Battery degradation model construction method based on Bayesian physical information neural network

Abstract

The invention discloses a battery degradation model construction method based on a Bayesian physical information neural network, and the method comprises the following steps: constructing a pseudo-two-dimensional battery data generation module based on a battery aging mechanism; designing a feature extraction network to extract IC feature parameters; constructing an aging parameter mapping network to describe the relationship between the IC characteristic parameters and the battery aging parameters; and constructing a Bayesian battery health state inference network in a parameter randomization mode. By adopting the battery degradation model construction method based on the Bayesian physical information neural network, while explicit mapping of IC features and aging parameters is realized, physical residual constraints are constructed by using an electrochemical equation in the battery, so that the battery degradation model has relatively high physical interpretability, and the risk of a black box model is avoided; through a Bayesian framework based on random parameter distribution modeling, uncertainty in a model modeling process is quantified, so that confidence quantification of model prediction is realized, and a risk sensitive decision is supported.

Description

Technical Field

[0001] The present invention relates to the technical field of battery degradation modeling, and in particular to a method for constructing a battery degradation model based on a Bayesian physical information neural network. Background Art

[0002] With the global energy transition and the rapid development of electric vehicles and renewable energy storage, battery technology, as one of the core supporting technologies, has garnered widespread attention. Battery performance directly impacts the range of electric vehicles, the endurance of electronic devices, and the energy storage efficiency of power grids. However, battery performance inevitably degrades over time, which not only shortens the battery's lifespan but can also pose safety risks. Therefore, accurately predicting battery degradation and optimizing battery usage and maintenance strategies have become critical challenges for the battery industry.

[0003] Currently, research on battery degradation models focuses on two main categories: physical models and data-driven models. Physical models describe the battery degradation process by establishing electrochemical reaction equations within the battery. Although they are highly interpretable, the model's accuracy and generalization capabilities are often limited due to the complex chemical reactions and material properties within the battery. Data-driven models, on the other hand, utilize deep learning technology to construct multi-layer neural networks based on a large amount of experimental data to automatically extract the complex features of the battery degradation process, and to a certain extent improve prediction accuracy. However, most existing deep learning models rely on deterministic predictions based on precise data and lack the ability to quantify uncertainty, which is particularly important in battery degradation prediction, as the battery degradation process is affected by a variety of uncertain factors, such as temperature, charge and discharge rate, and degree.

[0004] To address the shortcomings of existing research, this paper proposes a method for constructing a battery degradation model based on a Bayesian physical information neural network. This method combines the powerful feature extraction capabilities of deep learning with the uncertainty quantification advantages of Bayesian inference, enabling more accurate predictions of battery degradation trends and providing uncertainty estimates for degradation predictions. This method can provide battery manufacturers, electric vehicle manufacturers, and energy storage system operators with a more accurate battery degradation prediction tool, thereby optimizing battery usage and maintenance strategies, extending battery life, reducing operating costs, and improving system safety and reliability. Summary of the Invention

[0005] The purpose of the present invention is to provide a battery degradation model construction method based on a Bayesian physical information neural network. While realizing the explicit mapping of IC characteristics and aging parameters, the electrochemical equations inside the battery are used to construct physical residual constraints, so that the battery degradation model has a high physical interpretability and avoids the risks of black box models. Through a Bayesian framework based on random parameter distribution modeling, the uncertainty in the model modeling process is quantified, thereby realizing the confidence quantification of model predictions and supporting risk-sensitive decision-making.

[0006] To achieve the above objectives, the present invention provides a method for constructing a battery degradation model based on a Bayesian physical information neural network, comprising the following steps: Step S1: Based on the battery aging mechanism, a pseudo two-dimensional battery data generation module is constructed to generate training data for the battery degradation model; Step S2: designing a feature extraction network for extracting IC characteristic parameters from the voltage-capacity curve; Step S3: constructing an aging parameter mapping network to characterize the relationship between IC characteristic parameters and battery aging parameters; Step S4: construct a Bayesian battery health status inference network by randomizing parameters, whose input is the battery aging parameter and output is the probability distribution of the battery SOH; Step S5: During model training, the Butler-Volmer equation and the diffusion equation are used to design a loss function, physical constraints are added to the aging parameter mapping network, and the stochastic gradient variational Bayesian method is used to estimate the posterior distribution of the Bayesian neural network parameters, ultimately obtaining a battery degradation model.

