Digital twin model credibility evaluation method and system

Abstract

The invention discloses a digital twin model credibility evaluation method and system, and belongs to the technical field of communication twin networks. The method comprises the following steps: constructing an index system framework for credibility evaluation of the digital twin model, and obtaining corresponding information required for credibility evaluation according to credibility evaluation indexes of the digital twin model; secondly, converting into credibility evaluation grades in combination with demand analysis and expert experience, converting the obtained credibility evaluation grades into evidences through a cloud model, and obtaining the support degree of each evidence for different credibility grades in the recognition framework; and finally, calculating conflicts among the evidences through a Jousseme distance, acquiring a corresponding evidence correction weight for the evidence conflicts and evidence source credibility or evidence importance by using Shapley, acquiring an evidence discount factor in combination with a Deng entropy, and fusing the evidences to acquire final credibility evaluation. According to the method, credibility evaluation can be carried out on the indexes of the model by integrating multiple factors, the accuracy of credibility evaluation is effectively improved, and the uncertainty is reduced.

Description

Technical Field

[0001] The present invention relates to a digital twin model credibility assessment method and system, and belongs to the technical field of communication twin networks. Background Art

[0002] With the rapid development of storage, computing and communication technologies, digital twin technology has been widely used in various fields such as industry and agriculture. By creating a virtual model that is highly consistent with the physical entity and connecting the two, digital twin technology realizes the interaction between the physical system and the virtual model. The virtual model simulates the actual behavior of the physical entity, realizes the simulation behavior such as simulation, analysis and prediction of the physical system, and reacts the obtained simulation results to the physical system, so as to realize the decision-making and optimization of the physical system. This technology can not only continuously improve the system performance, but also accelerate the cycle of system iteration and optimization, and significantly reduce the R&D cost. However, with the expansion of system scale and increasing complexity, there are complex interactions between the internal mechanisms and components of the system, and the credibility requirements of the digital twin model are also increased.

[0003] In the industrial communication twin network, it is crucial to ensure the high credibility of the model. Any error may lead to communication interruption and thus losses. The credibility assessment of the communication twin model is to ensure that it can truly reflect the state and behavior of the physical network. Whether the digital twin system can be correctly applied depends on its credibility level. A highly credible communication twin model can provide a reliable basis for decision-making, optimize communication stability and information security, and reduce uncertainty in practical applications. Therefore, the credibility assessment of the digital twin model is one of the key steps to ensure the application of the digital twin system.

[0004] At present, many methods have emerged in the research on the credibility assessment of simulation models at home and abroad. Ran Peizhi from Harbin Institute of Technology proposed a credibility assessment method for simulation models based on improved fuzzy comprehensive evaluation. When performing simulation credibility assessment based on the traditional fuzzy comprehensive evaluation method, the weights are usually judged based on expert experience, and the fuzzy operation selection is unreasonable. This leads to inaccurate credibility assessment methods. The weights of indicators are determined based on the minimum membership weighted average deviation. In the fuzzy synthesis stage, the membership weighted average deviation is used to obtain a comprehensive evaluation, so as to obtain accurate simulation credibility assessment results; ("Simulation credibility assessment method based on improved fuzzy comprehensive evaluation", Ran Peizhi et al., Journal of System Simulation, Vol. 32, No. 12, pp. 2469-2474, August 2020.);

[0005] Ye Ziheng proposed a simulation model credibility assessment method based on deviation propagation for simulation model reuse in two different application environments: physical objects and numerical solutions. However, as the complexity of model combination changes, it may not be able to adapt to all types of complex combinations (“Credibility Assessment Method for Reuse Models of Complex Simulation Systems”, Ye Ziheng et al., Journal of System Simulation, Vol. 32, No. 12, pp. 2475-2484, August 2020); Wei Li proposed a multivariate output and uncertainty model validation method based on evidence theory, which extracts data features of system response under uncertainty, represents the validation data as evidence theory, and then uses improved traditional evidence distance to measure the consistency between simulation output and experimental observation. However, this method cannot solve the multivariate validation problem with correlated outputs well (“An evidence theory-based validation method for models with multivariate outputs and uncertainty”, Li Wei, SIMULATION-TRANSACTIONS OF THE SOCIETY FOR MODELING AND SIMULATION INTERNATIONAL, Vol. 97, No. 12, pp. 821-834, December 2021. );

[0006] In order to evaluate the credibility of simulation models with high dynamic characteristics, a study proposed a pattern-based validation method to extract hidden patterns existing in the simulation model and its reference data, and evaluate the model performance from different perspectives. However, as the number of simulation outputs increases, the computational complexity and time consumption of the proposed method will increase ("A pattern-based validation method for the credibility evaluation of simulation models", Y Laili, L Zhang, Y Luo, SIMULATION-TRANSACTIONS OF THE SOCIETY FOR MODELING AND SIMULATION INTERNATIONAL, Vol. 96, No. 2, pp. 151-167, February 2020);

[0007] Existing technologies have made great progress in the research of theories and methods for credibility assessment of traditional simulation models, and can effectively assess the reliability and accuracy of simulation models. In the construction and optimization of digital twin systems, simulation technology is still the key support, and twin models can actually be regarded as simulations of real systems. However, digital twin systems are very different from traditional simulations. Twin models are high-fidelity restorations of physical systems. Therefore, in order to achieve higher accuracy, models need to be more refined, which places higher demands on simulation technology. In addition, digital twin systems require real-time interaction between virtual models and physical systems, which makes the credibility assessment of digital twin models more complicated than traditional simulations. Credibility assessment not only involves more data processing, but also needs to consider more indicator dimensions to ensure the accuracy of the overall system credibility assessment. Summary of the invention

