A method of overheat management of a battery

Abstract

The application relates to the technical field of battery management, and particularly discloses a battery overheat management method, which comprises the following steps: determining an initial temperature gradient field of a battery in an operation process; determining a final temperature gradient field and a heat flow velocity distribution matrix according to the initial temperature gradient field; if it is judged from the final temperature gradient field that the temperature is unevenly distributed, taking the final temperature gradient field and the heat flow velocity distribution matrix as inputs, performing prediction through a pre-trained convolutional neural network, outputting a predicted temperature distribution matrix, and determining a local overheating area according to the predicted temperature distribution matrix; performing weighted fusion on the number of the local overheating areas and the temperature average value of all the local overheating areas to determine an overheating risk coefficient, and acquiring corresponding control parameters of a heat dissipation system according to the interval where the overheating risk coefficient is located. The battery overheat management method solves the problem that uneven temperature distribution in the operation of the battery may cause local overheating and further cause a safety hazard.

Description

Technical Field

[0001] This invention relates to the field of battery management technology, and more specifically, to a method for managing battery overheating. Background Technology

[0002] With the widespread application of energy storage devices such as UPS, effectively controlling battery temperature and avoiding risks caused by overheating has become a core issue for the industry. Battery overheating can not only lead to performance degradation but also pose safety hazards; therefore, researching efficient thermal management methods is particularly urgent.

[0003] Currently, although many thermal management solutions have been implemented, a common problem exists: most methods focus more on the overall battery temperature while neglecting the dynamic processes of heat generation and transfer within the battery. This neglect means that thermal management strategies often fail to adapt to the battery's actual thermal state under different operating conditions, thus affecting heat dissipation efficiency and battery stability. However, the heat distribution inside a battery is not uniform; the rate of temperature change and heat flow in different areas is affected by various factors, such as the characteristics of the cell materials and changes in operating conditions. Therefore, the rate of heat dissipation and the path of heat transfer inside the battery are difficult to accurately capture and analyze, making real-time response impossible for thermal management. For example, when the battery is operating at high power, some areas may heat up rapidly, but because the heat transfer path is unclear, the cooling system cannot determine suitable parameters, resulting in the inability to dissipate heat in time, leading to localized overheating and even affecting the safety of the entire battery pack. Summary of the Invention

[0004] In order to overcome the defects of the prior art, the present invention provides a battery overheat management method, which aims to solve the problems in the prior art.

[0005] The technical solution adopted by this invention to solve its technical problem is: a battery overheat management method, comprising the following steps: S1: Determine the initial temperature gradient field inside the battery during operation based on real-time multi-point temperature data; S2: Based on the initial temperature gradient field, a thermal simulation model is used to calculate the heat conduction and convection process in the battery material, and the final temperature gradient field and heat flow velocity distribution matrix are determined. S3: If the temperature is determined to be non-uniform from the final temperature gradient field, the final temperature gradient field and the heat flow velocity distribution matrix are used as inputs, and a pre-trained convolutional neural network is used to make predictions. The predicted temperature distribution matrix is ​​output, and the local overheated area is determined based on the predicted temperature distribution matrix. S4: Weighted fusion of the number of local overheated areas and the average temperature of all local overheated areas is performed to determine the overheating risk coefficient. Based on the interval in which the overheating risk coefficient is located, the corresponding control parameters of the heat dissipation system are obtained.

[0006] Preferably, in step S1, multi-point temperature monitoring signals are obtained from a temperature sensor array deployed inside the battery, the multi-point temperature monitoring signals including temperature value T and corresponding spatial coordinate X; For each point p, obtain all points k adjacent to point p from the multi-point temperature monitoring signal, and calculate the temperature difference. ,in Let k be the temperature value at point k. Let p be the temperature value, and the distance be... coordinates With coordinates The Euclidean distance, where Let k be the spatial coordinates. Let P be the spatial coordinates of point P. Calculate the gradient magnitude between point P and point K. Divide by The average of the gradient magnitudes of all points p and k is calculated as the magnitude of the temperature gradient feature. Points k with the largest gradient magnitude are extracted and paired with point p. Based on the coordinates... With coordinates The direction of the point pair is calculated as the direction of the temperature gradient feature, and the temperature gradient feature of point p is obtained. Based on the temperature gradient characteristics of all points, the points are arranged according to their coordinates to construct the initial temperature gradient field of the battery thermal distribution.

[0007] Specifically, in step S1, a spatial coordinate grid is determined from the multi-point temperature monitoring signal, and the temperature value of each grid point is calculated using Kriging interpolation to generate the processed multi-point temperature monitoring signal. For each point p in the processed multi-point temperature monitoring signal, determine its corresponding temperature gradient characteristics.