[0007] Preferably, in step S1, based on the battery aging mechanism, a pseudo two-dimensional battery data generation module is constructed to generate multi-dimensional training data for the battery degradation model. The specific process is as follows: Step S11, determining battery and operating parameters; Step S111, battery parameters, specifically including: positive / negative electrode materials, active material volume fraction, initial lithium concentration; Step S112: operating parameters, specifically including: charge and discharge rate, temperature, and number of cycles; Step S12: using the above parameters and the PyBamm toolkit, SEI growth and active material loss aging mechanisms are introduced to improve the pseudo two-dimensional battery model; Step S13: By combining different working condition parameters, scan and generate multiple groups of aging status data, each group contains: time series voltage data , capacity decay curve , voltage-capacity curve; Step S14: numerically differentiate the voltage-capacity curve to calculate the IC characteristic parameter label for subsequent feature extraction network training, specifically including: peak voltage , peak area , half-peak width .

[0008] Preferably, in step S2, a feature extraction network is designed to extract IC characteristic parameters from the voltage-capacity curve. The specific process is as follows: Step S21: Clean and interpolate the voltage-capacity curve to obtain a uniformly spaced voltage V and capacity Q sequence; Step S22: construct a feature extraction network, which specifically includes an input layer, a 1D convolutional layer, a maximum pooling layer, a physical constraint layer, and a fully connected layer; Step S23: input the voltage-capacity curve processed in step S21 into the input layer of the feature extraction network; Step S24: extracting local voltage-capacity correlation features in the 1D convolutional layer; Step S25: Using a maximum pooling layer to reduce the dimension of the extracted local voltage-capacity correlation features and retain significant features; Step S26: constraining the mathematical characteristics of the network output IC curve by designing a loss function at the physical constraint layer; Step S27: Directly regress IC feature parameters through the fully connected layer 、 、 .

[0009] Preferably, in step S26, a loss function is designed, specifically including: Step S261: Basic loss :The difference between the predicted IC parameters and the label data is calculated through the mean square error MSE; Step S262: Physical constraints : Using the Savitsky-Golay differential operator, a differential relation constraint is added to the loss function to ensure that the mathematical characteristics of the network intermediate features are consistent with the IC curve, as shown below: ; in, Capacity series predicted for the network; is the real capacity sequence, i.e. label data; is the Savitzky-Golay differential operation; is the total number of data points; Step S263: Total loss function , as shown below: ; in, is a hyperparameter that balances the weight of the mean squared error loss and the physical constraint term.

[0010] Preferably, in step S3, an aging parameter mapping network is constructed to characterize the relationship between IC characteristic parameters and battery aging parameters. The specific process is as follows: Step S31, determining the relationship between IC characteristic parameters and battery aging parameters; Step S311: Battery aging causes the exchange current density to Drop, overpotential rise, eventually leading to offset; Step S312: SEI growth or particle rupture causes the active material concentration to Decline, directly leading to reduce; Step S313: Aging causes pore clogging, which results in a decrease in the diffusion coefficient Decline, leading to rise; Step S32: Use a multi-layer MLP to construct an aging parameter mapping network, the structure of which includes an input layer, a hidden layer, and an output layer; Step S33: input IC characteristic parameters, and according to the relationship between the IC characteristic parameters and the battery aging parameters, add constraints to the mapping network to achieve mapping output of the battery aging parameters.

[0011] Preferably, in step S33, the Butler-Volmer equation is used to construct the residual constraint , as shown below: ; in, is the exchange current density; and is the symmetry factor; is the Faraday constant; is the overpotential; is the ideal gas constant; is the absolute temperature; is the label current density; is the L2 norm; Total loss function , as shown below: ; in, is the residual constraint The weight coefficient of .

[0012] Preferably, in step S4, a Bayesian battery health status inference network is constructed by parameter randomization, the input of which is the battery aging parameter and the output is the probability distribution of the battery SOH. The specific process is as follows: Step S41: Define Bayesian network node parameters using Gaussian distribution The prior distribution of is as follows: ; ; in, is the weight of the network node parameter; is the bias of the network node parameters; represents variance; Represents the weight parameter The prior probability density function of ; Represents the weight parameter The prior probability density function of ; represents Gaussian distribution; Step S42: Assume that the noise of the data follows a normal distribution and select Gaussian distribution to define the likelihood function. The likelihood function represents the probability of observing correct data given the network parameters, as shown below: ; in, represents the aging parameter dataset; represents the likelihood function; Indicates the true value SOH; Represents the model predicted value SOH; Step S43: Input the battery aging parameters and output the probability distribution of the battery health status ( , ), used to quantify the uncertainty of the prediction; where represents the mean, Represents standard deviation.

