Method and device for quickly estimating internal temperature of battery

Abstract

The invention relates to a method and device for quickly estimating the internal temperature of a battery, and belongs to the field of lithium battery temperature detection.The method comprises the steps that a battery electrochemical-thermal coupling model with impedance output is established to simulate impedance output, and EIS data of the battery in different states are obtained to verify the model; dRT conversion is carried out after the EIS data are verified through K-K, and high-frequency peak area characteristics highly related to the internal temperature of the battery are obtained through correlation analysis; according to the method, low-frequency data is removed, a set frequency sampling range is selected, so that the data volume and the calculation complexity are reduced, a multi-head attention mechanism prediction model based on Transform is established, a complex nonlinear relation is captured, and rapid lossless estimation of the internal temperature of the battery is achieved. According to the invention, an efficient, accurate and non-intrusive temperature monitoring means can be provided, can be conveniently integrated into the existing BMS as a part of temperature monitoring and safety management, and does not need additional complex hardware.

Description

Technical Field

[0001] The present invention relates to the field of lithium battery temperature detection, and in particular to a method and device for quickly estimating the internal temperature of a battery. Background Art

[0002] Lithium-ion batteries have been widely used in new energy industries such as electric vehicles and energy storage power stations to ensure safe and efficient operation of batteries throughout their entire generation cycle. The internal temperature of the battery is closely related to safety and efficiency. However, it is impractical to directly measure the internal temperature of the battery. Commonly used temperature monitoring only obtains the surface temperature through sensors. Electrochemical impedance spectroscopy (EIS) is a non-destructive testing method that can collect microscopic information of the battery and monitor the internal temperature of the battery. However, the internal temperature detection of EIS relies on the impedance characteristics of the battery at a specific frequency. Once the battery ages, the impedance changes significantly, which seriously interferes with the accuracy of the internal temperature estimation. Therefore, how to quickly estimate the internal temperature of the battery is a problem that needs to be considered at present. Summary of the invention

[0003] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a method and device for quickly estimating the internal temperature of a battery, thereby solving the deficiencies of the prior art.

[0004] The object of the present invention is achieved by the following technical solution: a method for quickly estimating the internal temperature of a battery, the method comprising: Step 1: Establish a battery electrochemical-thermal coupling model with impedance output to simulate impedance output, and obtain EIS data of the battery under different states to verify the model; Step 2: After verifying the EIS data through KK, perform DRT conversion and obtain the high-frequency peak area characteristics related to the high internal temperature of the battery through correlation analysis; Step 3: Remove low-frequency data and select a set frequency sampling range to reduce the amount of data and computational complexity. Establish a Transformer-based multi-head attention mechanism prediction model to capture complex nonlinear relationships and achieve fast and lossless estimation of the internal temperature of the battery.

[0005] The obtaining of EIS data of the battery in different states to verify the model includes: Carry out cycle aging experiment on the selected power battery to be tested. The battery is left to stand for a period of time after each charge and discharge, and the temperature sensor is used to collect the battery surface temperature data; Measure EIS data at every 5°C in the range of 10°C to 50°C and every 25%SOC in the range of 0%SOC to 100%SOC, and measure EIS data at every 5%SOH in the range of 100%SOH to 80%SOH; The measured EIS data at different temperatures, SOC and SOH are used to verify the battery electrochemical-thermal coupling model with impedance output.

[0006] The second step comprises: The KK relationship is used to evaluate the EIS data, and then the impedance 𝑍( f ) and DRT( t ) is expressed as , and the DRT of the battery impedance spectrum is calculated using the regularization method, where f is the frequency at which impedance testing is required, t is the relaxation time constant, j is an imaginary unit, R 0 means ohmic impedance, R pol is the polarization impedance; The key features related to the internal temperature of the battery in DRT are extracted through the feature extraction algorithm. The DRT curve is differentiated and the value of the derivative is selected as 0 to obtain different peak values ​​and peak center positions. The peak area in DRT is calculated by the integration method.

