A lithium battery pack temperature field prediction method

Abstract

The application discloses a lithium battery pack temperature field prediction method, and relates to the technical field of batteries, and comprises the following steps: S100, data acquisition and preprocessing; S200, data vector cross coding; S300, hidden vector propagation; S400, total error calculation; S500, temperature prediction neural network training; and S600, temperature field prediction.The lithium battery pack temperature field prediction method based on the cross attention mechanism coupled operator neural network introduces time-varying system iteration of latent space, reduces the operation time, and reduces the error; the application introduces physical information into the neural network, optimizes the parameters of the deviation of the actual physical scene partial differential equation and the deviation of the data together, and improves the explainability of the neural network.

Description

Technical Field

[0001] This invention relates to the field of battery technology, and in particular to a method for predicting the temperature field of lithium battery packs. Background Technology

[0002] Lithium-ion batteries, as a crucial modern energy source, are widely used in electric vehicles and energy storage systems due to their high energy density, long cycle life, low self-discharge, and low maintenance costs. For current lithium-ion battery technology, due to their inherent chemical and thermal properties, research indicates that their optimal operating temperature should be maintained between 15°C and 45°C. Otherwise, extreme conditions may lead to capacity degradation, electrode degradation, and even potential safety issues. The State of Temperature (SOT) is a vital indicator for assessing battery temperature changes and their impact on battery performance. As battery usage time increases, the SOT changes, which can affect charge / discharge efficiency and lifespan. Therefore, accurate and reliable estimation of the SOT is crucial in battery management technology.

[0003] However, because lithium batteries in electric vehicles or energy storage systems are typically densely connected in series or parallel with hundreds or thousands of individual cells, the overall battery temperature is usually difficult to monitor completely. Temperature changes in the battery under different operating conditions and environments can affect the effectiveness of temperature control. Therefore, accurate temperature prediction of the entire lithium battery pack is a prerequisite for battery temperature control.

[0004] Numerous researchers both domestically and internationally have conducted extensive research on methods for predicting the temperature of lithium-ion batteries. Examples include: 1) Establishing instantaneous temperature models for individual lithium-ion battery cells based on the Newman and Bernardi heat generation rate models. This involves calculating and collecting battery data, extracting key parameters, and then simulating the thermal behavior of individual cells to obtain predicted battery temperatures under different operating conditions. However, this method requires lengthy simulation calculations, is difficult to extract key battery parameters, and is not applicable to real-time situations. 2) Developing equivalent reduced-order thermal models for lithium-ion batteries based on data-driven methods (such as overlimit learning machines). This involves decoupling and reducing the order of the thermal process of lithium-ion batteries using the Karhunene-Loeve method, then integrating machine learning methods to analyze the data and ultimately predict the battery temperature. However, the order reduction process may lead to information loss, and the resulting model has limited generalizability. Furthermore, these methods primarily predict the thermal behavior of individual lithium-ion battery cells and may not be applicable to the temperature field formed by actual lithium-ion battery packs.

[0005] Therefore, those skilled in the art are dedicated to developing a method for predicting the temperature field of lithium battery packs. Summary of the Invention

[0006] In view of the above-mentioned deficiencies of the prior art, the technical problem to be solved by the present invention is how to use the operator neural network learning method to accurately predict the temperature field of lithium battery packs.

[0007] Predicting the temperature field of a lithium-ion battery pack is essentially solving a spatiotemporal partial differential equation concerning temperature conduction. This is equivalent to an operator that takes the physical information *u* of the initial and boundary conditions as input, and discretizes the target point's data coordinate parameter vector *d* to find the predicted temperature *T(u)(d)* at the corresponding point. The operator is implemented using a cross-attention mechanism for cross-encoding of information. The input battery pack parameters (voltage, current, initial temperature) sequentially pass through an encoder, processor, and decoder to obtain the battery pack temperature at a predetermined time step. The encoder extracts high-dimensional features of the battery pack parameters and performs cross-encoding; the processor iterates through these features at each time step; and the decoder restores the features to the output space to obtain the battery pack temperature.

