A quantitative decision method for three-level evaluation system of electric vehicle fire safety

Abstract

This invention discloses a quantitative decision-making method for a three-level evaluation system for electric vehicle fire safety. The method includes: Step 1, constructing a three-level evaluation index system; Step 2, using multiple large language models to quantitatively score each of the three levels of indicators, obtaining an expert comprehensive scoring vector for each large language model; and Step 3, calculating the subjective weight vector W for each level of indicator based on the expert comprehensive scoring vector. (x) If the consistency ratio CR does not meet the consistency requirements, W is corrected using the power method iterative correction. (x) Step 4: Introduce the fire news gathering agent to generate a vehicle model deduction term ΔG and an accident cause amplification coefficient (1+δ) to obtain a dynamically corrected subjective weight vector system W′; Step 5: Calculate the comprehensive score S. final The system classifies fire risk levels and generates visual reports. This invention constructs a three-level indicator system and introduces an intelligent agent for fire news collection, significantly improving the system's timeliness and adaptability.

Description

Technical Field

[0001] This invention relates to the field of electric vehicle safety evaluation technology, and in particular to a quantitative decision-making method for a three-level fire safety evaluation system for electric vehicles. Background Technology

[0002] With the rapid popularization of new energy vehicles, especially electric vehicles, while they have made significant progress in power performance and energy efficiency, the risk of fire accidents has also gradually attracted widespread attention. Power batteries are prone to combustion or even explosion under extreme conditions such as thermal runaway, short circuits, and collisions, posing a serious threat to vehicle, personnel, and public safety. Existing fire safety assessment methods for electric vehicles generally have the following shortcomings:

[0003] Staticity and lag: Most evaluation systems are based on fixed indicator weights and historical accident data, lacking the ability to dynamically respond to emerging risk factors and failing to reflect the latest safety situation in a timely manner.

[0004] The opinions of experts are too singular: Traditional methods often rely on a small number of domain experts to subjectively assign weights, which means that the evaluation results are affected by individual experience and cognitive biases, and the stability and objectivity of the weighting results are insufficient.

[0005] Lack of utilization of public opinion data: The current system has a low utilization of unstructured data such as real-time accident reports and public opinion, and fails to incorporate external data sources such as news and announcements into weight adjustment and risk assessment.

[0006] Therefore, there is an urgent need for a quantitative decision-making method and system that can integrate expert opinions from multiple sources, dynamically adjust indicator weights, and reflect the level of vehicle fire risk in real time, so as to improve the scientificity, accuracy and timeliness of electric vehicle fire safety evaluation and provide reliable decision-making basis for regulatory agencies, vehicle manufacturers and users. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the existing electric vehicle fire safety evaluation system, which is static and has a strong lag. It proposes a quantitative decision-making method for a three-level evaluation system for electric vehicle fire safety. This quantitative decision-making method integrates artificial intelligence multi-expert modeling and dynamic public opinion for correction, so as to realize the scientific, accurate and real-time safety evaluation.

[0008] The technical solution adopted to achieve the purpose of this invention is:

[0009] A quantitative decision-making method for a three-level fire safety evaluation system for electric vehicles includes the following steps:

[0010] Step 1: Construct a three-level evaluation index system according to the target layer, criterion layer, and sub-criterion layer. The target layer is the comprehensive evaluation of electric vehicle safety, the criterion layer is the first-level index, and the sub-criterion layer is the second-level and third-level index.

[0011] Step 2: Select multiple large language models to simulate experts and quantify and score each indicator in the first, second and third level indicators in Step 1. Each large language model outputs a comprehensive score vector.

[0012] Step 3: Construct a quantitative decision-making model for the indicator evaluation system: Based on the comprehensive scoring vector of each major language model in Step 2, introduce the authority weight coefficient γ of each major language model. k By using a weighted geometric mean, the importance of any two indicators at each level of the large language model is compared pairwise to obtain the comprehensive judgment matrix element a. ij Based on a ij Construct a comprehensive judgment matrix A, and calculate the subjective weight vector W for each level of indicator based on A. (x) It then determines whether the consistency ratio CR meets the consistency requirements. If not, it uses a power-law iterative correction method to adjust W. (x) Thus, an initial weight vector system W that meets the consistency requirements is obtained;

[0013] Step 4: Introduce the fire news gathering agent to generate a vehicle model deduction term ΔG and an accident cause amplification coefficient (1+δ). Use the accident cause amplification coefficient (1+δ) to evaluate the matrix element a in the comprehensive judgment matrix A from Step 3. ij Dynamic correction is performed to obtain the dynamically corrected subjective weight vector system W′, and to generate the dynamic correction coefficient α=1-δ for the three-level indicator scores;

[0014] Step 5: Manually score each of the three-level indicators from Step 1 based on actual vehicles to obtain the actual vehicle scoring results. The third-level indicator score is calculated by combining the dynamic correction coefficient α obtained in step 4 with the third-level indicator score. The scores of the second-level and first-level indicators are calculated sequentially based on the third-level indicator scores and W′. Finally, the comprehensive score S is calculated by combining the vehicle model deduction item ΔG. final This serves as a comprehensive safety evaluation for electric vehicles.