[0007] The step three comprises: The EIS frequency range in the DRT analysis was set to 10 Hz-1000 Hz to obtain the peaks appearing in the high-frequency region, and the sampling density was set to 5 sampling points per decade after removing the low-frequency data points; The DRT high-frequency peak area under different SOC and SOH conditions is used as input, and the simulated battery internal temperature is used as output. The machine learning model for battery internal temperature prediction is trained through the Transformer architecture to obtain a prediction model. The DRT data is converted into a high-dimensional vector form through an embedding layer to make it suitable for processing by the Transformer architecture. The SOC and SOH data are also processed through the embedding layer and used as input to the prediction model together with the DRT data. Each Transformer encoder layer fully captures the relationship between different features through a multi-head attention mechanism, and each Transformer encoder layer is followed by a feedforward neural network to process the features extracted by the self-attention mechanism; The processed features are converted into the final predicted value, i.e. the internal temperature of the battery, through a feedforward neural network and a fully connected layer.

[0008] A device for quickly estimating the internal temperature of a battery, the device comprising an electrochemical impedance spectroscopy measurement module, a relaxation time data conversion module and a model training and prediction module; The electrochemical impedance spectroscopy measurement module is used to establish a battery electrochemical-thermal coupling model with impedance output to simulate impedance output, and obtain EIS data of the battery in different states to verify the model; The relaxation time data conversion module is used to perform DRT conversion after verifying the EIS data through KK, and obtain the high-frequency peak area characteristics related to the internal temperature of the battery through correlation analysis; The model training and prediction module is used to remove low-frequency data, select and set the frequency sampling range to reduce the amount of data and computational complexity, establish a Transformer-based multi-head attention mechanism prediction model, capture complex nonlinear relationships, and achieve fast and lossless estimation of the internal temperature of the battery.

[0009] The obtaining of EIS data of the battery in different states to verify the model includes: Carry out cycle aging experiment on the selected power battery to be tested. The battery is left to stand for a period of time after each charge and discharge, and the temperature sensor is used to collect the battery surface temperature data; Measure EIS data at every 5°C in the range of 10°C to 50°C and every 25%SOC in the range of 0%SOC to 100%SOC, and measure EIS data at every 5%SOH in the range of 100%SOH to 80%SOH; The measured EIS data at different temperatures, SOC and SOH are used to verify the battery electrochemical-thermal coupling model with impedance output.

[0010] The relaxation time data conversion module specifically includes the following contents: The KK relationship is used to evaluate the EIS data, and then the impedance 𝑍( f ) and DRT( t ) is expressed as , and the DRT of the battery impedance spectrum is calculated using the regularization method, where f is the frequency at which impedance testing is required, t is the relaxation time constant, j is an imaginary unit, R 0 means ohmic impedance, R pol is the polarization impedance; The key features related to the internal temperature of the battery in DRT are extracted through the feature extraction algorithm. The DRT curve is differentiated and the value of the derivative is selected as 0 to obtain different peak values ​​and peak center positions. The peak area in DRT is calculated by the integration method.

[0011] The model training and prediction module specifically includes the following contents: The EIS frequency range in the DRT analysis was set to 10 Hz-1000 Hz to obtain the peaks appearing in the high-frequency region, and the sampling density was set to 5 sampling points per decade after removing the low-frequency data points; The DRT high-frequency peak area under different SOC and SOH conditions is used as input, and the simulated battery internal temperature is used as output. The machine learning model for battery internal temperature prediction is trained through the Transformer architecture to obtain a prediction model. The DRT data is converted into a high-dimensional vector form through an embedding layer to make it suitable for processing by the Transformer architecture. The SOC and SOH data are also processed through the embedding layer and used as input to the prediction model together with the DRT data. Each Transformer encoder layer fully captures the relationship between different features through a multi-head attention mechanism, and each Transformer encoder layer is followed by a feedforward neural network to process the features extracted by the self-attention mechanism; The processed features are converted into the final predicted value, i.e. the internal temperature of the battery, through a feedforward neural network and a fully connected layer.

