A vehicle whole-process digital design method

Abstract

This invention provides a full-process digital design method for vehicles, belonging to the field of vehicle design and digital technology. The method includes: S1: Multi-dimensional top-level requirement modeling at the system level; S2: System-to-system level indicator mapping and SysML structured modeling; S3: System-level vehicle concept design and logical architecture digital modeling; S4: Precise decomposition of system-to-subsystem indicators and targeted requirement distribution; S5: Subsystem professional simulation modeling and cross-subsystem coupling and integration verification; S6: System-level indicator trade-off analysis and optimal solution selection. The purpose of this invention is to construct a full-process digital design chain, break down data barriers between system requirements, system design, and subsystem implementation stages, achieve traceability and iterative design content, and solve the problem of fragmentation in traditional design.

Description

Technical Field

[0001] This invention relates to the field of vehicle design and digital technology, and in particular to a digital design method for the entire vehicle process. Background Technology

[0002] The global automotive industry is currently undergoing a profound transformation towards electrification, intelligence, and connectivity. Market and user demands for vehicles have shifted from simply optimizing performance to optimizing overall efficiency under specific tasks and scenarios. Examples include low energy consumption and long range in urban commuting scenarios, high passability and power stability in off-road scenarios, and high-precision perception and decision-making response in intelligent driving scenarios. Traditional vehicle design models have significant drawbacks: First, the design process is fragmented, with stages such as system requirements analysis, system concept design, and subsystem professional design being disconnected from each other, lacking a unified digital link. This results in untraceable design content and a disconnect between early requirements and later implementation. Second, cross-stage data interaction is inefficient, with significant differences in design tools and data formats used at each stage, a lack of standardized interfaces, severe data redundancy and conflicts, and high costs and long cycles for design changes. Third, the decomposition and verification of indicators lack scientific algorithm support, relying on empirical allocation, leading to indicator imbalances, poor subsystem coordination, and difficulty in meeting the multi-objective optimization needs of new vehicles. Fourth, existing digital design tools often focus on a single stage or a single subsystem, lacking a closed-loop verification mechanism for the entire process. This makes it impossible to predict cross-subsystem coupling problems in advance, resulting in frequent failures during the real vehicle testing phase and a surge in R&D costs.

[0003] While some companies have attempted to introduce digital design tools, core shortcomings remain: Firstly, there is a lack of precise methods for extracting multi-dimensional top-level requirements, leading to redundant requirements and wasted design resources. Secondly, the allocation of cross-stage indicators and the fusion of multi-source data lack efficient algorithmic support, making it difficult to achieve accurate indicator transmission and effective data utilization. Furthermore, the optimization of system architecture and subsystem models is often single-objective-oriented, neglecting the balance between performance, stability, and design efficiency, resulting in insufficient practicality of the solutions. Therefore, there is an urgent need to build a comprehensive digital design and verification methodology that is fully integrated across the entire process, collaborative across stages, fully supported by algorithms, and has a complete verification loop. This methodology aims to address the pain points of traditional design models, adapt to the transformation needs of the automotive industry, and improve the R&D efficiency and design accuracy of new vehicles. Summary of the Invention

[0004] This invention provides a digital design method for the entire vehicle process, the core objective of which is: Build a full-process digital design chain, break down data barriers between system requirements, system design, and subsystem implementation at each stage, realize the traceability and iterability of design content, and solve the problem of fragmentation in traditional design; By introducing efficient algorithm combinations, we can accurately extract multi-dimensional top-level requirements, scientifically allocate cross-stage indicators, effectively integrate multi-source simulation data, and optimize the selection of multi-objective solutions, thereby improving the scientific nature and accuracy of design decisions. Establish a cross-subsystem coupling verification and system-level closed-loop evaluation mechanism to predict subsystem coordination problems in advance, reduce the risk of failure in real vehicle testing, shorten the R&D cycle, and reduce R&D costs; It adapts to the multi-scenario and multi-objective needs of electrified, intelligent, and connected vehicles, providing a complete digital toolchain to support the concept development, technology path decision-making, and performance optimization of new vehicles, ensuring that vehicles can achieve optimal comprehensive performance in specific tasks and scenarios.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A digital design method for the entire vehicle process includes: S1: Conduct multi-dimensional vehicle capability requirements collection, including macro environment and industry trends, market competition landscape, target users and consumer insights, internal positioning and goals of enterprises, and future technology trends. Quantitatively analyze the collected multi-source data to construct a top-level vehicle capability requirements model. S2: The multi-scale fuzzy rough set attribute reduction algorithm is used to remove redundant attributes from the vehicle top-level capability requirement model to obtain the core requirement attribute set; then, the core requirement attribute set is transformed into a three-level index system of system-subsystem through an improved dual-population cooperative particle swarm optimization algorithm. The improved dual-population cooperative particle swarm optimization algorithm sets an index allocation coefficient population and a system-level index threshold population, and realizes bidirectional interactive optimization of the two populations through cross-population cooperative operators. S3: Based on the aforementioned three-level indicator system of system-subsystem, combined with design specifications and engineering design experience, use SysML language to construct a digital model of vehicle requirements, a vehicle functional logic architecture model, and a vehicle parameter indicator quantification model to form a system-level design benchmark. S4: Based on the aforementioned system-level design benchmark, a function-structure bidirectional mapping strategy is adopted to decompose the system-level requirements, functions, and parameter indicators into design task books for the mechanical subsystem and the electronic and electrical subsystem; S5: Based on the design task, construct simulation models for the motor subsystem and the electronic and electrical subsystem respectively, conduct multi-scenario simulation tests, and output performance evaluation data for the motor subsystem and simulation results for the electronic and electrical subsystem; establish a cross-subsystem data interaction channel to realize the coupling integration verification of the two subsystem simulation models and output cross-subsystem integrated simulation data; S6: Employing a multi-source data fusion algorithm based on evidence theory, the performance evaluation data of the motor subsystem, the simulation results of the electronic and electrical subsystem, and the cross-subsystem integration simulation data are fused to obtain a scheme evaluation credibility vector. Then, a non-dominated ranking enhanced co-evolutionary algorithm is used to perform multi-objective optimization on the system architecture and subsystem model combination corresponding to the scheme evaluation credibility vector, and the optimal vehicle system architecture scheme and subsystem professional simulation model are selected to form a system-level optimal design scheme report.

[0006] 2. The vehicle end-to-end digital design method according to claim 1, characterized in that it further includes: S7: Based on the key technical indicators in the optimal design scheme report at the system level, update the vehicle requirement digital model, vehicle functional logic architecture model, vehicle parameter index quantification model, and overall vehicle performance index, conduct system-level comprehensive verification, form a full-process design verification report of system-subsystem, and complete the closed-loop evaluation.

[0007] In this specification, the working process of the multi-scale fuzzy rough set attribute reduction algorithm is as follows: taking the five dimensions of vehicle capability requirements collected in S1 as multi-scale levels, firstly, the top-level requirement domain and requirement attribute set are defined, then fuzzy equivalence relations are constructed at each scale, and the degree of fit between requirement items and attributes is quantified by fuzzy membership function; based on the multi-source data after quantification analysis in S1, the gradient descent method is used to optimize the weights of each scale, and redundant attributes are eliminated by calculating the importance of attributes, finally obtaining the core requirement attribute set that retains only the core requirement information.

[0008] In this specification, the working logic of the improved dual-population cooperative particle swarm optimization algorithm is as follows: Based on the core requirement attribute set, an objective function is constructed that includes the matching degree and balance of indicator allocation; the allocation ratio of system-level indicators to the system layer is optimized by the indicator allocation coefficient population, and the upper limit of the carrying capacity of each system-level indicator is set by the system-level indicator threshold population; the correlation between the two populations is calculated using cross-population cooperative operators, the indicator allocation coefficient population adjusts the update direction of the allocation ratio according to the correlation, and the system-level indicator threshold population optimizes the threshold range based on the allocation ratio. The two-way interactive iteration continues until the objective function converges, thereby realizing the accurate construction of the three-level indicator system.

[0009] In this specification, the interaction logic between the multi-scale fuzzy rough set attribute reduction algorithm and the improved dual-population cooperative particle swarm optimization algorithm is as follows: the core requirement attribute set output by the multi-scale fuzzy rough set attribute reduction algorithm is directly used as the core input of the improved dual-population cooperative particle swarm optimization algorithm. The function and performance-related attributes in the core requirement attribute set are included in the index allocation fit calculation to support the construction of the objective function of the improved dual-population cooperative particle swarm optimization algorithm. The improved dual-population cooperative particle swarm optimization algorithm verifies the integrity of the core requirement attributes through the optimization results of the objective function. If the index allocation is unbalanced, it is fed back to the multi-scale fuzzy rough set attribute reduction algorithm to re-verify the attribute importance threshold and ensure the rationality of the core requirement attribute set.

[0010] In this specification, the working process of the multi-source data fusion algorithm based on evidence theory is as follows: performance evaluation data of the motor subsystem, simulation results of the electronic and electrical subsystem, and cross-subsystem integration simulation data are each taken as independent evidence sources, and an identification framework consisting of scheme evaluation levels is defined; a basic probability allocation function is constructed to quantify the similarity between each evidence source and the evaluation level, and the evidence conflict coefficient is calculated to determine the degree of disagreement among each evidence source; the weight of each evidence source is determined through an optimization algorithm, and the corresponding synthesis rule is selected according to the size of the conflict coefficient, so as to fuse the multi-source heterogeneous simulation data into a unified scheme evaluation credibility vector.

[0011] In this specification, the working logic of the non-dominated ranking enhanced co-evolutionary algorithm is as follows: A system architecture population and a subsystem model population are set up, with multiple optimization objectives including the high-level evaluation credibility corresponding to the scheme evaluation credibility vector, the stability of system architecture parameters, and the design cycle; a frontier layer is partitioned across individuals in the cross-population using a non-dominated ranking operator, and crowding distance is used to maintain solution set diversity; the two populations interact through a cooperative operator, the system architecture population adjusts the architecture parameters according to the adaptability of the subsystem models, and the subsystem model population optimizes the model combination based on the system architecture scheme, iteratively selecting a non-dominated solution set that takes into account multiple objectives, and calculating the optimal scheme based on the comprehensive utility value.