[0008] In one embodiment of the present invention, a method for predicting the temperature field of a lithium battery pack is provided, comprising the following steps:

[0009] S100, Data Acquisition and Preprocessing: Acquire raw data of lithium battery pack operation, spatially discretize the temperature field of lithium battery pack, randomly collect temperature data u at sampling points at different heights, and preprocess the raw data and temperature data u.

[0010] S200, data vector cross-coding, creating a feature extraction encoder, cross-coding the extracted data vectors, approximating the operator, performing high-dimensional mapping and filtering, and obtaining the latent vector after residual connection;

[0011] S300, Latent Vector Propagation, creates a time-varying system processor, which iteratively propagates latent vectors in the latent space, with a predetermined time step being a fixed time interval, until the target number of steps is reached;

[0012] S400, total error calculation, create decoder, map the latent vectors in the latent space back to the output space to obtain the preliminary temperature prediction value, calculate the physical equation error through partial differential equation, and add it to the prediction error to obtain the total error Loss;

[0013] S500: Train the temperature prediction neural network. Repeat steps S200 to S500. Calculate the total error Loss for each training round. Backpropagate the total error Loss. Optimize the network parameters to reduce the total error Loss based on the error gradient obtained from backpropagation. Continue until the total error Loss is less than the set target error value. Stop training to obtain the trained temperature prediction neural network.

[0014] S600 Temperature Field Prediction: Uses a trained temperature prediction neural network to predict the temperature at any point during the operation of the lithium battery pack.

[0015] Optionally, in the lithium battery pack temperature field prediction method in the above embodiments, step S100 includes:

[0016] S110. Collect raw data of lithium battery pack operation, take different locations in the lithium battery pack where the temperature needs to be predicted as collection points, and collect the operating data of the lithium battery pack at a fixed frequency under predetermined charging and discharging conditions.

[0017] S120. Collect temperature data u from sampling points, and randomly collect temperature data u from sampling points at different heights;

[0018] S130. Data preprocessing: The raw data is preprocessed into a lithium battery pack coordinate parameter vector d, as shown in the following formula:

[0019] d = [x,y,z,I] t U t ,T am [SOC] T

[0020] Where x, y, z are the coordinates in the temperature field space, I t U is the battery current at time t. t Let T be the battery voltage at time t. am The ambient temperature around the battery is denoted as , SOC is the state of charge of the battery, and t represents time.

[0021] S140. Calculate the sampling vector X. Preprocess the temperature data u into multiple two-dimensional matrices according to the height of the battery pack. Each point in the matrix represents a discrete sampling temperature point of a specific height layer. Perform multilayer perceptron encoding on u to obtain the sampling vector X containing physical information.

[0022] X = L*(x i )=L*f ω (u i ) 1≤i≤N

[0023] Where f ω For a multilayer perceptron (MLP), L is the number of layers in the height direction, and x i For sampling vectors on different layers, u i N represents the temperature data for different layers, where N is the number of temperature collection points for each layer.

[0024] Furthermore, in the lithium battery pack temperature field prediction method of the above embodiments, the lithium battery pack coordinate parameter vector d is encoded using a multilayer perceptron and mapped to a high-dimensional query vector Y:

[0025] Y = (y i )=f ω (d i ) 1≤i≤L

[0026] Among them, f ω For a Multiple Layer Perceptron (MLP), y i Let d be the row vector of Y. i These are the preprocessed coordinate parameters.

[0027] Optionally, in the lithium battery pack temperature field prediction method in any of the above embodiments, the predetermined charge and discharge conditions are 2C charging and 0.5C discharging.

[0028] Optionally, in the lithium battery pack temperature field prediction method of any of the above embodiments, the fixed frequency is once per second.

[0029] Optionally, in the lithium battery pack temperature field prediction method in the above embodiments, the feature extraction encoder is based on a cross-attention mechanism.

[0030] Furthermore, in the lithium battery pack temperature field prediction method in the above embodiments, the feature extraction encoder cross-fuses the high-dimensional query vector encoded by the lithium battery pack coordinate parameters and the sampling vector containing physical information to obtain a latent vector that can propagate in the latent space.

[0031] Furthermore, in the lithium battery pack temperature field prediction method in the above embodiments, step S200 includes:

[0032] S210. Create a feature extraction encoder, which includes a cross-attention mechanism layer, a residual connection layer, and a point-to-point fully connected layer.