[0008] It is worth noting that in step S2, the inside of the battery is divided into grids, the boundaries of each grid after grid division are obtained, and the temperature gradient features of all points of the corresponding grid after grid division are extracted from the initial temperature gradient field based on the grid boundaries. The average value of the temperature gradient characteristic magnitude of all points in each grid is calculated as the temperature gradient vector magnitude of that grid; the temperature gradient direction at the center point of the grid is taken as the temperature gradient vector direction of that grid. The final temperature gradient field is obtained by arranging the temperature gradient vectors of all grids according to the grid arrangement order.

[0009] Preferably, in step S2, the final temperature gradient field is imported into OpenFOAM as the initial field, the heat flow velocity vector corresponding to each grid is initialized to a zero vector, the heat flow velocity vector is iteratively updated using the PISO algorithm until the residual is less than a preset residual threshold, and all heat flow velocity vectors are arranged according to the grid arrangement order to obtain the heat flow velocity distribution matrix.

[0010] Optionally, in step S2, a pre-built battery geometric model is obtained from finite element analysis software, and the thickness of each layer of the battery multilayer composite material is extracted. Create a corresponding free tetrahedral mesh for each layer of composite material from the battery geometry model to obtain a coarse mesh distribution containing multiple first meshes; For each first grid in the coarse grid distribution, the maximum grid cell size is determined based on the thickness of the thickest layer, and the minimum grid cell size is determined based on the thickness of the thinnest layer. The grid cell sizes of other layers are linearly interpolated according to the thickness ratio, where the thickness ratio is defined as the ratio of the current layer thickness to the thickness of the thickest layer, to obtain the second grid size corresponding to each layer. The first grid in the coarse grid distribution is then divided according to the second grid size to obtain the preliminary grid division. The average value of the magnitude of the temperature gradient feature of all points in each second grid after the initial grid division is obtained from the initial temperature gradient field. For the second grid whose average magnitude is greater than or equal to the gradient threshold, the second grid is divided into third grids with a size 0.5 times the original size, thus obtaining the final grid division.

[0011] Specifically, in step S3, the temperature gradient characteristic modulus of each grid is obtained from the final temperature gradient field, and the variance of the temperature gradient characteristic modulus is calculated. If the variance exceeds a preset uniformity threshold, it is determined that the temperature is non-uniformly distributed. Then, the heat flow velocity vector magnitude for each grid in the final grid division is extracted from the heat flow velocity distribution matrix, and the temperature gradient feature magnitude for each grid in the final grid division is extracted from the final temperature gradient field. These features are then paired and combined into two-dimensional vector features, and finally arranged into a vector matrix according to the grid arrangement order. The vector matrix is ​​input into a pre-trained convolutional neural network, and the temperature value of each grid in the final grid division is calculated through forward propagation to determine the temperature distribution matrix in the next time step. The average temperature of each grid and all its neighboring grids is calculated from the predicted temperature distribution matrix as the local average temperature. If the local average temperature exceeds the preset overheating threshold, the grid is determined to be a local overheating region.

[0012] It is worth noting that in step S4, the number of local overheated areas is obtained, the average local temperature of each local overheated area is used as the temperature parameter of that area, and the average temperature value is obtained by calculating the average of the temperature parameters of all local overheated areas. Based on the number and average temperature of the aforementioned local overheated areas, a weighted fusion calculation is performed using preset weights to obtain the overheating risk coefficient; Based on the range of the overheating risk coefficient, obtain the corresponding control parameters of the heat dissipation system from the heat dissipation parameter table.

[0013] The beneficial effects of this invention are as follows: In the battery overheat management method, an initial temperature gradient field is constructed by real-time acquisition of multi-point temperature data. Combined with a thermal simulation model, heat conduction and convection processes are calculated to generate a final temperature gradient field and a heat flow velocity distribution matrix. When a non-uniform temperature distribution is detected, a pre-trained convolutional neural network is used to predict the temperature distribution matrix, accurately locating local overheating areas. An overheating risk coefficient is calculated by weighted fusion of the number of overheating areas and the average temperature. Finally, the heat dissipation control parameters are dynamically adjusted based on the risk coefficient range. This solution achieves intelligent management of the entire process from temperature gradient field analysis to overheating risk assessment and heat dissipation regulation, significantly improving the safety and stability of battery operation. Attached Figure Description

[0014] Figure 1 This is a flowchart of a battery overheat management method.

[0015] Figure 2 This is a flowchart of the steps in step S2.

