Lithium battery relaxation time distribution determination method and device, equipment and storage medium

Abstract

The embodiment of the invention provides a lithium battery relaxation time distribution determination method and device, equipment and a storage medium. The method comprises the steps that an initial Bayesian neural network model is obtained, the input of the initial Bayesian neural network model is a randomly initialized scalar, the output of the initial Bayesian neural network model is a vector composed of a weight vector of a radial basis function and an impedance estimated value of a lithium battery, the radial basis function is used for discretizing continuous relaxation time distribution, and the relaxation time distribution is obtained; the weight of the radial basis function is a discrete point of relaxation time distribution, and the loss function is an error function between an impedance value and an impedance estimated value of the electrochemical impedance spectroscopy of the lithium battery; training the initial Bayesian neural network model based on the electrochemical impedance spectroscopy to obtain a trained Bayesian neural network model; on the basis of the weight vector of the radial basis function output by the trained Bayesian neural network model, the relaxation time distribution of the lithium battery is determined, and the relaxation time distribution is accurately extracted from the electrochemical impedance spectrum of the lithium battery.

Description

Technical Field

[0001] The present application relates to the field of lithium batteries, and more specifically, to a method, device, equipment and storage medium for determining relaxation time distribution of lithium batteries. Background Art

[0002] State of charge (SOC) is an important indicator in the battery management system. High-precision SOC can improve the available capacity, service life and safety of the battery system. The SOC estimation algorithm based on the battery model has certain robustness and is the most widely used, but its estimation accuracy depends on the accuracy of the battery model. The traditional battery fractional-order modeling method usually determines the structure of the battery model by visually analyzing the electrochemical impedance spectrum (EIS), but the main electrochemical reactions of lithium batteries are usually concentrated in a limited bandwidth, which makes the EIS characteristic peaks overlap and difficult to separate, resulting in ambiguity and uncertainty in the diagnostic analysis results of EIS, and the battery model established based on this will also deviate from the real battery model.

[0003] The distribution of relaxation times (DRT) characterizes the time required for a variable to tend to a steady state from a transient state. It regards the electrochemical system as an ohmic resistor and an infinite number of polarization processes in series. The DRT analysis method can approximate the impedance model of any electrochemical system. Therefore, extracting DRT from EIS will help analyze the main chemical reaction process of lithium batteries and establish an accurate battery model based on this. The noise of EIS data is ubiquitous, but there is still no more accurate method to extract DRT from noisy EIS data. Summary of the invention

[0004] The present application provides a method, device, equipment and storage medium for determining the relaxation time distribution of a lithium battery, so as to accurately extract the relaxation time distribution from the electrochemical impedance spectrum of the lithium battery.

[0005] In a first aspect, the present application provides a method for determining relaxation time distribution of a lithium battery, comprising:

[0006] Obtaining an initial Bayesian neural network model, wherein the input of the initial Bayesian neural network model is a randomly initialized scalar, and the output is a vector composed of a weight vector of a radial basis function and an impedance estimate of the lithium battery, wherein the radial basis function is used to discretize a continuous relaxation time distribution, and the weight of the radial basis function is a discrete point of the relaxation time distribution, and the loss function of the initial Bayesian neural network model is based on an error function between an impedance value of an electrochemical impedance spectrum of the lithium battery and the impedance estimate;

[0007] Training the initial Bayesian neural network model based on the electrochemical impedance spectroscopy so that the loss value of the loss function meets a preset condition to obtain a trained Bayesian neural network model;

[0008] The relaxation time distribution of the lithium battery is determined based on the weight vector of the radial basis function output by the trained Bayesian neural network model.

[0009] In one implementation, the method further includes:

[0010] Obtaining a first relationship between an impedance value of an electrochemical impedance spectroscopy and a relaxation time distribution;

[0011] The first relationship is discretized by using a radial basis function to obtain a second relationship between the impedance value of the electrochemical impedance spectroscopy and the weight vector of the radial basis function;

[0012] An error function between the impedance value of the electrochemical impedance spectroscopy and the impedance estimation value is determined based on the second relationship.