[0012] The present invention has the following advantages: A method and device for quickly estimating the internal temperature of a battery can provide an efficient, accurate, non-invasive temperature monitoring method, which can be easily integrated into an existing BMS as part of temperature monitoring and safety management without the need for additional complex hardware. The BMS can set a temperature threshold to monitor whether the internal temperature exceeds a safe range. Once the internal temperature is too high, the BMS can automatically take measures to avoid overheating or thermal runaway of the battery, which helps to improve the safety, performance and service life of the battery system. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a schematic diagram of the process of the present invention; Figure 2 This is a structural diagram of the Transformer model. DETAILED DESCRIPTION

[0014] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the detailed description of the embodiments of the present application provided below in conjunction with the drawings is not intended to limit the scope of protection of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application. The present invention is further described below in conjunction with the drawings.

[0015] The present invention specifically relates to a method for quickly estimating the internal temperature of a battery. The introduction of the distribution of relaxation time (DRT) as a tool for interpreting EIS (electrochemical impedance spectroscopy) can effectively separate the internal processes of various electrochemical systems. First, by constructing an electrochemical-thermal coupling mechanism model that can output EIS, a large amount of battery EIS data under different working conditions is obtained to provide data support for data-driven. Then, high-frequency features that are highly correlated with the internal temperature are obtained through correlation analysis. In order to optimize the measurement efficiency, low-frequency data is removed, and a frequency range of 10Hz to 1000Hz is selected for data sampling, thereby improving the efficiency of measurement and calculation. Finally, an advanced Transformer machine learning model is used, combined with high-frequency feature data, to establish a prediction model to achieve rapid and high-precision prediction of the internal temperature of the battery.

[0016] like Figure 1 As shown, specifically including the following: 1. First, a battery coupling mechanism model that can output electrochemical impedance spectroscopy is constructed. The model formula is shown in Table 1. By simulating the electrochemical and thermodynamic reactions of the battery, the model can generate EIS data under different working conditions such as temperature, SOC, and SOH, and reflect the changes in these working conditions by controlling variables. The model with impedance data output can accurately simulate EIS data under different conditions, simulate the complex interaction between temperature and electrochemical state, and efficiently establish an EIS database.

[0017] Table 1. Coupling model equation table Most electrochemical parameters and aging parameters are affected by temperature, and the electrochemical heat generation, ohmic heat, polarization heat, and aging side reaction heat of lithium-ion batteries during the cycle process are considered to be the main heat sources. Considering that temperature plays an important role in establishing the battery model, the battery temperature is expressed by the following equation: (1), Where reversible entropy heat ( ) is generated by the entropy change of the electrode active material caused by the electrochemical reaction and only exists in the positive and negative porous electrodes. Ohmic heat in irreversible heat ( ) is caused by the charge transport in the solid conductor material and the electrolyte, and the polarization heat ( ) is caused by overpotential, and the aging side reaction heat ( ) is caused by the increase of SEI film due to battery aging. , , and Respectively expressed as: (2), (3), (4), (5), Among them, the equation parameters and formulas (1)-(5) in Table 1 are existing equations, which are the basic electrochemical mechanism equations in the electrochemical P2D model. Many parameters in them are electrochemical mechanism parameters. These parameters include thermal effects, that is, the changes in parameters are affected by temperature (Arrhenius formula).

[0018] The above equations are incorporated into the electrochemical-thermal coupling model, a key feature of which is that all parameters related to the reaction kinetics and transport processes are functions of the battery temperature. In the present invention, the diffusion coefficients of lithium ions in the solid phase and the solution phase are D s,i and the reaction rate constant s i etc. are all calculated using the Arrhenius-law: (6), in is a temperature related parameter, is the reference temperature T ref , the reference value under ，E act is the activation energy, R is the universal gas constant, and T is the battery temperature. In this study, The reference temperature was set to 25°C.

