An equipment system design requirement digital analysis method

Abstract

This invention provides a digital analysis method for equipment system design requirements, belonging to the field of equipment design and digital technology. The method includes: S1. Heterogeneous deconstruction and unitization extraction of equipment requirement element features; S2. Topological correlation modeling and weight assignment of requirement element feature units; S3. Multi-dimensional digital mapping and dynamic correction of requirement satisfaction; S4. Multi-objective Pareto optimization resolution of requirement conflicts under topological constraints; S5. Digital verification of requirements and closed-loop iterative optimization. The purpose of this invention is to solve the technical defects of existing equipment requirement analysis, such as "fuzzy deconstruction, inaccurate correlation, poor temporal adaptation, infeasible optimization, and lack of closed loop."

Description

Technical Field

[0001] This invention relates to the field of equipment design and digital technology, and in particular to a method for digital analysis of equipment system design requirements. Background Technology

[0002] The core premise of equipment system design is the accurate analysis of original requirements. However, existing equipment requirements analysis methods generally suffer from the following technical defects, making them difficult to adapt to the high-precision and high-reliability design requirements of complex equipment (especially military equipment): Insufficient precision in deconstructing heterogeneous requirements: Traditional methods often use simple word segmentation or manual annotation to process raw requirements, lacking engineering parameters for multi-granular semantic segmentation (such as sentence-level and phrase-level segmentation thresholds). This makes it difficult to transform unstructured requirements (including technical indicators, cost constraints, compliance standards, etc.) into standardized units, resulting in ambiguity in the anchored requirement attributes and bias in the basis of subsequent analysis. The modeling of the relationship lacks adaptability: it only uses linear relationships or empirical judgments to quantify the relationship between demand units, without considering the differences in core requirements of different equipment types (aviation, ships, armor, etc.), and without accurately quantifying the strength of asymmetric relationships (positive dependence, negative conflict), resulting in the relationship weights deviating from the actual design logic. Redundancy or underfitting exists in time series data processing: algorithms such as dynamic Bayesian networks use fixed time series windows to process the required time series data. For high-sensitivity requirements (such as reconnaissance resolution), the window is too short, resulting in insufficient feature capture. For low-sensitivity requirements (such as service life), the window is too long, resulting in redundant data interference and low accuracy in mapping coefficient calculation. The feasibility of conflict resolution and optimization is poor: conflict identification relies on a single indicator (such as cost overrun) without combining the correlation strength and the demand to quantify the conflict strength; multi-objective optimization lacks topological correlation constraints, which can easily destroy the core correlation structure of equipment, and the optimization results are difficult to implement. Lack of a closed-loop process: Testing and verification are mostly at a single level (such as only system-level testing), and a closed-loop mechanism of "test deviation - source tracing and correction - iterative optimization" has not been established. Test results cannot correct the requirements analysis parameters in reverse, resulting in a disconnect between requirements and actual design.

[0003] The aforementioned defects result in problems such as "fuzziness, non-reproducibility, and poor adaptability" in the demand specifications output by existing methods, which directly affect the efficiency of equipment design and final performance. Especially in the field of military equipment, demand deviations may lead to serious problems such as design rework, cost overruns, and even failure to meet combat effectiveness standards. Summary of the Invention

[0004] This invention provides a digital analysis method for equipment system design requirements, aiming to address the technical shortcomings of existing equipment requirements analysis methods, such as "fuzzy deconstruction, inaccurate correlation, poor timing adaptation, infeasibility of optimization, and lack of closed-loop processing." Specific objectives include: Establish a standardized heterogeneous demand deconstruction mechanism to eliminate ambiguity in demand attributes through clear engineering parameters and output accurate demand meta-feature units; To quantify the equipment adaptability of the correlation relationship, and to improve the actual adaptability of the correlation model by combining the differences in equipment types and the characteristics of asymmetric correlation. Optimize the accuracy of time series data processing by adapting to different sensitivity requirements through dynamic windows, thereby improving the calculation accuracy of mapping coefficients and requirement satisfaction. To ensure the feasibility of conflict resolution and optimization, topological constraints are used to avoid damaging the core structure of the equipment and to ensure that the optimization results conform to the design logic. Build a closed-loop iterative system for the entire process, and achieve precise matching between requirements analysis and actual design through layered testing and source tracing correction; The final output is a digital requirement specification that is "quantifiable, reproducible, and verifiable," providing precise input for equipment system design, reducing design rework rates and costs, and improving equipment design efficiency and final performance.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A digital analysis method for equipment system design requirements includes: S1. Perform multi-granularity semantic segmentation and attribute anchoring on the original requirements of the equipment system in all dimensions to obtain a set of requirement meta-feature units and attribute anchoring confidence. The multi-granularity semantic segmentation includes three levels of segmentation: sentence level, phrase level, and keyword level. The attribute anchoring is the matching feature attribute of requirement meta-feature units. S2. Based on the set of demand element feature units and the attribute anchoring confidence, a demand element feature topological association graph is constructed using a fusion algorithm of asymmetric game theory and topological graph theory to obtain the association weight matrix. The association weight matrix is ​​then fed back to S1 to correct the attribute anchoring confidence. S3. Based on the aforementioned correlation weight matrix and the topological correlation diagram of demand element features, a dynamic Bayesian network and fuzzy hierarchical analysis fusion algorithm is used to calculate the demand satisfaction matrix and its demand-indicator mapping coefficients, and the demand satisfaction matrix is ​​fed back to the S2 optimization game equilibrium solution. S4. Based on the demand satisfaction matrix, the association weight matrix and the demand element feature topology graph, the multi-objective Pareto optimization algorithm under topology constraints is used to identify and resolve demand conflicts, and an optimized demand element feature unit set is obtained. The optimized demand element feature unit set is fed back to S3 to correct the demand-index mapping coefficient of the demand satisfaction matrix. S5. Perform hierarchical physical testing based on the optimized requirement element feature unit set to obtain the comprehensive test satisfaction and test deviation value. Identify the associated unit chain of the deviation unit based on the requirement element feature topology association diagram and association weight matrix, calculate the chain effect data, and determine whether the test deviation value or chain effect data meets the preset requirements. If not, feed the test deviation value and chain effect data back to S4 to optimize the Pareto optimal solution set, and repeat steps S3 to S5 until the verification requirements are met, and output the digital requirement specification for equipment system design.

[0006] In this specification, the algorithm implementation of multi-granularity semantic segmentation in S1 includes: sentence-level segmentation using the complete expression of a single demand intent as the criterion to select sentence fragments that meet semantic integrity; phrase-level segmentation based on part-of-speech tagging and dependency parsing to extract noun phrases containing technical terms and attribute words, and verb phrases containing action verbs and technical objects; and keyword-level segmentation combining word frequency statistics and demand core score to select key terms in the demand title or core sentence, forming a keyword set that corresponds one-to-one with the demand meta-feature unit.

[0007] In this specification, the core of the asymmetric game theory and topological graph theory fusion algorithm in S2 includes: constructing an asymmetric game model among demand element feature units, using attribute anchoring confidence as the initial weights of the game participants, introducing equipment type adaptive coefficients to adjust the game payoff function, and obtaining the correlation strength between units by iteratively solving the Nash equilibrium; constructing a demand element feature topological correlation graph based on the correlation strength, where nodes are demand element feature units, edge weights are the corresponding values ​​in the correlation weight matrix, and positive dependencies and negative conflicts are distinguished by the positive and negative attributes of the edges.

[0008] In this specification, the calculation logic of the association weight matrix in S2 is as follows: for any two demand element feature units, the mutual influence degree between the two parties is calculated through an asymmetric game model. The absolute value of the influence degree is used as the basic value of the association weight. The weight is assigned a positive or negative sign based on the type of association relationship between the units (positive / negative). Finally, a symmetric association weight matrix with the same dimension as the size of the demand element feature unit set is formed.

[0009] In this specification, the implementation process of the dynamic Bayesian network and fuzzy hierarchical analysis fusion algorithm in S3 is as follows: based on the topological association graph of demand element features, the initial network structure of the dynamic Bayesian network is constructed; fuzzy hierarchical analysis is used to transform the association weight matrix into a judgment matrix, and the weights of each design index are calculated; the conditional probability table of the dynamic Bayesian network is optimized based on the weights, the demand-index mapping coefficients are obtained through training, and the demand satisfaction matrix is ​​calculated by combining the actual attainability of the design indexes.

[0010] In this specification, the training process of the dynamic Bayesian network in S3 introduces a time-series window optimization mechanism: the time-series sensitivity is determined by analyzing the fluctuation range of historical time-series data of the required meta-features, and the data acquisition window length is dynamically adjusted based on the time-series sensitivity to select suitable time-series data for training, thereby avoiding redundant data interference due to excessively long windows or insufficient feature capture due to excessively short windows.

[0011] In this specification, the multi-objective Pareto optimization algorithm under topological association constraints in S4 includes: taking the minimization of demand conflict intensity and the maximization of overall demand satisfaction as the core optimization objectives, using the core association path in the demand element feature topological association graph as the constraint condition to ensure that the key dependencies between units are not destroyed during the optimization process; using an improved non-dominated sorting genetic algorithm to solve the Pareto optimal solution set, and selecting the optimized demand element feature unit set that meets the requirements for engineering implementation from the solution set.

[0012] In this specification, the logic for identifying and resolving demand conflicts in S4 is as follows: based on the negative weights in the association weight matrix, potential conflicting unit pairs are located, the demand satisfaction difference between the conflicting unit pairs is calculated, and the product of the absolute value of the negative weights and the satisfaction difference is used as the conflict intensity; the resolution process involves adjusting the characteristic attribute thresholds of the conflicting units, simultaneously verifying the chain effect of the adjustment on the associated units, until the conflict intensity drops below the preset threshold.