[0013] In one implementation, the weight vector of the radial basis function outputted by the trained Bayesian neural network model is used to determine the relaxation time distribution of the lithium battery, including:

[0014] The relaxation time distribution of the lithium battery is determined based on the weight vector of the radial basis function output by the trained Bayesian neural network model, the first relationship and the second relationship.

[0015] In one implementation, the method further includes:

[0016] For the initial Bayesian neural network model, the Bayesian rule is used to determine the posterior probability distribution of neuron parameters, and variational inference is used to obtain the approximate distribution of the posterior probability distribution of the neuron parameters;

[0017] Based on the KL divergence between the posterior probability distribution of the neuron parameters and the approximate distribution, and the error function, a loss function of the initial Bayesian neural network model is determined.

[0018] In one implementation, the method further includes: during the training process, limiting the variance of the approximate distribution to a value greater than zero.

[0019] In one implementation, the method further includes: during the training process, performing gradient clipping on the gradient after each training to limit the gradient value to a preset range.

[0020] In one implementation, the method further includes:

[0021] The electrochemical impedance spectrum is preprocessed to remove high-frequency impedance frequency points in the electrochemical impedance spectrum.

[0022] In one implementation, the preprocessing of the electrochemical impedance spectrum to remove high-frequency impedance frequency points in the electrochemical impedance spectrum includes:

[0023] A threshold value is determined for the imaginary part of the electrochemical impedance spectrum, and frequency points where the imaginary part is a positive number are removed.

[0024] In a second aspect, the present application provides a device for determining relaxation time distribution of a lithium battery, comprising:

[0025] An acquisition module is used to acquire an initial Bayesian neural network model, wherein the input of the initial Bayesian neural network model is a randomly initialized scalar, and the output is a vector composed of a weight vector of a radial basis function and an impedance estimate of a lithium battery, wherein the radial basis function is used to discretize a continuous relaxation time distribution, and the weight of the radial basis function is a discrete point of the relaxation time distribution, and the loss function of the initial Bayesian neural network model is based on an error function between an impedance value of an electrochemical impedance spectrum of the lithium battery and the impedance estimate;

[0026] A training module, used for training the initial Bayesian neural network model based on the electrochemical impedance spectrum so that the loss value of the loss function satisfies a preset condition to obtain a trained Bayesian neural network model;

[0027] The determination module is used to determine the relaxation time distribution of the lithium battery based on the weight vector of the radial basis function output by the trained Bayesian neural network model.

[0028] In one implementation, the acquisition module is further used for:

[0029] Obtaining a first relationship between an impedance value of an electrochemical impedance spectroscopy and a relaxation time distribution;

[0030] The first relationship is discretized by using a radial basis function to obtain a second relationship between the impedance value of the electrochemical impedance spectroscopy and the weight vector of the radial basis function;

[0031] An error function between the impedance value of the electrochemical impedance spectroscopy and the impedance estimation value is determined based on the second relationship.

[0032] In one implementation, the determining module is used to:

[0033] The relaxation time distribution of the lithium battery is determined based on the weight vector of the radial basis function output by the trained Bayesian neural network model, the first relationship and the second relationship.

[0034] In one implementation, the acquisition module is further used for:

[0035] For the initial Bayesian neural network model, the Bayesian rule is used to determine the posterior probability distribution of neuron parameters, and variational inference is used to obtain the approximate distribution of the posterior probability distribution of the neuron parameters;

[0036] Based on the KL divergence between the posterior probability distribution of the neuron parameters and the approximate distribution, and the error function, a loss function of the initial Bayesian neural network model is determined.

[0037] In one implementation, the training module is used to: during the training process, limit the variance of the approximate distribution to a value greater than zero.

[0038] In one implementation, the training module is used to: during the training process, perform gradient clipping on the gradient after each training to limit the gradient value to a preset range.

[0039] In one implementation, the acquisition module is further used for:

[0040] The electrochemical impedance spectrum is preprocessed to remove high-frequency impedance frequency points in the electrochemical impedance spectrum.

[0041] In one implementation, the acquisition module is used to:

[0042] A threshold value is determined for the imaginary part of the electrochemical impedance spectrum, and frequency points where the imaginary part is a positive number are removed.