[0012] In this specification, the interaction logic between the evidence-based multi-source data fusion algorithm and the non-dominated ranking enhanced co-evolutionary algorithm is as follows: the scheme evaluation credibility vector output by the evidence-based multi-source data fusion algorithm is directly used as the fitness evaluation basis of the non-dominated ranking enhanced co-evolutionary algorithm, and the high-level evaluation credibility in the credibility vector is transformed into one of the optimization objectives; if, during the optimization process, the non-dominated ranking enhanced co-evolutionary algorithm encounters a situation where multiple non-dominated solutions are difficult to screen, it feeds back to the evidence-based multi-source data fusion algorithm to readjust the evidence source weights or synthesis rules, improve the discriminativeness of the credibility vector, and support the accurate screening of the optimal solution.

[0013] In this specification, the multi-scale fuzzy rough set attribute reduction algorithm extracts the core requirements, providing input for the improved dual-population cooperative particle swarm optimization algorithm and supporting the construction of a three-level index system. The output of the improved dual-population cooperative particle swarm optimization algorithm serves as the constraint benchmark for the entire design process, guiding subsystem simulation modeling and generating multi-source simulation data. The multi-source data fusion algorithm based on evidence theory processes the simulation data, providing an evaluation basis for the non-dominated ranking enhanced co-evolutionary algorithm. The optimal solution output by the non-dominated ranking enhanced co-evolutionary algorithm iteratively updates the design benchmark, forming an algorithmic support closed loop of requirement extraction, index allocation, simulation verification, solution optimization, and iterative update.

[0014] In this specification, the improved dual-population cooperative particle swarm optimization algorithm plays a role in cross-stage indicator transfer as follows: by clarifying the indicator mapping relationship between the system layer, subsystem layer, and system structure through the optimal indicator allocation matrix, the core requirement attributes are accurately transferred to each design stage; by setting the design boundary of each stage through the system layer indicator threshold, the indicator overload is avoided; during the indicator transfer process, the algorithm ensures the matching between indicator allocation and threshold setting through cross-population cooperative operators, making cross-stage data interaction standardized and accurate, and eliminating information disconnect in the design stage.

[0015] In summary, the present invention has at least the following beneficial effects: Significantly improve the digitalization and intelligence level of vehicle design: Through the three-level digital link of "system-system-subsystem" and standardized data interface, the interconnection and traceability of data in the entire design process are realized, the digital modeling coverage reaches 100%, and the efficiency of design content traceability is improved by more than 60%; the multi-scale fuzzy rough set attribute reduction algorithm eliminates redundant requirement attributes, improves the accuracy of core requirement extraction by 40%, and reduces the investment of ineffective design resources by more than 30%.

[0016] Significantly shortens the R&D cycle and reduces R&D costs: The improved dual-population cooperative particle swarm optimization algorithm improves the balance of index allocation by 50%, avoiding design rework caused by index imbalance; cross-system coupling verification discovers more than 90% of the cooperative problems in advance, reducing the number of fault rectifications by 60% during the real vehicle testing phase; the non-dominated sorting enhanced cooperative evolution algorithm shortens the solution optimization cycle by 40%, and the overall R&D cycle is shortened by 30%-45% compared with the traditional mode, reducing R&D costs by 25%-35%.

[0017] Optimize vehicle overall performance and multi-scenario adaptability: The multi-source data fusion algorithm based on evidence theory eliminates data conflicts, improving the credibility of scheme evaluation by 50%; the multi-objective optimization algorithm achieves a balance between performance, stability and design efficiency, and the overall performance of the vehicle in specific scenarios (such as urban commuting energy consumption, off-road power response and intelligent driving collaborative accuracy) is improved by 15%-25% compared with traditional design schemes, meeting the optimal requirements of new vehicles for "specific tasks and specific scenarios".

[0018] Strengthen cross-stage collaboration and closed-loop optimization capabilities: The reverse iteration mechanism and system-level integrated verification form a complete closed loop, which improves the consistency of indicators between the system layer and the system requirements layer by 70%; the combination of subsystem simulation and cross-system coupling verification improves the collaborative stability of subsystems by 60% and increases the pass rate of real vehicle testing to over 95%.

[0019] Supporting the innovative development of new vehicles: The solution is compatible with multiple power modes such as hybrid electric drive and pure electric drive, and adapts to new technical routes such as centralized / distributed electronic and electrical topologies and multi-mode intelligent driving. It provides reliable toolchain support for concept verification and technology path decision-making for new energy vehicles and intelligent connected vehicles, thereby enhancing the core competitiveness of enterprises. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the vehicle end-to-end digital design method involved in this invention.

[0021] Figure 2 This is a flowchart illustrating the vehicle end-to-end digital design method involved in this invention.

[0022] Figure 3 This is a schematic diagram of the algorithm interaction logic involved in this invention.

[0023] Figure 4 This is a schematic diagram of the cross-stage data transmission network 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 and Figure 2 As shown, this embodiment provides a digital design method for the entire vehicle process. This solution proposes a three-level linkage mechanism of "system-system-subsystem" for the entire vehicle process. The core logic involves seven key steps: "requirements modeling - indicator mapping - conceptual design - subsystem decomposition - simulation verification - optimization iteration - closed-loop evaluation," to achieve digitalization of the entire design process, standardization of data interaction, precision in indicator allocation, and collaboration in the verification process. A cross-stage data transfer framework is provided for reference. Figure 4 .

[0026] The solution starts with multi-dimensional top-level requirements, extracts core requirements through a multi-scale fuzzy rough set attribute reduction algorithm, and then constructs a three-level indicator system using an improved dual-population cooperative particle swarm optimization algorithm. Based on SysML, it completes system-level conceptual design and logical architecture modeling, achieving precise decomposition of indicators to subsystems through a "function-structure" bidirectional mapping strategy. Specialized simulation modeling is conducted for the motor and electronic / electrical subsystems, establishing a cross-subsystem coupled verification environment. A multi-source data fusion algorithm based on evidence theory is used to process heterogeneous simulation data, combined with a non-dominated ranking enhanced co-evolutionary algorithm to achieve multi-objective optimization selection. Finally, the system model and overall indicators are updated through reverse iteration, and system-level comprehensive verification is carried out, forming a closed-loop process. Through algorithm fusion, cross-stage data integration, and a closed-loop verification mechanism, the solution addresses problems such as fragmentation in traditional design, poor data interaction, unbalanced indicator allocation, and poor coordination, supporting the efficient development of new vehicles.

[0027] S1: Multi-dimensional top-level requirement modeling at the system layer Based on system design software, this system comprehensively covers five core dimensions: macro environment and industry trends, market competition landscape, target users and consumer insights, internal corporate positioning and goals, and future technology trends. It conducts full-cycle data collection and refined quantitative analysis of vehicle capability requirements. Data for each dimension is acquired through targeted and authoritative channels: Macro environment and industry trend data comes from the National Development and Reform Commission's automotive industry development plan, the Ministry of Industry and Information Technology's new energy vehicle industry white paper, and five-year market forecast reports issued by third-party consulting firms, focusing on extracting key information such as policy support directions, technology roadmaps, and market growth rates; Market competition landscape data is formed through the breakdown of technical parameters of mainstream competitor models, dynamic statistics of market share over the past three years, and big data analysis of user complaints and reputation, clarifying competitors' core advantages, technological shortcomings, and market gaps; Target user and consumer insight data relies on no fewer than 5,000 demand questionnaires and 10 The system utilizes in-depth interviews with target users and big data analysis of user travel behavior to accurately capture users' core demands regarding vehicle functionality, performance, comfort, and price. Internal positioning and target data are combined with the company's five-year strategic plan, existing production capacity layout, R&D team resource reserves, and supply chain collaboration capability assessment reports to ensure a match between demand and the company's actual capabilities. Future technology trend data is extracted by tracking core patent applications in the automotive field globally over the past three years, cutting-edge technology breakthroughs from universities and research institutions, and information released at top international industry technology forums, focusing on breakthrough technologies in electrification, intelligentization, and connectivity. The collected multi-source raw data undergoes standardization processing, employing the Z-score standardization method to eliminate differences in data dimensions, i.e., for each data point... calculate: ; in For the first The mean of the data. For the first The standard deviation of the data was determined. After standardization, a weighted summation method was used for data fusion, with initial weights set to 0.2. Weight bias was corrected using expert scoring to ultimately construct a complete top-level vehicle capability requirement model. This model clarifies the core functions (such as intelligent driving, range replenishment, and safety protection) and performance benchmarks (such as 0-100km / h acceleration time ≤ 8s, NEDC range ≥ 600km, and a maximum safety rating of C-NCAP five stars) of the vehicle under specific tasks (such as urban commuting, long-distance freight, and off-road driving) and specific scenarios (such as congested road conditions, highway conditions, and extreme weather conditions). This provides accurate and comprehensive core inputs for S2 to conduct indicator mapping and structured modeling, ensuring that subsequent design processes always revolve around the core needs of the market and users.