[0008] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a method and system for evaluating the credibility of a digital twin model. The purpose of implementing the credibility evaluation of a communication twin model is to ensure that the virtual model can accurately reflect the performance, behavior, and change trend of real communications, thereby improving the overall performance and reliability of the communication network. In order to achieve the above purpose / solve the above technical problems, the present invention is implemented by adopting the following technical solutions: A first aspect: A method for evaluating the credibility of a digital twin model, the method comprising: Construct an indicator system framework for the credibility evaluation of digital twin models, and determine the corresponding evaluation indicators for model credibility evaluation based on the indicator system framework according to the characteristics and target requirements of digital twins; Obtain corresponding credibility analysis raw data according to corresponding evaluation indicators of model credibility evaluation; Based on the credibility analysis, the original data is analyzed and processed by combining qualitative analysis methods and quantitative analysis methods, and a reasonable credibility level evaluation interval is constructed according to different evaluation indicators; Determine the corresponding basic probability distribution of evidence through the cloud model according to the credibility level evaluation interval; The basic probability distribution of the evidence obtained is combined with Deng entropy according to the evidence conflict and the credibility of the evidence source to obtain the corresponding discount factor of the evidence; Discounting the evidence by using the discount factor to generate evidence for fusion; Through the fusion rules of evidence theory, the evidence used for fusion is fused to obtain the basic probability distribution after the fusion of evidence; Construct a judgment matrix and calculate the weights of the corresponding evaluation indicators for model credibility assessment; Through the weights between the evaluation indicators, the basic probability distribution obtained by each evaluation indicator is fused again, and the credibility evaluation level of the digital twin model is evaluated based on the basic probability distribution after the fusion of evidence.

[0009] Optionally, the qualitative analysis method is: conducting qualitative analysis in combination with demand and expert experience.

[0010] Optionally, the formula used in the quantitative analysis method is: ; Where: represents the cumulative distribution of the model simulation, represents the empirical cumulative distribution of experimental data, represents the error measure between two distributions, is the cumulative distribution function of the simulation output, is the empirical cumulative distribution function of the experimental observation data.

[0011] Optionally, the digital features of the cloud model are represents the expected value, represents entropy, It represents the super entropy, which reflects the discrete degree of entropy and the expected value ,entropy The calculation is as follows: , ; Where: Indicates the maximum boundary value in different level intervals. It represents the minimum boundary value in different level intervals. The membership of the credibility evaluation level is calculated as follows: ; Since the Gaussian cloud model is universal, the normal cloud model evolved from the Gaussian distribution is used, where: , Represents a random number, represents the exponential function e, express expectations, A normal random number representing the variance, Represents a normal distribution.

[0012] Optionally, the basic probability distribution of the obtained evidence is combined with Deng entropy according to the evidence conflict and the credibility of the evidence source to obtain the corresponding discount factor of the evidence, including: The conflict between evidences is calculated by Jousselme distance. The formula for evidence distance conflict is as follows: ; ; Among them, D is a The matrix of It is the number of propositions in the identification indicator system framework; represents the Jousselme distance between evidence m1 and m2, , is the basic probability distribution of the two bodies of evidence, , is the vector of basic probability assignments; D(s1, s2) represents the elements of the matrix D, , is a non-empty subset in the recognition frame, Represents the transpose of a matrix; The expert credibility is defined as the ratio of the expert's weight value to the maximum weight among experts according to the expert's experience level. The relative weight of each expert is , where the sum of each expert weight is 1, then the credibility of the evidence is: ; Among them, Crd represents the credibility of evidence, represents the relative weight of the experts, It represents the largest relative weight of experts, i=1 to n, which means there are n experts with corresponding relative weights. Indicates n One of them; The Shapely value is used to weigh the distance conflict between evidence and the credibility of evidence to obtain the modified weight. The Shapely value calculation formula is as follows: ; In the formula, Represents Shapely value, represents all subsets that do not contain player i, represents the size of the subset S, Represents the total number of players, represents the payoff of subset S, Indicates the payoff of player i in subset S. represents factorial; Calculate the Deng entropy of each piece of evidence using the following formula: ; In the formula, represents the Deng entropy, Is a collection The base number, Express Basic support level; By combining the modified weight and Deng entropy, the final evidence discount value w1 is obtained.

[0013] Optionally, the weighting of the distance conflict between the evidence and the credibility of the evidence by using the Shapely value includes: (1) Treat the evidence as a cooperative or adversarial player. When the conflict between two pieces of evidence is small, the evidence is considered to be cooperative. This generates a positive gain, which enhances the impact of the two pieces of evidence on the overall evidence fusion. (2) When there is a large conflict of evidence, the credibility gap between the evidence is large, which will enhance the influence of the evidence with high credibility on the overall integration, while reducing the influence of the evidence with low credibility; (3) When the evidence conflict is large, but the credibility gap between the two pieces of evidence is small, the evidence is considered to be confrontational. At this time, the fusion of the two pieces of evidence will have a greater impact on the final result, while reducing the impact between the two pieces of evidence; Set the parameters according to steps (1) to (3), calculate the evidence conflict and the credibility of the evidence source, and calculate the correction weight corresponding to each piece of evidence through the Shapely value.