[0043] In a third aspect, the present application provides an electronic device, including: a memory and a processor;

[0044] The memory is used to store computer programs;

[0045] The processor is used to execute a computer program stored in the memory, and when the computer program is run, the processor is enabled to execute the method according to the first aspect.

[0046] In a fourth aspect, the present application provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the processor executes the method described in the first aspect.

[0047] In the relaxation time distribution determination method, device, equipment and storage medium of the lithium battery provided in the present application, the radial basis function is used to discretize the continuous relaxation time distribution for interpolation processing to construct the optimization problem of the weight vector of the DRT radial basis function from the EIS; a Bayesian neural network model is built, and the loss function of the Bayesian neural network model for extracting DRT is obtained by combining the radial basis function; the weight vector of the radial basis function is obtained by training the model, and the DRT of the lithium battery is calculated based on the radial basis function. This method can effectively overcome the interference of EIS data noise and realize the accurate extraction of DRT. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0049] Figure 1 This is a schematic diagram of a method for determining the relaxation time distribution of a lithium battery provided in an embodiment of the present application. Figure 1 ;

[0050] Figure 2 This is a schematic diagram of a method for determining the relaxation time distribution of a lithium battery provided in an embodiment of the present application. Figure 2 ;

[0051] Figure 3 It is a structural schematic diagram of a device for determining relaxation time distribution of a lithium battery provided in an embodiment of the present application;

[0052] Figure 4 It is a schematic block diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution 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 part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0054] With the development of online EIS technology, the application scenarios of EIS will no longer be limited to offline testing scenarios such as laboratories. However, due to problems such as insufficient equipment accuracy and complex scene conditions, noise in online EIS data is common. Extracting DRT from EIS data is an ill-posed optimization problem and is very sensitive to noise. Even small noise may lead to incorrect DRT results. Currently, there is still no more accurate solution to extract DRT from noisy EIS data.

[0055] In view of this, a method for extracting the DRT of a lithium battery based on a Bayesian Neural Network (BNN) is proposed in an embodiment of the present application. The radial basis function is used to discretize and interpolate the continuous relaxation time distribution, and an optimization problem of solving the weight vector of the DRT radial basis function from the EIS is constructed; a Bayesian neural network model is built, and the variational inference method is combined with the radial basis function to obtain the loss function of the Bayesian neural network model to extract the DRT; the weight vector of the radial basis function is obtained by training the model, and the DRT of the lithium battery is calculated based on the radial basis function. This method can effectively overcome the interference of EIS data noise and realize the accurate extraction of DRT. Combined with Figure 1 Provide explanation.

[0056] S101, pre-processing the EIS data of the lithium battery to remove high-frequency impedance frequency points.

[0057] The high-frequency characteristics of EIS can be ignored when modeling and analyzing batteries, so the EIS data is preprocessed to remove high-frequency frequency data.

[0058] Optionally, the pre-processing process may include:

[0059] The frequency points with positive imaginary parts in the EIS data were located using the threshold judgment method, and the EIS frequency point data located by the threshold judgment were deleted.

[0060] Assume [Zre k Zim k ], k=1,2...N is the EIS data vector, where N is the total number of EIS data frequency points, k is the position of the frequency point in the vector, Zre k , Zim k The real and imaginary parts of the corresponding frequency points, and the imaginary part Zim of each frequency point k Compare with the threshold 0 and record the frequency point position k where the imaginary part is greater than 0 sign . After that, delete the recorded position k sign The frequency points of EIS are reordered to obtain the preprocessed EIS data [Zre k Zim k], k = 1, 2…n, where n is the number of EIS frequency points after preprocessing. By preprocessing the original EIS data, the preprocessed EIS data is obtained, which is convenient for the subsequent DRT extraction.

[0061] S102, using radial basis functions to perform discretization interpolation processing on the continuous DRT, and constructing an optimization problem for solving the weight vector of the DRT radial basis function from the EIS.

[0062] First, the relationship expression between the impedance value of EIS and DRT is determined.