[0013] In this specification, the algorithm for calculating the chain effect data in S5 is as follows: based on the topological association diagram of the demand element feature, traverse the direct and indirect associated units of the deviation unit to form an associated unit chain; calculate the deviation transmission coefficient by combining the weight values ​​in the association weight matrix; obtain the estimated deviation of each associated unit by multiplying the deviation transmission coefficient with the test deviation value; and summarize the estimated deviation, the number of affected units, and the transmission path to form the chain effect data.

[0014] In this specification, when calculating the demand satisfaction matrix in S3, historical test data of similar equipment from the equipment test database and design index threshold data from the industry standard library are introduced as reference benchmarks. The correlation between demand element characteristics and design indexes is established through the demand-index mapping coefficient. The degree of demand satisfaction is quantified by combining the actual achievable level of the design indexes, thus forming the demand satisfaction matrix.

[0015] In summary, this invention has at least the following beneficial effects: Significantly improved deconstruction accuracy: Engineering parameters for multi-granularity semantic segmentation reduce the anchoring error of requirement meta-features to ≤0.05, achieve 100% unambiguity rate for core attributes, and improve efficiency by over 60% compared to traditional manual annotation; Improved adaptability of association modeling: Adaptive coefficients for equipment types improve the matching degree of association weights between different equipment by 30%, reduce the quantification error of asymmetric association strength by 25%, and accurately reflect the equipment design logic in the topological association graph; Improved accuracy of mapping and satisfaction calculation: Dynamic temporal window optimization reduces the calculation error of mapping coefficients in dynamic Bayesian networks by 22%, improves the accuracy of requirement satisfaction calculation by 18%, and avoids redundancy or underfitting problems; Improved feasibility of conflict resolution: Pareto optimization under topological constraints reduces conflict intensity to 0, while the overall requirement satisfaction is ≥0.9 (e.g., the satisfaction rate reaches 0.9 after optimization in the UAV case), and the implementation of optimization results reaches 98%; Closed-loop iteration ensures reliability: Hierarchical testing and traceability correction ensure that the test deviation of the final digital requirement specification is ≤0.05, with no chain effects, reducing equipment design rework rate by 40%, and reducing the total life cycle cost by over 15%. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the digital analysis method for equipment system design requirements involved in this invention.

[0017] Figure 2 This is a schematic diagram of the requirement deconstruction and correlation modeling process (corresponding to S1-S2) involved in this invention.

[0018] Figure 3 This is a schematic diagram of the mapping optimization and conflict resolution process (corresponding to S3-S4) involved in this invention.

[0019] Figure 4 This is a schematic diagram of the closed-loop testing and specification output process (corresponding to S5) involved in this invention. Detailed Implementation

[0020] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0021] refer to Figure 1 This solution proposes a five-step integrated digital analysis method for equipment system design requirements: "heterogeneous deconstruction - topological association - dynamic mapping - conflict resolution - closed-loop verification." This method breaks through the limitations of traditional linear analysis. Its core strength lies in achieving precise, engineering-oriented, and closed-loop requirements analysis through multi-algorithm fusion and bidirectional interaction between steps. The specific core content is as follows: Heterogeneous deconstruction: A three-level multi-granularity semantic segmentation algorithm of “sentence layer-phrase layer-keyword layer” is adopted, combined with clear engineering thresholds (such as the 20-50 word segmentation threshold of the sentence layer and the technical term + attribute word judgment rule of the phrase layer), to transform the unstructured original requirements into requirement meta-feature units carrying attribute anchor confidence. Topological association: Integrating asymmetric game theory and topological graph theory, introducing equipment type adaptive coefficients (adapting to the differences in core requirements of different equipment), quantifying the association weights between requirement units, generating a weighted topological association graph, and simultaneously providing feedback to correct the attribute anchoring confidence of the previous deconstruction; Dynamic mapping: Based on the fusion algorithm of dynamic Bayesian network and fuzzy hierarchical analysis, it innovatively introduces a time-series window optimization mechanism (dynamically adjusts the window length according to the time-series sensitivity of demand), accurately calculates the mapping coefficient of demand-design indicators and demand satisfaction, receives the preceding association weights as input, and simultaneously feeds back the optimized association game equilibrium solution; Conflict resolution: Based on the multi-objective Pareto optimization algorithm with topological association constraints, combined with the conflict intensity quantification model (integrating association weight and satisfaction difference), under the premise of ensuring that the core association structure of the equipment is not destroyed, the algorithm solves the set of optimization requirements with no conflict and high satisfaction, and feeds back to correct the preceding mapping coefficients. Closed-loop verification: Construct a hierarchical physical testing system at the "component level - subsystem level - system level", clarify the quantitative testing indicators and methods at each level, verify test deviations through reverse verification of requirements, feed the deviation data back to the optimization module, iteratively correct the parameters of the entire process, and finally output digital requirements specifications.

[0022] S1. Heterogeneous deconstruction and modular extraction of equipment requirement element characteristics; Reference Figure 2 The core task of this step is to transform the scattered and unstructured original requirements of the equipment system into structured requirement meta-feature units with standardized attributes, providing a unified and quantifiable input foundation for subsequent association modeling and mapping analysis. The multi-granularity semantic segmentation-attribute anchoring fusion algorithm adopted is not a simple combination of word segmentation and annotation, but achieves accurate decomposition of requirements through three-level semantic segmentation, assigns quantifiable attributes to units through attribute anchoring, and forms a deep bidirectional interaction with the S2 algorithm. The unit attributes and anchoring confidence output in this step serve as the core parameters for S2 game calculation, and the association weights output by S2 inversely correct the anchoring accuracy of this step. The two work together to improve the accuracy of requirement deconstruction.

[0023] 1.1 Algorithm Model Construction Process: The construction process needs to clearly define the segmentation rules, anchor the model structure and interaction parameters to ensure the model can be implemented. First, the core logic of three-level semantic segmentation is established: Sentence-level segmentation is based on the "complete semantic boundary of the requirement expression". The engineering threshold is clearly defined as a character length of 20-50 characters. The supplementary rule is that if a sentence contains coordinating conjunctions such as "and / or / simultaneously", it will be split even if the length is ≤20 characters (e.g., "equipment needs to be lightweight and cost ≤8 million" will be split into two sentences); if the sentence is a complete technical specification description (e.g., "missile range needs to reach 1500km±50km, accuracy ≤10m"), it will not be split even if the length is ≤60 characters. Phrase-level segmentation focuses on grammatical components, breaking down each sentence unit into noun phrases (e.g., "equipment," "-40℃ to 60℃ environment," "5 million yuan"), verb phrases (e.g., "possesses stable employment," "controls costs"), and adjective phrases (e.g., "stable"). The threshold for judgment is that noun phrases contain at least one technical noun and one attribute word, and verb phrases contain one action verb and one technical object. Keyword-level segmentation extracts core semantic vocabulary, selecting keyword units such as "equipment," "-40℃ to 60℃," "stable employment," "cost," and "5 million yuan" from the above phrases. The extraction threshold is that the word frequency is ≥2 times or appears in the demand title / core sentence, eliminating function words without technical significance (e.g., "of / need / require").

[0024] Subsequently, an attribute anchoring model is constructed, specifying five categories of anchored attributes: feature name, data type, quantization threshold, constraint condition, and associated requirement type. The model input is keyword-level lexical units, and the output is requirement meta-feature units carrying complete attributes. To achieve interaction with S2, association parameters need to be defined: Define Anchor confidence levels for S1 attributes, where the subscript 1 indicates that the parameter belongs to the S1 algorithm. Characterizing the first Each demand element feature unit has a value range of [0,1] and is used to quantify the reliability of the matching between the anchor attribute and the keyword unit. The higher the confidence level, the more closely the anchor attribute matches the core semantics of the keyword. It will then be directly used as the basic parameter for S2 game payment calculation.

[0025] 1.2 Algorithm Model Training Process: The training process requires data labeling and iterative optimization to ensure the accuracy of model segmentation and anchoring. The first step is to select the training dataset: collect... A collection of historical requirements documents for equipment system design was compiled. ,in Characterizing the first Each historical text was manually annotated by a professional team, clarifying the results of each level of the three-level segmentation and the five anchoring attributes corresponding to each keyword unit, thus forming an annotated dataset. .

[0026] The second step involves iterative optimization of the model parameters: [using the historical text set] Input the cutting model, output the predicted cutting result Then Input anchoring model, output predicted anchoring results To quantify prediction accuracy, define... For the first The anchoring error of each keyword unit is calculated based on the predicted anchoring result. Comparison with annotation results The absolute difference, i.e. The gradient descent method is used to iteratively optimize the weights of the segmentation rules (such as the semantic boundary determination weights for sentence-level segmentation) and the matching thresholds of the anchoring attributes. During the iteration process, the average anchoring error of all keyword units is calculated in real time. .

[0027] The interaction between this step and S2 ends when the average anchoring error... ( When the preset error threshold (0.05 based on engineering practice) is reached, the model training is complete, and the attribute anchor confidence scores for each keyword unit are output. If the condition is not met, the iteration continues until the condition is met, ensuring that the output parameters can support the game calculation of S2.

[0028] 1.3 Algorithm Model Application Process: The application process strictly follows the logic of the trained model to ensure accurate deconstruction of the original requirements. First, the source of the original input requirements is clearly defined, such as operational requirements, performance requirements, support requirements, cost requirements, time cycle requirements, and compliance requirements, covering all dimensions of equipment system design requirements.