[0019] The selected power battery to be tested is subjected to a cycle aging experiment. After each charge and discharge, the battery is left to stand for a period of time, and the battery surface temperature data is collected using a temperature sensor. The electrochemical impedance of the battery is measured by applying a small current disturbance every 5°C in the temperature range of 10-50°C and every 25% SOC in the range of 0%-100% battery state of charge (SOC). The EIS data of the battery under different conditions, such as temperature, SOC, and battery health state (SOH), are obtained. The experimental data obtained are used to verify the electrochemical mechanism model with impedance output.

[0020] 2. The data collected by EIS include sampling point frequency, real impedance and imaginary impedance; (7), In the formula, is the angular frequency, is the phase angle, is the real impedance, is the imaginary impedance, U and I are the amplitudes of the excitation voltage and response current respectively.

[0021] By establishing a mechanism model with impedance output, the EIS data under different working conditions are output to provide data support for the next step of DRT conversion. Before DRT conversion, in order to ensure the quality of the data, the KK relationship is used to evaluate the EIS data to confirm the causal, linear and time-invariant characteristics. Then the impedance 𝑍( f ) and DRT( t ) is expressed as: (8), In the formula, f is the frequency at which impedance testing is required, t is the relaxation time constant, j is an imaginary unit, R 0 means ohmic impedance, R pol is the polarization impedance. The DRT conversion process is described as the integral part in the above equation (8), which unwinds the polarization resistance component, which is usually composed of a combination of semicircles in the Nyquist plot. This process effectively converts these overlapping semicircles into a distribution of time constants, facilitating the unambiguous interpretation of the impedance spectroscopy data.

[0022] Since determining the DRT function from the EIS result data is an inherently ill-posed problem, the Tikhonov regularization method is used to calculate the DRT of the battery impedance spectrum. This method uses a discretization based on the radial basis function (RBF) and the M-centered characteristic scale. The formula is as follows: (9), in x mrepresents the unknown parameters estimated by fitting the experimental impedance data, M represents the number of central features of the radial basis function (RBF), f μ is the Gaussian radial basis function, τ m is the mth relaxation time constant, and the difference between the two determines the output of the RBF, which is used to describe the impedance at the center. The motivation for choosing an RBF-based method is its advantages in achieving fast convergence and detecting peaks at the boundaries of frequency data collection, ensuring that the DRT peak is fully captured rather than being truncated at high or low frequency extremes.

[0023] The impedance spectrum in a specific frequency range is converted into DRT through the data processing module, and a feature extraction algorithm is integrated to extract key features related to the internal temperature of the battery in DRT, such as the position, height and area of ​​the peak. By differentiating the DRT curve and selecting the value of the derivative as 0, different peak values ​​and peak center positions can be obtained, and the peak area in DRT can be calculated by the integration method.

[0024] The Pearson correlation coefficient method is used to analyze the correlation between each feature and temperature change. The calculation of the Pearson correlation coefficient is shown as follows: (10), In the formula, X and Y are the characteristic value and quantitative value of the aging mechanism respectively. m X and m Y Represent the average value of the feature and the average value of the temperature information, n is the number of samples, cov(X, Y) is the covariance, σ X and σ Y is the standard deviation, when r When it tends to 1 or -1, the two are strongly correlated, while 0 means that the two are independent of each other. r If the absolute value of is greater than 0.9, it is considered that X and Y are highly correlated. Further, the highly correlated features represent the internal temperature of the battery.

[0025] Through analysis, it is found that the features that are highly correlated with the internal temperature of the battery are mainly the peak area in the high-frequency region. Therefore, the model will select the high-frequency peak area as the input feature for training. Since experiments show that temperature has a high correlation with the DRT features in the high-frequency band, the low-frequency region features have little effect on the change of temperature. At the same time, the impedance change of the electrochemical system in the low-frequency band is more complicated and the calculation time is longer, so the sampling frequency points are optimized.

[0026] To this end, the EIS frequency range in the relaxation time distribution (DRT) analysis was set to 10 Hz to 1000 Hz, with the main goal of obtaining the peak that appears in the highest frequency region. After removing the low-frequency data points, the sampling density was set to 5 sampling points per decade. This sampling method can significantly reduce the time and computational burden of electrochemical impedance measurement while ensuring measurement accuracy.