[0063] In DRT analysis, the impedance of the battery can be regarded as a series connection of resistance R0 and an infinite number of polarization processes, which can be expressed as follows:

[0064]

[0065] In formula (1), Z(f) is the impedance value of EIS, R0 is the ohmic resistance, τ is the time constant, and g(lnτ) is the distribution density function of polarization resistance in the logarithmic time constant domain, namely DRT.

[0066] Afterwards, considering the limited number of EIS frequency points, the radial basis function is used to perform discretization interpolation processing on formula (1) to obtain the relationship expression between the EIS impedance value and the weight vector of the radial basis function.

[0067] Select radial basis function to discretize formula (1), and we can get:

[0068]

[0069] In formula (2), is a radial basis function with shape parameter u, and the weight x of the radial basis function m is the discrete point of the function g(lnτ), τ m is the central time scale parameter of the radial basis function.

[0070] Further writing formula (2) into matrix form, we can get:

[0071]

[0072] In formula (3), the real coefficient matrix A' nm and the imaginary coefficient matrix A' n ' m The definition is as follows:

[0073]

[0074]

[0075] In formulas (4) and (5), f n is the frequency of the actual EIS frequency, f m is the frequency of the interpolation frequency point, b = ln(τ-τ m ).

[0076] Finally, based on formula (3), the error function between the impedance value of EIS and the impedance estimation value is further constructed to form an optimization problem for solving DRT from EIS.

[0077] Based on formula (3), the error function between the EIS impedance value and the impedance estimation value is constructed:

[0078]

[0079] In formula (6), v is the estimated value of the ohmic resistance R0, Z exp , A', A" and x ridge is Z(f), A' in formula (3) nm , A″ nm and x m The vector matrix representation of .

[0080] At this point, the problem of solving DRT from EIS becomes an optimization problem of minimizing the error E. The radial basis function is used to discretize the continuous DRT, so that the DRT frequency range obtained based on the finite EIS frequency points is wider and the result is more accurate.

[0081] S103, building a BNN according to the number of frequency points of the preprocessed EIS data, using variational inference combined with radial basis function to derive the loss function of BNN for extracting DRT.

[0082] First, the BNN structure is determined, and the number of nodes in each layer is determined according to the frequency point N of the preprocessed EIS data.

[0083] Optionally, the BNN network structure includes four layers, including an input layer, two hidden layers and an output layer. The input of the BNN is a randomly initialized scalar x, and the output is a vector y=[x ridge v]. Optionally, the activation function of the first three layers of BNN is ReLU function, and the activation function of the last layer is Sigmoid function. The number of neurons in each node is determined according to the number of frequency points N of EIS. The specific structural parameters are shown in the table below.

[0084] Table 1

[0085]

[0086] Afterwards, the Bayesian rule is used to derive the expression of the posterior probability of the neuron parameters. In order to solve the problem that the posterior probability is difficult to obtain, variational inference is used to define a new distribution to approximate the required posterior probability.

[0087] Assuming that the parameter of BNN is w, for a given sample data (x, y), the training process of BNN is equivalent to using Bayes' rule to infer the posterior probability of neuron parameters, as shown in the following formula:

[0088]

[0089] In formula (7), p(y|x,w) is the likelihood estimate of the observed value, which follows the Gaussian distribution N(y pred ,δ l 2 I), y pred is the output of BNN, p(w) is the prior estimate of the neural network parameters, and it follows the Gaussian distribution N(0,δ r 2 I).

[0090] However, the integral ∫p(y|x,w)p(w)dw is difficult to calculate and solve, which makes p(w|x,y) unable to be solved by formula (7). Therefore, a Gaussian distribution q(w|θ) is defined, and the KL divergence between the distribution q(w|θ) and p(w|x,y) is minimized to make the defined Gaussian distribution q(w|θ) close to the posterior distribution p(w|x,y) that needs to be solved.

[0091] At this point, the problem of solving p(w|x,y) becomes the problem of finding the most suitable θ to minimize the value of KL(q(w|θ)||p(w|x,y)). The calculation expression of KL(q(w|θ)||p(w|x,y)) is as follows:

[0092]

[0093] Finally, for the given sample data, the Monte Carlo method is used and combined with the error function (6) to derive the loss function of BNN.