[0016] Figure 3 This is a schematic diagram of a temperature sensor array deployed inside a battery. Detailed Implementation

[0017] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings. It should be noted that these descriptions are for the purpose of aiding understanding the present invention, but do not constitute a limitation thereof. Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0018] Combination Figures 1 to 3 The method for managing overheating of a battery, as shown, includes the following steps: S1: Determine the initial temperature gradient field inside the battery during operation based on real-time multi-point temperature data; S2: Based on the initial temperature gradient field, a thermal simulation model is used to calculate the heat conduction and convection process in the battery material, and the final temperature gradient field and heat flow velocity distribution matrix are determined. S3: If the temperature is determined to be non-uniform from the final temperature gradient field, the final temperature gradient field and the heat flow velocity distribution matrix are used as inputs, and a pre-trained convolutional neural network is used to make predictions. The predicted temperature distribution matrix is ​​output, and the local overheated area is determined based on the predicted temperature distribution matrix. S4: Weighted fusion of the number of local overheated areas and the average temperature of all local overheated areas is performed to determine the overheating risk coefficient. Based on the interval in which the overheating risk coefficient is located, the corresponding control parameters of the heat dissipation system are obtained.

[0019] In the battery overheat management method, an initial temperature gradient field is constructed by real-time acquisition of multi-point temperature data. This is combined with a thermal simulation model to calculate heat conduction and convection processes, generating a final temperature gradient field and a heat flow velocity distribution matrix. When a non-uniform temperature distribution is detected, a pre-trained convolutional neural network is used to predict the temperature distribution matrix, accurately locating local overheating areas. An overheating risk coefficient is calculated by weighted fusion of the number of overheating areas and the average temperature. Finally, the heat dissipation control parameters are dynamically adjusted based on the risk coefficient range. This solution achieves intelligent management throughout the entire process, from temperature gradient field analysis to overheating risk assessment and heat dissipation regulation, significantly improving the safety and stability of battery operation.

[0020] It is worth noting that in step S1, multi-point temperature monitoring signals are obtained from the temperature sensor array deployed inside the battery. The multi-point temperature monitoring signals include temperature value T and corresponding spatial coordinate X. For each point p, obtain all points k adjacent to point p from the multi-point temperature monitoring signal, and calculate the temperature difference. ,in Let k be the temperature value at point k. Let p be the temperature value, and the distance be... coordinates With coordinates The Euclidean distance, where Let k be the spatial coordinates. Let P be the spatial coordinates of point P. Calculate the gradient magnitude between point P and point K. Divide by The average of the gradient magnitudes of all points p and k is calculated as the magnitude of the temperature gradient feature. Points k with the largest gradient magnitude are extracted and paired with point p. Based on the coordinates... With coordinates The direction of the point pair is calculated as the direction of the temperature gradient feature, and the temperature gradient feature of point p is obtained. Based on the temperature gradient characteristics of all points, the points are arranged according to their coordinates to construct the initial temperature gradient field of the battery thermal distribution.

[0021] For example, deploying a high-density thin-film temperature sensor array inside the battery enables real-time capture of thermal dynamics at different levels of the cell. In this solution, a prismatic lithium iron phosphate battery is used. For instance, inside the prismatic lithium iron phosphate battery, a 3x3x3 sensor matrix is ​​arranged at 5mm intervals, with each point outputting data including temperature and three-dimensional spatial coordinates. In one possible implementation, when the battery is in fast-charging mode, the target point at the center has coordinates of 10,10,10 and a measured temperature of 45.2 degrees Celsius. A search algorithm is used to find its six nearest Euclidean neighbors. If one of these neighbors has coordinates of 10,10,15 and a temperature of 48.2 degrees Celsius, the calculated temperature difference is 3.0 degrees Celsius, the distance is 5.0mm, and the gradient magnitude in that direction is 0.6 degrees per millimeter. In another possible implementation, by traversing all neighboring points to calculate the gradient magnitude and taking the average, local fluctuations caused by sensor noise can be effectively smoothed. For example, if the gradient magnitudes around the target point are 0.6, 0.4, 0.5, etc., and their average value of 0.5 is calculated as the characteristic magnitude of the temperature gradient at that point, this quantization method can reflect the intensity of heat conduction in that region.

[0022] It should be noted that, in order to determine the core direction of heat flow, the point pairs corresponding to the maximum gradient magnitude are extracted. If the gradient magnitude between the target point and its adjacent point reaches the maximum value, the vector pointing from the target point to that adjacent point is taken as the temperature gradient feature direction of that point. Specifically, combining the calculated gradient magnitude with this direction vector completely defines the temperature gradient feature of that point. In one possible implementation, the above calculation process is repeated for all points within the battery to obtain a multidimensional dataset containing spatial coordinates and temperature gradient features. It is worth noting that if there are multiple point pairs corresponding to the maximum gradient magnitude, the x-coordinate of point k is compared first, and the point k with the smallest x-coordinate is selected. If there are still multiple corresponding point pairs, the y-coordinate of the selected point k is compared, and the point k with the smallest y-coordinate is selected. If there are still multiple corresponding point pairs, the z-coordinate of the selected point k is compared, and the point k with the smallest z-coordinate is selected. Finally, the direction is calculated for the point pair corresponding to the unique selected point k.