[0027] The curves selected in the range of 10Hz-1000Hz correspond to the charge transfer process of lithium ion diffusion and migration in SEI, and the battery state of charge mainly affects the voltage at the electrode. Under normal circumstances, the impedance of lithium ion diffusion and migration in SEI does not change with voltage. This shows that the SOC of the battery does not affect the charge transfer impedance of the SEI impedance. At the same time, problems such as the growth and thickening of SEI caused by increased SOH are not the main factors affecting the impedance in this frequency band. Therefore, in the frequency range of 10Hz to 1000Hz, the virtual impedance spectrum can not only be used to evaluate the internal temperature of the battery, but also to decouple the effects of SOC and SOH to a certain extent, thereby improving the prediction accuracy and robustness of the model.

[0028] The time constants of different electrochemical reactions determine their positions and shapes in the DRT peaks. The height of the peak reflects the intensity of the electrochemical reaction, and the position of the peak reflects the response speed of the electrochemical reaction.

[0029] 3. By taking the simulated internal temperature of the battery as output and the high-frequency peak area of ​​DRT under different SOC and SOH conditions as input, a battery internal temperature prediction model is trained using machine learning methods. The model can capture the complex nonlinear relationship between these characteristics and temperature, and accurately predict the internal temperature of the battery by considering the impact of SOC and SOH on electrochemical behavior.

[0030] Specifically, the Transformer architecture is used to train a machine learning model for predicting the internal temperature of the battery. Compared with traditional recurrent neural networks (RNNs) or long short-term memory networks (LSTMs), Transformer's calculations can be parallelized, which means that it can process large-scale DRT data faster, especially when the amount of data is very large, and handle complex data relationships through its unique self-attention mechanism.

[0031] like Figure 2As shown, the Transformer is composed of multiple layers of self-attention and feedforward neural networks; first, the DRT data is converted into a high-dimensional vector form through an embedding layer, and the embedding dimension is set to 128 to make it suitable for processing by the Transformer architecture. The SOC and SOH data are also processed through the embedding layer and used as the input of the model together with the DRT features. Each Transformer encoder layer uses a multi-head self-attention mechanism. In order to fully capture the relationship between different features, the Transformer encoder in the present invention is configured with 4 attention heads, which can analyze the dependencies between input features from different angles. The calculation formula for each head is as follows: (11), The multi-head attention calculation formula is as follows: (12), (13), In the formula, Q, K, and V are query, key, and value respectively. d k is the dimension of the key. is the weight matrix learned by each head.

[0032] Each Transformer encoder layer is followed by a feed-forward neural network for further processing of the features extracted by the self-attention mechanism. The feed-forward neural network consists of two linear transformation layers and a ReLU activation function: (14), Among them, W1 and W2 are learnable parameters, x is the input feature data, b1 and b2 are the corresponding bias terms, and max is the relu activation function used to introduce nonlinearity. Through the feedforward neural network and the fully connected layer, these processed features are converted into the final predicted value, that is, the internal temperature of the battery.

[0033] In model training, the Adam optimizer is used for gradient descent. The Adam optimizer combines the advantages of momentum and adaptive learning rate to accelerate convergence and avoid falling into local optimality. Its update rule can be described by the following formula: (15), (16), m t is the first-order momentum estimate, representing the average of past gradients, v t is the second-order momentum estimate, representing the average of the squared gradients, β1 and β2 are momentum parameters, and g t is the current gradient, m t-1and v t-1 The historical information used in each time step update. The initial learning rate is set to 0.001, and the learning rate scheduler is used to dynamically adjust the learning rate to ensure that the learning rate can adapt to changes during training, helping the model maintain a stable convergence speed at different training stages.