[0094] For a given sample data (x, y), using the Monte Carlo rule on formula (8), we can get:

[0095]

[0096] In formula (9), m′ is the number of sampling points. Taking the calculation of neuron weights as an example, assuming that BNN has n′ weights w i＝1,2,…,n′ , the Gaussian distribution parameter of the i-th weight is θ i ={μ i ,δ i}, and assuming that the distributions of different weights are independent of each other, then:

[0097]

[0098] Substituting formula (10) into formula (9), we can get the loss function of BNN:

[0099]

[0100] Combined with the error function (6), in formula (11), logq(w i |θ i )、logp(w i ), the calculation expression of logp(y|w,x) is as follows:

[0101]

[0102]

[0103]

[0104] The variational inference method is used to solve the posterior probability of the parameters, which overcomes the problem that the posterior probability of BNN parameters is difficult to obtain.

[0105] S104 , training the BNN based on the EIS to obtain a weight vector of the radial basis function, and calculating the DRT based on the weight vector of the radial basis function.

[0106] Based on the impedance value of EIS, namely Z exp , train the BNN, complete the training when the loss function meets the preset conditions, and after the training is completed, the BNN outputs the weight vector x of the radial basis function ridge Based on formula (2) and formula (3), the inverse discretization interpolation calculation can be performed to obtain g(lnτ), that is, DRT.

[0107] Among them, during the training process, for the BNN in the process of back propagation parameter training, it is possible that the parameter variance δ i The problem of being less than 0 leads to an error in the calculation of the log function. In the embodiment of the present application, δ i Special processing is performed as follows:

[0108]

[0109] Among them, ρ i Set it as required.

[0110] In addition, since the parameter expression of BNN contains log terms, the gradient may reach a maximum value during the training process, causing a gradient explosion. Therefore, after each training, the gradient is clipped to constrain the gradient value within a small range.

[0111] In combination with the Bayesian neural network and the loss function construction process introduced above, an embodiment of the present application provides a method for determining the relaxation time distribution of a lithium battery.

[0112] Figure 2 1 is a flow chart of a method for determining the relaxation time distribution of a lithium battery provided in an embodiment of the present application. Figure 2 As shown, the method includes:

[0113] S201. Obtain an initial Bayesian neural network model. The input of the initial Bayesian neural network model is a randomly initialized scalar. The output is a vector composed of a weight vector of a radial basis function and an impedance estimate of the lithium battery. The radial basis function is used to discretize a continuous relaxation time distribution. The weight of the radial basis function is a discrete point of the relaxation time distribution. The loss function of the initial Bayesian neural network model is an error function between the impedance value of the electrochemical impedance spectrum of the lithium battery and the impedance estimate.

[0114] Optionally, the initial Bayesian neural network model can refer to the structure in the aforementioned embodiment, that is, the initial Bayesian neural network model includes four layers, including an input layer, two hidden layers and an output layer, the activation function of the first three layers is a ReLU function, and the activation function of the last layer is a Sigmoid function. The number of neurons in each node is determined according to the number of frequency points N of the EIS, and the specific details can be referred to the aforementioned Table 1.

[0115] Optionally, for the initial Bayesian neural network model, the Bayesian rule is used to determine the posterior probability distribution of neuron parameters, and variational inference is used to obtain the approximate distribution of the posterior probability distribution of neuron parameters; based on the KL divergence between the posterior probability distribution of neuron parameters and the approximate distribution, and the error function, the loss function of the initial Bayesian neural network model is determined.

[0116] Optionally, a first relationship between the impedance value of the electrochemical impedance spectrum and the relaxation time distribution is obtained; the first relationship is discretized using a radial basis function to obtain a second relationship between the impedance value of the electrochemical impedance spectrum and a weight vector of the radial basis function; and an error function between the impedance value of the electrochemical impedance spectrum and the impedance estimation value is determined based on the second relationship.

