A method for evaluating the fire resistance and thermal insulation performance of energy storage containers

Abstract

This application provides a method for evaluating the fire resistance and thermal insulation performance of an energy storage container, comprising: acquiring the layout, temperature, electrolyte state, and pressure changes of the battery pack inside the energy storage container, and preprocessing them to obtain the initial triggering characteristics of battery thermal runaway; acquiring the material distribution and structural parameters of the firewall in the jet erosion damage area, simulating the erosion effect of high-temperature electrolyte jet on the firewall using a finite element analysis algorithm to obtain the damage distribution range of the firewall; acquiring the temperature gradient distribution data of the firewall in the risk area where fire breaks through the isolation, calculating the thermal stress difference caused by the temperature gradient, and obtaining the degree of structural fatigue accumulation; predicting the remaining service life of the firewall structure based on the degree of structural fatigue accumulation, and adjusting the material thickness and layered structure of the firewall according to the design service life requirements; dynamically updating the material ratio and structural parameters of the firewall based on the probability of firewall crack formation, and generating the final firewall design scheme.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method for evaluating the fire resistance and heat insulation performance of energy storage containers. Background Technology

[0002] Energy storage technology is crucial in the new energy field, providing key support for solving the imbalance between power supply and demand. Especially in the large-scale application of renewable energy, energy storage containers, as core equipment, directly impact system safety and efficiency. However, the fire risk caused by battery thermal runaway has become a major bottleneck restricting its development. Currently, the design of firewalls inside energy storage containers mostly uses traditional refractory materials and simple partition structures. While these can delay the spread of fire to some extent, their effectiveness is limited in the face of the complex characteristics of battery thermal runaway. These methods are insufficient to effectively cope with the erosion and damage to the partition walls caused by high-temperature electrolyte spray, nor can they adequately suppress the propagation of deflagration pressure waves, making it easy for the fire to break through the partition walls and affect adjacent battery packs. Furthermore, uneven temperature distribution in the partition walls further weakens their fire-resistant performance, increasing the risk of fire runaway. The core challenge of fire isolation in energy storage containers lies in the extreme destructiveness of thermal runaway. When batteries experience thermal runaway, high-temperature electrolyte spray directly impacts the partition walls, causing material erosion and structural damage. This localized damage weakens the integrity of the partition walls, making them unable to withstand the pressure wave impact generated by subsequent deflagration. The rapid propagation of pressure waves further exacerbates uneven stress on the partition walls, leading to cracks or deformation, which in turn allows the fire to breach the isolation and spread to other battery packs. This problem is exacerbated by the uneven temperature gradient distribution caused by uneven stress on the partition walls, potentially leading to structural fatigue or localized failure due to differences in thermal stress at different locations. These factors are interconnected and collectively constitute a complex challenge for fire isolation. Therefore, optimizing the structural design of the fireproof walls of the compartments to effectively resist the erosion of high-temperature electrolyte spray, suppress the propagation of deflagration pressure waves, and improve the unevenness of temperature gradient distribution has become a key issue in improving the fire isolation effect of energy storage containers. Summary of the Invention

[0003] This invention provides a method for evaluating the fire resistance and thermal insulation performance of energy storage containers, mainly including:

[0004] The layout, temperature, electrolyte state and pressure changes of the battery pack inside the energy storage container are obtained and preprocessed to obtain the initial triggering characteristics of battery thermal runaway.

[0005] The initial triggering characteristics of battery thermal runaway are extrapolated and threshold judgment is performed to predict the time point of full thermal runaway. The jet trajectory of high-temperature electrolyte is obtained by combining the battery pack layout, and the erosion and damage area is determined based on the jet trajectory.

[0006] The material distribution and structural parameters of the fire wall of the compartment in the area damaged by jet erosion were obtained. The erosion effect of high-temperature electrolyte jet on the partition wall was simulated by the finite element analysis algorithm to obtain the distribution range of damage to the partition wall.

[0007] The initial pressure wave parameters generated by the thermal runaway and deflagration of batteries within the distribution range of the damaged partition wall were obtained. A deflagration pressure wave propagation model was constructed to simulate the propagation path and intensity of the pressure wave in the partition wall cracks. Based on the propagation path and intensity, the influence range of the pressure wave propagation on the adjacent battery pack was identified, and the risk area of ​​fire breaking through the isolation was determined.

[0008] Obtain temperature gradient distribution data of the partition wall in the risk zone where fire breaks through the isolation, calculate the difference in thermal stress caused by the temperature gradient, and obtain the degree of structural fatigue accumulation;

[0009] The remaining service life of the partition wall structure is predicted based on the degree of cumulative structural fatigue, and the material thickness and layering structure of the partition wall are adjusted according to the design service life requirements.

[0010] Real-time monitoring data of deflagration pressure waves were obtained after adjusting the material thickness and layered structure of the partition wall. The pressure release channels inside the partition wall were adjusted, and a multi-physics field coupled simulation algorithm was used to obtain the probability of partition wall crack generation.

[0011] Based on the probability of cracks forming in the partition wall, the material ratio and structural parameters of the partition wall are dynamically updated to generate the final firewall design scheme.

[0012] Furthermore, the layout, temperature, electrolyte state, and pressure changes of the battery packs inside the energy storage container are acquired and preprocessed to obtain the initial triggering characteristics of battery thermal runaway. This includes: acquiring spatial coordinate data and inter-pack distance data of the battery packs from sensors inside the energy storage container; processing the sensor data using a depth sensing algorithm to obtain a three-dimensional layout parameter set for the battery packs, which includes battery pack coordinate values, battery pack surface area, and inter-pack distance values. Real-time temperature field data is acquired from the battery pack surface using an infrared sensor array; the temperature field data is then denoised and filtered to obtain a battery pack temperature dataset, which includes battery pack surface temperature values, temperature gradient values, and temperature distribution uniformity values. Electrolyte vapor pressure data is acquired around the battery pack, electrolyte level data is acquired from the bottom of the battery pack, and surface stress distribution data is acquired; the acquired data is standardized to obtain a battery pack pressure state feature set. A temperature change rate matrix is ​​calculated based on the battery pack temperature dataset; when any element in the temperature change rate matrix exceeds a preset temperature abrupt change threshold, the corresponding position of that element is marked as a temperature anomaly point. A pressure change rate matrix is ​​calculated based on the battery pack pressure state feature set. When any element in the pressure change rate matrix exceeds a preset pressure mutation threshold, the corresponding position of that element is marked as a pressure anomaly. The electrolyte level drop rate is calculated based on the electrolyte level height data. When the electrolyte level drop rate exceeds a preset leakage threshold, it is marked as a leakage anomaly. A heat transfer matrix is ​​constructed based on the location of temperature anomalies and the battery pack 3D layout parameter set. A random forest algorithm is used to perform correlation analysis on temperature anomalies, pressure anomalies, and leakage anomalies, outputting the location of the point with the highest probability of thermal runaway triggering and the triggering time series.

[0013] Furthermore, the initial triggering characteristics of battery thermal runaway are extrapolated and threshold-based to predict the time point of full-scale thermal runaway. The jet trajectory of the high-temperature electrolyte is obtained by combining the battery pack layout. The erosion and damage area is determined based on the jet trajectory, including: constructing a thermal runaway parameter matrix based on battery pack temperature and pressure data. This parameter matrix includes temperature change rate, pressure change rate, and thermal diffusivity coefficient. When the temperature change rate exceeds a preset temperature mutation threshold or the pressure change rate exceeds a preset pressure mutation threshold, the mutation time and location are recorded. A long short-term memory network is used to fit the mutation parameters to obtain the thermal runaway propagation rate value. A thermal runaway propagation path map is constructed based on the thermal runaway propagation rate value and battery pack layout parameters. This path map records the spatiotemporal trajectory of thermal runaway propagation. The heat transfer direction is obtained by calculating the temperature and pressure gradients between adjacent battery packs. Computational fluid dynamics methods are used to calculate electrolyte jet parameters. Combined with the temperature field distribution in the thermal runaway propagation path map, the electrolyte jet trajectory curve is calculated. This trajectory curve characterizes the electrolyte diffusion range. Position sequence data is extracted from the electrolyte jet trajectory curve. Fourier transforms are performed on the temperature and pressure gradient data within the position sequence. The time point of full-scale thermal runaway is determined by comparing the frequency domain feature values ​​with a preset thermal runaway spectrum threshold. A damage prediction feature vector is constructed based on the electrolyte jet trajectory curve and the time point of full-scale thermal runaway. This feature vector includes temperature field distribution, pressure field distribution, and thermal runaway propagation velocity. A neural network is used to predict the battery pack damage probability distribution. High-probability region boundary points are extracted from the damage probability distribution map to determine the erosion and damage areas.

[0014] Furthermore, the material distribution and structural parameters of the firewall in the jet erosion damage area are obtained. The erosion effect of high-temperature electrolyte jet on the firewall is simulated using a finite element analysis algorithm to determine the damage distribution range. This includes: obtaining firewall structural mesh data based on 3D scanning data; retrieving material parameters for each layer of the firewall from a material database; collecting surface material distribution data using a spectrometer; and identifying the material layer structure using a deep learning algorithm to obtain a firewall structural parameter dataset. A finite element mesh model is constructed based on this dataset. Thermodynamic and mechanical parameter sets are retrieved from the material database to obtain a firewall material property dataset. A quadrilateral meshing method is used to decompose the firewall surface into elements. The force direction and magnitude of each mesh element are calculated based on the electrolyte jet trajectory curve, constructing a stress distribution matrix. This matrix records the normal and shear stress values ​​of each mesh element. The temperature field distribution matrix is ​​calculated using the heat conduction equation, recording the temperature and temperature gradient values ​​of each mesh element. The stress intensity value of each mesh element is calculated using the Mises stress criterion to obtain a stress intensity distribution matrix. The set of material strength threshold parameters for each layer of the firewall is obtained from a material database. When the stress intensity value of a mesh element exceeds the corresponding material strength threshold, it is marked as a damaged element. The expansion range of the damaged element is calculated using crack propagation theory, and a damage state feature vector is constructed. The feature vector includes stress intensity factor, crack propagation rate, and crack propagation direction. A convolutional neural network is used to predict the damage distribution map of the firewall, obtaining the coordinate set of boundary points of the damaged distribution range of the firewall.

