Modular division method for vehicle open electronic and electrical architecture based on MBSE and MOSA fusion

Abstract

This invention provides a modular partitioning method for open vehicle electronic and electrical architecture based on the integration of MBSE and MOSA. It belongs to the interdisciplinary technical fields of vehicle electronic and electrical architecture, system integration, and model-based systems engineering (MBSE). The method includes: S1: modular requirement layering and weight assignment; S2: function-structure architecture mapping modeling; S3: element-link semantic enhancement annotation; S4: dynamic construction of design structure matrix; S5: multi-constraint priority clustering analysis; S6: constraint conflict resolution and module optimization. The core objective of this invention is to solve the core problems in existing vehicle electronic and electrical architecture design: separation of modular design and evaluation, insufficient integration of MBSE and MOSA, and difficulty in resolving constraint conflicts.

Description

Technical Field

[0001] This invention relates to the interdisciplinary fields of vehicle electronic and electrical architecture, system integration, and model-based systems engineering (MBSE), and in particular to a modular partitioning method for an open electronic and electrical architecture of vehicles based on the integration of MBSE and MOSA. Background Technology

[0002] As the automotive industry undergoes a profound transformation towards electrification, intelligence, and connectivity, the drawbacks of traditional distributed electronic and electrical architecture are becoming increasingly apparent: deep coupling between hardware and software leads to system rigidity, making it unable to adapt to rapidly iterating intelligent driving algorithms and in-vehicle software; upgrades and maintenance are difficult, as updating a single component requires adjustments to multiple related systems, resulting in extremely high costs for later modifications; and the supply chain is singular, with a closed architecture design limiting the compatibility and replacement of components from different suppliers, resulting in weak risk resistance.

[0003] To address these challenges, the Modular Open Systems Approach (MOSA) and Model-Based Systems Engineering (MBSE) have become industry focuses. MOSA enhances system flexibility through open standards and modular design, enabling cost savings and rapid deployment of new technologies. MBSE, on the other hand, integrates requirement, structure, behavior, and parameter models to form an authoritative information source, supporting the development of complex systems. However, in existing technologies, the two have not achieved deep integration: on the one hand, existing MBSE methods do not systematically integrate modular design, analysis, and evaluation into the architecture development process, resulting in the separation and post-implementation of modular design and evaluation. Early in the architecture design phase, it is impossible to quantitatively evaluate and optimize the degree of modularity, making it difficult to balance functional requirements with maintainability, upgradeability, and other lifecycle value attributes. On the other hand, traditional modular partitioning methods rely on manual experience, failing to form a closed-loop process of "requirements-model-algorithm-constraints-evaluation." Furthermore, clustering algorithms are often general-purpose and do not incorporate engineering constraints of vehicle electronic and electrical architecture (such as supply chain, thermal management, and upgradeability), resulting in insufficient practicality of the modular partitioning results and failing to meet the core partitioning requirements of open and modular architecture.

[0004] Based on this, the industry urgently needs a method that can deeply integrate the MOSA design concept with the modeling and integration advantages of MBSE, and realize modular design, quantitative evaluation and optimization closed-loop iteration in the early stage of architecture design, so as to efficiently generate highly modular and open vehicle electronic and electrical architecture partitioning schemes. Summary of the Invention

[0005] This invention provides a modular partitioning method for an open electronic and electrical architecture of vehicles based on the integration of MBSE and MOSA. The core purpose is to solve the core problems in the existing vehicle electronic and electrical architecture design, namely, "separation of modular design and evaluation, insufficient integration of MBSE and MOSA, and difficulty in resolving constraint conflicts".

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A modular partitioning method for an open electronic and electrical architecture for vehicles based on the fusion of MBSE and MOSA includes: S1: Collect technical specifications from vehicle manufacturers, adaptation requirements from component suppliers, end-user feedback, industry technology upgrade trend documents, and policy and regulatory compliance standards, summarize them into an original requirements list, classify them into functional and non-functional requirements, calculate the weight value of each requirement, and generate a modular requirements hierarchy table. S2: Based on the modular requirements layering table, construct the functional architecture and structural architecture model in the MBSE tool, define all system elements to be modularized and the links between elements, clarify the core attributes of elements and links, and generate the structural architecture model; S3: Import the SysML modular configuration file, which is independently constructed based on the vehicle modular domain ontology, add semantic attributes to system elements and links, establish the traceability relationship between modular requirements and elements and links, and generate a semantically enhanced structural architecture model. S4: Extract data from the semantically enhanced structural architecture model, fill in the missing data using the sparse Bayesian matrix filling algorithm, and then fuse the requirement weight values ​​using the dynamic weight coupling matrix decomposition algorithm to construct the dynamic design structure matrix; S5: Collect actual engineering constraints, calculate constraint compatibility coefficients through multi-constraint collaborative optimization algorithm, combine with dynamic design structure matrix, use improved gravity search clustering algorithm for clustering, and generate initial module partitioning list and modularity index report; S6: Review the initial module partitioning list and modularity index report, identify constraint conflicts and formulate resolution strategies, rerun the improved gravity search clustering algorithm and multi-constraint collaborative optimization algorithm, generate an optimized module partitioning list, and complete the modular partitioning.

[0007] In this specification, in S4, a sparse Bayesian matrix is ​​used to construct a prior distribution model and a posterior distribution model. The missing values ​​in the original data are filled in by iteratively optimizing the accuracy parameters and noise variance. During the iteration process, the parameter change is less than the convergence threshold as the stopping condition, and finally a filled matrix with an average error ≤5% is generated.

[0008] In this specification, in S4, the dynamic weighted coupled matrix decomposition algorithm decomposes the filling matrix into a product of row feature matrix, dynamic weight matrix and column feature matrix. The dynamic weight matrix is ​​dynamically adjusted by the demand weight value. At the same time, an L1 regularization term is introduced to avoid overfitting and constrain the reconstruction error of the matrix decomposition to ≤0.01. Finally, the dynamic design structure matrix that integrates demand priorities is output.

[0009] In this specification, in S4, the sparse Bayesian matrix filling algorithm and the dynamic weighted coupled matrix decomposition algorithm form a two-way interactive mechanism: the filling matrix output by the sparse Bayesian matrix filling algorithm serves as the input data for the dynamic weighted coupled matrix decomposition algorithm, determining the initial feature distribution of the row feature matrix and the column feature matrix; the row feature matrix and the column feature matrix obtained by the dynamic weighted coupled matrix decomposition algorithm are fed back to the sparse Bayesian matrix filling algorithm, dynamically adjusting its accuracy parameter update formula, so that the accuracy of filling missing values ​​in the filling matrix matches the element feature distribution better, and finally achieving collaborative optimization of the two sets of algorithms, with the element value error of the dynamically designed structure matrix ≤3%.

[0010] In this specification, in S5, the improved gravity search clustering algorithm introduces a constraint compatibility coefficient as a gravity adjustment factor. It calculates the gravity value between system elements through the mechanism of gravity constant decaying with iteration, and updates the acceleration, velocity and position of elements by combining element mass and Euclidean distance. Iteratively, it achieves a clustering effect of high cohesion within modules and low coupling between modules. The number of clusters is determined by the elbow rule.

[0011] In this specification, in S5, the multi-constraint collaborative optimization algorithm constructs an objective function that includes constraint satisfaction and matrix smoothness. It assigns weights according to the priority of necessary decoupling constraints, supply chain constraints, thermal management constraints, upgradeability constraints, necessary coupling constraints, and maintenance cycle constraints. The optimal constraint compatibility coefficient is solved by a sequential quadratic programming algorithm, thereby improving the constraint satisfaction to over 98%.

[0012] In this specification, in S5, the improved gravity search clustering algorithm and the multi-constraint collaborative optimization algorithm form a two-way interaction mechanism: the constraint compatibility coefficient output by the multi-constraint collaborative optimization algorithm is injected into the gravity calculation model of the improved gravity search clustering algorithm as a gravity adjustment factor to correct the gravity values ​​between elements; the clustering results of the improved gravity search clustering algorithm are fed back to the multi-constraint collaborative optimization algorithm to dynamically adjust the priority weights of various constraints, so that the weights of high-satisfaction constraints are further increased and the weights of low-satisfaction constraints are adaptively reduced. Through interaction, the dual optimization of constraint satisfaction and cluster rationality is achieved, and finally the constraint satisfaction rate is ≥98% and the average interaction strength of elements within the cluster is ≥4.0.

[0013] In this specification, during S6, when rerunning the improved gravitational search clustering algorithm and the multi-constraint collaborative optimization algorithm, the initial modularity index output from S5 is introduced as a feedback parameter for algorithm iteration: if the initial modularity index is ≥0.65, the decay rate of the initial gravitational constant of the improved gravitational search clustering algorithm is reduced to minimize significant changes in the clustering structure and preserve reasonable partitioning logic; if the initial modularity index is <0.65, the initial weight ratio of necessary decoupling constraints and supply chain constraints in the multi-constraint collaborative optimization algorithm is increased to prioritize the satisfaction of core constraints; at the same time, the temporary modularity index is calculated in real time during the algorithm iteration process, and the iteration is automatically stopped when the difference between the temporary indices of two consecutive iterations is <0.01, thereby improving the efficiency of re-clustering by more than 20%, and the optimized modularity index is stable above 0.6.

[0014] In this specification, in S4, the interaction process between the sparse Bayesian matrix filling algorithm and the dynamic weighted coupled matrix decomposition algorithm also includes a dynamic parameter adjustment mechanism: when the reconstruction error of the dynamic weighted coupled matrix decomposition algorithm exceeds the preset threshold of 0.01, the accuracy parameter of the sparse Bayesian matrix filling algorithm is automatically optimized by increasing the number of iterations and adjusting the convergence threshold to optimize the filling matrix; when the filling error of the sparse Bayesian matrix filling algorithm is less than 3%, it is fed back to the dynamic weighted coupled matrix decomposition algorithm to reduce the regularization coefficient to improve the expressive power of the feature matrix. Through this dynamic parameter adjustment, the element value error of the dynamically designed structure matrix is ​​further reduced to ≤2%.