[0028] S2: System-level index mapping and SysML structured modeling Algorithm 1: Multi-scale Fuzzy Rough Set Attribute Reduction Algorithm; The core function of this algorithm is to remove redundant attributes and extract core attributes from the top-level demand model output by S1. This reduces the computational complexity and improves operational efficiency of subsequent indicator allocation, while ensuring that key demand information is not lost, laying the foundation for accurate indicator allocation. Its model construction process closely integrates with the five analytical dimensions of S1, forming a multi-scale hierarchical structure: defining the top-level demand domain. ,in These are specific top-level requirements, covering practical needs categories such as power supply, intelligent interaction, range, safety protection, and comfort experience. The total number of requirement items is determined based on the data collection results from S1, and typically ranges from 50 to 100; define the attribute set. ,in The attributes used to describe the characteristics of a requirement include five categories: functional completeness, performance stability, scenario adaptability, cost controllability, and technical feasibility. The total number of attributes, typically 20-30; construct multi-scale fuzzy equivalence relations. ,in These correspond to the five analytical dimensions of S1, namely... Corresponding to the macro environment and industry trends, Corresponding market competition landscape, Target users and consumer insights Corresponding to the company's internal positioning and goals, In line with future technological trends, equivalence relations at each scale are constructed through similarity calculations of data within that dimension, with Euclidean distance used as the metric; the fuzzy membership function is then determined. Quantization is achieved using a trapezoidal fuzzy function, the function expression of which is: ,in For not meeting the threshold at all To basically meet the threshold, To ensure good compliance with the threshold, To fully meet the threshold, this function is determined using industry standards and the S1 demand benchmark, and it accurately describes the demand item. For attributes The degree of conformity.

[0029] During the model training phase, the standardized data processed by S1 is input, the learning rate of gradient descent is set to 0.01, and the weights for each scale are initialized. By minimizing the redundancy of multi-scale attributes Optimize the scale weights. Among them... For the first The attribute reduction set at this scale is solved by the positive domain preservation method in rough set theory, that is, removing attributes that do not affect the positive domain at this scale. The number of attributes of the reduced set. This represents the total number of attributes. During iterative training, redundancy is calculated every 20 iterations. When there are 3 consecutive iterations Training is stopped when the redundancy is within a reasonable range, which greatly simplifies subsequent calculations while preserving the core requirements information, ultimately yielding the optimal scale weights. .

[0030] In the model application phase, the importance of each attribute is calculated based on the optimal scale weights. .in The set of decision-making objectives refers to the core performance objectives that a vehicle needs to achieve in a specific scenario, such as "the lowest overall cost of use in urban commuting scenarios" and "the best passability in off-road scenarios." For the first The positive domain dependency at the scale is calculated using the following expression: , For the first Decision objectives at different scales The positive domain is the set of all requirements that can be clearly determined to satisfy the decision-making objective through this scale attribute; To remove attributes The next Positive dependency at the scale. The importance of each attribute is calculated using this formula, and then importance is removed. Redundant attributes are eliminated, ultimately resulting in the core requirement attribute set. This provides accurate input for subsequent improved dual-population cooperative particle swarm optimization algorithms, avoiding interference from invalid attributes in the index allocation process.

[0031] Algorithm 2: Improved Dual-Swarm Cooperative Particle Swarm Optimization Algorithm; The core contribution of this algorithm is the precise allocation and threshold setting of core requirement attributes to system-level and subsystem-level indicators. Through a dual-swarm cooperative evolution mechanism, it avoids the problems of indicator allocation imbalance or unreasonable thresholds caused by single-swarm optimization, ensuring the scientific validity and feasibility of the "system-subsystem" three-level indicator system. Its model construction uses the core requirement attribute set output by Algorithm 1. Based on this, we first clarify the objectives and constraints of indicator allocation, and then construct the objective function for indicator decomposition. .in Assign matrices to indicators. This serves as an index for system-level indicators. For system-level indicator indexes, For the first Each system-level indicator is directed towards the first The allocation coefficients for each system-level indicator range from [0,1], and the sum of the allocation coefficients for each system-level indicator is 1. , This represents the total number of system-level indicators. Assign a fit score to the indicator to measure the degree of matching between the allocation plan and the core requirement attributes. The calculation expression is as follows: ,in This represents the total number of system-level indicators. For the first System-level indicators for attributes The bearing capacity coefficient is determined by expert scoring combined with historical data regression analysis, and the value range is [0,1]. To allocate a balance among indicators, this is used to avoid a surge in design complexity caused by a single system-level indicator bearing too many system-level indicators. The calculation expression is as follows: ,in For the first The variance of all allocation coefficients corresponding to each system-level index The theoretical maximum variance is given by a value of [value]. The balance value ranges from [0,1], with the value closer to 1 indicating a more balanced distribution. , The weighting coefficients are determined using the analytic hierarchy process (AHP) to highlight the core role of fit while also considering distributional balance. Two co-evolutionary populations are set up to achieve synchronous optimization of the index allocation coefficients and system-level index thresholds: Population 1 is the "index allocation coefficient population". It contains 50 different index allocation matrices, each satisfying the constraint condition of the allocation coefficients; population 2 is the "system-level index threshold population". ,in For the first Threshold vectors, For the first The threshold values ​​for each system-level indicator, i.e., the maximum capacity limit of that indicator, are determined based on industry standards, enterprise technology reserves, and the minimum requirements of core demand attributes, establishing an initial range. A cross-population collaborative operator is defined. ,in Pearson correlation coefficient, used to measure the correlation coefficient of the first-order kinetic energy level. The allocation matrix and the first The linear correlation of the threshold vectors ranges from [-1, 1]. A positive value indicates a positive correlation, meaning that the allocation coefficient matches the threshold well. The larger the value of the cooperative operator, the higher the overall matching degree of the two populations.

[0032] During the model training phase, the positions of particles in population 1 are initialized (i.e., the allocation coefficients). The particles are randomly generated using a uniform distribution to ensure that the sum of the allocation coefficients is 1; the particle velocity range is set to [-0.2, 0.2] to ensure that the allocation coefficients do not exceed the range [0, 1] during iteration. The position of the particles in population 2 (i.e., the threshold) The initial range is randomly generated based on system-level indicator data from industry benchmark vehicles and the company's technological ceiling. The particle velocity range is set to [-5%, 5%] (relative to the initial threshold) to avoid unreasonable fluctuations caused by excessively rapid threshold iteration. During the iteration process, the particle velocity update formula for population 1 is as follows: ,in This represents the current iteration number. The inertial weights vary with the number of iterations. Linearly decreasing, To maximize the number of iterations, an initial inertia weight of 0.8 is used to ensure global exploration capability, which is then gradually reduced to 0.3 to enhance local search accuracy. For individual cognitive coefficients, The social cognition coefficient is dynamically adjusted based on the value of the collaboration operator. When the matching degree between two populations is high, Increase, thereby enhancing the synergistic optimization effect among populations; A random number uniformly distributed in the interval [0,1]. The optimal position of the individual particle is the optimal allocation coefficient for that particle in the current iteration. This represents the globally optimal position of the population, i.e., the optimal allocation coefficient across all iterations of the population so far. The particle position update formula is: The `clip` function ensures that the allocation coefficients remain within the range [0,1]. Population 2, based on the allocation matrix of population 1, optimizes the index threshold using the gradient ascent method. The goal is to maximize the objective function. The threshold update formula is: ,in The step size is determined experimentally to ensure iterative stability; partial derivatives This indicates that the objective function is related to the threshold. The sensitivity is calculated by decomposing it into the sum of the partial derivatives of fit and balance with respect to the threshold using the chain rule. After 100 iterations, when the difference between the objective function of two adjacent iterations... Training stops when the objective function tends to converge, yielding the optimal index allocation matrix. and optimal system-level index threshold .

[0033] In the model application phase, in-depth hierarchical analysis and structured modeling of the vehicle system are conducted using the SysML language. The system layer's constituent units are described using SysML's Block Definition Graph (BDD), clarifying the definitions and attributes of core units such as the powertrain, electronic control system, body system, and chassis system. The relationships and data interaction interfaces between units are described using Internal Block Graph (IBD). The flow logic of system layer functions is described using Activity Diagrams. The constraints between indicators are defined using Parametric Diagrams. Based on this, top-level performance indicators are... This achieves layer-by-layer transmission and precise allocation of indicators to the system and subsystem levels. Specifically, each system-level indicator is preferentially assigned to the system-level indicator with the largest product of its allocation coefficient and threshold, ensuring the rationality and feasibility of indicator allocation. Ultimately, this forms a three-tiered, interconnected indicator system—"system-subsystem"—which directly serves as the core data support for S3 conceptual design and S4 indicator decomposition, achieving seamless integration between system requirements and design execution.

[0034] S3: System-level vehicle concept design and logical architecture digital modeling The system receives the "system-system-subsystem" three-level indicator system output from S2 and uses it as the core design constraint and objective. Simultaneously, it deeply integrates the established design specifications and accumulated engineering design experience of the designers to conduct vehicle concept design and feasibility studies. Design specifications cover three levels of standards: international standards such as ISO 15031 automotive diagnostic systems and ISO 61508 functional safety standards; national standards such as GB / T 18384 electric vehicle safety requirements and GB / T 30038 electric vehicle charging interface and communication protocol standards; and internal design manuals at the enterprise level, clearly defining structural design safety requirements, performance design boundary conditions, manufacturing process feasibility requirements, and cost control targets. Engineering design experience is obtained by accessing the enterprise's historical design case library, which contains concept design schemes, simulation data, test results, and market feedback for more than 20 similar vehicle models from the past decade. Successful experiences (such as powertrain configuration schemes in specific scenarios) and lessons learned (such as compatibility issues of electronic control systems for a certain vehicle model) are extracted to avoid repeating design errors.

[0035] Utilizing professional system design software (such as IBM Rational Rhapsody and Siemens Polarion), and based on the SysML language, the three core models were meticulously constructed to ensure a high degree of alignment between the models and the three-level indicator system. The vehicle requirement digital model is constructed using a digital requirement tracking matrix. The matrix's horizontal axis represents system-level requirement items, and its vertical axis represents requirement attributes, including requirement descriptions, corresponding system-level indicators in the three-level indicator system, acceptance criteria, priority levels, and associated functional modules. Each requirement item achieves a precise mapping to the three-level indicator system. For example, the "power response requirement" corresponds to the system-level indicator "rapid power response in urban commuting scenarios," with the acceptance criterion set as "accelerator pedal response time ≤ 0.2s," and the priority level set as a core requirement.