[0117] The approximate distribution can refer to the Gaussian distribution q(w|θ) in the aforementioned embodiment, the first relationship between the impedance value of the electrochemical impedance spectrum and the relaxation time distribution can refer to formula (1) in the aforementioned embodiment, the second relationship obtained after discretization processing can refer to formula (2) in the aforementioned embodiment, and the error function can refer to formula (6) in the aforementioned embodiment.

[0118] The KL divergence between the approximate distribution and the posterior probability distribution, as well as the corresponding loss function, can be parameterized in formulas (8) to (14) in the aforementioned embodiments.

[0119] S202, training the initial Bayesian neural network model based on the electrochemical impedance spectroscopy so that the loss value of the loss function meets the preset conditions, and obtaining a trained Bayesian neural network model.

[0120] Among them, the electrochemical impedance spectrum is preprocessed to remove the high-frequency impedance frequency points in the electrochemical impedance spectrum. For example, the imaginary part of the electrochemical impedance spectrum is threshold judged to remove the frequency point data with a positive imaginary part. After that, the randomly initialized scalar is input into the initial Bayesian neural network model to obtain the output of the model. Based on the impedance value of the electrochemical impedance spectrum and the aforementioned loss function, the parameters of the model are adjusted. After multiple trainings until the loss value of the loss function meets the preset conditions, the training is determined to be completed.

[0121] During the training process, optionally, the variance of the approximate distribution is limited to a value greater than zero to avoid errors in the log function calculation. Optionally, in order to prevent gradient explosion, the gradient after each training is gradient clipped to limit the gradient value to a preset range.

[0122] S203, determining the relaxation time distribution of the lithium battery based on the weight vector of the radial basis function output by the trained Bayesian neural network model.

[0123] Based on the weight vector of the radial basis function output by the trained Bayesian neural network model, the first relationship and the second relationship, the relaxation time distribution of the lithium battery is determined, that is, based on the weight vector of the radial basis function, an inverse discretization interpolation calculation is performed to obtain the relaxation time distribution of the lithium battery.

[0124] The relaxation time distribution determination method of the lithium battery provided in the embodiment of the present application, by pre-processing the electrochemical impedance spectrum data, remove the high-frequency impedance frequency point;For the characteristics of the limited number of electrochemical impedance frequency points, the continuous relaxation time distribution is discretized and interpolated by radial basis function, and the optimization problem of solving the relaxation time distribution radial basis function weight vector from the electrochemical impedance spectrum is constructed;According to the number of frequency points of the electrochemical impedance spectrum to be processed, a Bayesian neural network is built, and variational inference is used, combined with radial basis function, to derive the loss function of the Bayesian neural network to extract the relaxation time distribution;Finally, the weight vector of the radial basis function is obtained by training using the Monte Carlo method, and the relaxation time distribution of the lithium battery is calculated based on the weight vector of the radial basis function. The method can accurately obtain the relaxation time distribution of the lithium battery from the electrochemical impedance spectrum containing noise, solves the problem that it is difficult to accurately obtain the relaxation time distribution from the electrochemical impedance spectrum containing noise in the electrochemical characteristic analysis of the lithium battery, and the relaxation time distribution extraction architecture based on the Bayesian neural network can effectively overcome the interference of electrochemical impedance spectrum data noise, and realize the accurate extraction of relaxation time distribution.

[0125] Figure 3 Schematic diagram of a relaxation time distribution determination device for a lithium battery provided in an embodiment of the present application. Figure 3 As shown, the relaxation time distribution determination device 300 of the lithium battery includes:

[0126] An acquisition module 301 is used to acquire an initial Bayesian neural network model, wherein the input of the initial Bayesian neural network model is a randomly initialized scalar, and the output is a vector composed of a weight vector of a radial basis function and an impedance estimate of the lithium battery. The radial basis function is used to discretize a continuous relaxation time distribution, and the weight of the radial basis function is a discrete point of the relaxation time distribution. The loss function of the initial Bayesian neural network model is an error function between an impedance value of an electrochemical impedance spectrum of the lithium battery and an impedance estimate.