[0023] Preferably, these features are arranged in a matrix according to their spatial coordinates to construct an initial temperature gradient field for the battery's thermal distribution. This refined field construction provides a more accurate input for battery thermal management strategies than single-point temperature, significantly improving the safety and heat dissipation efficiency of the battery system.

[0024] Preferably, in step S1, a spatial coordinate grid is determined from the multi-point temperature monitoring signal, and the temperature value of each grid point is calculated using Kriging interpolation to generate the processed multi-point temperature monitoring signal. For each point p in the processed multi-point temperature monitoring signal, determine its corresponding temperature gradient characteristics.

[0025] For acquiring multi-point temperature monitoring signals in a temperature sensor array, these sensors collect temperature values ​​and corresponding three-dimensional spatial coordinates in real time. For example, one sensor located at coordinates (10,10,10) measures a temperature of 45.2℃, while another located at (10,10,15) measures 48.2℃, thus forming a set of discrete points.

[0026] Kriging interpolation is a geostatistical method based on spatial statistics. Its principle is to estimate the value of unknown points by using the temperature values ​​of known grid points and their spatial correlations. Specifically, it first calculates the semivariogram values ​​of the distances h between multiple known points. , Let h be the number of point pairs that are h distances from a known point i. Given the temperature at point i, The temperature of a known point at a distance h from a known network point i; a Gaussian model is fitted using these semivariogram values. Result in the value of the gold nugget Structural variance and variable range The value of is then calculated using a Gaussian model to determine the semivariogram value between known point i and known point j. and the semivariogram values ​​of known point i and the point to be interpolated 0 The distance between known point i and known point j (i.e., h in the Gaussian model) is obtained by calculating the Euclidean distance between known point i and known point j, and the distance between known point i and the insertion point 0 (i.e., h in the Gaussian model) is obtained by calculating the Euclidean distance between known point i and the insertion point 0. Then, the formula is used... Find the weight of a given point i. ,in, Let i be the semivariogram value between known point i and known point j. Given the semivariogram values ​​of point i and the point to be interpolated 0, These are Lagrange multipliers. Finally, calculate the temperature at the interpolation point 0. ,in , … All temperatures are known points. , … The weights are assigned to the corresponding known points. This method considers spatial autocorrelation, is more accurate than simple linear interpolation, and can effectively handle regions with large temperature gradients. Finally, the temperature values ​​of all points are combined to generate a processed multi-point temperature monitoring signal to reflect the temperature situation inside the battery at that moment.

[0027] Optionally, in step S2, the inside of the battery is divided into grids, the boundaries of each grid are obtained, and the temperature gradient features of all points of the corresponding grid after grid division are extracted from the initial temperature gradient field based on the grid boundaries. The average value of the temperature gradient characteristic magnitude of all points in each grid is calculated as the temperature gradient vector magnitude of that grid; the temperature gradient direction at the center point of the grid is taken as the temperature gradient vector direction of that grid. The final temperature gradient field is obtained by arranging the temperature gradient vectors of all grids according to the grid arrangement order.

[0028] In one possible implementation, when meshing the interior of the battery, finite element analysis software can be used to generate a structured mesh. During the process of obtaining the boundaries of each mesh after meshing, the boundary surface coordinates of each mesh are extracted through the finite element analysis software interface, thereby defining its geometric boundary. Specifically, this boundary extraction facilitates subsequent data mapping; for example, the boundary of a mesh consists of four faces, ensuring alignment with the initial temperature gradient field.

[0029] In one possible implementation, when extracting the temperature gradient features of all points within a given grid from an initial temperature gradient field based on the grid's boundaries, a pre-defined temperature gradient distribution data is first loaded. Then, for each grid, all points within it are traversed, and the temperature gradient features at the same coordinates as these points are obtained. This extraction method yields a list of temperature gradient features for all points within the grid, providing foundational data for averaging calculations. The key to this method lies in boundary constraints, ensuring that only points within the grid are processed, avoiding external interference.

[0030] When calculating the average of the temperature gradient feature magnitudes of all points in each grid as the temperature gradient vector magnitude of that grid, the magnitudes of the temperature gradient features in each temperature gradient feature list are calculated, and then averaged. This average value is used as the temperature gradient vector magnitude of that grid. In one possible implementation, when the temperature gradient direction at the center point of the grid is used as the temperature gradient vector direction of that grid, for example, if the gradient at the center point points to the positive x-axis, then the direction of the entire grid is defined accordingly.