[0015] In this specification, in step S5, the constraint priority weight update of the multi-constraint collaborative optimization algorithm depends on the clustering result feedback of the improved gravity search clustering algorithm: if the satisfaction rate of a certain type of constraint in the clustering result is less than 95%, the priority weight of that type of constraint is automatically increased by 0.05; if the satisfaction rate of a certain type of constraint is ≥99% for two consecutive iterations, its priority weight is appropriately reduced by 0.02, while keeping the priority weights of necessary decoupling constraints and supply chain constraints no less than 0.25. Through dynamic iterative optimization of constraint weights, the rationality of clustering and the satisfaction rate of constraints are improved in both directions, and finally the average interaction strength of elements within the cluster is ≥4.2 and the constraint satisfaction rate is ≥99%.

[0016] In summary, the present invention has at least the following beneficial effects: Modular design and evaluation are integrated and implemented early: By embedding modular design principles and quantitative evaluation algorithms into the entire MBSE modeling and iteration process, modularization is transformed from a late-stage evaluation activity into a core activity throughout the architecture design cycle. The modularity index can be quantified early in the architecture design process (eventually stabilizing above 0.6), and modular defects can be identified in advance, reducing the cost of later changes by more than 40% compared to traditional methods.

[0017] Modular decision-making is scientifically traceable: Quantitative analysis is achieved based on the dynamic design structure matrix and two sets of collaborative algorithms (SBMP+DWCMF, IGSA-C+MCOA). All design decisions, constraints and their impact on modular indicators are recorded in the system model. The proportion of decision-making relying on human experience is reduced by 70%, and the traceability and reproducibility of decisions reach 100%.

[0018] Deep balance between functional and non-functional requirements: Through the "interactive evaluation and constraint injection" and "multi-constraint collaborative optimization" processes, non-functional requirements such as supply chain, thermal management, and upgradeability are integrated into the module partitioning algorithm in the form of engineering constraints. The P1 level constraint satisfaction rate reaches 100%, and the P2 level constraint satisfaction rate reaches 97%, achieving a precise balance between functional optimization and the full life cycle value target.

[0019] Design efficiency and partitioning quality are significantly improved: automated data extraction and analysis avoid tedious and error-prone manual operations; closed-loop optimization within the core partitioning process supports rapid exploration of multiple architectural alternatives; a unified model ensures consistency of requirements, design, analysis, and evaluation information; and the modular partitioning cycle is shortened by more than 30%. The final output optimized module partitioning list is logically clear, with an average module cohesion of ≥4.0 and an average inter-module coupling of ≤1.5, fully meeting the core requirements of modular partitioning for open electronic and electrical architectures in vehicles, and providing a high-quality foundation for subsequent interface design, architecture implementation, and other stages. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the modular partitioning method for an open electronic and electrical architecture of a vehicle based on the fusion of MBSE and MOSA, as described in this invention.

[0021] Figure 2 This is a schematic diagram illustrating the modular partitioning process of the open electronic and electrical architecture for vehicles based on the fusion of MBSE and MOSA involved in this invention.

[0022] Figure 3 This is a schematic diagram of the interaction process between SBMP and DWCMF involved in this invention.

[0023] Figure 4 This is a schematic diagram of the interaction process between IGSA-C and MCOA involved in this invention. Detailed Implementation

[0024] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0025] like Figure 1 As shown, this embodiment provides a modular partitioning method for an open electronic and electrical architecture for vehicles based on the fusion of MBSE and MOSA, as detailed below: S1: Modular Requirements Layering and Weight Assignment; Requirements are the source of architecture design. The core objective of this step is to transform the scattered requirements of stakeholders into a structured and quantifiable basis for modular design, providing clear guidance for subsequent architecture modeling, constraint setting, and module evaluation.

[0026] 1. Requirements Acquisition: Comprehensive collection of all requirements to ensure coverage of all dimensions of vehicle electronic and electrical architecture design. Specific requirements sources include: technical specifications from OEMs, clearly defining the performance indicators and integration requirements the architecture must meet; adaptation requirements from component suppliers, involving key parameters such as interface standards and installation dimensions; user needs raised by end-users through surveys, test drive feedback, etc., such as ease of operation and functional expandability; industry technology upgrade trend documents published by professional organizations, predicting future (e.g., 3-5 years) technological evolution directions; and policy, regulatory, and compliance standards, clarifying mandatory safety and environmental requirements the architecture must meet. All requirements are aggregated and organized through a unified requirements collection platform to form an original requirements list, which is then checked one by one to ensure no requirements are omitted.

[0027] 2. Requirements Hierarchy and Classification: A two-tier classification method of "functional requirements - non-functional requirements" is used to systematically decompose the original requirements, ensuring clear and comprehensive classification logic. Functional requirements focus on the realization of core vehicle functions, specifically including power output control requirements, responsible for the distribution and adjustment of vehicle power; intelligent driving perception and decision-making requirements, supporting vehicle environmental perception, path planning, and decision execution; vehicle body state adjustment requirements, covering the control of body-related functions such as door and window control and seat adjustment; infotainment interaction requirements, including user interaction functions such as audio and video playback and navigation services; and chassis attitude control requirements, involving the control of chassis systems such as braking, steering, and suspension adjustment. Non-functional requirements focus on the performance of the architecture throughout its entire lifecycle, specifically including upgradeability requirements, ensuring the architecture can adapt to future technology iterations; maintainability requirements, facilitating later troubleshooting and component replacement; supply chain adaptability requirements, meeting the compatibility and replacement of components from different suppliers; thermal management adaptability requirements, ensuring the stable operation of the architecture under different temperature environments; vibration isolation requirements, reducing the impact of vibration on precision components; cost control requirements, controlling the architecture design and manufacturing costs; and lifecycle adaptability requirements, specifying that the architecture must meet the requirement of a service life of more than 5 years.

[0028] 3. Weight Assignment Calculation: To quantify the importance of different requirements, the Analytic Hierarchy Process (AHP) is used for weight assignment. First, a requirement weight judgment matrix is ​​constructed, and the importance of each requirement is compared pairwise. A consistency check is performed to ensure the rationality of the judgment matrix; the check criterion is CR ≤ 0.1. When this condition is met, the matrix has logical consistency and subsequent weight calculations can proceed. The final weight values ​​for each requirement were calculated. Intelligent driving perception and decision-making requirement has a weight of 0.22, ranking as the highest priority as a core functional requirement; upgradeability requirement has a weight of 0.18, reflecting the architecture's core need to cope with technological iterations; power output control requirement has a weight of 0.15, ensuring basic vehicle driving functions; supply chain adaptability requirement has a weight of 0.12, supporting supply chain flexibility; thermal management adaptability requirement has a weight of 0.10, ensuring the architecture's operational stability; vehicle body state adjustment requirement has a weight of 0.08, meeting basic user comfort needs; infotainment interaction requirement has a weight of 0.07, improving user experience; chassis attitude control requirement has a weight of 0.06, ensuring driving safety and stability; maintainability requirement has a weight of 0.05, reducing total lifecycle costs; vibration isolation requirement has a weight of 0.03, protecting precision components; cost control requirement has a weight of 0.02, balancing functionality and cost; and lifecycle adaptability requirement has a weight of 0.02, ensuring the long-term effectiveness of the architecture.

[0029] 4. Output: Generates a modular requirement hierarchy table, which records in detail the name, category, weight value and corresponding source basis of each requirement, clearly presenting the structured and quantitative results of the requirements.

[0030] S2: Function-Structure Architecture Mapping Modeling; This step builds upon the structured requirements output from S1 to construct a mapping relationship from function to structure, transforming abstract requirements into concrete architectural model elements, laying the foundation for subsequent semantic annotation and matrix construction.

[0031] 1. Functional Architecture Construction: Based directly on the modular requirement hierarchy table of S1, the top-level functional architecture is constructed using SysML activity diagrams in the MBSE tool to ensure a one-to-one correspondence between functional domains and requirements, achieving precise implementation from requirements to functions. Specifically, the functional domains include: the powertrain functional domain, corresponding to power output control requirements and responsible for the functional integration of the powertrain system; the intelligent driving functional domain, corresponding to intelligent driving perception and decision-making requirements, integrating environmental perception, decision planning, and control execution functions; the body control functional domain, corresponding to body state adjustment requirements, coordinating the control logic of body functions such as doors, windows, and seats; the infotainment functional domain, corresponding to infotainment interaction requirements, integrating audio, video, navigation, and other interactive functions; and the chassis control functional domain, corresponding to chassis attitude control requirements, coordinating chassis functions such as braking, steering, and suspension. Each functional domain is further broken down into three levels of sub-functions. Taking the intelligent driving functional domain as an example, the first-level function is broken down into the perception and acquisition sub-function, the decision-making and planning sub-function, and the control and execution sub-function. The perception and acquisition sub-function is further broken down into second-level sub-functions such as lidar acquisition, millimeter-wave radar acquisition, and camera acquisition. The second-level sub-functions are further broken down into third-level sub-functions such as signal reception and data preprocessing. At the same time, the data flow interaction relationship between each level of sub-function is clearly defined to ensure the continuity of functional logic.

[0032] 2. System Element Definition: Based on the three-level sub-function decomposition results of the functional architecture, define all "system elements" to be modularized. Each system element corresponds to a specific sub-function implementation to ensure accurate mapping between functions and elements. The system elements specifically include: an AI computing chip responsible for the computation and data processing of intelligent driving algorithms; a safety MCU, responsible for executing vehicle safety-related control logic; a LiDAR sensor interface chip, enabling the transmission and conversion of LiDAR data; a millimeter-wave radar sensor interface chip, responsible for the interaction of millimeter-wave radar data; a camera sensor interface chip, processing data collected by the camera; an ultrasonic sensor interface chip, enabling the transmission of ultrasonic sensor data; a power management chip, coordinating the power distribution and management of the architecture; DDR5 memory, providing data storage support for the AI ​​computing chip, etc.; a PCIe 5.0 high-speed bus switch, responsible for high-speed data interaction and forwarding; a brake control actuator, executing brake control commands; a steering control actuator, responding to steering control needs; a suspension adjustment actuator, realizing the adjustment of suspension attitude; door and window control actuators, controlling the opening and closing of doors and windows; a seat adjustment actuator, adjusting the position and attitude of the seat; an air conditioning control module, managing the operation of the vehicle's air conditioning system; a vehicle display control chip, controlling the content output of the vehicle's display screen; a navigation and positioning module, providing accurate navigation and positioning services; a Bluetooth communication module, enabling short-range wireless communication; and a 5G communication module, supporting high-speed long-range wireless communication. Each system element is labeled with a unique identification ID, such as the AI ​​computing chip ID being SYE-001, the security MCU ID being SYE-002, and so on, to ensure that each system element can be uniquely identified and that no element is omitted.