[0015] Furthermore, the initial pressure wave parameters generated by the thermal runaway deflagration of batteries within the damaged partition wall distribution area are obtained, and a deflagration pressure wave propagation model is constructed to simulate the propagation path and intensity of the pressure wave in the partition wall cracks. Based on the propagation path and intensity, the impact range of the pressure wave propagation on adjacent battery packs is identified, and the risk area of ​​fire breaching the isolation is determined. This includes: obtaining a set of crack morphology parameters based on the partition wall damage distribution area data, including crack length, crack width, crack depth, and crack orientation angle; collecting deflagration pressure wave data through a pressure sensor array; performing spectral analysis on the pressure wave data using Fourier transform to obtain a pressure wave feature dataset; extracting the pressure wave frequency and amplitude components from the pressure wave feature dataset to construct an initial pressure wave parameter matrix; constructing a computational fluid dynamics field based on the crack morphology parameter set and the initial pressure wave parameter matrix, including a pressure field, velocity field, and density field; and calculating the pressure wave propagation process in the cracks by solving the Navier-Stokes equations; using an adaptive mesh refinement method to divide the crack region into meshes; constructing a pressure wave propagation path map based on the pressure and velocity values ​​at the mesh nodes; and extracting the pressure wave intensity attenuation curve from the pressure wave propagation path map. Stress wave data from adjacent battery packs is collected using acoustic emission sensors. This data includes amplitude, arrival time, and duration. A long short-term memory network is used to predict the stress response of the battery packs, resulting in a stress distribution cloud map. The stress exceedance range is calculated based on the stress distribution cloud map. A set of structural strength parameters for the battery packs is extracted from a materials database, and the probability of damage is determined by comparing the stress exceedance value with the structural strength threshold. A fire breach prediction model is constructed using a support vector machine. The input features of this model include the stress exceedance range, damage probability distribution, and pressure wave intensity attenuation rate, resulting in a set of coordinates for the boundary points of the fire breach isolation risk zone.

[0016] Furthermore, the pressure wave's effective range is calculated based on its propagation path, and the probability of thermal runaway triggering in adjacent battery packs is analyzed. Based on this probability, a set of fire breach areas is identified. Weighted risk analysis is performed on this set to assess the failure risk of isolation facilities, calculate the isolation risk distribution, obtain the risk area distribution, generate a spatial mapping of the isolation risk, visualize the risk areas, and determine the risk areas. This includes: calculating the pressure wave attenuation curve based on the pressure wave propagation path data; calculating the temperature field distribution matrix within the pressure wave's effective range using the heat conduction equation; and predicting the probability of thermal runaway triggering in adjacent battery packs using a deep neural network. The input features of the deep neural network include pressure wave intensity, temperature field, and stress distribution values, resulting in a trigger probability distribution matrix. Based on the trigger probability distribution matrix, probability threshold point coordinates are extracted to construct the fire breach area outline. A region segmentation algorithm is used to divide the area within the outline into a grid, resulting in a set of regional grid units. A spatial clustering algorithm is used to classify the regional grid units. This clustering algorithm calculates the clustering distance based on the grid unit's pressure wave intensity, temperature field, and trigger probability values, and sets a clustering threshold to classify risk level areas. A regional weight matrix is ​​constructed based on the risk level regions. The weight coefficients in this matrix are obtained through normalization calculations of pressure wave intensity, temperature field, and trigger probability values, resulting in a weighted risk distribution map. A stress distribution matrix for the isolation facility is calculated based on this weighted risk distribution map. The stress state at each point is calculated using the equations of mechanics of materials, and the failure probability of the isolation facility is determined using fracture mechanics criteria, which are based on a comparison between the stress intensity factor and the critical strength. A random forest algorithm is used to optimize the risk region boundaries. The input features of the random forest algorithm include stress state values, failure probability values, and regional weight values. A set of risk contour lines is obtained through spatial interpolation. A spatial coordinate mapping relationship is established based on the risk contour line set to construct a spatial distribution dataset for the risk region. This dataset includes boundary point coordinates, risk level values, and spatial location relationships.

[0017] Furthermore, the temperature gradient distribution data of the partition wall in the risk area where fire breaks through the isolation barrier is obtained, and the thermal stress difference caused by the temperature gradient is calculated to obtain the cumulative degree of structural fatigue. This includes: obtaining the temperature field distribution of the partition wall based on the data of the risk area where fire breaks through the isolation barrier; using an infrared sensor array to scan the temperature of the partition wall surface; constructing a three-dimensional temperature distribution matrix through a temperature field reconstruction algorithm; the temperature distribution matrix records the temperature value at each point of the partition wall; and calculating the spatial partial derivative of the temperature distribution matrix to obtain a temperature gradient field dataset. A thermal expansion strain matrix is ​​calculated based on the temperature gradient field dataset; the strain matrix includes normal strain components and tangential strain components; and a thermal stress tensor is calculated using the thermoelastic constitutive equation; the thermal stress tensor includes normal stress components and shear stress components. A deep neural network is used to predict the spatial distribution of thermal stress; the neural network input features include temperature gradient values, strain component values, and material parameter values; and the output is a thermal stress difference distribution map, which records the thermal stress gradient values. A stress cycle spectrum is constructed based on the thermal stress difference distribution map; structural stress fluctuation data is collected using an acoustic emission sensor; and stress peak-valley value sequences are extracted through stress waveform analysis; the sequences record the maximum and minimum values ​​of stress fluctuations. A rainflow counting method is used to statistically analyze the stress peak-valley sequence. The rainflow counting is based on the stress amplitude and the number of cycles, resulting in a stress cycle counting matrix. This matrix records the number of cycles at different amplitude levels. The cumulative damage value is calculated based on the stress cycle counting matrix and corrected using a linear cumulative damage criterion, which considers both stress amplitude and average stress effects. A fuzzy neural network is used to predict the structural fatigue degree. The neural network's input features include the cumulative damage value, the number of stress cycles, and material fatigue characteristic parameters. The output is structural fatigue cumulative distribution data, which records the spatial distribution of fatigue damage.

[0018] Furthermore, the remaining service life of the partition wall structure is predicted based on the cumulative degree of structural fatigue. The material thickness and layered structure of the partition wall are adjusted according to the design service life requirements. This includes: constructing a fatigue life prediction matrix based on the cumulative degree of structural fatigue data; acquiring structural stress wave data using an acoustic emission sensor; obtaining stress wave amplitude, frequency, and duration values ​​through a stress waveform feature extraction algorithm; and constructing a stress feature vector. A long short-term memory network is used to predict the remaining service life of the structure. The network input features include the stress feature vector, cumulative fatigue value, and material parameter values, and outputs a life prediction distribution map, which records the remaining service life value at each point. The material strength attenuation curve is calculated based on the life prediction distribution map. The distribution map of internal defects in the material is obtained using an acoustic attenuation detection method. The material strength degradation rate is calculated using defect evolution theory, based on defect size, density, and distribution. A material thickness optimization objective function is constructed based on the material strength degradation rate. This objective function includes stress distribution terms, service life constraint terms, and thickness constraint terms. The stress distribution map of each layer of material is calculated using a finite element reinforcement algorithm. A genetic algorithm is used to optimize the thickness of the partition wall material. This algorithm iteratively optimizes based on stress distribution maps and life constraints to obtain a material thickness distribution scheme, which includes the thickness values ​​of each layer. The layered structural parameters are optimized according to the material thickness distribution scheme, and a stress transfer function is constructed using material mechanics calculation methods. This function describes the interlayer stress transfer characteristics. The interlayer stress distribution is calculated using stress wave propagation theory, resulting in a set of layered structural parameters for the partition wall.

[0019] Furthermore, real-time monitoring data of the deflagration pressure wave was obtained after adjusting the material thickness and layered structure of the partition wall. The pressure release channels inside the partition wall were adjusted, and a multiphysics coupling simulation algorithm was used to obtain the probability of crack formation in the partition wall. This included: obtaining a set of pressure monitoring point coordinates based on the adjusted partition wall structure data; collecting deflagration pressure wave data through a pressure sensor array; and constructing a pressure field distribution map using a deep neural network. The neural network input features included pressure values, location coordinate values, and time series values. Temperature field data was collected through a temperature sensor array, and a temperature field distribution map was constructed. A multiphysics dataset was obtained using a field value superposition method. Pressure release channel parameters were calculated based on the multiphysics dataset, and the pressure distribution within the channel was calculated using fluid dynamics equations. The channel parameters included channel width, channel length, and channel cross-sectional area. A dynamic controller was used to adjust the channel opening in real time. The controller performed feedback control based on pressure difference and flow rate values, and channel flow rate data was obtained through a flow sensor to obtain a channel control parameter matrix. A multiphysics coupling model is constructed based on the channel control parameter matrix. Stress field data is collected from a stress sensor array, and the thermal stress distribution is obtained through thermal stress calculation methods. The coupling model includes the interaction relationships between the pressure field, temperature field, and stress field. A random forest algorithm is used to extract features from the coupled field intensity distribution map, including field intensity values, field gradients, and field fluctuation amplitudes. The crack propagation driving force is calculated using fracture mechanics criteria. Crack initiation conditions are determined based on the material fracture toughness threshold, and the crack nucleation probability is calculated using probabilistic statistical methods, resulting in a crack generation probability distribution matrix for the partition wall. This matrix records the crack generation probability value at each point.

