A simulation data association and reproduction method for typical working conditions of a diesel engine

Abstract

The application discloses a simulation data correlation and reproduction method for typical working conditions of diesel engines and relates to the technical field of computer simulation data processing, and comprises the following steps: standardizing simulation basic information and generating a standardized simulation element set; complementing historical correlation gaps and generating a reliable simulation correlation chain; processing through an improved Perceiver IO network and generating sensitive and historical environment representation; restoring simulation conditions and re-solving to generate simulation reproduction results; comparing the reproduction results with target results to generate reproduction deviation representation; performing reproduction deviation step-by-step attribution to generate a candidate biasing element set; performing historical environment closed-loop rechecking to generate a closed-loop rechecking result; determining reliable environment and biasing elements to generate simulation correlation reproduction results. Through historical correlation complementation, version consistent evolution, sensitive factor routing and historical environment reverse restoration, the application improves the historical simulation condition tracing, result reproduction and deviation attribution capabilities.

Description

Technical Field

[0001] This invention relates to the field of computer simulation data processing technology, and in particular to a method for correlation and reproduction of simulation data for typical diesel engine operating conditions. Background Technology

[0002] With the development of digital design and computer-aided simulation technology for diesel engines, the number of simulation tasks generated under typical operating conditions continues to increase. The simulation process involves multiple types of information, including operating condition information, input parameters, model version, solver configuration, and simulation results. Current simulation management methods mostly use file directories, task numbers, or database records to store simulation data generated at different stages, and rely on manual registration to establish relationships between parameters, models, and results. When reviewing historical simulation results, technicians need to re-search for the corresponding operating conditions, input parameters, model version, and solver configuration, then rebuild the simulation environment and execute the solution. The completeness of the historical record directly affects the reproduction process.

[0003] Existing technologies lack the ability to handle multi-level relationships, version evolution, and missing historical records among simulation elements. When historical relationships are missing, broken, or conflicting, it is difficult to reconstruct a complete historical simulation environment from the target simulation result. When the simulation reproduction results generated by the resolution differ from the target simulation results, existing methods often only compare numerical values, lacking hierarchical correlation analysis based on input parameters, model version, and solver configuration, making it difficult to pinpoint the specific simulation elements causing the deviation. After identifying the deviating factors, there is a lack of a closed-loop mechanism for verification through updating simulation elements, rebuilding the environment, and resolving, affecting the traceability, reproducibility, and accuracy of deviation location of historical simulation results.

[0004] Therefore, how to provide a method for correlation and reproduction of simulation data for typical diesel engine operating conditions is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a simulation data association and reproduction method for typical diesel engine operating conditions. This invention integrates simulation data standardization processing, reliable association completion, improved Perceiver IO network, historical environment reverse restoration, stepwise deviation attribution, and closed-loop verification methods to achieve historical simulation environment restoration, simulation result reproduction, and deviation-causing factor location. It has the advantages of complete association tracing, accurate simulation reproduction, clear deviation location, and reliable closed-loop verification.

[0006] A method for correlation and reproduction of simulation data for typical diesel engine operating conditions according to an embodiment of the present invention includes:

[0007] Collect basic simulation information under typical diesel engine operating conditions, perform standardization processing, and generate a standardized simulation element set;

[0008] Based on a standardized set of simulation elements, historical correlation gaps are identified, reliable gap filling is performed, multi-level correlation relationships are established, and a reliable simulation correlation chain is generated.

[0009] The simulation results to be reproduced are determined from the trusted simulation association chain as the target simulation results. An improved Perceiver IO network is constructed, and a version domain consistent evolution layer, a sensitive factor routing layer, and a historical environment reverse restoration layer are set. The trusted simulation association chain and the target simulation results are input to generate sensitive routing potential representations and historical simulation environment representations.

[0010] Based on the historical simulation environment characterization, the corresponding simulation conditions are restored, the historical simulation environment is reconstructed, and the simulation solution is executed to generate simulation reproduction results;

[0011] The simulation results are correlated and compared with the target simulation results. The differences in the results indicators are analyzed, and a characterization of the reproduction deviation is generated.

[0012] Based on the reproducible deviation characterization, the sensitive route potential characterization, and the credible simulation association chain, the reproducible deviation is attributed stepwise to generate a candidate set of bias-causing factors.

[0013] Based on the candidate biased feature set, perform historical environment closed-loop verification, update the simulation features and perform simulation solution again to generate closed-loop verification results.

[0014] Based on the closed-loop verification results, the credible historical simulation environment and the biased simulation elements are determined, and simulation correlation reproduction results are generated.

[0015] Optionally, the simulation basic information includes typical working condition information, simulation input parameter information, model and version information, solver configuration information, simulation result information, and simulation task association information.

[0016] Optionally, generating the standardized simulation element set includes:

[0017] Acquire typical working condition information, simulation input parameter information, model and version information, solver configuration information, simulation result information and simulation task association information, and collect them according to the simulation task identifier to generate simulation task element groups;

[0018] Perform format conversion, field name unification, and unit of measurement standardization on the simulation task element group to generate a unified format simulation element group;

[0019] Based on a unified format simulation element group, extract the working condition identifier, parameter identifier, model version identifier, solution configuration identifier, and result identifier to establish a simulation element object index;

[0020] The simulation element object indexes are time-aligned according to the execution time corresponding to the simulation task, and associated with a unified format simulation element group to generate a standardized simulation element set.

[0021] Optionally, generating a trusted simulation association chain includes:

[0022] Based on the standardized simulation element set, read the simulation task identifier and simulation element object index corresponding to the completed simulation task, match the association records between typical working conditions, simulation input parameters, model version, solution configuration and simulation results, and generate a historical association relationship set;

[0023] Perform a completeness check on the historical association set to identify missing, broken, and conflicting associations, and generate a historical association gap set;

[0024] Based on the standardized simulation element set, retrieve the simulation elements of the same task corresponding to the historical associated gap set, combine the time sequence, version connection and result correspondence to generate candidate completion relationships, calculate the credibility of candidate completion, and generate a set of credible completion relationships;

[0025] Write the trusted completion relation set into the historical relation set, and reorganize the relation paths according to the hierarchical relationship of working conditions, parameters, model version, solution configuration and simulation results to generate a trusted simulation relation chain.

[0026] Optionally, the generation of sensitive route potential representations and historical simulation environment representations includes:

[0027] Determine the simulation result to be reproduced from the trusted simulation association chain, and mark the simulation result to be reproduced as the target simulation result;

[0028] An improved Perceiver IO network is constructed, which includes a version-domain consistent evolution layer, a sensitivity factor routing layer, and a historical environment reverse restoration layer.

[0029] The trusted simulation association chains are encoded according to the simulation element category and association level, and then mapped to the latent space through cross-attention to generate the simulation latent representation.

[0030] The simulation latent representation is input into the version domain consistent evolution layer, which divides the operating condition domain, parameter domain, model domain, configuration domain and result domain. The latent variable evolution is performed in combination with the version inheritance relationship and branch relationship to generate the version consistent latent representation.

[0031] Input the version-consistent potential representation into the sensitive factor routing layer, calculate the sensitive factor based on the response relationship between the changes in simulation elements and the changes in result indicators under typical working conditions, adjust the propagation path and propagation weight of latent variables, and generate the sensitive routing potential representation.

[0032] The potential representations of sensitive routes and the target simulation results are input into the historical environment reverse restoration layer. Based on the target simulation results, the potential representation reverse query is performed to extract the associated simulation condition representations and generate the historical simulation environment representations.

[0033] The improved Perceiver IO network is trained, historical reliable simulation correlation data and historical environment and simulation result annotations are obtained, and training set and validation set are divided. The training set is input into the improved Perceiver IO network to train the parameters of the version domain consistency evolution layer, the sensitive factor routing layer and the historical environment reverse restoration layer. The parameters are updated based on the environment restoration error, the sensitive bias, and the version consistency error. The validation loss is obtained using the validation set. Training is terminated when the validation loss is lower than the training convergence threshold or the training epoch reaches the upper limit. The model parameters with the lowest validation loss are saved.

[0034] Optionally, generating the simulation reproduction result includes:

[0035] Read the operating conditions, input parameters, model version and solution configuration from the historical simulation environment representation, restore the correspondence between each simulation condition according to the trusted simulation association chain, and generate historical environment restoration items;

[0036] Load the corresponding model version based on the historical environment restoration item, write the input parameters and solution configuration, and complete the simulation environment initialization according to the working conditions to generate the historical simulation environment;

[0037] Establish simulation solution tasks based on historical simulation environments, execute typical diesel engine operating condition simulation calculations according to the solution configuration, record the solution process and result indicators, and generate a reproducible solution record.

[0038] Extract the result indicators corresponding to the target simulation results from the reproduced solution records, associate and organize them according to the result identifiers, and generate the simulation reproduction results.

[0039] Optionally, the generation of the reproducibility deviation characterization includes:

[0040] Read the result indicators and corresponding result values ​​from the simulation reproduction results and the target simulation results, establish the correspondence between the indicators according to the result identifiers, and generate a set of result indicator pairs;

[0041] The numerical differences between corresponding result indicators are calculated based on the result indicator pairing set, and the direction of the difference is determined according to the increase or decrease relationship of the result values ​​to generate the result indicator difference set.

[0042] Based on the statistical analysis of the difference set of the outcome indicators, the magnitude and direction of the differences of each outcome indicator are statistically analyzed to form the change characteristics of the outcome indicators;

[0043] The characteristics of changes in related result indicators, the paired set of result indicators, and the target simulation results are characterized and organized according to the order of result indicators to generate a representation of the reproducibility deviation.

[0044] Optionally, generating the candidate biased feature set includes:

[0045] Read the change characteristics of the result indicators in the reproduced deviation characterization, associate them with the potential characterization of sensitive routes, determine the sensitive simulation elements corresponding to each result indicator, and generate a deviation-sensitive association set;

[0046] Based on the reliable simulation association chain, the input parameters, model version and solution configuration corresponding to the deviation-sensitive association set are traced, and deviation attribution paths are established according to the association level to generate a candidate attribution path set.

[0047] The attribution correlation degree between each associated simulation element and the change characteristics of the result index is calculated along the candidate attribution path set, and the attribution ranking is determined by combining the solution of the sensitivity relationship in the potential representation of sensitive routes, and the ranking result of the biased elements is generated.

[0048] Based on the ranking results of bias-causing factors, we screen the associated simulation factors that meet the attribution conditions, and associate them with the corresponding bias attribution paths and the change characteristics of the result indicators to generate a candidate set of bias-causing factors.

[0049] Optionally, generating the closed-loop verification result includes:

[0050] Read the associated simulation elements, deviation attribution paths, and result index change characteristics in the candidate bias factor set, associate them with historical simulation environment characterization, and generate a closed-loop retest task.

[0051] Based on the closed-loop verification task, the correlation and closure relationship between candidate bias-causing factors and operating conditions, input parameters, model version and solution configuration are checked, and a closure verification result is generated.

[0052] Based on the closure check results, update the simulation conditions corresponding to the candidate bias-causing factors, reconstruct the correction simulation environment, execute the simulation solution, and generate the correction simulation results.

[0053] The calibration simulation results are paired with the target simulation results for result index matching and deviation comparison. The deviation change is calculated by combining the reproduction deviation characterization and the closed-loop verification results are generated.

[0054] Optionally, the generation of simulation-related reproduction results includes:

[0055] Read the correction simulation results and deviation changes in the closed-loop retest results, correlate them with the target simulation results, determine the correction simulation environment that meets the retest judgment conditions, and generate a reliable historical simulation environment;

[0056] Based on the closed-loop retest results, the deviation changes corresponding to the candidate bias-causing factors are extracted, and the associated simulation factors confirmed by the retest are screened according to the deviation reduction relationship to generate bias-causing simulation factors.