[0033] 3. Element Link Definition: Based on the data flow and interaction logic between sub-functions, define the "links" between system elements, clarifying the interaction methods, strengths, and key parameters between each element, providing detailed interaction data for subsequent semantic annotation and matrix construction. The link types and corresponding attributes are as follows: PCIe bus connection: Primarily used for high-speed data transmission scenarios, the interaction objects include AI computing chips and DDR5 memory, and AI computing chips and PCIe 5.0 high-speed bus switches. The interaction strength is set at 5 points, which is the strongest among the 1-5 strength levels, reflecting the requirement of this type of connection for high-speed and high-reliability data transmission; the data transmission rate reaches 32GB / s, meeting the needs of rapid exchange of large amounts of data between AI computing chips and memory, and bus switches; the reliability level is 3, the highest among the 1-3 reliability levels, ensuring the stability of core data transmission.

[0034] Ethernet connectivity: Suitable for medium-to-high-speed data transmission scenarios, interacting with PCIe 5.0 high-speed bus switches and LiDAR sensor interface chips, and PCIe 5.0 high-speed bus switches and camera sensor interface chips. The interaction strength is set to 4 points, indicating high-intensity interaction; the data transmission rate is 10GB / s, meeting the high-speed transmission requirements of sensor data; the reliability level is 3, ensuring the accuracy of sensor data transmission.

[0035] CANFD bus connection: Primarily used for vehicle control data transmission. Interaction objects include the safety MCU with brake control actuators, the safety MCU with steering control actuators, and the safety MCU with suspension adjustment actuators. The interaction strength is set to 3 points, representing medium-intensity interaction; the data transmission rate is 10Mbps, meeting the real-time transmission requirements of control commands; the reliability level is 3, ensuring stable transmission of control commands and preventing control failures due to data loss.

[0036] Power Line Connection: Provides power to all system elements. The interaction objects are the power management chip and all other system elements. The interaction strength is set to 5 points, which is the strongest interaction, because power supply is the foundation for the normal operation of all elements; there is no data transmission rate attribute; the reliability level is 3, ensuring a continuous and stable power supply and avoiding system failure due to power outages.

[0037] Data stream interaction links: used for non-real-time data stream transmission. Interaction objects include the navigation and positioning module and the vehicle display control chip, the Bluetooth communication module and the vehicle display control chip, and the 5G communication module and the AI ​​computing chip. The interaction intensity is set at 2 points, belonging to low-to-medium intensity interaction; the data transmission rate is 1Mbps, meeting the transmission requirements of ordinary data streams; the reliability level is 2, ensuring basic stability of data transmission.

[0038] Mechanical fixing link: Used for physical fixing and heat dissipation between elements. The interaction objects are power management chip and thermal management heat sink, and AI computing chip and thermal management heat sink. The interaction intensity is set to 3 points, which is a medium intensity interaction; there is no data transmission rate attribute; the reliability level is 2, ensuring the sturdiness of the physical fixing and the effectiveness of heat dissipation.

[0039] 4. Output: Generates a functional architecture diagram and a structural architecture model. The functional architecture diagram clearly presents the hierarchical relationship of the three-level sub-functions and the data flow interaction path; the structural architecture model includes a system element list and a link list. The system element list clearly defines the ID, name, and corresponding function of each element, while the link list records in detail the interaction object, type, strength, data transmission rate, and reliability level of each link.

[0040] S3: Element-Link Semantic Enhancement Annotation; This step introduces a unified semantic annotation standard to semantically enhance the structural architecture model constructed in S2, clarifying the attribute characteristics of elements and links and their traceability relationship with requirements, providing data support rich in semantic information for the accurate construction of the subsequent design structure matrix.

[0041] 1. Import of Annotation Standards: A pre-developed SysML modular configuration file is introduced. This configuration file is independently constructed based on the vehicle modular ontology. The ontology system has been verified through extensive engineering practice and can accurately cover the core semantic requirements of modular design of vehicle electronic and electrical architecture. The ontology contains four core semantic classes: system elements, links, modules, and modular requirements. The system element semantic class defines the core attribute dimensions of elements, the link semantic class clarifies the key feature descriptions of interactions, the module semantic class standardizes the definition and attributes of modules, and the modular requirement semantic class establishes the association rules between requirements and elements and links. This configuration file provides a unified and standardized semantic standard for subsequent semantic annotation, ensuring the consistency and usability of annotation results.

[0042] 2. System Element Semantic Annotation: Based on the system element list output by S2, semantic attributes are added to each system element in the attribute panel of the MBSE tool. The value of each attribute is determined by considering the element's functional characteristics, engineering design requirements, and actual supply chain conditions to ensure the accuracy and practicality of the attribute annotation. The specific semantic attributes of each system element are as follows: AI computing chip (SYE-001): Manufacturer A has a technological advantage in the high-end chip field; maintenance cycle is 5 years, determined by the chip's technology update cycle and usage scenarios; technology update rate is once a year, in line with the industry trend of rapid iteration of intelligent driving chips; power consumption level is high (300W), determined by the chip's computing power and process; temperature sensitivity is high (≤85℃), exceeding this temperature will affect chip performance; supply chain priority is P1 (highest), ensuring the stability of the core chip supply.

[0043] Safety MCU (SYE-002): Manufacturer B has extensive experience in the automotive safety MCU field; maintenance cycle is 8 years, determined based on the MCU's operating environment and reliability design; technology update rate is once every 3 years, which is relatively low for safety chips; power consumption level is low (10W), meeting the low power consumption requirements of safety control modules; temperature sensitivity is medium (≤105℃), with a wide temperature adaptability range; supply chain priority is P1, ensuring the supply of safety-related components.

[0044] LiDAR sensor interface chip (SYE-003): Manufacturer is C, which focuses on the research and development of sensor interface chips; the maintenance cycle is 6 years, which is determined in combination with the life cycle of the sensor; the technology update rate is once every 2 years, following the iteration rhythm of LiDAR technology; the power consumption level is medium (20W), balancing performance and power consumption; the temperature sensitivity is medium (≤105℃), adapting to different vehicle usage environments; the supply chain priority is P2.

[0045] Millimeter-wave radar sensor interface chip (SYE-004): Manufacturer is C, from the same supplier as the lidar sensor interface chip, facilitating supply chain management; maintenance cycle is 6 years, consistent with the lidar sensor interface chip; technology update rate is once every 2 years, following the development of millimeter-wave radar technology; power consumption level is medium (20W), comparable to the lidar sensor interface chip; temperature sensitivity is medium (≤105℃); supply chain priority is P2.

[0046] Camera sensor interface chip (SYE-005): Manufacturer is Manufacturer D, with mature technology in the field of image sensor interfaces; maintenance cycle is 6 years, matching the usage cycle of cameras; technology update rate is once every 2 years, following the iteration of camera technology; power consumption level is medium (25W), due to slightly higher image data processing requirements than other sensor interface chips; temperature sensitivity is medium (≤105℃); supply chain priority is P2.

[0047] Ultrasonic sensor interface chip (SYE-006): Manufacturer is Manufacturer D, from the same supplier as the camera sensor interface chip; maintenance cycle is 6 years, matching the usage cycle of the ultrasonic sensor; technology update rate is once every 2 years; power consumption level is low (15W), due to the small data volume of the ultrasonic sensor and low power consumption requirements; temperature sensitivity is medium (≤105℃); supply chain priority is P3.

[0048] Power management chip (SYE-007): Manufacturer is E, with high reliability in the automotive power management field; maintenance cycle is 8 years, based on the core position and reliability design of the power management chip; technology update rate is once every 3 years, and power technology updates are relatively stable; power consumption level is medium (50W), meeting the overall power management requirements; temperature sensitivity is low (≤125℃), with strong temperature adaptability; supply chain priority is P1, ensuring overall power supply stability.

[0049] DDR5 memory (SYE-008): Manufacturer A, from the same supplier as the AI ​​computing chip, ensuring better compatibility; maintenance cycle of 5 years, matching the update cycle of the AI ​​computing chip; technology update rate once a year, keeping pace with the rapid iteration of memory technology; power consumption level of medium (30W), balancing capacity and power consumption; temperature sensitivity of medium (≤95℃); supply chain priority of P2.

[0050] PCIe 5.0 High-Speed ​​Bus Switch (SYE-009): Manufacturer is F, a leader in high-speed bus technology; maintenance cycle is 6 years, based on the operating environment and reliability of bus switches; technology update rate is once every 2 years, keeping pace with bus technology development; power consumption level is medium (40W), meeting the power consumption requirements of high-speed data transmission; temperature sensitivity is medium (≤105℃); supply chain priority is P2.

[0051] Brake control actuator (SYE-010): Manufacturer is G, which has extensive experience in the field of automotive braking systems; maintenance cycle is 8 years, based on the core safety position of the braking system; technology update rate is once every 4 years, the brake actuator technology is mature and the update frequency is low; power consumption level is low (15W); temperature sensitivity is low (≤125℃); supply chain priority is P1 to ensure braking safety.

[0052] Steering control actuator (SYE-011): Manufacturer is G, which is from the same supplier as the brake control actuator; maintenance cycle is 8 years, matching the service life of the steering system; technology update rate is once every 4 years; power consumption level is low (15W); temperature sensitivity is low (≤125℃); supply chain priority is P1 to ensure steering safety.