[0020] Furthermore, based on the probability of crack formation in the partition wall, the material ratio and structural parameters of the partition wall are dynamically updated to generate the final firewall design scheme. This includes: constructing a material ratio optimization objective function based on the crack formation probability data, wherein the objective function includes material strength, thermal insulation performance, and fire resistance performance; obtaining a material component parameter set from a material database, wherein the parameter set includes the content of refractory components, thermal insulation components, and structural components; using a genetic algorithm to dynamically optimize the material ratio, wherein the algorithm iteratively optimizes based on the crack formation probability value and material performance indicators to obtain a material composition ratio scheme, wherein the scheme includes the optimal ratio value of each component; constructing a structural parameter matrix based on the material composition ratio scheme; obtaining stress distribution data through finite element method calculation; using a deep neural network to predict the thickness value of each layer of the firewall, wherein the neural network input features include stress distribution value, temperature distribution value, and crack probability value, and outputs the optimal thickness value of each layer; calculating interlayer bonding parameters based on the thickness value of each layer of the firewall; obtaining the acoustic characteristics of the interlayer interface through ultrasonic detection method; and calculating the interlayer bonding strength value and interface stress distribution value using interface mechanics theory. An isolation performance evaluation matrix was constructed based on the interlayer stress distribution values, and the thermal conductivity of the firewall was obtained through thermal parameter calculation methods. Structural integrity parameters, including structural damage degree, crack density, and strain distribution, were obtained using acoustic emission detection methods to generate a firewall design dataset. This dataset includes material proportioning parameters, structural dimensional parameters, and performance index parameters.

[0021] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0022] This invention discloses a method for evaluating the fire resistance and thermal insulation performance of energy storage containers. The method acquires data on battery pack layout, temperature, electrolyte state, and pressure changes, preprocesses this data to obtain initial thermal runaway triggering characteristics, and estimates the time point of full-scale thermal runaway and the trajectory of high-temperature electrolyte jetting. Subsequently, finite element analysis is used to simulate the erosion effect of electrolyte jetting on the partition wall, constructing a deflagration pressure wave propagation model to identify risk areas where fire can breach the isolation barrier. Based on thermal stress analysis, the degree of structural fatigue accumulation is calculated, the remaining service life of the partition wall is predicted, and the material thickness and layered structure are adjusted. Finally, a dynamic control algorithm is used to adjust the pressure release channel, and multiphysics coupling simulation is employed to obtain the probability of crack formation in the partition wall, dynamically updating the partition wall design. This invention achieves accurate evaluation and optimization of the fire isolation performance of energy storage containers, improving system safety. Attached Figure Description

[0023] Figure 1 This is a flowchart of a method for evaluating the fire resistance and heat insulation performance of an energy storage container according to the present invention.

[0024] Figure 2 This is a schematic diagram of a method for evaluating the fire resistance and heat insulation performance of an energy storage container according to the present invention. Detailed Implementation

[0025] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] To facilitate understanding of the embodiments of the present invention, the key technical concepts involved in the present invention will be briefly described below:

[0027] Energy storage containers are modular energy storage devices that integrate battery packs, cooling systems, and control units, used for storing and releasing electrical energy. Battery thermal runaway refers to the uncontrollable temperature rise, electrolyte leakage, and deflagration caused by overheating, overcharging, or internal short circuits in the battery, which can potentially lead to a fire. Firewalls are fire-resistant structures inside energy storage containers used to separate battery packs and prevent the spread of fire. The destructive characteristics of thermal runaway include high-temperature electrolyte spray, deflagration pressure waves, and thermal stress induced by temperature gradients; these factors place extremely high demands on the materials and structural performance of the firewalls.

[0028] In existing technologies, partition wall designs are mostly based on static fire resistance performance tests, which are insufficient to cope with the dynamic failure process of thermal runaway. This invention comprehensively evaluates the fire isolation performance of partition walls through multiphysics coupling analysis and dynamic optimization, and proposes targeted design improvement schemes to enhance the safety of energy storage containers. Figure 1-2 This embodiment of a method for evaluating the fire resistance and thermal insulation performance of an energy storage container may specifically include:

[0029] S101 acquires the layout parameters, temperature field data, electrolyte state and pressure characteristics of the battery pack inside the energy storage container, and extracts the initial triggering characteristics of battery thermal runaway through data preprocessing.

[0030] In this embodiment of the invention, multiple types of sensors are deployed inside the energy storage container to collect real-time data on the battery pack's operating status. Data acquisition covers the battery pack's spatial layout, temperature distribution, electrolyte state, and pressure changes, aiming to provide comprehensive foundational data for extracting thermal runaway characteristics. It is understood that the sensor types and arrangement can be flexibly adjusted according to actual scenario requirements, such as using infrared sensors, pressure sensors, or liquid level sensors.

[0031] S1011 acquires the three-dimensional layout parameters and temperature field data of the battery pack, and constructs the spatial distribution model and temperature distribution characteristics of the battery pack.

[0032] In this embodiment of the invention, the coordinates and gap data of the battery packs collected by sensors are processed using a depth sensing algorithm to generate a three-dimensional layout parameter set containing the battery pack positions, surface areas, and spacing. For example, the coordinates of 500 battery packs inside a container, the surface area of ​​each battery pack (approximately 250 square centimeters), and the 20-millimeter gap distance can constitute the basic layout data. An infrared sensor array is used to collect temperature field data, with a sensor coverage area of ​​100 square centimeters and a collection interval of 100 milliseconds. After Gaussian filtering and noise reduction, the collected data forms a temperature dataset containing surface temperature values, temperature gradients, and uniformity. The temperature range is typically between 20 and 45 degrees Celsius, the gradient does not exceed 3 degrees Celsius per centimeter, and the uniformity value remains above 0.9. This data provides a spatial and thermal basis for thermal runaway analysis.

[0033] S1012 collects electrolyte state and pressure data, extracts abnormal features, and constructs a heat transfer model.

[0034] In this embodiment of the invention, the electrolyte vapor pressure is monitored by a vapor pressure sensor, which is distributed in a grid around the battery pack. The normal pressure range is 0.8 to 1.2 atmospheres. A liquid level sensor is located at the bottom of the battery pack, with a reference liquid level height of 80 mm. An abnormal signal is triggered when the liquid level is below 75 mm. A piezoelectric pressure sensor measures the stress distribution on the surface of the battery pack, and the stress state feature set is formed after standardization.

[0035] Based on temperature and pressure data, a rate of change matrix is ​​calculated. Anomalies are identified when the temperature change rate exceeds 8 degrees Celsius every 5 seconds or the pressure change rate exceeds 0.3 atmospheres every 2 seconds. An electrolyte leakage anomaly is identified when the liquid level drop rate exceeds 2 millimeters per second. Combining the anomaly locations and 3D layout parameters, a heat transfer matrix is ​​constructed, considering a thermal conductivity of 0.6 watts per meter Kelvin and a thermal diffusivity of 0.8 millimeters per second, to describe the heat conduction characteristics between battery cells.

[0036] S1013 uses the random forest algorithm to analyze the correlation of outliers and determine the point with the highest probability of thermal runaway and its time series.

[0037] In this embodiment of the invention, the random forest algorithm takes the location, time series, and heat transfer matrix of temperature anomalies, pressure anomalies, and leakage anomalies as input, and outputs the probability that each anomaly will become a thermal runaway trigger source. When the trigger probability at a certain location exceeds 85%, that location is determined as the initial trigger point for thermal runaway. Simultaneously, it records the entire process data from the occurrence of the anomaly to the trigger determination, including the change curves of temperature, pressure, and liquid level. This data provides a reliable basis for subsequent prediction of thermal runaway trends. This embodiment of the invention does not limit the specific parameters of the algorithm and can be optimized by technicians according to actual needs.

[0038] In this embodiment of the invention, multi-dimensional sensor data acquisition and in-depth analysis are used to accurately extract the initial triggering characteristics of thermal runaway, including sudden temperature changes, electrolyte leakage, and rapid pressure increases. Compared with traditional single-parameter monitoring, this method can comprehensively characterize the early signs of thermal runaway, significantly improve the accuracy of trigger point identification, and provide a solid foundation for subsequent fire propagation prediction and partition wall optimization, thereby effectively reducing the risk of fire runaway and improving the safety of energy storage containers.

[0039] S102 performs dynamic trend analysis and threshold determination based on the initial triggering characteristics of battery thermal runaway, predicts the time of full-scale thermal runaway, derives the trajectory of high-temperature electrolyte spray based on battery pack layout, and determines the erosion and damage area of ​​the partition wall.

[0040] In this embodiment of the invention, through in-depth analysis of the initial triggering characteristics of thermal runaway, combined with multidimensional data processing and model prediction, the dynamic evolution process of thermal runaway is accurately deduced, electrolyte jet trajectory is generated, and damaged areas of the partition wall are identified. This embodiment of the invention does not impose excessive limitations on the specific algorithm implementation details, which can be optimized and adjusted by technicians according to actual scenarios.

[0041] S1021 constructs a thermal runaway parameter matrix, extracts the characteristics of temperature and pressure abrupt changes, and determines the thermal runaway propagation rate.