[0013] Preferably, in step S5, the constructed battery degradation model is trained, and the stochastic gradient variational Bayesian method is used to realize the posterior distribution estimation of the Bayesian neural network parameters, and finally the battery degradation model is obtained. The specific process includes: Step S51: data preprocessing; Step S52: Optimization algorithm; Step S53: setting training parameters; In step S52, the specific process of performing the optimization algorithm is as follows: Step S521: Based on the Bayesian inference theorem, use variational inference to construct a set of parameters Parameterized variational distribution ( ), used to fit the posterior distribution to be estimated; Among them, the posterior distribution based on the Bayesian inference theorem is: ; in, is the evidence normalization factor, the product of the likelihood function and the prior distribution is Integration over space; is the prior distribution; is the posterior distribution to be estimated; The variational distribution used ( ) is the mean field Gaussian distribution, as shown below: ; in, Network node parameters After decomposition, it is divided into multiple independent sub-variables; is the mean parameter of the variational distribution; is the variance parameter of the variational distribution; Step S522: Maximizing the evidence lower bound ELBO is used as the optimization goal of the posterior distribution fitting, that is, minimizing the negative ELBO, as shown below: ; ; in, Represents the variational distribution Below, the logarithmic expectation of the data likelihood; KL represents the variational distribution With prior distribution KL divergence of represents the joint probability distribution; ; in, represents the standard deviation of the prior distribution; represents the standard deviation of the variational distribution; represents the mean of the prior distribution; represents the mean of the variational distribution; Step S523: Using the stochastic gradient variational Bayesian method, the gradient is calculated by the reparameterization technique, specifically including: First, Expressed as a deterministic function: ; in, is element-wise multiplication, is standard Gaussian noise; is the random parameter vector generated by reparameterization; is the mean of the variational distribution; is the standard deviation of the variational distribution; Then, estimate gradient , as shown below: ; in, Indicates the total number of sampling times; , indicating the Subsampling parameters; is the variational parameter; Indicates the given parameter Time data probability; Represents the variational distribution in The probability density of Representation parameters The prior probability density of ; Step S524: Minimizing the negative ELBO is used as the loss calculation method, and the back propagation algorithm is used to update the gradient. Finally, the variational distribution when the loss converges is used as an approximation of the posterior distribution.

[0014] A battery degradation model construction system based on a Bayesian physical information neural network, including a pseudo-two-dimensional battery data generation module, a feature extraction network module, a battery aging parameter mapping module, and a Bayesian battery health status inference module; Among them, the pseudo two-dimensional battery data generation module is used to generate training data for the battery degradation model; Feature extraction network module, used to extract IC characteristic parameters from voltage-capacity curve; The battery aging parameter mapping module is used to explicitly map aging parameters. It also uses the physical partial differential equations of the battery degradation process to construct physical constraint losses and incorporate them into the model training process, making the battery degradation model conform to physical laws and physically interpretable. The Bayesian battery health state inference module is used for randomized parameter modeling of the model. It uses battery aging parameters and prior knowledge to achieve battery health inference and model uncertainty quantification.

[0015] A computer device includes a memory and a processor; the memory stores a computer program, and the processor executes the computer program to implement the steps of a method for constructing a battery degradation model based on a Bayesian physical information neural network.

[0016] Therefore, the present invention adopts the above-mentioned battery degradation model construction method based on the Bayesian physical information neural network. While realizing the explicit mapping of IC characteristics and aging parameters, it uses the electrochemical equations inside the battery to construct physical residual constraints, so that the battery degradation model has a high physical interpretability and avoids the risks of the black box model; through the Bayesian framework based on random parameter distribution modeling, the uncertainty in the model modeling process is quantified, thereby realizing the confidence quantification of model predictions and supporting risk-sensitive decision-making.

