Electric vehicle transportation fire cause analysis method based on coupling model

Abstract

The invention discloses an electric vehicle transportation fire cause analysis method based on a coupling model, and the method comprises the steps: collecting a fire scene ablation form, combustion product data and a transportation condition log, and carrying out the fusion to obtain a scene feature data set; a convolutional neural network is used for extracting features to determine a fire development path, vehicle attributes are matched when thermal diffusion is abnormal, electrochemical reaction data are integrated, and a multi-physical-field initial coupling model is constructed; vibration and temperature influences are simulated through finite element analysis, a thermal runaway trigger point is positioned, a refining cause mechanism is obtained through a reverse mapping calibration model, if the transportation working condition is a dominant factor, a preventive threshold value is determined, and the method is applied to real-time monitoring and fusion model output to achieve risk dynamic assessment. The method can accurately position a trigger point, analyzes a multi-physical field disaster-causing mechanism, improves the cause analysis precision and the fire prevention rate, and is suitable for multi-scene electric vehicle transportation safety management and control.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of information technology, and particularly relates to a fire cause analysis method for electric vehicle transportation based on a coupling model. BACKGROUND

[0002] With the rapid development of the new energy vehicle industry, the frequency of fire accidents of electric vehicles during transportation has significantly increased, which has become a major hidden danger threatening traffic safety and the development of the logistics industry. Electric vehicle fires have characteristics such as high burning temperature, difficulty in extinguishing, and release of toxic gases. Once a fire occurs during transportation, it often causes serious casualties and property losses.

[0003] Current analysis methods for electric vehicle transportation fires mainly rely on traditional fire investigation techniques. These methods have obvious shortcomings in dealing with the complex combustion mechanism specific to electric vehicles. Existing technologies often analyze fire identification separately from vehicle design parameters, and cannot accurately identify the characteristic traces left by vehicle inherent properties such as battery energy density and thermal diffusion characteristics at the fire scene, resulting in a lack of scientific basis for determining the cause of the accident. At the same time, traditional methods ignore the key influence of transportation conditions on fire triggering and do not fully consider how transportation environmental factors such as vibration, temperature changes, and other factors interact with the electrochemical reactions inside the vehicle.

[0004] This analysis limitation brings two core technical problems. There is a complex relationship between the ablation morphology, combustion product distribution at the fire scene, and vehicle design and manufacturing parameters, but there is a lack of effective reverse analysis means to accurately infer the fire starting location and development process from the fire residues. More critically, mechanical stress during transportation can cause changes in the internal structure of the battery, which in turn affects the thermal stress distribution and electrochemical reaction process. The synergistic mechanism of this multi-physical field has not been thoroughly analyzed. For example, when a transportation vehicle is driving on a bumpy road, the impact on the battery pack can cause micro-damage to the separator, and subsequent temperature fluctuations can accelerate the decomposition of the electrolyte, ultimately triggering thermal runaway under certain conditions, but existing technologies cannot accurately capture this gradual disaster-causing process.

[0005] Therefore, how to establish the coupling relationship between the characteristics of the fire scene and the inherent parameters of the vehicle, and accurately analyze the synergistic mechanism of multi-physical field factors under transportation conditions, has become a key problem for the scientific identification and effective prevention of electric vehicle transportation fires. SUMMARY

[0006] To solve the above technical problems, the present application proposes a fire cause analysis method for electric vehicle transportation based on a coupling model to solve the problems existing in the prior art.

[0007] To achieve the above purpose, the present application provides a fire cause analysis method for electric vehicle transportation based on a coupling model, comprising: Collecting ablation morphology, combustion product data and residue spatial distribution of electric vehicle transportation fire scene, fusing vibration and temperature change logs in transportation working condition record to obtain scene characteristic data set; According to the scene characteristic data set, the texture pattern of ablation morphology and the chemical composition distribution characteristics of combustion products are extracted, and the preliminary outline of fire development path is determined; if the preliminary outline shows abnormal heat diffusion, the corresponding vehicle inherent attribute is matched and the electrochemical reaction simulation data is integrated to obtain a multi-physical field initial coupling model; The multi-physical field initial coupling model is processed by finite element analysis to simulate the impact of vibration and jolt on the internal structure of the battery and the amplification effect of temperature change on thermal stress, and to determine the location of the thermal runaway trigger point; According to the thermal runaway trigger point, the ablation morphology in the scene characteristic data set is reversely mapped, and an iterative optimization algorithm is used to calibrate the coupling model parameters to obtain a refined fire cause mechanism description; if the refined fire cause mechanism description includes the dominant factor of transportation working condition, the vibration and temperature change mode of similar cases is obtained from the historical accident database to determine the preventive threshold setting; The preventive threshold setting is applied to the real-time transportation monitoring system, and the output of the multi-physical field coupling model is fused to obtain the dynamic evaluation result of electric vehicle transportation fire risk.

[0008] Optionally, the process of obtaining the scene characteristic data set includes: High-resolution scanning of electric vehicle transportation fire scene is performed using an image scanning device to obtain residue spatial distribution information of ablation morphology and combustion products, and a first feature image is generated; The vibration frequency and temperature change amplitude data are extracted from the electric vehicle transportation working condition record, and a time series analysis method is used to extract features of the vibration frequency and temperature fluctuation to obtain a working condition feature vector; According to the resolution parameters of the first feature image, a data fusion algorithm is used to perform multi-modal feature alignment processing on the first feature image and the working condition feature vector to generate a second feature data set; According to the second feature data set, a convolutional neural network algorithm is used to classify and identify the spatial distribution characteristics of the ablation morphology and combustion products to determine the characteristic category attribute of the fire scene, and a scene characteristic data set is obtained.

[0009] Optionally, according to the scene characteristic data set, the process of extracting the texture pattern of ablation morphology and the chemical composition distribution characteristics of combustion products to determine the preliminary outline of fire development path includes: The data preprocessing method is used to standardize the field feature data set to obtain a preprocessed data set; the convolutional neural network is used to extract features from the preprocessed data set to generate an ablation morphology texture feature set and a chemical composition distribution feature set; the spatial distribution analysis method is used to process the ablation morphology texture feature set to obtain a spatial distribution pattern of the texture; the combustion process simulation method is used to analyze the chemical composition distribution feature set to generate a dynamic distribution feature of combustion products; if the spatial overlap degree of the spatial distribution pattern of the texture and the dynamic distribution feature of the combustion products is higher than a preset threshold, the path prediction model is used to fuse the two types of features to generate a preliminary contour of the fire development path.