[0053] Suspension Adjustment Actuator (SYE-012): Manufacturer is Manufacturer H, which has professional technology in the field of suspension systems; maintenance cycle is 8 years, based on the usage environment of the suspension system; technology update rate is once every 4 years; power consumption level is medium (25W), which meets the power requirements of suspension adjustment; temperature sensitivity is low (≤125℃); supply chain priority is P2.

[0054] Door and window control actuator (SYE-013): Manufacturer is Manufacturer I, offering high cost-performance in the field of vehicle body actuators; maintenance cycle is 10 years, the working environment of door and window actuators is relatively mild, resulting in a long lifespan; technology update rate is once every 5 years; power consumption level is low (5W), with extremely low power consumption requirements; temperature sensitivity is low (≤125℃); supply chain priority is P3.

[0055] Seat adjustment actuator (SYE-014): Manufacturer is Manufacturer I, from the same supplier as the door and window control actuator; maintenance cycle is 10 years, matching the lifespan of the seat; technology update rate is once every 5 years; power consumption level is low (8W), slightly higher than the door and window control actuator; temperature sensitivity is low (≤125℃); supply chain priority is P3.

[0056] Air Conditioning Control Module (SYE-015): Manufacturer is J, with mature technology in the field of automotive air conditioning control; maintenance cycle is 8 years, based on the usage frequency and reliability of the air conditioning system; technology update rate is once every 3 years; power consumption level is medium (60W), with higher power consumption due to the cooling and heating requirements of the air conditioning system; temperature sensitivity is low (≤125℃); supply chain priority is P2.

[0057] Vehicle infotainment display control chip (SYE-016): Manufacturer is K, which has advantages in the field of display control chips; maintenance cycle is 5 years, following the update cycle of the vehicle infotainment system; technology update rate is once a year, and vehicle infotainment display technology iterates quickly; power consumption level is medium (35W), which meets the performance requirements of display control; temperature sensitivity is medium (≤95℃); supply chain priority is P3.

[0058] Navigation and Positioning Module (SYE-017): Manufacturer is L, which has a high-precision advantage in the field of navigation and positioning; maintenance cycle is 5 years, matching the update cycle of navigation technology; technology update rate is once a year; power consumption level is low (10W); temperature sensitivity is low (≤125℃); supply chain priority is P3.

[0059] Bluetooth communication module (SYE-018): Manufacturer is L, the same supplier as the navigation and positioning module; maintenance cycle is 5 years, following the updates of wireless communication technology; technology update rate is once a year; power consumption level is low (5W), Bluetooth communication power consumption is extremely low; temperature sensitivity is low (≤125℃); supply chain priority is P4.

[0060] 5G Communication Module (SYE-019): Manufacturer is Manufacturer M, a leader in 5G communication technology; maintenance cycle is 5 years, following the iteration rhythm of 5G technology; technology update rate is once a year; power consumption level is medium (20W), meeting the power consumption requirements of high-speed communication; temperature sensitivity is low (≤125℃); supply chain priority is P2.

[0061] 3. Link Semantic Annotation: Based on the link list output by S2, semantic attributes are added to each link individually. Attribute values ​​are consistent with the link characteristics defined in S2 to ensure data continuity and consistency. Link semantic attributes include: interaction type attribute (using the six categories defined in S2: PCIe bus connection, Ethernet connection, CANFD bus connection, power line connection, data flow interaction link, and mechanical fixed link); interaction strength value attribute (using the 1-5 score range set in S2); data transmission rate attribute (using the specific values ​​set in S2, with links without data transmission marked as "none"); and reliability level attribute (using the 1-3 level range set in S2). Semantic annotation further clarifies the core characteristics of the links, providing a basis for numerical calculations in subsequent matrix construction.

[0062] 4. Modular Requirements Traceability and Labeling: Utilizing the traceability function of the MBSE tool, establish the association between each requirement in the modular requirements hierarchy table output by S1 and its corresponding system elements or links, achieving traceability from requirements to design. Specific associations are as follows: Intelligent driving perception and decision-making requirements are associated with AI computing chips, LiDAR sensor interface chips, millimeter-wave radar sensor interface chips, camera sensor interface chips, ultrasonic sensor interface chips, and their corresponding PCIe bus connections and Ethernet connections; Upgradeability requirements are associated with AI computing chips, DDR5 memory, vehicle display control chips, navigation and positioning modules, Bluetooth communication modules, 5G communication modules, and their corresponding PCIe bus connections and data flow interaction links; Power output control requirements are associated with power management chips, brake control actuators, steering control actuators, and their corresponding power line connections and CANFD bus connections; Supply chain adaptation requirements are associated with all system elements and their corresponding links, with a focus on elements and links from the same manufacturer; Thermal management adaptation requirements are associated with AI computing chips, power management chips, air conditioning control modules, and their corresponding mechanical fixing links and power lines. The system is designed to meet various needs, including: road connectivity; vehicle body status adjustment requirements (connecting door and window control actuators, seat adjustment actuators, suspension adjustment actuators, and corresponding CANFD bus connections and power line connections); infotainment interaction requirements (connecting vehicle display control chips, navigation and positioning modules, Bluetooth communication modules, 5G communication modules, and corresponding data stream interaction links); chassis attitude control requirements (connecting brake control actuators, steering control actuators, suspension adjustment actuators, and corresponding CANFD bus connections); maintainability requirements (connecting all system elements and their corresponding links, with a focus on elements with short maintenance cycles); vibration isolation requirements (connecting precision components such as safety MCUs and LiDAR sensor interface chips, and their corresponding mechanical fixing links); cost control requirements (connecting all system elements and their corresponding links, with a focus on high-power, high-cost elements); and lifecycle adaptation requirements (connecting all system elements and their corresponding links to ensure long-term stable operation of the architecture).

[0063] 5. Output: Generate a semantically enhanced structural architecture model, which includes a system element annotation dataset, a link annotation dataset, and a requirement-element / link traceability table. The system element annotation dataset records the ID, name, and all semantic attributes of each element in detail; the link annotation dataset clarifies the semantic attributes of each link, such as the interaction object, type, strength, data transmission rate, and reliability level; the requirement-element / link traceability table clearly presents the elements and links corresponding to each requirement.

[0064] S4: Dynamic Construction of Design Structure Matrix; The core objective of this step is to transform the semantically enhanced data output from S3 into a structured design structure matrix. By introducing two sets of collaborative algorithms, the problem of missing data is solved and the requirement weights are integrated to construct a dynamic matrix that can accurately reflect the interaction characteristics of elements and the priority of requirements, providing high-quality input data for subsequent clustering analysis.

[0065] Algorithm 1: Sparse Bayesian Matrix Imputation (SBMP) Algorithm; In practical engineering scenarios, the link attribute data of structural architecture models often has some missing parts. For example, the transmission rate data of low-frequency interaction links may not be fully collected, directly affecting the accuracy of subsequent matrix calculations. Based on Bayesian theory and the sparsity assumption, the SBMP algorithm can accurately imput missing values ​​without destroying the original data distribution, providing a complete and reliable data foundation for subsequent matrix decomposition. Its core advantage lies in balancing data fitting and model complexity through prior distribution and posterior inference, avoiding overfitting or underfitting.

[0066] Model Construction Process: The core idea of ​​SBMP is to assume that the original matrix is ​​sparse, meaning that most elements have low interaction strength, while a few core elements have high interaction strength. By deriving the posterior distribution of missing values ​​using Bayes' theorem, accurate imputation is achieved. The original matrix to be imputed is defined as follows: ,in Corresponding to the 19 system elements defined in S2, Represents system elements With system elements The original link attribute composite value between them is obtained by weighting the interaction strength, data transmission rate, and reliability level after normalization. Missing values ​​are recorded as follows when not filled. The prior distribution model of SBMP is defined as follows: ; In the formula, This is a precision parameter used to characterize the sparsity of matrix elements. The larger the value, the stronger the sparsity of the matrix. ; Noise variance is used to characterize random errors during the data acquisition process. ; Indicates a normal distribution. The mean of the distribution is 0, which is taken here. It is the covariance matrix; for A dimensional identity matrix is ​​used to ensure the positive definiteness of the covariance matrix. Based on Bayes' theorem, the posterior distribution is derived by combining the prior distribution and the likelihood function: In the formula, for The set of non-missing values ​​already observed in the dataset. and They are respectively and The prior distribution is assumed to be uninformative, ensuring that the data itself dominates the posterior inference.

[0067] Model training process: 1. Input data: Import the interaction strength values, data transmission rate, and reliability level from the link annotation dataset output by S3, and integrate them into the original matrix through weighted normalization (weights are 0.4, 0.3, and 0.3 respectively). Statistics show that approximately 10% of the matrix contains missing values, primarily concentrated in the transmission rate data of some data stream interaction links and mechanically fixed links. 2. Initialization Parameters: Set initial precision parameters. Initial noise variance Number of iterations Convergence threshold This ensures that the iterative process can both fully optimize the parameters and avoid over-iteration. 3. Iterative update process: The first... The next iteration ( From 1 to First, calculate the log-likelihood function of the posterior distribution, transforming the product form into a summation form to simplify the calculation: ;right and By taking the partial derivatives separately and setting them to zero, we obtain the analytical solution for the parameter update: ; In the formula, for The number of elements in for The set of indices For the first The missing value filling result obtained from the next iteration.

[0068] Calculate the change in parameters ,like If the parameters have stabilized, the iteration stops; otherwise, proceed to the next iteration. 4. Training Results: Iteration to... When, parameter change The process converges iteratively, ultimately yielding the accuracy parameters. noise variance Generate the filled complete matrix After sampling verification, the average error between the filled value and the actual engineering test value is ≤5%, which meets the requirements for subsequent calculations.

[0069] Model application process: The padded complete matrix Directly used as input data for the DWCMF algorithm, its core contribution lies in filling in missing values ​​in the original data, avoiding distortion of matrix decomposition results due to incomplete data, and ensuring the accuracy of subsequent dynamic design structure matrix construction. For example, in the original link data between system elements SYE-001 (AI computing chip) and SYE-019 (5G communication module), the data transmission rate was not collected due to testing limitations. After being filled in by the SBMP algorithm, the value was 2.5GB / s. Subsequent engineering tests verified that the error between this filled value and the actual value was only 3.2%, fully demonstrating the effectiveness of the algorithm.