[0057] Write back the trusted historical simulation environment and biased simulation elements to the trusted simulation association chain, update the environmental association relationship and bias attribution relationship of the corresponding simulation task, and generate the re-verification association chain.

[0058] The correlation and verification of the correlation chain, the credible historical simulation environment, the biased simulation elements and the target simulation results are linked and the reproduction records are organized according to the simulation task identifier to generate the simulation correlation and reproduction results.

[0059] The beneficial effects of this invention are:

[0060] This invention improves the integrity of historical simulation data association and the accuracy of simulation environment restoration by constructing a process that standardizes simulation elements for typical diesel engine operating conditions, reliably fills historical correlation gaps, reversely restores historical environments, reproduces simulation results, and attributes reproduction deviations step by step. Compared to simulation management methods that rely on manual registration or file directories, this invention generates a reliable simulation association chain based on the multi-level correlation between operating conditions, input parameters, model versions, solver configurations, and simulation results. Furthermore, by improving the version domain consistency evolution layer, solver sensitivity factor routing layer, and historical environment reverse restoration layer in the Perceiver IO network, it jointly processes version relationships, result response sensitivity relationships, and target simulation results, thereby improving the ability to trace back historical simulation conditions and the consistency of simulation reproduction.

[0061] This invention pairs simulation results with target simulation results and characterizes deviations. It performs hierarchical attribution along the associated path of input parameters, model version, and solution configuration to screen candidate bias factors. It then combines historical environment closed-loop verification to update simulation elements, reconstruct and correct the simulation environment, and re-execute the simulation solution to verify and confirm the candidate bias factors. This forms a processing chain of historical tracing, environment restoration, result reproduction, deviation location, and closed-loop verification, which improves the accuracy of bias factor location, the credibility of simulation result reproduction, and the efficiency of historical simulation task verification. It has the advantages of complete association tracing, accurate environment restoration, clear deviation location, and reliable closed-loop verification. Attached Figure Description

[0062] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0063] Figure 1 This is a flowchart of a simulation data association and reproduction method for typical diesel engine operating conditions proposed in this invention;

[0064] Figure 2 This is a schematic diagram of the improved Perceiver IO network structure in a simulation data association and reproduction method for typical diesel engine operating conditions proposed in this invention.

[0065] Figure 3 This is a flowchart of the step-by-step attribution of reproduction deviations and closed-loop verification of historical environment in a simulation data association and reproduction method for typical diesel engine operating conditions proposed in this invention. Detailed Implementation

[0066] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0067] refer to Figure 1 , Figure 2 and Figure 3 A method for correlation and reproduction of simulation data for typical diesel engine operating conditions, including:

[0068] Collect basic simulation information under typical diesel engine operating conditions, perform standardization processing, and generate a standardized simulation element set;

[0069] Based on a standardized set of simulation elements, historical correlation gaps are identified, reliable gap filling is performed, multi-level correlation relationships are established, and a reliable simulation correlation chain is generated.

[0070] The simulation results to be reproduced are determined from the trusted simulation association chain as the target simulation results. An improved Perceiver IO network is constructed, and a version domain consistent evolution layer, a sensitive factor routing layer, and a historical environment reverse restoration layer are set. The trusted simulation association chain and the target simulation results are input to generate sensitive routing potential representations and historical simulation environment representations.

[0071] Based on the historical simulation environment characterization, the corresponding simulation conditions are restored, the historical simulation environment is reconstructed, and the simulation solution is executed to generate simulation reproduction results;

[0072] The simulation results are correlated and compared with the target simulation results. The differences in the results indicators are analyzed, and a characterization of the reproduction deviation is generated.

[0073] Based on the reproducible deviation characterization, the sensitive route potential characterization, and the credible simulation association chain, the reproducible deviation is attributed stepwise to generate a candidate set of bias-causing factors.

[0074] Based on the candidate biased feature set, perform historical environment closed-loop verification, update the simulation features and perform simulation solution again to generate closed-loop verification results.

[0075] Based on the closed-loop verification results, the credible historical simulation environment and the biased simulation elements are determined, and simulation correlation reproduction results are generated.

[0076] In this embodiment, the simulation basic information includes typical working condition information, simulation input parameter information, model and version information, solver configuration information, simulation result information, and simulation task association information, wherein:

[0077] Typical operating condition information records the operating condition identifier, operating status category, boundary condition identifier, and simulation execution time corresponding to the diesel engine simulation task, which is used to distinguish the simulation tasks corresponding to different typical operating conditions.

[0078] The simulation input parameter information records the parameter identifier, parameter name, parameter value, unit of measurement, and parameter version identifier of the input parameters in the simulation task, and establishes a correspondence between the simulation task identifier and typical working condition information;

[0079] The model and version information records the model file identifier, model version identifier, version generation time, and inheritance and branching relationships between model versions called by the simulation task. A correspondence is established between the simulation task identifier and the simulation input parameter information.

[0080] The solution configuration information records the solution configuration identifier, solution method, calculation step size, iteration settings, and convergence judgment settings used in the simulation task, and establishes a correspondence between the simulation task identifier and the model and version information;

[0081] Simulation result information records the result identifier, result index name, result index value and result generation time of the simulation task output. A correspondence is established between the simulation task identifier and typical working condition information, simulation input parameter information, model and version information and solution configuration information.

[0082] The simulation task association information record records the correspondence between the simulation task identifier and the operating condition identifier, parameter identifier, model version identifier, solution configuration identifier, and result identifier, forming the basic association record corresponding to a single diesel engine typical operating condition simulation task.

[0083] In this embodiment, generating the standardized simulation element set includes:

[0084] Acquire typical operating condition information, simulation input parameter information, model and version information, solver configuration information, simulation result information, and simulation task association information, and aggregate them according to the simulation task identifier to generate a simulation task element group, wherein:

[0085] The simulation task identifier is a unique identification code generated by the simulation management platform when creating a typical diesel engine operating condition simulation task. It is written into the corresponding record along with typical operating condition information, simulation input parameter information, model and version information, solution configuration information and simulation result information, and is used to mark the same simulation task to which various types of basic simulation information belong.

[0086] Perform format conversion, field name unification, and unit of measurement standardization on the simulation task element group to generate a unified format simulation element group;

[0087] Based on a unified format simulation element group, the working condition identifier, parameter identifier, model version identifier, solver configuration identifier, and result identifier are extracted to establish a simulation element object index, wherein:

[0088] Establish an index for simulation element objects, specifically as follows:

[0089] Using the simulation task identifier as the primary index key, the working condition identifier, parameter identifier, model version identifier, solution configuration identifier, and result identifier corresponding to the same simulation task are written into the index record corresponding to the primary index key;

[0090] A secondary index item is established based on the working condition identifier, parameter identifier, model version identifier, solver configuration identifier, and result identifier. Each secondary index item points to the typical working condition information, simulation input parameter information, model and version information, solver configuration information, and simulation result information in the simulation element group in a unified format.

[0091] The index hierarchy is organized according to the correspondence between the simulation task identifier and each secondary index item, and the affiliation relationship between each secondary index item is recorded to generate a simulation element object index.

[0092] The simulation feature object indexes are time-aligned according to the execution time corresponding to the simulation task, and associated with a unified format simulation feature group to generate a standardized simulation feature set, wherein:

[0093] The standardized simulation element set is as follows:

[0094] This includes the simulation task identifier after completion time alignment, simulation task execution time, working condition identifier and corresponding typical working condition information, parameter identifier and corresponding simulation input parameter information, model version identifier and corresponding model and version information, solver configuration identifier and corresponding solver configuration information, result identifier and corresponding simulation result information, simulation task association information, and simulation element object index;

[0095] Organize the various contents according to the simulation task identifier, associate the typical working condition information, simulation input parameter information, model and version information, solution configuration information and simulation result information corresponding to the same simulation task with the corresponding execution time, and retain the index relationship and task affiliation relationship between various types of information to form a standardized simulation element set.

[0096] In this embodiment, generating a trusted simulation association chain includes:

[0097] Based on the standardized simulation element set, read the simulation task identifier and simulation element object index corresponding to the completed simulation task, match the association records between typical working conditions, simulation input parameters, model version, solution configuration and simulation results, and generate a historical association relationship set;

[0098] Perform an integrity check on the historical association set to identify missing, broken, and conflicting associations, generating a historical association gap set, where:

[0099] Generate a set of historical correlation gaps, specifically as follows:

[0100] Read the corresponding working condition identifier, parameter identifier, model version identifier, solution configuration identifier, and result identifier in the historical association set one by one according to the simulation task identifier, and check the association existence status, association continuity status, and association unique status between each identifier;

[0101] Records lacking corresponding associated objects are marked as missing associations; records lacking continuous association paths between adjacent levels are marked as broken associations; and records with multiple inconsistent associated objects corresponding to the same type of simulation element in the same simulation task are marked as conflicting associations.

[0102] Extract the simulation task identifier, association location, association type, and related simulation element identifiers corresponding to missing associations, broken associations, and conflict associations, and aggregate them according to the simulation task identifiers to generate a historical association gap set;

[0103] Based on the standardized simulation element set, simulated elements of the same task corresponding to the historical associated gap set are retrieved. Candidate completion relationships are generated by combining time sequence, version continuity, and result correspondence. The confidence level of the candidate completions is calculated, and a set of reliable completion relationships is generated, where:

[0104] Generate a reliable set of complete relations, specifically as follows:

[0105] Read the simulation task identifier, associated location, associated type and related simulation element identifier recorded in the historical associated gap set, and retrieve typical working conditions, simulation input parameters, model version, solution configuration and simulation results with the same simulation task identifier from the standardized simulation element set to form a set of simulation elements corresponding to the gap;

[0106] The order of simulation elements is determined by the execution time of the simulation task in the set of simulation elements corresponding to the gap. The version connection status is determined by the inheritance and branching relationship in the model and version information. The result ownership status is determined by the correspondence between the simulation task identifier and the result identifier. Simulation elements that satisfy the gap association position are combined into candidate completion relationships.

[0107] The time sequence consistency, version connection consistency, and result attribution consistency of the candidate completion relationship are judged respectively. The judgment result that satisfies the corresponding relationship is recorded as 1, and the judgment result that does not satisfy the corresponding relationship is recorded as 0. The average value of each judgment result is calculated to obtain the candidate completion confidence.

[0108] Candidate completion relationships corresponding to the same historical correlation gap are sorted according to their completion credibility. The candidate completion relationship with the highest completion credibility is selected and associated with the corresponding simulation task identifier, associated location and related simulation element identifier to generate a set of credible completion relationships.

[0109] The trusted completion relation set is written into the historical relation set. The relation paths are reorganized according to the hierarchical relationship of working conditions, parameters, model version, solver configuration, and simulation results to generate a trusted simulation relation chain, where:

[0110] Generate a reliable simulation association chain, specifically as follows:

[0111] Read the simulation task identifier, associated position and related simulation element identifier in the trusted completion relation set, write the trusted completion relation into the corresponding historical association gap position in the historical association relation set, and form a completed historical association relation;

[0112] Following the order of operating conditions, parameters, model version, solution configuration, and simulation results, hierarchical connections are established for the operating condition identifier, parameter identifier, model version identifier, solution configuration identifier, and result identifier in the completed historical association relationship, and the task affiliation of each hierarchical associated object is limited according to the simulation task identifier.

[0113] The hierarchical connections corresponding to the same simulation task are organized into associated paths according to the execution time of the simulation task. The candidate completion credibility and associated position corresponding to the credibility completion relationship are retained to form a task-level simulation associated path.

[0114] The simulation association paths at each task level are grouped according to the simulation task identifier, and associated with the typical working conditions, simulation input parameters, model and version information, solution configuration information and simulation result information corresponding to the standardized simulation element set to generate a reliable simulation association chain.