[0014] Optionally, the discounting of the evidence by using a discount factor includes: The conflict between evidences is calculated by using the Josselme distance, the evidence conflict matrix is ​​obtained, the evidence similarity matrix is ​​obtained, and the support is obtained; The mean of the support is calculated and combined with Deng entropy to obtain the evidence discount value w2. When the evidence discount value w1 is greater than or equal to the evidence discount value w2, it is considered that there is a good consistency between the evidence conflict and the evidence credibility, and the discount factor is taken. , when the evidence discount value w1 is less than the evidence discount value w2, it is considered that there is no good consistency between the evidence conflict and the evidence credibility, and the discount factor is ; .

[0015] Optionally, the evidence is discounted by a discount factor to generate a formula for fusion evidence: ; In the formula, Express The basic level of support represents the total number of evidence bodies, is the evidence after the discount factor has discounted the evidence, As the conflict factor, for The average basic trust distribution over all evidence is given by: , .

[0016] Optionally, the evidence for fusion is fused by using a fusion rule of evidence theory to obtain a basic probability distribution after the fusion of evidence, including: The fusion is performed K-1 times through the fusion rule of evidence theory to obtain the final fusion result and the credibility evaluation level. , Two sets of basic probability distribution functions for the same recognition framework, based on Dempster's combination rule: , ; in, , is the basic probability distribution of the two bodies of evidence, is the conflict coefficient, Representation Proposition and The intersection between and Represents different sets of beliefs.

[0017] A second aspect: A digital twin model credibility assessment system, the system comprising: The evaluation index determination model is used to build an index system framework for the credibility evaluation of the digital twin model. Based on the index system framework, the corresponding evaluation indexes for the credibility evaluation of the model are determined according to the characteristics and target requirements of the digital twin. A data analysis module is used to obtain corresponding credibility analysis raw data according to the corresponding evaluation indicators of the model credibility evaluation; The grade analysis module is used to analyze and process the original credibility analysis data in combination with qualitative analysis methods and quantitative analysis methods, and to construct a reasonable credibility grade evaluation interval based on different evaluation indicators; A cloud processing module is used to determine the corresponding basic probability distribution of evidence through a cloud model according to the credibility level evaluation interval; The first calculation module is used to distribute the basic probability of the obtained evidence, and obtain the corresponding discount factor of the evidence according to the evidence conflict and the credibility of the evidence source combined with Deng entropy; A generation module, used to discount the evidence by using a discount factor to generate evidence for fusion; The evidence fusion module is used to fuse the evidence used for fusion through the fusion rules of evidence theory to obtain the basic probability distribution after the fusion of evidence; The second calculation module is used to construct a judgment matrix and calculate the weights of the corresponding evaluation indicators of the model credibility evaluation; The evaluation calculation module is used to fuse the basic probability distribution obtained by each evaluation indicator again through the weights between the evaluation indicators, and evaluate the credibility evaluation level of the digital twin model based on the basic probability distribution after the fusion of evidence.

[0018] Compared with the prior art, the present invention has the following beneficial effects: The method of the present invention firstly constructs an indicator system framework for credibility evaluation of digital twin models, and obtains the information required for corresponding credibility evaluation according to the credibility evaluation indicators of digital twin models; secondly, it converts the credibility evaluation level into credibility evaluation level in combination with demand analysis and expert experience, and converts the obtained credibility evaluation level into evidence through a cloud model to obtain the support of the credibility evaluation level of each piece of evidence; finally, the conflict between evidences is calculated through the Josselme distance, and the corresponding evidence correction weight is obtained by Shapely for the evidence conflict and the credibility of the evidence source or the importance of evidence, and the evidence discount factor is obtained in combination with Deng entropy, and the final credibility evaluation is obtained by fusing the evidence; The method proposed in this invention can evaluate the credibility of the model's indicators by comprehensively considering various factors, effectively improve the accuracy of the credibility evaluation, and reduce its uncertainty, especially showing good application effects in the communication twin model. Through this method, the credibility of the communication twin model can be evaluated more accurately, thereby ensuring its effective application in complex communication environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 Flow chart of the digital twin model credibility assessment method of the present invention; Figure 2 It is a schematic diagram of the indicator system framework of the digital twin model credibility assessment method of the present invention; Figure 3 It is a discount factor judgment flow chart of the digital twin model credibility assessment method of the present invention; Figure 4 It is a comprehensive credibility assessment flow chart of the credibility assessment of the digital twin model of the present invention; Figure 5 It is the basic probability distribution comparison for the indicator credibility evaluation of the present invention. DETAILED DESCRIPTION

[0020] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods.

[0021] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and the like are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, features defined as "first", "second", and the like may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more.

[0022] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood by specific circumstances.

[0023] Embodiment 1, as Figure 1 As shown, the present invention discloses a method for evaluating the credibility of a digital twin model, the method comprising: Step 1: Construct an indicator system framework for the credibility evaluation of the digital twin model, and determine the corresponding evaluation indicators for the model credibility evaluation based on the indicator system framework according to the characteristics and target requirements of the digital twin; Step 2, obtaining corresponding credibility analysis raw data according to the corresponding evaluation index of the model credibility evaluation; Step 3: Based on the credibility analysis, the original data is analyzed and processed in combination with qualitative analysis methods and quantitative analysis methods, and a reasonable credibility level evaluation interval is constructed according to different evaluation indicators; Step 4: Determine the corresponding basic probability distribution of evidence through the cloud model according to the credibility level evaluation interval; Step 5: Assign the basic probability of the evidence obtained, according to the evidence conflict and the credibility of the evidence source, combined with Deng entropy, to obtain the corresponding discount factor of the evidence; Step 6, discount the evidence by using the discount factor to generate evidence for fusion; Step 7, using the fusion rule of evidence theory, the evidence to be fused is fused to obtain the basic probability distribution after the fusion of evidence; Step 8: construct a judgment matrix and calculate the weights of the corresponding evaluation indicators for model credibility evaluation; Step 9: Through the weights between the evaluation indicators, the basic probability distribution obtained by each evaluation indicator is fused again, and the credibility evaluation level of the digital twin model is evaluated based on the basic probability distribution after the fusion of evidence.