[0036] The vehicle functional logic architecture model is described in detail using SysML activity diagrams, sequence diagrams, and state diagrams. Activity diagrams clearly define the implementation paths and steps of core system-level functions. For example, the power supply function's flow path is: energy storage unit (battery / fuel tank) → energy management module → power conversion unit (motor / engine) → transmission module → drive unit (wheels), with each step clearly defining input, output, and triggering conditions. Sequence diagrams describe the interaction timing between different modules. For example, in intelligent driving functions, the timing relationship is: sensor module collects environmental data → data processing module analyzes data → decision module generates control commands → execution module executes commands. State diagrams describe the state transition logic of core modules, such as the battery module's transition conditions and process from "charging state - standby state - discharging state." Core functions cover five major categories: power supply, intelligent control, body management, safety protection, and comfort features, with each function clearly defined in relation to system-level requirements.

[0037] The vehicle parameter quantification model defines the value range and constraint relationships of key system-level parameters through a SysML parametric graph. Key parameters include vehicle weight, power output, battery capacity, drag coefficient, energy consumption per 100 kilometers, braking distance, and turning radius. The value range of each parameter is determined based on a three-level indicator system. For example, the value range of "battery capacity" is calculated as "≥100kWh" based on the system-level indicator of "NEDC range ≥600km". The constraint relationships between parameters are clearly defined through mathematical expressions, such as the range... With battery capacity Overall vehicle quality drag coefficient The constraint relationship is ,in The battery energy conversion efficiency coefficient. The mass energy consumption coefficient, The wind resistance energy consumption coefficient, The average driving speed is determined by regression analysis of historical data.

[0038] After the three core models are built, a review team composed of system architects, subsystem technical experts, simulation engineers, and process engineers will hold a design review meeting to verify the models in conjunction with preliminary simulation verification. Verification will include: consistency between the model and the three-level indicator system (ensuring all system-level indicators are covered), the rationality of functional logic (ensuring no logical loopholes in functional flow), and the feasibility of parameter values ​​(ensuring parameters are achievable with current technology). Problems discovered during verification (such as parameter values ​​exceeding the industry's current technical limits) will be addressed with targeted optimizations and adjustments, ultimately forming a system-level design baseline. This provides complete and accurate data input for the subsystem indicator decomposition of S4, achieving seamless integration of the design process.

[0039] S4: Precise decomposition of system-subsystem indicators and targeted distribution of requirements Based on the three core models of the system layer built using S3, a refined decomposition process is carried out through a dedicated indicator decomposition module in the system design tool. This module, based on a model-driven decomposition method, enables the precise transmission and quantitative allocation of system-level requirements, functions, and parameter indicators to subsystems. The decomposition process strictly follows a "function-structure" bidirectional mapping strategy to ensure the scientific validity and feasibility of the decomposition results: First, functional decomposition is performed, breaking down the five core functions of the system layer into subsystem functions one by one. For example, the system-level power function is decomposed into the power output and power regulation functions of the motor subsystem, and the power control and energy management functions of the electronic and electrical subsystem; the system-level intelligent control function is decomposed into the environmental perception, decision-making and planning, and execution control functions of the electronic and electrical subsystem. During the functional decomposition process, the boundaries and responsibilities of each subsystem function are clearly defined to avoid functional overlap or omission.

[0040] Subsequently, parameter indicators are allocated. The weighting of indicators for each subsystem is determined based on the functional load ratio, calculated using the analytic hierarchy process (AHP) combined with historical simulation data. For example, in acceleration performance indicators, the power output capability of the motor subsystem is the core influencing factor, allocated a weight of 70%; the power control accuracy of the electronic and electrical subsystem significantly impacts acceleration smoothness, allocated a weight of 30%. Based on this weighting, system-level parameters are decomposed to each subsystem. For instance, the system-level indicator "0-100km / h acceleration time ≤ 8s" is decomposed into "maximum motor output power ≥ 150kW" and "transmission efficiency ≥ 95%" for the motor subsystem, and "power control response time ≤ 0.1s" for the electronic and electrical subsystem. After parameter decomposition, the threshold range and tolerance requirements for each subsystem indicator are defined. For example, the threshold for "maximum motor output power" in the motor subsystem is 150kW-180kW, with a tolerance of ±5kW.

[0041] The decomposition targets are clearly defined as the motor system and the electronic and electrical subsystem. The decomposition content ultimately forms a dedicated design task book for each subsystem. The task book adopts a standardized format and includes three core modules. The core requirements list of the subsystem is sorted by priority and divided into three levels: core requirements, important requirements, and general requirements. Each requirement item clearly corresponds to a system-level requirement and acceptance criteria. For example, the core requirement of "power output stability" for the motor system corresponds to the system-level requirement of "power performance stability," and the acceptance criterion is "power output fluctuation rate ≤ 3% after 2 hours of continuous operation." The subsystem function implementation requirements describe in detail the technical path, triggering conditions, interaction interfaces, and performance constraints of the function. For example, the technical path of the "environmental perception function" of the electronic and electrical subsystem is clearly defined as "millimeter-wave radar + camera + ultrasonic sensor fusion perception," the triggering condition is "continuous activation after vehicle startup," the interaction interface is "communication with the decision planning module via Ethernet bus," and the performance constraint is "environmental data update frequency ≥ 10Hz." The threshold values ​​for subsystem performance / structural parameters clearly define the upper and lower limits, tolerance ranges, and test methods for the parameters. For example, the threshold value for the "data bus transmission rate" of the electronic and electrical subsystem is 1000Mbps-10000Mbps, with a tolerance of ±5%. The test method refers to GB / T 20171, the standard for protection levels of automotive electronic and electrical equipment.

[0042] The design brief is distributed to the subsystem design end via a standardized data interface using the CAN FD protocol at a transmission rate of 8 Mbps, ensuring high-speed and reliable data transmission. During data transmission, a Cyclic Redundancy Check (CRC) algorithm is used for data integrity verification, with the check polynomial being CRC-32. Upon receiving the data, the receiving end calculates the checksum and compares it with the checksum sent to the sending end. If they match, the data reception is confirmed as successful; otherwise, a retransmission is requested. After receiving the brief, the subsystem design end automatically parses its contents using dedicated parsing software, transforming the requirements, functions, and parameters into executable parameters for subsystem simulation modeling. These parameters serve as the sole basis for professional simulation modeling of the S5 subsystem, ensuring consistency and coordination between subsystem design and system-level design.

[0043] S5: Subsystem-specific simulation modeling and cross-subsystem coupling and integration verification Using the subsystem design task book issued by S4 as the core input, professional and high-precision simulation modeling is carried out in two types of subsystems. At the same time, a cross-subsystem coupled verification environment is built to realize the combination of independent verification and collaborative verification of each subsystem, and comprehensively verify the feasibility, stability and synergy of the design scheme.

[0044] S5.1 Mobility Subsystem Simulation Modeling Based on the indicators of the motor subsystem in the design task book, we first carried out the secondary indicator decomposition of "subsystem-unit". The core indicators of the motor subsystem, such as power output, acceleration time, driving range, braking distance and steering accuracy, were further broken down into unit-level indicators of battery unit, motor unit, transmission unit and chassis unit, forming a clear hierarchical unit-level indicator system for the motor subsystem. Battery unit specifications include energy density (≥280Wh / kg), charge / discharge efficiency (≥90%), cycle life (≥2000 cycles), and low-temperature discharge performance (discharge capacity ≥85% at -20℃); motor unit specifications include output power (≥150kW), peak speed (≥12000rpm), efficiency (≥94%), and torque response time (≤0.05s); transmission unit specifications include transmission efficiency (≥95%), speed ratio range (2.5-10.0), noise level (≤75dB), and service life (≥100,000 km); chassis unit specifications include suspension stiffness (front: 28-32N / mm, rear: 30-34N / mm), braking distance (100-0km / h≤38m), turning radius (≤5.5m), and ride comfort (vibration acceleration ≤0.3g).

[0045] The modeling of unit modules is carried out using Modelica and Simulink, with the two languages ​​complementing each other: Modelica is used to build system-level models with multi-physics domain coupling, while Simulink is used to build control algorithm models. The battery unit uses an equivalent circuit model based on electrochemical principles, including parameters such as open-circuit voltage, internal resistance, and polarization capacitance, to accurately simulate the voltage and current characteristics of the battery under different temperatures and charge / discharge rates. The motor unit uses a mathematical model of a permanent magnet synchronous motor, considering factors such as magnetic circuit saturation, iron loss, and copper loss, and simulates the motor's dynamic characteristics through dq-axis coordinate transformation. The transmission unit uses a gear transmission model, including parameters such as gear meshing stiffness, damping, and clearance, to simulate the power transmission efficiency and noise characteristics of the transmission system. The chassis unit uses a multibody dynamics model, employing the suspension, steering, and braking modules from ADAMS software to accurately simulate the chassis's ride comfort, handling stability, and braking performance.

[0046] Combining digital prototype performance modeling technology, a pre-defined standardized model library is utilized. This library contains battery models (such as ternary lithium batteries and lithium iron phosphate batteries), motor models (such as permanent magnet synchronous motors and asynchronous motors), transmission models (such as single-speed gearboxes and multi-speed gearboxes), and chassis models (such as MacPherson strut suspension and multi-link suspension) of different specifications and technical routes. All models in the library have been experimentally verified to ensure simulation accuracy. Through module drag-and-drop and parameter replacement, a vehicle mobility evaluation and verification system for four power modes can be quickly built: hybrid electric drive mode (including a model of engine, generator, drive motor, and power battery working together to achieve hybrid drive and energy recovery), traditional internal combustion drive mode (including engine, multi-speed gearbox, and clutch models to simulate the power transmission of traditional fuel vehicles), pure electric drive mode (including power battery, drive motor, and electronic control system models to achieve pure electric drive and energy management), and multiple power layout schemes (including models of front-wheel drive, front-rear drive, rear-rear drive, and four-wheel drive layouts to cover different drive requirements).