[0115] In this embodiment, the generation of potential sensitive route representations and historical simulation environment representations includes:

[0116] The simulation result to be reproduced is determined from the trusted simulation association chain, and the simulation result to be reproduced is marked as the target simulation result, wherein:

[0117] Mark the simulation result to be reproduced as the target simulation result, specifically:

[0118] Read the result identifier, result indicator name, result indicator value and corresponding simulation task identifier from the simulation result information in the trusted simulation association chain, and locate the corresponding simulation result according to the result identifier that needs to be reproduced;

[0119] Read the working condition identifier, parameter identifier, model version identifier and solution configuration identifier under the corresponding simulation task along the trusted simulation association chain, and check whether the association path between the simulation results and the previous simulation elements is complete.

[0120] The simulation result with a complete associated path to be reproduced is identified as the target simulation result, and the result identifier and simulation task identifier corresponding to the target simulation result are retained;

[0121] An improved Perceiver IO network is constructed, comprising a version-domain consistent evolution layer, a sensitivity factor routing layer, and a historical environment reverse recovery layer, wherein:

[0122] The improved Perceiver IO network is constructed as follows:

[0123] The original Perceiver IO network retains the input preprocessing structure, encoder-side cross-attention, latent variable array, latent variable self-attention processing structure, and decoder-side output query cross-attention structure. The encoder-side cross-attention structure is set to receive the subsequently formed simulation-related input representation and write the encoding result into the latent variable array.

[0124] A version-domain consistency evolution layer is added inside the original latent variable self-attention processing structure. The latent variables are divided into the operating condition domain, parameter domain, model domain, configuration domain, and result domain. After the latent variables are updated, the model version inheritance relationship and branch relationship are read, and the version consistency judgment is performed on the cross-domain propagation results to form a version-consistent latent representation.

[0125] After the version domain consistency evolution layer, a solution sensitivity factor routing layer is added. The corresponding records of version consistency potential representation and typical working conditions, simulation element changes and result index changes in the trusted simulation association chain are read. The solution sensitivity factor is calculated, and the propagation weight of latent variables from the parameter domain, model domain and configuration domain to the result domain is adjusted according to the solution sensitivity factor to generate the sensitivity routing potential representation.

[0126] The original decoding end retains the process of performing cross-attention reading on latent variables by outputting queries. The original general output query construction process is replaced by a historical environment reverse query process driven by the target simulation results. A historical environment reverse restoration layer is added before the cross-attention of the output query at the decoding end. The target simulation results and potential representations of sensitive routes are read, and the reverse query vectors corresponding to the working conditions, input parameters, model version and solution configuration are constructed. The corresponding latent variables are read through the cross-attention of the decoding end, and the historical simulation environment representation is output.

[0127] The three newly added structures are connected to the original main paths of input reading, latent variable processing, and decoding output to form an improved Perceiver IO network;

[0128] The reliable simulation association chains are encoded according to the simulation element category and association level, and mapped to the latent space through cross-attention to generate the simulation latent representation, where:

[0129] Generate the simulation latent representation, specifically:

[0130] Read typical working conditions, simulation input parameters, model version, solution configuration and simulation results from the trusted simulation association chain, and extract simulation element content, simulation element identifier and simulation task identifier according to the working condition layer, parameter layer, model layer, configuration layer and result layer to form a hierarchical simulation element sequence;

[0131] For continuous values ​​in the hierarchical simulation element sequence, read the historical values ​​of the corresponding fields in all completed simulation tasks from the standardized simulation element set, and calculate the minimum and maximum values ​​of the historical values ​​of the same field respectively.

[0132] When the historical maximum value and the historical minimum value of a field are not equal, subtract the historical minimum value of the field from the current consecutive value, and then divide by the difference between the historical maximum value and the historical minimum value of the field to obtain the numerical code.

[0133] When the historical maximum value of a field is equal to its historical minimum value, the corresponding consecutive numerical values ​​will be encoded as 0.

[0134] The category field, simulation element identifier, and simulation task identifier are mapped to fixed-dimensional vectors. The numerical code, category field vector, simulation element identifier vector, simulation task identifier vector, simulation element category code, and association hierarchy code are concatenated and linearly mapped to obtain the simulation association input representation.

[0135] The latent variable array is mapped to the query vector through the query matrix, and the simulation associated input representation is mapped to the key vector and value vector through the key matrix and value matrix, respectively.

[0136] Calculate the dot product of each query vector and each key vector, divide the dot product result by the square root of the key vector dimension, and perform exponential normalization on all dot product results corresponding to the same query vector to obtain the cross attention weights corresponding to each simulation element.

[0137] Multiply each cross-attention weight by its corresponding value vector, and sum the weighted value vectors corresponding to the same query vector to obtain the aggregated result of a single latent variable;

[0138] All aggregation results in the latent variable array are arranged according to the latent variable position, and a simulation latent representation is generated through linear mapping and residual connection. The simulation latent representation is then input into the latent variable domain index area of ​​the version domain consistent evolution layer.

[0139] The simulation latent representation is input into the version-domain consistent evolution layer, which divides the domain into operating condition, parameter, model, configuration, and result domains. Latent variable evolution is then performed based on version inheritance and branching relationships to generate a version-consistent latent representation, where:

[0140] Version-domain consistent evolution layer, specifically including:

[0141] Latent variable domain index area, version relationship orchestration channel, branch consistency determination kernel, cross-domain controlled evolution slot and consistency evolution output end;

[0142] Generate version-consistent potential representations, specifically:

[0143] Input the simulation potential representation into the latent variable domain index area, read the working condition identifier, parameter identifier, model version identifier, solution configuration identifier and result identifier corresponding to the simulation potential representation, and write the latent variables into the working condition domain, parameter domain, model domain, configuration domain and result domain respectively according to the simulation element category to generate a domain latent variable set;

[0144] Input the set of domain latent variables into the version relationship orchestration channel, read the model version identifier, version generation time, inheritance relationship and branch relationship from the model and version information in the trusted simulation association chain, establish the front and back order of model versions according to the version generation time, establish vertical evolution connection for model versions with inheritance relationship, establish branch connection for model versions with branch relationship, and associate the parameter domain latent variables and configuration domain latent variables under the corresponding simulation task to generate version evolution relationship;

[0145] Input the version evolution relationship into the branch consistency determination kernel. For any two latent variables to be propagated, read the corresponding model version identifier, simulation task identifier and branch identifier. When the two model versions are in the same inheritance path or the same branch path, record the version consistency determination value as 1. When the two model versions are in different branches and there is no inheritance connection, record the version consistency determination value as 0 and generate the version consistency determination result.

[0146] Input the version consistency determination result and the set of domain latent variables into the cross-domain controlled evolution slot, calculate the dot product correlation value between the latent variables of the working condition domain, parameter domain, model domain, and configuration domain and the latent variables of the result domain, multiply the dot product correlation value by the corresponding version consistency determination value to obtain the cross-domain propagation weight, perform exponential normalization on the propagation weight corresponding to the same target latent variable, multiply the normalized propagation weight by the source domain latent variable and sum them to obtain the cross-domain aggregated latent variable;

[0147] The cross-domain aggregated latent variables are added to the corresponding original latent variables, and after linear mapping, they are input into the consistent evolution output. The output is then rearranged in the order of the operating condition domain, parameter domain, model domain, configuration domain, and result domain to generate a version-consistent latent representation.

[0148] The version-consistent latent representation is input into the sensitive factor routing layer for solving. Based on the response relationship between changes in simulation elements and changes in result indicators under typical operating conditions, the sensitive factors are calculated, and the propagation paths and weights of latent variables are adjusted to generate a sensitive routing latent representation. Where:

[0149] Solving the sensitivity factor routing layer specifically includes:

[0150] The response relationship reading area, the sensitive factor construction kernel, the domain routing adjustment channel, the high-sensitivity propagation holding slot, and the sensitive route output terminal;

[0151] Generate potential representations of sensitive routes, specifically:

[0152] The potential characterization input-response relationship reading area with consistent version is used to read the simulation input parameters, model version, solution configuration and simulation results corresponding to the same typical working condition from the trusted simulation association chain according to the typical working condition identifier. The response pairing record between simulation elements and result indicators is established according to parameter identifier, model version identifier, solution configuration identifier and result identifier.

[0153] The response pairing records are input into the sensitivity factor to construct the kernel. For the input parameters, the parameter values ​​and result index values ​​corresponding to adjacent simulation tasks are read. The absolute value of the difference between parameter values ​​and the absolute value of the difference between result index values ​​are calculated. The absolute value of the difference between result index values ​​is divided by the absolute value of the difference between parameter values ​​to obtain the parameter sensitivity value.

[0154] For the model version and the solution configuration, read the values ​​of the result indicators before and after the version change and the solution configuration change, respectively, calculate the absolute value of the difference between the corresponding result indicator values, and perform proportional normalization under the same typical working condition and within the same result indicator range to obtain the model sensitivity value and the configuration sensitivity value.

[0155] Divide the parameter sensitivity value, model sensitivity value, and configuration sensitivity value corresponding to the same result indicator by the sum of all sensitivity values ​​to obtain the parameter sensitivity factor, model sensitivity factor, and configuration sensitivity factor.

[0156] Input the version consistency latent representation into the domain routing adjustment channel, read the parameter domain latent variables, model domain latent variables, configuration domain latent variables and result domain latent variables respectively, calculate the dot product of each parameter domain latent variable, model domain latent variable and configuration domain latent variable with the corresponding result domain latent variable, divide the dot product result by the square root of the latent variable dimension, and multiply it by the version consistency judgment value corresponding to the version consistency latent representation generated by the version domain consistency evolution layer to obtain the cross-domain correlation value;

[0157] Perform an exponential operation on all cross-domain correlation values ​​pointing to the same result domain latent variable, and divide the exponential result corresponding to a single cross-domain correlation value by the sum of the exponential results of all cross-domain correlation values ​​to obtain the original propagation weights from the parameter domain to the result domain, from the model domain to the result domain, and from the configuration domain to the result domain.

[0158] The original propagation weights corresponding to the parameter domain, model domain, and configuration domain are multiplied by the parameter sensitivity factor, model sensitivity factor, and configuration sensitivity factor, respectively, to obtain the sensitivity moderating propagation weights. All sensitivity moderating propagation weights pointing to the latent variables of the same result domain are then subjected to exponential normalization to obtain the normalized sensitivity propagation weights.

[0159] The normalized sensitive propagation weights are input into the high-sensitivity propagation retention slot. Propagation paths are arranged according to the magnitude of the normalized sensitive propagation weights corresponding to the same result index. High-weight propagation paths are retained, the propagation ratio of low-weight propagation paths is reduced, and the working condition constraint connections between latent variables in the working condition domain and latent variables in the parameter domain, model domain, and configuration domain are retained to form a set of sensitive propagation paths.

[0160] Reduce the propagation proportion of low-weight propagation paths, specifically:

[0161] Calculate the average value of all normalized sensitive propagation weights corresponding to the same result indicator, and use the average value as the propagation retention threshold;

[0162] Propagation paths with normalized sensitive propagation weights greater than or equal to the propagation retention threshold are identified as high-weight propagation paths, while the corresponding normalized sensitive propagation weights remain unchanged.

[0163] Propagation paths with normalized sensitive propagation weights lower than the propagation retention threshold are identified as low-weight propagation paths. The ratio of the normalized sensitive propagation weight to the propagation retention threshold is calculated, and the corresponding normalized sensitive propagation weight is multiplied by the ratio to obtain the attenuated propagation weight.

[0164] The weight reduction of the low-weight propagation path is the original normalized sensitive propagation weight minus the attenuation propagation weight.