[0031] Finally, the magnitude and direction of the temperature gradient vector are combined into a single temperature gradient vector. The temperature gradient vectors of all meshes are then arranged according to their grid order to obtain the final temperature gradient field, which is used for subsequent heat transfer simulations. This process improves the accuracy of the temperature gradient field analysis.

[0032] Preferably, in step S2, the final temperature gradient field is imported into OpenFOAM as the initial field, the heat flow velocity vector corresponding to each grid is initialized to a zero vector, the heat flow velocity vector is iteratively updated using the PISO algorithm until the residual is less than a preset residual threshold, and all heat flow velocity vectors are arranged according to the grid arrangement order to obtain the heat flow velocity distribution matrix.

[0033] In the battery thermal simulation, OpenFOAM software was used to process the final temperature gradient field data. OpenFOAM is an open-source computational fluid dynamics toolkit that supports multiphysics coupling simulations, including heat transfer and fluid dynamics analysis.

[0034] In one possible implementation, the final temperature gradient field is imported as the initial field using the open-source computational fluid dynamics software OpenFOAM. Specifically, OpenFOAM defines initial conditions through field files in its 0 directory. For example, the temperature gradient field data is written to a T file in ASCII or binary format, where the gradient vector components of each grid cell, such as 0.8 degrees per micrometer in the x-direction and 0.5 degrees per micrometer in the y-direction, are precisely recorded, thus providing an initial distribution for the battery heat transfer simulation. Then, the heat flow velocity vector corresponding to each grid cell is initialized to a zero vector. For example, the initial value of the U field is set to (000) in the system / controlDict file, ensuring that the velocity components of all grid cells inside the battery are zero, which establishes a static ground state for subsequent iterations.

[0035] Specifically, the heat flow velocity vector is iteratively updated using the PISO algorithm until the residual is less than a preset residual threshold, such as 0.000001. The PISO algorithm is a pressure implicit splitting operator method used to handle the pressure-velocity coupling problem in the unsteady Navier-Stokes equations. Its core includes multi-stage iterations such as prediction step, non-orthogonal correction step, and pressure correction step.

[0036] In battery simulation, velocity prediction is first initiated based on the buoyancy source term driven by the temperature gradient field. For example, a Boussinesq buoyancy model is added in fvOptions to convert the temperature difference into velocity perturbations. Then, pressure is solved, and the GAMG solver is used to optimize the sparse matrix. Next, the velocity field is corrected to satisfy the continuity equation. This process is repeated, with each step advancing the time by 0.001 seconds, until the residual monitor defined in fvSolution shows that the velocity residual has dropped below 0.000001. This iteration ensures that the thermal convection velocity vector accurately captures the flow trend from the high-temperature region to the low-temperature region of the battery. For example, the mesh velocity vector in a certain high-temperature region is updated to 0.2 micrometers per second and diffuses into the mesh in a certain low-temperature region. Finally, all thermal flow velocity vectors are arranged according to the mesh arrangement to obtain the thermal flow velocity distribution matrix.

[0037] Specifically, in step S2, a pre-constructed battery geometric model is obtained from the finite element analysis software, and the thickness of each layer of the battery multilayer composite material is extracted. Create a corresponding free tetrahedral mesh for each layer of composite material from the battery geometry model to obtain a coarse mesh distribution containing multiple first meshes; For each first grid in the coarse grid distribution, the maximum grid cell size is determined based on the thickness of the thickest layer, such as 0.1 times the thickness of the thickest layer. The minimum grid cell size is determined based on the thickness of the thinnest layer, such as 0.01 times the thickness of the thinnest layer. The grid cell sizes of other layers are linearly interpolated according to the thickness ratio, where the thickness ratio is defined as the ratio of the current layer thickness to the thickness of the thickest layer. This yields the second grid size for each layer. The first grid in the coarse grid distribution is then divided according to the second grid size to obtain the preliminary grid division. The average value of the magnitude of the temperature gradient feature of all points in each second grid after the initial grid division is obtained from the initial temperature gradient field. For the second grids whose average magnitude is greater than or equal to the gradient threshold, the second grids are divided into third grids with a size 0.5 times the original size, to obtain the final grid division. The gradient threshold is 1.5 times the average value of the magnitude of the temperature gradient feature of all points in the entire temperature gradient field.

[0038] In one possible implementation, the battery geometry model obtained from finite element analysis software such as COMSOL Multiphysics typically includes a multi-layered structure comprising a positive electrode layer, a separator, a negative electrode layer, and a current collector. Specifically, the thickness of each layer, as a key geometric parameter, directly affects the fineness of the subsequent mesh generation. For example, the positive electrode layer thickness might be 80 micrometers, the separator 20 micrometers, the negative electrode layer 70 micrometers, and the aluminum current collector 15 micrometers. These thickness data form the basis for determining the mesh size range.