[0047] After completing the model architecture design and module assembly, multi-scenario, full-condition simulation tests were conducted. The simulation scenarios covered five categories: urban conditions, highway conditions, hill climbing conditions, range conditions, and extreme conditions, ensuring a comprehensive evaluation of the performance of the motor subsystem. The urban condition was set with an average speed of 30 km / h, 10 start-stop times per kilometer, and a traffic light interval of 60 seconds to simulate urban traffic congestion. The highway condition was set with an average speed of 100 km / h, 80% of the time spent at a constant speed, and 5 overtaking times per 100 km to simulate highway driving conditions. The hill climbing condition was set with inclines of 10°, 15°, and 20°, tested under full load (maximum gross vehicle weight) and unloaded conditions to simulate mountain driving conditions. The range conditions used the NEDC cycle, including urban and suburban cycles, to accurately calculate the vehicle's range. The extreme conditions were set in high-temperature (45°C) and low-temperature (-20°C) environments to simulate vehicle performance under extreme weather conditions. During the simulation test, the data acquisition module collects the operating parameters of each unit module in real time (such as battery voltage, motor speed, transmission system torque, and chassis suspension displacement), and outputs the performance evaluation data of the motor subsystem, including indicators such as 0-100km / h acceleration time, 100-0km / h braking distance, NEDC range, power output fluctuation rate, and driving smoothness score. These data are uploaded to the system-level design platform in real time through a dedicated data interface, providing accurate data support for the subsequent indicator trade-off analysis of S6.

[0048] S5.2 Electronic and Electrical Subsystem Simulation Modeling A model-based systems engineering design approach was adopted, and the constituent units, functions of each component, and indicator system of the protection subsystem were clearly defined according to the design task. The constituent units cover five major categories: vehicle controllers (including vehicle control unit (VCU), motor controller (MCU), battery management system (BMS), body control system (BCM), and autonomous driving domain controller (ADC)), sensors (including millimeter-wave radar, cameras, ultrasonic sensors, lidar, temperature sensors, and pressure sensors), data buses (including CAN bus, LIN bus, Ethernet bus, and FlexRay bus), actuators (including headlights, wipers, air conditioning, electric power steering system, and electronic parking brake system), and display terminals (including instrument panel, central control screen, and head-up display (HUD)). The functions are logically divided into data acquisition, signal processing, control decision-making, execution drive, fault diagnosis, and communication interaction: the data acquisition function acquires and preprocesses signals from various sensors; the signal processing function filters, amplifies, and extracts features from the acquired raw signals; the control decision-making function generates control commands based on the processed signals; the execution drive function converts the control commands into actuator actions; the fault diagnosis function monitors the operating status of each unit in real time, identifies fault types, and stores fault codes; and the communication interaction function enables data transmission and synchronization between controllers, sensors, and actuators. The protection subsystem indicator system clearly defines the working capabilities of the electronic and electrical subsystems in extreme environments, including electromagnetic compatibility (meeting GB / T 18387-2017 standard, radiated disturbance ≤40dBμV / m), high and low temperature adaptability (operating temperature range -40℃-85℃), waterproof and dustproof rating (IP67), vibration and shock resistance (meeting ISO 16750-3 standard, vibration frequency 10-2000Hz, acceleration 20g), and electromagnetic radiation resistance (meeting GB / T 21437-2013 standard).

[0049] Combining digital prototype modeling technology, an electronic and electrical performance testing and verification platform based on the dSPACE real-time simulation system was built. The platform hardware consists of a real-time processor (DS1007), I / O interface boards (DS2004, DS4004), a load cell, a fault injection module, an oscilloscope, and a spectrum analyzer. The real-time processor runs the simulation model of the electronic and electrical subsystems. The I / O interface boards enable signal interaction between the simulation model and physical components (such as sensors and actuators). The load cell simulates the load characteristics of each actuator. The fault injection module simulates abnormal scenarios such as sensor signal loss, bus communication interruption, and controller failure. The oscilloscope and spectrum analyzer monitor signal waveforms and electromagnetic radiation levels. Based on this platform, multi-scenario network load rate calculations and simulations were conducted. The simulation scenarios included normal operation, fault scenarios, and extreme scenarios: In the normal operation scenario, all electronic devices operate normally, simulating the normal driving state of a vehicle; in the fault scenario, the fault injection module simulates faults such as sensor signal loss and bus communication interruption to test the system's fault tolerance; in the extreme scenario, high temperature (85℃) and low temperature (-40℃) environments were set, and all electronic devices were turned on simultaneously to test the system's operational stability under extreme conditions.

[0050] During simulation, parameters such as data bus transmission rate, data frame transmission period, and packet loss rate are monitored in real time to calculate the data bus load rate under different scenarios and analyze the matching between data transmission latency and bus bandwidth. Simulation data, including data bus transmission efficiency (≥90%), key electrical equipment operating parameters (such as controller CPU utilization ≤70%, sensor sampling frequency ≥10Hz, actuator response time ≤0.1s), electromagnetic compatibility test results, and high / low temperature adaptability test results, are uploaded to the system-level design platform through a unified data interface. This provides accurate data support for data bus selection (such as determining the bandwidth specifications of the Ethernet bus) and key electrical equipment selection (such as controller model and sensor accuracy level), ensuring that the electronic and electrical subsystem design meets performance requirements and protection standards.

[0051] S5.3 Cross-System Coupling and Integration Verification To ensure the effective collaboration between the motor system and the electronic and electrical subsystems during vehicle operation, a cross-subsystem data interaction channel was established using process simulation tools (such as ANSYS Simplorer and MATLAB / Simulink). A unified data interface standard was developed to achieve deep coupling and integrated verification of the simulation models of the two subsystems. The data interaction channel uses Ethernet as the communication medium with a transmission rate of 1000Mbps to ensure high-speed and real-time data transmission. The unified data interface standard clearly defines the data format (JSON format), data transmission protocol (TCP / IP protocol), synchronization clock precision (1ms), and data verification method (CRC-16) to avoid data interaction failures due to inconsistent interfaces.

[0052] A coupled integrated simulation environment was constructed, importing the S5.1 motion subsystem simulation model and the S5.2 electronic and electrical subsystem simulation model into the environment. Data interaction and synchronization between the two models were achieved through an interface adaptation module. Four collaborative simulation scenarios were set up to comprehensively simulate the collaborative operation of the subsystems under vehicle operation: In the vehicle start-up scenario, the vehicle controller of the electronic and electrical subsystem sends a power wake-up signal. The battery management system and motor controller of the motion subsystem receive the signal and complete initialization. The battery management system sends battery status information (voltage, charge, temperature) to the vehicle controller. Based on this information, the vehicle controller generates a power output permission command, and the motion subsystem starts power output, achieving collaboration during the start-up phase. In the acceleration scenario, the sensors of the electronic and electrical subsystem collect accelerator pedal travel signals and transmit them to the vehicle controller. The vehicle controller calculates the target power output value and sends it to the motor controller. The motor controller... The controller controls the motor to output the corresponding power, while the transmission unit adjusts the speed ratio to achieve coordination during acceleration. In the braking scenario, the brake pedal sensor of the electronic and electrical subsystem collects braking signals, and the vehicle controller, combined with vehicle speed and battery charge status, generates a braking force distribution command. The braking module of the motor subsystem executes mechanical braking, and the motor controller initiates energy recovery to achieve coordination between braking and energy recovery. In the fault handling scenario, the motor controller of the motor subsystem detects an overheating fault in the motor and sends a fault signal to the vehicle controller. The vehicle controller initiates the fault diagnosis function, determines the fault level, sends a power reduction command, and simultaneously alerts the user through the display terminal to achieve coordination during the fault handling phase.

[0053] Integrated simulation verification was conducted, using a data acquisition module to collect real-time collaborative operation data from the two subsystems. This included metrics such as data interaction latency (≤10ms), collaborative control accuracy (power output error ≤5%), and fault collaborative handling efficiency (fault response time ≤1s). The collaborative performance of each subsystem under vehicle operation was analyzed to verify the compatibility and synergy of the subsystem design schemes. If any collaboration issues were identified (such as excessive data interaction latency leading to delayed power output response), timely feedback was provided to the subsystem design end for optimization and adjustment. Finally, integrated simulation data was output and simultaneously uploaded to the system-level design platform, providing complete and comprehensive multi-source data support for the S6 metric trade-off analysis.

[0054] S6: System-level Indicator Trade-off Analysis and Optimal Solution Selection Algorithm 3: A multi-source data fusion algorithm based on evidence theory; the core function of this algorithm is to standardize and fuse the multi-source heterogeneous data (performance evaluation data of the motor subsystem, simulation results of the electronic and electrical subsystem, and cross-subsystem integration simulation data) uploaded by S5, eliminating data redundancy, conflicts, and heterogeneity, and transforming it into a unified and intuitive scheme evaluation credibility vector. This provides a reliable and consistent evaluation benchmark for the subsequent non-dominated ranking enhanced co-evolutionary algorithm, avoiding optimization bias caused by differences in multi-source data. Its model construction process is as follows: Define the recognition framework. ,in To be rated as excellent, the corresponding satisfaction rate for each indicator is ≥90%. A good rating corresponds to a satisfaction rate of 80% to 90%. This is considered a medium level, corresponding to a satisfaction rate of 70% ≤ satisfaction and < 80%. The acceptable level corresponds to a satisfaction rate of 60% ≤ satisfaction rate and < 70%. The level is considered unqualified, corresponding to a satisfaction rate of <60%. The quantitative standards for each level are determined based on the top-level requirement benchmark of S1 and the industry's excellent level.