[0165] The normalized sensitive propagation weights corresponding to high-weight propagation paths and the decayed propagation weights corresponding to low-weight propagation paths are re-normalized proportionally to obtain the final propagation weights, and a set of sensitive propagation paths is formed according to the final propagation weights.

[0166] Input the set of sensitive propagation paths into the sensitive route output terminal, multiply the final propagation weight corresponding to each propagation path by the corresponding source domain latent variable, and sum the weighted source domain latent variables pointing to the same result domain latent variable to obtain the sensitive route aggregate latent variable;

[0167] The latent variables of sensitive routes are aggregated and added to the latent variables of the corresponding result domains. The latent variables are then reorganized according to the operating condition domain, parameter domain, model domain, configuration domain, and result domain to generate a potential representation of sensitive routes.

[0168] The potential representations of sensitive routes and the target simulation results are input into the historical environment reverse reconstruction layer. Based on the target simulation results, a reverse query of the potential representations is performed to extract the associated simulation condition representations and generate the historical simulation environment representations, where:

[0169] The historical environment reverse restoration layer specifically includes:

[0170] The system includes a target result query area, a reverse latent variable retrieval channel, an environmental element mapping kernel, an association closure check slot, and a historical environment output terminal.

[0171] Generate a historical simulation environment representation, specifically:

[0172] Input the target simulation result into the target result query area, read the result identifier, result indicator name, result indicator value and simulation task identifier corresponding to the target simulation result, map the result identifier, result indicator name and simulation task identifier into fixed dimension vectors respectively, and concatenate them with the result indicator value code to form the target result representation through linear mapping;

[0173] After representing the target results, add operating condition domain identifiers, parameter domain identifiers, model domain identifiers, and configuration domain identifiers respectively to form operating condition query vectors, parameter query vectors, model query vectors, and configuration query vectors.

[0174] The four types of query vectors are input into the reverse latent variable retrieval channel. The sensitive route potential representations are converted into key vectors and value vectors respectively through the key mapping matrix and the value mapping matrix. The dot product of the four types of query vectors with each key vector is calculated. The dot product result is divided by the square root of the key vector dimension. Then, exponential normalization is performed on all results corresponding to the same query vector to obtain the reverse query weight.

[0175] Each reverse query weight is multiplied by its corresponding value vector, and the weighted value vectors corresponding to the same query vector are summed to obtain the working condition candidate representation, parameter candidate representation, model candidate representation, and configuration candidate representation, respectively.

[0176] Four types of candidate representations are input into the environmental element mapping kernel. The working condition identifier, parameter identifier, model version identifier, and solution configuration identifier corresponding to the simulation task to which the target simulation result belongs are read from the trusted simulation association chain. The simulation elements corresponding to each identifier are encoded into element vectors that are consistent with the candidate representation dimensions.

[0177] Calculate the cosine similarity between the candidate representation of the working condition and each working condition element vector, the candidate representation of the parameter and each parameter element vector, the candidate representation of the model and each model element vector, and the candidate representation of the configuration and each configuration element vector. Divide the dot product of the candidate representation and the element vector by the product of their vector magnitudes to obtain the environmental element matching value.

[0178] Within each simulation element category, select the operating condition element, parameter element, model version element, and solution configuration element with the highest environmental element matching value to form a characterization of the associated simulation conditions.

[0179] The associated simulation conditions are input into the associated closed verification slot. The connection relationship between the corresponding working condition identifier, parameter identifier, model version identifier, solver configuration identifier and the result identifier corresponding to the target simulation result in the trusted simulation association chain is read. The association path is checked step by step according to the working condition layer, parameter layer, model layer, configuration layer and result layer.

[0180] For operating condition elements, parameter elements, model version elements, and solution configuration elements with continuous associated paths, retain the corresponding environmental element matching values. For environmental elements that lack continuous associated paths, record the environmental element matching value as 0, and reselect the simulation element with the highest environmental element matching value and continuous associated path from the same type of candidate elements.

[0181] The operating conditions, input parameters, model version, and solution configuration verified by correlation closure are input to the historical environment output terminal. They are then spliced ​​together according to the order of the environment components of operating conditions, input parameters, model version, and solution configuration, and a historical simulation environment representation is formed through linear mapping.

[0182] The improved Perceiver IO network is trained by acquiring historical reliable simulation correlation data and annotations of historical environments and simulation results. A training set and a validation set are then created. The training set is input into the improved Perceiver IO network to train the parameters of the version domain-consistent evolution layer, the sensitivity factor routing layer, and the historical environment reverse restoration layer. The parameters are updated based on the environment restoration error, sensitivity bias, and version consistency error. The validation loss is obtained using the validation set. Training is terminated when the validation loss falls below the training convergence threshold or the training epochs reach their maximum limit. The model parameters at which the validation loss is lowest are saved.

[0183] The improved Perceiver IO network is trained as follows:

[0184] Historical credible simulation association data is extracted from the credible simulation association chain corresponding to the completed diesel engine typical working condition simulation tasks. The working conditions, input parameters, model version, solution configuration and simulation results are read and organized into historical environment labels and simulation result labels. After deduplication according to the simulation task identifier, the training set and the validation set are divided into 80% and 20% respectively.

[0185] The training set is input into the improved Perceiver IO network to train the latent variable domain index area, version relationship arrangement channel, branch consistency decision kernel, cross-domain controlled evolution slot and consistency evolution output of the version domain consistency evolution layer. The sensitive factor construction kernel, domain routing adjustment channel and sensitive routing output of the sensitive factor routing layer are solved, as well as the trainable parameters of the target result query area, reverse latent variable retrieval channel, environmental element mapping kernel and historical environment output of the historical environment reverse restoration layer.

[0186] The predicted historical simulation environment characterization is compared with the historical environment annotation to calculate the environment restoration error;

[0187] The sensitivity bias is calculated by comparing the parameter sensitivity factor, model sensitivity factor, and configuration sensitivity factor with the reference sensitivity factor obtained based on the changes in simulation elements and result indicators.

[0188] Compare cross-domain propagation relationships with version inheritance and branching relationships to calculate version consistency error;

[0189] The training loss is obtained by multiplying the environment restoration error, solution sensitivity bias and version consistency error by 0.5, 0.3 and 0.2 respectively and summing them. The model parameters are then updated using backpropagation, with the initial learning rate set to 0.0001.

[0190] After each training round, the validation set is input into the improved Perceiver IO network to calculate the validation loss. Training is terminated when the validation loss is below 0.005 for 10 consecutive rounds or when the training rounds reach 300. The model parameters corresponding to the round with the lowest validation loss are saved.

[0191] This invention structurally optimizes the traditional Perceiver IO network, forming a potential representation method for the association of simulation data under typical diesel engine operating conditions and the restoration of historical environments. In terms of version relationship modeling, a version-domain consistent evolution layer is introduced. The simulation potential representation is divided into operating condition domain, parameter domain, model domain, configuration domain, and result domain through a latent variable domain index area. A version relationship orchestration channel is used to establish model version inheritance paths and branch paths. Cross-version propagation relationships are identified through a branch consistency decision kernel, and the propagation of latent variables between different simulation element domains is constrained through a cross-domain controlled evolution slot. The consistent evolution output generates a version-consistent potential representation, reducing the interference caused by model version mismatch and cross-branch parameter configuration mixing on historical environment restoration.

[0192] In terms of modeling the influence relationship of simulation elements, a solution sensitivity factor routing layer is introduced. The response records between input parameters, model version, solution configuration and result indicators under typical working conditions are extracted through the response relationship reading area. The parameter sensitivity factor, model sensitivity factor and configuration sensitivity factor are calculated through the sensitivity factor construction kernel. The latent variable propagation weight is adjusted by the domain routing adjustment channel, and the high sensitivity propagation path is retained by the high sensitivity propagation holding slot. The sensitive routing potential representation is generated through the sensitive routing output end to improve the model's ability to identify key biased simulation elements.

[0193] In terms of historical environment restoration, a historical environment reverse restoration layer is introduced. Based on the target simulation results, a reverse query representation is constructed through the target result query area. The working conditions, parameters, models, and configuration candidate representations are extracted through the reverse latent variable retrieval channel. These are then mapped to the corresponding simulation conditions by the environmental element mapping kernel. A correlation closure check slot is used to check the continuous correlation path between environmental elements and the target simulation results. Finally, a historical simulation environment representation is generated through the historical environment output terminal, improving the accuracy and completeness of historical simulation condition reverse tracing. The improved Perceiver IO network enhances version association maintenance, sensitive element identification, and historical environment restoration capabilities, providing a reliable representation foundation for subsequent simulation reproduction, deviation attribution, and closed-loop verification.

[0194] In this embodiment, generating the simulation reproduction result includes:

[0195] The system reads the operating conditions, input parameters, model version, and solver configuration from the historical simulation environment representation, restores the correspondence between each simulation condition according to the trusted simulation association chain, and generates historical environment restoration items, including:

[0196] The historical environment restoration item is generated as follows:

[0197] Read the operating conditions, input parameters, model version and solution configuration from the historical simulation environment representation, and extract the corresponding operating condition identifier, parameter identifier, model version identifier and solution configuration identifier respectively. Bind the four types of identifiers to the simulation task identifier corresponding to the target simulation result.

[0198] Following the trusted simulation association chain in the order of working condition layer, parameter layer, model layer, configuration layer and result layer, the hierarchical connections between working condition identifier and parameter identifier, parameter identifier and model version identifier, model version identifier and solver configuration identifier, and solver configuration identifier and the result identifier corresponding to the target simulation result are retrieved to restore the task affiliation relationship and association order among the four types of simulation conditions.

[0199] The restored operating conditions, input parameters, model version, and solution configuration are associated with the corresponding simulation task identifier, simulation element identifier, and hierarchical connection relationship to form an environmental element restoration record;

[0200] The environmental element recovery records are organized according to the environmental composition order of working conditions, input parameters, model version and solution configuration, and the relationship between the association hierarchy and task affiliation of each environmental element is preserved to generate historical environmental recovery items.

[0201] Based on the historical environment restoration item, the corresponding model version is loaded, input parameters and solver configuration are written, and the simulation environment is initialized according to the operating conditions to generate the historical simulation environment, wherein:

[0202] The historical simulation environment is generated as follows:

[0203] Read the model version identifier in the historical environment restoration item, and locate the corresponding model file from the model and version information in the standardized simulation element set based on the model version identifier, load the corresponding model version, and establish a simulation model running instance;

[0204] Read the parameter identifiers and corresponding input parameters from the historical environment restoration items, and write each input parameter into the parameter position corresponding to the simulation model running instance according to the parameter name, parameter value and unit of measurement to form the parameter configuration result;

[0205] Read the solver configuration identifier and corresponding solver configuration from the historical environment recovery item, and write the solver method, calculation step size, iteration settings and convergence judgment settings into the solver configuration location corresponding to the simulation model running instance to form the solver environment configuration result;

[0206] Read the operating condition identifier and corresponding operating condition from the historical environment recovery item, write the operating condition into the operating condition setting location corresponding to the simulation model running instance, and associate the parameter configuration result and the solution environment configuration result to complete the simulation environment initialization;

[0207] The simulation model running instance, operating conditions, input parameters, model version, and solution configuration that have been initialized are associated with the corresponding simulation task identifier to generate a historical simulation environment;

[0208] A simulation task is established based on the historical simulation environment. Typical diesel engine operating condition simulation calculations are performed according to the solution configuration. The solution process and result indicators are recorded, and a reproducible solution record is generated. Among these:

[0209] Generate a reproducible solution record, specifically as follows:

[0210] Read the simulation model running instances, operating conditions, input parameters, model version and solution configuration from the historical simulation environment, establish a reproduction solution task according to the simulation task identifier corresponding to the target simulation result, and generate a reproduction task identifier for the reproduction solution task;

[0211] Based on the solution method, calculation step size, iteration settings and convergence judgment settings in the solution configuration of the reproduced solution task, the simulation model running instance in the historical simulation environment is called to perform typical diesel engine operating condition simulation calculation under the corresponding working conditions and input parameters.