[0015] In the above technical solution, the primary indicators mentioned in step 1 include battery safety, electrical safety, and mechanical safety.

[0016] In the above technical solution, the subjective weight vector W at each level mentioned in step 3 (x) Represented as:

[0017] W (x) =[w1,w2,…,w n ]

[0018] In the formula, w i Let be the subjective weight of index i, and n be the matrix order;

[0019] The subjective weight w of index i i The calculation formula is:

[0020]

[0021] In the formula, a ij These are the elements of the comprehensive judgment matrix for indicator i and indicator j;

[0022] The comprehensive judgment matrix element a of indicator i and indicator j ij The calculation formula is:

[0023]

[0024] In the formula, γ is the importance ratio of indicator i to indicator j for the k-th large language model. k Let n be the authority weight coefficient of the k-th large language model, and n be the matrix order.

[0025] In the above technical solution, the formula for calculating the consistency ratio CR in step 3 is as follows:

[0026]

[0027] In the formula, RI is the random consistency index for the corresponding order, and CI is the consistency index.

[0028] The formula for calculating the consistency index CI is as follows:

[0029]

[0030] In the formula, λ max The subjective weight vector W at each level (x) The largest eigenvalue, where n is the matrix order.

[0031] In the above technical solution, the fire news collection intelligent agent mentioned in step 4 is built based on natural language processing technology and includes a data acquisition module, a text preprocessing module, an information extraction module, and a weight correction module.

[0032] In the above technical solution, step 5 is based on the comprehensive score S final Classify electric vehicle fire risk levels and generate preset fire emergency suggestions, while also generating visual reports;

[0033] The electric vehicle fire risk level is divided into three levels: high risk, medium risk, and low risk.

[0034] High risk: Sfinal If the score is less than 60, the system will automatically generate fire emergency suggestions, including real-time monitoring, increasing insulation measures, and replacing high-risk components in advance.

[0035] Medium risk: 60≤S final For scores below 80, we provide general safety guidance and recommendations for regular inspections.

[0036] Low risk: S final If the score is ≥80, it is recommended to perform maintenance according to the regular maintenance cycle.

[0037] In the above technical solution, the comprehensive score S mentioned in step 5 final The calculation formula is:

[0038] S final =S base -ΔG

[0039] In the formula, S base ΔG represents the base score for the electric vehicle to be evaluated, with ΔG being the deduction for the vehicle model.

[0040] The formula for calculating the deduction ΔG for the vehicle model is as follows:

[0041] ΔG=c×(fT)

[0042] In the formula, f is the actual frequency, T is the threshold, and c is the penalty coefficient;

[0043] The baseline score S of the electric vehicle to be evaluated base The calculation formula is:

[0044]

[0045] In the formula, Scoring for each primary indicator, The weight of the primary indicator.

[0046] The scores of each primary indicator The calculation formula is:

[0047]

[0048] In the formula, This represents the weight of the secondary indicator relative to its primary indicator h. Scoring for each secondary indicator;

[0049] The scores of each secondary indicator The calculation formula is:

[0050]

[0051] In the formula, The scores for each of the three levels of indicators are as follows. Let be the weight of the m-th tertiary indicator relative to its secondary indicator l;

[0052] The revised scores of each of the three-level indicators The calculation formula is:

[0053]

[0054] In the formula, Let α be the actual score of the vehicle for the m-th tertiary indicator. m is the dynamic correction coefficient for the score of the m-th tertiary indicator.

[0055] In another aspect of the present invention, an electronic device includes: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the quantitative decision-making method for the three-level evaluation system of electric vehicle fire safety.

[0056] In another aspect, the invention stores computer-executable instructions, which, when executed, are used to implement the quantitative decision-making method for the three-level fire safety evaluation system of electric vehicles.

[0057] In another aspect of the present invention, a computer program product includes computer-executable instructions, which, when executed, are used to implement the quantitative decision-making method for the three-level evaluation system of electric vehicle fire safety.