[0033] S220, cross-coding, performs high-dimensional mapping and filtering. In the cross-attention mechanism layer, parameter matrices Wq, Wk, and Wv are created. Wq is multiplied by Y to obtain the query vector Q. Wk and Wv are multiplied by X to obtain the key vector K and value vector V, respectively. The query vector Q and key vector K are then normalized.

[0034]

[0035]

[0036] Where q i and k i To query the row vectors of vector Q and key vector K, n e The number of columns for query vector Q and key vector K;

[0037] S230. Calculate the latent vector, and then obtain the latent vector z corresponding to the query vector yi through cross multiplication. i :

[0038]

[0039] Where, q t Given a query vector as input, v i Let q be the row vector of the value vector V. t Derived from a high-dimensional query vector Y,k containing coordinate parameter information i and v i The implicit vector Z, which contains physical information, is derived from the sampled vector X and approximates the operator through cross-multiplication. i .

[0040] S240. Calculate the actual latent vector. Since the input X has L layers, and the feature output corresponds to only one layer, sum all L layers and add the query vector q. t The residual input was used to obtain q. t The corresponding actual latent vector z t :

[0041]

[0042] S250. Calculate the latent vector output. After passing through the cross-attention mechanism layer, in order to make the output features consistent with the input space, the input q is then... t Adding it to a fully connected layer forms a residual connected layer, resulting in the latent vector output.

[0043]

[0044] Optionally, in the lithium battery pack temperature field prediction method in any of the above embodiments, the time-varying system processor is based on a point-type multilayer sensor.

[0045] Optionally, in the lithium battery pack temperature field prediction method of any of the above embodiments, step S300 includes:

[0046] S310, Create a time-varying system processor;

[0047] S320, Latent vector output propagation, for latent vectors Propagation occurs in latent space;

[0048] S330, Time Iteration: Iterate the time to the target number of steps before outputting the result.

[0049]

[0050] After iterative propagation to reach the target number of steps, the hidden vector used for output is obtained.

[0051] Preferably, in the lithium battery pack temperature field prediction method in the above embodiments, the target number of steps is 5.

[0052] Preferably, in the lithium battery pack temperature field prediction method in the above embodiments, the fixed time interval is 1 second.

[0053] Optionally, in the lithium battery pack temperature field prediction method in any of the above embodiments, the operating data of the lithium battery pack includes voltage, current, and state of charge.

[0054] Optionally, in the lithium battery pack temperature field prediction method of any of the above embodiments, step S400 includes:

[0055] S410. Create a decoder, which is based on a self-attention mechanism;

[0056] S420, Calculate the preliminary temperature forecast value, Using parameter matrices respectively Multiply to obtain the query matrix. Key matrix Sum matrix The parameter matrix used here is not shared with the parameters used in S220. The output vector H is then calculated using the Softmax function, and finally normalized through a residual connection to obtain the predicted temperature output. The formula is as follows:

[0057]

[0058]

[0059] in, For the predicted temperature at the query point, the FFN is a fully connected layer and does not share parameters with the FFN that has passed through the cross attention mechanism layer;

[0060] S430. Calculate the error in the physical equations, specifically the error L in the physical equations for network parameter optimization. phy The error in a partial differential equation is expressed in the following form:

[0061]

[0062]

[0063] Wherein, ρ, C p k and Q represent the battery's density, specific heat capacity, and average thermal conductivity in each direction, respectively. batt V represents the equivalent heat generated during battery operation. battE is the volume of the temperature field region. oc This is the open-circuit voltage of the battery;

[0064] S440. Calculate the total error; the difference between the predicted temperature value and the actual temperature value yields the data error L. data , with L phy The sums are the total error Loss.

[0065] Optionally, in the lithium battery pack temperature field prediction method of any of the above embodiments, step S500 includes:

[0066] S510. Calculate the total backpropagation error. For the obtained total error Loss, the loss value is distributed to each network layer of the neural network encoder, processor, and decoder through backpropagation to calculate the error gradient.