[0212] During the simulation calculation, the solution time, number of iterations, convergence status and solution status are recorded according to the calculation step size. Each record is associated with the reproduction task identifier and the corresponding solution configuration identifier to form a solution process record.

[0213] After the simulation calculation is completed, the result index name and result index value with the same result identifier category as the target simulation result are read from the output results of the simulation model running instance, and associated with the corresponding working condition identifier, parameter identifier, model version identifier and solution configuration identifier to form a record of the reproduced result index;

[0214] The solution process records and reproduction result index records are collected according to the reproduction task identifier, and associated with the simulation task identifier corresponding to the target simulation result to generate the reproduction solution record;

[0215] Extract the result indicators corresponding to the target simulation results from the reproduced solution records, associate and organize them according to the result identifiers, and generate simulation reproduction results, including:

[0216] The simulation reproduction results are generated as follows:

[0217] Read the reproduction result index record in the reproduction solution record, and read the result identifier, result index name and result index value in the target simulation result. Filter the corresponding result index in the reproduction result index record according to the result identifier.

[0218] The filtered result indicator names are matched one by one with the result indicator names in the target simulation results. The result indicators with the same name and their corresponding result indicator values ​​are retained, and the reproduction task identifier, working condition identifier, parameter identifier, model version identifier and solution configuration identifier are associated.

[0219] The matched result indicators are sorted according to the result identifier. The result indicator name, the value of the reproduced result indicator and the simulation task identifier corresponding to the target simulation result are collected to form a reproduced result entry.

[0220] All reproduction result items are organized according to the reproduction task identifier, and associated with the corresponding working conditions, input parameters, model version and solution configuration to generate simulation reproduction results.

[0221] In this embodiment, the generation of the reproducibility deviation characterization includes:

[0222] Read the result indicators and corresponding result values ​​from the simulation reproduction results and the target simulation results, establish the indicator correspondence according to the result identifier, and generate a result indicator pairing set, where:

[0223] Generate a set of paired result metrics, specifically:

[0224] Read the result identifier, result indicator name, and reproduction result indicator value from the simulation reproduction result, and read the result identifier, result indicator name, and target result indicator value from the target simulation result to form the reproduction result indicator sequence and the target result indicator sequence, respectively.

[0225] The reproduced result indicator sequence and the target result indicator sequence are grouped according to the result identifier, and two groups of result indicators with the same result identifier are assigned to the same indicator matching area.

[0226] Within the same indicator matching area, each result indicator is compared according to its name. A pairing relationship is established between the value of the reproduced result indicator with the value of the target result indicator that has the same result indicator name. The corresponding result identifier, result indicator name, reproduced result indicator value and target result indicator value are recorded to form a result indicator pairing record.

[0227] Records with identical result identifiers but inconsistent result indicator names will be separated and excluded from the result indicator pairing records;

[0228] Arrange all result indicator pairing records according to result identifier and result indicator name, and associate them with the reproduction task identifier and the simulation task identifier corresponding to the target simulation result to generate a result indicator pairing set;

[0229] The numerical differences between corresponding outcome indicators are calculated based on the pairing set of outcome indicators, and the direction of the difference is determined according to the increasing or decreasing relationship of the outcome values, generating a set of outcome indicator differences, where:

[0230] The result index difference set is generated as follows:

[0231] Read the result identifier, result indicator name, reproduced result indicator value and target result indicator value from the result indicator pairing set. For each pair of result indicators, subtract the target result indicator value from the reproduced result indicator value to obtain the result indicator difference.

[0232] The absolute value of the difference between the result indicators is calculated to obtain the magnitude of the difference between the result indicators.

[0233] When the difference in the result index is greater than 0, the direction of the difference is marked as a positive deviation;

[0234] When the difference in the result index is less than 0, the direction of the difference is marked as negative deviation;

[0235] When the difference in the result index is equal to 0, the direction of the difference is marked as unbiased;

[0236] The result identifier, result indicator name, reproduced result indicator value, target result indicator value, result indicator difference, result indicator difference magnitude and difference direction are associated to form a result indicator difference record;

[0237] All result index difference records are arranged according to result identifier and result index name, and associated with the reproduction task identifier and the simulation task identifier corresponding to the target simulation result to generate a result index difference set;

[0238] Based on the statistical analysis of the difference set of outcome indicators, the magnitude and direction of the differences among each outcome indicator are statistically analyzed to form the change characteristics of the outcome indicators, among which:

[0239] The characteristics of the changes in the resulting indicators are as follows:

[0240] Read the result identifier, result indicator name, result indicator difference magnitude and difference direction from the result indicator difference set, and collect the corresponding difference records according to the result indicator name;

[0241] For each outcome indicator, read the value of the corresponding target outcome indicator, divide the difference in the outcome indicator by the absolute value of the target outcome indicator value, and obtain the relative difference in the outcome indicator.

[0242] When the target result indicator is 0, the difference in the result indicator is used as the representation of the difference in the corresponding result indicator.

[0243] The difference direction of positive deviation is encoded as 1, the difference direction of negative deviation is encoded as -1, and the difference direction of no deviation is encoded as 0, thus obtaining the result index direction code;

[0244] Multiply the relative difference magnitude of the result indicator by the directional code of the result indicator to obtain the directional difference value;

[0245] The result identifier, result indicator name, result indicator difference range, result indicator relative difference range, result indicator direction code, and direction difference value are associated to form a record of changes in a single result indicator;

[0246] Organize all individual result indicator change records according to result identifier and result indicator name to form result indicator change characteristics;

[0247] The characteristics of changes in related outcome indicators, the paired set of outcome indicators, and the target simulation results are characterized and organized according to the order of the outcome indicators to generate a representation of the reproducibility deviation, wherein:

[0248] The reproducibility bias characterization is generated as follows:

[0249] Read the result identifier, result indicator name, result indicator difference range, result indicator relative difference range, result indicator direction code, and direction difference value from the result indicator change characteristics, and read the values ​​of the reproduced result indicators and the target result indicators from the result indicator pair set;

[0250] Establish a correspondence between the change characteristics of the result indicators and the pairing set of the result indicators according to the result identifier and the result indicator name. Encode and collect the values ​​of the reproduced result indicators, the values ​​of the target result indicators, the relative difference of the result indicators and the direction of the result indicators corresponding to the same result indicator into single-item deviation information.

[0251] Read the order of the result indicators in the target simulation results, and sort all the individual deviation information according to the order of the result indicators recorded in the target simulation results to form an ordered deviation sequence;

[0252] The values ​​of the reproducible result indicators, the values ​​of the target result indicators, the relative difference magnitude of the result indicators, the direction encoding of the result indicators, and the difference values ​​with direction are all aligned in dimension and then concatenated according to the name of the result indicator to form a single result deviation vector.

[0253] All individual result deviation vectors are combined according to an ordered deviation sequence, and associated with the reproduction task identifier, the simulation task identifier and result identifier corresponding to the target simulation result to generate a reproduction deviation representation.

[0254] In this embodiment, generating the candidate biased feature set includes:

[0255] Read the change characteristics of the result indicators in the reproduced deviation characterization, associate them with the potential characterization of sensitive routes, determine the sensitive simulation elements corresponding to each result indicator, and generate a deviation-sensitive association set, where:

[0256] Generate bias-sensitive association sets, specifically as follows:

[0257] Read the result identifier, result indicator name, relative difference magnitude of result indicator, direction code of result indicator and direction difference value from the reproduction deviation characterization, and determine the result indicator with deviation according to the result indicator name;

[0258] Read the latent variables corresponding to the parameter domain, model domain, configuration domain and result domain in the latent representation of sensitive routes, and solve the parameter sensitivity factor, model sensitivity factor, configuration sensitivity factor and final propagation weight formed in the sensitivity factor routing layer;

[0259] For each result indicator with bias, read the corresponding latent variable in the result domain, calculate the dot product correlation value between the parameter domain latent variable, the model domain latent variable, the configuration domain latent variable and the result domain latent variable respectively, and multiply the dot product correlation value by the corresponding final propagation weight to obtain the parameter correlation strength, model correlation strength and configuration correlation strength.

[0260] Multiply the parameter correlation strength, model correlation strength, and configuration correlation strength by the corresponding parameter sensitivity factor, model sensitivity factor, and configuration sensitivity factor, respectively, and then multiply by the relative difference magnitude of the corresponding result index to obtain the deviation sensitivity value of each simulation element to the current result index.

[0261] Arrange the simulation elements corresponding to the parameter domain, model domain, and configuration domain according to the deviation sensitivity value from high to low, and determine the simulation elements whose deviation sensitivity value is greater than the average of all deviation sensitivity values ​​under the same result index as sensitive simulation elements.

[0262] Associate the result identifier, result indicator name, result indicator direction code, parameter identifier corresponding to the sensitive simulation element, model version identifier or solver configuration identifier, deviation sensitivity value and final propagation weight to form a single deviation sensitivity record;

[0263] All individual deviation-sensitive records are aggregated according to the result identifier and result indicator name to generate a deviation-sensitive association set;

[0264] Based on the input parameters, model version, and solution configuration corresponding to the deviation-sensitive association set, a deviation attribution path is established according to the association hierarchy, generating a candidate attribution path set, where:

[0265] The candidate attribution path set is generated as follows:

[0266] Read the result identifier, result index name, parameter identifier, model version identifier, solver configuration identifier, and deviation sensitivity value from the bias-sensitive association set, and extract the corresponding sensitive input parameters, sensitive model versions, and sensitive solver configurations according to the result index names;

[0267] Starting from the parameter identifier corresponding to the sensitive input parameter, the hierarchical connections between the parameter identifier and the model version identifier, the model version identifier and the solver configuration identifier, and the solver configuration identifier and the result identifier are sequentially retrieved in the trusted simulation association chain to form the parameter attribution path.

[0268] Starting from the model version identifier corresponding to the sensitive model version, the hierarchical connections between the model version identifier and the solution configuration identifier, and between the solution configuration identifier and the result identifier are sequentially retrieved in the trusted simulation association chain to form the model attribution path.

[0269] Using the solution configuration identifier corresponding to the sensitive solution configuration as the starting point of the path, the hierarchical connection between the solution configuration identifier and the result identifier is retrieved in the trusted simulation association chain to form a configuration attribution path;

[0270] The parameter attribution path, model attribution path, and configuration attribution path are associated with the corresponding result indicator name, result indicator direction code, bias sensitivity value, and simulation task identifier, respectively, to form a candidate attribution path record;

[0271] All candidate attribution path records are collected according to the result identifier and result indicator name, and the input parameters, model version, solution configuration and hierarchical connection order in each path are retained to generate a candidate attribution path set;

[0272] The attribution correlation degree between each associated simulation element and the change characteristics of the result index is calculated along the candidate attribution path set. The attribution ranking is then determined by combining the solution sensitivity relationship in the potential representation of sensitive routes, generating the ranking result of biased elements. Where:

[0273] The following is a list of the results that generate the ranking of biased elements:

[0274] Read the parameter attribution path, model attribution path, and configuration attribution path from the candidate attribution path set, and read the corresponding result index name, bias sensitivity value, result index direction code, and simulation task identifier.

[0275] For each candidate attribution path, read the association source corresponding to each level connection in the path, record the association confidence value corresponding to the original level connection in the historical association relationship set as 1, take the corresponding candidate completion confidence value of the association confidence value corresponding to the level connection formed by the confidence completion relationship, and multiply all association confidence values ​​in the same path to obtain the path confidence.