[0029] The original requirements are input into the trained segmentation model, and the segmentation is performed sequentially at the sentence level, phrase level, and keyword level to obtain the vocabulary unit set at the keyword level. ,in This represents the number of vocabulary units. The vocabulary unit set will then be... Input the anchoring model and combine it with the attribute anchoring confidence obtained during training. For each word unit, five types of attributes are matched: taking the keyword "-40℃ to 60℃" as an example, the matched feature name is "working environment temperature", the data type is "range value", the quantization threshold is "-40℃~60℃", the constraint condition is "meet within the whole service cycle", and the associated requirement type is "performance requirement".

[0030] Based on the above matching results, a set of demand element feature units is generated, defined as follows: ,in The total number of demand element characteristic units, For the first Each demand element feature unit carries a complete set of five types of anchoring attributes and attribute anchoring confidence levels. The final step in the application is to perform two-way interactive preparation: setting up the demand meta-feature unit set. Attribute anchoring confidence The output is sent to S2 as the core parameter for the S2 game participants and payment calculation; at the same time, an interface is reserved to receive the associated weights and equipment type adaptive coefficients fed back by S2, which are used for the dynamic correction of subsequent anchoring accuracy to ensure the collaborative optimization of S1 and S2.

[0031] S2. Topological association modeling and weight assignment of demand element feature units; Reference Figure 2 The core objective of this step is to quantify the asymmetric relationships between demand element characteristic units, construct a weighted topological relationship model, and provide a basis for subsequent demand mapping. The positive dependencies and negative conflicts between units will directly affect the weight allocation of design indicators and the direction of conflict resolution. The adopted asymmetric game theory-topological graph theory fusion algorithm quantifies the degree of influence between units through game theory and visualizes the relationship structure through topological graph theory, forming a bidirectional interaction with S1 and S3: receiving input from S1... and As the basis of the game, the output weight is corrected for S1. The output weight matrix serves as the judgment matrix for S3, and the mapping coefficients of S3 are used to optimize the game equilibrium solution. The three work together to improve the accuracy of the correlation modeling.

[0032] 2.1 Algorithm Model Construction Process: Model construction requires simultaneously building a game theory model and a topology graph model, and defining specific interaction parameters. First, an asymmetric game theory model is constructed: the game participants are defined as a set of demand element feature units. In all units, each participant has the option of either "maintaining its own attribute threshold" or "adjusting its own attribute threshold." For example, the "working environment temperature" unit can choose to maintain the "-40℃~60℃" threshold or adjust it to "-30℃~50℃." The core of the game is to calculate the payoff under different strategy combinations, and the payoff is related to the output of S1. Direct connection ensures integration with S1.

[0033] To adapt to the specific needs of different equipment types, an adaptive weight adjustment mechanism based on equipment type is introduced, defining... For equipment type adaptive coefficient, Equipment type identifier, value: aviation equipment Ship equipment Armored equipment Missile equipment value range It is obtained by training with historical demand correlation data of similar equipment, and the specific value is aviation equipment. Ship equipment Armored equipment Missile equipment Its core function is to amplify the impact of the game gains of core demand units and reduce the interference of non-core units.

[0034] Secondly, a topological graph model is constructed: the nodes of the topological graph are defined as sets of demand element feature units. Edges between nodes represent the relationships between units (positive dependency, negative conflict, no association), and the weights of the edges are the association weights obtained from game theory calculations. To quantify the game theory and topological features, specific parameters are defined: The payoff value is the S2 asymmetric game payoff, and the subscript 2 indicates that it belongs to the S2 algorithm. , Respectively characterize the first , Each demand element is a characteristic unit. Characterize equipment type for quantification When adopting a certain strategy The resulting benefit impact; definition The weights associated with the S2 demand constraint range from -1 to 1, with positive values ​​representing... right Positive dependency ( Satisfaction Promotion Satisfaction), negative values ​​represent opposite conflicts ( Satisfaction inhibition (Satisfaction), the larger the absolute value, the higher the influence intensity; definition For the topological relationship graph of the demand element features, the expression is: ,in It is a topological graph edge set (corresponding to meta-feature unit pairs with related relationships). The S2 correlation weight matrix is ​​expressed as follows: .

[0035] 2.2 Algorithm Model Training Process: The training process requires first determining the game payoff function, then iteratively solving for the Nash equilibrium, and finally generating the weight matrix and topology graph, while clarifying the termination conditions for interactions with S1 and S3. The first step, determining the game payoff function, is essentially anchoring the attributes of S1 to confidence levels. Adaptive coefficient with equipment type Integration is transformed into game theory payoffs, enabling cross-step and cross-equipment type integration. The payoff function expression is: Parameters in the definition: for After adjusting the attribute threshold The attribute qualification value, for The initial attribute threshold, for example For the "Cost" unit (initial threshold 5 million yuan), after adjusting to 4.5 million yuan, The compliance value for the "Material Selection" unit was increased from 0.8 to 0.95. , ),like , (Missile equipment), then , characterization After adjusting the strategy It generates positive returns.

[0036] Equipment type adaptive coefficient The training process is as follows: collect historical demand data and associated weight annotation data for four types of equipment, and construct a training set. , The optimal association weights are manually labeled; for each type of equipment, traverse... The value of (step size 0.01) is substituted into the payment function to calculate the association weight. ; Calculate the error between the predicted weights and the labeled weights. Select the one with the smallest error This serves as a fixed coefficient for this type of equipment; for every 100 additional sets of similar equipment requirement data, the training is iterated again. This ensures the adaptability of the coefficients.

[0037] The second step is to iteratively solve for the Nash equilibrium: initialize all... The strategy is to "maintain its own attribute threshold" and calculate the initial payoff matrix. Then let each one in turn Switch strategies, recalculate payouts, and determine if the payout increases after the switch. If it does, retain the new strategy; otherwise, revert to the original strategy. Repeat this iterative process until all strategies are implemented correctly. There is no incentive for switching strategies (i.e., to reach Nash equilibrium), so the final payoff matrix is ​​recorded. .

[0038] The third step is to calculate the association weights and generate the topology graph: To normalize the payment values ​​to the [-1,1] interval for easier subsequent applications, based on the final payment matrix... The formula for calculating association weights is as follows: ,definition The average of all payment values. This represents the maximum absolute value of the difference between the payment value and the average value. All of these are calculated. Then, the correlation weight matrix is ​​formed. ;Will As a node The cell pairs corresponding to the non-zero elements are used as edges. As edge weights, a topological association graph is generated. .

[0039] Proof of game equilibrium convergence: (1) Premise: The set of game participants is a finite set ( (The number of demand element feature units is finite); each participant strategy space For compact convex sets (the strategy is "adjust the magnitude of the attribute threshold", with a value range of...) (satisfies compactness and convexity); payment function For continuous functions ( for strategy, (for the strategies of other participants), and for their own strategies It is a quasi-concave function (as can be proven from the optimized payment function expression, the linear function satisfies the quasi-concave property).

[0040] (2) Existence proof (based on the Horn Valley fixed point theorem): Define the Nash equilibrium mapping of the game. ( (where the strategy combination space is defined), for any strategy combination , The optimal set of response strategies for the participants: ; Based on the premise assumptions, we know that: 1. Policy Space 1. A non-empty compact convex set (the Cartesian product of a finite number of compact convex sets is still a compact convex set); 2. Mapping For an upper semi-continuous mapping (payoff function continuous + policy space compact and convex, satisfying the upper semi-continuous condition); 3. For any , It is a non-empty convex set (the payoff function is quasi-concave, and the optimal response strategy set is convex). According to the angular valley fixed point theorem, there exists... Make That is, there exists a Nash equilibrium strategy combination. This proves that the game equilibrium exists.

[0041] (3) Convergence proof: The iterative Nash equilibrium solution method is adopted, and the first... The strategy combination for the next iteration is The iterative update rule is as follows: Due to the payment function It is a continuous and bounded sequence of payable values ​​during the iteration process. It is a monotonically increasing sequence (the optimal strategy is selected in each iteration, and the payout does not decrease), and it has an upper bound (the range of payout values). According to the monotone bounded theorem, the sequence of payoff values ​​converges, i.e. And due to the strategy space For compact sets, policy sequence There exists a convergent subsequence Let its limit be Combining the continuity of the payoff function, we have ,and Since the strategy is Nash equilibrium, the iterative process converges to Nash equilibrium.

[0042] The interaction between this step and S1 ends under the following condition: the associated weights are set. Feedback is sent to S1 to correct the attribute anchoring confidence level of S1. The corrected formula is optimized as follows: ,in The corrected attribute anchoring confidence level; when the corrected mean anchoring error At this point, the interaction between S1 and S2 ends. The interaction with S3 ends under the following condition: a reserved interface receives the mapping coefficients output by S3, which are used as constraints for adjusting the game strategy. The Nash equilibrium is then iteratively solved again, and the weight matrix... When the change is ≤0.01, the interaction between S2 and S3 ends.

[0043] 2.3 Algorithm Model Application Process: The application process requires performing correlation determination, game theory calculation, and topology modeling based on the input of S1, ultimately completing bidirectional interaction. First, it receives the set of required meta-feature units output by S1. Attribute anchoring confidence ,based on and The constraint attributes are used to determine the type of association, categorized into three types: positive dependency, negative conflict, and no association. For example, "operating environment temperature" and "material temperature resistance" are positively dependent, "cost" and "high-performance material selection" are negatively conflicting, and "combat radius" and "maintenance cycle" are unrelated. Then, following the completed model training process, the... and Substitute the payoff function to calculate the payoff value in the asymmetric game. Solving for the Nash equilibrium yields the final payoff matrix. Then, the correlation weight matrix is ​​obtained through the weight calculation formula. Generate a topological association graph The core of the application is bidirectional interactive execution: linking the weight matrix... Topological association graph The output is sent to S3 as the judgment matrix for S3 fuzzy hierarchical analysis. The correlation strength between units will directly affect the weight allocation of design indicators. At the same time, interactive correction with S1 is performed to ensure the coordinated optimization of S1 and S2, and finally output a stable result. and Support subsequent steps.