[0024] In the specific implementation process of this embodiment, step 1 is further explained. The indicator framework for the credibility evaluation of the digital twin model is as follows: Figure 2 As shown in the figure, verification, validation and accreditation (VV&A) of the simulation model is an important method to test whether the model is credible. Combined with VV&A, a comprehensive digital twin model credibility evaluation index system can be constructed according to the application purpose and requirements of the twin system. This system covers the entire process of the system from the initial concept design to the operation of the model. In the conceptual design stage, it is first necessary to ensure that the design goals and requirements of the digital twin model are clear. At the same time, appropriate parameters are selected according to the design requirements to ensure that the model can theoretically achieve the expected functions. In the design and implementation stage, it is necessary to verify whether the model meets the design specifications and requirements, and to ensure the correctness of each functional module of the model through unit testing and integration testing. In the evaluation of the system operation process, it is necessary to evaluate the stability, robustness and flexibility of the model under different tasks. And finally, the simulation results are compared and analyzed to ensure that the model output meets the expected goals and is consistent with the actual data. Through the construction of these evaluation indicators, a comprehensive and accurate credibility evaluation can be achieved.

[0025] In the specific implementation process of this embodiment, the step 3 is further explained, and the qualitative analysis method is: qualitative analysis is performed in combination with demand and expert experience. Due to the diversity of indicators and complex hierarchical structures, different evaluation methods need to be used for different evaluation indicators. The evaluation method mainly includes qualitative evaluation and quantitative evaluation, and the two are usually combined for analysis. Qualitative evaluation is mainly obtained through expert evaluation. Quantitative evaluation is mainly error analysis, which quantifies the difference between simulation data and actual data for consistency analysis. Since actual data is difficult to obtain and belongs to small sample data, and simulation data is easy to obtain, the cumulative distribution function (CDF) between the two data is calculated, and then the error is measured. The area metric is used to measure the difference between the simulation model and the experimental observation data. The calculation formula is as follows: ; In the formula, represents the cumulative distribution of the model simulation, represents the empirical cumulative distribution of experimental data, represents the error measure between two distributions, is the cumulative distribution function of the simulation output, The area metric takes into account the differences in the entire probability distribution and can more accurately reflect the consistency between the model prediction and the experimental data. The combination of qualitative and quantitative methods can comprehensively evaluate the credibility of the simulation model.

[0026] Through the combination of qualitative and quantitative methods, the appropriate credibility evaluation level is determined according to the actual target requirements. The credibility evaluation level is converted into a basic probability distribution function through the cloud model, so that the conclusions drawn from different types of data and different evaluation methods can be integrated to obtain a comprehensive and reliable credibility evaluation.

[0027] In the specific implementation process of this embodiment, step 4 is further explained. The cloud model is an uncertainty conversion model between a qualitative concept and its quantitative representation represented by a natural language value. The digital feature adopts ( , , ), respectively represent the expected value, entropy and super entropy. Super entropy reflects the discrete degree of entropy, generally taken as 0.01, , The calculation is as follows: ; ; In the formula Indicates the boundary values ​​in different level intervals. The membership of the credibility evaluation level is calculated as follows: ; Since the Gaussian cloud model is universal, the normal cloud model evolved from the Gaussian distribution is used. , Represents a random number, represents the exponential function e, express expectations, A normal random number representing the variance, Represents a normal distribution.

[0028] In step 5, since the credibility assessment of a single data source is often incomplete, it is necessary to evaluate from multiple perspectives and subjective evaluation requires multiple experts to evaluate, and multiple evaluation results are fused to obtain an accurate and comprehensive credibility assessment. However, in the process of evidence fusion, there may be inconsistencies or contradictions in the evidence from different evidence sources. If there is unreliable evidence, this will reduce the accuracy of the final fusion result. Therefore, the conflict between evidence must be dealt with.

[0029] Common evidence conflict handling methods are usually based on distance measurement between evidence, that is, evaluating the distance between each piece of evidence and other evidence, and then reducing the impact of evidence with large conflicts on the fusion results; however, this method often ignores the importance or credibility of the evidence source itself; in extreme cases, a relatively credible piece of evidence may have a large conflict with other evidence, while the distance between other less credible or less important evidence is smaller; at this time, relying solely on distance measurement will reduce the impact of credible evidence and increase the impact of uncredible evidence, resulting in inaccurate final fusion results.

[0030] Although this situation is not encountered during evidence fusion, due to the different sources of evidence, credibility and emphasis, it is necessary to properly weigh the credibility or importance of the evidence source and the degree of conflict between the evidence when dealing with evidence conflicts to obtain accurate evidence correction weights, and then combine Deng entropy to obtain the final evidence discount factor to discount the evidence, thereby reducing the impact of unreliable evidence and increasing the impact of credible evidence on the fusion results, and obtaining accurate credibility evaluation results.