[0042] In this embodiment of the invention, a thermal runaway parameter matrix is ​​constructed based on real-time acquired battery pack temperature and pressure data, including the rate of temperature change, the rate of pressure change, and the thermal diffusivity. During normal operation, the battery pack temperature fluctuates between 20 and 40 degrees Celsius, with a temperature change rate less than 0.1 degrees Celsius per second, and the pressure ranges from 0.9 to 1.1 standard atmospheres, with a pressure change rate less than 0.05 standard atmospheres per second. When the rate of temperature change exceeds 0.5 degrees Celsius per second or the rate of pressure change exceeds 0.05 standard atmospheres per second, the time and location of the abrupt change are recorded. A long short-term memory network is used to perform time-series fitting on the abrupt change data to calculate the thermal runaway propagation rate, which typically fluctuates within the range of 0.5 to 2 centimeters per second.

[0043] The thermal diffusivity increases with increasing temperature, reaching 0.8 mm² / s for example at 35 degrees Celsius, providing a quantitative basis for thermal runaway propagation. This step, through matrix processing, clearly characterizes the thermodynamic properties of thermal runaway, laying the foundation for diffusion path analysis.

[0044] S1022 generates a thermal runaway propagation path diagram based on battery pack layout and abrupt change characteristics, and derives the direction of heat transfer.

[0045] In this embodiment of the invention, a thermal runaway propagation path diagram with 500 nodes is constructed by combining the thermal runaway propagation rate and a three-dimensional layout parameter set. Each node represents a battery pack, and the connection weights between nodes are determined by the temperature and pressure gradients of adjacent battery packs. When the temperature gradient reaches 5 degrees Celsius per centimeter, the heat transfer rate increases significantly, triggering a thermal runaway propagation chain. For example, the temperature of eight adjacent battery packs around the thermal runaway source point rises by more than 15 degrees Celsius within 120 seconds, forming a propagation chain. The path diagram records the thermal runaway propagation path through spatiotemporal trajectories and determines the heat transfer direction by combining gradient analysis, providing spatial constraints for the derivation of the electrolyte jet trajectory. It is understood that the number of nodes and the weight calculation method of the path diagram can be adjusted according to the container size.

[0046] S1023 uses computational fluid dynamics to simulate the electrolyte jetting behavior and generate jetting trajectory curves.

[0047] In this embodiment of the invention, based on the temperature field distribution in the path diagram, computational fluid dynamics is used to simulate the injection parameters of the high-temperature electrolyte, including injection velocity, angle, and pressure. The electrolyte viscosity drops to 50% of that at room temperature at 85 degrees Celsius, the initial injection velocity can reach 3 meters per second, the angle deflects vertically from 15 to 75 degrees, and the peak pressure reaches 3.5 atmospheres.

[0048] Based on the simulation results, an electrolyte jet trajectory curve was generated, representing a circular area with a diameter of approximately 80 cm and a radial trajectory distribution. The jet parameters were calculated considering the hydrodynamic properties of the electrolyte and the influence of the temperature field to ensure the accuracy of the trajectory curve. This step, through dynamic simulation, accurately describes the potential erosion path of the electrolyte on the partition wall.

[0049] S1024 predicts the timing of thermal runaway and the affected area through frequency domain analysis and neural networks.

[0050] In this embodiment of the invention, position sequence data is extracted from the jet trajectory curve, and Fourier transforms are performed on the temperature and pressure gradients in the sequence to generate frequency domain feature values. During normal operation, the spectral energy is concentrated below 0.1 Hz; when thermal runaway occurs, significant energy accumulation occurs in the range of 0.5 to 2 Hz. When the feature value exceeds a preset threshold of 0.8, it is determined to be a precursor to a full-scale thermal runaway, with an outbreak interval of approximately 180 seconds. The threshold is calibrated based on historical fault data.

[0051] Further, a damage prediction feature vector incorporating temperature, pressure, and diffusion velocity was constructed, and a convolutional neural network was used to predict the probability distribution of battery pack damage. The prediction results showed that the probability of damage within 30 cm of the thermal runaway source exceeded 90%, while the probability decreased from 40% to 90% in the 30-60 cm range, forming an elliptical high-risk area with a major axis of 80 cm and a minor axis of 60 cm. This area represents the erosion and damage zone of the partition wall. This step, combining frequency domain analysis and machine learning, improved the accuracy of the burst time prediction and clarified the boundaries of the damaged area.

[0052] Compared with traditional static analysis, this method can capture the spatiotemporal evolution characteristics of thermal runaway, generate accurate electrolyte spray trajectories, and identify damaged areas of the partition wall, providing a reliable basis for partition wall performance evaluation, thereby effectively improving the pertinence and reliability of fire isolation design.

[0053] S103 obtains the material distribution and structural parameters of the firewall of the energy storage container compartment, and uses finite element analysis to simulate the erosion effect of high-temperature electrolyte spray on the firewall, generating the distribution range of damage to the firewall.

[0054] In this embodiment of the invention, a finite element model of the firewall is constructed using high-precision three-dimensional scanning and material property analysis of the electrolyte jet erosion area. Combining thermodynamic and mechanical parameters, the erosion process is dynamically simulated, accurately predicting the extent of damage to the firewall. Compared to traditional static testing, this method can meticulously characterize the material failure behavior under the coupling of jet impact and thermal stress, providing data support for optimizing firewall design. This embodiment of the invention does not strictly limit specific model parameters or algorithm hyperparameters; they can be flexibly adjusted according to actual needs.

[0055] S1031 Obtain firewall structure mesh data and material layering characteristics, and construct a structural parameter dataset.

[0056] In this embodiment of the invention, a 3D scanner is used to scan the surface of a firewall with a precision of 0.1 mm, generating structural mesh data with a thickness of 120 mm. The firewall consists of a refractory layer, a thermal insulation layer, and a load-bearing layer, with thicknesses of 30 mm, 60 mm, and 30 mm, respectively. A spectral analyzer collects surface material distribution data. The ceramic material of the refractory layer exhibits characteristic spectral peaks in the 750-850 nm wavelength range, the infrared transmittance of the mineral wool in the thermal insulation layer is less than 5%, and the steel in the load-bearing layer has metallic reflective properties. A deep convolutional neural network is used to analyze the spectral and mesh data, identify the material layer interfaces, and generate a structural parameter dataset containing the coordinates, thickness, and distribution characteristics of each layer. The material database provides the following data: refractory layer density 2.8 g / cm³, Young's modulus 180 GPa, Poisson's ratio 0.28; thermal insulation layer density 0.12 g / cm³; load-bearing layer density 7.8 g / cm³, yield strength 355 MPa. These data provide an accurate geometric and material basis for subsequent finite element modeling, ensuring the reliability of the simulation results.

[0057] S1032 constructs a finite element mesh model, integrates thermodynamic and mechanical parameters, and generates a data set of material properties.

[0058] In this embodiment of the invention, a finite element model is constructed using a quadrilateral meshing method based on a structural parameter dataset. The surface of the firewall is divided into 96,000 elements with a side length of 5 mm. Thermodynamic parameters are retrieved from a material database, including the specific heat capacity of the refractory layer (850 J / kg Kelvin), coefficient of thermal expansion (8.5 × 10⁻⁶ Kelvin), and thermal conductivity (1.8 W / m Kelvin); the thermal conductivity of the insulation layer (0.035 W / m Kelvin) and specific heat capacity (1200 J / kg Kelvin); and the thermal conductivity of the load-bearing layer (45 W / m Kelvin). Mechanical parameters include the fracture toughness of the refractory layer (3.5 MPa²), the fatigue strength of the load-bearing layer (160 MPa), and the yield strength (355 MPa). These parameters are then integrated to form a firewall material property dataset, recording quantitative indicators of the thermodynamic and mechanical behavior of each layer, providing comprehensive parameter support for simulating jet erosion effects.

[0059] S1033 simulates the impact and heat conduction effects of electrolyte jetting, and calculates stress and temperature distribution.

[0060] In this embodiment of the invention, the jet impact load is calculated based on the electrolyte jet trajectory curve. The maximum pressure is 4 MPa, the duration is 300 milliseconds, and the stressed region is an ellipse with a major axis of 80 cm and a minor axis of 60 cm. A stress distribution matrix is ​​constructed to record the normal stress and shear stress of each grid element. The normal stress in the central region reaches 320 MPa, and the maximum shear stress at the edge is 85 MPa. The temperature field distribution matrix is ​​calculated using the heat conduction equation. The temperature in the central region reaches 650 degrees Celsius, decreasing by approximately 50 degrees Celsius every 10 cm outwards. The temperature gradient significantly affects the thermal stress of the material. The Mises stress criterion is used to calculate the stress intensity of each element. The intensity value in the central region reaches 375 MPa, exceeding the compressive strength of the refractory layer (280 MPa) and the yield strength of the load-bearing layer (355 MPa). This step, through multiphysics coupling analysis, realistically reproduces the synergistic damage mechanism of jet impact and high-temperature erosion.

[0061] S1034 Based on crack propagation theory and neural network prediction of damage distribution, the damage boundary of the partition wall is determined.

[0062] In this embodiment of the invention, when the stress intensity of a mesh element exceeds a material strength threshold, such as the compressive strength of the refractory layer or the yield strength of the load-bearing layer, it is marked as a damaged element. The damaged area is calculated using crack propagation theory, with a crack propagation rate of approximately 0.8 mm / s and a 75-degree angle between the main crack direction and the maximum principal stress. A damaged state feature vector containing the stress intensity factor, crack propagation rate, and direction is constructed and input into a convolutional neural network for prediction. The convolutional neural network outputs a damaged distribution map, showing a severely damaged area as a circular region with a diameter of 40 cm, and a moderately damaged area extending to an ellipse with a major axis of 90 cm and a minor axis of 70 cm. The coordinate set of the boundary points of the distribution map is extracted to accurately define the damaged area of ​​the partition wall.