[0070] Algorithm 2: Dynamic Weighted Coupling Matrix Factorization (DWCMF); Original matrix While reflecting the interaction characteristics between elements, it does not incorporate the demand weights determined by S1, and therefore cannot reflect the impact of demand priority on module partitioning. The DWCMF algorithm couples element features with dynamic weights through matrix factorization, and introduces a regularization term to avoid overfitting. It can dynamically integrate demand weights into the matrix construction process while extracting element interaction features, so that the final dynamic design structure matrix reflects both the element interaction strength and demand priority, providing inputs that are more in line with the design goals for subsequent cluster analysis.

[0071] Model building process: The core idea of ​​DWCMF is to use the padded matrix The matrix decomposition model is decomposed into a product of row feature matrices, dynamic weight matrices, and column feature matrices. Demand weights are incorporated into the dynamic weight matrix, thus coupling interactive features with demand priorities. The matrix decomposition model is defined as follows: In the formula, For row characteristic matrices, The feature dimension is determined by the elbow rule, balancing feature expressiveness and computational efficiency. Each row corresponds to the system element's own attribute characteristics, such as computing power, power consumption, reliability, etc. It is a column feature matrix, where each column corresponds to the interaction requirement feature of a system element, such as data transmission requirements, collaborative function requirements, etc. This is a dynamic weight matrix, obtained by dynamically adjusting the demand weights of S1, and its elements... ,in The value range is 1 to , This is the coupling coefficient, used to balance the influence of different dimensions of demand weights. For the first in S1 Weight values ​​for different types of requirements; The error matrix is ​​used to characterize the reconstruction error of matrix decomposition and to constrain... ,in For the error threshold, The Frobenius norm is used to ensure that the reconstruction error is within an acceptable range. To avoid overfitting, an L1 regularization term is introduced, and the objective function is defined as: ; In the formula, The regularization coefficient is used to control the sparsity of the feature matrix and prevent overfitting. It is determined through cross-validation. ; Using the L1 norm, the sparsity of the feature matrix is ​​enhanced by penalizing large elements in the feature matrix, thereby improving the model's generalization ability.

[0072] Model training process: 1. Input data: Import the filling matrix output by the SBMP algorithm. and the 12 demand weight values ​​output by S1. 1. Ensure data integrity and consistency. 2. Initialize parameters: initialize the row feature matrix. and column characteristic matrix Initialize the element to conform Uniformly distributed random matrix, dynamic weight matrix Initialize to an initial matrix calculated based on demand weights, and set the number of iterations. Convergence threshold 3. Iterative update process: Iterative optimization is performed using alternating least squares method. Two matrices are fixed, and the third matrix is ​​updated. This process is repeated until convergence. ,renew : Pair the objective function Taking the partial derivative and setting it to 0, we get The analytical update formula is as follows: ;fixed ,renew Similarly, the objective function is applied to... Taking the partial derivative and setting it to 0, we get The analytical update formula is as follows: ; Calculate reconstruction error ,like If the model has converged, stop iterating; otherwise, continue iterating. 4. Training results: Iteration to... When, the change in reconstruction error is The model converges, and the final row feature matrix is ​​obtained. Column feature matrix With dynamic weight matrix Reconstruction error This satisfies the error constraint requirements.

[0073] Model application process: 1. Dynamic design and construction of structure matrix: based on training. , and By combining the average value of the demand weights, the final dynamic design structure matrix is ​​constructed. The calculation formula is: In the formula, for The Row vectors, representing system elements The final attribute characteristics; for The Column vectors, representing system elements The final interaction requirements characteristics; This is the demand weight correction coefficient, used to adjust the degree of influence of the average demand weight on the matrix elements; For elements in S1 , The average of all associated demand weights is used to further reinforce the impact of demand priority on element interactions.

[0074] 2. Algorithm Interaction Process: SBMP and DWCMF form a bidirectional interactive optimization mechanism: Forward Propagation: The filling matrix output by SBMP Complete input data is provided for DWCMF. The distribution characteristics directly determine and The initial feature extraction direction ensures that matrix factorization is based on real and complete interaction data; reverse feedback: the feature matrix obtained from DWCMF training. and Feedback is sent to the SBMP to dynamically adjust its accuracy parameters. The formula has been updated and is now adjusted as follows: ,in and They are respectively and The Frobenius norm is adjusted to make the SBMP filling results more closely match the true feature distribution of the elements, improving filling accuracy. Through this bidirectional interaction, the two sets of algorithms work together to optimize, ultimately enabling the dynamic design of the structure matrix. The element value error is ≤3%, accurately reflecting the element interaction characteristics and demand priority.

[0075] Output: Generates a dynamic design structure matrix. The matrix contains specific values ​​in 19×19 cells, each value accurately reflecting the interaction strength and priority of the corresponding system element pair; it also outputs an element ID index table, clarifying the correspondence between the matrix row / column index and the system element ID; in addition, it outputs a record of algorithm interaction parameters, detailing the training parameters, convergence process, and interaction adjustment details of SBMP and DWCMF.

[0076] S5: Multi-constraint priority clustering analysis; This step, based on the dynamic design structure matrix output by S4, introduces two sets of collaborative algorithms to solve the problem that traditional clustering algorithms cannot take into account both element interaction characteristics and multi-constraint conflicts, thereby achieving module division that meets actual engineering needs. At the same time, it quantifies the degree of modularity and provides a basis for subsequent module optimization.

[0077] Algorithm 3: Improved Gravity Search Clustering Algorithm (IGSA-C); Traditional clustering algorithms (such as K-Means) only cluster based on the distance between elements, without considering the impact of engineering constraints on module partitioning, which may lead to clustering results that do not meet actual engineering needs. The IGSA-C algorithm is an improvement on the gravity search algorithm, introducing a constraint compatibility gravity term, which transforms the degree of constraint satisfaction into a gravity adjustment factor. In the clustering process, it considers both the interaction characteristics (distance) between elements and constraint compatibility, making the initial clustering results more in line with engineering practice. Its core advantage lies in simulating the "attraction" and "repulsion" between elements through a gravity mechanism, achieving the clustering goal of high cohesion within modules and low coupling between modules.

[0078] Model building process: The core idea of ​​IGSA-C is to regard each system element as a "particle" in the feature space. The gravitational force between particles is determined by the interaction strength (mass), distance and constraint compatibility of elements. By simulating the motion of particles in the gravitational field, the natural clustering of particles is achieved.

[0079] The gravity calculation model is defined as follows: In the formula, For the first System elements in the next iteration For system elements The gravitational value; the larger the gravitational value, the stronger the element. right The stronger the attraction; The gravitational constant decays exponentially with the iteration process, and its calculation formula is: ,in Let be the initial gravitational constant. The total number of iterations is used to ensure that the particle exploration range is wide in the early stage of the iteration and gradually converges in the later stage. For system elements The quality is obtained by normalizing the row vectors of the dynamic design structure matrix, and the calculation formula is: ,in For dynamic design of structure matrix The For row vectors, the greater the mass, the stronger the "attraction" of the elements. For system elements and The Euclidean distance in the feature space is calculated using the following formula: The smaller the distance, the stronger the attraction between elements; To find the minimum value, avoid a denominator of 0; To constrain compatibility coefficients, the MCOA algorithm outputs values ​​that characterize system elements. and The degree to which engineering constraints are met. , The larger the value, the better the constraint compatibility and the stronger the attraction.

[0080] The formulas for particle acceleration, velocity, and position update are as follows: ; ; In the formula, For system elements The acceleration of a particle is directly proportional to the total gravitational force it experiences and inversely proportional to its mass. For system elements speed, The inertia coefficient is used to balance the particle's exploration and convergence capabilities; For system elements At the new location in the feature space, the initial location The position update reflects the movement trend of elements in the gravitational field.

[0081] Model training process: 1. Input data: Import the dynamic design structure matrix output by S4. And the initial constraint compatibility coefficients pre-calculated by the MCOA algorithm. 1. Initial constraint compatibility coefficients are initially assigned based on engineering constraint rules. 2. Initialization parameters: Set the initial positions of system elements. initial velocity (Zero vector); Determining the number of clusters based on the elbow rule This clustering number ensures both the functional integrity of the modules and avoids excessive subdivision; the number of iterations is set. The convergence threshold is the change in element position. This ensures the stability of the clustering results.

[0082] 3. Iterative update process: The first... Second iteration: First, calculate the mass of each system element. It is obtained by normalizing the row vectors; the Euclidean distance between elements is calculated. This reflects the similarity of element interaction features. It receives the current iteration constraint compatibility coefficient from the MCOA algorithm. Substitute into the gravity formula to update the gravitational forces between each element. Calculate the acceleration of each element based on the total gravitational force. The element velocity is updated by combining the inertia coefficient with the current velocity. This updates the element's position in the feature space. The positional distribution of elements in the statistical feature space is analyzed, and the element with the densest position vectors is selected. Each point serves as a new cluster center. Each element is assigned to the nearest cluster center, forming Initially, cluster the elements. Calculate the sum of squared errors within each cluster (SSE), which is the sum of the squared distances from each element to its cluster center, reflecting the compactness of the elements within the cluster. If the change in SSE is less than the convergence threshold or the number of iterations reaches a certain threshold, the cluster is considered closed. If the iteration stops, proceed to the next iteration; otherwise, continue to the next iteration.

[0083] 4. Training Results: Iterated to When, the change in SSE is The convergence threshold was set, and the clustering converged, generating an initial clustering result of 7 modules. The SSE within the cluster was 0.85, indicating that the elements within the cluster were densely distributed. According to the constraint satisfaction statistics, the constraint satisfaction of the initial clustering result was 85%, indicating that some constraints were violated and further optimization is needed.