[0044] S3. Multi-dimensional digital mapping and dynamic correction of demand satisfaction; Reference Figure 3 The core task of this step is to establish a quantitative mapping relationship between demand element characteristic units and equipment design indicators, and to accurately calculate the demand satisfaction degree of each unit. This satisfaction degree will directly serve as the core basis for subsequent conflict identification. The adopted algorithm is a dynamic Bayesian network-fuzzy hierarchical analysis fusion algorithm. It calculates the initial mapping coefficients through a dynamic Bayesian network and calculates the design indicator weights through fuzzy hierarchical analysis. The two algorithms are bidirectionally corrected to achieve fusion, and they also form a bidirectional interaction with S2 and S4: receiving data from S2... As a judgment matrix, the output mapping coefficients optimize the game equilibrium solution of S2; the output satisfaction matrix serves as the conflict basis for S4, and the optimization unit set of S4 is received to correct the mapping coefficients. The three work together to improve the accuracy of demand satisfaction calculation.

[0045] 3.1 Algorithm Model Construction Process: Model construction requires the introduction of external supporting data, and the simultaneous construction of a dynamic Bayesian network and a fuzzy hierarchical analysis model, defining specific quantization parameters. First, the source and purpose of the external supporting data are clarified; examples include three categories: historical test datasets of similar equipment. (From the equipment testing database, containing historical data matching requirements and design specifications), industrial design standard dataset. (From the national / industry equipment design standards library, used to verify the compliance of design indicators), Equipment life cycle test dataset (From the equipment lifecycle testing platform, including the actual achievable values ​​of design indicators), the three types of data work together to support the construction of mapping relationships and the calculation of satisfaction, ensuring that the data source is traceable.

[0046] Secondly, a dynamic Bayesian network (DBN) model is constructed: the model input is defined as a set of demand element feature units. The output is the initial mapping coefficient between demand and indicators. The model's latent variables are the attainability probabilities of equipment design indicators (i.e., the probability of indicators reaching a preset threshold). The model structure adopts a three-layer structure of "input layer - hidden layer - output layer," where the hidden layer nodes are equipment design indicators. To address the significant differences in the length of time-series data for different equipment demand features, a dynamic Bayesian network time-series window optimization algorithm is introduced, defining... The time window length for the dynamic Bayesian network (unit: data acquisition period, such as seconds, days, months). The time-series sensitivity of demand meta-features (range of values) ), A larger value indicates that the demand is more sensitive to changes in time (e.g., "instantaneous thrust of the engine"). "Equipment service life" The data is obtained by training on demand types and historical time-series data fluctuation amplitude.

[0047] Constructing a fuzzy hierarchical analysis (FAHP) model: Define the model's decision matrix as the association weight matrix output by S2. The correlation strength between units will be transformed into the importance weight of design indicators, and the model output will be a weight vector of equipment design indicators.

[0048] To quantize the mapping and satisfaction features, define specific parameters: Define This is the S3 demand-indicator mapping coefficient, with the subscript 3 indicating that it belongs to the S3 algorithm. Characterizing the first Each demand element is a characteristic unit. Characterizing the first Equipment design indicators are used for quantification. With the Equipment design specifications The degree of matching; definition Design index weights for S3 to represent Importance in equipment system design; definition The actual achievable value of the S3 design index is taken from... middle Test verification value; definition S3 demand satisfaction, characterizing The degree to which the equipment design specifications are met; the requirement satisfaction matrix is ​​defined as follows. .

[0049] 3.2 Algorithm Model Training Process: The training process requires training the dynamic Bayesian network and the fuzzy hierarchical analysis model separately, and then achieving fusion through bidirectional correction, clarifying the termination conditions for interactions with S2 and S4. The first step is to train the dynamic Bayesian network model: from... Extract historical demand feature units and matching data with design indicators to construct a training sample set. ,in This represents historical matching results (values ​​0 or 1, where 0 indicates no match and 1 indicates a match).

[0050] Calculate the time-series sensitivity for each demand element feature. Extracting historical time-series data Calculate the volatility coefficient ( Standard deviation, (where the mean is) then Substitute into the timing window optimization function Calculate the optimal window length, where , , This is the floor function.

[0051] sample set Based on optimal window length Extract time-series data, input it into a dynamic Bayesian network model, and set the initial conditional probability of the model to be... (i.e., initial assumptions) and With a matching probability of 50%, the conditional probability is updated iteratively using the maximum likelihood estimation method, with the iterative formula being: ,in The attribute anchor confidence level is set after S1 correction. for under conditions The prior probability is used, and this formula incorporates the confidence level of S1 into the probability update, achieving cross-step fusion. Iteration continues until the conditional probability converges (change ≤ 0.01), outputting the initial demand-indicator mapping coefficient. .

[0052] Convergence proof of dynamic Bayesian networks: (1) Model definition and premise: The parameters of the dynamic Bayesian network are the parameters in the conditional probability table (CPT). ,in For the first The node state at any given moment. For nodes The parent node set; the training data is a time-series sample set. ( (where is the time series length), assuming the samples are independent and identically distributed.

[0053] (2) Core proposition of convergence: When the training sample size At that time, the DBN parameter estimates based on maximum likelihood estimation (MLE) are... Converging to the true parameters ,Right now (Converges with probability).

[0054] (3) Proof process: 1. The objective function of maximum likelihood estimation is the log-likelihood function: In the formula This is an indicator function (it takes the value 1 if the condition is met, otherwise it takes the value 0).

[0055] 2. Regarding Taking the partial derivative and setting it to 0, we obtain the MLE estimate: ; 3. According to the law of large numbers, when hour: ; ; 4. Combining the above two equations, we can obtain: That is, the DBN parameters estimated by MLE converge to the true parameters in a probabilistic manner.

[0056] The temporal window optimization algorithm ensures sufficient and non-redundant training samples for each node by truncating time-series data to the optimal length, further improving the convergence speed; it recalculates every 50 new time-series data sets. and Update the window length of the sub-model.

[0057] The second step is to train the fuzzy hierarchical analysis model: This involves processing the correlation weight matrix output by S2. As the judgment matrix for fuzzy hierarchical analysis, a fuzzy consistency test is first performed to calculate the consistency ratio. ,like (Indicating inconsistency in the judgment matrix logic), then for Make corrections, the correction formula is as follows ,in The corrected association weights; if Then proceed directly to weight calculation. The weighted sum method is used to calculate the design index weights, and the formula is: ,in Given the total number of equipment design indicators, this formula transforms the correlation strength between units into indicator weights, ensuring integration with S2.

[0058] The third step is to perform bidirectional fusion optimization: This involves combining the results obtained from fuzzy hierarchical analysis... Input the dynamic Bayesian network model and adjust the conditional probabilities. The higher the weight of the indicator, the greater the adjustment magnitude of its corresponding conditional probability. The adjustment formula is as follows: Based on the optimized conditional probability, calculate the final demand-indicator mapping coefficient. At the same time, Feedback is sent to S2 as a constraint on the adjustment of the game strategy in S2. The higher the mapping coefficient of a unit pair, the higher the priority of its game strategy adjustment, thus optimizing the Nash equilibrium solution of S2.

[0059] The interaction between this step and S2 ends when the consistency ratio of the judgment matrix of the fuzzy hierarchical analysis reaches a certain level. When the conditional probability of the dynamic Bayesian network converges, the interaction between S2 and S3 ends; the interaction with S4 ends when the reserved interface receives the optimized requirement meta-feature unit set output by S4, and after correcting the mapping coefficients, the satisfaction calculation error is ≤0.05.

[0060] 3.3 Algorithm Model Application Process: The application process requires calculation of the requirement satisfaction degree and completion of bidirectional interaction based on the input of S2 and external data. First, the correlation weight matrix output by S2 is received. Topological association graph Receive the demand element feature unit set after S1 correction Simultaneously calling external support data , , .

[0061] The core element is the quantitative calculation of demand satisfaction, based on training data. , and In Construct the satisfaction calculation formula: The core function of this formula is to integrate three types of key parameters: Ensure the accuracy of the mapping (the higher the match, the greater the contribution). This reflects the importance of the indicator (the higher the weight, the greater the contribution). To ensure the objectivity of the calculations (based on actual test data), these three elements work together to achieve accurate quantification of requirement satisfaction. For example... This is the "Operating Ambient Temperature" unit. (Matching degree with the "material temperature resistance" index) = 0.9 (Weight of "Material Temperature Resistance") = 0.15 (Actual temperature resistance compliance rate) = 0.95, then this index is important for... The contribution is 0.9 × 0.15 × 0.95 = 0.12825. After summing the contributions of all design parameters, we get... .

[0062] Satisfaction based on individual demand element feature units Generate a demand satisfaction matrix The final step in the application is bidirectional interactive execution: the requirement satisfaction matrix is ​​then processed. The output is sent to S4, serving as the core basis for S4 conflict identification. The satisfaction difference between conflicting unit pairs directly quantifies the conflict intensity. Simultaneously, a reserved interface is provided to receive the optimized requirement meta-feature unit set output from S4. ,when When adjusting the quantization threshold of the middle unit, the formula is corrected. Corrected mapping coefficients This ensures that the satisfaction calculation matches the optimized requirements, achieving collaborative optimization between S3 and S4.