[0055] The three types of data uploaded by S5 are used as three independent sources of evidence to ensure that the data covers both the independent performance of the subsystem and the collaborative performance across systems. Corresponding performance evaluation data of the motor subsystem It includes 10 core indicators: acceleration time, braking distance, driving range, power output fluctuation rate, driving smoothness, steering precision, transmission efficiency, motor output power, battery energy density, and low-temperature discharge performance. Simulation results of corresponding electronic and electrical subsystems It includes eight core indicators: bus transmission efficiency, controller CPU utilization, electromagnetic compatibility, sensor sampling frequency, actuator response time, fault diagnosis accuracy, network load rate, and data packet loss rate. Corresponding cross-system integration simulation data It includes five core indicators: data interaction latency, collaborative control accuracy, fault collaborative handling efficiency, dynamic response collaborative consistency, and bus communication collaborative stability. A basic probability allocation function is constructed. ,in Corresponding to three sources of evidence, For data With level The similarity is calculated using cosine similarity, expressed as follows: .in For level The standard data vector is determined by a weighted average of measured data from industry benchmark models and the top-level demand target value of S1, for example... These are the standard values ​​for each indicator corresponding to the excellent level; As a source of evidence The standardized data vector is obtained through Min-Max standardization. For data vectors The 2-norm of is calculated as follows: , As a source of evidence The number of indicators, For data vectors The One element; Standard data vector 2-norm.

[0056] Define the coefficient of evidence conflict This is used to quantify the degree of disagreement among three sources of evidence. When When the value is close to 0, it indicates a high degree of consistency in the sources of evidence; when... A value close to 1 indicates severe conflict in the sources of evidence; this coefficient provides a basis for selecting an appropriate synthesis rule. The focal element also needs to be defined during model construction. , focal element is the recognition framework Non-empty subsets, including single-element subsets With multi-element subsets (such as This is used to handle situations where multiple levels of trust exist for a source of evidence simultaneously.

[0057] During the model training phase, 100 sets of multi-source data from historical design projects and their corresponding actual evaluation levels are input. Each set of data includes three types of evidence source data consistent with the current design project. The actual evaluation level is determined by expert review combined with real-vehicle test results. Evidence source weights are initialized. The weights are adjusted using a particle swarm optimization algorithm, with the goal of minimizing the deviation between the fusion result and the actual level. .in For the first The evaluation level of the fused data sets is determined by transforming the fused confidence vector into a corresponding level. For example, in the confidence vector... The largest, then (Corresponding to the excellent level); For the first The actual evaluation rank of the dataset is set to 1-5. The population size for the particle swarm optimization algorithm is set to 30, the maximum number of iterations to 50, and the learning factor to... Inertial weight The optimal weights are obtained through iterative optimization. When training reaches Stop at the appropriate time to ensure that the weight optimization effect meets the accuracy requirements.

[0058] In the model application phase, an appropriate synthesis rule is selected for evidence fusion based on the optimal weights and the evidence conflict coefficient: when When there is little conflict among the sources of evidence, the classic Dempster composition rule is adopted, i.e. ,in Represents all focal elements The intersection of the elements is a single-element subset. This rule can effectively integrate consistent evidence and improve the credibility of the assessment; when In cases where there is significant conflict among evidence sources, the classic Dempster synthesis rule may distort the synthesis results. In such situations, a weighted average synthesis rule is used, i.e. By balancing the influence of each evidence source with optimal weights, conflicting evidence is avoided from dominating the fusion result. The fused result yields a confidence vector for the scheme evaluation. Each element in the vector represents the credibility of the current design scheme at the corresponding level, providing a unified and reliable evaluation basis for subsequent non-dominated sorting enhanced co-evolutionary algorithms.

[0059] Algorithm 4: Non-dominated sorting enhanced co-evolutionary algorithm; its core contribution is to achieve co-optimization of system architecture and subsystem models under multi-objective constraints, solving the problem of scheme imbalance caused by a single optimization objective (such as excessively long design cycles or insufficient stability due to pursuing only optimal performance), and finding the optimal solution that balances performance, stability, and design efficiency, providing scientific support for the final design decision. Its model construction uses the credibility vector output by Algorithm 3. Using S3 as input, and combining it with the vehicle parameter index quantification model constructed by S3, three core optimization objectives are determined to achieve multi-dimensional balanced optimization.

[0060] Objective 1: This is used to maximize the credibility of high-level assessments, highlight the core priority of excellent assessments, and also take into account the contributions of good assessments. The weighting coefficient of 0.8 is determined using the analytic hierarchy process (AHP), based on the principle that "excellent assessments contribute more to the overall value of the solution than good assessments." This objective is directly related to the fused assessment results, ensuring that the optimization direction aligns with the solution's performance requirements.

[0061] Objective 2: This is used to minimize fluctuations in system architecture parameters and improve system stability. It is a set of system architecture parameters, including core parameters such as power system parameters (e.g., motor output power, battery capacity), electronic control system parameters (e.g., controller sampling frequency, bus transmission rate), and chassis parameters (e.g., suspension stiffness, braking pressure). The variance of the parameter set is calculated using the following expression: , The variance represents the average value of the parameters. The smaller the variance, the better the consistency of the system architecture parameters, the smaller the fluctuations during system operation, and the higher the stability. The objective function is the reciprocal of the variance, which transforms minimizing the variance into maximizing the objective function, making optimization calculations easier.

[0062] Objective 3: This is used to minimize the design cycle and improve design efficiency. The design cycle includes the time required for subsystem modeling. Simulation test time Iteration optimization time ,Right now . The complexity of the subsystem model and the efficiency of the modelers are determined. Based on the determination of the number of simulation scenarios and the simulation step size The design efficiency is determined based on the number of iterations and the optimization time for each iteration. A shorter design cycle results in higher design efficiency; the objective function is the reciprocal of the design cycle to maximize design efficiency.

[0063] Two co-evolutionary populations are constructed, corresponding to the system architecture and subsystem models respectively, to achieve synchronous optimization and adaptation of the two: Population A is the "system architecture population". It includes 60 different system architecture solutions, each solution It consists of core elements such as powertrain layout (e.g., front-wheel drive, four-wheel drive), electronic and electrical topology (e.g., centralized, distributed), and chassis structural parameters (e.g., suspension type, braking system configuration). The value range of each element is determined based on the parameter index quantification model of S3. Population B is the "subsystem model population". It contains 60 different subsystem model combinations, each combination It consists of a dynamic mode model of the motor subsystem (such as pure electric drive and hybrid drive) and a control strategy model of the electronic and electrical subsystem (such as PID control and model predictive control). The model combination is based on the selection of the S5 simulation model library.

[0064] Define non-dominated sorting operators Used for cross-population individuals Sort the solutions and filter out non-dominated solutions. The sorting process consists of two steps: First, calculate the three objective function values ​​for each cross-population individual. Then, for each individual, the number of individuals dominating it is counted. If individual A's three objective function values ​​are all better than individual B's, then individual A is said to dominate individual B; individuals with 0 dominations are in the first frontier layer (optimal frontier layer), individuals with 1 domination are in the second frontier layer, and so on. The second step is to calculate the crowding distance for each individual to maintain the diversity of the solution set. The crowding distance refers to the distance between an individual and its neighboring individuals in its frontier layer, and the calculation expression is: ,in For the first The maximum value of each objective function. For the first The minimum value of each objective function. The larger the crowding distance, the sparser the distribution of individuals in the region where the individual is located. Retaining such individuals can prevent the solution set from being concentrated in a certain region and ensure the diversity of optimization results.

[0065] During the model training phase, individuals in populations A and B are initialized: the system architecture parameters of population A are generated using Latin hypercube sampling to ensure uniform distribution of parameters within their range, avoiding insufficient optimization due to initial population concentration; the subsystem model combinations of population B are generated using random sampling, randomly selecting different power mode models and control strategy models from the model library for combination. During iteration, populations A and B achieve information exchange and collaborative optimization through cooperative operators: population A receives model adaptation feedback from population B, adjusting system architecture parameters (e.g., reducing the adjustment frequency of power output) based on the performance limitations of the subsystem models (e.g., the upper limit of the response speed of a certain control strategy model); population B optimizes the subsystem model combinations based on the architecture scheme of population A (e.g., selecting a control strategy model adapted to four-wheel drive control if the system architecture adopts a four-wheel drive layout).

[0066] Introducing an adaptive mutation operator This is used to prevent the algorithm from getting trapped in local optima. For variable asynchronous length, This represents the current iteration number. The maximum number of iterations is determined by the variable asynchronous length, which decreases linearly with the number of iterations. The initial value of 0.1 ensures the global exploration capability in the early stage of the algorithm, and is reduced to 0.01 in the later stage to enhance the accuracy of local search. Standard normal distribution random numbers are used to ensure the randomness and rationality of mutation. Population A uses simulated binary crossover with a crossover probability of 0.9; population B uses single-point crossover with a crossover probability of 0.8. The maximum number of iterations is set to 150. When the change in the number of individuals in the non-dominated solution set is less than 5% after 10 consecutive iterations, it indicates that the solution set tends to stabilize, training is stopped, and the non-dominated solution set is obtained. The solution set contains all the optimal solutions that take into account the three optimization objectives.

[0067] In the model application phase, the comprehensive utility value of each individual within the non-dominated solution set is calculated using the vehicle parameter index quantification model constructed with S3. .in The target weights are determined using the analytic hierarchy process (AHP) combined with enterprise design priorities, highlighting the core position of performance objectives while also considering stability and design efficiency. The individual with the highest overall utility value is selected. Determine the optimal vehicle system architecture scheme and professional simulation models of each subsystem A system-level optimal design report is generated, including detailed architectural parameters, subsystem model configuration lists, simulation verification results, performance indicator compliance, design cycle estimates, and cost estimates. This report serves as the core input for the S7 back-end iteration. (Algorithm interaction logic reference) Figure 3 .