[0039] For example, after COMSOL Multiphysics creates a free tetrahedral coarse mesh for each material layer, the mesh cell size needs to be dynamically adjusted according to the layer thickness ratio. Understandably, the thickest layer (e.g., 80 micrometers) determines the global maximum mesh cell size, for example, set to 8 micrometers; the thinnest layer (e.g., 15 micrometers) determines the minimum mesh cell size, for example, set to 0.15 micrometers. For a membrane layer with a thickness of 20 micrometers, the thickness ratio is 0.25, and the mesh cell size calculated through linear interpolation is approximately 2.075 micrometers. This scaling method ensures that thinner regions have a sufficiently dense mesh to capture changes in the physical field, while thicker regions use a sparser mesh to balance computational resources.

[0040] It should be noted that adaptive mesh refinement based on the gradient characteristics of the initial temperature gradient field is an iterative process. For example, after the initial mesh generation, the average temperature gradient magnitude of all points within a mesh located at a tab junction is calculated. If this value is 1.2 degrees per millimeter, and the global gradient threshold is set to 0.75 degrees per millimeter, then this mesh is marked as needing refinement. Then, its mesh element size is halved and used as the size of the third mesh to further refine the second mesh, resulting in the final mesh generation. This process automatically clusters the mesh in regions of drastic temperature changes while maintaining a sparser mesh in regions with stable temperatures. Specifically, for further mesh generation, Gmsh, a three-dimensional finite element mesh generator, can be used to further refine the second mesh.

[0041] It is worth noting that in step S3, the temperature gradient characteristic modulus of each grid is obtained from the final temperature gradient field, and the variance of the temperature gradient characteristic modulus is calculated. If the variance exceeds a preset uniformity threshold, it is determined that the temperature is non-uniformly distributed. Then, the heat flow velocity vector magnitude for each grid in the final grid division is extracted from the heat flow velocity distribution matrix, and the temperature gradient feature magnitude for each grid in the final grid division is extracted from the final temperature gradient field. These features are then paired and combined into two-dimensional vector features, and finally arranged into a vector matrix according to the grid arrangement order. The vector matrix is ​​input into a pre-trained convolutional neural network, and the temperature value of each grid in the final grid division is calculated through forward propagation to determine the temperature distribution matrix in the next time step. The average temperature of each grid and all its neighboring grids is calculated from the predicted temperature distribution matrix as the local average temperature. If the local average temperature exceeds the preset overheating threshold, the grid is determined to be a local overheating region.

[0042] In one embodiment, calculating the variance of the temperature gradient feature modulus involves statistical principles. First, the average value of the temperature gradient feature modulus across all grids is calculated. Then, the average of the squared differences between each temperature gradient feature modulus and the average value is calculated. If the variance exceeds a preset uniformity threshold, such as 0.5, it is determined to be a non-uniform temperature distribution, resulting in a non-uniform temperature label. For example, for a discharging battery, if the variance is 1.2, exceeding the preset uniformity threshold of 0.5, a non-uniform temperature is identified, thereby triggering temperature prediction.

[0043] In one embodiment, for the identification of non-uniform temperature, the heat flow velocity vector modulus of each grid is extracted from the heat flow velocity distribution matrix, combined with the temperature gradient feature modulus to form a two-dimensional vector feature, and then arranged into a vector matrix according to the grid order, in order to capture the dynamics of heat transfer.

[0044] In one embodiment, the vector matrix is ​​input to a pre-trained convolutional neural network. The temperature value of each grid in the next time step is calculated through forward propagation to determine the temperature distribution matrix. The convolutional neural network is a deep learning model that includes convolutional layers, pooling layers, and fully connected layers. The weights are optimized through historical thermal data during pre-training. The feature maps are calculated layer by layer through forward propagation, and finally, the temperature prediction is output.

[0045] For example, in battery thermal management, the vector matrix size is 32×32, the network structure is 3 layers of convolution, each layer has a kernel of 5×5, the activation function is ReLU, the forward propagation process is as follows: the input vector matrix is ​​convolved to extract edge features, then pooled to reduce dimensionality, the fully connected layer outputs temperature values, such as the grid predicting 50.5°C from the current temperature gradient feature magnitude and heat flow velocity vector magnitude, the temperature values ​​of all grids form a temperature distribution matrix, thereby simulating thermal evolution and predicting temperature distribution.

[0046] In one embodiment, the average temperature of each grid and its neighboring grids is calculated from the predicted temperature distribution matrix as the local average temperature. If the average temperature exceeds a preset overheating threshold, such as 80°C, it is identified as a local overheating region. This is based on neighborhood average smoothing noise to achieve thermal anomaly detection. For example, the neighborhood includes 8 adjacent grids, and the average value is calculated as (this grid T + ∑neighborhood T) / 9. If the result is 85°C, which exceeds the preset overheating threshold, overheating is identified, thereby activating subsequent adjustments.

