An intelligent verification method and system for electronic control parameters based on multimodal AI fusion

Abstract

The present invention discloses a method and system for intelligent verification of electric control parameters based on multimodal AI fusion, which relates to the field of industrial automation technology, including preprocessing parameters in different formats, performing format compliance checks through the initial verification layer, verifying whether the data naming complies with standard naming specifications, and comparing parameter values ​​with the safety range defined by the enterprise; parsing parameter description information through the semantic understanding layer, and using the knowledge graph to query the requirement document to verify whether the parameters contain necessary constraints; performing parameter dependency verification and system-level constraint checks through the association analysis layer; and performing comprehensive scoring of calibration parameters through the dynamic evaluation layer. Therefore, the use of the above-mentioned method and system for intelligent verification of electric control parameters based on multimodal AI fusion can overcome the problems of low verification efficiency, delayed rule update, and insufficient complex parameter association analysis capabilities in the prior art, and realize intelligent verification of electric control parameters.

Description

Technical Field

[0001] The present invention relates to the field of industrial automation technology, and in particular to a method and system for intelligent verification of electric control parameters based on multimodal AI fusion. Background Art

[0002] ECU calibration technology is crucial for precise vehicle performance tuning. It requires meticulous calibration of the ECU's operating and control parameters. Especially during the development phase, key parameters (such as constants a and b in fuel injection calculations) require technical adjustments to meet specific performance requirements.

[0003] Currently, engineers typically rely on Flash and RAM-based calibration methods, as well as AUTOSAR's pointer-based calibration concept, to ensure accurate execution of control algorithms. Existing calibration tools (such as INCA and CANape) support communication protocols such as CCP or XCP. These tools not only manage calibration data but also provide key functions such as data acquisition, calibration, and ECU flash programming, enabling comprehensive ECU testing and calibration.

[0004] However, the parameter calibration process relies on individual experience. With the increasing number of product calibration projects, individuals are often responsible for only one area. New hires are unfamiliar with the rest of the team, leading to inadequate verification. This leads to rework, poor quality, and management difficulties, making it difficult to accumulate technical expertise. Therefore, there is an urgent need to develop a method and approach that can accumulate expert calibration experience in specific areas and parameters, as well as effectively verify participation, to improve the quality of ECU calibration. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for intelligent verification of electronic control parameters based on multimodal AI fusion, which can overcome the problems of low verification efficiency, delayed rule update, insufficient ability to analyze complex parameter associations, etc. in the prior art, and realize intelligent verification of electronic control parameters.

[0006] To achieve the above objectives, the present invention provides an intelligent verification system for electronic control parameters based on multimodal AI fusion, comprising:

[0007] The initial verification layer, based on the basic rule base, performs format compliance checks and value range boundary verification on the verification benchmark;

[0008] The semantic understanding layer parses parameter description information based on the semantic understanding of the big model and verifies the requirement-parameter mapping through the fine-tuned domain big model;

[0009] The association analysis layer verifies parameter dependencies and checks system-level constraints based on the semantic knowledge graph, and obtains a compromise solution through a conflict rule self-negotiation strategy.

[0010] The dynamic assessment layer assesses risk probability based on a dynamic experience database and provides a comprehensive score based on basic compliance, logic, system security, and optimization potential, enabling case-driven adaptive verification. Furthermore, based on the aggregated verification feedback, the verification strategy evolution module uses reinforcement learning to update the corresponding strategy.

[0011] The initial verification layer, semantic understanding layer, association analysis layer, and dynamic evaluation layer are connected in sequence to process the input calibration parameters and obtain corresponding security scores based on historical cases; and the evaluation results of the dynamic evaluation layer are used to update the rules of the initial verification layer.

[0012] Preferably, the semantic knowledge graph includes building an ontology model based on parameter metadata, using a graph embedding algorithm to learn the association relationship between parameters, and dynamically updating the graph edge weights using actual verification result feedback.

[0013] Preferably, the dynamic experience library constructs a parameter-result mapping relationship database by collecting historical calibration cases.

[0014] Preferably, the verification strategy evolution module includes extracting a target parameter set from a verification request, performing multi-hop queries in a semantic knowledge graph, obtaining implicit association rules, and predicting potential conflicts using a graph neural network.

[0015] The present invention also provides an intelligent verification method for electric control parameters based on multimodal AI fusion, which uses the above-mentioned intelligent verification system for electric control parameters based on multimodal AI fusion, including the following steps:

[0016] S1. Normalize parameters in different formats, establish an enterprise-level parameter alias library, extract corresponding parameter features and associated features through parameter sensitivity analysis, and form parameter description information;

[0017] S2. Perform compliance checks on parameter formats through the initial validation layer, verify whether data naming complies with standard naming conventions, and compare parameter values ​​with the enterprise-defined safety range.