[0017] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flow chart of the method for constructing a battery degradation model based on a Bayesian physical information neural network of the present invention; Figure 2 This is a flow chart of the design of a feature extraction network for extracting IC feature parameters; Figure 3 This is a flowchart of the present invention for constructing an aging parameter mapping network to depict the relationship between IC characteristic parameters and battery aging parameters. DETAILED DESCRIPTION

[0019] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0020] like Figure 1 As shown, the method for constructing a battery degradation model based on a Bayesian physical information neural network of the present invention includes the following steps: Step S1: constructing a pseudo-two-dimensional (P2D) battery data generation module based on the battery aging mechanism to generate training data for the battery degradation model; Step S2: designing a feature extraction network to extract IC characteristic parameters from the voltage-capacity (VQ) curve; Step S3: constructing an aging parameter mapping network to characterize the relationship between IC characteristic parameters and battery aging parameters; Step S4: construct a Bayesian SOH (battery state of health) inference network by parameter randomization, whose input is the battery aging parameter and output is the probability distribution of the battery SOH; Step S5: During model training, the Butler-Volmer equation and the diffusion equation are used to design a loss function to add physical constraints to the aging parameter mapping network. The stochastic gradient variational Bayesian (SGVB) method is used to estimate the posterior distribution of the Bayesian neural network parameters, and finally a battery degradation model is obtained.

[0021] Example 1

[0022] Step S1: Based on the battery aging mechanism, a pseudo two-dimensional (P2D) battery data generation module is constructed to generate multi-dimensional training data for a battery degradation model.

[0023] Step S11: Determine battery and operating parameters.

[0024] Step S111: Battery parameters, specifically including: (1) The positive electrode material is , the active material volume fraction is 0.65, and the initial lithium concentration is 2000 mol / m 3 .

[0025] (2) The negative electrode material is graphite, the active material volume fraction is 0.70, and the initial lithium concentration is 0 mol / m 3 .

[0026] Step S112: operating parameters, specifically including: (1) Charge and discharge rate: 1C, 2C, 3C.

[0027] (2) Temperature: 25℃, 35℃, 45℃.

[0028] (3) Number of cycles: 50 times, 100 times, 150 times, 200 times.

[0029] Step S12: Using the above parameters and the PyBamm toolkit, aging mechanisms such as SEI growth and active material loss are introduced to improve the P2D battery model.

[0030] SEI film growth is described by the Butler-Volmer equation, with parameters including the reaction rate constant and activation energy. Active material loss simulates the decay of active material concentration due to particle rupture, which is exponentially related to the number of cycles.

[0031] Step S13: By combining different working condition parameters, 3×3×4=36 groups of aging status data are scanned and generated, each group including: (1) Time series voltage data , the sampling frequency is 10Hz.

[0032] (2) Capacity decay curve , record once every 5 cycles.

[0033] (3) Voltage-capacity curve (VQ curve), that is, the curve showing the change of voltage with capacity during the charging and discharging process.

[0034] Step S14: numerically differentiate the VQ curve, set the window size of the Savitzky-Golay filter to 7, and calculate the IC characteristic parameter label, specifically including: (1) Peak voltage is the voltage corresponding to the peak value of the differential curve; (2) Peak area is the integrated area of ​​the peak region of the differential curve; (3) Half-peak width is the width at half the height of the peak.

[0035] The calculated IC feature parameter labels are used for subsequent feature extraction network training, so that the model is not limited to a specific data set, and the neural network mapping method has high computational efficiency.

[0036] Step S2: Design a feature extraction network to extract IC characteristic parameters from the voltage-capacity (VQ) curve, such as Figure 2 shown.

[0037] Step S21: perform data cleaning and interpolation on the VQ curve to obtain V and Q sequences with uniform and equal intervals.

[0038] Step S22: Construct a feature extraction network, which specifically includes an input layer, a 1D convolution layer, a maximum pooling layer, a physical constraint layer, and a fully connected layer, as shown in Table 1.

[0039] Table 1 Feature extraction network architecture ;

[0040] Step S23: input the preprocessed VQ curve into the feature extraction network input layer.

[0041] Step S24: extract local voltage-capacity correlation features in the 1D convolution layer.

[0042] Step S25: Reduce the dimension of the extracted local voltage-capacity correlation features through the maximum pooling layer and retain the significant features.

[0043] Step S26: Design a loss function at the physical constraint layer to constrain the network output to conform to the mathematical characteristics of the IC characteristic curve; wherein the loss function includes a true value and a predicted value.

[0044] Step S261: Basic loss : The difference between the predicted IC parameters and the label data is calculated through the mean square error (MSE).

[0045] Step S262: Physical constraints : Using the Savitsky-Golay differential operator, a differential relation constraint is added to the loss function to ensure that the mathematical characteristics of the network intermediate features are consistent with the IC curve, as shown below: ; in, Capacity series predicted for the network; is the real capacity sequence, i.e. label data; is the Savitzky-Golay differential operation; is the total number of data points.