[0031] In the specific implementation process of this embodiment, the method for calculating the evidence discount factor is as follows: 1) Calculate the conflict between evidences through the Jousselme distance. The evidence distance conflict measurement formula is as follows: ; ; 2) Determine the credibility of the evidence source according to different data conditions, such as sensor accuracy, expert experience level, uncertainty, etc. If the credibility of the evidence source is set according to the expert level, the relative weight of each expert is , where the sum of each expert's weight is 1, and the expert's credibility is defined as the ratio of the expert's weight value to the maximum weight among the experts, then the credibility of the evidence is: ; 3) The Shapely value is used to weigh the distance conflict between the evidence and the credibility of the evidence to obtain the modified weight. The Shapely value calculation formula is as follows: ; In the formula, represents all subsets that do not contain player i, represents the size of the subset S, Represents the total number of players, represents the payoff of subset S, represents the benefit after adding player i to subset S.

[0032] In this embodiment, the specific calculation steps based on Shapely value are as follows: (1) Treat the evidence as a cooperative or adversarial player. When the conflict between two pieces of evidence is small, the evidence is considered to be cooperative. This generates a positive gain, which enhances the impact of the two pieces of evidence on the overall evidence fusion. (2) When there is a large conflict of evidence, the credibility gap between the evidence is large, which will enhance the influence of the evidence with high credibility on the overall integration, while reducing the influence of the evidence with low credibility; (3) When the evidence conflict is large, but the credibility gap between the two pieces of evidence is small, the evidence is considered to be confrontational. In this case, the fusion of the two pieces of evidence will have a greater impact on the final result, while reducing the impact between the two pieces of evidence. According to the above rules, set appropriate parameters, calculate evidence conflicts and the credibility of evidence sources, and calculate the correction weight corresponding to each piece of evidence through Shapely values; 4) Calculate the Deng entropy of each piece of evidence. Deng entropy, as a generalization of Shannon entropy, is an effective way to measure uncertainty. The larger the cardinality in the evidence, the larger the Deng, so that the evidence contains more information. When a piece of evidence has a larger Deng entropy, it can be better supported by other evidence, that is, the evidence plays an important role in the final fusion; ; In the formula, is an assumption in the basic probability distribution function, Is a collection By combining the correction weight and Deng entropy, and then obtaining the final evidence discount value w1, the unreliable evidence is discounted to reduce its impact on the final fusion result, and an accurate system simulation credibility evaluation is obtained.

[0033] like Figure 3 As shown in Figure 1, in step 5, after calculating the evidence discount factor, it is compared with the original value of resolving the evidence conflict to determine whether the evidence needs to be discounted. The conflict between the evidence is calculated by the Josselme distance, the evidence conflict matrix is ​​obtained, the evidence similarity matrix is ​​obtained, the support is obtained, and then the mean of the support is obtained and combined with the Deng entropy to obtain the evidence discount value w2. When the evidence discount value w1 is greater than or equal to the evidence discount value w2, it is considered that there is a good consistency between the evidence conflict and the evidence credibility, and the discount factor is taken. , when the evidence discount value w1 is less than the evidence discount value w2, it is considered that there is no good consistency between the evidence conflict and the evidence credibility, and the discount factor is ; .

[0034] In step 6 and step 7, after discounting the evidence according to the discount factor, the influence of unreliable evidence is reduced, and the influence of reliable evidence on the overall result is strengthened, so as to improve the accuracy of the final fusion result. Multiple pieces of evidence are synthesized through the following fusion rules to obtain the corrected evidence: ; In the formula, Express The basic level of support represents the total number of evidence bodies, is the evidence after the discount factor has discounted the evidence, As the conflict factor, for The average basic trust distribution over all evidence is given by: ; ; Then obtain the final corrected evidence, fuse it K-1 times through the fusion rules of evidence theory, obtain the final fusion result, and get the credibility evaluation level. Assume that m1 and m2 are two sets of basic probability distribution functions of the same recognition framework, based on Dempster combination rule: ; ; Among them, k is the conflict coefficient, A represents the proposition and After obtaining multiple credibility evaluations of each sub-indicator and fusing them to obtain the final evidence, the credibility evaluation levels of the corresponding sub-indicators are obtained through the above method.

[0035] like Figure 4 As shown, in step 8, a judgment matrix is ​​constructed to obtain the weight of each indicator. The weight calculation needs to focus on different indicators according to different needs. After the indicator weight is calculated, each indicator is integrated through the above method to obtain the overall credibility evaluation of the system, as follows: (1) Construct corresponding indicator systems according to different digital twin models, obtain relevant information on credibility evaluation for each indicator system, and obtain corresponding credibility evaluation levels based on target requirements and expert experience; (2) The obtained credibility evaluation information is converted into the corresponding basic probability distribution function through the cloud model method, and each evaluation information generates a set of evidence; (3) The conflict between evidences is calculated by the Jousselme distance. The larger the distance between evidences, the greater the conflict between evidences. (4) Based on the information of the evidence source obtained, and according to the specific circumstances of the data obtained, such as the expert experience level, sensor accuracy, data uncertainty, etc., obtain the credibility of the evidence corresponding to the corresponding evidence; (5) After obtaining the evidence conflict and evidence credibility, the correction weight of the corresponding evidence is calculated using the Shapely value. The accurate evidence correction weight is obtained by considering the conflict between the evidence and other evidence and the credibility of the evidence itself. (6) Deng entropy is an effective method to measure uncertainty. The evidence correction weight is combined with Deng entropy to obtain the final evidence discount factor to discount the evidence. (7) The conflict between evidences is calculated by using the Josselme distance, the evidence conflict matrix is ​​obtained, the evidence similarity matrix is ​​obtained, the support is obtained, and then the evidence weight is obtained by combining Deng entropy. When there is a good consistency between the evidence conflict and the evidence credibility, the evidence is directly obtained without discounting. Otherwise, the evidence is discounted by the discount factor to obtain the discounted evidence; (8) Synthesize the directly obtained evidence and the discounted evidence to obtain a new revised evidence, which is used as the revised evidence in the evidence theory evidence fusion to obtain the final credibility level evaluation; (9) The new evidence is fused K-1 times through evidence theory to obtain the final evidence credibility evaluation. The specific formula is as follows: ; .