[0058] Compared with the prior art, the beneficial effects of the present invention are:

[0059] 1. A multi-level indicator system and strict consistency control fill evaluation gaps and ensure evaluation stability. This invention constructs a three-level indicator system covering key dimensions such as battery safety, electrical safety, and mechanical safety. It innovatively incorporates forward-looking indicators such as early warning systems for thermal runaway, emergency response and protection, and redundant safety design into the evaluation framework, effectively filling several gaps in existing systems. A four-level hierarchical evaluation system is established for quantitative decision-making (target layer—criteria layer—sub-criteria layer—scheme layer). The geometric mean method is used to calculate the weight vectors of the first, second, and third-level indicators, and the consistency verification standard is tightened to CR < 0.08 to ensure the logical consistency and stability of the multi-level indicator system. The inconsistency matrix is ​​iteratively corrected using a power method to avoid distortion of evaluation results due to weight fluctuations, providing a stable and scientific benchmark framework for subsequent dynamic corrections.

[0060] 2. This invention introduces a multi-expert scoring system based on large language models. It selects the top ten high-performance models from authoritative rankings and combines a general knowledge base with a specialized knowledge base customized for the characteristics of each language model. Through prompt word engineering, it sets expert identities and tasks to achieve quantitative scoring of each of the three levels of indicators. This design ensures both the professionalism of the scoring and the authority of the data, and by recording the model versions, knowledge base content, and authority weight allocations for each language model, it effectively reduces the subjective bias and non-repeatability inherent in traditional expert weighting methods.

[0061] 3. A dynamic correction mechanism driven by public opinion enables real-time response to emerging risks. This invention designs a fire news collection intelligent agent based on natural language processing, which can automatically capture relevant news and accident reports daily, extract key information such as vehicle type, accident type, and cause of fire, form a multi-dimensional reporting frequency matrix, and make real-time corrections to the AHP weights according to the "two-layer transmission deduction mode". For high-frequency vehicle types, a vehicle type deduction item ΔG is generated; for high-frequency accident causes, the judgment matrix elements are dynamically adjusted and a correction coefficient α is generated, so that the comprehensive score can reflect the latest risk situation in a timely manner. This mechanism effectively overcomes the defects of long update cycles and strong lag in traditional evaluation systems, realizes daily updates of safety evaluations and immediate response to emergencies, and significantly improves the timeliness and adaptability of the system.

[0062] 4. Full-process automated calculation and executable risk output reduce labor costs and enhance practicality. This invention achieves end-to-end processing from indicator scoring and weight adjustment to comprehensive score and risk level output through automated calculation, completing large-scale data collection, analysis, and evaluation without human intervention. The final result not only includes quantified risk scores and classification results, but also generates fire emergency recommendations for different risk levels (such as thermal management improvement schemes for high-risk models, routine safety inspection guidelines, etc.). Compared with traditional methods that rely on manual weighting and offline aggregation, this invention improves efficiency by more than 30% and can be directly used by regulatory agencies, automakers, and consumers in structured formats (such as JSON, report documents), realizing the efficient transmission and practical application of safety assessment results and promoting the digital upgrade of safety management in the electric vehicle industry. Attached Figure Description

[0063] Figure 1 The diagram shows a quantitative decision-making method for the three-level evaluation system for electric vehicle fire safety of the present invention.

[0064] Figure 2 The diagram shows the process of scoring various indicators using a large language model in this invention. Detailed Implementation

[0065] The present invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0066] Example 1

[0067] A quantitative decision-making method for a three-level fire safety evaluation system for electric vehicles includes the following steps:

[0068] Step 1: Establish a three-tiered safety evaluation index system for electric vehicles to support the construction and subsequent calculation of a quantitative decision-making model for electric vehicle fire safety. This method combines fire accident characteristics, safety risk distribution patterns, national standards, and industry norms to form a structured, measurable, and operable multi-tiered index system.

[0069] 1.1 Overall Framework of the Indicator System

[0070] Reference Figure 1 A quantitative decision-making method for a three-level fire safety evaluation system for electric vehicles includes the following steps:

[0071] Step 1: Construct a three-level hierarchical evaluation index system, including:

[0072] Target layer: Comprehensive safety evaluation of electric vehicles;

[0073] Criteria Level (Level 1 Indicators): Includes three major categories: battery safety, electrical safety, and mechanical safety, covering the main risk sources of the powertrain, electrical system, and vehicle body structure in a fire scenario;

[0074] Sub-criteria layer (secondary and tertiary indicators): Primary indicators are further subdivided into secondary indicators, and each secondary indicator is further refined into multiple quantifiable tertiary indicators, with clear detection methods and judgment criteria.