[0084] Model Application Process: The core contribution of IGSA-C lies in generating initial clustering results with high cohesion within modules and low coupling between modules, based on the interaction characteristics of the dynamic design structure matrix and combined with constraint compatibility. In this initial clustering result, the average interaction strength of elements within each module is ≥4.0 (out of 5), and the average interaction strength of elements between modules is ≤1.5, fully reflecting the core requirements of modular design. The specific initial clustering results are as follows: Cluster 1 (AI Computing Core Module): Includes SYE-001 (AI computing chip) and SYE-008 (DDR5 memory). The elements within this module mainly undertake the functions of intelligent driving algorithm calculation and data storage, with an average interaction strength of 4.91 and tight coupling. Cluster 2 (Vehicle Control Core Module): Includes SYE-002 (safety MCU), SYE-010 (brake control actuator), and SYE-011 (steering control actuator). The elements within this module are responsible for the core vehicle control functions, with an average interaction strength of 4.23. Cluster 3 (General Sensing Interface Module): Includes SYE-003 (LiDAR sensor interface chip), SYE-004 (millimeter-wave radar sensor interface chip), SYE-005 (camera sensor interface chip), and SYE-006 (ultrasonic sensor interface chip). Elements within this module are responsible for the transmission and conversion of various sensor data, with an average interaction strength of 3.89. Cluster 4 (Power and Thermal Management Module): Includes SYE-007 (power management chip) and SYE-015 (air conditioning control module). Elements within this module are responsible for power and thermal management, with an average interaction strength of 3.76. Cluster 5 (Body Control Module): Includes SYE-012 (suspension adjustment actuator), SYE-013 (door and window control actuator), and SYE-014 (seat adjustment actuator). Elements within this module are responsible for adjusting the vehicle's status, with an average interaction strength of 3.52. Cluster 6 (Infotainment and Communication Module): Includes SYE-016 (Vehicle Display Control Chip), SYE-017 (Navigation and Positioning Module), SYE-018 (Bluetooth Communication Module), and SYE-019 (5G Communication Module). Elements within this module are responsible for infotainment and communication functions, with an average interaction strength of 3.68. Cluster 7 (High-Speed ​​Bus Interaction Module): Includes SYE-009 (PCIe 5.0 High-Speed ​​Bus Switch), responsible for high-speed data exchange and forwarding. Due to its independent function, it forms a separate module.

[0085] Algorithm 4: Multi-Constraint Collaborative Optimization (MCOA) Algorithm; Initial clustering results may contain some constraint violations, and conflicts may exist between multiple constraints (such as supply chain constraints and upgradeability constraints). Traditional optimization algorithms struggle to efficiently resolve these conflicts. The MCOA algorithm constructs a multi-constraint collaborative optimization objective function, incorporating constraint satisfaction and matrix smoothness into the optimization objective. While resolving constraint conflicts, it ensures the rationality of module partitioning. Its core advantage lies in reflecting constraint priority through weight allocation, achieving the optimization objective of prioritizing the satisfaction of high-priority constraints and adaptively adjusting low-priority constraints.

[0086] Model building process: The core idea of ​​MCOA is to transform various engineering constraints into constraint satisfaction indices, construct an objective function that includes constraint satisfaction and matrix smoothness, and obtain the optimal constraint compatibility coefficients through optimization, providing a basis for dynamic adjustment of IGSA-C.

[0087] The objective function is defined as: The constraints are: ( ),in: To constrain the compatibility coefficient matrix, Used to characterize elements and The degree to which constraints are satisfied; Number 6 corresponds to 6 types of engineering constraints, namely, necessary decoupling constraints, supply chain constraints, thermal management constraints, upgradeability constraints, necessary coupling constraints, and maintenance cycle constraints. To assign priority weights to constraints, values ​​are allocated based on the importance of the constraints, and are set to a value of [value missing]. Among them, the necessary decoupling constraints and supply chain constraints (P1 level) have the highest weight (0.3), followed by thermal management constraints and upgradeability constraints (P2 level) (0.15), and the necessary coupling constraints and maintenance cycle constraints (P3 level) have the lowest weight (0.05), reflecting the priority of constraints; For the first The satisfaction degree of class constraints is calculated using the following formula: , The closer the value is to 1, the higher the degree of constraint satisfaction. It is a matrix smoothing coefficient used to balance the constraint satisfaction and the smoothness of the dynamic design structure matrix, avoiding matrix distortion caused by excessive pursuit of constraint satisfaction; For dynamic design of structure matrix gradient, The L2 norm of the gradient is used to characterize the smoothness of the matrix; the smaller the gradient, the smoother the matrix and the more reasonable the module partitioning. Constraint satisfaction. The specific calculation rules are as follows: Decoupling constraints are required ( If element and If the necessary decoupling requirements such as "separation of high-power-consumption elements from high-temperature-sensitive elements" and "isolation of vibration-sensitive components" are met, then... ,otherwise ,final Supply chain constraints If element and If the supply chain requirements such as "elements of supply chain priority P1 are independent modules" and "elements of the same manufacturer are coupled" are met, then... ,otherwise , The calculation method is the same as above; thermal management constraints ( If element and If thermal management requirements such as "centralized partitioning of high-power elements" are met, then ,otherwise Upgradeability constraints ( If element and If the upgradeability requirements such as "centralizing elements with high technology update rates" are met, then... ,otherwise Required coupling constraints ( If element and If the necessary coupling requirements such as "co-maintaining periodic element coupling" are met, then ,otherwise Maintenance cycle constraints ( If element and If the maintenance requirements such as "centralized partitioning of elements with a maintenance cycle of ≤6 years" are met, then ,otherwise .

[0088] Model Training Process: 1. Input Data: Import the specific requirements of 6 types of engineering constraints, as shown in the examples below: Required Decoupling Constraints: High-power elements (AI computing chips) and high-temperature sensitive elements (safety MCUs, vehicle display control chips) need to be assigned to different modules; vibration-sensitive components (LiDAR sensor interface chips, camera sensor interface chips) need to be isolated from high-vibration elements. Supply Chain Constraints: Elements with supply chain priority P1 (AI computing chips, safety MCUs, power management chips, brake control actuators, steering control actuators) need to be assigned to independent modules to avoid mixing with elements with priority P3 and below. Thermal Management Constraints: Elements with "high" power consumption levels (AI computing chips) need to be centrally assigned to the same module for easy integration of heat dissipation design. Upgradeability Constraints: Elements with a technology update rate of "once a year" (AI computing chips, DDR5 memory, vehicle display control chips, navigation and positioning modules, Bluetooth communication modules, 5G communication modules) need to be assigned to the same module to ensure that independent upgrades do not affect other elements. Required Coupling Constraints: Elements from the same manufacturer (LiDAR sensor interface chips and millimeter-wave radar sensor interface chips, brake control actuators and steering control actuators, door and window control actuators and seat adjustment actuators, navigation and positioning modules and Bluetooth communication modules) must be grouped into the same module. Maintenance Cycle Constraints: Elements with a maintenance cycle ≤ 6 years (AI computing chips, LiDAR sensor interface chips, millimeter-wave radar sensor interface chips, camera sensor interface chips, ultrasonic sensor interface chips, DDR5 memory, PCIe 5.0 high-speed bus switch, vehicle display control chip, navigation and positioning module, Bluetooth communication module, 5G communication module) must be grouped into the same module for centralized maintenance.

[0089] Simultaneously import the initial constraint compatibility coefficient Based on the initial assignment of the above constraint rules, if the P1 level constraints are satisfied... If the P2 level constraints are satisfied, then If the P3 level constraints are satisfied, then Violation of any constraint .

[0090] 2. Optimization Solution: The objective function is solved using the Sequential Quadratic Programming (SQP) algorithm, with the following steps: Initialize the constraint compatibility coefficient matrix Set the number of iterations Convergence threshold .

[0091] No. The next iteration: First, construct a quadratic programming subproblem, approximating the objective function as a quadratic form: ,in Let Hessian matrix be the objective function. Let be the gradient vector of the objective function.

[0092] Solving the quadratic programming subproblem yields the coefficient update amount. The step size is determined by line search. Update the constraint compatibility coefficient matrix: Calculate the objective function value. .like If the objective function has converged, the iteration stops; otherwise, proceed to the next iteration.

[0093] 3. Training Results: Iterated to When, the change in the objective function is The iteration converges, and the final objective function value is... The constraint satisfaction rate increased from the initial 85% to 98%, generating an optimized constraint compatibility coefficient matrix. .

[0094] Model application process: 1. Algorithm interaction process: IGSA-C and MCOA form a bidirectional collaborative optimization mechanism to ensure that the clustering results satisfy the constraints while maintaining high cohesion within the modules: Forward propagation: Optimized constraint compatibility coefficient matrix output by MCOA The data is fed back to IGSA-C in real time as an adjustment factor for gravity calculations, correcting the gravitational values ​​between elements. For element pairs that satisfy high-priority constraints, Larger values ​​increase attraction and promote element clustering; for pairs of elements that violate high-priority constraints, Smaller values ​​reduce the gravitational pull and prevent element clustering.

[0095] Reverse feedback: The clustering results of IGSA-C are fed back to MCOA to dynamically adjust the constraint priority weights. Adjust the formula to ,in For the first The current degree of satisfaction of class constraints. This adjustment, set to the maximum value of all constraint satisfaction, further increases the weight of constraints with high satisfaction and appropriately reduces the weight of constraints with low satisfaction, ensuring that resources are tilted towards key constraints.

[0096] Collaborative optimization example: In the initial IGSA-C clustering, the AI ​​computing chip (SYE-001) was classified into the infotainment and communication module due to its high interaction intensity with the vehicle display control chip (SYE-016). However, this violates supply chain constraints (SYE-001 is a P1 priority element and should be an independent module). After identifying this conflict, MCOA outputs... (Low compatibility) After receiving the data, the IGSA-C recalculates the attraction. The attraction between the AI ​​computing chip and the vehicle display control chip is significantly weakened, while the attraction with the DDR5 memory (SYE-008) remains strong. Ultimately, the AI ​​computing chip is adjusted to an independent module, and the constraint satisfaction rate increases from 85% to 98%.