[0046] Step S263: Total loss function , as shown below: ; in, is a hyperparameter that balances the weight of the mean square error loss and the physical constraint term. .

[0047] Step S27: Directly regress IC feature parameters through the fully connected layer 、 、 .

[0048] Step S3: construct an aging parameter mapping network to describe the relationship between IC characteristic parameters and battery aging parameters, such as Figure 3 shown.

[0049] Step S31: Determine the relationship between the IC characteristic parameters and the battery aging parameters.

[0050] Step S311: Battery aging causes the exchange current density to Drop, overpotential rise, eventually leading to offset; Step S312: SEI growth or particle rupture causes the active material concentration to Decline, directly leading to reduce; Step S313: Aging causes pore blockage and other phenomena that make the diffusion coefficient Decline, leading to rise.

[0051] Step S32: Use a multi-layer MLP to construct an aging parameter mapping network, the structure of which is as follows: (1) Input layer: 3 neurons, corresponding to IC feature parameters 、 、 ; (2) Hidden layer: 2 layers, 32 neurons in each layer, using ReLU activation function; (3) Output layer: 3 neurons, corresponding to aging parameters 、 、 .

[0052] Step S33: input IC characteristic parameters, and add constraints to the mapping network according to the relationship between the IC characteristic parameters and the battery aging parameters to achieve mapping output of the battery aging parameters.

[0053] Using the Butler-Volmer equation, construct residual constraints , as shown below: ; in, is the exchange current density; and is the symmetry factor, take ; is the Faraday constant; is the overpotential, calculated from the voltage data; is the ideal gas constant; is the absolute temperature; is the label current density; is the L2 norm.

[0054] Total loss function , as shown below: ; in, is the residual constraint The weight coefficient is .

[0055] Step S4: By randomizing parameters, a Bayesian SOH (battery state of health) inference network is constructed, whose input is the battery aging parameter and output is the probability distribution of the battery SOH.

[0056] Step S41: Define Bayesian network node parameters using Gaussian distribution The prior distribution of is as follows: ; ; in, is the weight of the network node parameter; is the bias of the network node parameters; represents variance; Represents the weight parameter The prior probability density function of ; Represents the weight parameter The prior probability density function of ; Represents a Gaussian distribution.

[0057] Step S42: Assuming that the noise of the data follows a normal distribution, a Gaussian distribution is selected to define the likelihood function. The likelihood function represents the probability of observing the correct data given the network parameters, as shown below: ; in, represents the aging parameter dataset; represents the likelihood function; Indicates the true value SOH; Represents the model predicted value SOH.

[0058] Step S43: Input battery aging parameters and output the probability distribution of SOH ( , ), used to quantify the uncertainty of the prediction; where represents the mean, Represents standard deviation.

[0059] Step S5: Based on the above process, the constructed battery degradation model is trained, and the stochastic gradient variational Bayesian (SGVB) method is used to realize the posterior distribution estimation of the Bayesian neural network parameters, and finally the battery degradation model is obtained.

[0060] Step S51: data preprocessing.

[0061] Step S511: interpolate the VQ curve to 500 equally spaced points; Step S512: normalize the IC characteristic parameters and aging parameters to the interval [0, 1].

[0062] Step S52: Optimize algorithm.

[0063] Step S521: Based on the Bayesian inference theorem, use variational inference to construct a set of parameters Parameterized variational distribution ( ), which is used to fit the posterior distribution to be estimated.

[0064] Among them, the posterior distribution to be fitted based on the Bayesian inference theorem is: ; in, is the evidence normalization factor, the product of the likelihood function and the prior distribution is Integration over space; is the prior distribution; is the posterior distribution to be estimated.

[0065] The variational distribution used ( ) is the mean field Gaussian distribution, as shown below: .

[0066] in, Network node parameters After decomposition, it is divided into multiple independent sub-variables; is the mean parameter of the variational distribution; is the variance parameter of the variational distribution.

[0067] Step S522: Maximizing the Evidence Lower Bound (ELBO) is used as the optimization objective for fitting the posterior distribution, that is, minimizing the negative ELBO, as shown below: ; ; in, Represents the variational distribution Below, the logarithmic expectation of the data likelihood; KL represents the variational distribution With prior distribution KL divergence of represents the joint probability distribution.

[0068] ; in, represents the standard deviation of the prior distribution (usually 1); represents the standard deviation of the variational distribution; represents the mean of the prior distribution; represents the mean of the variational distribution; Step S523: Use the stochastic gradient variational Bayes (SGVB) method to calculate the gradient through the reparameterization technique.