[0034] The model training data includes processed DRT data, SOC, SOH and target temperature values. The data set is divided into 80% training set and 20% test set to ensure the generalization ability of the model. In order to prevent the model from overfitting, the K-fold cross-validation method is adopted. During the training process, cross-validation can effectively evaluate the generalization performance of the model to ensure that the model can perform stably under different working conditions and data distributions. At the same time, cross-validation can further adjust the model parameters to improve its adaptability under different working conditions. The mean square error (MSE) is used as the loss function to measure the gap between the predicted temperature and the actual temperature. MSE is chosen as the loss function because it is more sensitive to larger errors and helps the model to optimize under extreme conditions (such as abnormal temperature). The loss function formula is: (17), In the formula is the predicted value, is the true value, and n is the number of samples.

[0035] Finally, the effectiveness of the model was verified, focusing on evaluating the prediction performance of the model under different SOC and SOH conditions. The evaluation indicators include root mean square error (RMSE) and mean absolute error (MAE), which are used to measure the prediction error of the model to ensure its accuracy and robustness in practical applications.

[0036] (18), (19), 4. To further verify the actual application effect of the model, the prediction model can be integrated into the battery management system (BMS) to monitor and predict the internal temperature of the battery in real time. The temperature data predicted by the model can timely identify potential battery abnormalities or overheating, thereby preventing safety problems. At the same time, the system can also help optimize the working efficiency of the battery and extend its service life.

[0037] The above is only a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the form disclosed herein, and should not be regarded as excluding other embodiments, but can be used for various other combinations, modifications and improvements, and can be modified within the scope of the concept described herein through the above teachings or the technology or knowledge of the relevant field. The changes and modifications made by those skilled in the art do not deviate from the spirit and scope of the present invention, and should be within the scope of protection of the claims attached to the present invention.

Claims

1. A method for quickly estimating the internal temperature of a battery, characterized in that: The method comprises: Step 1: Establish a battery electrochemical-thermal coupling model with impedance output to simulate impedance output, and obtain EIS data of the battery under different states to verify the model; Step 2: After verifying the EIS data through KK, perform DRT conversion and obtain the high-frequency peak area characteristics related to the high internal temperature of the battery through correlation analysis; Step 3: Remove low-frequency data and select a set frequency sampling range to reduce the amount of data and computational complexity. Establish a Transformer-based multi-head attention mechanism prediction model to capture complex nonlinear relationships and achieve fast and lossless estimation of the internal temperature of the battery.

2. A method for rapidly estimating the internal temperature of a battery according to claim 1, characterized in that: The obtaining of EIS data of the battery in different states to verify the model includes: Carry out cycle aging experiment on the selected power battery to be tested. The battery is left to stand for a period of time after each charge and discharge, and the temperature sensor is used to collect the battery surface temperature data; Measure EIS data at every 5°C in the range of 10°C to 50°C and every 25%SOC in the range of 0%SOC to 100%SOC, and measure EIS data at every 5%SOH in the range of 100%SOH to 80%SOH; The measured EIS data of different temperatures, SOC and SOH are used to verify the battery electrochemical-thermal coupling model with impedance output.

3. A method for rapidly estimating the internal temperature of a battery according to claim 1, characterized in that: The second step comprises: The KK relationship is used to evaluate the EIS data, and then the impedance 𝑍( f ) and DRT( τ ) is expressed as , and the DRT of the battery impedance spectrum is calculated using the regularization method, where f is the frequency at which impedance testing is required, τ is the relaxation time constant, j is an imaginary unit, R 0 means ohmic impedance, R pol is the polarization impedance; The key features related to the internal temperature of the battery in DRT are extracted through the feature extraction algorithm. The DRT curve is differentiated and the value of the derivative is selected as 0 to obtain different peak values ​​and peak center positions. The peak area in DRT is calculated by the integration method.