[0067] S520. Calculate the error gradient. Using the chain rule, take the partial derivative of the loss value of each network layer with respect to the parameters in the network to obtain the error gradient of each network layer.

[0068] S530. Optimize the total error. For the error gradient obtained from backpropagation, use stochastic gradient descent to optimize the parameters of the neural network. Then repeat steps S200 to S500 until the total error Loss is lower than the set value 1*e. -5 Thus, a well-trained temperature prediction neural network is obtained.

[0069] This invention presents a lithium battery pack temperature field prediction method based on a cross-attention mechanism coupled with an operator neural network. By training the neural network and utilizing existing temperature information and predicted temperature, the method achieves a significant speed improvement and enhanced real-time performance compared to finite element simulation. Furthermore, this invention introduces time-varying system iteration in the latent space, performing multiple iterations over multiple time steps based on only one feature extraction, thus reducing computation time and errors. Finally, this invention incorporates physical information into the neural network, optimizing parameters by combining deviations in the partial differential equations of the actual physical scenario with data deviations, thereby improving the interpretability of the neural network.

[0070] The following will further explain the concept, specific structure, and technical effects of the present invention in conjunction with the accompanying drawings, so as to fully understand the purpose, features, and effects of the present invention. Attached Figure Description

[0071] Figure 1 This is a flowchart illustrating a method for predicting the temperature field of a lithium battery pack according to an exemplary embodiment. Detailed Implementation

[0072] The following description, with reference to the accompanying drawings, illustrates several preferred embodiments of the present invention to make its technical content clearer and easier to understand. The present invention can be embodied in many different forms, and the scope of protection of the present invention is not limited to the embodiments mentioned herein.

[0073] In the accompanying drawings, components with the same structure are indicated by the same numerical designation, and components with similar structures or functions are indicated by similar numerical designations. The dimensions and thicknesses of each component shown in the drawings are arbitrary, and the present invention does not limit the dimensions and thicknesses of each component. To make the illustrations clearer, the thickness of components is schematically exaggerated in some places in the drawings.

[0074] The inventors have provided a method for predicting the temperature field of lithium battery packs, such as... Figure 1 As shown, it includes the following steps:

[0075] S100, Data Acquisition and Preprocessing: Acquire raw data from the lithium battery pack operation; spatially discretize the temperature field of the lithium battery pack; randomly collect temperature data u at sampling points at different heights; and preprocess the raw data and temperature data u; including:

[0076] S110. Collect raw data on the operation of the lithium battery pack. Using different locations within the lithium battery pack where the temperature needs to be predicted as collection points, collect the operating data of the lithium battery pack at a fixed frequency under predetermined charge and discharge conditions, including voltage, current, state of charge, and predetermined charge and discharge parameters.

[0077] The operating conditions are 2C charging and 0.5C discharging, with a fixed frequency of once per second.

[0078] S120. Collect temperature data u from sampling points, and randomly collect temperature data u from sampling points at different heights;

[0079] S130. Data preprocessing: The raw data is preprocessed into a lithium battery pack coordinate parameter vector d, as shown in the following formula:

[0080] d = [x,y,z,I] t U t ,T am [SOC] T

[0081] Where x, y, z are the coordinates in the temperature field space, I t U is the battery current at time t. t Let T be the battery voltage at time t. am The ambient temperature surrounding the battery is denoted as , SOC as the battery's state of charge is denoted as , and t represents time. The coordinate parameter vector d of the lithium battery pack is encoded using a multilayer perceptron and mapped to a high-dimensional query vector Y.

[0082] Y = (yi )=f ω (d i ) 1≤i≤L

[0083] Among them, f ω For a Multiple Layer Perceptron (MLP), y i Let d be the row vector of Y. i These are the preprocessed coordinate parameters.

[0084] S140. Calculate the sampling vector X. Preprocess the temperature data u into multiple two-dimensional matrices according to the battery pack's height, where each point in the matrix represents a discrete sampling temperature point at a specific height level. Perform multilayer perceptron encoding on u to obtain the sampling vector containing physical information.

[0085] Quantity X:

[0086] X = L*(x i )=L*f ω (u i ) 1≤i≤N

[0087] Where f ω For a multilayer perceptron (MLP), L is the number of layers in the height direction, and x i For sampling vectors on different layers, u i N represents the temperature data for different layers, where N is the number of temperature collection points for each layer.