[0075] 1.2 Setting of Primary Indicators

[0076] Battery safety: This encompasses the structural design, thermal management capabilities, protection functions, and anomaly response capabilities of the power battery system. Examples include: battery pack system safety; battery protection and anomaly response functions.

[0077] Electrical safety: Covers the insulation, protection, leakage current, and electrical protection of the vehicle's high-voltage system. For example: circuit design and insulation performance (insulation resistance ≥ the value specified by national standards, such as ≥100Ω / V as required by GB / T 18384); leakage current and short circuit protection (with real-time leakage current detection and automatic disconnection functions).

[0078] Mechanical safety: This covers the impact and vibration resistance of the vehicle body and related structures. For example: mechanical structure design (reasonable battery pack structure, reliability of mechanical connectors, etc.); thermal management and intelligent monitoring (post-collision thermal runaway early warning, fire emergency response capabilities).

[0079] 1.3 Refinement of Secondary and Tertiary Indicators

[0080] Under each primary indicator, several secondary indicators are set, and further refined into tertiary indicators to ensure that the indicators are quantifiable, measurable, and reproducible. For example: battery pack thermal safety characteristics: tertiary indicators include the timeliness of thermal alarm and the safety margin for aging and degradation; circuit design: tertiary indicators include the rationality of circuit layout and circuit fixation and protection; mechanical fault protection: tertiary indicators include vibration and shock protection, foreign object intrusion protection, and mechanical stress detection.

[0081] 1.4 Testing Methods and Judgment Criteria

[0082] To ensure data consistency and result comparability, standardized testing methods and judgment criteria are defined for each tertiary indicator:

[0083] Test methods: Specify the test equipment model (e.g., thermal imager resolution, sampling frequency), environmental conditions (ambient temperature, humidity), test procedures (heating rate, impact energy level), etc.

[0084] Judgment criteria: Refer to the qualified range specified in GB, ISO and other standards. For example, the insulation resistance should not be less than 100Ω / V; the thermal runaway peak temperature should be ≤800℃ to be judged as "excellent".

[0085] The electric vehicle safety three-level evaluation index system is referenced in Table 1-3.

[0086] Table 1 Battery Safety Evaluation Indicators

[0087]

[0088]

[0089]

[0090] Table 2 Electrical Safety Evaluation Indicators

[0091]

[0092] Table 3 Mechanical Safety Evaluation Indicators

[0093]

[0094]

[0095] Step 2 aims to utilize a large language model to simulate senior experts in the field of electric vehicle fire safety evaluation, quantifying and scoring each indicator in the primary, secondary, and tertiary levels of indicators from Step 1. To ensure the scientific rigor, professionalism, and traceability of the scoring, refer to... Figure 2 Taking a large language model as an example, the specific steps include:

[0096] 2.1 Expert Roles and Cue Keyword Construction. A core cue keyword framework was designed, assigning specific expert roles to each language model, such as "Chief Expert of the China Electric Vehicle Safety Committee" or "Specially Appointed Safety Assessment Consultant of a National Laboratory." Their task was clearly defined as: independently and unbiasedly, quantitatively scoring each of the Level 1, Level 2, and Level 3 indicators using a 1-10 point system. Necessary citation requirements were embedded in the cue keywords, ensuring that each language model's evaluation was based on authoritative information such as the Ministry of Industry and Information Technology's safety standards, laboratory testing data, historical accident cases, and technical white papers. The scoring reasons and confidence levels were provided simultaneously in the output for subsequent tracking and auditing.

[0097] Example of key keywords:

[0098] You are a senior expert with extensive experience in the field of new energy vehicle safety. You are now required to score the importance of each level of indicators in the "Electric Vehicle Fire Safety Three-Level Evaluation System". This evaluation system includes:

[0099] Primary indicators: battery safety, electrical safety, and mechanical safety;

[0100] Secondary indicators include battery pack system safety, circuit design, and insulation performance.

[0101] Level 3 indicators include early warning of thermal runaway, intelligent monitoring of mechanical stress, and fire emergency response.

[0102] Based on general and specialized knowledge bases, and combined with real-time online data (latest accident news, technology updates, and public opinion information), please score each primary, secondary, and tertiary indicator on a scale of 1-10. The score should reflect the importance of the indicator to overall fire safety; a higher score indicates greater importance. Additional requirements:

[0103] Ensure the scoring process is interpretable; each score must have a corresponding source of evidence.

[0104] Return "N / A" with an explanation when there is insufficient data;

[0105] Pay closer attention to recent hot-button risk events (such as the frequency of battery thermal runaway accidents in the past year);

[0106] Output should avoid subjective language, and data and reasoning should be based on facts.