[0063] In this embodiment of the invention, through high-precision data acquisition and multi-scale simulation, the system can dynamically capture the erosive effect of electrolyte spray on the firewall and generate a detailed damage distribution map. The resulting boundary coordinate set provides a direct basis for subsequent partition wall optimization design, effectively guiding the selection of refractory materials and structural reinforcement, significantly reducing the risk of fire breaking through the partition wall, thereby ensuring the safe operation of the energy storage container.

[0064] S104 collects the initial pressure wave parameters generated by the thermal runaway and deflagration of batteries in the damaged area of ​​the partition wall of the energy storage container, constructs a deflagration pressure wave propagation model to simulate its propagation path and intensity in the partition wall cracks, analyzes the impact range of the pressure wave on the adjacent battery pack, and identifies high-risk areas where the fire breaks through the isolation.

[0065] In this embodiment of the invention, high-frequency pressure sensors and acoustic emission sensors are used to collect deflagration pressure wave and stress wave data. Combined with crack morphology analysis and multiphysics modeling, the propagation behavior of pressure waves is accurately simulated to predict the risk area of ​​fire breaking through the partition wall.

[0066] S1041 Obtain the morphological parameters of the partition wall cracks and the characteristics of the deflagration pressure wave to construct the initial parameter matrix.

[0067] In this embodiment of the invention, based on the damage distribution data of the partition wall, crack morphology parameters are extracted, including a main crack length of approximately 250 mm, a maximum width of 1.8 mm, a depth penetrating a 120 mm thick partition wall, and a crack orientation at an angle of 65 degrees to the horizontal plane. A pressure sensor array is arranged around the crack at a sampling frequency of 10 kHz, recording a peak deflagration pressure wave of 12 MPa for a duration of 85 milliseconds.

[0068] Fast Fourier Transform (FFT) was used to perform spectral analysis on the pressure wave data, extracting the dominant frequency of 800-1200 Hz, the maximum amplitude of 8.5 MPa (located at 950 Hz), and the second harmonic frequency of 1900 Hz (amplitude 2.8 MPa). The initial root mean square (RMS) value of the pressure wave was 6.5 MPa, decaying exponentially over time with a decay coefficient of 0.15 per millisecond. The frequency components, amplitude components, and crack parameters were integrated to construct an initial parameter matrix for the pressure wave, providing a quantitative basis for subsequent propagation simulation. This step, through high-precision data acquisition and frequency domain analysis, clearly characterized the dynamic characteristics of the pressure wave.

[0069] S1042 constructs a computational fluid dynamics field to simulate the propagation path and intensity attenuation of pressure waves in a crack.

[0070] In this embodiment of the invention, a computational fluid dynamics model including pressure, velocity, and density fields is constructed by combining crack morphology parameters and the initial pressure wave parameter matrix. The initial pressure field is 12 MPa, a shock wave is formed at the crack inlet, the airflow velocity reaches 320 m / s, and the compressed gas density is 6.8 times that of standard atmospheric pressure. An adaptive mesh refinement technique is used to divide the crack region into 85,000 mesh cells with a minimum size of 0.2 mm, and the pressure wave propagation process is calculated by solving the Navier-Stokes equations.

[0071] The propagation path diagram shows that the pressure wave propagates at a speed of 320 m / s in the main fracture. Due to wall friction and reflection, its intensity decreases by 35% every 100 mm, dropping to 2.8 MPa at the fracture exit. The intensity attenuation curve quantifies the spatial attenuation law of the pressure wave, providing a basis for assessing the impact of nearby battery packs. This step, through refined modeling, realistically reproduces the propagation behavior of pressure waves in complex fractures.

[0072] S1043 collects stress wave data from nearby battery packs to predict stress distribution and damage probability.

[0073] In this embodiment of the invention, an acoustic emission sensor records stress waves on the surface of the adjacent battery pack, with an amplitude of 1.5 MPa and a duration of 45 milliseconds. The arrival time difference reflects the propagation characteristics of the medium. A long short-term memory network is used to predict the temporal stress wave data, generating a stress distribution cloud map. This shows that the battery pack within 150 mm of the crack exit is significantly affected, with a maximum stress of 280 MPa (located 80 mm from the exit). The battery pack casing strength parameters are extracted from the material database: tensile strength 320 MPa, compressive strength 385 MPa, and shear strength 195 MPa. Comparison reveals that the shear stress of some battery packs exceeds the limit, increasing the risk of damage. The cloud map, combined with strength thresholds, calculates the range of stress exceeding the limit, quantifying the degree of mechanical damage caused by the pressure wave and providing input for fire breakthrough prediction.

[0074] S1044 Constructs a fire breach prediction model to identify the boundaries of high-risk areas.

[0075] In this embodiment of the invention, a fire breakthrough prediction model is constructed using a support vector machine. Input features include a stress exceedance area of ​​420 square centimeters, a damage probability distribution slope of 0.08 per centimeter, and a pressure wave intensity attenuation rate of 0.15 per millisecond. The prediction results show that a fan-shaped high-risk area with a radius of 180 millimeters and an angle of 85 degrees is formed around the crack exit, with an internal battery pack chain reaction probability exceeding 75%. Further, a deep neural network is used to predict the thermal runaway trigger probability. Inputs include pressure wave intensity, temperature field (850 degrees Celsius at the center, decreasing by 120 degrees Celsius every 50 millimeters), and stress distribution, outputting a trigger probability distribution matrix. Regions with a probability exceeding 85% are extracted, constructing an irregular elliptical contour line with a major axis of 350 millimeters and a minor axis of 280 millimeters. A region segmentation algorithm divides the contour line into 2500 grid cells with 10-millimeter side lengths. Based on a spatial clustering algorithm, high, medium, and low risk levels are defined using pressure wave intensity of 0.8 to 12 MPa, temperature of 120 to 850 degrees Celsius, and trigger probability of 15% to 95%. The high-risk area covers a center of 180 mm, with pressure exceeding 6 MPa, temperature exceeding 600 degrees Celsius, and a trigger probability exceeding 80%.

[0076] S1045 assesses the risk of isolation facility failure and generates a spatial mapping of the risk area.

[0077] In this embodiment of the invention, a weight matrix is ​​constructed based on risk level regions. The weight coefficients are calculated by normalizing the pressure wave intensity, temperature field, and trigger probability, with weights of 0.4, 0.35, and 0.25, respectively, generating a weighted risk distribution map with risk values ​​ranging from 0 to 1. The stress distribution of the isolation facility is calculated using the equations of mechanics of materials, with maximum tensile stress of 350 MPa, compressive stress of 420 MPa, and shear stress of 180 MPa. Fracture mechanics criteria are used to compare the stress intensity factor with the square root of the material fracture toughness of 32 MPa to determine regions with a failure probability exceeding 90%. A random forest algorithm, using stress state, failure probability, and region weights as input, optimizes the risk boundary through 1000 iterations, generating a set of risk contour lines containing five contour lines corresponding to risk levels of 20%, 40%, 60%, 80%, and 95%. A spatial distribution dataset containing the coordinates of 3600 boundary points is constructed, recording spatial location, risk level, and inter-point connectivity. This data is visualized using geographic information system (GIS) technology, forming an intuitive spatial mapping of risk regions.

[0078] S105 collects temperature gradient distribution data of the partition wall within the risk zone of fire penetration of the energy storage container, quantifies the difference in thermal stress caused by the temperature gradient through thermal stress analysis algorithm, and assesses the cumulative degree of fatigue of the partition wall structure.

[0079] In this embodiment of the invention, high-precision infrared and acoustic emission sensors are used to collect temperature and stress fluctuation data of the partition wall. Combined with thermoelastic theory and machine learning models, the thermal stress distribution and fatigue damage evolution process are dynamically analyzed. Through multi-dimensional data fusion and refined calculation, the mechanical response of the partition wall under high-temperature conditions is accurately characterized, providing a scientific basis for improving the fire isolation performance of energy storage containers. This embodiment of the invention does not strictly limit the algorithm hyperparameters or sensor configurations; technicians can optimize them according to actual scenarios.

[0080] S1051 uses an infrared sensor array to scan the surface temperature of the partition wall, constructs a three-dimensional temperature distribution matrix, and extracts the temperature gradient field.

[0081] In this embodiment of the invention, an infrared sensor array is arranged on the surface of the partition wall at 5 mm intervals, with a sampling frequency of 10 Hz, to collect temperature data of the fire penetration area. The temperature in the central region reaches as high as 850 degrees Celsius, decreasing by approximately 120 degrees Celsius every 100 mm outwards. A temperature field reconstruction algorithm is used to generate a three-dimensional temperature distribution matrix containing 96,000 grid nodes based on sensor data interpolation, with each node recording a precise temperature value. Spatial partial derivative operations are performed on the matrix to generate a temperature gradient field dataset, with a maximum gradient of 15 degrees Celsius per millimeter, concentrated at the edge of the heat source. The temperature gradient field quantifies the degree of uneven heating of the partition wall, providing crucial input for subsequent thermal expansion and stress analysis. This step, through high-resolution data acquisition and mathematical modeling, ensures the accuracy of the temperature distribution description.

[0082] S1052 calculates the thermal expansion strain and thermal stress tensor, and generates a thermal stress difference distribution map.