[0276] Within the same result index range, the normalized deviation sensitivity value is obtained by dividing the deviation sensitivity value corresponding to a single associated simulation element by the sum of the deviation sensitivity values ​​of all candidate associated simulation elements.

[0277] Read the relative difference magnitude of the result indicators in the change characteristics of the result indicators, multiply the normalized deviation sensitivity value, path confidence and the relative difference magnitude of the result indicators to obtain the attribution correlation degree between the associated simulation elements and the corresponding result indicators;

[0278] Read the parameter sensitivity factors, model sensitivity factors, or configuration sensitivity factors corresponding to the associated simulation elements from the potential representation of sensitive routes, and read the final propagation weights formed by solving the sensitive factor routing layer. Multiply the corresponding solved sensitive factors with the final propagation weights to obtain the solved sensitivity relationship values.

[0279] Within the same result index range, the normalized solution sensitivity value is obtained by dividing the individual solution sensitivity value by the sum of the solution sensitivity values ​​of all candidate correlation simulation elements.

[0280] The attribution correlation score is obtained by multiplying the normalized sensitivity correlation value by the attribution correlation score.

[0281] When the same associated simulation element corresponds to multiple result indicators, the sum of the individual attribution scores corresponding to each result indicator is divided by the number of corresponding result indicators to obtain the comprehensive attribution score.

[0282] The input parameters, model versions, and solution configurations are arranged from high to low according to the comprehensive attribution score, and the corresponding candidate attribution paths, result indicator names, and result indicator direction codes are associated to generate the ranking results of biased factors.

[0283] Based on the ranking results of bias-causing factors, relevant simulation factors that meet the attribution conditions are selected, and the corresponding bias attribution paths and result index change characteristics are associated to generate a candidate set of bias-causing factors, wherein:

[0284] Generate a candidate biased feature set, specifically as follows:

[0285] Read the input parameters, model version, solution configuration, comprehensive attribution score, candidate attribution path, result index name, and result index direction code from the ranking results of biased factors;

[0286] Calculate the average value of the comprehensive attribution scores of all related simulation elements within the same result index range, and calculate the maximum value among all comprehensive attribution scores. Use the median value between the average value and the maximum value of the comprehensive attribution scores as the attribution screening threshold.

[0287] Input parameters, model versions, and solution configurations with a comprehensive attribution score greater than or equal to the attribution screening threshold are identified as related simulation elements that meet the attribution conditions, while related simulation elements with a comprehensive attribution score lower than the attribution screening threshold are removed from the current candidate range.

[0288] Read the candidate attribution paths corresponding to the associated simulation elements that meet the attribution conditions, and read the relative difference magnitude, direction code and direction difference value of the corresponding result indicators from the change characteristics of the result indicators. Establish the correspondence between the associated simulation elements, the deviation attribution paths and the change characteristics of the result indicators.

[0289] When the same associated simulation element corresponds to multiple result indicators, the corresponding multiple deviation attribution paths and result indicator change characteristics are aggregated, and the comprehensive attribution score corresponding to each result indicator is retained;

[0290] The associated simulation elements that meet the attribution conditions, their corresponding comprehensive attribution scores, bias attribution paths, result index names, relative difference magnitudes of result indices, result index direction codes, and simulation task identifiers are arranged in descending order of comprehensive attribution scores to generate a candidate set of bias-causing elements.

[0291] In this embodiment, generating the closed-loop verification result includes:

[0292] Read the associated simulation elements, deviation attribution paths, and result index change characteristics from the candidate bias factor set, correlate them with historical simulation environment characterization, and generate a closed-loop retest task, wherein:

[0293] The closed-loop verification task is generated as follows:

[0294] Read the associated simulation elements, comprehensive attribution score, bias attribution path, result index name, relative difference magnitude of result index, result index direction code and simulation task identifier from the candidate bias factor set;

[0295] According to the simulation task identifier, the corresponding working conditions, input parameters, model version and solution configuration are read from the historical simulation environment representation, and the position of the associated simulation elements in the historical simulation environment representation is determined.

[0296] For each associated simulation element, the corresponding deviation attribution path is read, and the input parameters, model version, solution configuration and result index connected by the deviation attribution path are extracted. The corresponding relationship is established with the relative difference magnitude of the result index, the direction code of the result index and the direction difference value in the change characteristics of the result index.

[0297] The re-verification order of related simulation elements is determined according to the comprehensive attribution score from high to low. Each related simulation element is identified as the current re-verification object, and the working conditions, input parameters, model version and solution configuration in the historical simulation environment representation other than the current re-verification object are recorded as fixed environment elements.

[0298] The current re-test object, fixed environmental elements, corresponding deviation attribution path, result index change characteristics, target simulation results and re-test sequence are associated to form a single re-test task record;

[0299] All individual re-verification task records are collected according to the re-verification order, and associated with the corresponding simulation task identifiers to generate closed-loop re-verification tasks.

[0300] Based on the closed-loop verification task, the correlation and closure relationship between candidate bias-causing factors and operating conditions, input parameters, model version, and solution configuration are checked, and a closure verification result is generated, in which:

[0301] The closure check result is generated as follows:

[0302] Read the current verification object, fixed environmental elements, deviation attribution path, target simulation results and simulation task identifier in the closed-loop verification task;

[0303] According to the simulation task identifier, the operating condition identifier, parameter identifier, model version identifier, solution configuration identifier and result identifier corresponding to the current verification object are read from the trusted simulation association chain, and a corresponding relationship is established with the operating conditions, input parameters, model version and solution configuration in the fixed environment elements;

[0304] Check sequentially along the deviation attribution path whether there are continuous hierarchical connections between the working condition identifier and the parameter identifier, the parameter identifier and the model version identifier, the model version identifier and the solver configuration identifier, and the solver configuration identifier and the result identifier;

[0305] Record a closed state value of 1 for adjacent simulation elements that have continuous hierarchical connections, and record a closed state value of 0 for adjacent simulation elements that lack hierarchical connections. Multiply all closed state values ​​of the same path to obtain the path closure value.

[0306] When the path closure value is 1, the current verification object is marked as associated closure;

[0307] When the path closure value is 0, read the associated position and simulation element identifier corresponding to the missing hierarchical connection and mark it as an unclosed association;

[0308] Associate the current verification object, path closure value, closure status, missing associated location, corresponding deviation attribution path, and simulation task identifier to generate closure verification results;

[0309] Based on the closure check results, the simulation conditions corresponding to the candidate bias-causing factors are updated, the correction simulation environment is reconstructed, and the simulation solution is executed to generate the correction simulation results, in which:

[0310] The calibration simulation results are generated as follows:

[0311] Read the current verification object, closure state, corresponding deviation attribution path and simulation task identifier from the closure verification result, and read the fixed environmental elements and target simulation results from the closed-loop verification task;

[0312] When the current verification object corresponds to an associated closure state, determine the input parameters, model version, or solver configuration category to which the current verification object belongs, and read the existing simulation condition values ​​corresponding to the current verification object in the historical simulation environment representation;

[0313] Based on the deviation attribution path corresponding to the current re-examination object, retrieve similar simulation elements that have a continuous correlation path with the target simulation result from the trusted simulation association chain, eliminate similar simulation elements that have the same values ​​as the existing simulation conditions, and form candidate update elements.

[0314] Read the candidate completion confidence and path confidence in the associated path of the candidate update element, sort them from high to low according to the path confidence, and determine the value corresponding to the candidate update element with the highest path confidence as the verification update value.

[0315] When the current verification object is an input parameter, the corresponding parameter value will be replaced with the verification update value;

[0316] When the current verification object is a model version, the corresponding model version will be replaced with the verification update value;

[0317] When the current verification object is the solution configuration, the corresponding solution configuration will be replaced with the verification update value;

[0318] The working conditions, input parameters, model version and solution configuration of the closed-loop verification task are kept unchanged except for the current verification object. The updated current verification object is recombined with the fixed environmental elements to form the calibration simulation conditions.

[0319] Load the corresponding model version according to the calibration simulation conditions, write the input parameters and solver configuration, and complete the simulation environment initialization according to the working conditions to rebuild the calibration simulation environment;

[0320] Establish a calibration solution task based on the calibration simulation environment, perform typical diesel engine operating condition simulation calculations according to the solution configuration, and record the result identifier, result index name and result index value corresponding to the target simulation result.

[0321] The calibration simulation environment, the result indicators corresponding to the calibration solution task, the current verification object, the verification update value, the deviation attribution path and the simulation task identifier are associated to generate calibration simulation results;

[0322] The calibration simulation results are paired with the target simulation results for result index matching and deviation comparison. The deviation change is calculated by combining the reproduced deviation characterization, generating closed-loop verification results, where:

[0323] The closed-loop verification results are generated as follows:

[0324] Read the result identifier, result index name, and result index value from the calibration simulation result and the target simulation result, and establish a pairing relationship according to the result identifier and result index name;

[0325] The absolute value of the difference between the corrected result index value and the target result index value is calculated to obtain the magnitude of the difference in the corrected result index.

[0326] When the target result indicator is not equal to 0, the difference magnitude of the correction result indicator is divided by the absolute value of the target result indicator to obtain the relative difference magnitude of the correction result indicator.

[0327] When the target result indicator is equal to 0, the difference magnitude of the correction result indicator is directly used as the difference characterization.

[0328] Read the original difference magnitude of the corresponding result index from the reproduction deviation characterization, calculate the difference between the original difference magnitude and the corrected difference magnitude, and obtain the reduction amount of the single deviation.

[0329] When the original difference magnitude is not equal to 0, divide the reduction amount of the individual deviation by the original difference magnitude to obtain the reduction ratio of the individual deviation.

[0330] When the original difference magnitude is equal to 0, the reduction ratio of the individual deviation is recorded as 0;

[0331] When the same current re-examination object corresponds to multiple result indicators, calculate the average of the reduction ratio of each individual deviation to obtain the overall deviation reduction ratio of the current re-examination object;

[0332] Read the calibration simulation environment associated with the calibration simulation results, and associate the current verification object, the difference magnitude of calibration result indicators, the relative difference magnitude of calibration result indicators, the reduction amount of individual deviations, the reduction ratio of individual deviations, the reduction ratio of comprehensive deviations, the calibration simulation environment, the deviation attribution path, and the simulation task identifier.

[0333] When the overall deviation reduction rate is greater than 0, the current re-inspection object is recorded as a valid re-inspection object;

[0334] When the overall deviation reduction rate is less than or equal to 0, the current re-inspection object will be recorded as an invalid re-inspection object.

[0335] All re-test records are collected based on the current re-test object, and a closed-loop re-test result is generated that includes the calibration simulation environment and the corresponding deviation changes.

[0336] In this embodiment, generating simulation-related reproduction results includes:

[0337] Read the correction simulation results and deviation changes from the closed-loop retest results, correlate them with the target simulation results, determine the correction simulation environment that meets the retest judgment conditions, and generate a reliable historical simulation environment, wherein:

[0338] The process of generating a reliable historical simulation environment is as follows:

[0339] Read the calibration simulation results, individual deviation reduction ratio, comprehensive deviation reduction ratio, current calibration object and corresponding calibration simulation environment from the closed-loop re-verification results, and read the result identifier, result index name and target result index value from the target simulation results;

[0340] According to the result identifier and result index name, establish a correspondence between the calibration simulation result and the target simulation result, read the relative difference of the calibration result index corresponding to each result index, and calculate the average value of the relative difference of all calibration result indexes under the same calibration simulation environment to obtain the environmental retest deviation value.

[0341] Read the relative difference magnitude of the corresponding result index in the reproduction deviation characterization, calculate the average of the relative difference magnitudes of all result indices, and obtain the original reproduction deviation value;

[0342] Corrected simulation environments that have a comprehensive deviation reduction ratio greater than 0 and an environmental re-verification deviation value less than the original reproduction deviation value are marked as corrected simulation environments that meet the re-verification judgment conditions.