[0010] Optionally, the process of obtaining the multi-physics initial coupling model comprises: If the thermal diffusion anomaly is detected, a parameter set matching the inherent properties of the vehicle is obtained from the battery density parameter library to obtain an initial battery density parameter; the reaction rate distribution is obtained by simulating the electrochemical reaction process using the finite element analysis method according to the initial battery density parameter and in combination with the vehicle operating state; if the reaction rate distribution exceeds a preset threshold, the electrochemical reaction parameters are optimized by adjusting the battery material properties to obtain an optimized reaction rate distribution; the coupling characteristics of thermal diffusion and electrochemical reaction are obtained by constructing a multi-physics model according to the optimized reaction rate distribution and in combination with the thermal conduction characteristics; if the thermal diffusion value in the coupling characteristics exceeds a safety threshold, the model parameters are optimized using a gradient descent algorithm to obtain an adjusted multi-physics model; the influence distribution of the abnormality is obtained by analyzing the influence of the thermal diffusion anomaly on the vehicle operating state through the adjusted multi-physics model; and the parameters of the initial coupling model are updated according to the influence distribution of the abnormality to obtain an optimized multi-physics initial coupling model.

[0011] Optionally, the process of determining the position of the thermal runaway trigger point comprises: The multi-physics initial coupling model is obtained by finite element analysis, the model is input with vibration bump data to obtain a first stress distribution; the first stress distribution is superimposed with temperature change data to obtain a second stress distribution; if the second stress distribution exceeds a preset threshold, the structure impact position is determined to obtain an impact area; the impact area is processed using finite element analysis and superimposed with an amplification effect to obtain a third stress distribution; the thermal stress concentration area is obtained by analyzing the third stress distribution through a convolutional neural network; and if the energy density of the thermal stress concentration area exceeds a preset threshold, the position of the thermal runaway trigger point is determined.

[0012] Optionally, the process of calibrating the coupling model parameters using an iterative optimization algorithm to obtain a refined fire cause mechanism description according to the ablation morphology of the thermal runaway trigger point in the field feature data set comprises: The preset mapping rule is obtained according to the thermal runaway trigger point, and the matching degree of the mapping rule and the field ablation morphology data set is determined. If the matching degree is higher than a preset threshold, a preliminary reverse mapping path is determined, and ablation morphology data association is extracted; The ablation morphology data association is reduced in dimension by using a principal component analysis algorithm to obtain a reduced dimension morphology data set; An initial coupling model parameter is constructed according to the reduced dimension morphology data set, and a deviation of the initial coupling model parameter from a thermal runaway trigger point is judged; If the deviation exceeds a preset threshold, the initial coupling model parameter is adjusted by an iterative optimization algorithm to obtain a calibrated coupling model parameter; The reverse mapping path is calculated according to the calibrated coupling model parameter, and the ablation morphology data set is integrated to generate a fire cause mechanism description; If the completeness of the fire cause mechanism description is lower than a preset threshold, the principal component analysis algorithm is used for optimization to obtain a refined fire cause mechanism description.

[0013] Optionally, the process of determining the preventive threshold setting comprises: Similar case original vibration patterns are obtained from a historical accident database; the similar case original vibration patterns are clustered by a K-nearest neighbor algorithm to obtain clustered vibration patterns; if the standard deviation of the clustered vibration patterns exceeds a preset threshold, an abnormal vibration pattern is judged to obtain the abnormal vibration pattern; a jolt pattern is extracted from the abnormal vibration pattern to obtain the extracted jolt pattern; the jolt pattern is reduced in dimension by principal component analysis to obtain a reduced dimension jolt pattern; a temperature change sequence is calculated according to the reduced dimension jolt pattern to obtain a temperature change sequence; if the slope of the temperature change sequence exceeds a preset threshold, a high-risk temperature sequence is judged to obtain the high-risk temperature sequence; the reduced dimension jolt pattern is fused with the high-risk temperature sequence to obtain a fused risk pattern; the fused risk pattern is classified by a random forest algorithm to obtain a classified risk pattern; a preventive threshold is determined according to the classified risk pattern to obtain a transportation preventive threshold; if the transportation preventive threshold is lower than a preset lower limit, the transportation preventive threshold is adjusted to obtain a final preventive threshold.

[0014] Optionally, the process of applying the preventive threshold setting to a real-time transportation monitoring system, fusing the output of a multi-physical field coupling model, and obtaining a dynamic evaluation result of an electric vehicle transportation fire risk comprises: Sensor data in the real-time transportation monitoring system is obtained; the output of the multi-physical field coupling model is calculated according to the sensor data; if the output of the multi-physical field coupling model exceeds a preset threshold, an initial state of a fire risk is judged; a preliminary risk distribution is obtained by fusing transportation system data with the initial state of the fire risk; the preliminary risk distribution is processed by a long short-term memory network to determine a time series risk change; if the time series risk change continues to rise, a risk level is classified by a support vector machine; the dynamic evaluation result is obtained by fusing electric vehicle location data according to the risk level.

[0015] Compared with the prior art, the present invention has the following advantages and technical effects: This invention collects data on ablation morphology and combustion products at fire scenes using image scanning equipment. It integrates vibration and temperature change logs from transportation conditions to construct a scene feature dataset. A convolutional neural network is then used to extract texture patterns and chemical composition distribution features to determine the fire development path. When an abnormal thermal diffusion is detected, vehicle attributes are matched with a battery density parameter library to establish a multiphysics coupling model. Finite element analysis is used to simulate the impact of vibration, shock, and temperature stress on the battery's internal structure, accurately determining the location of the thermal runaway trigger point. Based on the trigger point, the model parameters are calibrated through inverse mapping to obtain a refined fire causation mechanism. Combined with a historical accident database, preventative thresholds are determined and applied to real-time transportation monitoring. This addresses the complex scenario of electric vehicles experiencing battery thermal runaway and subsequent fires during transportation due to vibration, shock, and temperature changes, enabling dynamic assessment and proactive prevention of fire risks during electric vehicle transportation. Attached Figure Description

[0016] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart of a method according to an embodiment of the present invention. Detailed Implementation

[0017] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0018] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0019] Example 1 like Figure 1 As shown, this embodiment provides a method for analyzing the causation of electric vehicle transportation fires based on a coupled model, including: Step S101: By collecting data on the ablation morphology and combustion products at the electric vehicle transportation fire scene, using image scanning equipment to capture the spatial distribution of residues, and integrating vibration and temperature change logs from the transportation condition records, a scene feature dataset is obtained.

[0020] The image scanning device is used to scan the electric vehicle transportation fire scene at high resolution to obtain the ablation morphology and the spatial distribution information of the residues of combustion products, and generate a first feature image; vibration frequency and temperature change amplitude data are extracted from the electric vehicle transportation working condition record, and a time series analysis method is used to extract features of the vibration frequency and temperature fluctuation to obtain a working condition feature vector; according to the resolution parameters of the first feature image, a data fusion algorithm is used to perform multi-modal feature alignment processing on the first feature image and the working condition feature vector to generate a second feature data set; according to the second feature data set, a convolutional neural network algorithm is used to classify and identify the spatial distribution features of the ablation morphology and the combustion products, judge the characteristic category attribute of the fire scene, and obtain a scene feature data set.