[0036] Figure 4 The paper describes the process of evaluating the system credibility by calculating each indicator using the proposed method and synthesizing the credibility. The process includes the following steps: (1) After obtaining the information required for the credibility evaluation of the corresponding indicator, Figure 3 The processes in the above respectively calculate the credibility evaluation of the corresponding indicators, and obtain the corresponding evidence for the corresponding indicators; (2) Construct a judgment matrix and obtain the importance of corresponding indicators according to the target requirements of system credibility evaluation. The judgment matrix and consistency test CI are: A = [ a 11 a 12 ⋯ a 1 n a 21 a 22 ⋯ a 2 n ⋮ ⋮ ⋱ ⋮ a n 1 a n 2 ⋯ a nn ] ; , where the consistency test checks the consistency of the judgment matrix pairwise. is the maximum eigenvalue, When , the judgment matrix is ​​considered acceptable, otherwise, the judgment matrix is ​​modified; (3) Discounting the evidence by calculating the evidence discount factor based on the above method by combining the importance of the evidence and the conflict between the evidence; (4) Obtain the corrected evidence for the final fusion, perform K-1 times of evidence fusion through evidence theory, and obtain the final system credibility evaluation.

[0037] Figure 5 The evidence of using different evidence conflict resolution methods to obtain indicator credibility evaluation is described. The horizontal axis represents the credibility evaluation level, and the vertical axis represents the support for the corresponding evaluation level.

[0038] The meaning of each evaluation level is explained in Table 1. Table 1 Credibility Evaluation Level Table: Rating meaning excellent The model can well represent the behavior of the real system. The error between the simulation results and the actual results is very small, and the interaction between the virtual and the real is very smooth. good The model can better represent the behavior of the real system. The simulation results have a certain deviation from the actual results, and the virtual-real interaction effect is normal. middle The model can express the behavior of the real system to a certain extent, but the simulation results are obviously different from the actual results, and the interaction between virtual and real is unstable. Difference The model cannot represent the behavior of the real system, the simulation results deviate significantly from the actual results, and there are problems with the interaction between the virtual and the real

[0039] First, according to the indicators, several corresponding groups of evidence for credibility evaluation are obtained from multiple angles and aspects. Each piece of evidence can be obtained from different sources, such as expert evaluation, sensor collection data, system output data, etc.

[0040] The obtained indicator evidence is shown in Table 2. Table 2 Basic probability distribution function of a certain indicator: Basic probability distribution excellent good middle Difference 1 0.15 0.69 0.16 0 2 0.71 0.16 0.13 0 3 0.53 0.29 0.18 0

[0041] The three pieces of evidence in Table 2 have different credibility evaluation levels because of their different sources. Since the credibility of different evidence sources is different, when dealing with evidence conflicts in credibility evaluation, it is necessary to consider the credibility of each source of evidence as the importance of evidence. When the credibility of one piece of evidence is much greater than the credibility of other evidence, even if the evidence conflicts greatly with the rest of the evidence, the impact of this piece of evidence on the fusion of evidence should be considered. Therefore, in order to intuitively analyze the credibility of the fused evidence, the credibility of the three sources of evidence is set to 0.8, 0.1, and 0.4.

[0042] The fusion results of different evidence conflict processing methods are shown in Table 3. Table 3 Credibility evaluation results of a certain indicator: Basic probability distribution excellent good middle Difference Method 1 0.4852 0.4662 0.0485 0 Method 2 0.5372 0.4577 0.0051 0 Method 3 0.7036 0.2713 0.0251 0 Method 4 0.7581 0.2156 0.0262 0 The present invention 0.1911 0.7857 0.0232 0

[0043] Since the credibility of the evidence sources of three pieces of evidence is set, and the credibility of evidence 2 is very low, the actual result should not consider the influence of this piece of evidence. From the fusion results in Table 3, it can be seen that the credibility evaluation of the first piece of evidence fusion result cannot be obtained, while the remaining three credibility evaluation results are consistent with evidence 2. However, since the credibility of evidence 2 is very low, the evaluation result cannot correctly reflect the true credibility evaluation. Although the remaining three methods can achieve good results when the credibility of the evidence source is not much different, when conducting credibility evaluation of complex models, the increase in system complexity and uncertainty may lead to inaccurate evaluation results. The credibility evaluation result obtained by the proposed method is good, which can comprehensively consider the influence of each valuable evidence. The credibility evaluation result obtained is very consistent with evidence 1, and the accuracy of the evaluation result is improved compared with the other methods.