[0107] 2.2 Selection and Authority Weight Allocation of Large Language Models. Referring to performance rankings of large language models (such as CompassArena), the top ten large models in terms of overall performance were selected (e.g., Qwen3-235B-A22B, Spark-X1, GPT-4o-20241120, etc.), and authority weight coefficients γ were assigned according to their rankings. k (First-ranked model has a weight of 0.182, last-ranked model has a weight of 0.018, and the middle models are distributed according to an arithmetic progression). The high-ranking models are assigned higher-level expert roles in the prompts to strengthen their guiding role in the overall scoring.

[0108] 2.3 External Knowledge Base Mechanism. To enhance the model's professional judgment capabilities in the field of electric vehicle fire safety, a two-layer external knowledge base mechanism is adopted, consisting of a general knowledge base and a specialized knowledge base.

[0109] General Knowledge Base: Adapted to all ten major language models, the content includes but is not limited to national and international electric vehicle safety standards (such as GB / T, ISO standards), annual industry technical white papers, industry annual reports, vehicle recall notices, publicly available fire accident statistics and cause analysis reports, standardized test method descriptions, etc., to ensure the consistency and authority of the major models in general domain information.

[0110] Dedicated knowledge base: Tailored to the context processing capabilities, information preferences, and reasoning strengths of different large-scale models, each model is equipped with differentiated, in-depth industry data. For example:

[0111] The accident case database (subdivided by vehicle model, year, and accident type) is adapted to large models with strong long context processing capabilities, so that more accident features can be associated in a single inference.

[0112] High-resolution laboratory test raw datasets (including temperature curves, gas concentration changes, structural stress data, etc.) are adapted to large models with strong numerical reasoning capabilities for refined index analysis.

[0113] The system includes records of policy and regulatory changes and revisions to technical standards, and is adapted to large models with strong legal and regulatory reasoning capabilities to evaluate regulatory compliance and risk assessment.

[0114] The data on accidents and recalls in overseas markets is compared and adapted to a large model with strong multilingual processing capabilities to provide international comparative reference.

[0115] The content of the dedicated knowledge base is obtained through public data interfaces of industry associations and regulatory authorities, laboratory test project results, academic papers and reports from third-party security evaluation agencies. Before being input into the large model, it is formatted and compressed to match the context length limit and optimal information density of the target large model.

[0116] 2.4 Parallel Execution and Result Aggregation. After loading their respective general and specialized knowledge bases, the ten selected large models independently complete the quantitative scoring of all three-level indicators based on the core prompts, and output the score, reason, and confidence level. To prevent "illusion" or unfounded inferences, the system sets an anti-illusion mechanism in the prompts. When a large model lacks sufficient evidence, it should output "N / A" and explain the reason for the lack of evidence in the reason.

[0117] 2.5 Scoring Result Processing. The system sequentially aggregates the scoring results of the 10 major models to form a complete "multi-model expert scoring result set," which consists of 10 independent expert model comprehensive scoring vectors. Each comprehensive scoring vector will serve as an important input for subsequent AHP judgment matrix construction and weight calculation, ensuring that the final weights and scoring system take into account expert experience, objective data, and model diversity in their sources.

[0118] Step 3: Based on the three-level evaluation index system constructed in Step 1, a quantitative decision-making model for the index evaluation system is built. The electric vehicle fire safety evaluation problem is decomposed into a four-level hierarchical structure: the target layer is "comprehensive evaluation of electric vehicle safety," the criterion layer is the first-level evaluation index (including battery safety, electrical safety, and mechanical safety), the sub-criterion layer is the second- and third-level detailed indexes under each first-level index, and the scheme layer is the initial weight vector system W constructed based on AHP. Specifically:

[0119] In step 2, the comprehensive scoring vectors of the 10 large language models for the evaluation index system established in step 1 were obtained; in this step, the authority weight coefficient γ of each model is further introduced. k The pairwise comparison results of 10 large models were fused using a weighted geometric average to construct a comprehensive judgment matrix A. The subjective weight vector W of each level of indicator was then calculated using the Analytic Hierarchy Process (AHP). (x) This serves as the basis for subsequent consistency checks and dynamic corrections.

[0120] First, a judgment matrix is ​​constructed between each layer using the 1–9 scaling method. The matrix elements reflect the relative importance ratio of two adjacent indicators under the target of the higher-level indicator. A judgment matrix is ​​constructed based on the pairwise comparison results of each large model. Let γ be the authority weight coefficient of the k-th large language model in step 2. k The importance ratio of indicator i to indicator j at each level is: Then, comprehensively judge the matrix element a ij The calculation is as follows:

[0121]

[0122] In obtaining the comprehensive judgment matrix A = (a) for each level of indicators ij ) n×nThen, the geometric mean method is used to calculate the subjective weight vector W for each level. (x) =[w1,w2,…,w n ], where x is the index level. The geometric mean method is less sensitive to extreme values ​​than the arithmetic mean method, and can significantly improve the stability of the weights.