[0018] S3. Parse the parameter description information through the semantic understanding layer and input the parsed parameter description text into the fine-tuned domain model. Use the semantic knowledge graph to query the requirement document and verify whether the parameters contain the necessary constraints.

[0019] S4. Check whether there are loop dependencies in the semantic knowledge graph through the association analysis layer, verify whether the parameters comply with physical laws, call predefined constraint rules to perform system-level constraint checks, and use the conflict rule self-negotiation strategy to obtain a compromise solution;

[0020] S5. Based on the dynamic experience database, the dynamic assessment layer evaluates the risk probability of the calibration parameters, and uses basic compliance, logic, system security and optimization potential to perform a comprehensive score, and then implements adaptive parameter calibration and verification through case-driven methods.

[0021] Preferably, step S1 includes:

[0022] S11. Parse parameter files of different formats, extract parameter information of various formats, and perform standardization to unify the unit format;

[0023] S12. Based on the normalized parameter information, extract the time series features, statistical features, and correlation features of the parameters.

[0024] Preferably, in step S3, the parameter description text is parsed by a large language model LLM.

[0025] Preferably, in step S5, the calculation method of the optimization potential includes obtaining the deviation of the current calibration parameter value by comparing with the historical optimal calibration data.

[0026] Preferably, in step S5, the expression of the comprehensive score is: comprehensive score = basic compliance × 0.3 + logical consistency × 0.25 + system security × 0.25 + optimization potential × 0.2.

[0027] Therefore, the present invention adopts the above-mentioned method and system for intelligent verification of electric control parameters based on multimodal AI fusion, which has the following technical effects:

[0028] (1) By constructing a three-level knowledge system consisting of a basic rule base, a dynamic experience base, and a semantic knowledge graph, the authority of the benchmark can be verified and the workload of manual review can be reduced. At the same time, the ability to analyze complex parameter associations can be enhanced, the detection rate of system-level constraint conflicts can be improved, and the accuracy of anomaly detection can be improved.

[0029] (2) Through a hybrid reasoning mechanism consisting of semantic understanding based on a large model, association reasoning based on a knowledge graph, and verification strategy evolution based on reinforcement learning, it is possible to discover implicit association rules, predict potential conflicts, and resolve cross-subsystem parameter conflicts that cannot be detected by traditional methods.

[0030] (3) Through parameter sensitivity analysis, the fluctuation degree of parameter values ​​under different working conditions can be obtained, and the parameter characteristics and their relationship characteristics can be obtained, providing a reference for the calibration of parameter values; the conflict rule self-negotiation strategy can realize the intelligent verification process, and the feedback learning mechanism of the verification strategy evolution can update the corresponding strategy according to the actual verification results, thereby realizing dynamic verification.

[0031] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a flow chart of the intelligent verification method of electronic control parameters based on multimodal AI fusion. DETAILED DESCRIPTION

[0033] The present invention can be explained in more detail by the following examples. The purpose of disclosing the present invention is to protect all changes and improvements within the scope of the present invention. The present invention is not limited to the following examples.

[0034] The present invention provides an intelligent verification system for electric control parameters based on multimodal AI fusion, comprising:

[0035] The initial validation layer, based on a basic rule base, checks format compliance and verifies value range boundaries of the validation benchmark, ensuring its authority and reducing manual review workload. The basic rule base is constructed by parsing industry standard documents using natural language processing (NLP) and structured storage of enterprise design specifications (in XML format).

[0036] The semantic understanding layer parses parameter description information based on the semantic understanding of the big model and verifies the requirement-parameter mapping through the fine-tuned domain big model.

[0037] The association analysis layer, based on the semantic knowledge graph, verifies parameter dependencies and checks system-level constraints, enhancing complex parameter association analysis capabilities and improving the detection rate of system-level constraint conflicts. It employs a conflict rule auto-negotiation strategy to output compromise solutions, enabling intelligent parameter verification. The semantic knowledge graph includes constructing an ontology model based on parameter metadata, using a graph embedding algorithm to learn parameter associations, and dynamically updating graph edge weights using feedback from actual verification results.

[0038] The dynamic assessment layer provides a comprehensive score based on basic compliance, logic, system security, and optimisation potential to generate a credibility score for the calibration parameters. At the same time, the verification strategy evolution module extracts the target parameter set from the verification request, performs multi-hop queries in the knowledge graph, discovers implicit association rules, and uses graph neural networks to predict potential conflicts.