[0044] Table 4 describes the credibility evaluation of the corresponding indicators for a simplified system model, and the final system credibility evaluation is obtained by the proposed method. Three systems are set up, and each system has three indicators. The credibility evaluation levels of the three groups of indicators of system 1 are "good", "excellent", and "excellent", respectively; the credibility evaluation levels of the three groups of indicators of system 2 are "medium", "good", and "good", respectively; the credibility evaluation levels of the three groups of indicators of system 3 are "poor", "medium", and "medium", respectively. The importance weights of the three groups of indicators are set to 0.5, 0.3, and 0.2, respectively. According to the proposed method, the basic belief distribution function of the credibility evaluation of each indicator is fused with evidence to obtain the final credibility evaluation result, as shown in Table 4: Table 4 Basic belief distribution function for credibility evaluation of each indicator: This probability distribution excellent good middle Difference System 1 0.2429 0.7530 0.0020 0.0021 System 2 0.0002 0.2903 0.7090 0.0005 System 3 0.0000 0.0008 0.3246 0.6746 As can be seen from Table 4, the final credibility evaluation results of each system can be correctly obtained based on the results of the credibility fusion of each system indicator. It can be concluded that the credibility evaluation level of system 1 is "good", which is higher than the credibility evaluation level of system 2, which is "medium", and the credibility evaluation of system 2 is higher than the credibility evaluation level of system 3, which is "poor". It is consistent with the credibility of each system, accurately obtains the system credibility evaluation results, and can correctly identify systems with different credibility.

[0045] Embodiment 2: The present invention further discloses a digital twin model credibility assessment system, the system comprising: The evaluation index determination model is used to build an index system framework for the credibility evaluation of the digital twin model. Based on the index system framework, the corresponding evaluation indexes for the credibility evaluation of the model are determined according to the characteristics and target requirements of the digital twin. A data analysis module is used to obtain corresponding credibility analysis raw data according to the corresponding evaluation indicators of the model credibility evaluation; The grade analysis module is used to analyze and process the original credibility analysis data in combination with qualitative analysis methods and quantitative analysis methods, and to construct a reasonable credibility grade evaluation interval based on different evaluation indicators; A cloud processing module is used to determine the corresponding basic probability distribution of evidence through a cloud model according to the credibility level evaluation interval; The first calculation module is used to distribute the basic probability of the obtained evidence, and obtain the corresponding discount factor of the evidence according to the evidence conflict and the credibility of the evidence source combined with Deng entropy; A generation module, used to discount the evidence by using a discount factor to generate evidence for fusion; The evidence fusion module is used to fuse the evidence used for fusion through the fusion rules of evidence theory to obtain the basic probability distribution after the fusion of evidence; The second calculation module is used to construct a judgment matrix and calculate the weights of the corresponding evaluation indicators of the model credibility evaluation; The evaluation calculation module is used to fuse the basic probability distribution obtained by each evaluation indicator again through the weights between the evaluation indicators, and evaluate the credibility evaluation level of the digital twin model based on the basic probability distribution after the fusion of evidence.

[0046] In summary, from the perspective of the credibility assessment of digital twin models, the advantages and disadvantages of several system simulation credibility assessment methods that exist in today's era are analyzed. On this basis, a credibility assessment method for digital twin models based on evidence theory is proposed, which solves some defects in the current model credibility assessment methods to a certain extent, improves the accuracy of credibility assessment, and reduces the uncertainty of credibility assessment. Therefore, the credibility assessment method for digital twin models based on evidence theory proposed in the present invention is feasible and effective.

[0047] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A digital twin model credibility assessment method, characterized in that: The method comprises: Construct an indicator system framework for the credibility evaluation of digital twin models, and determine the corresponding evaluation indicators for model credibility evaluation based on the indicator system framework according to the characteristics and target requirements of digital twins; Obtain corresponding credibility analysis raw data according to corresponding evaluation indicators of model credibility evaluation; Based on the credibility analysis, the original data is analyzed and processed by combining qualitative analysis methods and quantitative analysis methods, and a reasonable credibility level evaluation interval is constructed according to different evaluation indicators; Determine the corresponding basic probability distribution of evidence through the cloud model according to the credibility level evaluation interval; The basic probability distribution of the evidence obtained is combined with Deng entropy according to the evidence conflict and the credibility of the evidence source to obtain the corresponding discount factor of the evidence; Discounting the evidence by using the discount factor to generate evidence for fusion; Through the fusion rules of evidence theory, the evidence used for fusion is fused to obtain the basic probability distribution after the fusion of evidence; Construct a judgment matrix and calculate the weights of the corresponding evaluation indicators for model credibility assessment; Through the weights between the evaluation indicators, the basic probability distribution obtained by each evaluation indicator is fused again, and the credibility evaluation level of the digital twin model is evaluated based on the basic probability distribution after the fusion of evidence.

2. The digital twin model credibility assessment method according to claim 1, characterized in that: The qualitative analysis method is: conducting qualitative analysis in combination with demand and expert experience.

3. The digital twin model credibility assessment method according to claim 1, characterized in that: The formula used in the quantitative analysis method is: ； Where: represents the cumulative distribution of the model simulation, represents the empirical cumulative distribution of experimental data, represents the error measure between two distributions, is the cumulative distribution function of the simulation output, is the empirical cumulative distribution function of the experimental observation data.

4. The digital twin model credibility assessment method according to claim 1, characterized in that: The digital features of the cloud model are represents the expected value, represents entropy, It represents the super entropy, which reflects the discrete degree of entropy and the expected value ,entropy The calculation is as follows: ， ； Where: Indicates the maximum boundary value in different level intervals. It represents the minimum boundary value in different level intervals. The membership of the credibility evaluation level is calculated as follows: ； Since the Gaussian cloud model is universal, the normal cloud model evolved from the Gaussian distribution is used, where: , Represents a random number, represents the exponential function e, express expectations, A normal random number representing the variance, Represents a normal distribution.