[0021] First, the spatial feature data of the scene is collected. A high-resolution image scanning device (such as an industrial-grade three-dimensional scanner, a high-definition camera array) is used to scan the electric vehicle transportation fire scene comprehensively, focusing on capturing the spatial distribution information of the ablation morphology (such as the depth of the ablation marks on the vehicle body frame and the battery pack shell, the damage morphology, and the residual structure contour) and the combustion products (such as electrolyte residue stains, metal oxide powder, and plastic combustion residues). The scanning process needs to cover all areas affected by the fire, including the vehicle interior battery compartment, the external vehicle body, and the surrounding ground residues, and finally generate a first feature image containing spatial coordinates, material morphology, color gray scale, and other information, providing an intuitive basis for the subsequent analysis of the scene space.

[0022] Second, the time sequence features of the transportation working condition are extracted. From the vehicle monitoring system or transportation management platform of the electric vehicle, the transportation working condition record of a period of time before the fire (such as 24 hours before the fire) is retrieved, and the vibration frequency (such as the number of jolts per unit time, the peak value of jolt acceleration) and the temperature change amplitude (such as the fluctuation range of the battery compartment temperature and the ambient temperature, the heating / cooling rate) data are selected. A time series analysis method (such as sliding window smoothing method, trend decomposition method) is used to preprocess these dynamic data, remove device noise interference, extract key time sequence features (such as vibration frequency mean, temperature fluctuation variance, extreme value occurrence time point), and convert them into structured working condition feature vectors to realize the quantitative expression of the transportation working condition data.

[0023] Then, multi-modal data fusion alignment is performed. According to the resolution parameters (such as pixel size, spatial sampling interval) of the first feature image, a multi-modal data fusion algorithm (such as the canonical correlation analysis algorithm of feature-level fusion) is used to establish the mapping relationship between the spatial dimension of the first feature image and the time dimension of the working condition feature vector. For example, the vibration and temperature data in a specific time period are associated with the possible ablation mark positions of the battery pack in that time period. Through coordinate calibration and time stamp matching, multi-modal alignment of spatial features and time sequence features is completed, generating a second feature data set that contains both the physical form and the dynamic changes of the working conditions, and solving the dimension difference problem of different types of data.

[0024] Finally, feature classification and data set determination are completed. Based on the second feature data set, a convolutional neural network algorithm is used to classify and identify the spatial distribution features of the ablation form and combustion products. The network extracts feature details through the convolution layer, compresses the data dimension through the pooling layer, and realizes classification through the full connection layer to determine the feature category attributes of the fire scene (such as local battery fire type and overall spread type, and electrolyte dominant type and plastic dominant type). The classification results are integrated with the original feature data to form a complete and clear on-site feature data set, ensuring that the data can accurately support subsequent fire development path analysis.

[0025] In step S102, according to the on-site feature data set, a convolutional neural network is used to extract the texture pattern of the ablation form and the chemical composition distribution feature of the combustion products, and a preliminary outline of the fire development path is determined.

[0026] The on-site feature data set is standardized by using a data preprocessing method to obtain a preprocessed data set; the preprocessed data set is feature-extracted by using a convolutional neural network to generate an ablation form texture feature set and a chemical composition distribution feature set; the ablation form texture feature set is processed by a spatial distribution analysis method to obtain a spatial distribution pattern of the texture; the chemical composition distribution feature set is analyzed by a combustion process simulation method to generate dynamic distribution features of the combustion products; if the spatial overlap degree of the spatial distribution pattern of the texture and the dynamic distribution features of the combustion products is higher than a preset threshold, a path prediction model is used to fuse the two types of features to generate a preliminary outline of the fire development path; the preliminary outline is adjusted by an iterative optimization method to obtain an optimized outline of the fire development path; and the spatiotemporal distribution features of the fire development path are generated according to the optimized outline to determine the final fire path outline.

[0027] Preprocess the field feature data. Due to the problems of inconsistent dimensions (e.g., spatial size units in millimeters, chemical composition concentration units in milligrams per cubic meter), data missing (e.g., feature missing due to local area scanning blind area), or outliers (e.g., incorrect temperature data generated by equipment failure), data preprocessing methods are used for standardization processing. Through Z-score standardization, the data is mapped to a unified numerical range (mean value of 0, standard deviation of 1), linear interpolation method is used to supplement the missing data, and 3σ principle is used to eliminate abnormal values beyond the normal range, to obtain a preprocessed data set with uniform data distribution and high reliability, avoiding the influence of data quality problems on the accuracy of subsequent feature extraction.

[0028] Extract the key feature set. The preprocessed data set is input into the convolutional neural network, and the network extracts two types of core features through different size convolution kernels (e.g., 3x3, 5x5): one is the texture feature of ablation morphology (e.g., the direction, density, and edge profile smoothness of ablation marks), and the other is the chemical composition distribution feature of combustion products (e.g., the concentration gradient, distribution range, and peak position of specific chemical components). After multiple convolution and pooling operations, the key features reflecting the nature of the fire are selected, generating structured ablation morphology texture feature set and chemical composition distribution feature set, realizing the transformation from raw data to high-dimensional features.

[0029] Analyze the spatial and dynamic properties of the features. The spatial distribution analysis method (e.g., spatial autocorrelation analysis, K-means clustering analysis) is used for the ablation morphology texture feature set to calculate the distribution density and aggregation pattern of the texture features in different regions, and determine the spatial distribution pattern of the texture. This pattern can directly reflect the spreading direction (e.g., from the battery compartment to the front of the vehicle body) and concentration area (e.g., the center position of the battery pack) of the fire in space. The combustion process simulation method (e.g., CHEMKIN simulation model based on chemical kinetics) is used for the chemical composition distribution feature set, combined with the combustion reaction rate and product diffusion law, to simulate the generation and diffusion process of combustion products over time, generating dynamic distribution features of combustion products, and reflecting the time dimension changes of fire development (e.g., the spatial distribution of carbon dioxide concentration at different time points).

[0030] The fusion feature generates a preliminary contour. The spatial distribution pattern of the texture and the dynamic distribution feature of the combustion product are calculated. The spatial overlap degree of the two types of features is measured by the IoU index. If the overlap degree is higher than the preset threshold (such as 80%), it indicates that the fire information reflected by the two types of features is highly consistent. At this time, the path prediction model (such as the path prediction model based on the time series convolution network) is used to fuse the two types of features, combine the spatial spreading direction and the time dynamic change, and generate the preliminary contour of the fire development path. The contour needs to clearly mark the core information such as the fire starting area, the main spreading path (such as from the battery cell to the battery module to the whole package), and the key diffusion node (such as the fire acceleration point caused by electrolyte leakage).

[0031] The optimized contour determines the final path. An iterative optimization method (such as the gradient descent iteration method) is used to adjust the preliminary contour, taking the measured data such as the actual residual distribution, the combustion product concentration peak position, and the vehicle body ablation depth as the reference, to continuously correct the spatial coordinates (such as adjusting the angle deviation of the spreading direction) and time nodes (such as correcting the time when the fire reaches a certain part of the vehicle body) of the path contour, and reduce the model prediction deviation. According to the optimized contour, the spatiotemporal distribution feature of the fire development path (such as the spatial range covered by the fire every 10 minutes) is further generated, and the final fire path contour that accurately reflects the fire propagation process is finally determined.