[0063] S4. Multi-objective Pareto optimization solution under topological constraints of demand conflict; Reference Figure 3The core objective of this step is to accurately identify the conflict relationships between demand element feature units. Under the premise of satisfying topological association constraints, it solves for the multi-objective optimal solution, generating an optimized set of demand element feature units. This optimized set must eliminate conflicts and ensure overall satisfaction. The multi-objective Pareto optimization algorithm under topological association constraints is used. It identifies conflicts through conflict intensity quantification, solves for the optimal solution through Pareto optimization, and forms a bidirectional interaction with S3 and S5: receiving input from S3... Calculate the conflict intensity and output. Correct the mapping coefficients of S3; output As the basis for S5 testing, the test deviation of S5 is used to optimize the Pareto solution set, and the three work together to improve the effectiveness of conflict resolution.

[0064] 4.1 Algorithm Model Construction Process: Model construction requires building a conflict identification model and a multi-objective optimization model, defining specific quantization parameters. First, the conflict identification model is constructed: the core logic is based on S2. Negative values ​​identify conflicting unit pairs. The larger the negative absolute value, the more significant the conflict; combined with S3... The intensity of conflict is quantified, and the conflict types are categorized into four types: performance-cost conflict, operation-support conflict, time-quality conflict, and compliance-flexibility conflict, covering the core conflict scenarios in equipment system design.

[0065] Secondly, a multi-objective optimization model is constructed: three main optimization objectives are defined: minimizing conflict intensity (eliminating inter-unit conflicts), maximizing overall demand satisfaction (ensuring demand fulfillment), and maximizing the attainability of design indicators (ensuring design feasibility); the optimization constraint is a topological relationship graph of S2. The associated path constraint means that the optimized unit cannot destroy the original core associated structure (such as the positive dependency relationship between "combat radius" and "power system power"), so as to ensure the integrity of the equipment system.

[0066] To quantify conflict and optimize features, define specific parameters: Define The S4 conflict strength is indicated by the subscript 4, which signifies belonging to the S4 algorithm. , Characterizing conflicting unit pairs for quantization and The degree of conflict; definition The S4 optimized requirement element feature set is the set of elements that are conflict-free after modification; definition The overall satisfaction threshold for S4 is set to 0.8, taking into account practical engineering needs (i.e., the optimized overall satisfaction must be ≥ 0.8); Define For overall demand satisfaction, To ensure the achievability of design metrics.

[0067] 4.2 Algorithm Model Training Process: The training process requires training both the conflict intensity calculation model and the Pareto optimization model. This bidirectional fusion improves optimization accuracy and clarifies the termination conditions for interactions with S3 and S5. The first step is to train the conflict intensity calculation model: selecting a historical conflict case dataset. ,in The conflict intensity is manually labeled (value range [0,1]). A conflict intensity calculation function is constructed, the core of which is to fuse the association weight of S2 with the satisfaction difference of S3. The function expression is as follows: This function embodies the logic that "the stronger the association conflict and the greater the difference in satisfaction, the higher the conflict intensity," for example... (Strong reverse conflict) (If the satisfaction level difference is large), then (Medium to high intensity conflict). The sample set... Input the model, calculate the error between the predicted conflict intensity and the labeled conflict intensity, iteratively optimize the function coefficients until the error is ≤0.05, and the model training is complete.

[0068] The second step is to train the Pareto optimization model: An improved non-dominated sorting genetic algorithm (NSGA-Ⅲ) is used, with core engineering parameters and an initial population size set. (i.e., the number of solutions in each iteration, based on the number of characteristic units of equipment requirements) Sure, when hour, when (Time), crossover probability (i.e., the probability of gene crossover; a single-point crossover strategy is adopted for crossovers with a quantified threshold based on demand), mutation probability. (i.e., the probability of gene mutation, for the mapping coefficient) The variation range is ±10%); the maximum number of iterations is 500; and the supplementary convergence criterion is that if the change in the optimal solution is ≤0.001 after 20 consecutive iterations, the iteration is terminated early.

[0069] Quantization threshold and mapping coefficient of demand element feature unit achievable design performance values As individuals within a population; the topological association graph The association path is used as a constraint condition. When adjusting an individual, the core association relationship must not be destroyed. The association path constraint threshold allows for the adjustment of attributes of a maximum of 3 association units; the conflict intensity is... Overall demand satisfaction Design target attainability As the fitness function (i.e. optimization objective), the population is iterated through selection (preserving high-quality individuals), crossover (merging the high-quality genes of two individuals), and mutation (introducing new genes to avoid local optima) until the population converges, resulting in a Pareto optimal solution set. This solution set contains multiple non-dominated solutions (i.e., it is impossible to improve all objectives at the same time; improving one objective requires sacrificing another).

[0070] The third step is to perform bidirectional fusion optimization: This involves optimizing the S3 requirement satisfaction matrix. Incorporate a fitness function to enhance the model's sensitivity to demand fulfillment; reserve an interface to receive test deviations from S5, and when the test deviation exceeds a threshold, adjust the weights of each objective in the fitness function (such as increasing the weight of "design metric attainability"), and iteratively optimize the Pareto solution set.

[0071] The interaction between this step and S3 ends under the following condition: After the mapping coefficients are corrected and fed back to S3, the interaction ends when the requirement satisfaction error is ≤0.05. The interaction with S5 ends when the test deviation from S5 is received and optimized. The interaction ends when the corresponding test deviation is ≤δ5.

[0072] 4.3 Algorithm Model Application Process: The application process requires inputs from S3 and S2 to perform conflict identification, optimization, and bidirectional interaction. First, it receives the requirement satisfaction matrix output by S3. Receive the correlation weight matrix output by S2 Topological association graph .

[0073] The first step is to perform conflict identification: traverse the weight matrix. All negative Identify all conflicting unit pairs; and In , Substituting into the conflict intensity calculation function, we obtain the values ​​for each conflict element pair. For example, the "Cost" and "High-Performance Material Selection" units , , ,but .

[0074] The second step is to perform multi-objective optimization: This involves resolving conflicting unit pairs. Overall demand satisfaction Design target attainability Input the trained Pareto optimization model, using the topological relationship graph. Using the associated paths as constraints, solve for the Pareto optimal solution set. Select the optimal solution from the solution set based on the following selection criteria: (No conflict) and (Overall satisfaction meets the standard); Based on the optimal solution, adjust the quantification threshold and constraints of the demand element characteristic unit. For example, the quantification threshold of the "cost" unit is adjusted from 5 million yuan to 5.2 million yuan, and the quantification threshold of the "high-performance material selection" unit is adjusted from "imported materials" to "imported core components + domestic auxiliary components" to eliminate the conflict between the two and ensure the overall satisfaction.

[0075] Based on the adjusted unit attributes, an optimized requirement meta-feature unit set is generated, defined as follows: ,in For the revised first Each requirement element is a characteristic unit. The final step in the application is bidirectional interactive execution: [The text abruptly ends here, likely due to an incomplete sentence or a formatting error.] The output is sent to S5, serving as the core basis for S5's hierarchical physical testing. The test object must be consistent with... The attributes match exactly; at the same time Feedback is sent to S3, and the formula is corrected. Correcting the mapping coefficients of S3 This ensures that the satisfaction calculation of S3 matches the optimized requirements, thereby achieving collaborative optimization between S3 and S4.

[0076] S5. Digital verification of requirements and closed-loop iterative optimization; Reference Figure 4 The core objective of this step is to verify the optimized set of requirement meta-feature units through hierarchical physical testing. The feasibility of the requirements is assessed by identifying deviations through reverse verification of demand tracing, forming a closed-loop iteration throughout the entire process to ensure that the final output of digital requirements is accurate and feasible. The hierarchical physical testing-demand tracing reverse verification algorithm used covers all dimensions of requirements through three levels of physical testing, links previous steps through tracing verification, and establishes bidirectional interaction with S4: receiving data from S4... As a basis for testing, the Pareto solution set of S4 is optimized by outputting test deviation and cascading effect data; at the same time, it is reversed to S3, S2, and S1 to form a closed loop of the entire process, with all steps being optimized in a coordinated manner.

[0077] 5.1 Algorithm Model Construction Process: Model construction requires building a hierarchical physical testing model and a traceability verification model, defining specific quantitative parameters and end conditions for the entire process interaction. First, a hierarchical physical testing model is constructed: based on the hierarchical structure of the equipment system, testing levels are divided, for example, into three categories: component-level testing, subsystem-level testing, and system-level testing. The test objects are respectively... Matching equipment component prototypes, subsystem prototypes, and system prototypes are implemented. For example, the "material temperature resistance" unit corresponds to component-level testing, the "power system coordination" unit corresponds to subsystem-level testing, and the "combat effectiveness" unit corresponds to system-level testing, ensuring that testing covers all requirement units. Secondly, a requirement sourcing and reverse verification model is constructed: based on the S2 topology graph. Clearly define the associated unit chain for each demand element characteristic unit. When a test deviation occurs in a certain unit, verify whether the deviation triggers a chain reaction through the associated chain. For example, if the deviation of the "material temperature resistance" unit is detected, it is necessary to verify whether its associated units such as "working environment temperature" and "power system heat dissipation" are affected.