[0083] In this embodiment of the invention, based on a temperature gradient field dataset, a thermal expansion strain matrix is ​​calculated, including normal strain (maximum value 0.015) and tangential strain (peak value 0.008), reflecting the deformation characteristics of the material at high temperatures. The thermoelastic constitutive equation is used to correlate the strain matrix with material parameters, such as the coefficient of thermal expansion (8.5 × 10⁻⁶). -6 The thermal stress tensor was calculated using a Kelvin-per-second method. The maximum normal stress component reached 380 MPa, occurring in the region of maximum temperature gradient; the peak shear stress was 180 MPa, concentrated at the interface between the refractory layer and the insulation layer. Further utilizing a deep neural network, with input temperature gradient, strain components, and material parameters, the spatial distribution of thermal stress was predicted, generating a thermal stress difference distribution map. The distribution map showed that the stress concentration region was an 80 mm wide annular band, with a stress gradient of 4.5 MPa per millimeter.

[0084] S1053 collects stress fluctuation data and performs cyclic statistics to construct a stress cycle counting matrix.

[0085] In this embodiment of the invention, an acoustic emission sensor monitors the stress fluctuations of the partition wall, recording a peak stress of 420 MPa at the initial heating stage, followed by stabilization within the range of 280 to 350 MPa. Stress waveform analysis extracts a peak-valley sequence containing 2800 data points, recording the stress amplitude variation. Rainflow counting is used to statistically analyze stress cycles, focusing on cycles with amplitudes from 35 to 85 MPa, with the 65 MPa amplitude showing the most cycles (850). The generated stress cycle counting matrix records the number of cycles at each amplitude level, quantifying the fatigue accumulation process of the partition wall under repeated thermal stress.

[0086] S1054 predicts the cumulative fatigue distribution of structures based on the linear cumulative damage criterion and fuzzy neural network.

[0087] In this embodiment of the invention, based on the stress cycle counting matrix, the initial cumulative damage value of 0.45 is calculated using the linear cumulative damage criterion. Considering the stress amplitude effect, this is corrected to 0.58, and further adjusted to 0.72 by combining the average stress effect, reflecting the intensification of damage under complex stress states. Material fatigue characteristic parameters show a fatigue limit of 180 MPa and a fatigue life curve slope of -0.12.

[0088] A fuzzy neural network was used to predict fatigue levels. Inputting cumulative damage values, cycle counts, and material parameters, the system outputs cumulative fatigue distribution data. Results showed that the fatigue damage level reached 0.85 in the central region, decreasing to 0.25 at the edges, forming a high-damage zone of 250 mm in diameter and a medium-damage zone of 400 mm in diameter. This step, through multi-level damage analysis, clearly revealed the spatial distribution characteristics of fatigue damage in the partition wall, providing a direct basis for firewall life assessment and optimized design. Through refined temperature and stress analysis, the system can dynamically capture the thermal stress and fatigue evolution process of the partition wall in the fire penetration zone, generating high-precision cumulative fatigue distribution data. The obtained distribution characteristics can guide the selection of partition wall materials and structural reinforcement, significantly reducing the failure risk under long-term high-temperature environments, thereby improving the fire isolation capability and operational safety of energy storage containers.

[0089] S106 predicts the remaining service life of the energy storage container partition structure based on the fatigue accumulation degree, and optimizes the material thickness and layered structure parameters in combination with the design service life requirements.

[0090] In this embodiment of the invention, the stress wave characteristics of the partition wall are monitored by an acoustic emission sensor. Combined with a long short-term memory network and a genetic algorithm, the remaining lifespan of the partition wall is dynamically predicted and its structural design is optimized. Through multi-dimensional data analysis and mechanical modeling, the impact of fatigue damage on the lifespan of the partition wall is accurately quantified, and targeted material and structural optimization schemes are generated, thereby extending the service life of the partition wall and enhancing the fire isolation capability of the energy storage container.

[0091] S1061 collects stress wave data of the partition wall and extracts features to construct a stress feature vector.

[0092] In this embodiment of the invention, acoustic emission sensors are arranged on the surface of the partition wall at a sampling frequency of 1 MHz, forming a monitoring network covering the entire damaged area. The collected stress wave signals show that the amplitude fluctuates between 0.5 and 4.5 volts, the main frequency is concentrated between 150 and 450 kHz, and the signal duration is between 20 and 150 microseconds. Using a waveform feature extraction algorithm, the raw signal is processed to generate a stress feature vector, containing key features such as a maximum amplitude of 2.8 volts, a center frequency of 280 kHz, and a cumulative duration of 85 microseconds. These features quantify the dynamic response of the partition wall under high temperature and stress, providing high-precision input data for subsequent life prediction.

[0093] S1062 uses a long short-term memory network to predict the remaining life of the partition wall, generates a life distribution map, and analyzes the material strength degradation.

[0094] In this embodiment of the invention, the stress feature vector, fatigue cumulative value of 0.72, and material parameters such as elastic modulus of 180 GPa are input into a long short-term memory network to predict the remaining service life of the partition wall. The network captures the long-term evolution of stress waves and fatigue damage through time series analysis and outputs a life prediction distribution map, showing that the remaining service life in the central region is only 35% of the design service life, while the remaining service life in the peripheral region reaches 85%.

[0095] To further assess material degradation, an acoustic attenuation detection method was used to scan the interior of the partition wall, generating a defect distribution map. This revealed the presence of microcracks with a maximum length of 0.8 mm in the central region, with a defect density of 8 per square centimeter, consistent with Weibull distribution characteristics. Based on defect evolution theory, the material strength degradation rate was calculated, showing a 5.8% strength decrease per 1000 hours in the central region and 2.2% in the edge region. The strength degradation curve provides a quantitative basis for optimized design. This step, combining machine learning and defect analysis, significantly improves the accuracy of life prediction.

[0096] S1063 constructs a thickness optimization objective function and uses a genetic algorithm to optimize the material thickness.

[0097] In this embodiment of the invention, based on material strength degradation rate and stress distribution data, an optimization objective function is constructed, comprising a stress distribution term (weight 0.45), a life constraint term (weight 0.35), and a thickness constraint term (weight 0.2). The stress distribution of each material layer is calculated; after optimization, the maximum stress is reduced by 25%, and the stress distribution is more uniform. A genetic algorithm is used to optimize the thickness. After 200 iterations, a thickness distribution scheme is generated: the fire-resistant layer is increased to 45 mm, the insulation layer is adjusted to 75 mm, and the load-bearing layer remains at 30 mm. This scheme balances mechanical performance and material cost, ensuring the partition wall maintains structural stability under high temperature and fatigue conditions.

[0098] S1064 optimizes the layered structural parameters of the partition wall, constructs the stress transfer function, and generates a description of its mechanical properties.

[0099] In this embodiment of the invention, the layered structure of the partition wall is further optimized based on the thickness distribution scheme. A stress transfer function is constructed using materials mechanics methods to describe the stress transfer characteristics between the fire-resistant layer, the insulation layer, and the load-bearing layer. The interface transfer coefficient between the fire-resistant layer and the insulation layer is 0.65, and that between the insulation layer and the load-bearing layer is 0.48. Combining stress wave propagation theory, the interlayer stress distribution is calculated, with wave velocities of 3200 m / s for the fire-resistant layer, 1800 m / s for the insulation layer, and 5400 m / s for the load-bearing layer.

[0100] A set of parameters for the layered structure was generated, including interlayer bond strength of 12 MPa, maximum interfacial normal stress of 280 MPa, peak shear stress of 95 MPa, and transfer coefficients ranging from 0.45 to 0.68. This parameter set comprehensively describes the mechanical behavior of the multi-layered structure, providing a precise basis for partition wall design optimization. This step, through mechanical modeling and parameter quantification, ensures the collaborative working capability of the layered structure. Through dynamic monitoring and multi-level optimization, the remaining lifespan of the partition wall can be accurately predicted, and scientific material and structural adjustment schemes can be generated. The optimized partition wall design significantly reduces the risk of fatigue failure, extends service life, and provides reliable protection for the safe operation of energy storage containers under extreme fire conditions.

[0101] S107 collects real-time monitoring data of deflagration pressure waves after adjusting the thickness and layered structure of the partition wall material of the energy storage container, optimizes the pressure release channel parameters, and uses multi-physics field coupling simulation to quantify the probability of partition wall crack generation.

[0102] By combining deep neural networks and fluid dynamics modeling, the pressure release channels of the partition wall are dynamically adjusted. At the same time, the risk of crack formation is predicted through fracture mechanics analysis. This enables precise control of the deflagration pressure wave and a comprehensive assessment of the mechanical response of the partition wall, effectively reducing the probability of fire breaching the isolation and providing technical support for the safe operation of energy storage containers.

[0103] S1071 uses a pressure sensor array to collect deflagration pressure wave data and constructs a pressure field distribution map.

[0104] In this embodiment of the invention, a pressure sensor array is uniformly arranged on the surface of the partition wall at 10 cm intervals, with a sampling frequency of 100 kHz, capturing a peak value of 12 MPa for the deflagration pressure wave, exhibiting a circular diffusion characteristic. The collected data includes pressure time series from 3600 monitoring points, with each series recording 1000 sampling points with millimeter-level precision. A deep neural network is used to process this data, with input features including pressure values, location coordinates, and time series, outputting a pressure field distribution map. The distribution map shows the attenuation law of the pressure wave propagating in space, with the highest pressure in the central region, gradually weakening outwards, with an attenuation rate of approximately 0.8 MPa per centimeter. This step, through the combination of high-frequency monitoring and machine learning, accurately characterizes the spatial dynamics of the deflagration pressure wave, providing a reliable foundation for subsequent multiphysics field analysis.

[0105] S1072 integrates temperature and pressure data to generate a multiphysics dataset and calculates the parameters of the pressure release channel.

[0106] In this embodiment of the invention, a temperature sensor array collects temperature field data on the surface of the partition wall. The temperature at the center of the deflagration reaches 850 degrees Celsius, decreasing by approximately 80 degrees Celsius every 50 millimeters outward, with a maximum temperature gradient of 15 degrees Celsius per millimeter. The temperature field and pressure field data are fused using a field value superposition method to generate a multiphysics dataset, which includes pressure field values, temperature field values, and their gradients.