[0032] In step S103, if the preliminary contour shows thermal diffusion anomaly, the corresponding vehicle inherent attribute is matched from the battery density parameter library, and the electrochemical reaction simulation data is integrated to obtain a multi-physical field initial coupling model.

[0033] If thermal diffusion anomaly is detected, the parameter set matched with the vehicle inherent attribute is obtained from the battery density parameter library to obtain the initial battery density parameter. According to the initial battery density parameter, combined with the vehicle running state, the finite element analysis method is used to simulate the electrochemical reaction process to obtain the reaction rate distribution. If the reaction rate distribution exceeds the preset threshold, the electrochemical reaction parameters are optimized by adjusting the battery material properties to obtain the optimized reaction rate distribution. According to the optimized reaction rate distribution, combined with the thermal conduction characteristics, a multi-physical field model is constructed to obtain the coupling characteristics of thermal diffusion and electrochemical reaction. If the thermal diffusion value in the coupling characteristics exceeds the safety threshold, the model parameters are optimized by using the gradient descent algorithm to obtain the adjusted multi-physical field model. The influence of thermal diffusion anomaly on the vehicle running state is analyzed by using the adjusted multi-physical field model to obtain the abnormal influence distribution. According to the abnormal influence distribution, the parameters of the initial coupling model are updated to obtain the optimized multi-physical field initial coupling model.

[0034] First, match the vehicle inherent attribute parameters. When the fire development path preliminary contour shows that the heat diffusion is abnormal (such as the heat diffusion rate is more than 5 times of the normal burning scene, and the heat distribution area is concentrated in the non-surface position of the battery pack), the vehicle inherent attribute parameter set completely matched with the model and production batch of the electric vehicle is retrieved from the battery density parameter library. The parameter set covers key indicators such as battery energy density (such as 180 Wh / kg), battery pack arrangement (such as 48 strings and 5 parallel), battery material thermal conductivity (such as positive material 1.2 W / (m·K)), shell structure strength (such as aluminum alloy shell yield strength 280 MPa), etc. Through the unique identification of vehicle VIN code, the initial battery density parameters obtained are completely consistent with the actual vehicle, providing real basic data for model construction.

[0035] Second, simulate the electrochemical reaction process. Based on the initial battery density parameters, combined with the vehicle running state before the fire occurs (such as battery remaining capacity 60%, discharge current 150A, voltage 380V), an electrochemical reaction simulation model is constructed using finite element analysis method (such as ANSYS software). The model needs to include the physical and chemical properties of the core components such as positive and negative electrodes, electrolyte, and separator in the battery. The model simulates the lithium ion migration, electron transfer, and side reactions (such as electrolyte decomposition, SEI film rupture) in the electrochemical reaction process, calculates the reaction rate distribution in different regions (such as positive material delithiation rate 0.02 mol / (L·s), electrolyte decomposition rate 0.005 mol / (L·s)), and obtains the reaction rate distribution data reflecting the spatial difference of electrochemical reaction intensity.

[0036] Then, optimize the electrochemical reaction parameters. Determine whether the reaction rate distribution exceeds the preset threshold (such as electrolyte decomposition rate threshold 0.003 mol / (L·s)). If it exceeds the threshold, it indicates that the electrochemical reaction has entered an abnormal state and there is a risk of thermal runaway. At this time, by adjusting the battery material properties (such as replacing LiPF6 electrolyte with higher stability, optimizing the positive material Ni-Co-Mn ratio), the electrochemical reaction parameters are recalculated (such as reaction activation energy reduction 5 kJ / mol, reaction order adjustment to 1.2), the abnormal reaction rate is reduced, and the optimized reaction rate distribution (such as electrolyte decomposition rate reduced to 0.002 mol / (L·s)) that meets the safety range is obtained, laying a foundation for constructing a stable multi-physical field model.

[0037] Next, a multi-physics coupling model is constructed. Based on the optimized reaction rate distribution, combined with the thermal conductivity characteristics of the battery materials (such as the thermal conductivity of the electrolyte 0.12 W / (m·K), the thermal conductivity of the separator 0.08 W / (m·K)), a multi-physics model containing electrochemical reaction field and thermal conduction field is constructed. The interaction between the two physical fields is described by coupling equations (such as the heat generation rate equation Q=I²R+Q_reaction, where Q_reaction is the electrochemical reaction heat), and the heat generation under different reaction rates (such as the reaction rate 0.002 mol / (L·s) corresponding to the heat generation rate 200 W / m³) and the reaction rate change under different temperature distributions (such as the temperature rise of 10℃ leading to the reaction rate increase of 15%) are calculated, and the coupling characteristics of thermal diffusion and electrochemical reaction are obtained.

[0038] Subsequently, the model parameters are optimized to adjust the physical field. Calculate whether the thermal diffusion value in the coupling characteristics (such as heat flux 500 W / m², temperature gradient 20 K / m) exceeds the safety threshold (such as the thermal diffusion value 450 W / m² corresponding to the battery material thermal runaway critical temperature 60℃), if it exceeds the threshold, use gradient descent algorithm to optimize the multi-physics model parameters (such as the thermal conductivity coefficient is adjusted to 1.3 W / (m·K), the reaction rate constant is reduced by 0.001). Through iterative calculation (such as 50 iterations), the parameter value is continuously adjusted to reduce the thermal diffusion intensity, until the thermal diffusion value falls back to the safety range (such as the heat flux decreases to 420 W / m²), and the adjusted multi-physics model is obtained.

[0039] After that, the abnormal influence distribution is analyzed. Using the adjusted multi-physics model, simulate the influence of thermal diffusion anomaly on vehicle running state, calculate the propagation range (such as the diffusion radius of 10 cm inside the battery pack) and the influence degree (such as the shell temperature rises to 45℃) of abnormal thermal diffusion in different parts of the battery pack, vehicle body structure, etc., and obtain the abnormal influence distribution. The distribution results need to mark the thermal damage level of each part (such as slight damage, moderate damage), and clarify the damage range of thermal diffusion anomaly to the key components of the vehicle.

[0040] Finally, the initial coupling model is updated. According to the abnormal influence distribution, further update the parameters of the initial coupling model (such as correcting the thermal conductivity coefficient of the damaged area separator to 0.05 W / (m·K), the structure parameters of the deformed area shell), through model verification (such as comparing the simulated thermal distribution with the field measured thermal trace), ensure the model accuracy, and finally obtain the optimized multi-physics initial coupling model which can accurately describe the coupling relationship between thermal diffusion anomaly and electrochemical reaction.

[0041] Step S104, the initial coupling model of multiple physical fields is processed by finite element analysis, the impact of vibration bump on the internal structure of the battery is simulated, and the amplification effect of temperature change on thermal stress is superimposed to determine the position of the thermal runaway trigger point.