[0047] Preferably, in step S4, the number of local overheated areas is obtained, the average local temperature of each local overheated area is used as the temperature parameter of that area, and the average temperature value is obtained by calculating the average of the temperature parameters of all local overheated areas. Based on the number and average temperature of the aforementioned local overheated areas, a weighted fusion calculation is performed using preset weights to obtain the overheating risk coefficient; Based on the range of the overheating risk coefficient, obtain the corresponding control parameters of the heat dissipation system from the heat dissipation parameter table.

[0048] In one embodiment, when obtaining the number of local overheated areas, it is first necessary to traverse all grids of the final grid distribution and count the total number of those identified as local overheated areas, thereby providing basic data support for subsequent risk assessment.

[0049] Next, the average local temperature of each local overheated region is used as the temperature parameter of that region. That is, the average of all temperature values ​​within the local overheated region is calculated. It should be noted that the temperature parameter of each region represents the average level of heat accumulation in that region. For example, the temperature parameter of one overheated region may be 85°C and another may be 82°C. These parameters are obtained by summing the temperatures of all grids in the region and dividing by the number of grids to ensure that the parameters reflect the true thermal state.

[0050] Based on these temperature parameters, the average temperature value is obtained by calculating the average of the temperature parameters of all local overheated areas. Specifically, the parameters of all areas are summed and then divided by the number of local overheated areas. For example, if the parameters of the number of 5 local overheated areas are 85°C, 82°C, 88°C, 80°C and 84°C, then the average temperature value is (85+82+88+80+84) / 5=83.8°C. This averaging operation helps to comprehensively assess the overall overheating trend. In the battery management system, this value is used to quantify system-level thermal risk, rather than isolated area analysis, and thus connects to the fusion calculation of risk coefficients.

[0051] The process of calculating the overheating risk coefficient by weighting and fusing the number and average temperature of the local overheated areas using preset weights requires a detailed explanation of the source and application of the preset weights. The weights are typically set based on experience or simulation data. For example, if the quantity weight is 0.4 and the average temperature weight is 0.6, then the risk coefficient = 0.4 × (quantity / maximum possible quantity) + 0.6 × (average temperature / maximum threshold). In one possible implementation, the maximum possible quantity is 10, and the maximum threshold is 90°C. If the quantity is 5 and the average temperature is 83.8°C, then the risk coefficient = 0.4 × (5 / 10) + 0.6 × (83.8 / 90) ≈ 0.4 × 0.5 + 0.6 × 0.931 ≈ 0.2 + 0.559 = 0.759. This weighted fusion emphasizes that the severity of temperature is greater than the influence of quantity, achieving dynamic risk quantification through this coefficient and supporting real-time adjustments.

[0052] In the process of obtaining the control parameters corresponding to the heat dissipation system from the heat dissipation parameter table based on the range of the overheating risk coefficient, the heat dissipation parameter table is a predefined mapping table. For example, the range [0, 0.3] corresponds to low-risk parameters such as fan speed of 30%, [0.3, 0.7] is medium-risk speed of 60%, and [0.7, 1.0] is high-risk speed of 100%. If the risk coefficient 0.759 falls in the range [0.7, 1.0], then the obtained control parameter is full-speed fan operation. This acquisition ensures that the parameters directly respond to the risk level, thereby optimizing the system response.

[0053] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and these variations still fall within the protection scope of the present invention.

Claims

1. A method for managing overheating of a battery, characterized in that, Includes the following steps: S1: Determine the initial temperature gradient field inside the battery during operation based on real-time multi-point temperature data; S2: Based on the initial temperature gradient field, a thermal simulation model is used to calculate the heat conduction and convection process in the battery material, and the final temperature gradient field and heat flow velocity distribution matrix are determined. S3: If the temperature is determined to be non-uniform from the final temperature gradient field, the final temperature gradient field and the heat flow velocity distribution matrix are used as inputs, and a pre-trained convolutional neural network is used to make predictions. The predicted temperature distribution matrix is ​​output, and the local overheated area is determined based on the predicted temperature distribution matrix. S4: Weighted fusion of the number of local overheated areas and the average temperature of all local overheated areas is performed to determine the overheating risk coefficient. Based on the interval in which the overheating risk coefficient is located, the corresponding control parameters of the heat dissipation system are obtained.