[0088] S200, Data Vector Cross-Encoding: A feature extraction encoder is created based on a cross-attention mechanism. The extracted data vectors are cross-encoded, approximating the operator and performing high-dimensional mapping and filtering. After residual connections, a latent vector is obtained. Specifically, the feature extraction encoder cross-fuses the high-dimensional query vector encoded with the lithium battery pack coordinate parameters and the sampled vector containing physical information to obtain a latent vector that can propagate in the latent space. This includes:

[0089] S210. Create a feature extraction encoder, which includes a cross-attention mechanism.

[0090] A force mechanism layer, a residual connection layer, and a point-to-point fully connected layer;

[0091] S220, cross-coding, performs high-dimensional mapping and filtering. In the cross-attention mechanism layer, parameter matrices Wq, Wk, and Wv are created. Wq is multiplied by Y to obtain the query vector Q. Wk and Wv are multiplied by X to obtain the key vector K and value vector V, respectively. The query vector Q and the key...

[0092] Normalize vector K:

[0093]

[0094]

[0095] Where q i and k i To query the row vectors of vector Q and key vector K, n e The number of columns for query vector Q and key vector K;

[0096] S230. Calculate the latent vector, and then obtain the latent vector corresponding to the query vector yi through cross-multiplication.

[0097] Quantity z i :

[0098]

[0099] Where, q t Given a query vector as input, v i Let q be the row vector of the value vector V. t Derived from a high-dimensional query vector Y,k containing coordinate parameter information i and v i Originating from a sampling vector X containing physical information, the operator is approximated through cross-multiplication to obtain the implicit vector z containing the temperature field information of the lithium battery pack. i .

[0100] S240. Calculate the actual latent vector. Since the input X has L layers, and the feature output corresponds to only one layer, sum all L layers and add the query vector q. t The residual input is obtained

[0101] When q t The corresponding actual latent vector z t :

[0102]

[0103] S250. Calculate the latent vector output. After passing through the cross-attention mechanism layer, in order to make the output features consistent with the input space, the input q is then... t Adding it to a fully connected layer (FFN) forms a residual connected layer, which yields the hidden vector output.

[0104]

[0105] S300, Latent Vector Propagation, creates a time-varying system processor to iteratively propagate latent vectors in the latent space, with a predetermined time step of a fixed interval of 1 second, until the target number of steps is 5; specifically including:

[0106] S310, Create a time-varying system processor;

[0107] S320, Latent vector output propagation, for latent vectors Propagation occurs in latent space;

[0108] S330, Time Iteration: Iterate the time to the target number of steps before outputting the result.

[0109]

[0110] After iterative propagation to reach the target number of steps, the hidden vector used for output is obtained.

[0111] S400, Total Error Calculation: A decoder is created to map latent vectors from the latent space back to the output space, obtaining preliminary temperature predictions. The error in the physical equations is calculated using partial differential equations, and this error is added to the prediction error to obtain the total error, Loss. Specifically, this includes:

[0112] S410. Create a decoder, which is based on a self-attention mechanism;

[0113] S420, Calculate the preliminary temperature forecast value, Using parameter matrices respectively Multiply to obtain the query matrix. Key matrix Sum matrix The parameter matrix used here is not shared with the parameters used in S220. The output vector H is then calculated using the Softmax function, and finally normalized through a residual connection to obtain the predicted temperature output. The formula is as follows:

[0114]

[0115]

[0116] in, For the predicted temperature at the query point, FFN is a fully connected layer, connected to the...

[0117] FFNs passing through the cross-attention mechanism layer do not share parameters;

[0118] S430. Calculate the error in the physical equations, specifically the error L in the physical equations for network parameter optimization. phy for

[0119] The error in a partial differential equation is expressed in the following form:

[0120]

[0121]

[0122] Wherein, ρ, C p k and Q represent the battery's density, specific heat capacity, and average thermal conductivity in each direction, respectively. batt V represents the equivalent heat generated during battery operation. batt Temperature field region

[0123] The volume of the domain, E oc This is the open-circuit voltage of the battery;

[0124] S440. Calculate the total error by subtracting the predicted temperature value from the actual temperature value to obtain the data error.