[0127] A training module 302 is used to train the initial Bayesian neural network model based on the electrochemical impedance spectroscopy so that the loss value of the loss function meets a preset condition to obtain a trained Bayesian neural network model;

[0128] The determination module 303 is used to determine the relaxation time distribution of the lithium battery based on the weight vector of the radial basis function output by the trained Bayesian neural network model.

[0129] In one implementation, the acquisition module 301 is further configured to:

[0130] Obtaining a first relationship between an impedance value of an electrochemical impedance spectroscopy and a relaxation time distribution;

[0131] The first relational expression is discretized by using a radial basis function to obtain a second relational expression between the impedance value of the electrochemical impedance spectroscopy and the weight vector of the radial basis function;

[0132] An error function between the impedance value of the electrochemical impedance spectroscopy and the impedance estimation value is determined based on the second relationship.

[0133] In one implementation, the determination module 303 is used to:

[0134] The relaxation time distribution of the lithium battery is determined based on the weight vector of the radial basis function output by the trained Bayesian neural network model, the first relationship and the second relationship.

[0135] In one implementation, the acquisition module 301 is further configured to:

[0136] For the initial Bayesian neural network model, the Bayesian rule is used to determine the posterior probability distribution of neuron parameters, and variational inference is used to obtain the approximate distribution of the posterior probability distribution of neuron parameters;

[0137] The loss function of the initial Bayesian neural network model is determined based on the KL divergence between the posterior probability distribution of the neuron parameters and the approximate distribution, as well as the error function.

[0138] In one implementation, the training module 302 is used to: during the training process, limit the variance of the approximate distribution to a value greater than zero.

[0139] In one implementation, the training module 302 is used to: during the training process, perform gradient clipping on the gradient after each training to limit the gradient value to a preset range.

[0140] In one implementation, the acquisition module 301 is further configured to:

[0141] The electrochemical impedance spectrum is preprocessed to remove high-frequency impedance frequency points in the electrochemical impedance spectrum.

[0142] In one implementation, the acquisition module 301 is used to:

[0143] The imaginary part of the electrochemical impedance spectrum is subjected to threshold judgment, and the frequency points where the imaginary part is a positive number are removed.

[0144] The device of the embodiment of the present application is used to implement the method in the aforementioned embodiment, and its implementation principle and technical effect are similar, which will not be repeated here.

[0145] Figure 4 is a schematic block diagram of an electronic device provided in an embodiment of the present application. Figure 4 As shown, the electronic device 400 may include at least one processor 401 for implementing the method provided in the embodiment of the present application.

[0146] Optionally, the electronic device 400 also includes at least one memory 402 for storing program instructions and / or data. The memory 402 is coupled to the processor 401. The coupling in the embodiment of the present application is an indirect coupling or communication connection between devices, units or modules, which can be an electrical coupling, a mechanical coupling or other form, for information exchange between devices, units or modules. The processor 401 may operate in conjunction with the memory 402. The processor 401 may execute program instructions stored in the memory 402. At least one of the at least one memory may be included in the processor.

[0147] Optionally, the electronic device 400 also includes a communication interface 403 for communicating with other devices via a transmission medium, so that the electronic device 400 can communicate with other devices. The communication interface 403 can be, for example, a transceiver, an interface, a bus, a circuit, or a device capable of implementing a transceiver function. The processor 401 can use the communication interface 403 to send and receive data and / or information, and is used to implement the method provided in the embodiment of the present application. Please refer to the detailed description in the previous embodiment for details, which will not be repeated here.

[0148] The specific connection medium between the processor 401, the memory 402 and the communication interface 403 is not limited in the embodiment of the present application. Figure 4 In the embodiment, the processor 401, the memory 402 and the communication interface 403 are connected via a bus 404. The bus 404 is Figure 4 The connection between other components is only for illustration and is not intended to be limiting. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 4 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0149] It should be understood that the processor in the embodiment of the present application can be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method embodiment can be completed by the hardware integrated logic circuit or software instructions in the processor. The above processor can be a general processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in the embodiment of the present application can be directly embodied as a hardware decoding processor to perform, or the hardware and software modules in the decoding processor can be combined to perform. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0150] It should also be understood that the memory in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct RAM (DR RAM). It should be noted that the memory of the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0151] The present application also provides a computer-readable storage medium, which stores a computer program (also referred to as code or instruction). When the computer program is executed, the computer executes the method in any of the above embodiments.