[0123] Among them, the subjective weight w of index i i The calculation formula is:

[0124]

[0125] To ensure the logical consistency of the judgment matrix, the subjective weight vector W of the first-level, second-level, and third-level indicators is calculated respectively. (x) The largest eigenvalue λ max Consistency indicators CI and consistency ratio CR:

[0126]

[0127] Where n is the matrix order, and RI is the random consistency index for the corresponding order. Considering the characteristics of the four-level indicator system, this embodiment tightens the traditional standard of CR < 0.1 to CR < 0.08. If CR ≥ 0.08, the power method is used to adjust the subjective weight vector W. (x) Iterative corrections are performed until the subjective weight vector converges and satisfies the consistency requirement. The subjective weight vector W... (x) The iterative formula is:

[0128]

[0129] in Let be the subjective weight vector for the t-th iteration, and represent the normalization operation.

[0130] The final stable and consistent initial weight vector system W, consisting of weights of indicators at various levels, is obtained, wherein W = [W (1) W (2) W (3) ], where W (1) For the first-level indicator weight vector, W (2) For the secondary indicator weight vector, W (3) The weight vector of the three-level indicators is stored in the format shown in Table 4 below, and is used for subsequent dynamic public opinion correction and comprehensive score calculation.

[0131] Table 4. Storage format of the initial weight vector system W

[0132]

[0133] Step 4: After obtaining the initial weight vector system W of the electric vehicle fire safety evaluation system in Step 3, a fire news collection intelligent agent is introduced to realize the dynamic correction of the AHP subjective weight vector based on public opinion information, ensuring that the evaluation system can keenly capture emerging safety risks and realize the online monitoring and dynamic optimization of the electric vehicle fire risk evaluation system.

[0134] The fire news gathering intelligent agent is built based on natural language processing (NLP) technology and mainly consists of a data acquisition module, a text preprocessing module, an information extraction module, and a weight correction module.

[0135] The data acquisition module periodically accesses news portals, industry association announcement platforms, and accident reports issued by government regulatory departments through a preset web crawler script and open data interface (API) to capture unstructured raw text data related to electric vehicle fires. The acquisition frequency can be set by day or hour, for example, a full update can be performed once a day at 0:00 UTC+8 time zone.

[0136] The text preprocessing module performs noise reduction, word segmentation, stop word filtering, and format standardization on the original text from the data acquisition module.

[0137] The information extraction module combines text classification and named entity recognition (NER) technologies to automatically extract key information such as "vehicle model name," "accident type," and "cause of fire" from the text processed by the text preprocessing module, and stores it in the database according to dimensions such as date, source, and keywords. Based on the extraction results, the system generates a report frequency matrix F. pq :

[0138] F pq =Frequency of reports on vehicle model p under accident cause q

[0139] The weight correction module adopts a two-layer transmission deduction mode of "global vehicle risk - specific cause risk":

[0140] Global vehicle model risk correction: Based on the report frequency matrix F generated by the information extraction module pq The frequency of a certain model's reports within a preset time window is obtained. When the frequency of a certain model's reports within the preset time window exceeds a threshold (the threshold is set to 5 times in this embodiment), a deduction item ΔG is generated for the model. The value can be calculated as ΔG = c × (fT), where f is the actual frequency, T is the threshold, and c is the penalty coefficient (e.g., 1.0 point / time).

[0141] Specific Cause Risk Correction: When the frequency of a certain type of accident cause exceeds a threshold (e.g., 20%), the importance of the relevant tertiary indicator in the AHP judgment matrix is ​​increased by an accident cause amplification factor (1+δ) (e.g., δ is the accident cause amplification ratio, δ=0.1). This means the importance of the corresponding matrix element a is adjusted accordingly.ij Multiply by (1+δ) and recalculate the weight element w. i A consistency test was performed (CR < 0.08) to obtain the subjective weight vector W of each level of indicators after dynamic correction. (x) This leads to the dynamically corrected subjective weight vector system W′. For the affected third-level indicators, a dynamic correction coefficient α = 1 - δ is generated based on the amplification ratio δ of the accident cause, which is used for subsequent score calculation.

[0142] Replace the initial subjective weight vector system W in step 3 with the dynamically corrected subjective weight vector system W′ to ensure that the comprehensive risk assessment results reflect the latest safety situation in real time, and store them in the database together with the dynamic correction coefficient α of the three-level indicator scores and the vehicle deduction item ΔG.