[0042] The initial coupling model of multiple physical fields is obtained by finite element analysis, vibration bump data is input into the model to obtain a first stress distribution; a second stress distribution is obtained by superimposing the first stress distribution with temperature change data; if the second stress distribution exceeds a preset threshold, the impact area is determined; a third stress distribution is obtained by processing the impact area using finite element analysis and superimposing the amplification effect; a thermal stress concentration area is obtained by analyzing the third stress distribution through a convolutional neural network; if the energy density of the thermal stress concentration area exceeds a preset threshold, the position of the thermal runaway trigger point is determined.

[0043] First step, input vibration data to obtain a first stress distribution. The initial coupling model of multiple physical fields is loaded by using finite element analysis method, the vibration bump data (such as bump acceleration, vibration frequency, duration) during transportation before the fire occurs is input into the model as the model input, and the impact of vibration bump on the internal structure of the battery is simulated. In the finite element model, the battery pack, the battery cell, the separator and other components are divided into multiple units, the stress and strain state (such as tensile stress, shear stress) of each unit under vibration impact is calculated, and the first stress distribution reflecting the stress distribution of the battery caused by vibration impact is obtained. The stress distribution needs to mark the stress size, direction and distribution area of each unit.

[0044] Second step, superimpose temperature data to obtain a second stress distribution. The temperature change data (such as environmental temperature fluctuation, battery cabin temperature rise curve) before the fire occurs is extracted from the transportation working condition record, and the temperature change data is superimposed on the first stress distribution to consider the influence of temperature change on the mechanical properties of battery materials (such as the decrease of material strength caused by temperature rise, additional stress generated by thermal expansion), and the stress state of each unit is recalculated to obtain the second stress distribution after superimposing the amplification effect of temperature change on thermal stress. In the calculation process, the thermal stress calculation formula needs to be introduced, combined with the material thermal expansion coefficient and the temperature change amount, to quantify the amplification degree of temperature on stress, so as to ensure that the second stress distribution can truly reflect the synergistic effect of vibration and temperature.

[0045] Third step, determine the weak points of the impact area. Set a stress preset threshold (this threshold is determined based on the yield strength, fracture strength and other mechanical indicators of battery materials, such as the maximum stress that the diaphragm material can withstand), compare the stress values of each unit in the second stress distribution with the preset threshold, if the stress value of a certain area unit exceeds the threshold, it indicates that this area has exceeded the material bearing limit under the action of vibration and temperature, and is judged as the impact position of the structure, and the impact area is formed by integrating these positions. The impact area is usually the key position of the weak structure inside the battery, such as the connection position between the battery cells, the contact area between the diaphragm and the electrode.

[0046] Fourth step, analyze the impact area to obtain the third stress distribution. Perform fine finite element analysis on the determined impact area, densify the unit grid in this area, and improve the calculation accuracy; at the same time, superimpose the further amplification effect of temperature change on thermal stress (such as after the material damage of the impact area, heat is more likely to accumulate, causing thermal stress to further rise), recalculate the stress distribution of the impact area, and obtain a third stress distribution with higher resolution and more accurate stress state. The third stress distribution needs to focus on depicting the stress concentration in the impact area, such as the stress peak position and stress gradient change.

[0047] Fifth step, identify the thermal stress concentration area. Input the third stress distribution data into the convolutional neural network, the network extracts local features of the stress distribution (such as the shape, size, and gray value change of the stress peak area) through convolutional layers, filters key features through pooling layers, and classifies features through fully connected layers. Finally, output the position information of the thermal stress concentration area. The thermal stress concentration area refers to the area where the stress value is much higher than the surrounding area and there is a trend of obvious heat accumulation. This area is the potential location where thermal runaway is most likely to occur.

[0048] Sixth step, determine the thermal runaway trigger point. Calculate the energy density of the thermal stress concentration area (energy density = stress work / unit volume, combined with material elastic potential energy and thermal energy), if the energy density exceeds the preset threshold (this threshold is determined based on experimental data of battery thermal runaway critical energy density, such as the energy density required for electrolyte decomposition), it indicates that this area has the energy condition for thermal runaway to occur, and at this time the thermal stress concentration area is determined as the thermal runaway trigger point position. The trigger point position needs to be clearly marked with specific spatial coordinates (such as X / Y / Z axis coordinates relative to the battery pack) to ensure accurate positioning.

[0049] Step S105, according to the determined thermal runaway trigger point, reversely map to the ablation morphology in the field feature data set, and use an iterative optimization algorithm to calibrate the coupling model parameters to obtain a refined fire cause mechanism description.

[0050] According to the thermal runaway trigger point, a preset mapping rule is obtained, the matching degree of the mapping rule and the field ablation morphology data set is judged, if the matching degree is higher than a preset threshold, a preliminary reverse mapping path is determined and ablation morphology data association is extracted, the ablation morphology data association is reduced dimension by principal component analysis algorithm, and a reduced dimension morphology data set is obtained, an initial coupling model parameter is constructed according to the reduced dimension morphology data set, the deviation of the initial coupling model parameter and the thermal runaway trigger point is judged, if the deviation exceeds a preset threshold, the initial coupling model parameter is adjusted by an iterative optimization algorithm to obtain a calibrated coupling model parameter, the reverse mapping path is calculated according to the calibrated coupling model parameter, and the ablation morphology data set is integrated to generate a fire cause mechanism description, if the completeness of the fire cause mechanism description is lower than a preset threshold, a refined fire cause mechanism description is obtained by optimization using principal component analysis algorithm.

[0051] The generation of the "refined fire cause mechanism description" reversely maps the thermal runaway trigger point to the field ablation morphology, calibrates the coupling model parameter, optimizes the mechanism description, and clearly determines the core reason for the occurrence of the fire, thereby providing a basis for subsequent preventive threshold setting.

[0052] Firstly, the mapping rule is matched with the ablation morphology. According to the determined thermal runaway trigger point position, the corresponding reverse mapping rule is retrieved from the preset mapping rule library. The mapping rule library contains the association relationship between different thermal runaway trigger points (such as the internal of the battery cell and the interface of the battery pack) and typical ablation morphologies (such as the ablation depth, residual composition and ablation range of the corresponding position). The rule is constructed based on a large number of historical fire case data and simulation experiment results. The matching degree (such as the similarity ratio of the typical ablation morphology in the rule and the actual ablation morphology on site) of the retrieved mapping rule and the field ablation morphology data set (such as the ablation image scanned on site and the residual composition detection data) is calculated, and it is judged whether the two are consistent.

[0053] Secondly, the mapping path is determined to extract the data association. If the matching degree of the mapping rule and the field ablation morphology data set is higher than a preset threshold (such as 90%), it indicates that the mapping rule can effectively reflect the association between the thermal runaway trigger point and the field ablation. At this time, the preliminary reverse mapping path is determined, which needs to clearly determine how the thermal runaway trigger point forms the field ablation morphology through the burning process (such as how the fire spreads to cause ablation in different regions after the trigger point catches fire). Based on the path, the associated information related to the thermal runaway trigger point (such as the ablation trace data around the trigger point and the combustion product concentration data of the corresponding area) in the field ablation morphology data is extracted, and the ablation morphology data association is formed to provide on-site basis for model parameter construction.

