feature fusion stage performs tensor normalization and injects domain knowledge rules to output an element list with confidence rating.

4. The method for intelligent extraction, recombination, and release of comprehensive elements in a complex multivariate combination model according to claim 1, characterized in that: In the steps of constructing the element knowledge graph, edge relation weights are calculated to establish an association network containing semantics such as "spatial assembly", "energy transfer", and "data dependency".

5. The method for intelligent extraction, recombination, and release of comprehensive elements in a complex multivariate combination model according to claim 1, characterized in that: In the simulation code generation and verification steps, the generated code triggers triple verification in real time: syntax verification calls the Modelica compiler, physical verification performs dimensional balance calculations, and functional verification uses an automated verification framework to perform test case coverage analysis.

6. The method for intelligent extraction, recombination, and release of comprehensive elements in a complex multivariate combination model according to claim 1, characterized in that: In the dynamic reorganization optimization step of the model, a Pareto front solution set is generated in each iteration, and the optimal reorganization scheme is selected by the Top-Optimal Solution Ranking (TOPSIS) decision method.

Images