[0143] Step 5: In step 4, the weight vectors W of each level of indicators, dynamically corrected by incorporating public opinion information, are obtained. (x) ′、Dynamic correction coefficients α for each relevant tertiary indicator m Based on the vehicle model deduction ΔG, and according to the evaluation content of each indicator in Table 1-3, each third-level indicator in step 1 is manually scored based on the actual vehicle to obtain the actual vehicle score. Based on the actual vehicle scoring results The comprehensive risk score and risk level are calculated and output for the vehicles to be evaluated.

[0144] 5.1 Indicator Score Calculation:

[0145] First, calculate the revised scores for each of the three levels of indicators.

[0146]

[0147] in, Let α be the actual score (0-100 points) of the m-th tertiary indicator for the vehicle. m is the dynamic correction coefficient for the m-th tertiary indicator.

[0148] Secondly, calculate the scores for each secondary indicator.

[0149]

[0150] In the formula, Let be the weight of the m-th tertiary indicator relative to its secondary indicator l.

[0151] Finally, the scores for each primary indicator are calculated.

[0152]

[0153] in, The weight of the secondary indicator relative to its primary indicator h.

[0154] 5.2 Base Score and Final Score:

[0155] Calculate the baseline score S of the electric vehicle to be evaluated base :

[0156]

[0157] in, The weight of the primary indicator.

[0158] Overall Score S final The calculation formula is:

[0159] S final =S base -ΔG

[0160] Among them, ΔG is the deduction item for the vehicle model, which is output from step 4.

[0161] 5.3 Risk Level Classification and Result Output:

[0162] The system is based on the comprehensive score S final Vehicles are classified into three levels: high-risk, medium-risk, and low-risk. Among them:

[0163] High risk: S final If the score is less than 60, the system will automatically generate fire emergency suggestions, including real-time monitoring, increasing insulation measures, and replacing high-risk components in advance.

[0164] Medium risk: 60≤S final For scores below 80, we provide general safety guidance and recommendations for regular inspections.

[0165] Low risk: S final If the score is ≥80, it is recommended to perform maintenance according to the regular maintenance cycle.

[0166] 5.4 Data Storage and Visualization:

[0167] Final score S final The scores, weighting coefficients, risk levels, and recommendations for each indicator are stored in the database in the form of structured records. An example table structure is shown in Table 5 below:

[0168] Table 5 Final Score S final Scores, weighting coefficients, risk levels, and suggested storage formats for each indicator.

[0169]

[0170]

[0171] The system generates visual reports (bar charts, radar charts, etc.) based on the final score, scores of each indicator, weighting coefficients, risk level, and recommendations, and provides PDF / HTML export functionality for regulatory agencies, manufacturing enterprises, and users to conduct risk analysis and decision-making references.

[0172] Example 2

[0173] An electronic device includes: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement a quantitative decision-making method for a three-level fire safety evaluation system for electric vehicles.

[0174] Example 3

[0175] A computer-readable storage medium storing computer-executable instructions, which, when executed, are used to implement a quantitative decision-making method for a three-level fire safety evaluation system for electric vehicles.

[0176] Example 4

[0177] A computer program product comprising computer-executable instructions, which, when executed, are used to implement a quantitative decision-making method for a three-level fire safety evaluation system for electric vehicles.

[0178] The above description is only a preferred embodiment of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A quantitative decision-making method for a three-level evaluation system of electric vehicle fire safety, characterized in that, Includes the following steps: Step 1: Construct a three-level evaluation index system according to the target layer, criterion layer, and sub-criterion layer. The target layer is the comprehensive evaluation of electric vehicle safety, the criterion layer is the first-level index, and the sub-criterion layer is the second-level and third-level index. Step 2: Select multiple large language models to simulate experts and quantify and score each indicator in the first, second and third level indicators in Step 1. Each large language model outputs a comprehensive score vector. Step 3: Construct a quantitative decision-making model for the indicator evaluation system: Based on the comprehensive scoring vector of each major language model in Step 2, introduce the authority weight coefficients of each major language model. By using a weighted geometric mean, the importance of any two indicators at each level of the large language model is compared pairwise to obtain the elements of the comprehensive judgment matrix. ,based on Construct a comprehensive judgment matrix ,based on Calculate the subjective weight vector of each level of indicator And determine the consistency ratio. CR If the consistency requirement is not met, then iterative correction using the power method is employed. This yields an initial weight vector system that meets the consistency requirements. ; Step 4: Introduce a fire news gathering AI agent to generate vehicle model deduction criteria. and the amplification factor of the cause of the accident 1+δ ,in, δ The amplification factor for the cause of the accident is used to represent the amplification ratio. 1+δ For the comprehensive judgment matrix in step 3 Matrix elements Dynamic correction yields the dynamically corrected subjective weight vector system. And generate dynamic correction coefficients for the three-level indicator scores. α= 1-δ; Step 5: Manually score each of the three-level indicators from Step 1 based on actual vehicles to obtain the actual vehicle scoring results. Combined with the dynamic correction coefficients of the three-level indicator scores obtained in step 4 α The scores of the three-level indicators are calculated, and the scores of the three-level indicators are combined with... Calculate the scores for the secondary and primary indicators sequentially, and then combine them with the deductions for vehicle models. Calculate the overall score This serves as a comprehensive safety evaluation for electric vehicles.