[0068] In some embodiments, the multi-scale fuzzy rough set attribute reduction algorithm interacts with an evidence-based multi-source data fusion algorithm. The interaction logic revolves around the guiding role of core requirement attributes in simulation data fusion. The multi-scale fuzzy rough set attribute reduction algorithm outputs the core requirement attribute set. In, the importance of each attribute The weights of evidence sources in multi-source data fusion algorithms based on evidence theory are directly used as the basis for adjusting the weights of corresponding evidence sources. For subsystem simulation data corresponding to core requirement attributes (such as performance evaluation data of the motor subsystem corresponding to core power attributes), the weights of the evidence sources need to be dynamically optimized in conjunction with attribute importance to ensure that the fusion process is tilted towards core requirements. The interaction correlation formula is: ,in In the multi-source data fusion algorithm based on evidence theory, the first... Adjusted weights for each source of evidence As the initial weights, For the first The importance of each source of evidence corresponds to the core attribute. The maximum importance of all core attributes. Interaction termination condition: When the deviation between the adjusted weights and the consistency of the multi-source data fusion algorithm based on evidence theory with the core requirement attributes is less than a preset threshold (e.g., deviation < 5%), and the fluctuation of the fusion results in three consecutive iterations is less than 1%, the weight adjustment stops, and the interaction is complete.

[0069] In some embodiments, the multi-scale fuzzy rough set attribute reduction algorithm interacts with the non-dominated ranking enhanced co-evolutionary algorithm. The core of this interaction is the weight allocation of multi-objective optimization guided by core requirement attributes. The multi-scale fuzzy rough set attribute reduction algorithm focuses on the importance of core requirement attributes. This is used to dynamically adjust the weights of multiple optimization objectives in a non-dominated sorting-enhanced co-evolutionary algorithm, especially when performance-related attributes account for a high proportion of the core requirements (such as...). If the importance of a certain attribute is high, then the weight of the "high-level evaluation credibility" objective in the non-dominated ranking enhanced co-evolutionary algorithm will be increased; if the importance of cost and cycle-related attributes is prominent, then the weight of the "design cycle" objective will be increased. The interaction correlation formula is: ,in In the non-dominated sorting enhanced co-evolutionary algorithm, the first... Adjusted weights for each optimization objective. As the initial target weight, In order to be with the first The sum of the importance of the core attributes related to each objective. Interaction termination condition: When the fluctuation of the comprehensive utility value of the non-dominated solution set corresponding to the adjusted objective weights in the enhanced co-evolutionary algorithm is less than 2%, and the core attribute satisfaction of the optimal solution is ≥90%, the objective weight adjustment stops, and the interaction terminates.

[0070] In some embodiments, the improved dual-population cooperative particle swarm optimization algorithm interacts with an evidence-based multi-source data fusion algorithm. The interaction logic involves the allocation relationship of a three-level index system providing a priority basis for simulation data fusion. The optimal index allocation matrix output by the improved dual-population cooperative particle swarm optimization algorithm is shown in the figure. In the system, the index allocation coefficients from the system level to the subsystem level. This is used to correct the similarity calculation between simulation data and evaluation levels in multi-source data fusion algorithms based on evidence theory. The higher the allocation coefficient of an indicator, the greater its similarity weight to the evaluation level. The interaction correlation formula is: ,in To adjust the similarity, For indicator functions (the first) The source of evidence corresponds to the first (Takes a value of 1 when the indicator mapping relationship is established, otherwise takes a value of 0). Interaction termination condition: The multi-source data fusion algorithm based on evidence theory evaluates the credibility vector of the scheme based on the adjusted similarity. The interaction ends when the evaluation level corresponding to the maximum credibility matches the target level of the three-level index system of the improved dual-population collaborative particle swarm optimization algorithm by ≥95%, and the credibility difference between two consecutive fusion results is less than 3%.

[0071] In some embodiments, the improved dual-population cooperative particle swarm optimization algorithm interacts with the non-dominated sorting enhanced co-evolutionary algorithm. The core of this interaction is the bidirectional calibration of the indicator system constraints and the optimization scheme. The system-level indicator threshold of the improved dual-population cooperative particle swarm optimization algorithm is... Initialization constraints for the system architecture population of the non-dominated sorting enhanced co-evolutionary algorithm (architectural parameters must satisfy) Meanwhile, the optimal system architecture parameters output by the non-dominated sorting enhanced co-evolutionary algorithm are... The feedback is fed back to the improved dual-population cooperative particle swarm optimization algorithm to correct the index assignment matrix and threshold. The interaction correlation formula is: ,in To correct the threshold values ​​of the system-level indicators, For the corresponding number in the optimal architecture The parameter values ​​of each indicator, This represents the average value of the corresponding parameter in the population of the non-dominated sorting enhanced co-evolutionary algorithm. The interaction termination condition is the modified objective function of the improved two-population co-evolutionary particle swarm optimization algorithm. Convergence (difference between adjacent iterations < Furthermore, the optimal architecture parameters of the non-dominated sorting enhanced co-evolutionary algorithm satisfy... At the same time, when the comprehensive utility value reaches the preset optimal threshold (e.g., ≥0.85), the bidirectional calibration terminates and the interaction is completed.

[0072] S7: Reverse Iterative Update and System-Level Synthesis Verification The key technical indicators from the optimal system-level design report output by S6 are transmitted back to the system layer via the architecture-system interface. This interface uses a data synchronization protocol (such as SyncML) to ensure the real-time performance and accuracy of indicator transmission. During transmission, key technical indicators are encrypted (using the AES-256 encryption algorithm) to ensure data security. Key technical indicators include vehicle power performance indicators (such as 0-100km / h acceleration time and NEDC range), electronic and electrical system indicators (such as bus transmission efficiency and controller response time), system stability indicators (such as parameter volatility and collaborative control accuracy), design cycle time, and cost indicators.

[0073] Based on the key technical indicators transmitted, the three core models constructed by S3 are updated and iterated in a targeted manner to ensure that the models are highly adapted to the optimal design scheme. The update of the vehicle requirement digital model mainly involves adjusting the acceptance criteria for requirement items. Based on the actual performance indicators of the optimal scheme, the original acceptance criteria are replaced with more precise and realistic standards. For example, the original acceptance criterion of "NEDC range ≥ 600km" is updated to "NEDC range ≥ 620km" based on the simulation results of the optimal scheme. Simultaneously, the fulfillment status of requirement items is updated to clarify whether each requirement has been achieved through the optimal scheme. The update of the vehicle functional logic architecture model focuses on adjusting the functional flow paths and module interaction relationships. For example, if a new power control strategy is adopted in the optimal scheme, the flow path of the power function needs to be adjusted accordingly, and the interaction logic in the sequence diagram and activity diagram needs to be updated to ensure that the functional implementation path is consistent with the optimal scheme. The update of the vehicle parameter index quantification model mainly involves adjusting the parameter value range and constraint relationship. Based on the architecture parameters of the optimal solution, the actual values ​​and upper and lower limits of the parameters are updated, and the constraint relationship expression between the parameters is corrected. For example, based on the battery capacity and vehicle weight of the optimal solution, the functional relationship coefficient between the driving range and the two is adjusted.

[0074] The overall vehicle performance indicators were iterated synchronously to form an updated overall performance indicator system. This system covers six major categories of indicators: vehicle power performance, range, intelligent control accuracy, system stability, design cycle, and cost. Each indicator has clearly defined numerical values ​​and tolerance requirements, providing clear objectives for system-level comprehensive verification. The updated overall performance indicators were imported into system-level comprehensive simulation analysis software (using a co-simulation platform built with Prescan and MATLAB / Simulink) to conduct system-level comprehensive capability assessment. The simulation scenarios fully covered the specific tasks and scenarios specified in S1, including urban commuting, long-distance freight, off-road driving, and scenarios such as congested road conditions, highway conditions, and extreme weather conditions, ensuring the comprehensiveness and relevance of the assessment.

[0075] The evaluation mainly includes three aspects: First, the consistency between the overall performance indicators and the vehicle top-level capability requirement model constructed by S1. By comparing the updated overall performance indicators with the top-level requirement benchmark, it is analyzed whether each indicator meets the core needs of the market and users. For example, whether the "comprehensive usage cost in urban commuting scenarios" has reached the target value set by the top-level requirements. Second, the reliability of the collaborative work of each subsystem. Through long-term (no less than 100 hours) continuous simulation testing, the collaborative stability of the subsystems under continuous operation is monitored, the number and type of failures are counted, and the fault tolerance and durability of the system are evaluated. Third, the economy and safety of the whole vehicle. The economy evaluation mainly calculates the energy consumption cost of the vehicle (such as the power consumption per 100 kilometers and the charging cost) and maintenance cost. The safety evaluation mainly simulates the vehicle safety performance (such as the body structure strength and the airbag triggering accuracy) under collision scenarios and extreme weather driving scenarios.

[0076] During the evaluation process, simulation data is recorded in real time through the data acquisition module. Changes in indicators before and after the update are compared to analyze the effectiveness of iterative optimization. For example, the 0-100km / h acceleration time before and after iteration is compared to assess the improvement in power performance; the number of system failures before and after iteration is compared to assess the improvement in system stability. If any indicator is found to be unsatisfactory during the evaluation (e.g., excessive range reduction under extreme weather conditions), it is promptly reported to S6 for secondary optimization and adjustment of the optimal solution. After the evaluation is completed, a full-process design verification report (system-subsystem) is generated. The report details the data, simulation results, iteration process, and evaluation conclusions of each design stage, clearly defining the advantages of the optimal design (e.g., performance exceeding the industry average by 10%) and improvement directions (e.g., room for improvement in range under extreme environments). This completes the closed-loop comprehensive evaluation of the vehicle system layer and the system requirements layer, ensuring that the design solution meets the optimal performance requirements for specific tasks and scenarios.

[0077] Algorithm Fusion and Collaboration Explanation In S2, the multi-scale fuzzy rough set attribute reduction algorithm and the improved dual-population cooperative particle swarm optimization algorithm form a close synergistic relationship. The former, through multi-scale hierarchical structure and fuzzy rough set theory, accurately eliminates redundant attributes in the top-level requirement model and extracts the core requirement attribute set. This approach reduces the computational complexity of the latter and avoids interference from invalid attributes in indicator allocation, ensuring the relevance and accuracy of indicator allocation. The latter takes the core requirement attribute set as input and achieves synchronous optimization of indicator allocation coefficients and system-level indicator thresholds through a dual-population co-evolution mechanism. This transforms abstract core requirements into a three-level indicator system of "system-system-subsystem" that can be implemented. The two work together to build a bridge between system requirements and design execution, ensuring that the design process always revolves around core requirements.