[0039] On the one hand, the initial verification layer, semantic understanding layer, association analysis layer, and dynamic assessment layer are connected in sequence to process the input calibration parameters and combine them with historical cases to derive a corresponding safety score. For example, after the calibration parameters pass the initial format verification, they enter the semantic understanding layer with a compliance mark, which parses out "This parameter is used for knock control under high-speed conditions." The association analysis layer then discovers its dependency on cooling system parameters in the knowledge graph. Finally, the dynamic assessment layer combines historical cases to determine the safety score.

[0040] On the other hand, the dynamic assessment layer updates the rules of the initial verification layer based on the dynamic assessment results. For example, if the dynamic assessment layer detects the same parameter anomaly three times in a row and the original rules are not covered, a new verification rule will be added to the initial verification layer. The hot rule update does not affect the current verification process.

[0041] like Figure 1 As shown, based on the above-mentioned electronic control parameter intelligent verification system based on multimodal AI fusion, the present invention also provides an electronic control parameter intelligent verification method based on multimodal AI fusion, which is suitable for parameter calibration and verification of embedded systems such as automotive electronic control units (ECUs) and industrial controllers, including the following steps:

[0042] S1. Preprocess the parameters, which includes two parts:

[0043] The first part normalizes parameters in different formats. This example parses configuration files in formats such as A2L, Hex, DCM, hex, s19, mot, smot, hmot, hcal, Excel, csv, and json, extracts corresponding parameter descriptions, and performs unified data processing and conversion on multiple parameters. Specifically, for files in A2L, Hex, DCM, hex, s19, mot, smot, hmot, and hcal formats, ECU descriptions are extracted, including parameter addresses, parameter types, and calibration quantities. For Excel / csv format files, table header fields and body values ​​are identified, including parameter names, values, units, and descriptions. For JSON format files, corresponding nested value information is extracted.

[0044] Through standardization, the unit format is unified, the parameter naming in different files is unified, and for parameters with time series characteristics, the interpolation method is used to unify the timestamps and establish an enterprise-level parameter alias library.

[0045] In the second part, through parameter sensitivity analysis, the time series characteristics, statistical characteristics and correlation characteristics of the parameters are extracted, and parameter description information is formed to provide a reference basis for the calibration of parameter values.

[0046] Among them, time series features are used for dynamic parameter calibration. For example, based on the sliding window difference method, the corresponding parameter values ​​are analyzed through a fixed window length, the main frequency component is obtained through FFT transformation, and the parameter change rate is calculated using linear regression as a derivative indicator of the time series features. Statistical features mainly calculate the distribution of parameter values ​​for steady-state parameter verification. For example, based on the 3σ principle, outliers that exceed ±3 times the standard deviation of the mean are marked and the coefficient of variation of the parameter within a fixed period is calculated. Correlation features use the DBSCAN algorithm to identify parameter groups with similar change patterns. Granger causality tests are used to determine the influence relationship between parameters for system-level verification.

[0047] S2. Perform compliance checks on the parameter format through the initial verification layer, verify whether the data naming complies with standard naming specifications, and compare parameter values ​​with the safety range defined by the enterprise.

[0048] S3. The semantic understanding layer parses the parameter description information through the large language model (LLM) to obtain the corresponding parameter description text, which is then input into the fine-tuned domain large model. The semantic knowledge graph is used to query the requirement document to verify whether the parameters contain the necessary constraints, i.e., requirement-parameter mapping verification.

[0049] S4: The association analysis layer checks for loop dependencies in the semantic knowledge graph, such as parameter A depending on B, and B depending on A. It also verifies that parameters conform to physical laws, such as the consistency of the formula "Turbine power = f(intake pressure, exhaust temperature)." It also invokes predefined constraint rules for system-level constraint checking and uses a conflicting rule auto-negotiation strategy to obtain a compromise solution. For example, if rule A requires an EGR rate ≥ 20% and rule B requires an EGR rate ≤ 15%, the compromise solution, EGR rate = 18%, is output and marked as "requires manual review."

[0050] S5. Based on the dynamic experience database, the dynamic assessment layer evaluates the risk probability of parameter values ​​and uses basic compliance, logic, system security and optimization potential for comprehensive scoring to achieve a case-driven adaptive and verification process.

[0051] The overall score = basic compliance × 0.3 + logical consistency × 0.25 + system security × 0.25 + optimization potential × 0.2. The optimization potential can be calculated by comparing the current parameter value with the historical optimal calibration data and calculating the deviation (Euclidean distance).