[0054] Then, the dimensionality reduction process simplifies the morphology data. Since the ablation morphology data associated may contain a large amount of redundant information (such as repeated ablation area data, noise interference data), the principal component analysis algorithm is used for dimensionality reduction processing. By calculating the covariance matrix of the data, the eigenvalues and eigenvectors are solved, and the principal components with high contribution rate (such as the first N principal components with a cumulative contribution rate of 95%) are selected to replace the original high-dimensional data, and a dimensionality-reduced morphology data set with lower dimension and more concentrated information is obtained, which reduces the complexity of subsequent model calculation while retaining key ablation information.

[0055] Next, the parameter judgment deviation calibration model is constructed. Based on the dimensionality-reduced morphology data set, combined with the structure of the initial coupling model, the initial coupling model parameters (such as the initial values of the thermal conductivity coefficient and the reaction rate constant) are constructed, these parameters are substituted into the model to simulate the formation process of the thermal runaway trigger point, and the deviation (such as spatial coordinate deviation, formation time deviation) between the simulated thermal runaway trigger point and the actual judged thermal runaway trigger point is calculated. If the deviation exceeds the preset threshold (such as spatial deviation exceeding 5mm, time deviation exceeding 10s), it indicates that there is a difference between the initial model parameters and the actual situation, and the parameter value needs to be adjusted through an iterative optimization algorithm (such as Newton iteration method, particle swarm optimization algorithm) until the deviation is reduced to within the threshold range, and the calibrated coupling model parameters are obtained.

[0056] Finally, the mechanism description is generated and optimized. According to the calibrated coupling model parameters, the reverse mapping path is recalculated, the complete process of the thermal runaway trigger point from formation to the present field ablation morphology is clarified, and the key information of the ablation morphology data set (such as the correlation between ablation degree and thermal runaway energy, the correspondence between combustion products and reaction path) is integrated to generate a fire cause mechanism description containing the fire starting cause, development process, and key influencing factors. The integrity of the description (such as whether it covers the trigger point formation reason, the influence of vibration temperature, and the role of thermal diffusion anomaly) is judged, and if the integrity is lower than the preset threshold (such as lacking specific description of the influence of vibration and jolt on the trigger point), the principal component analysis algorithm is used to optimize the description content and supplement the key information, and finally a refined fire cause mechanism description with logical integrity and sufficient basis is obtained.

[0057] Step S106, if the refined fire cause mechanism description contains the dominant factor of transportation working conditions, the vibration and jolt and temperature change pattern of similar cases are obtained from the historical accident database to determine the preventive threshold setting.

[0058] The original vibration mode of a similar case is obtained according to a historical accident database; the original vibration mode of the similar case is clustered through a K nearest neighbor algorithm to obtain a clustered vibration mode; if a standard deviation of the clustered vibration mode exceeds a preset threshold value, an abnormal vibration mode is judged, and the abnormal vibration mode is obtained; a jounce mode is extracted according to the abnormal vibration mode, and the extracted jounce mode is obtained; the jounce mode is extracted through principal component analysis dimension reduction, and a dimension-reduced jounce mode is obtained; a temperature change related sequence is calculated according to the dimension-reduced jounce mode, and a temperature change sequence is obtained; if a slope of the temperature change sequence exceeds a preset threshold value, a high-risk temperature sequence is judged, and the high-risk temperature sequence is obtained; the dimension-reduced jounce mode is fused according to the high-risk temperature sequence, and a fused risk mode is obtained; the fused risk mode is classified through a random forest algorithm, and a classified risk mode is obtained; a prevention threshold value is determined according to the classified risk mode, and a transportation prevention threshold value is obtained; if the transportation prevention threshold value is lower than a preset lower limit, the transportation prevention threshold value is adjusted, and a final prevention threshold value is obtained.

[0059] In the first step, the vibration mode of the case is obtained and clustered. Similar cases are selected from the historical accident database according to the vehicle model, transportation route, and transportation environment (such as climate conditions and road conditions) of the current fire case, and the original vibration mode data (such as vibration acceleration time sequence and jounce frequency distribution) before the fire of these similar cases is extracted. The K nearest neighbor algorithm is used for clustering analysis of the original vibration mode, and the vibration data is divided into multiple clusters according to the similarity (such as the difference between different case vibration modes calculated by the Euclidean distance), each cluster represents a vibration mode with similar characteristics, and the clustered vibration mode is obtained. The optimal K value is determined by indicators such as the silhouette coefficient to ensure the rationality of the clustering results.

[0060] In the second step, the abnormal vibration mode is identified. The standard deviation of each clustered vibration mode (reflecting the fluctuation degree of the vibration mode) is calculated, and the standard deviation is compared with a preset vibration standard deviation threshold value (which is determined based on vehicle transportation safety standards and battery structure tolerance limits). If the standard deviation of a certain clustered vibration mode exceeds the threshold value, it indicates that the vibration mode fluctuates violently and is easy to cause damage to the battery structure, and it is judged as an abnormal vibration mode. The abnormal vibration mode needs to record key features such as vibration acceleration peak value, duration, and fluctuation frequency in detail, which serves as the basis for subsequent jounce mode extraction.

[0061] Third step, extract and reduce dimension of jolt pattern. Based on abnormal vibration pattern, extract the jolt pattern directly related to battery damage (such as specific jolt frequency that causes micro-damage to battery separator, time interval of high-intensity jolt), jolt pattern extraction needs to be combined with battery structure mechanics analysis to screen out the vibration component that has the greatest impact on the internal structure of the battery. Principal component analysis algorithm is used to reduce the dimension of the extracted jolt pattern, remove redundant features (such as highly correlated jolt frequency data), and retain the principal components that can reflect the core risk of jolt, to obtain the reduced dimension jolt pattern, and reduce the complexity of subsequent data processing.

[0062] Fourth step, analyze temperature change sequence. According to the time interval corresponding to the reduced dimension jolt pattern, extract the temperature change data (such as time series of battery compartment temperature, ambient temperature) of the same period from the transportation working condition records of similar cases, calculate the temperature change related sequence (such as temperature change rate, temperature fluctuation amplitude), and obtain the temperature change sequence. The slope of the temperature change sequence is calculated by linear regression method, and the slope reflects the trend of temperature change (such as positive slope indicating continuous temperature rise), and the slope is compared with the preset temperature slope threshold (determined based on the temperature change rule of battery thermal runaway), if the slope exceeds the threshold, it indicates that the temperature rises too fast, and there is a risk of thermal runaway, which is judged as a high-risk temperature sequence.

[0063] Fifth step, fuse risk patterns and classify. Fuse the high-risk temperature sequence with the reduced dimension jolt pattern, integrate the two types of risk features by feature splicing (such as using temperature change rate and jolt acceleration peak value as the dimension of fused features), obtain the fused risk pattern, and the fused pattern needs to reflect the coordinated risk of vibration and temperature. Random forest algorithm is used to classify the fused risk pattern, which is divided into different risk levels (such as low risk, medium risk, high risk), and the algorithm parameters (such as the number of decision trees, node splitting criteria) are optimized during the classification process to ensure the classification accuracy, and the classified risk pattern that can accurately reflect the risk level is obtained.