[0069] Will Expressed as a deterministic function: ; in, is element-wise multiplication, is standard Gaussian noise; is the random parameter vector generated by reparameterization (as the sampling output of the variational distribution, it is the parameter actually input into the model network node); is the mean of the variational distribution; is the standard deviation of the variational distribution.

[0070] estimate gradient , as shown below: ; in, Indicates the total number of sampling times; , indicating the Subsampling parameters; is the variational parameter; Indicates the given parameter Time data probability; The variation distribution is represented by The probability density of Representation parameters The prior probability density of .

[0071] Step S524: Minimizing the negative ELBO is used as the loss calculation method, and the back propagation algorithm is used to update the gradient. Finally, the variational distribution when the loss converges is used as an approximation of the posterior distribution.

[0072] Step S53: Set training parameters: (1) Batch size = 32, epoch = 200; (2) Learning rate = 1e-3, using Adam optimizer; (3) Verify the physical residual and prediction accuracy every 10 epochs.

[0073] Example 2

[0074] This embodiment constructs a test scenario for experimentally verifying the physical interpretability of the model building method proposed in the present invention.

[0075] 1. Experimental environment.

[0076] (1) Hardware configuration, as shown in Table 2.

[0077] Table 2 Hardware configuration ;

[0078] (2) Software implementation, as shown in Table 3.

[0079] Table 3 Software Configuration

[0080] ;

[0081] 2. Set training parameters.

[0082] A set of operating conditions that were not used in the training was selected: temperature = 30°C, charge / discharge rate = 2C, and number of cycles = 180. The key indicators compared with the existing model are shown in Table 4.

[0083] Table 4 Key indicators ;

[0084] (1) VQ curve comparison: The IC characteristics (peak voltage shift, peak area reduction) extracted by the model of the present invention are highly consistent with the actual aging data.

[0085] (2) SOH uncertainty distribution: The output includes the mean (e.g., 85.3%) and standard deviation (e.g., 1.2%), providing a basis for risk decision-making for the battery management system.

[0086] This embodiment achieves high-precision prediction and confidence assessment of the battery degradation model by integrating physical constraints and Bayesian uncertainty quantification. It has high physical interpretability and is suitable for battery health management in scenarios such as electric vehicles and energy storage systems.

[0087] Based on the above-mentioned battery degradation model construction method based on the Bayesian physical information neural network proposed by the present invention, a battery degradation model construction system based on the Bayesian physical information neural network is also proposed, including a pseudo two-dimensional (P2D) battery data generation module, a feature extraction network module, a battery aging parameter mapping module and a Bayesian battery health status inference module.

[0088] Among them, the pseudo two-dimensional (P2D) battery data generation module is used to generate training data for the battery degradation model.

[0089] Feature extraction network module, used to extract IC characteristic parameters from voltage-capacity (VQ) curve.

[0090] The battery aging parameter mapping module is used to explicitly map aging parameters. At the same time, it uses the physical partial differential equations in the battery degradation process to construct physical constraint losses and incorporates them into the model training process, so that the battery degradation model conforms to physical laws and has physical interpretability.

[0091] The Bayesian battery health state inference module is used for randomized parameter modeling of the model. It uses battery aging parameters and prior knowledge to achieve battery health inference and model uncertainty quantification.

[0092] Therefore, the present invention adopts the above-mentioned battery degradation model construction method based on the Bayesian physical information neural network. While realizing the explicit mapping of IC characteristics and aging parameters, it uses the electrochemical equations inside the battery to construct physical residual constraints, so that the battery degradation model has a high physical interpretability and avoids the risks of the black box model; through the Bayesian framework based on random parameter distribution modeling, the uncertainty in the model modeling process is quantified, thereby realizing the confidence quantification of model predictions and supporting risk-sensitive decision-making.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for constructing a battery degradation model based on a Bayesian physical information neural network, characterized in that: The following steps are involved: Step S1: Based on the battery aging mechanism, a pseudo two-dimensional battery data generation module is constructed to generate training data for the battery degradation model; Step S2: designing a feature extraction network for extracting IC characteristic parameters from the voltage-capacity curve; Step S3: constructing an aging parameter mapping network to characterize the relationship between IC characteristic parameters and battery aging parameters; Step S4: construct a Bayesian battery health status inference network by randomizing parameters, whose input is the battery aging parameter and output is the probability distribution of the battery SOH; Step S5: During model training, the Butler-Volmer equation and the diffusion equation are used to design a loss function, physical constraints are added to the aging parameter mapping network, and the stochastic gradient variational Bayesian method is used to estimate the posterior distribution of the Bayesian neural network parameters, ultimately obtaining a battery degradation model.