[0097] 2. Modularity Index Calculation: To quantify the rationality of module division, a modularity index is calculated. The calculation formula is: In the formula, For the first A collection of elements from each module. This represents the sum of interactions between modules. This represents the total number of interactions within the system. The range of values ​​is , The closer the value is to 1, the higher the degree of modularity. After co-optimization by IGSA-C and MCOA, the initial modularity index... This indicates that the module division is highly reasonable.

[0098] Output results: Generates an initial module partitioning list, a modularity index report, and a constraint satisfaction report. The initial module partitioning list clearly defines the name of each module and the IDs and names of the system elements it contains; the modularity index report records in detail the calculation process, results, and significance of the modularity index; the constraint satisfaction report statistically analyzes the satisfaction status of various constraints, with a 100% satisfaction rate for P1-level constraints, a 97% satisfaction rate for P2-level constraints, and a 95% satisfaction rate for P3-level constraints, fully reflecting the requirements of constraint priority.

[0099] In some embodiments, the Sparse Bayesian Matrix Filling (SBMP) algorithm and the Multi-Constraint Cooperative Optimization (MCOA) algorithm achieve bidirectional interaction: the SBMP output ensures the completeness of the filled matrix. The percentage of cells with no missing values ​​is fed back to MCOA in real time, determining the adjustment range of non-core constraint weights; MCOA calculates the core constraint satisfaction rate. (Requires decoupling and supply chain constraints) Feedback is sent to SBMP to tighten the data completion accuracy of the corresponding elements.

[0100] 1. Formula for adjusting the weights of non-core constraints: ; The weight reduction for non-core constraints (such as maintenance cycle constraints) only occurs when... The adjustment is triggered at any time, and the adjustment amount increases linearly with the integrity gap. 2. Filling error threshold tightening formula: ; The SBMP fill error threshold for core constraint associated elements, only when The tightening can be as low as 1%.

[0101] Termination condition (formulaic): Or cumulative number of interaction iterations (The first one to be satisfied shall prevail).

[0102] In some embodiments, the Sparse Bayesian Matrix Filling (SBMP) algorithm and the Improved Gravitational Search Clustering (IGSA-C) algorithm achieve bidirectional interaction: the inter-element filling error output by SBMP... (Completion error of interaction data between elements i and j), used as a correction factor for the initial distance of IGSA-C; the mean value of intra-cluster interaction strength output by IGSA-C. (The k-th cluster) determines the data completion priority of SBMP.

[0103] 1. Initial distance correction formula: ; The initial distance between elements i and j after correction. The initial distance is the original Euclidean distance. The larger the filling error, the larger the initial distance and the lower the cluster correlation. 2. Complete the priority weight formula: ; The completion priority (0-100) is used for the k-th cluster. The smaller the value, the higher the priority. SBMP prioritizes filling in missing data within the cluster.

[0104] Termination condition (formulaic): or ( This represents the current iteration number. This represents the overall average filling error. (This represents the average cluster strength).

[0105] In some embodiments, the Dynamic Weighted Coupling Matrix Factorization (DWCMF) algorithm and the Multi-Constraint Collaborative Optimization (MCOA) algorithm achieve bidirectional interaction: the similarity of the feature matrices output by DWCMF... The overlap of row / column eigenvectors is fed back to the MCOA, triggering necessary decoupling constraint weight adjustments; the constraint compatibility coefficient matrix output by the MCOA... (Constraint matching degree of elements i and j), which is the core parameter of DWCMF regularization term.

[0106] 1. Formula for adjusting the weights of the necessary decoupling constraints: ; The decoupling constraint weights are necessary for the next round. This indicates that the element's function is redundant and that decoupling constraints need to be strengthened. 2. DWCMF regularization term formula: ; For regularization terms, The regularization coefficient is . Let i be the eigenvectors of elements i and j. The lower the value, the greater the regularization penalty, thus avoiding feature confusion after decomposition.

[0107] Termination condition (formulaic): ( (mean of compatibility coefficient) or ( (This refers to the reconstruction error of DWCMF, which showed no significant decrease after two consecutive iterations).

[0108] In some embodiments, the Dynamic Weighted Coupling Matrix Factorization (DWCMF) algorithm and the Improved Gravitational Search Clustering (IGSA-C) algorithm achieve bidirectional interaction: the element eigenvectors output by DWCMF... It replaces the traditional Euclidean distance as the core distance metric of IGSA-C; the clustering results output by IGSA-C (The cluster label to which the element belongs) provides feedback to adjust the decomposition dimensions of DWCMF.

[0109] 1. Cluster distance metric formula: ; The clustering distance between elements i and j is calculated using the complementary value of the cosine similarity of the feature vectors. The higher the similarity, the smaller the distance, and the easier it is for them to be classified into the same cluster. 2. Decomposition Dimension Adjustment Formula: ; For the next round of DWCMF decomposition dimensions, This represents the current number of clusters in IGSA-C. The decomposition dimension is adjusted synchronously when the number of clusters changes to ensure that the feature representation matches the cluster structure.

[0110] Termination condition (formulaic): ( The degree of matching between the clustering results and the feature vector distribution. (Reconstruction error of DWCMF) or cumulative number of interactions .

[0111] S6: Constraint Conflict Resolution and Module Optimization; Although the initial module partitioning list output from S5 satisfies most constraint requirements, a small number of constraint conflicts still exist, and the modularity index has room for further optimization. This step involves interactive review by architects to identify conflicts and formulate resolution strategies. The collaborative algorithm is then re-run to optimize the module partitioning, ensuring that the module partitioning both conforms to engineering practice and has a high degree of modularity.

[0112] 1. Initial Result Review: The architect comprehensively reviewed the initial module partitioning list, modularity index report, and constraint satisfaction report output by S5 using the visual interface of the MBSE tool, focusing on constraint conflicts and the rationality of the modularity index. The review identified two key constraint conflicts: Conflict 1: Supply chain constraints (P1 level) require P1 priority elements to be independent modules, while upgradeability constraints (P2 level) require elements with high technology update rates to be grouped into modules. This creates a conflict. Specifically, the AI ​​computing chip (SYE-001) is both a P1 priority element requiring an independent module and a high-update-rate element with an annual update rate. According to upgradeability constraints, it needs to be grouped into a module with other high-update-rate elements (DDR5 memory, vehicle display control chips, etc.). Strictly meeting supply chain constraints would violate upgradeability constraints; conversely, meeting upgradeability constraints would violate supply chain constraints.

[0113] Conflict 2: The mandatory coupling constraint (P3 level) requires navigation and positioning modules (SYE-017) and Bluetooth communication modules (SYE-018) from the same manufacturer to be coupled. This does not conflict with the maintenance cycle constraint (P3 level) requiring elements with a maintenance cycle ≤ 6 years to be grouped into modules, but it partially overlaps with the upgradeability constraint (P2 level) requiring high update rate elements to be grouped together. Both the navigation and positioning module and the Bluetooth communication module are high update rate elements and need to be grouped together with AI computing chips, etc. However, due to supply chain constraints, the AI ​​computing chip needs to be a separate module, making complete grouping of the two impossible.

[0114] 2. Conflict Resolution Strategy Formulation: Based on the constraint priority principle, a strategy of "prioritizing high-priority constraints and adaptively adjusting low-priority constraints" is adopted to formulate targeted conflict resolution solutions, ensuring that core constraints are satisfied while maximizing the consideration of secondary constraints. Regarding Conflict 1: Supply chain constraints are P1 level, and upgradeability constraints are P2 level. Prioritize meeting supply chain constraints by designating the AI ​​computing chip (SYE-001) as a separate core module to ensure the independence of P1 priority elements. For upgradeability constraints, other high update rate elements (DDR5 memory, vehicle display control chip, navigation and positioning module, Bluetooth communication module, 5G communication module) are assigned to two adjacent modules (the high-speed bus interaction module adjacent to the AI ​​computing core module, and the infotainment and communication module). Standardized interface design ensures upgrade compatibility between high update rate elements, satisfying upgradeability requirements without violating supply chain constraints.

[0115] Regarding Conflict 2: Both the necessary coupling constraint and the maintenance cycle constraint are at level P3, while the upgradeability constraint is at level P2. Priority is given to satisfying the upgradeability constraint, and the navigation and positioning module and the Bluetooth communication module are incorporated into the infotainment and communication module (a module with a high update rate of elements). Since both belong to the same manufacturer, the necessary coupling constraint is satisfied. As for the maintenance cycle constraint, all elements in the infotainment and communication module have a maintenance cycle of ≤6 years, thus satisfying the maintenance cycle constraint and achieving the coordinated satisfaction of the three types of constraints.

[0116] 3. Module Adjustment and Re-clustering: The conflict resolution scheme is transformed into new constraint parameters, the rule of "P1 level constraints take precedence over P2 level constraints" is supplemented, and the IGSA-C and MCOA collaborative algorithm is re-run. The algorithm inputs the dynamic design structure matrix of S4 and the adjusted constraint parameters. Through bidirectional interactive optimization, an optimized module partitioning list is generated, as follows: AI Computing Core Module: Includes AI computing chip (SYE-001) and DDR5 memory (SYE-008). DDR5 memory, as a supporting component of the AI ​​computing chip, is tightly coupled with the AI ​​computing chip and does not violate supply chain constraints. Including it in this module improves module cohesion. Vehicle Control Core Module: Includes safety MCU (SYE-002), brake control actuator (SYE-010), and steering control actuator (SYE-011). All are P1 priority elements, forming independent modules, satisfying supply chain constraints, and the elements within the module are responsible for the core vehicle control functions, with close interaction. The general-purpose sensor interface module includes a lidar sensor interface chip (SYE-003), a millimeter-wave radar sensor interface chip (SYE-004), a camera sensor interface chip (SYE-005), and an ultrasonic sensor interface chip (SYE-006). Elements of the same functional type are grouped together, satisfying necessary coupling constraints and maintenance cycle constraints. The power and thermal management module includes a power management chip (SYE-007) and an air conditioning control module (SYE-015), responsible for power supply and thermal management, satisfying thermal management constraints. Both are P2 priority elements and are independent modules. The body control module includes door and window control actuators (SYE-013), seat adjustment actuators (SYE-014), and suspension adjustment actuators (SYE-012). Elements of the same functional type are grouped together, satisfying necessary coupling constraints and maintenance cycle constraints. Infotainment and Communication Module: This module includes the vehicle display control chip (SYE-016), navigation and positioning module (SYE-017), Bluetooth communication module (SYE-018), and 5G communication module (SYE-019). High-update-rate elements are concentrated here, meeting upgradeability constraints. Furthermore, the navigation and positioning module and Bluetooth communication module are from the same manufacturer, satisfying the necessary coupling constraint. High-Speed ​​Bus Interaction Module: This module includes a PCIe 5.0 high-speed bus switch (SYE-009), responsible for high-speed data exchange between modules. It is functionally independent and forms a separate module.