2. The quantitative decision-making method for the three-level evaluation system of electric vehicle fire safety as described in claim 1, characterized in that, The primary indicators mentioned in step 1 include battery safety, electrical safety, and mechanical safety.

3. The quantitative decision-making method for the three-level fire safety evaluation system for electric vehicles as described in claim 1, characterized in that, The subjective weight vectors at each level mentioned in step 3 Represented as: In the formula, As an indicator i Subjective weighting, n The order of the matrix; The indicators i Subjective weight The calculation formula is: In the formula, As an indicator i With indicators j The comprehensive judgment matrix elements; The indicators i With indicators j Comprehensive judgment matrix elements The calculation formula is: In the formula, For the first k Large language models for indicators i With indicators j Importance ratio For the first k The authority weight coefficients of each large language model n is the order of the matrix.

4. The quantitative decision-making method for the three-level evaluation system of electric vehicle fire safety as described in claim 1, characterized in that, The consistency ratio mentioned in step 3 CR The calculation formula is: In the formula, RI For the corresponding order of random consistency index, CI As a consistency indicator; The consistency index CI The calculation formula is: In the formula, Subjective weight vectors at each level The largest eigenvalue, n is the order of the matrix.

5. The quantitative decision-making method for the three-level fire safety evaluation system for electric vehicles as described in claim 1, characterized in that, The fire news gathering agent described in step 4 is built based on natural language processing technology and includes a data acquisition module, a text preprocessing module, an information extraction module, and a weight correction module.

6. The quantitative decision-making method for the three-level evaluation system of electric vehicle fire safety as described in claim 1, characterized in that, In step 5, based on the overall score Classify electric vehicle fire risk levels and generate preset fire emergency suggestions, while also generating visual reports; The electric vehicle fire risk level is divided into three levels: high risk, medium risk, and low risk. High risk: S final At 60 points, the system automatically generates fire emergency recommendations, including real-time monitoring, increasing insulation measures, and replacing high-risk components in advance. Medium risk: 60≤ S final 80 points, providing routine safety guidance and recommendations for regular inspections; Low risk: S final If the score is ≥80, it is recommended to perform maintenance according to the regular maintenance cycle.

7. The quantitative decision-making method for the three-level fire safety evaluation system for electric vehicles as described in claim 1, characterized in that, The comprehensive score mentioned in step 5 The calculation formula is: In the formula, This is the baseline score for the electric vehicle to be evaluated. This is a negative factor for the vehicle model. The vehicle model's deduction items The calculation formula is: In the formula, f This refers to the actual frequency. T For the threshold, c This is the penalty coefficient; The basic score of the electric vehicle to be evaluated The calculation formula is: In the formula, Scoring for each primary indicator, Weights for primary indicators; The scores of each primary indicator The calculation formula is: In the formula, Secondary indicators are relative to their primary indicators. h The weight, Scoring for each secondary indicator; The scores of each secondary indicator The calculation formula is: In the formula, The revised scores for each of the three-level indicators. For the first m Each tertiary indicator relative to its tertiary indicator l The weights; The revised scores of each of the three-level indicators The calculation formula is: In the formula, , For the first m The dynamic correction coefficient for the scores of each tertiary indicator.

8. An electronic device, characterized in that, include: One or more processors; A memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the quantitative decision-making method for the three-level evaluation system of electric vehicle fire safety as described in claim 1.

9. A computer-readable storage medium, characterized in that, It stores computer-executable instructions, which, when executed, are used to implement the quantitative decision-making method of the three-level evaluation system for electric vehicle fire safety as described in claim 1.

10. A computer program product, characterized in that, The aforementioned computer program product includes computer-executable instructions, which, when executed, are used to implement the quantitative decision-making method for the three-level evaluation system of electric vehicle fire safety as described in claim 1.

Images