2. The method for constructing a battery degradation model based on a Bayesian physical information neural network according to claim 1, characterized in that: In step S1, based on the battery aging mechanism, a pseudo-two-dimensional battery data generation module is constructed to generate multi-dimensional training data for the battery degradation model. The specific process is as follows: Step S11, determining battery and operating parameters; Step S111, battery parameters, specifically including: positive / negative electrode materials, active material volume fraction, initial lithium concentration; Step S112: operating parameters, specifically including: charge and discharge rate, temperature, and number of cycles; Step S12: using the above parameters and the PyBamm toolkit, SEI growth and active material loss aging mechanisms are introduced to improve the pseudo two-dimensional battery model; Step S13: By combining different working condition parameters, scan and generate multiple groups of aging status data, each group contains: time series voltage data , capacity decay curve , voltage-capacity curve; Step S14: numerically differentiate the voltage-capacity curve to calculate the IC characteristic parameter label for subsequent feature extraction network training, specifically including: peak voltage , peak area , half-peak width .

3. The method for constructing a battery degradation model based on a Bayesian physical information neural network according to claim 1, characterized in that: In step S2, a feature extraction network is designed to extract IC characteristic parameters from the voltage-capacity curve. The specific process is as follows: Step S21: Clean and interpolate the voltage-capacity curve to obtain a uniformly spaced voltage V and capacity Q sequence; Step S22: construct a feature extraction network, which specifically includes an input layer, a 1D convolutional layer, a maximum pooling layer, a physical constraint layer, and a fully connected layer; Step S23: input the voltage-capacity curve processed in step S21 into the input layer of the feature extraction network; Step S24: extracting local voltage-capacity correlation features in the 1D convolutional layer; Step S25: Using a maximum pooling layer to reduce the dimension of the extracted local voltage-capacity correlation features and retain significant features; Step S26: constraining the mathematical characteristics of the network output IC curve by designing a loss function at the physical constraint layer; Step S27: Directly regress IC feature parameters through the fully connected layer 、 、 .

4. The method for constructing a battery degradation model based on a Bayesian physical information neural network according to claim 3, characterized in that: In step S26, a loss function is designed, specifically including: Step S261: Basic loss :The difference between the predicted IC parameters and the label data is calculated through the mean square error MSE; Step S262: Physical constraints : Using the Savitsky-Golay differential operator, a differential relation constraint is added to the loss function to ensure that the mathematical characteristics of the network intermediate features are consistent with the IC curve, as shown below: ； in, Capacity series predicted for the network; is the real capacity sequence, i.e. label data; is the Savitzky-Golay differential operation; is the total number of data points; Step S263: Total loss function , as shown below: ； in, is a hyperparameter that balances the weight of the mean squared error loss and the physical constraint term.

5. The method for constructing a battery degradation model based on a Bayesian physical information neural network according to claim 1, characterized in that: In step S3, an aging parameter mapping network is constructed to characterize the relationship between IC characteristic parameters and battery aging parameters. The specific process is as follows: Step S31, determining the relationship between IC characteristic parameters and battery aging parameters; Step S311: Battery aging causes the exchange current density to Drop, overpotential rise, eventually leading to offset; Step S312: SEI growth or particle rupture causes the active material concentration to Decline, directly leading to reduce; Step S313: Aging causes pore clogging, which results in a decrease in the diffusion coefficient Decline, leading to rise; Step S32: Use a multi-layer MLP to construct an aging parameter mapping network, the structure of which includes an input layer, a hidden layer, and an output layer; Step S33: input IC characteristic parameters, and according to the relationship between the IC characteristic parameters and the battery aging parameters, add constraints to the mapping network to achieve mapping output of the battery aging parameters.

6. The method for constructing a battery degradation model based on a Bayesian physical information neural network according to claim 5, characterized in that: In step S33, the Butler-Volmer equation is used to construct the residual constraint , as shown below: ； in, is the exchange current density; and is the symmetry factor; is the Faraday constant; is the overpotential; is the ideal gas constant; is the absolute temperature; is the label current density; is the L2 norm; Total loss function , as shown below: ； in, is the residual constraint The weight coefficient of .