[0343] When there are multiple calibration simulation environments that meet the re-verification criteria, sort them by environment re-verification deviation value from smallest to largest, and select the calibration simulation environment with the smallest environment re-verification deviation value as the target recovery environment.

[0344] Read the operating conditions, input parameters, model version and solver configuration in the target recovery environment, and associate them with the corresponding simulation task identifier, current verification object, deviation attribution path, correction simulation results and environment verification deviation value to generate a reliable historical simulation environment;

[0345] Based on the closed-loop verification results, the deviation changes corresponding to the candidate bias-causing factors are extracted. The associated simulation factors confirmed by the verification are then selected according to the deviation reduction relationship to generate bias-causing simulation factors, where:

[0346] The following parameters are generated to induce biased simulation elements:

[0347] Read the current verification object, relative difference of correction result indicators, reduction of single deviation, reduction ratio of single deviation, reduction ratio of comprehensive deviation and verification status in the closed-loop verification results, and establish a correspondence with the candidate bias-causing factor set according to the parameter identifier, model version identifier or solution configuration identifier corresponding to the current verification object;

[0348] For each candidate bias factor, read the corresponding comprehensive deviation reduction ratio and the individual deviation reduction ratio corresponding to each result indicator, and include the candidate bias factors with a comprehensive deviation reduction ratio greater than 0 and a re-examination status of valid re-examination object into the re-examination confirmation scope.

[0349] When a candidate bias factor corresponds to multiple outcome indicators, count the number of outcome indicators with a single bias reduction ratio greater than 0, divide the number of outcome indicators with a single bias reduction ratio greater than 0 by the total number of outcome indicators corresponding to the candidate bias factor to obtain the bias improvement ratio.

[0350] The associated simulation elements with a comprehensive deviation reduction ratio greater than 0 and a deviation improvement ratio greater than or equal to 0.5 are identified as the associated simulation elements that have been verified.

[0351] Read the parameter identifier, model version identifier, or solver configuration identifier corresponding to the verified and confirmed related simulation elements, and associate them with the comprehensive attribution score, comprehensive deviation reduction ratio, deviation improvement ratio, deviation attribution path, and corresponding result indicator name to form a deviation-causing element confirmation record.

[0352] All deviation-causing factor confirmation records are arranged from high to low according to the overall deviation reduction ratio, and deviation-causing simulation factors are generated.

[0353] The trusted historical simulation environment and bias-causing simulation elements are written back to the trusted simulation association chain, updating the environmental association relationships and bias attribution relationships of the corresponding simulation tasks, and generating a re-verification association chain, in which:

[0354] The process of generating a verification chain is as follows:

[0355] Read the operating conditions, input parameters, model version, solver configuration, corresponding simulation task identifier and environment verification deviation value in the trusted historical simulation environment, and read the parameter identifier, model version identifier or solver configuration identifier, comprehensive attribution score, comprehensive deviation reduction ratio and deviation attribution path in the deviation simulation elements.

[0356] According to the simulation task identifier, locate the corresponding task-level simulation association path in the trusted simulation association chain, write the working conditions, input parameters, model version and solution configuration in the trusted historical simulation environment into the corresponding positions of the working condition layer, parameter layer, model layer and configuration layer respectively, and retain the target simulation results in the corresponding result layer to form a verification environment association record.

[0357] Locate the corresponding node of the simulation element causing deviation in the task-level simulation association path according to the parameter identifier, model version identifier, or solver configuration identifier. Write the comprehensive attribution score, comprehensive deviation reduction ratio, and corresponding deviation attribution path into the association record between the corresponding node and the result index to form a re-verification deviation attribution record.

[0358] Associate the retest environment association record with the retest deviation attribution record, update the environment association relationship from the working condition layer to the parameter layer, the parameter layer to the model layer, the model layer to the configuration layer, and the configuration layer to the result layer, and add the deviation attribution relationship from the simulation elements that cause deviations to the corresponding result indicators.

[0359] The updated task-level simulation association path is reorganized according to the simulation task identifier, retaining the original historical association relationships, trusted completion relationships, verification environment association records and verification deviation attribution records, and generating a verification association chain.

[0360] The correlation and verification chain, the trusted historical simulation environment, the biased simulation elements, and the target simulation results are linked and verified. Reproduction records are organized according to the simulation task identifier to generate simulation correlation and reproduction results, including:

[0361] The simulation-related reproduction results are generated as follows:

[0362] Read the calibration simulation results in the closed-loop verification results, read the task-level simulation association path, verification environment association record and verification deviation attribution record in the verification association chain, and read the operating conditions, input parameters, model version, solution configuration and environment verification deviation value in the trusted historical simulation environment.

[0363] Read the parameter identifier, model version identifier or solver configuration identifier, comprehensive attribution score, comprehensive deviation reduction ratio, deviation improvement ratio and deviation attribution path from the simulation elements that cause deviation, and read the result identifier, result index name and target result index value from the target simulation result;

[0364] According to the simulation task identifier, the trusted historical simulation environment, the biased simulation elements and the target simulation results are matched with the corresponding task-level simulation association paths in the re-verification association chain to establish the correspondence between environment recovery information, biased element information, target result information and re-verification association information.

[0365] The working conditions, input parameters, model version, solution configuration, target simulation results, corrected simulation results, environmental verification deviation values, simulation factors causing deviations, comprehensive attribution scores, and comprehensive deviation reduction ratios corresponding to the same simulation task are organized into a single simulation reproduction record.

[0366] All single simulation reproduction records are collected according to the simulation task identifier, and the association paths between the working condition layer, parameter layer, model layer, configuration layer and result layer are retained, as well as the deviation attribution paths from the simulation elements that cause deviations to the corresponding result indicators, and simulation association reproduction results are generated.

[0367] Example 1: To verify the feasibility of this invention in practice, it was applied to a simulation data reproduction verification scenario of a typical diesel engine operating condition. The simulation management platform received 6200 sets of completed simulation task records. Each task set included typical operating condition information, simulation input parameter information, model and version information, solver configuration information, and simulation result information. The 6200 task sets contained a total of 6200 operating condition identifiers, 37200 input parameter records, 18 model versions, 27 types of solver configurations, and 24800 result indicator records. Complete task records were manually retained as a verification benchmark, and 742 correlation anomalies were set from the historical records, including 316 missing correlations, 287 broken correlations, and 139 conflicting correlations, to simulate file migration, version switching, and missing correlation record issues that occur during long-term simulation data accumulation.

[0368] After the simulation data enters the processing flow, the system collects the records according to the simulation task identifier. A certain training sample corresponds to an engine speed of 1800 r / min, a load rate of 75%, an intake pressure of 0.241 MPa, a fuel injection quantity of 61.8 mg / cycle, a fuel injection advance angle of 8.4°CA, and a coolant temperature of 358.2 K. The model version is MV-12, the solution step size is 0.10°CA, the upper limit of iteration is 1200, the convergence threshold is 1.0×10^-5, and the corresponding peak in-cylinder pressure is 15.72 MPa, the exhaust temperature is 751.6 K, the fuel consumption rate is 208.4 g / (kW·h), and the turbocharger speed is 101320 r / min. Another training sample corresponds to an engine speed of 2000 r / min, a load rate of 85%, an intake pressure of 0.258 MPa, an injection quantity of 67.3 mg / cycle, an injection advance angle of 9.1°CA, and a coolant temperature of 360.4 K. The model version is MV-14, the solution step size is 0.08°CA, and the corresponding four result indicators are 17.26 MPa, 779.8 K, 214.7 g / (kW·h) and 108640 r / min, respectively.

[0369] After the fields were standardized, the engine speed, pressure, fuel injection quantity, injection advance angle, and temperature were normalized according to the minimum and maximum values ​​of the fields in the 6200 completed tasks. The historical range of engine speed is 1200–2400 r / min, with 1800 r / min corresponding to a value code of 0.500; the historical range of intake pressure is 0.185–0.292 MPa, with 0.241 MPa corresponding to a value code of 0.523; and the historical range of fuel injection quantity is 42.6–76.8 mg / cycle, with 61.8 mg / cycle corresponding to a value code of 0.561. After coding, a sequence of 6200 stratified simulation elements was formed, with each group of input representation dimensions standardized to 256.

[0370] The system checks the correlations between the operating condition layer, parameter layer, model layer, configuration layer, and result layer according to the simulation task identifier. Candidate completion relationships are generated for 742 correlation anomalies. For a certain model version break record, there are three candidate versions: MV-11, MV-12, and MV-15. The time sequence consistency criteria are 1, 1, and 0 respectively; the version continuity consistency criteria are 0, 1, and 0 respectively; and the result attribution consistency criteria are 1, 1, and 1 respectively. The corresponding candidate completion confidence levels are 0.67, 1.00, and 0.33 respectively. The system selects MV-12 for completion. Of the 742 correlation anomalies, 718 completion results are consistent with the reserved baseline, achieving a completion accuracy of 96.77%. The traditional method of directly matching according to the simulation task identifier and file generation order only recovers 629 anomalies, with an accuracy of 84.77%.

[0371] After completing the reliable simulation association chain, 4960 sets were selected from 6200 sets of data as the training set and 1240 sets as the validation set. The training set contains approximately 29760 input parameter records and 19840 result index records. The number of latent variables and the dimension of the latent variables in the improved Perceiver IO network were set to 256, and the initial learning rate was set to 0.0001. The validation loss decreased to 0.0049 by the 164th epoch and to 0.0046 by the 173rd epoch. After continuously satisfying the convergence condition, the model parameters of the 173rd epoch were saved. When training the same samples using the original Perceiver IO network, the lowest validation loss was 0.0083.

[0372] In the target task reproduction experiment, one set of historical results was selected as the target simulation result. The target environment corresponds to an engine speed of 1800 r / min, a load rate of 75%, an intake pressure of 0.238 MPa, a fuel injection quantity of 62.4 mg / cycle, a fuel injection advance angle of 8.6°CA, a model version of MV-13, and a solution step size of 0.10°CA. The target in-cylinder peak pressure is 15.84 MPa, the exhaust temperature is 748.3 K, the fuel consumption rate is 207.6 g / (kW·h), and the turbocharger speed is 101800 r / min. The model version association in the historical records was broken. The traditional method restored it to MV-11 based on the nearest file and incorrectly used a solution step size of 0.20°CA. The reproduced results were 14.91 MPa, 783.5 K, 218.1 g / (kW·h), and 96940 r / min, with relative deviations of 5.87%, 4.70%, 5.06%, and 4.77%, respectively.

[0373] After the trusted simulation association chain is encoded, it enters the version domain consistency evolution layer. In the model domain corresponding to the target task, the version consistency determination value between MV-13 and MV-12 on the inheritance path is 1, while the version consistency determination value between MV-13 and another branch MV-16 is 0. After cross-domain controlled evolution, the propagation weight corresponding to MV-13 increases from 0.34 to 0.48, and the propagation weight corresponding to MV-16 decreases from 0.27 to 0. The sensitivity factor routing layer calculates the injection advance angle sensitivity factor (0.31), injection quantity sensitivity factor (0.22), model version sensitivity factor (0.29), and configuration sensitivity factor (0.18). After high-sensitivity propagation is maintained, the final propagation weights corresponding to injection advance angle, model version, and injection quantity are 0.36, 0.32, and 0.21, respectively.

[0374] The historical environment reverse restoration layer performs a reverse query based on the target results. The matching value between the candidate operating condition and the environmental elements of the 1800 r / min, 75% load operating condition is 0.94; the matching values ​​between the candidate parameter and the injection quantity of 62.4 mg / cycle and the injection advance angle of 8.6°CA are 0.91 and 0.93, respectively; the matching value between the candidate model and MV-13 is 0.96; and the matching value between the candidate solution configuration and the 0.10°CA configuration is 0.92. After closure verification, the restored operating conditions, input parameters, model version, and solution configuration are all consistent with the reserved real environment.