[0078] To quantify and verify features, define specific parameters: Define The S5 component-level test satisfaction level is represented by the subscript 5, which indicates belonging to the S5 algorithm and 1, which indicates component-level. To ensure S5 subsystem-level test satisfaction, S5 system-level test satisfaction; definition , , The weights for the S5 test level are determined based on the topological relationship graph of S2. The hierarchical association strength of the middle unit is determined to satisfy... For example, units with high component-level correlation strength. Set to 0.4 at the subsystem level, 0.3 at the subsystem level, and 0.3 at the system level; define. To characterize the S5 comprehensive test satisfaction level. Final test satisfaction; definition The S5 test deviation value represents the difference between the overall test satisfaction and the S3 requirement satisfaction; defined... The S5 deviation threshold is set to 0.05 based on practical engineering considerations; (Definition) The scope of the S5 chain effect is characterized by... The number of related unit deviations caused by the deviation; the end condition for the entire process interaction is defined as: the deviation of all units... and (Without any chain reaction) At this point, the verification is successful, and the final digital requirement specification is output.

[0079] 5.2 Algorithm Model Training Process: The training process requires determining the test level weights and the chain effect judgment function to ensure the accuracy of testing and validation. The first step is to determine the test level weights: based on the S2 topology graph. Calculate the correlation strength of each demand element feature unit at different levels, and define... For component-level correlation strength ( (a subset of component-level units) For subsystem-level correlation strength ( (a subset of subsystem-level units) For system-level correlation strength ( (For system-level unit subsets). Weights are calculated based on association strength, using the following formula: , , The calculation logic is consistent. This process transforms the topological association strength of S2 into test weights, ensuring that the test focus matches the association features.

[0080] The second step is to train the chain effect judgment function: select a historical test deviation case dataset. ,in This represents the manually labeled scope of cascading effects. A cascading effect judgment function is constructed, with the expression: This function embodies the logic that "the higher the correlation strength and the greater the deviation, the wider the cascading effect," for example... For the "material temperature resistance" unit, Associated units If the sum is 1.2, then Based on the threshold (0.1), it was determined that there was no significant linkage effect. The sample set... Input the model, iteratively optimize the function coefficients until the chain effect prediction accuracy is ≥95%, and the model training is complete.

[0081] 5.3 Algorithm Model Application Process: The application process requires performing hierarchical testing, deviation calculation, source tracing verification, and closed-loop iteration based on the input of S4, ultimately outputting a digital requirement specification that meets the requirements. First, it receives the optimized requirement meta-feature unit set output by S4. Receive the topology graph output by S2 The requirement satisfaction matrix of receiving S3 output .

[0082] The first step is to conduct tiered physical testing, clarifying the testing details and complete quantitative indicators for each tier: Component-level testing: targeting The test indicators for key component performance units (such as "material temperature resistance" and "bearing life") include: component parameter compliance rate (quantification threshold ≥ 98%), component environmental adaptability (quantification threshold: stable operation ≥ 1000h at -50℃~70℃, humidity 10%-95%, vibration frequency 10-2000Hz), component reliability (quantification threshold: mean time between failures (MTBF) ≥ 2000h), and corrosion resistance (only for ship / armored equipment components, quantification threshold: no rust after ≥ 1000h salt spray test, in accordance with GB / T10125-2021 standard). The test method involves placing the component in a comprehensive environmental test chamber to simulate extreme environments and continuously monitoring its operating status. Data is collected using high-precision sensors, and calculations are performed. For example, the compliance rate of the "material temperature resistance" unit in an environment of -50℃ is 0.98. .

[0083] Subsystem-level testing: targeting The focus is on units that coordinate with subsystems (such as "power system coordination" and "navigation-communication coordination"), combined with S2. Verify the constraint satisfaction between subsystems; examples of test metrics include subsystem collaboration efficiency (quantization threshold ≥ 95%), subsystem interface compatibility (quantization threshold: interface adaptation success rate 100%), subsystem fault tolerance rate (quantization threshold ≥ 90%), and collaboration latency (quantization threshold ≤ 50ms); the test method involves building a subsystem integration platform, simulating collaboration scenarios, using a high-speed oscilloscope to collect signal transmission delays between subsystems, collecting collaboration data, and calculating... .

[0084] System-level testing: targeting The focus is on the units of the overall capability of the equipment system (such as "combat effectiveness" and "life cycle cost"), combined with S3. Verify the overall requirement fulfillment; examples of test indicators include system combat effectiveness (quantification threshold ≥90%, test method is to construct a real combat simulation scenario and calculate the success rate of equipment in completing combat missions), system life cycle cost (quantification threshold ≤105% of budget, statistical analysis of costs throughout the entire stages of R&D, production, operation and maintenance, and decommissioning), system compliance (quantification threshold: 100% compliance with corresponding industry standards, item-by-item verification against GB, GJB, and other standards), and system response speed (quantification threshold: response time after receiving an instruction ≤1 second, recording response time through the command and control system); the test method is to build a full-system simulation test field, conduct actual equipment testing, collect system-level data, and calculate... .

[0085] The second step is to calculate the overall test satisfaction and deviation values: The formula for calculating the overall test satisfaction is as follows: The results of the three levels of testing are integrated to ensure the comprehensiveness of the evaluation; the formula for calculating the test deviation value is as follows: The gap between the quantitative test results and the theoretical satisfaction.

[0086] The third step is to perform reverse verification of requirements sourcing: [This will...] Substitute into the chain reaction judgment function and calculate the chain reaction effect of each unit. Combined with the topological relationship diagram of S2 Tracing the source of the deviation, for example The unit exceeding the standard is "power system coordination," which can be traced back to the mapping coefficient of S3 through the correlation chain. The value is too low, which can be further traced back to the association weight of S2. Calculate the deviation and identify its root cause.

[0087] The fourth step involves closed-loop iterative optimization, strictly adhering to the end conditions of the entire process interaction: If all units and The entire interaction process is complete and the verification is successful; Combination The requirements are transformed into digital requirements specifications for equipment system design. The specifications include the attributes, quantification thresholds, and test verification results of each requirement element characteristic unit, serving as the final input basis for equipment system design.

[0088] If there are units or The termination condition was not met; and Feedback is sent to S4 to adjust the fitness function weights of the Pareto optimization model and to re-solve for the optimal solution; the optimized solution... The mapping coefficients are then passed to S3, S2, and S1 respectively to correct them in turn. Association weight Attribute anchoring confidence Repeat steps S1 to S5 until all units meet the requirements. and The entire interactive process has ended.

[0089] Overview of bidirectional interaction and data association between steps: 1. S1 ↔ S2: Output of S1 and Up to S2, supporting game theory calculations; S2 output and Correction of S1 The interaction ends under the corrected condition. 2. S2 ↔ S3: S2 output S3 is used as the FAHP judgment matrix; S3 output The optimal game equilibrium solution for S2 is determined by the following condition: And the conditional probability converges. 3. S3↔S4: S3 output Calculation up to S4 S4 output Correction of S3 The interaction ends when the satisfaction error is ≤0.05. 4. S4 ↔ S5: S4 output S5 serves as the basis for testing; S5 output and The Pareto solution set of S4 is optimized, and the end condition for the entire interactive process is: and 5. Closed-loop process: Feedback data from S5 is integrated into S1-S4, enabling deep fusion and iterative optimization of algorithms across all steps, ensuring that the final output of digital requirements is standardized, accurate, and feasible.

[0090] Simplified Case Study on Requirements Analysis of Small Tactical Reconnaissance UAVs: This case study selects small tactical reconnaissance UAVs (aviation equipment, Taking as the analysis object, four core requirement element feature units were extracted.

[0091] Case premise definition: 1. Original requirements for UAVs: "Small tactical reconnaissance UAVs need to have an endurance of ≥4 hours, a reconnaissance resolution of ≤0.1m, a cost of ≤200,000 yuan, anti-interference performance that meets the GJB151B-2013 standard, and a life cycle of ≤5 years." This sentence is 38 characters long, which meets the threshold requirement of 20-50 characters for sentence segmentation in S1.

[0092] 2. The equipment type is aviation equipment, and the equipment type adaptive coefficient is... .

[0093] 3. Number of core requirement meta-feature units The four units are respectively Battery life requirement, quantified threshold ≥4h, constraint condition is the entire service life; The reconnaissance resolution requirement is a quantization threshold of ≤0.1m, and the constraint is a visible light reconnaissance scenario. Cost requirements, with a quantitative threshold of ≤200,000 yuan, and constraints covering the entire R&D and production stages; The anti-interference requirement is quantified to meet the GJB151B-2013 standard, and the constraint is a complex electromagnetic environment.

[0094] 4. Number of design indicators The three indicators are respectively Battery capacity, actual achievable value ; Optical lens accuracy, actual achievable value ; Anti-interference module selection, actual achievable value This value represents the module compliance rate.

[0095] S1. Heterogeneous deconstruction and modular extraction of equipment requirement meta-characteristics 1.1 Three-level semantic segmentation and attribute anchoring: In the sentence-level segmentation stage, the original requirement is a complete sentence of 38 characters, meeting the threshold and requiring no further splitting. In the phrase-level segmentation stage, noun phrases including battery life, reconnaissance resolution, cost, anti-interference capability, and GJB151B standard are extracted from the original requirement; verb phrases including "must meet," "≤," and "≥" are extracted; and attribute phrases including "4h," "0.1m," and "200,000 yuan" are extracted. All phrases meet the threshold of "noun phrases containing technical terms + attribute words, and verb phrases containing action verbs + technical objects." In the keyword-level segmentation stage, core keywords are extracted: battery life, 4h, reconnaissance resolution, 0.1m, cost, 200,000 yuan, anti-interference capability, and GJB151B. Although these keywords all have a frequency of 1, they all appear in the core sentence of the requirement, meeting the extraction threshold. In the attribute anchoring stage, five types of attributes—feature name, data type, quantification threshold, constraint condition, and associated requirement type—are matched to the four requirement meta-feature units. After anchoring, the initial attribute anchoring confidence score is output. corresponding , corresponding , corresponding , corresponding .