7. The method for constructing a battery degradation model based on a Bayesian physical information neural network according to claim 1, characterized in that: In step S4, a Bayesian battery health status inference network is constructed by parameter randomization. Its input is the battery aging parameter and its output is the probability distribution of the battery SOH. The specific process is as follows: Step S41: Define Bayesian network node parameters using Gaussian distribution The prior distribution of is as follows: ； ； in, is the weight of the network node parameter; is the bias of the network node parameters; represents variance; Represents the weight parameter The prior probability density function of ; Represents the weight parameter The prior probability density function of ; represents Gaussian distribution; Step S42: Assume that the noise of the data follows a normal distribution and select Gaussian distribution to define the likelihood function. The likelihood function represents the probability of observing correct data given the network parameters, as shown below: ； in, represents the aging parameter dataset; represents the likelihood function; Indicates the true value SOH; Represents the model predicted value SOH; Step S43: Input the battery aging parameters and output the probability distribution of the battery health status ( , ), used to quantify the uncertainty of the prediction; where represents the mean, Represents standard deviation.

8. The method for constructing a battery degradation model based on a Bayesian physical information neural network according to claim 1, characterized in that: In step S5, the constructed battery degradation model is trained using the stochastic gradient variational Bayesian method to estimate the posterior distribution of the Bayesian neural network parameters, and finally the battery degradation model is obtained. The specific process includes: Step S51: data preprocessing; Step S52: Optimization algorithm; Step S53: setting training parameters; In step S52, the specific process of performing the optimization algorithm is as follows: Step S521: Based on the Bayesian inference theorem, use variational inference to construct a set of parameters Parameterized variational distribution ( ), used to fit the posterior distribution to be estimated; Among them, the posterior distribution based on the Bayesian inference theorem is: ； in, is the evidence normalization factor, the product of the likelihood function and the prior distribution is Integration over space; is the prior distribution; is the posterior distribution to be estimated; The variational distribution used ( ) is the mean field Gaussian distribution, as shown below: ； in, Network node parameters After decomposition, it is divided into multiple independent sub-variables; is the mean parameter of the variational distribution; is the variance parameter of the variational distribution; Step S522: Maximizing the lower bound of evidence ELBO is used as the optimization goal of the posterior distribution fitting, that is, minimizing the negative ELBO, as shown below: ； ； in, Represents the variational distribution Below, the logarithmic expectation of the data likelihood; KL represents the variational distribution With prior distribution KL divergence of represents the joint probability distribution; ； in, represents the standard deviation of the prior distribution; represents the standard deviation of the variational distribution; represents the mean of the prior distribution; represents the mean of the variational distribution; Step S523: Using the stochastic gradient variational Bayesian method, the gradient is calculated by the reparameterization technique, specifically including: First, Expressed as a deterministic function: ; in, is element-wise multiplication, is standard Gaussian noise; is the random parameter vector generated by reparameterization; is the mean of the variational distribution; is the standard deviation of the variational distribution; Then, estimate gradient , as shown below: ； in, Indicates the total number of sampling times; , indicating the Subsampling parameters; is the variational parameter; Indicates the given parameter Time data probability; Represents the variational distribution in The probability density of Representation parameters The prior probability density of ; Step S524: Minimizing the negative ELBO is used as the loss calculation method, and the back propagation algorithm is used to update the gradient. Finally, the variational distribution when the loss converges is used as an approximation of the posterior distribution.

9. The system for constructing a battery degradation model based on a Bayesian physical information neural network according to any one of claims 1 to 8, characterized in that: It includes a pseudo-two-dimensional battery data generation module, a feature extraction network module, a battery aging parameter mapping module, and a Bayesian battery health status inference module; Among them, the pseudo two-dimensional battery data generation module is used to generate training data for the battery degradation model; Feature extraction network module, used to extract IC characteristic parameters from voltage-capacity curve; The battery aging parameter mapping module is used to explicitly map aging parameters. It also uses the physical partial differential equations of the battery degradation process to construct physical constraint losses and incorporate them into the model training process, making the battery degradation model conform to physical laws and physically interpretable. The Bayesian battery health state inference module is used for randomized parameter modeling of the model. It uses battery aging parameters and prior knowledge to achieve battery health inference and model uncertainty quantification.

10. A computer device, characterized in that: The method comprises a memory and a processor; the memory stores a computer program, and the processor executes the computer program to implement the steps of the method according to any one of claims 1 to 8.

Images