[0064] Sixth step, determine and optimize the prevention threshold. According to the classified risk pattern, formulate the corresponding prevention threshold for different risk levels (such as vibration acceleration threshold, temperature change rate threshold corresponding to high-risk pattern), the prevention threshold needs to ensure that when the vibration and temperature during transportation reach the threshold, an early warning signal can be sent in time, and the transportation prevention threshold is obtained. Determine whether the transportation prevention threshold is lower than the preset lower limit (the lower limit is determined based on the limit value of normal vehicle transportation working condition, such as lower than the maximum vibration value of normal transportation), if lower than the lower limit, the threshold needs to be adjusted appropriately, balance the warning sensitivity and false alarm rate, and finally obtain the reasonable and feasible final prevention threshold.

[0065] Step S107, by applying the determined preventive threshold setting to the real-time transportation monitoring system, the output of the multi-physical field coupling model is fused to obtain the dynamic assessment result of the electric vehicle transportation fire risk.

[0066] Obtain sensor data in real-time transportation monitoring system, calculate multi-physical field coupling model output according to sensor data, if multi-physical field coupling model output exceeds preset threshold, judge fire risk initial state, fuse transportation system data through fire risk initial state to obtain preliminary risk distribution, use long short-term memory network to process preliminary risk distribution to determine time series risk change, if time series risk change continues to rise, classify risk level through support vector machine, fuse electric vehicle position data according to risk level to obtain dynamic assessment result.

[0067] Firstly, real-time monitoring sensor data is collected. The real-time transportation monitoring system includes sensors (such as vibration sensors, temperature sensors, current sensors, voltage sensors) deployed at key parts of the electric vehicle body, battery pack, and transport carriage, which collect vibration and jolt data (such as real-time acceleration, vibration frequency), temperature data (such as real-time battery compartment temperature, ambient temperature), and battery operation data (such as real-time current, voltage, and remaining capacity) during transportation in real time. Sensor data needs to be continuously collected at a preset sampling frequency (such as 1 Hz) to ensure real-time and continuity of the data. The collected data is transmitted to the data processing center through the wireless communication module.

[0068] Secondly, the model output is calculated and the initial risk is determined. The real-time collected sensor data is input into the constructed multi-physical field coupling model (optimized model after pre-calibration), which simulates the current working condition of the battery's electrochemical reaction state, thermal diffusion state, and stress distribution state according to the input data, calculates the output results of the multi-physical field coupling model (such as thermal diffusion rate, reaction rate, stress value), and compares the model output results with the preset model output threshold (determined based on historical risk cases and safety standards). If the output result exceeds the threshold, it indicates that there is a fire risk at present, and the initial state of fire risk is determined (such as the initial risk level is medium risk); if it does not exceed the threshold, it is determined as a low risk state.

[0069] Then, the data is fused to obtain the preliminary risk distribution. The fire risk initial state is fused with the transportation system data (such as the road condition information of the current transportation route, weather conditions, and transportation cargo weight), the risk contribution of the two types of data is integrated using a weighted fusion method (such as a higher weight for poor road conditions), and the risk values of each part are calculated based on the risk sensitivity of different parts of the electric vehicle (such as the risk weight of the battery pack is higher than that of the vehicle body) to obtain the preliminary risk distribution. The preliminary risk distribution needs to be presented in the form of a spatial distribution map, which marks the real-time risk level of each part and intuitively reflects the spatial distribution difference of the risk.

[0070] Next, analyze the time series risk change trend. Use the long short-term memory network (LSTM) to perform time series analysis on the preliminary risk distribution. The LSTM network can effectively capture the long-term dependence of time series data. By inputting the risk distribution data in the historical time period, the model learns the risk change rule and predicts the risk change trend in the future period, determining the time series risk change (such as risk continuously rising, remaining stable, gradually declining). During the analysis process, the training data needs to be updated in real time to ensure that the model can adapt to the dynamic changes of the transportation conditions and improve the prediction accuracy.

[0071] Then, classify the risk level. If the time series risk change shows that the risk is continuously rising (such as the risk level rising for 5 consecutive minutes), it indicates that the risk is intensifying, and further accurate classification of the risk level is needed. Use the support vector machine (SVM) algorithm to classify the current risk data (such as real-time vibration, temperature, model output results) and divide the risk into more detailed levels (such as low risk, medium risk, high risk, and extremely high risk). In the classification process, the kernel function optimization algorithm is used to improve the performance to ensure accurate differentiation of different risk levels and provide clear basis for subsequent response measures.

[0072] Finally, generate dynamic evaluation results. Fuse the classified risk level with the real-time location data of the electric vehicle (obtained through the GPS module) and mark the current location and corresponding risk level on the electronic map. Combine the risk change trend to predict the risk state of the future location and generate dynamic evaluation results including real-time risk level, risk distribution, future risk warning, and recommended response measures (such as reducing speed, stopping for inspection). The evaluation results need to be pushed to the transportation monitoring center and driver terminal in the form of visual reports in real time to ensure that relevant personnel can timely grasp the risk situation and take effective control measures.

[0073] The above is only the preferred specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or replacements within the technical scope disclosed in the present application can be easily thought of by those skilled in the art, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for causal analysis of electric vehicle transportation fires based on a coupled model, characterized in that, Includes the following steps: Data on ablation morphology, combustion products, and spatial distribution of residues at the scene of an electric vehicle transport fire were collected and integrated with vibration and temperature change logs from the transport condition records to obtain a scene feature dataset. Based on the on-site feature dataset, the texture pattern of the ablation morphology and the chemical composition distribution characteristics of the combustion products are extracted to determine the preliminary outline of the fire development path; if the preliminary outline shows abnormal thermal diffusion, the corresponding vehicle inherent attributes are matched and the electrochemical reaction simulation data are integrated to obtain the multiphysics initial coupling model. The initial coupling model of the multiphysics field is processed by finite element analysis to simulate the impact of vibration and turbulence on the internal structure of the battery and the amplification effect of temperature change on thermal stress, and to determine the location of the thermal runaway trigger point. Based on the ablation morphology in the field feature dataset inversely mapped from the thermal runaway trigger point, the coupling model parameters are calibrated using an iterative optimization algorithm to obtain a refined description of the fire cause mechanism. If the refined description of the fire cause mechanism includes the dominant factors of transportation conditions, vibration and temperature change patterns of similar cases are obtained from the historical accident database to determine the setting of preventive thresholds. By applying the aforementioned preventative threshold setting to the real-time transportation monitoring system and integrating the output of the multiphysics coupling model, a dynamic assessment result of the fire risk in electric vehicle transportation can be obtained.