5. The digital twin model credibility assessment method according to claim 1, characterized in that: The basic probability distribution of the evidence is obtained, according to the evidence conflict and the credibility of the evidence source, combined with Deng entropy, to obtain the corresponding discount factor of the evidence, including: The conflict between evidences is calculated by Jousselme distance. The formula for evidence distance conflict is as follows: ， ； in, For one The matrix of It is the number of propositions in the identification indicator system framework; Show evidence and The joussselme distance between , is the basic probability distribution of the two bodies of evidence, , A vector of assignments to base probabilities; ( , ) represents the matrix Elements of , is a non-empty subset in the recognition frame, Represents the transpose of a matrix; The expert credibility is defined as the ratio of the expert's weight value to the maximum weight among experts according to the expert's experience level. The relative weight of each expert is , where the sum of each expert weight is 1, then the credibility of the evidence is: ； Among them, Crd represents the credibility of evidence, represents the relative weight of the experts, Represents the largest relative weight of experts, i=1 to n, indicating that there are n experts with corresponding relative weights of n experts. Indicates n One of them; The Shapely value is used to weigh the distance conflict between evidence and the credibility of evidence to obtain the modified weight. The Shapely value calculation formula is as follows: ； In the formula, Represents Shapely value, represents all subsets that do not contain player i, represents the size of the subset S, Represents the total number of players, represents the payoff of subset S, Indicates the payoff of player i in subset S. represents factorial; Calculate the Deng entropy of each piece of evidence using the following formula: ； In the formula, represents the Deng entropy, Is a collection The base number, Express Basic support level; By combining the modified weight and Deng entropy, the final evidence discount value w1 is obtained.

6. The digital twin model credibility assessment method according to claim 5, characterized in that: The Shapely value is used to weigh the distance conflict between evidences and the credibility of evidences and assign weights, including: (1) Treat the evidence as a cooperative or confrontational player. When the conflict between two pieces of evidence is small, the evidence is considered to be cooperative. This generates a positive gain, which enhances the impact of the two pieces of evidence on the overall evidence fusion. (2) When there is a large conflict of evidence, the credibility gap between the evidence is large, which will enhance the influence of the evidence with high credibility on the overall integration, while reducing the influence of the evidence with low credibility; (3) When the evidence conflict is large, but the credibility gap between the two pieces of evidence is small, the evidence is considered to be confrontational. At this time, the fusion of the two pieces of evidence will have a greater impact on the final result, while reducing the impact between the two pieces of evidence; Set the parameters according to steps (1) to (3), calculate the evidence conflict and the credibility of the evidence source, and calculate the correction weight corresponding to each piece of evidence through the Shapely value.

7. The digital twin model credibility assessment method according to claim 6, characterized in that: The discounting of evidence by using a discount factor includes: The conflict between evidences is calculated by using the Josselme distance, the evidence conflict matrix is ​​obtained, the evidence similarity matrix is ​​obtained, and the support is obtained; The mean of the support is calculated and combined with Deng entropy to obtain the evidence discount value w2. When the evidence discount value w1 is greater than or equal to the evidence discount value w2, it is considered that there is a good consistency between the evidence conflict and the evidence credibility, and the discount factor is taken. , when the evidence discount value w1 is less than the evidence discount value w2, it is considered that there is no good consistency between the evidence conflict and the evidence credibility, and the discount factor is ; 。 8. The digital twin model credibility assessment method according to claim 7, characterized in that: The formula for generating evidence for fusion by discounting the evidence through the discount factor is: ； In the formula, Express The basic level of support represents the total number of evidence bodies, is the evidence after the discount factor has discounted the evidence, As the conflict factor, for The average basic trust distribution over all evidence is given by: ， 。 9. The digital twin model credibility assessment method according to claim 8, characterized in that: The fusion rule of the evidence theory is used to fuse the evidence to obtain the basic probability distribution after the fusion of evidence, including: The fusion is performed K-1 times through the fusion rule of evidence theory to obtain the final fusion result and the credibility evaluation level. , Two sets of basic probability distribution functions for the same recognition framework, based on Dempster's combination rule: ， ； in, , is the basic probability distribution of the two bodies of evidence, is the conflict coefficient, Representation Proposition and The intersection between and Represents different sets of beliefs.

10. A digital twin model credibility assessment system, characterized in that: The system comprises: The evaluation index determination model is used to build an index system framework for the credibility evaluation of the digital twin model. Based on the index system framework, the corresponding evaluation indexes for the credibility evaluation of the model are determined according to the characteristics and target requirements of the digital twin. A data analysis module is used to obtain corresponding credibility analysis raw data according to the corresponding evaluation indicators of the model credibility evaluation; The grade analysis module is used to analyze and process the original credibility analysis data in combination with qualitative analysis methods and quantitative analysis methods, and to construct a reasonable credibility grade evaluation interval based on different evaluation indicators; A cloud processing module is used to determine the corresponding basic probability distribution of evidence through a cloud model according to the credibility level evaluation interval; The first calculation module is used to distribute the basic probability of the obtained evidence, and obtain the corresponding discount factor of the evidence according to the evidence conflict and the credibility of the evidence source combined with Deng entropy; A generation module, used to discount the evidence by using a discount factor to generate evidence for fusion; The evidence fusion module is used to fuse the evidence used for fusion through the fusion rules of evidence theory to obtain the basic probability distribution after the fusion of evidence; The second calculation module is used to construct a judgment matrix and calculate the weights of the corresponding evaluation indicators of the model credibility evaluation; The evaluation calculation module is used to fuse the basic probability distribution obtained by each evaluation indicator again through the weights between the evaluation indicators, and evaluate the credibility evaluation level of the digital twin model based on the basic probability distribution after the fusion of evidence.

Images