[0052] Therefore, the present invention adopts the above-mentioned method and system for intelligent verification of electronic control parameters based on multimodal AI fusion, which can overcome the problems of low verification efficiency, delayed rule update, insufficient ability to analyze complex parameter associations, etc. in the existing technology, and realize intelligent verification of electronic control parameters.

[0053] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. An intelligent verification system for electric control parameters based on multimodal AI fusion, characterized in that: include: The initial verification layer, based on the basic rule base, performs format compliance checks and value range boundary verification on the verification benchmark; The semantic understanding layer parses parameter description information based on the semantic understanding of the big model and verifies the requirement-parameter mapping through the fine-tuned domain big model; The association analysis layer verifies parameter dependencies and checks system-level constraints based on the semantic knowledge graph, and obtains a compromise solution through a conflict rule self-negotiation strategy. The dynamic assessment layer assesses risk probability based on a dynamic experience database and provides a comprehensive score based on basic compliance, logic, system security, and optimization potential, enabling case-driven adaptive verification. Furthermore, based on the aggregated verification feedback, the verification strategy evolution module uses reinforcement learning to update the corresponding strategy. The initial verification layer, semantic understanding layer, association analysis layer, and dynamic evaluation layer are connected in sequence to process the input calibration parameters and obtain corresponding security scores based on historical cases; and the evaluation results of the dynamic evaluation layer are used to update the rules of the initial verification layer.

2. The electronic control parameter intelligent verification system based on multimodal AI fusion according to claim 1 is characterized in that: The semantic knowledge graph includes building an ontology model based on parameter metadata, using a graph embedding algorithm to learn the association relationship between parameters, and dynamically updating the graph edge weights using actual verification result feedback.

3. The electronic control parameter intelligent verification system based on multimodal AI fusion according to claim 1 is characterized in that: The dynamic experience database collects historical calibration cases and builds a parameter-result mapping database.

4. The electronic control parameter intelligent verification system based on multimodal AI fusion according to claim 1 is characterized in that: The verification strategy evolution module includes extracting the target parameter set from the verification request, performing multi-hop queries in the semantic knowledge graph, obtaining implicit association rules, and predicting potential conflicts using graph neural networks.

5. The electronic control parameter intelligent verification system based on multimodal AI fusion according to claim 1 is characterized in that: The verification strategy evolution module includes generating new verification strategies through genetic algorithms based on the summary of verification feedback results.

6. An intelligent verification method for electric control parameters based on multimodal AI fusion, using an intelligent verification system for electric control parameters based on multimodal AI fusion according to any one of claims 1 to 5, characterized in that: The following steps are involved: S1. Normalize parameters in different formats, establish an enterprise-level parameter alias library, extract corresponding parameter features and associated features through parameter sensitivity analysis, and form parameter description information; S2. Perform compliance checks on parameter formats through the initial validation layer, verify whether data naming complies with standard naming conventions, and compare parameter values ​​with the enterprise-defined safety range. S3. Parse the parameter description information through the semantic understanding layer and input the parsed parameter description text into the fine-tuned domain model. Use the semantic knowledge graph to query the requirement document and verify whether the parameters contain the necessary constraints. S4. Check whether there are loop dependencies in the semantic knowledge graph through the association analysis layer, verify whether the parameters comply with physical laws, call predefined constraint rules to perform system-level constraint checks, and use the conflict rule self-negotiation strategy to obtain a compromise solution; S5. Based on the dynamic experience database, the dynamic assessment layer evaluates the risk probability of the calibration parameters, and uses basic compliance, logic, system security and optimization potential to perform a comprehensive score, and then implements adaptive parameter calibration and verification through case-driven methods.

7. The method for intelligent verification of electric control parameters based on multimodal AI fusion according to claim 6 is characterized in that: Step S1 includes: S11. Parse parameter files of different formats, extract parameter information of various formats, and perform standardization to unify the unit format; S12. Based on the normalized parameter information, extract the time series features, statistical features, and correlation features of the parameters.

8. The method for intelligent verification of electric control parameters based on multimodal AI fusion according to claim 6 is characterized in that: In step S3, the parameter description text is parsed by the large language model LLM.

9. The method for intelligent verification of electric control parameters based on multimodal AI fusion according to claim 6, characterized in that: In step S5, the optimization potential is calculated by comparing historical optimal calibration data to obtain the deviation of the current calibration parameter value.

10. The method for intelligent verification of electric control parameters based on multimodal AI fusion according to claim 6, characterized in that: In step S5 , the expression of the comprehensive score is: comprehensive score = basic compliance × 0.3 + logical consistency × 0.25 + system security × 0.25 + optimization potential × 0.2.

Images