[0117] 4. Modularity Index Update: Based on the optimized module partitioning list, the modularity index is recalculated. The sum of interactions between modules is the sum of the values ​​of the dynamic design structure matrix among all different module elements, calculated to be 128.6; the total sum of system interactions is the sum of the values ​​of all cells in the dynamic design structure matrix, which is 338.4; substituting these values ​​into the modularity index formula... Although this value is slightly lower than the initial value (0.68), it is still within a reasonable range (0.6-0.8 represents a high degree of modularity) and meets more important engineering constraints, significantly improving the practicality and rationality of the overall module division.

[0118] 5. Output Results: Generate an optimization module partitioning list, an updated modular index report, and a conflict resolution report. The optimization module partitioning list clearly defines the name of each module and the system elements it contains, ensuring no elements are omitted; the updated modular index report records the index calculation process and results in detail, explaining the reasons for index changes; the conflict resolution report records the complete process of conflict identification, strategy formulation, and optimization execution, as well as the constraint satisfaction status after conflict resolution.

Claims

1. A modular partitioning method for an open electronic and electrical architecture of vehicles based on the fusion of MBSE and MOSA, characterized in that, include: S1: Collect technical specifications from vehicle manufacturers, adaptation requirements from component suppliers, end-user feedback, industry technology upgrade trend documents, and policy and regulatory compliance standards, summarize them into an original requirements list, classify them into functional and non-functional requirements, calculate the weight value of each requirement, and generate a modular requirements hierarchy table. S2: Based on the modular requirements layering table, construct the functional architecture and structural architecture model in the MBSE tool, define all system elements to be modularized and the links between elements, clarify the core attributes of elements and links, and generate the structural architecture model; S3: Import the SysML modular configuration file, which is independently constructed based on the vehicle modular domain ontology, add semantic attributes to system elements and links, establish the traceability relationship between modular requirements and elements and links, and generate a semantically enhanced structural architecture model. S4: Extract data from the semantically enhanced structural architecture model, fill in the missing data using the sparse Bayesian matrix filling algorithm, and then fuse the requirement weight values ​​using the dynamic weight coupling matrix decomposition algorithm to construct the dynamic design structure matrix; S5: Collect actual engineering constraints, calculate constraint compatibility coefficients through multi-constraint collaborative optimization algorithm, combine with dynamic design structure matrix, use improved gravity search clustering algorithm for clustering, and generate initial module partitioning list and modularity index report; S6: Review the initial module partitioning list and modularity index report, identify constraint conflicts and formulate resolution strategies, rerun the improved gravity search clustering algorithm and multi-constraint collaborative optimization algorithm, generate an optimized module partitioning list, and complete the modular partitioning.

2. The modular partitioning method for an open electronic and electrical architecture of a vehicle based on the fusion of MBSE and MOSA as described in claim 1, characterized in that, In step S4, the sparse Bayesian matrix constructs the prior distribution model and the posterior distribution model. The missing values ​​in the original data are filled by iteratively optimizing the accuracy parameter and noise variance. During the iteration process, the parameter change is less than the convergence threshold as the stopping condition, and finally a filled matrix with an average error ≤ 5% is generated.

3. The modular partitioning method for an open electronic and electrical architecture of a vehicle based on the fusion of MBSE and MOSA as described in claim 2, characterized in that, In step S4, the dynamic weighted coupled matrix decomposition algorithm decomposes the filling matrix into a product of row feature matrix, dynamic weight matrix and column feature matrix. The dynamic weight matrix is ​​dynamically adjusted by the demand weight value. At the same time, an L1 regularization term is introduced to avoid overfitting and constrain the reconstruction error of the matrix decomposition to ≤0.

01. Finally, the dynamic design structure matrix that integrates demand priorities is output.

4. The modular partitioning method for an open electronic and electrical architecture of a vehicle based on the fusion of MBSE and MOSA as described in claim 3, characterized in that, In step S4, the sparse Bayesian matrix filling algorithm and the dynamic weighted coupled matrix decomposition algorithm form a two-way interaction mechanism: the filling matrix output by the sparse Bayesian matrix filling algorithm serves as the input data for the dynamic weighted coupled matrix decomposition algorithm, determining the initial feature distribution of the row feature matrix and the column feature matrix; the row feature matrix and the column feature matrix obtained by the dynamic weighted coupled matrix decomposition algorithm are fed back to the sparse Bayesian matrix filling algorithm, dynamically adjusting its accuracy parameter update formula, so that the accuracy of filling missing values ​​in the filling matrix matches the element feature distribution better, and finally achieving collaborative optimization of the two sets of algorithms, with the element value error of the dynamically designed structure matrix ≤3%.

5. The modular partitioning method for an open electronic and electrical architecture of a vehicle based on the fusion of MBSE and MOSA as described in claim 1, characterized in that, In step S5, the improved gravity search clustering algorithm introduces a constraint compatibility coefficient as a gravity adjustment factor. It calculates the gravity value between system elements through the mechanism of gravity constant decaying with iteration. It updates the acceleration, velocity and position of elements by combining element mass and Euclidean distance, and iteratively achieves a clustering effect of high cohesion within modules and low coupling between modules. The number of clusters is determined by the elbow rule.

6. The modular partitioning method for an open electronic and electrical architecture of a vehicle based on the fusion of MBSE and MOSA as described in claim 5, characterized in that, In step S5, the multi-constraint collaborative optimization algorithm constructs an objective function that includes constraint satisfaction and matrix smoothness. Weights are assigned according to the priority of necessary decoupling constraints, supply chain constraints, thermal management constraints, upgradeability constraints, necessary coupling constraints, and maintenance cycle constraints. The optimal constraint compatibility coefficient is solved by a sequential quadratic programming algorithm, thereby increasing the constraint satisfaction to over 98%.

7. The modular partitioning method for an open electronic and electrical architecture of a vehicle based on the fusion of MBSE and MOSA as described in claim 6, characterized in that, In step S5, the improved gravity search clustering algorithm and the multi-constraint collaborative optimization algorithm form a two-way interaction mechanism: the constraint compatibility coefficient output by the multi-constraint collaborative optimization algorithm is injected into the gravity calculation model of the improved gravity search clustering algorithm as a gravity adjustment factor to correct the gravity values ​​between elements; the clustering results of the improved gravity search clustering algorithm are fed back to the multi-constraint collaborative optimization algorithm to dynamically adjust the priority weights of various constraints, so that the weights of high-satisfaction constraints are further increased and the weights of low-satisfaction constraints are adaptively reduced. Through interaction, the dual optimization of constraint satisfaction and cluster rationality is achieved, and finally the constraint satisfaction rate is ≥98% and the average interaction strength of elements within the cluster is ≥4.

0.

8. The modular partitioning method for an open electronic and electrical architecture of a vehicle based on the fusion of MBSE and MOSA as described in claim 1, characterized in that, In S6, when rerunning the improved gravitational search clustering algorithm and the multi-constraint collaborative optimization algorithm, the initial modularity index output by S5 is introduced as a feedback parameter for algorithm iteration: if the initial modularity index is ≥0.65, the decay rate of the initial gravitational constant of the improved gravitational search clustering algorithm is reduced to minimize large changes in the clustering structure and retain reasonable partitioning logic; if the initial modularity index is <0.65, the initial weight ratio of the necessary decoupling constraints and supply chain constraints in the multi-constraint collaborative optimization algorithm is increased to prioritize the satisfaction of core constraints; at the same time, the temporary modularity index is calculated in real time during the algorithm iteration process, and the iteration is automatically stopped when the difference between the temporary indices of two consecutive iterations is <0.01, which improves the efficiency of re-clustering by more than 20%, and the optimized modularity index is stable above 0.

6.

9. The modular partitioning method for an open electronic and electrical architecture of a vehicle based on the fusion of MBSE and MOSA as described in claim 4, characterized in that, In S4, the interaction process between the sparse Bayesian matrix filling algorithm and the dynamic weighted coupled matrix decomposition algorithm also includes a dynamic parameter adjustment mechanism: when the reconstruction error of the dynamic weighted coupled matrix decomposition algorithm exceeds the preset threshold of 0.01, the accuracy parameter of the sparse Bayesian matrix filling algorithm is automatically optimized by increasing the number of iterations and adjusting the convergence threshold to optimize the filling matrix; when the filling error of the sparse Bayesian matrix filling algorithm is less than 3%, it is fed back to the dynamic weighted coupled matrix decomposition algorithm to reduce the regularization coefficient to improve the expressive power of the feature matrix. Through this dynamic parameter adjustment, the element value error of the dynamically designed structure matrix is ​​further reduced to ≤2%.

10. The modular partitioning method for an open electronic and electrical architecture of a vehicle based on the fusion of MBSE and MOSA as described in claim 7, characterized in that, In S5, the constraint priority weight update of the multi-constraint collaborative optimization algorithm depends on the clustering result feedback of the improved gravity search clustering algorithm: if the satisfaction of a certain type of constraint in the clustering result is less than 95%, the priority weight of that type of constraint is automatically increased by 0.05; if the satisfaction of a certain type of constraint is ≥99% for two consecutive iterations, its priority weight is appropriately reduced by 0.02, while keeping the priority weights of necessary decoupling constraints and supply chain constraints no less than 0.

25. Through dynamic iterative optimization of constraint weights, the rationality of clustering and the satisfaction of constraints are improved in both directions, and finally the average interaction strength of elements within the cluster is ≥4.2 and the constraint satisfaction rate is ≥99%.

Images