[0375] After resolving based on the restored environment, the four results were 15.63 MPa, 753.4 K, 210.0 g / (kW·h), and 100960 r / min, respectively, with corresponding relative deviations reduced to 1.33%, 0.68%, 1.16%, and 0.83%. The system further constructed a characterization of the reproduced deviation, with comprehensive attribution scores for injection advance angle, model version, solution configuration, and injection quantity being 0.84, 0.78, 0.63, and 0.47, respectively. Within the same result index range, the average comprehensive attribution score was 0.68, the maximum was 0.84, and the dynamic attribution screening threshold was calculated to be 0.76. Injection advance angle and model version were included in the candidate bias-causing factor set.

[0376] During the closed-loop retest, all environmental factors except the current retest object remained unchanged. After restoring the injection advance angle from 8.1°CA to 8.6°CA, the average result deviation decreased from 1.00% to 0.61%, with a comprehensive deviation reduction of 39.00%. After restoring the model version from MV-11 to MV-13, the average result deviation further decreased to 0.28%, with a comprehensive deviation reduction of 72.00%. The final results were 15.80 MPa, 749.8 K, 208.1 g / (kW·h), and 101520 r / min, with deviations from the target results of 0.25%, 0.20%, 0.24%, and 0.28%, respectively. All four results showed improvement, with a deviation improvement rate of 1.00%.

[0377] A comparison was conducted on 1240 verification tasks using the same process. The historical association recovery accuracy of the present invention was 96.61%, compared to 84.35% for the traditional method; the historical environment integrity recovery rate was 95.89%, compared to 81.77% for the traditional method; the average result reproduction deviation was 0.82%, compared to 4.56% for the traditional method; and the accuracy rate for identifying biased factors was 92.74%, compared to 76.45% for the traditional method based on single-item difference sorting. The average processing time for a single abnormal task from data reading to completion of verification was 8.4 minutes, while the average processing time for traditional manual file retrieval, version verification, and repeated solving was 39.7 minutes. The comparative results show that the present invention can recover a reliable historical simulation environment under conditions of missing associations, version mismatch, and broken solution configuration, and can identify the actual biased simulation factors through deviation attribution and closed-loop verification.

[0378] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for correlation and reproduction of simulation data for typical diesel engine operating conditions, characterized in that, include: Collect basic simulation information under typical diesel engine operating conditions, perform standardization processing, and generate a standardized simulation element set; Based on a standardized set of simulation elements, historical correlation gaps are identified, reliable gap filling is performed, multi-level correlation relationships are established, and a reliable simulation correlation chain is generated. The simulation results to be reproduced are determined from the trusted simulation association chain as the target simulation results. An improved Perceiver IO network is constructed, and a version domain consistent evolution layer, a sensitive factor routing layer, and a historical environment reverse restoration layer are set. The trusted simulation association chain and the target simulation results are input to generate sensitive routing potential representations and historical simulation environment representations. Based on the historical simulation environment characterization, the corresponding simulation conditions are restored, the historical simulation environment is reconstructed, and the simulation solution is executed to generate simulation reproduction results; The simulation results are correlated and compared with the target simulation results. The differences in the results indicators are analyzed, and a characterization of the reproduction deviation is generated. Based on the reproducible deviation characterization, the sensitive route potential characterization, and the credible simulation association chain, the reproducible deviation is attributed stepwise to generate a candidate set of bias-causing factors. Based on the candidate biased feature set, perform historical environment closed-loop verification, update the simulation features and perform simulation solution again to generate closed-loop verification results. Based on the closed-loop verification results, the credible historical simulation environment and the biased simulation elements are determined, and simulation correlation reproduction results are generated.

2. The method for correlation and reproduction of simulation data for typical diesel engine operating conditions according to claim 1, characterized in that, The simulation basic information includes typical working condition information, simulation input parameter information, model and version information, solver configuration information, simulation result information, and simulation task association information.

3. The method for simulation data association and reproduction for typical diesel engine operating conditions according to claim 1, characterized in that, The generation of the standardized simulation element set includes: Acquire typical working condition information, simulation input parameter information, model and version information, solver configuration information, simulation result information and simulation task association information, and collect them according to the simulation task identifier to generate simulation task element groups; Perform format conversion, field name unification, and unit of measurement standardization on the simulation task element group to generate a unified format simulation element group; Based on a unified format simulation element group, extract the working condition identifier, parameter identifier, model version identifier, solution configuration identifier, and result identifier to establish a simulation element object index; The simulation element object indexes are time-aligned according to the execution time corresponding to the simulation task, and associated with a unified format simulation element group to generate a standardized simulation element set.

4. The method for correlation and reproduction of simulation data for typical diesel engine operating conditions according to claim 1, characterized in that, The generation of the trusted simulation association chain includes: Based on the standardized simulation element set, read the simulation task identifier and simulation element object index corresponding to the completed simulation task, match the association records between typical working conditions, simulation input parameters, model version, solution configuration and simulation results, and generate a historical association relationship set; Perform an integrity check on the historical association set to identify missing, broken, and conflicting associations, and generate a historical association gap set; Based on the standardized simulation element set, retrieve the simulation elements of the same task corresponding to the historical associated gap set, combine the time sequence, version connection and result correspondence to generate candidate completion relationships, calculate the credibility of candidate completion, and generate a set of credible completion relationships; Write the trusted completion relation set into the historical relation set, and reorganize the relation paths according to the hierarchical relationship of working conditions, parameters, model version, solution configuration and simulation results to generate a trusted simulation relation chain.

5. The method for correlation and reproduction of simulation data for typical diesel engine operating conditions according to claim 1, characterized in that, The generation of sensitive route potential representations and historical simulation environment representations includes: Determine the simulation result to be reproduced from the trusted simulation association chain, and mark the simulation result to be reproduced as the target simulation result; An improved Perceiver IO network is constructed, which includes a version-domain consistent evolution layer, a sensitivity factor routing layer, and a historical environment reverse restoration layer. The trusted simulation association chains are encoded according to the simulation element category and association level, and then mapped to the latent space through cross-attention to generate the simulation latent representation. The simulation latent representation is input into the version domain consistent evolution layer, which divides the operating condition domain, parameter domain, model domain, configuration domain and result domain. The latent variable evolution is performed in combination with the version inheritance relationship and branch relationship to generate the version consistent latent representation. Input the version-consistent potential representation into the sensitive factor routing layer, calculate the sensitive factor based on the response relationship between the changes in simulation elements and the changes in result indicators under typical working conditions, adjust the propagation path and propagation weight of latent variables, and generate the sensitive routing potential representation. The potential representations of sensitive routes and the target simulation results are input into the historical environment reverse restoration layer. Based on the target simulation results, the potential representation reverse query is performed to extract the associated simulation condition representations and generate the historical simulation environment representations. The improved Perceiver IO network is trained, historical reliable simulation correlation data and historical environment and simulation result annotations are obtained, and training set and validation set are divided. The training set is input into the improved Perceiver IO network to train the parameters of the version domain consistency evolution layer, the sensitive factor routing layer and the historical environment reverse restoration layer. The parameters are updated based on the environment restoration error, the sensitive bias, and the version consistency error. The validation loss is obtained using the validation set. Training is terminated when the validation loss is lower than the training convergence threshold or the training epoch reaches the upper limit. The model parameters with the lowest validation loss are saved.

6. The method for simulation data association and reproduction for typical diesel engine operating conditions according to claim 1, characterized in that, The generated simulation reproduction results include: Read the operating conditions, input parameters, model version and solution configuration from the historical simulation environment representation, restore the correspondence between each simulation condition according to the trusted simulation association chain, and generate historical environment restoration items; Load the corresponding model version based on the historical environment restoration item, write the input parameters and solution configuration, and complete the simulation environment initialization according to the working conditions to generate the historical simulation environment; Establish simulation solution tasks based on historical simulation environments, execute typical diesel engine operating condition simulation calculations according to the solution configuration, record the solution process and result indicators, and generate a reproducible solution record. Extract the result indicators corresponding to the target simulation results from the reproduced solution records, associate and organize them according to the result identifiers, and generate simulation reproduction results.

7. The method for simulation data association and reproduction for typical diesel engine operating conditions according to claim 1, characterized in that, The generation of the reproducibility bias characterization includes: Read the result indicators and corresponding result values ​​from the simulation reproduction results and the target simulation results, establish the correspondence between the indicators according to the result identifiers, and generate a set of result indicator pairs; The numerical differences between corresponding result indicators are calculated based on the result indicator pairing set, and the direction of the difference is determined according to the increase or decrease relationship of the result values ​​to generate the result indicator difference set. Based on the statistical analysis of the difference set of the outcome indicators, the magnitude and direction of the differences of each outcome indicator are statistically analyzed to form the change characteristics of the outcome indicators; The characteristics of changes in related result indicators, the paired set of result indicators, and the target simulation results are characterized and organized according to the order of result indicators to generate a representation of the reproducibility deviation.

8. The method for simulation data association and reproduction for typical diesel engine operating conditions according to claim 1, characterized in that, The generation of the candidate biased feature set includes: Read the change characteristics of the result indicators in the reproduced deviation characterization, associate them with the potential characterization of sensitive routes, determine the sensitive simulation elements corresponding to each result indicator, and generate a deviation-sensitive association set; Based on the reliable simulation association chain, the input parameters, model version and solution configuration corresponding to the deviation-sensitive association set are traced, and deviation attribution paths are established according to the association level to generate a candidate attribution path set. The attribution correlation degree between each associated simulation element and the change characteristics of the result index is calculated along the candidate attribution path set, and the attribution ranking is determined by combining the solution of the sensitivity relationship in the potential representation of sensitive routes, and the ranking result of the biased elements is generated. Based on the ranking results of bias-causing factors, we screen the associated simulation factors that meet the attribution conditions, and associate them with the corresponding bias attribution paths and the change characteristics of the result indicators to generate a candidate set of bias-causing factors.

9. The method for correlation and reproduction of simulation data for typical diesel engine operating conditions according to claim 1, characterized in that, The generation of closed-loop verification results includes: Read the associated simulation elements, deviation attribution paths, and result index change characteristics in the candidate bias factor set, associate them with historical simulation environment characterization, and generate a closed-loop retest task. Based on the closed-loop verification task, the correlation and closure relationship between candidate bias-causing factors and operating conditions, input parameters, model version and solution configuration are checked, and a closure verification result is generated. Based on the closure check results, update the simulation conditions corresponding to the candidate bias-causing factors, reconstruct the correction simulation environment, execute the simulation solution, and generate the correction simulation results. The calibration simulation results are paired with the target simulation results for result index matching and deviation comparison. The deviation change is calculated by combining the reproduction deviation characterization and the closed-loop verification results are generated.

10. The method for correlation and reproduction of simulation data for typical diesel engine operating conditions according to claim 1, characterized in that, The generated simulation-related reproduction results include: Read the correction simulation results and deviation changes in the closed-loop retest results, correlate them with the target simulation results, determine the correction simulation environment that meets the retest judgment conditions, and generate a reliable historical simulation environment; Based on the closed-loop retest results, the deviation changes corresponding to the candidate bias-causing factors are extracted, and the associated simulation factors confirmed by the retest are screened according to the deviation reduction relationship to generate bias-causing simulation factors. Write back the trusted historical simulation environment and biased simulation elements to the trusted simulation association chain, update the environmental association relationship and bias attribution relationship of the corresponding simulation task, and generate the re-verification association chain. The correlation and verification of the correlation chain, the credible historical simulation environment, the biased simulation elements and the target simulation results are linked and the reproduction records are organized according to the simulation task identifier to generate the simulation correlation and reproduction results.

Images