2. The method for analyzing the causation of electric vehicle transportation fires based on a coupled model according to claim 1, characterized in that, The process of obtaining a field feature dataset includes: High-resolution scanning of the electric vehicle transportation fire scene was performed using image scanning equipment to obtain information on the ablation morphology and the spatial distribution of combustion product residues, generating a first feature image. Vibration and turbulence frequency and temperature change amplitude data are extracted from the electric vehicle transportation condition records. Time series analysis is used to extract features from the vibration frequency and temperature fluctuations to obtain a condition feature vector. Based on the resolution parameters of the first feature image, a data fusion algorithm is used to perform multimodal feature alignment processing between the first feature image and the working condition feature vector to generate a second feature dataset. Based on the second feature dataset, a convolutional neural network algorithm is used to classify and identify the spatial distribution characteristics of the ablation morphology and combustion products, determine the feature category attributes of the fire scene, and obtain the scene feature dataset.

3. The method for analyzing the causation of electric vehicle transportation fires based on a coupled model according to claim 1, characterized in that, Based on the on-site feature dataset, the process of using convolutional neural networks to extract texture patterns of ablation morphology and chemical composition distribution features of combustion products to determine the preliminary outline of the fire development path includes: The on-site feature dataset is standardized using a data preprocessing method to obtain a preprocessed dataset. A convolutional neural network is used to extract features from the preprocessed dataset, generating an ablation morphology texture feature set and a chemical composition distribution feature set. The ablation morphology texture feature set is processed using a spatial distribution analysis method to obtain the spatial distribution pattern of the texture. The chemical composition distribution feature set is analyzed using a combustion process simulation method to generate the dynamic distribution characteristics of combustion products. If the spatial overlap between the spatial distribution pattern of the texture and the dynamic distribution characteristics of the combustion products is higher than a preset threshold, a path prediction model is used to fuse the two types of features to generate a preliminary outline of the fire development path.

4. The method for analyzing the causation of electric vehicle transportation fires based on a coupled model according to claim 1, characterized in that, The process of obtaining the initial coupling model of a multiphysics system includes: If an abnormal thermal diffusion is detected, a parameter set matching the vehicle's inherent properties is retrieved from the battery density parameter library to obtain initial battery density parameters. Based on the initial battery density parameters and the vehicle's operating state, the electrochemical reaction process is simulated using finite element analysis to obtain the reaction rate distribution. If the reaction rate distribution exceeds a preset threshold, the electrochemical reaction parameters are optimized by adjusting the battery material properties to obtain an optimized reaction rate distribution. Based on the optimized reaction rate distribution and thermal conductivity characteristics, a multiphysics model is constructed to obtain the coupling characteristics between thermal diffusion and electrochemical reaction. If the thermal diffusion value in the coupling characteristics exceeds a safety threshold, the model parameters are optimized using a gradient descent algorithm to obtain an adjusted multiphysics model. The impact of the abnormal thermal diffusion on the vehicle's operating state is analyzed using the adjusted multiphysics model to obtain the abnormal impact distribution. Based on the abnormal impact distribution, the parameters of the initial coupling model are updated to obtain an optimized multiphysics initial coupling model.

5. The method for analyzing the causation of electric vehicle transportation fires based on a coupled model according to claim 1, characterized in that, The process of determining the location of the thermal runaway trigger point includes: The initial coupling model of the multiphysics field is obtained through finite element analysis. Vibration and turbulence data are input into the model to obtain a first stress distribution. The first stress distribution is superimposed with temperature change data to obtain a second stress distribution. If the second stress distribution exceeds a preset threshold, the impact location of the structure is determined to obtain the impact region. The impact region is processed by finite element analysis and amplification effect is superimposed to obtain a third stress distribution. The third stress distribution is analyzed by a convolutional neural network to obtain a thermal stress concentration region. If the energy density of the thermal stress concentration region exceeds a preset threshold, the location of the thermal runaway trigger point is determined.

6. The method for analyzing the causation of electric vehicle transportation fires based on a coupled model according to claim 1, characterized in that, The process of inversely mapping the thermal runaway trigger point to the ablation morphology in the on-site feature dataset, and using an iterative optimization algorithm to calibrate the coupled model parameters to obtain a refined description of the fire causation mechanism includes: Based on the thermal runaway trigger point, a preset mapping rule is obtained, and the matching degree between the mapping rule and the on-site ablation morphology dataset is determined. If the matching degree is higher than a preset threshold, a preliminary reverse mapping path is determined and ablation morphology data association is extracted. Principal component analysis was used to reduce the dimensionality of the ablation morphology data and correlate them to obtain a dimensionality-reduced morphology dataset. Based on the reduced-dimensional dataset, construct initial coupling model parameters and determine the deviation between the initial coupling model parameters and the thermal runaway trigger point. If the deviation exceeds a preset threshold, the initial coupling model parameters are adjusted through an iterative optimization algorithm to obtain the calibrated coupling model parameters; The inverse mapping path is calculated based on the parameters of the calibration coupling model, and the ablation morphology dataset is integrated to generate a description of the fire causation mechanism. If the completeness of the fire cause mechanism description is lower than a preset threshold, then principal component analysis algorithm is used to optimize it and obtain a refined fire cause mechanism description.

7. The method for analyzing the causation of electric vehicle transportation fires based on a coupled model according to claim 1, characterized in that, The process of determining the preventive threshold setting includes: The system obtains original vibration patterns from similar cases based on a historical accident database. These patterns are then clustered using the K-nearest neighbor algorithm to obtain clustered vibration patterns. If the standard deviation of the clustered vibration patterns exceeds a preset threshold, they are identified as abnormal vibration patterns. Tumble patterns are extracted from these abnormal vibration patterns. Principal component analysis is used to reduce the dimensionality of these turbulence patterns, resulting in dimensionality-reduced turbulence patterns. Temperature change correlation sequences are calculated based on these dimensionality-reduced turbulence patterns to obtain temperature change sequences. If the slope of the temperature change sequences exceeds a preset threshold, they are identified as high-risk temperature sequences. These high-risk temperature sequences are then fused with the dimensionality-reduced turbulence patterns to obtain fused risk patterns. Random forest algorithms are used to classify and fuse these risk patterns to obtain classified risk patterns. Prevention thresholds are determined based on these classified risk patterns to obtain transportation prevention thresholds. If the transportation prevention threshold is lower than a preset lower limit, it is adjusted to obtain the final prevention threshold.

8. The method for causal analysis of electric vehicle transportation fires based on a coupled model according to claim 1, characterized in that, The process of applying the aforementioned preventative threshold setting to a real-time transportation monitoring system and fusing the output of a multi-physics coupling model to obtain a dynamic assessment result of the fire risk in electric vehicle transportation includes: The system acquires sensor data from a real-time transportation monitoring system; calculates the output of a multiphysics coupling model based on the sensor data; if the output of the multiphysics coupling model exceeds a preset threshold, it determines the initial state of fire risk; it integrates transportation system data with the initial state of fire risk to obtain a preliminary risk distribution; it uses a long short-term memory network to process the preliminary risk distribution and determine the time-series risk changes; if the time-series risk changes continue to rise, it classifies the risk level using a support vector machine; and it integrates electric vehicle location data based on the risk level to obtain a dynamic assessment result.

Images