[0096] 1.2 Output Data: This step outputs the set of requirement meta-feature units. and attribute anchor confidence sequence All of the above data are used as input parameters for S2 to support subsequent game calculations and correlation modeling.

[0097] S2. Topological association modeling and weight assignment of demand element feature units 2.1 Association Relationship Determination: For each pair of units in the demand element feature unit set, the association type is determined. The specific determination results are as follows: Unit Pair This is a positive dependency relationship, determined by the fact that longer endurance leads to longer reconnaissance time and makes it easier to meet reconnaissance resolution requirements; unit pair This is a conflicting relationship, determined by the fact that extended battery life requires larger capacity batteries, directly leading to increased costs; cell pair The relationship is one of reverse conflict, determined by the fact that improved reconnaissance resolution requires higher precision optical lenses, and higher lens precision results in higher procurement costs; unit pair The positive dependency relationship is determined by the requirement for stable transmission of high-resolution reconnaissance data, necessitating improved anti-jamming capabilities of the equipment; unit to unit The determination is based on the absence of any correlation between battery life and anti-interference capability; there is no direct logical constraint between the unit pairs. The relationship is considered to be one of reverse conflict, based on the fact that the procurement cost of high-performance anti-interference modules is relatively high, which will directly increase the overall cost of the equipment.

[0098] 2.2 Game Payoff and Weight Calculation: First, set the initial attribute thresholds for each demand element feature unit. ,in The initial threshold is 4h. The initial threshold is 0.1m. The initial threshold is 200,000 yuan. The initial threshold is to conform to the GJB151B standard; then historical test data is selected as the compliance value after unit adjustment. ,For example When the threshold is adjusted to 5h, The acceptable value is 0.9. The target value is 0.7.

[0099] Substitute the equipment type adaptive coefficient, attribute anchoring confidence level, initial threshold, and adjusted target value into the optimized payoff function. Calculate the game payoff for typical unit pairs: Calculate the unit pair The payment value, substituted into the numerical value, yields... This value is positive, indicating that... right Generates positive returns; computing units for The payment value, substituted into the numerical value, yields... This value is negative, indicating right This results in negative payoffs. By iteratively solving for the Nash equilibrium, the final game payoff matrix is ​​obtained, and then the correlation weight matrix is ​​calculated based on this matrix. The specific matrix form is as follows: The association weight matrix is ​​fed back to S1 to correct the attribute anchoring confidence. The correction formula is as follows: .by For example, the sum of the absolute values ​​of their association weights Substituting into the formula, we can obtain Since the confidence level ranges from [0,1], the value is truncated to 1.0.

[0100] 2.3 Output Data: This step outputs the correlation weight matrix. Topological association diagram with demand element features The nodes of the topological association graph are sets of demand element feature units. edge set For pairs of units that are related, the edge weights are the association weight matrix. The non-zero elements in the above data are all used as input parameters for S3 to support subsequent dynamic mapping and satisfaction calculation.

[0101] S3. Multi-dimensional digital mapping and dynamic adjustment of demand satisfaction 3.1 Timing Window Optimization: First, determine the timing sensitivity of each requirement element's characteristic unit, where... Time sensitivity of battery life requirements This falls under the category of low-sensitivity requirements because the time-series data on battery life shows relatively small fluctuations. Temporal sensitivity of reconnaissance resolution requirements This is a highly sensitive requirement because time-series data with high resolution is subject to significant fluctuations due to environmental factors.

[0102] Substitute timing sensitivity into the window optimization function Calculate the optimal timing window length for each unit: Calculate The optimal window length can be obtained by substituting the values. The data collection period is 1 day; calculation The optimal window length can be obtained by substituting the values. The data collection cycle is hours.

[0103] 3.2 Mapping Coefficient and Satisfaction Calculation: Based on the optimal time-series window length, historical data is extracted, a dynamic Bayesian network model is trained, and the initial demand-indicator mapping coefficient is output. The specific values ​​are: and The initial mapping coefficient for battery capacity is 0.9. and The initial mapping coefficient for optical lens accuracy is 0.1. and The initial mapping coefficient of the anti-interference module is 0.0; and The initial mapping coefficient for battery capacity is 0.1. and The initial mapping coefficient for optical lens accuracy is 0.95. and The initial mapping coefficient of the anti-interference module is 0.2; and The initial mapping coefficient for battery capacity is 0.8. and The initial mapping coefficient for optical lens accuracy is 0.85. and The initial mapping coefficient of the anti-interference module is 0.7; and The initial mapping coefficient for battery capacity is 0.0. and The initial mapping coefficient for optical lens accuracy is 0.2. and The initial mapping coefficient of the anti-interference module is 0.9.

[0104] The correlation weight matrix output by S2 As the judgment matrix for fuzzy hierarchical analysis, the weights of the design indicators are calculated. ,in Battery capacity has a weight of 0.3. The weight of optical lens accuracy is 0.4. The weight of the anti-interference module is 0.3. Substitute the design index weights into the mapping coefficient correction formula. Correct the initial mapping coefficients. and Taking the mapping coefficient as an example, the corrected value is Substitute the revised mapping coefficients, design index weights, and actual achievable values ​​into the requirement satisfaction calculation formula. Calculate the requirement satisfaction level of each unit. For example, substituting the values, we can obtain After normalizing the value, we get Similarly, the demand satisfaction level of other units is calculated, ultimately yielding the demand satisfaction matrix. .

[0105] 3.3 Output Data: This step outputs the requirement satisfaction matrix. This matrix serves as the input parameter for S4, supporting subsequent conflict intensity calculations and multi-objective optimization.

[0106] S4. Multi-objective Pareto optimization solution under topological constraints of demand conflict 4.1 Conflict Intensity Calculation: Substitute the negative weight values ​​from the correlation weight matrix output by S2 and the values ​​from the demand satisfaction matrix output by S3 into the conflict intensity calculation formula. Calculate the conflict intensity of typical conflict element pairs. Calculate element pairs. The intensity of the conflict can be obtained by substituting the values. ; Computing unit The intensity of the conflict can be obtained by substituting the values. .

[0107] 4.2 Multi-objective optimization solution: Three core objectives are set for this optimization: minimizing conflict intensity, overall demand satisfaction, etc. Maximize the accessibility of design metrics, and simultaneously output the topology graph from S2. The associated paths in the model are used as optimization constraints to ensure that the optimized units do not disrupt the core interconnected structure of the equipment system. Engineering parameters for Pareto optimization are set, including the population size. Crossover probability Probability of mutation The maximum number of iterations is 500. The conflict intensity, overall requirement satisfaction, and design index attainability are used as fitness functions, and the improved non-dominated sorting genetic algorithm is used for iterative solving to obtain the Pareto optimal solution set.

[0108] The optimal solution that meets the requirements is selected from the set of optimal solutions, and the corresponding adjustment scheme for the characteristic units of the demand element is as follows: The cost threshold has been relaxed to 220,000 yuan. The reconnaissance resolution threshold has been adjusted to 0.12m. The battery life threshold remains unchanged at 4 hours. The anti-interference threshold remains unchanged, conforming to the GJB151B standard. Based on this adjustment scheme, an optimized set of requirement element feature units is generated. Calculate the overall demand satisfaction after optimization. This value is ≥0.8, and the conflict intensity of all conflicting element pairs is... This satisfies the optimization objective.

[0109] 4.3 Output Data: This step outputs the set of optimized requirement meta-feature units. This unit set serves as the input parameter for S5, supporting subsequent hierarchical physical testing and closed-loop verification.

[0110] S5. Digital Validation and Closed-Loop Iterative Optimization of Requirements 5.1 Layered Testing Execution: Component-level testing was performed, targeting three types of components matched with the optimized unit set: batteries, optical lenses, and anti-interference modules. Test indicators and corresponding thresholds were: battery life ≥ 4 hours, optical lens resolution ≤ 0.12 μm, and anti-interference performance conforming to GJB151B standard. Additionally, component reliability was tested, requiring a mean time between failures (MTBF) ≥ 2000 hours. Test results showed that the actual battery life was 4.2 hours, the actual optical lens resolution was 0.11 μm, and the anti-interference performance compliance rate was 98%. All indicators met the threshold requirements. Based on the test results, the component-level test satisfaction rate was calculated. corresponding , corresponding , corresponding , corresponding .

[0111] Subsystem-level testing was conducted, targeting two types of subsystems: the flight control-reconnaissance subsystem and the communication-anti-jamming subsystem. The test indicators and corresponding thresholds were: subsystem coordination latency ≤ 50ms, interface compatibility 100%, coordination efficiency ≥ 95%, and fault tolerance ≥ 90%. Test results showed that the actual subsystem coordination latency was 35ms, the interface compatibility pass rate was 100%, the coordination efficiency was 96%, and the fault tolerance rate was 92%, with all indicators meeting the threshold requirements. Based on the test results, the subsystem-level test satisfaction level was calculated. corresponding , corresponding , corresponding , corresponding .

[0112] System-level testing was conducted on a small tactical reconnaissance UAV. The test indicators and corresponding thresholds were: system operational effectiveness ≥90%, total lifecycle cost ≤220,000 RMB, system response speed ≤1 second, and 100% compliance with industry standards. Test results showed that the actual system operational effectiveness was 92%, the actual total lifecycle cost was 215,000 RMB, the actual system response speed was 0.8 seconds, and the compliance test pass rate was 100%. All indicators met the threshold requirements. Based on the test results, the system-level test satisfaction level was calculated. corresponding , corresponding , corresponding , corresponding .