[0125] L data , with L phy The sums are the total error Loss.

[0126] S500: Training the temperature prediction neural network. Repeat steps S200 to S500, calculating the total error Loss for each training round, backpropagating this total error Loss, and optimizing the network parameters to reduce the total error Loss based on the error gradient obtained from backpropagation, until the total error Loss is less than the set target error value, then stopping training to obtain the trained temperature prediction neural network; specifically including:

[0127] S510. Calculate the total backpropagation error. For the obtained total error Loss, the loss value is distributed to each network layer of the neural network encoder, processor, and decoder through backpropagation to calculate the error gradient.

[0128] S520. Calculate the error gradient. Using the chain rule, take the partial derivative of the loss value of each network layer with respect to the parameters in the network to obtain the error gradient of each network layer.

[0129] S530. Optimize the total error. For the error gradient obtained from backpropagation, use stochastic gradient descent to optimize the parameters of the neural network. Then repeat steps S200 to S500 until the total error Loss is lower than the set value 1*e. -5 Thus, a well-trained temperature prediction neural network is obtained.

[0130] S600 Temperature Field Prediction: Uses a trained temperature prediction neural network to predict the temperature at any point during the operation of the lithium battery pack.

[0131] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for predicting the temperature field of a lithium battery pack, characterized in that, Includes the following steps: S100. Data Acquisition and Preprocessing: Acquire raw data from the lithium battery pack operation, spatially discretize the temperature field of the lithium battery pack, and randomly collect temperature data at sampling points at different heights. u Regarding the raw data and the temperature data u Preprocessing is required; S200, data vector cross-coding, creating a feature extraction encoder, cross-coding the extracted data vectors, approximating the operator, performing high-dimensional mapping and filtering, and obtaining the latent vector after residual connection; S300, Latent vector propagation: Create a time-varying system processor to iteratively propagate the latent vector in the latent space, with a predetermined time step being a fixed time interval, until the target number of steps is reached; S400. Total error calculation: Create a decoder to map the latent vectors in the latent space back to the output space to obtain a preliminary temperature prediction value. Calculate the physical equation error using partial differential equations and add it to the prediction error to obtain the total error. Loss ; S500, Temperature prediction neural network training: Repeat steps S200 to S500, and calculate the total error for each training round. Loss The total error is backpropagated. Loss Based on the error gradient obtained from backpropagation, the network parameters are optimized to reduce the total error. Loss until the total error Loss If the error value is less than the set target value, stop training and obtain the trained temperature prediction neural network. S600 Temperature field prediction: Using the trained temperature prediction neural network, predict the temperature at any point during the operation of the lithium battery pack; Step S200 includes: S210. Create a feature extraction encoder, which includes a cross-attention mechanism layer, a residual connection layer and a point-to-point fully connected layer. S220, cross-encoding, performs high-dimensional mapping and filtering, and creates a parameter matrix in the cross-attention mechanism layer. Wq , Wk , Wv ,Will Wq and Y Multiply to obtain the query vector Q , Wk , Wv respectively with X Multiplication yields the key vector K Sum value vector V For the query vector Q and the key vector K Normalization is performed: ； ； in and query vector Q and key vector K The row vector, query vector Q and key vector K The number of columns; S230. Calculate the latent vector, and then obtain the query vector through cross-multiplication. yi The corresponding hidden vector : ； in, Given a query vector as input, Value vector V The row vector, Derived from a high-dimensional query vector containing coordinate parameter information Y , and Derived from a sampled vector containing physical information X The operator was approximated by cross-multiplication, resulting in the implicit vector containing the temperature field information of the lithium battery pack. ; S240. Calculate the actual implicit vector, since... X Input has L Each layer has a feature output corresponding to only one layer. L Add up all the layers plus the query vector The residual input is obtained. The corresponding actual latent vector : ； S250. Calculate the latent vector output. After passing through the cross-attention mechanism layer, in order to make the output features consistent with the input space, the input... The residual connection layer is formed by adding a fully connected layer to it, and the hidden vector output is obtained. : 。 2. The lithium battery pack temperature field prediction method as described in claim 1, characterized in that, Step S100 includes: S110. Collect raw data of the lithium battery pack operation, take different locations in the lithium battery pack where the temperature needs to be predicted as collection points, and collect the operating data of the lithium battery pack at a fixed frequency under predetermined charging and discharging conditions. S120, Collect temperature data from sampling points u Temperature data were randomly collected at sampling points at different altitudes. u ; S130. Data preprocessing: The original data is preprocessed into a lithium battery pack coordinate parameter vector. d The formula is as follows: ； in These are the coordinates in the temperature field space. Let be the battery current at time t. Let be the battery voltage at time t. The ambient temperature around the battery. SOC This refers to the battery's state of charge. t Indicates time; S140, Calculate the sampling vector X， The temperature data u The battery pack is preprocessed into multiple two-dimensional matrices according to its height, with each point in the matrix representing a discrete sampling temperature point of a specific height layer. u Perform multilayer perceptron encoding to obtain a sampling vector containing physical information. X : in For multilayer perceptrons, L The number of layers in the height direction. For sampling vectors on different layers, Temperature data for different layers, N The number of temperature sampling points for different layers.