[0107] Based on fluid dynamics equations, the parameters of the pressure relief channel were calculated, and a variable cross-section channel was designed with an inlet width of 80 mm, an outlet width of 120 mm, and a length of 400 mm. The cross-sectional area was increased from 50 cm² to 75 cm². Simulation of the pressure distribution within the channel showed a significant pressure gradient at the inlet and stable flow at the outlet. The dynamic controller employs a fuzzy PID strategy, adjusting based on the pressure difference and the maximum flow rate of 80 m³ / min monitored by the flow sensor. When the pressure difference exceeds 2 MPa, the channel opening is automatically increased.

[0108] S1073 constructs a multiphysics coupling model and extracts field intensity features to predict the probability of crack formation.

[0109] In this embodiment of the invention, a multi-physics coupling model is constructed by integrating the maximum stress value of 420 MPa collected by the stress sensor array and the additional stress peak of 180 MPa calculated by thermal stress, and constructing a model that includes a pressure field, a temperature field, and a stress field. This model considers the compressive heating effect of the pressure field on the temperature field, the softening effect of the temperature field on the material strength, and the interaction between the stress field and the pressure field.

[0110] A random forest algorithm was used to extract features from the coupled field intensity distribution map, including field intensity values, gradients, and fluctuation amplitudes, particularly in stress concentration regions where the field value fluctuation amplitude reached its peak. Based on fracture mechanics criteria, the crack propagation driving force was calculated; when the driving force exceeded the square root of 32 MPa, the material entered a damaged state. Probabilistic statistical analysis showed that the crack formation probability in stress concentration regions exceeded 85%, exhibiting a 60 mm wide band distribution. High-risk areas covered 15% of the grid points, mainly located at channel corners and stress concentration points. A crack formation probability distribution matrix was generated, recording the risk values ​​of 15,000 grid points. This step, through multi-field coupling and feature extraction, achieved accurate quantification of crack risk.

[0111] S108 dynamically adjusts the material ratio and structural parameters based on the probability of crack formation in the partition wall of the energy storage container to generate the final firewall design scheme.

[0112] In this embodiment of the invention, based on crack generation probability data, combined with a material database and a multi-level optimization algorithm, the material ratio and layered structure of the partition wall are dynamically adjusted to generate a firewall design scheme that takes into account fire resistance, heat insulation and structural stability. This significantly improves the crack resistance and long-term reliability of the partition wall, providing a scientific guarantee for the safe operation of energy storage containers.

[0113] S1081 constructs a material ratio optimization objective function and uses a genetic algorithm to generate the optimal ratio scheme.

[0114] In this embodiment of the invention, based on the probability data of crack generation in the partition wall, an optimization objective function is constructed, comprising a material strength term (weight 0.4), a thermal insulation performance term (weight 0.35), and a fire resistance performance term (weight 0.25). A parameter set is extracted from the material database, consisting of: refractory component aluminosilicate fiber (thermal conductivity 0.04 W / m Kelvin, density 180 kg / m³), thermal insulation component ceramic fiber felt (thermal conductivity 0.035 W / m Kelvin, density 120 kg / m³), and structural component stainless steel (elastic modulus 210 GPa, Poisson's ratio 0.3). A genetic algorithm is used for proportion optimization, with a population size of 200, 500 iterations, a crossover probability of 0.8, and a mutation probability of 0.1. The comprehensive performance of each proportion scheme is evaluated using a fitness function. The final optimal proportion scheme is generated: refractory component 35%, thermal insulation component 45%, structural component 20%, with the proportion error controlled within 2%.

[0115] S1082 generates the structural parameter matrix and calculates the stress distribution through finite element analysis.

[0116] In this embodiment of the invention, a structural parameter matrix is ​​constructed based on the optimal mix design to record the equivalent physical parameters of the composite material, including a density of 160 kg / m³, an elastic modulus of 85 GPa, and a Poisson's ratio of 0.28. The finite element method is used to simulate the stress distribution of the partition wall under deflagration pressure waves and high-temperature environments. The maximum stress value is 250 MPa, and the distribution is relatively uniform, with the stress gradient in the concentrated area being less than 2 MPa per millimeter. The finite element model is divided into 15,000 mesh elements, each with a side length of 5 millimeters, effectively capturing the spatial heterogeneity of the stress field.

[0117] S1083 predicts firewall layer thickness and calculates interlayer bonding characteristics.

[0118] In this embodiment of the invention, a deep neural network is used to predict the thickness of each layer of the firewall. The input features include stress distribution, temperature distribution (850 degrees Celsius at the center, decreasing by 80 degrees Celsius every 50 millimeters outward), and crack probability value. The output is the optimal thickness configuration: 45 mm for the fire-resistant layer, 75 mm for the heat insulation layer, and 30 mm for the structural layer, with the thickness tolerance controlled within 1 mm.

[0119] Further analysis of the interlayer interface acoustic properties using ultrasonic testing (5 MHz probe) yielded a sound velocity of 3200 m / s and an attenuation coefficient of 0.8 dB / mm, indicating a tight interlayer bond. Based on interface mechanics theory, the calculated interlayer bond strength reached 15 MPa, with uniform interface stress distribution and a maximum stress concentration factor of 1.8. This step, combining machine learning and non-destructive testing, optimized the layered structure and ensured the reliability of the interlayer mechanical properties.

[0120] S1084 evaluates isolation performance and generates a dataset of final firewall design schemes.

[0121] In this embodiment of the invention, an isolation performance evaluation matrix is ​​constructed, encompassing fire resistance, thermal insulation, and structural stability. The fire resistance limit is 1200 degrees Celsius, the equivalent thermal conductivity is 0.045 W / m Kelvin, and the maximum stress for structural stability is the deformation at 250 MPa. Structural integrity is assessed using acoustic emission testing with a 150 kHz sensor. The recorded structural damage level is below 0.15, crack density is less than 0.8 cracks per square centimeter, and maximum strain is 0.2%. A firewall design dataset is generated, including material proportion parameters, structural dimension parameters, and performance index parameters. The material proportion parameters represent the content of each component, and the structural dimension parameters represent the thickness of each layer, comprehensively meeting the fire integrity requirements. This scheme achieves excellent thermal insulation and fire resistance while ensuring structural strength, with reliable interlayer bonding and stable overall performance. This step, through multi-dimensional performance evaluation, has formed a complete firewall design scheme, providing a scientific basis for improving the fire isolation performance of energy storage containers, effectively reducing the risk of crack formation, improving the fire resistance and heat insulation of the partition wall, and providing a solid guarantee for the safe operation of energy storage containers under extreme fire conditions.

[0122] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for evaluating the fire resistance and thermal insulation performance of energy storage containers, characterized in that, The method includes: The layout, temperature, electrolyte state and pressure changes of the battery pack inside the energy storage container are obtained and preprocessed to obtain the initial triggering characteristics of battery thermal runaway. The initial triggering characteristics of battery thermal runaway are extrapolated and threshold judgment is performed to predict the time point of full thermal runaway. The jet trajectory of high-temperature electrolyte is obtained by combining the battery pack layout, and the erosion and damage area is determined based on the jet trajectory. The material distribution and structural parameters of the fire wall of the compartment in the area damaged by jet erosion were obtained. The erosion effect of high-temperature electrolyte jet on the partition wall was simulated by the finite element analysis algorithm to obtain the distribution range of damage to the partition wall. The initial pressure wave parameters generated by the thermal runaway and deflagration of batteries within the distribution range of the damaged partition wall were obtained. A deflagration pressure wave propagation model was constructed to simulate the propagation path and intensity of the pressure wave in the partition wall cracks. Based on the propagation path and intensity, the influence range of the pressure wave propagation on the adjacent battery pack was identified, and the risk area of ​​fire breaking through the isolation was determined. Obtain temperature gradient distribution data of the partition wall in the risk zone where fire breaks through the isolation, calculate the difference in thermal stress caused by the temperature gradient, and obtain the degree of structural fatigue accumulation; The remaining service life of the partition wall structure is predicted based on the degree of cumulative structural fatigue, and the material thickness and layering structure of the partition wall are adjusted according to the design service life requirements. Real-time monitoring data of deflagration pressure waves were obtained after adjusting the material thickness and layered structure of the partition wall. The pressure release channels inside the partition wall were adjusted, and a multi-physics field coupled simulation algorithm was used to obtain the probability of partition wall crack generation. Based on the probability of cracks forming in the partition wall, the material ratio and structural parameters of the partition wall are dynamically updated to generate the final firewall design scheme.

2. The method for evaluating the fire resistance and thermal insulation performance of an energy storage container according to claim 1, characterized in that, The process of acquiring the battery pack layout, temperature, electrolyte state, and pressure changes inside the energy storage container, and preprocessing them to obtain the initial triggering characteristics of battery thermal runaway, includes: The battery pack's three-dimensional layout parameter set is obtained by processing data collected by sensors inside the energy storage container through a deep perception algorithm. The temperature field data is then processed by noise reduction and filtering to obtain the battery pack temperature dataset, which includes surface temperature values ​​and temperature gradient values. Calculate the rate of change matrix based on the battery pack temperature dataset and pressure state feature set. When an element in the rate of change matrix exceeds a preset mutation threshold, mark the corresponding position as an anomaly. A heat transfer matrix is ​​constructed based on the location of the anomaly and the three-dimensional layout parameter set of the battery pack. The random forest algorithm is used to perform correlation analysis on the anomaly to obtain the location of the point with the highest probability of thermal runaway, thus forming the initial triggering feature of battery thermal runaway.