[0113] 5.2 Comprehensive Test Satisfaction and Deviation Calculation: Based on the topology graph output by S2, the test level weights are calculated, and the component level weights are finally determined. Subsystem-level weights System-level weights The sum of the three weights is 1.

[0114] Substitute the test satisfaction scores and corresponding weights at each level into the overall test satisfaction calculation formula. Calculate the overall test satisfaction of each unit. For example, substituting the values, we can obtain Similarly, the overall test satisfaction scores for other units are calculated, ultimately yielding the overall test satisfaction score sequence. Substitute the overall test satisfaction level and the optimized requirement satisfaction level from S3 output into the deviation calculation formula. The test deviation value of each unit is calculated, and finally the deviation value sequence is obtained. .

[0115] 5.3 Closed-loop iteration and termination determination: Setting a deviation threshold By comparing the deviation value sequence, we can see that... The corresponding deviation value is 0.059. The corresponding deviation value is 0.086. Both values ​​exceed the preset threshold, while the deviation values ​​of the remaining units meet the requirements.

[0116] By tracing back to the source of demand, we can trace the root cause of deviations. The cost discrepancy stems from fluctuations in the purchase price of anti-interference modules. The deviation stemmed from insufficient accuracy in electromagnetic environment simulation during the testing phase. The deviation value and cascading effect data were fed back to S4, and the fitness function weights of the Pareto optimization model were adjusted to increase the weight of cost indicators. Simultaneously, the anti-interference module selection scheme was optimized, replacing it with a more cost-effective domestic module.

[0117] Based on the adjusted parameters, steps S3 to S5 were re-executed. After the second test, the deviation values ​​of all units were ≤0.05, and the cascading effect range was limited. This satisfies the conditions for ending the entire interaction process.

[0118] Case Data Flow Overview: S1 outputs the set of requirement meta-feature units and attribute anchoring confidence, which is input to S2 to support correlation modeling; S2 outputs the correlation weight matrix and topological correlation graph, which is input to S3 to support dynamic mapping, while also providing feedback to correct the confidence of S1; S3 outputs the requirement satisfaction matrix, which is input to S4 to support conflict resolution, while also providing feedback to optimize the game equilibrium solution of S2; S4 outputs the optimized unit set, which is input to S5 to support physical testing, while also providing feedback to correct the mapping coefficients of S3; S5 outputs the deviation value and cascading effect data, which is fed back to S4 to optimize the Pareto solution set, ultimately forming a closed-loop iteration throughout the entire process, outputting the digital requirement specification for small tactical reconnaissance UAVs.

[0119] Key findings of this case study: This method achieves precise digital analysis of the requirements for small tactical reconnaissance UAVs through core innovations such as adaptive weight adjustment based on equipment type, dynamic Bayesian network time-series window optimization, and bidirectional interactive closed-loop between steps. The optimized requirement scheme eliminates the core conflict between cost and performance while meeting the core performance requirements of tactical reconnaissance missions. Real-world data validates the engineering practicality and innovation of the method, distinguishing it from traditional linear requirement analysis processes.

Claims

1. An equipment system design requirement digitalization analysis method, characterized by, include: S1. Perform multi-granularity semantic segmentation and attribute anchoring on the original requirements of the equipment system in all dimensions to obtain a set of requirement meta-feature units and attribute anchoring confidence. The multi-granularity semantic segmentation includes three levels of segmentation: sentence level, phrase level, and keyword level. The attribute anchoring is the matching feature attribute of requirement meta-feature units. S2. Based on the set of demand element feature units and the attribute anchoring confidence, a demand element feature topological association graph is constructed using a fusion algorithm of asymmetric game theory and topological graph theory to obtain the association weight matrix. The association weight matrix is ​​then fed back to S1 to correct the attribute anchoring confidence. S3. Based on the aforementioned correlation weight matrix and the topological correlation diagram of demand element features, a dynamic Bayesian network and fuzzy hierarchical analysis fusion algorithm is used to calculate the demand satisfaction matrix and its demand-indicator mapping coefficients, and the demand satisfaction matrix is ​​fed back to the S2 optimization game equilibrium solution. S4. Based on the demand satisfaction matrix, the association weight matrix and the demand element feature topology graph, the multi-objective Pareto optimization algorithm under topology constraints is used to identify and resolve demand conflicts, and an optimized demand element feature unit set is obtained. The optimized demand element feature unit set is fed back to S3 to correct the demand-index mapping coefficient of the demand satisfaction matrix. S5. Perform hierarchical physical testing based on the optimized requirement element feature unit set to obtain the comprehensive test satisfaction and test deviation value. Identify the associated unit chain of the deviation unit based on the requirement element feature topology association diagram and association weight matrix, calculate the chain effect data, and determine whether the test deviation value or chain effect data meets the preset requirements. If not, feed the test deviation value and chain effect data back to S4 to optimize the Pareto optimal solution set, and repeat steps S3 to S5 until the verification requirements are met, and output the digital requirement specification for equipment system design.

2. The method of claim 1, wherein, The algorithm implementation of multi-granularity semantic segmentation in S1 includes: sentence-level segmentation using the complete expression of a single demand intent as the criterion to select sentence fragments that meet semantic integrity; phrase-level segmentation based on part-of-speech tagging and dependency parsing to extract noun phrases containing technical terms and attribute words, and verb phrases containing action verbs and technical objects; and keyword-level segmentation combining word frequency statistics and demand core score to select key terms in the demand title or core sentence, forming a keyword set that corresponds one-to-one with the demand meta-feature unit.

3. The method of claim 1, wherein the equipment system design requirement digitalization analysis method is characterized by, The core of the asymmetric game theory and topological graph theory fusion algorithm in S2 includes: constructing an asymmetric game model among demand element feature units, using attribute anchoring confidence as the initial weights of the game participants, introducing equipment type adaptive coefficients to adjust the game payoff function, and obtaining the correlation strength between units by iteratively solving the Nash equilibrium; constructing a demand element feature topological correlation graph based on the correlation strength, where nodes are demand element feature units, edge weights are the corresponding values ​​in the correlation weight matrix, and positive dependencies and negative conflicts are distinguished by the positive and negative attributes of the edges.

4. The digital analysis method for equipment system design requirements according to claim 1, characterized in that, The calculation logic of the association weight matrix in S2 is as follows: For any two demand element feature units, the mutual influence degree between the two parties is calculated through an asymmetric game model. The absolute value of the influence degree is used as the basic value of the association weight. The positive / negative sign of the weight is assigned in combination with the positive / negative type of the association relationship between the units. Finally, a symmetric association weight matrix with the same dimension as the size of the demand element feature unit set is formed.

5. The digital analysis method for equipment system design requirements according to claim 1, characterized in that, The implementation process of the dynamic Bayesian network and fuzzy hierarchical analysis fusion algorithm in S3 is as follows: based on the topological association graph of demand element features, the initial network structure of the dynamic Bayesian network is constructed; fuzzy hierarchical analysis is used to transform the association weight matrix into a judgment matrix and calculate the weight of each design index. Based on the conditional probability table of the dynamic Bayesian network optimized by the weight, the demand-indicator mapping coefficients are obtained through training, and then the demand satisfaction matrix is ​​calculated by combining the actual achievability of the design indicators.

6. The digital analysis method for equipment system design requirements according to claim 1, characterized in that, The training process of the dynamic Bayesian network in S3 introduces a time-series window optimization mechanism: the time-series sensitivity is determined by analyzing the fluctuation range of historical time-series data of the required meta-features, and the data acquisition window length is dynamically adjusted based on the time-series sensitivity to select suitable time-series data for training, thereby avoiding redundant data interference caused by excessively long windows or insufficient feature capture caused by excessively short windows.

7. The digital analysis method for equipment system design requirements according to claim 1, characterized in that, The multi-objective Pareto optimization algorithm under topological association constraints in S4 includes: taking the minimization of demand conflict intensity and the maximization of overall demand satisfaction as the core optimization objectives, using the core association path in the topological association graph of demand element features as the constraint condition to ensure that the key dependencies between units are not destroyed during the optimization process; using an improved non-dominated sorting genetic algorithm to solve the Pareto optimal solution set, and selecting the optimized demand element feature unit set that meets the requirements of engineering implementation from the solution set.

8. The digital analysis method for equipment system design requirements according to claim 1, characterized in that, The logic for identifying and resolving demand conflicts in S4 is as follows: based on the negative weights in the association weight matrix, potential conflicting unit pairs are located, the demand satisfaction difference between the conflicting unit pairs is calculated, and the product of the absolute value of the negative weights and the satisfaction difference is used as the conflict intensity; the resolution process involves adjusting the characteristic attribute thresholds of the conflicting units and simultaneously verifying the chain effect of the adjustment on the associated units until the conflict intensity drops below the preset threshold.

9. The digital analysis method for equipment system design requirements according to claim 1, characterized in that, The algorithm for calculating the chain effect data in S5 is as follows: based on the topological association graph of the demand element feature, traverse the direct and indirect associated units of the deviation unit to form an associated unit chain; calculate the deviation transmission coefficient by combining the weight values ​​in the association weight matrix; obtain the estimated deviation of each associated unit by multiplying the deviation transmission coefficient with the test deviation value; and summarize the estimated deviation, the number of affected units and the transmission path to form the chain effect data.

10. The digital analysis method for equipment system design requirements according to claim 1, characterized in that, In S3, when calculating the demand satisfaction matrix, historical test data of similar equipment from the equipment test database and design index threshold data from the industry standard library are introduced as reference benchmarks. The correlation between demand element features and design indexes is established through the demand-index mapping coefficient. The degree of demand satisfaction is quantified by combining the actual achievable level of the design indexes, thus forming the demand satisfaction matrix.

Images