[0078] In S6, the evidence-based multi-source data fusion algorithm and the non-dominated ranking enhanced co-evolutionary algorithm form a closed-loop collaborative relationship of "data support - optimization decision-making". The former transforms multi-source heterogeneous simulation data into a unified scheme evaluation credibility vector, eliminating data conflicts and redundancy, and providing the latter with a reliable and consistent fitness evaluation basis, avoiding optimization bias caused by differences in multi-source data; the latter uses the credibility vector as the core input, and through multi-objective optimization and dual-population co-evolution, achieves collaborative optimization of system architecture and subsystem models. Its optimization objective is directly related to the fused credibility vector, ensuring that the optimization direction is consistent with the scheme performance requirements. The interaction between the two forms a complete closed loop of "data fusion - multi-objective optimization - scheme selection", which not only guarantees the performance advantage of the optimal scheme, but also takes into account system stability and design efficiency, significantly improving the scientificity and practicality of the design scheme.

Claims

1. A vehicle full-process digital design method, characterized in that, include: S1: Conduct multi-dimensional vehicle capability requirements collection, including macro environment and industry trends, market competition landscape, target users and consumer insights, internal positioning and goals of enterprises, and future technology trends. Quantitatively analyze the collected multi-source data to construct a top-level vehicle capability requirements model. S2: The multi-scale fuzzy rough set attribute reduction algorithm is used to remove redundant attributes from the vehicle top-level capability requirement model to obtain the core requirement attribute set; then, the core requirement attribute set is transformed into a three-level index system of system-subsystem through an improved dual-population cooperative particle swarm optimization algorithm. The improved dual-population cooperative particle swarm optimization algorithm sets an index allocation coefficient population and a system-level index threshold population, and realizes bidirectional interactive optimization of the two populations through cross-population cooperative operators. S3: Based on the aforementioned three-level indicator system of system-subsystem, combined with design specifications and engineering design experience, use SysML language to construct a digital model of vehicle requirements, a vehicle functional logic architecture model, and a vehicle parameter indicator quantification model to form a system-level design benchmark. S4: Based on the aforementioned system-level design benchmark, a function-structure bidirectional mapping strategy is adopted to decompose the system-level requirements, functions, and parameter indicators into design task books for the mechanical subsystem and the electronic and electrical subsystem; S5: Based on the design task, construct simulation models for the motor subsystem and the electronic and electrical subsystem respectively, conduct multi-scenario simulation tests, and output performance evaluation data for the motor subsystem and simulation results for the electronic and electrical subsystem; establish a cross-subsystem data interaction channel to realize the coupling integration verification of the two subsystem simulation models and output cross-subsystem integrated simulation data; S6: Employing a multi-source data fusion algorithm based on evidence theory, the performance evaluation data of the motor subsystem, the simulation results of the electronic and electrical subsystem, and the cross-subsystem integration simulation data are fused to obtain a scheme evaluation credibility vector. Then, a non-dominated ranking enhanced co-evolutionary algorithm is used to perform multi-objective optimization on the system architecture and subsystem model combination corresponding to the scheme evaluation credibility vector, and the optimal vehicle system architecture scheme and subsystem professional simulation model are selected to form a system-level optimal design scheme report.

2. The vehicle end-to-end digital design method according to claim 1, characterized in that, Also includes: S7: Based on the key technical indicators in the optimal design scheme report at the system level, update the vehicle requirement digital model, vehicle functional logic architecture model, vehicle parameter index quantification model, and overall vehicle performance index, conduct system-level comprehensive verification, form a full-process design verification report of system-subsystem, and complete the closed-loop evaluation.

3. The vehicle end-to-end digital design method according to claim 1, characterized in that, The working process of the multi-scale fuzzy rough set attribute reduction algorithm is as follows: taking the five dimensions of vehicle capability requirements collected in S1 as multi-scale levels, firstly define the top-level requirement domain and requirement attribute set, then construct fuzzy equivalence relations at each scale, and quantify the degree of fit between requirement items and attributes through fuzzy membership functions; based on the multi-source data after quantification analysis in S1, the gradient descent method is used to optimize the weights of each scale, and redundant attributes are eliminated by calculating the importance of attributes, finally obtaining the core requirement attribute set that retains only the core requirement information.

4. The vehicle end-to-end digital design method according to claim 3, characterized in that, The working logic of the improved dual-population cooperative particle swarm optimization algorithm is as follows: Based on the core requirement attribute set, construct an objective function that includes the matching degree and balance of index allocation; optimize the allocation ratio of system layer indicators to system layer through the index allocation coefficient population; and set the upper limit of each system layer indicator through the system layer indicator threshold population. The correlation between two populations is calculated using a cross-population collaborative operator. The population that assigns the index allocation coefficient adjusts the update direction of the allocation ratio based on the correlation. The population that assigns the index threshold at the system level optimizes the threshold range based on the allocation ratio. The bidirectional interactive iteration continues until the objective function converges, thus achieving the accurate construction of a three-level index system.

5. The vehicle end-to-end digital design method according to claim 4, characterized in that, The interaction logic between the multi-scale fuzzy rough set attribute reduction algorithm and the improved dual-population cooperative particle swarm optimization algorithm is as follows: the core requirement attribute set output by the multi-scale fuzzy rough set attribute reduction algorithm is directly used as the core input of the improved dual-population cooperative particle swarm optimization algorithm. The function and performance-related attributes in the core requirement attribute set are included in the index allocation fit calculation to support the construction of the objective function of the improved dual-population cooperative particle swarm optimization algorithm. The improved dual-population cooperative particle swarm optimization algorithm verifies the integrity of core requirement attributes by back-checking the optimization results of the objective function. If there is an imbalance in the allocation of indicators, it is fed back to the multi-scale fuzzy rough set attribute reduction algorithm to re-verify the attribute importance threshold and ensure the rationality of the core requirement attribute set.

6. The vehicle end-to-end digital design method according to claim 1, characterized in that, The working process of the multi-source data fusion algorithm based on evidence theory is as follows: performance evaluation data of the motor subsystem, simulation results of the electronic and electrical subsystem, and cross-subsystem integration simulation data are taken as independent evidence sources, and an identification framework consisting of scheme evaluation levels is defined; a basic probability allocation function is constructed to quantify the similarity between each evidence source and the evaluation level, and the evidence conflict coefficient is calculated to determine the degree of disagreement among each evidence source; the weight of each evidence source is determined by the optimization algorithm, and the corresponding synthesis rule is selected according to the size of the conflict coefficient, so as to fuse the multi-source heterogeneous simulation data into a unified scheme evaluation credibility vector.

7. The vehicle end-to-end digital design method according to claim 6, characterized in that, The working logic of the non-dominated ranking enhanced co-evolutionary algorithm is as follows: A system architecture population and a subsystem model population are set up, with multiple optimization objectives including the high-level evaluation credibility corresponding to the scheme evaluation credibility vector, the stability of system architecture parameters, and the design cycle. A frontier layer is partitioned across individuals in the cross-population using a non-dominated ranking operator, and crowding distance is used to maintain solution set diversity. The two populations interact through a cooperative operator. The system architecture population adjusts the architecture parameters based on the adaptability of the subsystem models, while the subsystem model population optimizes model combinations based on the system architecture scheme. Iterative selection yields a non-dominated solution set that balances multiple objectives, and the optimal scheme is determined by combining the comprehensive utility value.

8. The vehicle end-to-end digital design method according to claim 7, characterized in that, The interaction logic between the evidence-based multi-source data fusion algorithm and the non-dominated ranking enhanced co-evolutionary algorithm is as follows: the scheme evaluation credibility vector output by the evidence-based multi-source data fusion algorithm is directly used as the fitness evaluation basis of the non-dominated ranking enhanced co-evolutionary algorithm, and the high-level evaluation credibility in the credibility vector is transformed into one of the optimization objectives; if multiple non-dominated solutions are difficult to screen during the optimization process, the non-dominated ranking enhanced co-evolutionary algorithm feeds back to the evidence-based multi-source data fusion algorithm to readjust the evidence source weights or synthesis rules, improve the discriminativeness of the credibility vector, and support the accurate screening of the optimal solution.

9. The vehicle end-to-end digital design method according to claim 1, characterized in that, The multi-scale fuzzy rough set attribute reduction algorithm extracts core requirements, providing input for the improved dual-population cooperative particle swarm optimization algorithm and supporting the construction of a three-level index system. The output of the improved dual-population cooperative particle swarm optimization algorithm serves as the constraint benchmark for the entire design process, guiding subsystem simulation modeling and generating multi-source simulation data. The multi-source data fusion algorithm based on evidence theory processes the simulation data, providing evaluation basis for the non-dominated ranking enhanced co-evolutionary algorithm. The optimal solution output by the non-dominated ranking enhanced co-evolutionary algorithm iteratively updates the design benchmark, forming an algorithmic support closed loop of requirement extraction, index allocation, simulation verification, solution optimization, and iterative update.

10. The vehicle end-to-end digital design method according to claim 1, characterized in that, The mechanism by which the improved dual-population cooperative particle swarm optimization algorithm plays a role in cross-stage indicator transfer is as follows: the optimal indicator allocation matrix clarifies the indicator mapping relationship between the system layer, subsystem layer, and system, accurately transferring core requirement attributes to each design stage; the system layer indicator threshold sets the design boundary of each stage to avoid indicator overload; during the indicator transfer process, the algorithm ensures the matching between indicator allocation and threshold setting through cross-population cooperative operators, making cross-stage data interaction standardized and accurate, and eliminating information disconnect in the design stage.

Images