3. The method for evaluating the fire resistance and thermal insulation performance of an energy storage container according to claim 1, characterized in that, The process involves trend extrapolation and threshold judgment of the initial triggering characteristics of battery thermal runaway to estimate the time point of full-scale thermal runaway. Combined with the battery pack layout, the jet trajectory of the high-temperature electrolyte is obtained, and the erosion and damage area is determined based on the jet trajectory, including: A thermal runaway parameter matrix is ​​constructed based on battery pack temperature and pressure data. The thermal runaway parameter matrix includes temperature change rate and pressure change rate. When the temperature change rate exceeds a preset temperature change threshold, the change time and change location data are obtained. The thermal runaway propagation rate is calculated using the data of the abrupt change time and location. A thermal runaway propagation path diagram is constructed based on the thermal runaway propagation rate and battery pack layout parameters. The direction of heat transfer is obtained by calculating the temperature gradient between adjacent battery packs. Based on the temperature field distribution in the thermal runaway diffusion path diagram, the electrolyte injection parameters are calculated, and the electrolyte injection trajectory curve is constructed using the injection parameters. Position sequence data is extracted from the electrolyte jet trajectory curve, Fourier transform is performed on the temperature gradient data on the position sequence, and the time point of full thermal runaway is determined by comparing the frequency domain feature value with the preset thermal runaway spectrum threshold. The probability distribution of battery pack damage is predicted by using a neural network, and the boundary points of high probability areas are extracted to determine the erosion and damage areas.

4. The method for evaluating the fire resistance and thermal insulation performance of an energy storage container according to claim 1, characterized in that, The process involves obtaining the material distribution and structural parameters of the bulkhead firewall in the area damaged by jet erosion, simulating the erosion effect of high-temperature electrolyte jet on the firewall using a finite element analysis algorithm, and obtaining the damage distribution range of the firewall, including: Firewall structural mesh data is obtained from the firewall 3D scan data. The structural mesh data is then used to identify the material layer structure map through deep learning algorithm to obtain the firewall structural parameter dataset. A finite element mesh model was constructed using the aforementioned firewall structural parameter dataset. The thermodynamic parameter set and mechanical parameter set were retrieved from the material database to obtain the firewall material property dataset. Based on the firewall material property dataset, the wall surface is decomposed into elements, the temperature field distribution matrix of the grid elements is calculated by the heat conduction equation, and the stress intensity value of the grid elements is calculated by the Mises stress criterion. When the stress intensity value of the mesh element exceeds the material strength threshold parameter, it is marked as a damaged element. The expansion range of the damaged element is calculated using crack propagation theory. The damage distribution map of the firewall is predicted by a convolutional neural network, and the coordinate set of the boundary points of the damaged distribution range of the partition wall is obtained.

5. The method for evaluating the fire resistance and thermal insulation performance of an energy storage container according to claim 1, characterized in that, The process involves obtaining the initial pressure wave parameters generated by the thermal runaway and deflagration of batteries within the damaged partition wall area, constructing a deflagration pressure wave propagation model, simulating the propagation path and intensity of the pressure wave in the partition wall cracks, identifying the impact range of the pressure wave propagation on adjacent battery packs based on the propagation path and intensity, and determining the risk area where the fire breaks through the isolation barrier, including: Collect deflagration pressure wave data to obtain a pressure wave characteristic dataset; Spectral analysis is performed on the pressure wave feature dataset to extract the pressure wave frequency components and amplitude components from the pressure wave feature dataset, construct the pressure wave initial parameter matrix, and generate the pressure wave propagation path map based on the pressure wave initial parameter matrix; Long Short-Term Memory (LSTM) networks are used to predict stress wave data of the battery pack to obtain stress distribution cloud maps, and the stress exceedance range is calculated based on the stress distribution cloud maps. A fire breakthrough prediction model is constructed using support vector machines. The input features of the prediction model include the stress exceedance range, the damage probability distribution, and the pressure wave intensity attenuation rate, resulting in the coordinate set of boundary points of the fire breakthrough isolation risk area.

6. The method according to claim 5, characterized in that, Also includes: The pressure wave's effective range is calculated based on its propagation path. The probability of thermal runaway triggering near adjacent battery packs is analyzed. Based on this probability, a set of fire breach areas is identified. Weighted risk analysis is performed on this set to assess the failure risk of isolation facilities, calculate the isolation risk distribution, obtain the risk area distribution, generate a spatial mapping of isolation risks, visualize the risk areas, and determine the specific risk areas, including: The pressure wave attenuation curve is obtained based on the pressure wave propagation path data. The temperature field distribution matrix within the pressure wave's effective range is calculated using the heat conduction equation. The trigger probability distribution matrix is ​​predicted using a deep neural network. The input features of the deep neural network include pressure wave intensity value, temperature field value, and stress distribution value. Based on the trigger probability distribution matrix, the coordinates of probability threshold points are extracted to construct the outline of the fire breakthrough area, and the area within the outline is divided into grids to obtain a set of grid cells. The set of grid cells is classified to obtain risk level regions. The spatial clustering algorithm calculates the clustering distance based on the grid cell pressure wave intensity value, temperature field value, and trigger probability value. The stress distribution matrix of the isolation facility is calculated based on the risk level area, and the failure probability of the isolation facility is determined by the fracture mechanics criterion. The fracture mechanics criterion determines the boundary of the risk area based on the comparison between the stress intensity factor and the critical strength.

7. The method for evaluating the fire resistance and thermal insulation performance of an energy storage container according to claim 1, characterized in that, The process of acquiring temperature gradient distribution data of the partition wall in the risk zone where fire breaks through the isolation, calculating the difference in thermal stress caused by the temperature gradient, and obtaining the cumulative degree of structural fatigue includes: Temperature scanning is performed on the surface of the partition wall, and a three-dimensional temperature distribution matrix is ​​obtained through a temperature field reconstruction algorithm. Spatial partial derivative calculation is performed based on the temperature distribution matrix to obtain a temperature gradient field dataset, which is used to calculate the thermal expansion strain matrix. The thermal stress tensor is obtained by calculating the thermal expansion strain matrix and the thermoelastic constitutive equation. The thermal stress tensor includes normal stress components and shear stress components. A deep neural network is used to predict the spatial distribution of the thermal stress tensor to obtain a thermal stress difference distribution map. Stress cycle statistics are then performed based on the thermal stress difference distribution map to obtain a stress cycle counting matrix. The cumulative fatigue distribution data of the structure is obtained by calculating using the stress cycle counting matrix and the linear cumulative damage criterion.

8. The method for evaluating the fire resistance and thermal insulation performance of an energy storage container according to claim 1, characterized in that, The method of predicting the remaining service life of the partition wall structure based on the cumulative degree of structural fatigue, and adjusting the material thickness and layering structure of the partition wall according to the design service life requirements, includes: Stress waveform features are extracted from structural stress wave data to obtain a stress feature vector composed of stress wave amplitude, frequency, and duration. Based on the stress feature vector, fatigue accumulation value and material parameter value, input to the long short-term memory network, a predicted distribution map of the remaining service life of the structure is obtained. The predicted distribution map records the remaining service life value at each point. The material internal defect distribution map is obtained by acoustic attenuation detection method, and the material strength attenuation curve is calculated based on the lifetime prediction distribution map and the defect distribution map. A genetic algorithm is used to optimize the material thickness. The genetic algorithm iteratively optimizes the material thickness distribution scheme based on the stress distribution map and life constraints to obtain the material thickness distribution scheme. The scheme includes the thickness values ​​of each layer. The layer structure parameters are optimized according to the material thickness distribution scheme.

9. The method for evaluating the fire resistance and thermal insulation performance of an energy storage container according to claim 1, characterized in that, The process involves acquiring real-time monitoring data of the deflagration pressure wave after adjusting the material thickness and layered structure of the partition wall, adjusting the pressure release channels inside the partition wall, and using a multi-physics coupling simulation algorithm to obtain the probability of crack formation in the partition wall, including: Pressure monitoring point data is acquired through a pressure sensor array. The pressure monitoring point data is processed by a deep neural network to obtain a pressure field distribution map. The input features of the neural network include pressure value, location coordinate value and time series value. A multiphysics dataset is obtained by overlaying the pressure field distribution map with data collected by the temperature sensor array. The pressure release channel parameter matrix is ​​then calculated based on the multiphysics dataset. The channel parameters include channel width, channel length, and channel cross-sectional area. Based on the channel parameter matrix, feature extraction is performed on the coupled field intensity distribution map, and the crack generation probability distribution matrix is ​​calculated using fracture mechanics criteria. The matrix records the crack generation probability value at each point.

10. The method for evaluating the fire resistance and thermal insulation performance of an energy storage container according to claim 1, characterized in that, The method of dynamically updating the material ratio and structural parameters of the partition wall based on the probability of crack generation, and generating the final firewall design scheme, includes: The contents of refractory components, thermal insulation components, and structural components are obtained from the material database. The component contents are used to construct an objective function for optimizing the material proportions. The objective function includes material strength, thermal insulation performance, and fire resistance. A genetic algorithm is used to iteratively optimize the objective function, and a material composition ratio scheme is obtained based on the crack generation probability value and material performance index. The ratio scheme includes the optimal ratio value of each component. A structural parameter matrix is ​​constructed based on the material composition ratio scheme, and stress distribution data is obtained through finite element method. The stress distribution data is calculated using a deep neural network. The input features of the neural network include stress distribution values, temperature distribution values, and crack probability values. The interlayer bonding strength value and interface stress distribution value are obtained through interface mechanics theory, and the final firewall design scheme is generated.

Images