4. A method for rapidly estimating the internal temperature of a battery according to claim 1, characterized in that: The step three comprises: The EIS frequency range in the DRT analysis was set to 10 Hz-1000 Hz to obtain the peaks appearing in the high-frequency region, and the sampling density was set to 5 sampling points per decade after removing the low-frequency data points; The DRT high-frequency peak area under different SOC and SOH conditions is used as input, and the simulated battery internal temperature is used as output. The machine learning model for battery internal temperature prediction is trained through the Transformer architecture to obtain a prediction model. The DRT data is converted into a high-dimensional vector form through an embedding layer to make it suitable for processing by the Transformer architecture. The SOC and SOH data are also processed through the embedding layer and used as input to the prediction model together with the DRT data. Each Transformer encoder layer fully captures the relationship between different features through a multi-head attention mechanism, and each Transformer encoder layer is followed by a feedforward neural network to process the features extracted by the self-attention mechanism; The processed features are converted into the final predicted value, i.e. the internal temperature of the battery, through a feedforward neural network and a fully connected layer.

5. A device for quickly estimating the internal temperature of a battery, characterized in that: The device comprises an electrochemical impedance spectroscopy measurement module, a relaxation time data conversion module and a model training and prediction module; The electrochemical impedance spectroscopy measurement module is used to establish a battery electrochemical-thermal coupling model with impedance output to simulate impedance output, and obtain EIS data of the battery in different states to verify the model; The relaxation time data conversion module is used to perform DRT conversion after verifying the EIS data through KK, and obtain the high-frequency peak area characteristics related to the internal temperature of the battery through correlation analysis; The model training and prediction module is used to remove low-frequency data, select and set the frequency sampling range to reduce the amount of data and computational complexity, establish a Transformer-based multi-head attention mechanism prediction model, capture complex nonlinear relationships, and achieve fast and lossless estimation of the internal temperature of the battery.

6. The device for rapidly estimating the internal temperature of a battery according to claim 5, characterized in that: The obtaining of EIS data of the battery in different states to verify the model includes: Carry out cycle aging experiment on the selected power battery to be tested. The battery is left to stand for a period of time after each charge and discharge, and the temperature sensor is used to collect the battery surface temperature data; Measure EIS data at every 5°C in the range of 10°C to 50°C and every 25%SOC in the range of 0%SOC to 100%SOC, and measure EIS data at every 5%SOH in the range of 100%SOH to 80%SOH; The measured EIS data at different temperatures, SOC and SOH are used to verify the battery electrochemical-thermal coupling model with impedance output.

7. The device for rapidly estimating the internal temperature of a battery according to claim 5, characterized in that: The relaxation time data conversion module specifically includes the following contents: The KK relationship is used to evaluate the EIS data, and then the impedance 𝑍( f ) and DRT( τ ) is expressed as , and the DRT of the battery impedance spectrum is calculated using the regularization method, where f is the frequency at which impedance testing is required, τ is the relaxation time constant, j is an imaginary unit, R 0 means ohmic impedance, R pol is the polarization impedance; The key features related to the internal temperature of the battery in DRT are extracted through the feature extraction algorithm. The DRT curve is differentiated and the value of the derivative is selected as 0 to obtain different peak values ​​and peak center positions. The peak area in DRT is calculated by the integration method.

8. The device for rapidly estimating the internal temperature of a battery according to claim 5, characterized in that: The model training and prediction module specifically includes the following contents: The EIS frequency range in the DRT analysis was set to 10 Hz-1000 Hz to obtain the peaks appearing in the high-frequency region, and the sampling density was set to 5 sampling points per decade after removing the low-frequency data points; The DRT high-frequency peak area under different SOC and SOH conditions is used as input, and the simulated battery internal temperature is used as output. The machine learning model for battery internal temperature prediction is trained through the Transformer architecture to obtain a prediction model. The DRT data is converted into a high-dimensional vector form through an embedding layer to make it suitable for processing by the Transformer architecture. The SOC and SOH data are also processed through the embedding layer and used as input to the prediction model together with the DRT data. Each Transformer encoder layer fully captures the relationship between different features through a multi-head attention mechanism, and each Transformer encoder layer is followed by a feedforward neural network to process the features extracted by the self-attention mechanism; The processed features are converted into the final predicted value, i.e. the internal temperature of the battery, through a feedforward neural network and a fully connected layer.

Images