3. The lithium battery pack temperature field prediction method as described in claim 2, characterized in that, For the coordinate parameter vector of the lithium battery pack d Encoding is performed using a multilayer perceptron, which is then mapped to a high-dimensional query vector. Y : in, For multilayer perceptrons, for Y The row vector, These are the preprocessed coordinate parameters.

4. The lithium battery pack temperature field prediction method as described in claim 1, characterized in that, The predetermined charge / discharge conditions are 2C charging and 0.5C discharging.

5. The lithium battery pack temperature field prediction method as described in claim 2, characterized in that, The feature extraction encoder is based on a cross-attention mechanism.

6. The lithium battery pack temperature field prediction method as described in claim 1, characterized in that, The time-varying system processor is based on a point-type multilayer perceptron.

7. The lithium battery pack temperature field prediction method as described in claim 5, characterized in that, Step S300 includes: S310, Create a time-varying system processor; S320, Latent vector output propagation, for the latent vector Propagation occurs in latent space; S330, Time Iteration: Iterate the time to the target number of steps before outputting the result. The implicit vector for output is obtained by iterative propagation to reach the target number of steps.

8. The lithium battery pack temperature field prediction method as described in claim 7, characterized in that, Step S400 includes: S410. Create a decoder, which is based on a self-attention mechanism; S420, Calculate the preliminary temperature forecast value, Using parameter matrices respectively q , k , v Multiply to obtain the query matrix. Key matrix Sum matrix Then through Softmax The function calculates the output vector. H The predicted temperature output is then obtained by passing the residuals through a single layer of normalization, as shown in the following formula: ； in, For the predicted temperature at the query point, FFN It is a fully connected layer, connected to a layer that passes through a cross-attention mechanism layer. FFN Parameters are not shared; S430. Calculate the error in the physical equations for network parameter optimization. The error in a partial differential equation is expressed in the following form: ； in, ρ , , k These are the battery's density, specific heat capacity, and average thermal conductivity in each direction. This refers to the equivalent heat generated during battery operation. Let V be the volume of the temperature field region. This is the open-circuit voltage of the battery; S440. Calculate the total error by subtracting the predicted temperature value from the actual temperature value to obtain the data error. ,and Adding them together gives the total error. Loss .

9. The method for predicting the temperature field of a lithium battery pack as described in claim 1, characterized in that, Step S500 includes: S510. Calculate the total backpropagation error. For the obtained total error... Loss Through backpropagation, the loss value is distributed to each network layer of the neural network encoder, processor, and decoder; S520. Calculate the error gradient by using the chain rule to obtain the error gradient of each network layer by taking the partial derivative of the loss value of each network layer with respect to the parameters in the network. S530. Optimize the total error. For the error gradient obtained through backpropagation, use stochastic gradient descent to optimize the parameters of the neural network. Then repeat steps S200 to S500 until the total error is reached. Loss Below the set value 1*e -5 Thus, a well-trained temperature prediction neural network is obtained.

Images