2. The battery overheat management method according to claim 1, characterized in that: In step S1, multi-point temperature monitoring signals are acquired from a temperature sensor array deployed inside the battery. The multi-point temperature monitoring signals include a temperature value T and a corresponding spatial coordinate X. For each point p, obtain all points k adjacent to point p from the multi-point temperature monitoring signal, and calculate the temperature difference. ,in Let k be the temperature value at point k. Let p be the temperature value, and the distance be... coordinates With coordinates The Euclidean distance, where Let k be the spatial coordinates. Let p be the spatial coordinates of point p. Calculate the gradient magnitude between point p and point k. Divide by The average of the gradient magnitudes of all points p and k is calculated as the magnitude of the temperature gradient feature. Points k with the largest gradient magnitude are extracted and paired with point p. Based on the coordinates... With coordinates The direction of the point pair is calculated as the direction of the temperature gradient feature, and the temperature gradient feature of point p is obtained. Based on the temperature gradient characteristics of all points, the points are arranged according to their coordinates to construct the initial temperature gradient field of the battery thermal distribution.

3. The battery overheat management method according to claim 2, characterized in that: In step S1, a spatial coordinate grid is determined from the multi-point temperature monitoring signal, and the temperature value of each grid point is calculated using Kriging interpolation to generate the processed multi-point temperature monitoring signal. For each point p in the processed multi-point temperature monitoring signal, determine its corresponding temperature gradient characteristics.

4. The battery overheat management method according to claim 1, characterized in that: In step S2, the inside of the battery is divided into grids, the boundaries of each grid are obtained, and the temperature gradient features of all points of the corresponding grid after grid division are extracted from the initial temperature gradient field based on the grid boundaries. The average value of the temperature gradient characteristic magnitude of all points in each grid is calculated as the temperature gradient vector magnitude of that grid; the temperature gradient direction at the center point of the grid is taken as the temperature gradient vector direction of that grid. The final temperature gradient field is obtained by arranging the temperature gradient vectors of all grids according to the grid arrangement order.

5. The battery overheat management method according to claim 4, characterized in that: In step S2, the final temperature gradient field is imported into OpenFOAM as the initial field, the heat flow velocity vector corresponding to each grid is initialized to a zero vector, and the heat flow velocity vector is iteratively updated through the PISO algorithm until the residual is less than the preset residual threshold. All heat flow velocity vectors are arranged according to the grid arrangement order to obtain the heat flow velocity distribution matrix.

6. The battery overheat management method according to claim 4, characterized in that: In step S2, a pre-constructed battery geometric model is obtained from the finite element analysis software, and the thickness of each layer of the battery multilayer composite material is extracted. Create a corresponding free tetrahedral mesh for each layer of composite material from the battery geometry model to obtain a coarse mesh distribution containing multiple first meshes; For each first grid in the coarse grid distribution, the maximum grid cell size is determined based on the thickness of the thickest layer, and the minimum grid cell size is determined based on the thickness of the thinnest layer. The grid cell sizes of other layers are linearly interpolated according to the thickness ratio, where the thickness ratio is defined as the ratio of the current layer thickness to the thickness of the thickest layer, to obtain the second grid size corresponding to each layer. The first grid in the coarse grid distribution is then divided according to the second grid size to obtain the preliminary grid division. The average value of the magnitude of the temperature gradient feature of all points in each second grid after the initial grid division is obtained from the initial temperature gradient field. For the second grid whose average magnitude is greater than or equal to the gradient threshold, the second grid is divided into third grids with a size 0.5 times the original size, thus obtaining the final grid division.

7. The battery overheat management method according to claim 1, characterized in that: In step S3, the temperature gradient characteristic modulus of each grid is obtained from the final temperature gradient field, and the variance of the temperature gradient characteristic modulus is calculated. If the variance exceeds a preset uniformity threshold, it is determined that the temperature is non-uniformly distributed. Then, the heat flow velocity vector magnitude for each grid in the final grid division is extracted from the heat flow velocity distribution matrix, and the temperature gradient feature magnitude for each grid in the final grid division is extracted from the final temperature gradient field. These features are then paired and combined into two-dimensional vector features, and finally arranged into a vector matrix according to the grid arrangement order. The vector matrix is ​​input into a pre-trained convolutional neural network, and the temperature value of each grid in the final grid division is calculated through forward propagation to determine the temperature distribution matrix in the next time step. The average temperature of each grid and all its neighboring grids is calculated from the predicted temperature distribution matrix as the local average temperature. If the local average temperature exceeds the preset overheating threshold, the grid is determined to be a local overheating region.

8. The battery overheat management method according to claim 1, characterized in that: In step S4, the number of local overheated areas is obtained, and the average local temperature of each local overheated area is used as the temperature parameter of that area. The average temperature value is obtained by calculating the average of the temperature parameters of all local overheated areas. Based on the number and average temperature of the aforementioned local overheated areas, a weighted fusion calculation is performed using preset weights to obtain the overheating risk coefficient; Based on the range of the overheating risk coefficient, obtain the corresponding control parameters of the heat dissipation system from the heat dissipation parameter table.

Images