[0152] The terms "unit", "module" and the like used in this specification may be used to represent a computer-related entity, hardware, firmware, a combination of hardware and software, software, or software in execution.

[0153] It will be appreciated by those skilled in the art that the various illustrative logical blocks and steps described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or in combination with computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application. In the several embodiments provided in this application, it should be understood that the disclosed devices, equipment and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the unit is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical coupling, mechanical coupling or other forms.

[0154] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0155] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0156] In the above-described embodiments, the functions of each functional unit can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions (programs). When the computer program instructions (programs) are loaded and executed on a computer, the flow or function according to the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable devices. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media integrated therein. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a magnetic tape), an optical medium (e.g., a digital video disc (DVD)), or a semiconductor medium (e.g., a solid state drive (SSD)), etc.

[0157] If this function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard drives, ROM, RAM, magnetic disks, or optical disks.

[0158] The user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0159] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for determining relaxation time distribution of a lithium battery, characterized in that: include: Obtaining an initial Bayesian neural network model, wherein the input of the initial Bayesian neural network model is a randomly initialized scalar, and the output is a vector composed of a weight vector of a radial basis function and an impedance estimate of the lithium battery, wherein the radial basis function is used to discretize a continuous relaxation time distribution, and the weight of the radial basis function is a discrete point of the relaxation time distribution, and the loss function of the initial Bayesian neural network model is based on an error function between an impedance value of an electrochemical impedance spectrum of the lithium battery and the impedance estimate; Training the initial Bayesian neural network model based on the electrochemical impedance spectroscopy so that the loss value of the loss function meets a preset condition to obtain a trained Bayesian neural network model; The relaxation time distribution of the lithium battery is determined based on the weight vector of the radial basis function output by the trained Bayesian neural network model.

2. The method according to claim 1, characterized in that Also includes: Obtaining a first relationship between an impedance value of an electrochemical impedance spectroscopy and a relaxation time distribution; The first relationship is discretized by using a radial basis function to obtain a second relationship between the impedance value of the electrochemical impedance spectroscopy and the weight vector of the radial basis function; An error function between the impedance value of the electrochemical impedance spectroscopy and the impedance estimation value is determined based on the second relationship.

3. The method according to claim 2, characterized in that The weight vector of the radial basis function outputted by the trained Bayesian neural network model is used to determine the relaxation time distribution of the lithium battery, including: The relaxation time distribution of the lithium battery is determined based on the weight vector of the radial basis function output by the trained Bayesian neural network model, the first relationship and the second relationship.

4. The method according to claim 2, characterized in that: Also includes: For the initial Bayesian neural network model, the Bayesian rule is used to determine the posterior probability distribution of neuron parameters, and variational inference is used to obtain the approximate distribution of the posterior probability distribution of the neuron parameters; Based on the KL divergence between the posterior probability distribution of the neuron parameters and the approximate distribution, and the error function, a loss function of the initial Bayesian neural network model is determined.

5. The method according to claim 4, characterized in that Also includes: During the training process, the variance of the approximate distribution is limited to a value greater than zero.

6. The method according to claim 4, characterized in that Also includes: During the training process, the gradient is clipped after each training to limit the gradient value to a preset range.

7. The method according to claim 1, characterized in that Also includes: The electrochemical impedance spectrum is preprocessed to remove high-frequency impedance frequency points in the electrochemical impedance spectrum.

8. The method according to claim 7, characterized in that The preprocessing of the electrochemical impedance spectrum to remove high-frequency impedance frequency points in the electrochemical impedance spectrum includes: A threshold value is determined for the imaginary part of the electrochemical impedance spectrum, and frequency points where the imaginary part is a positive number are removed.

9. A device for determining relaxation time distribution of a lithium battery, characterized in that: include: An acquisition module is used to acquire an initial Bayesian neural network model, wherein the input of the initial Bayesian neural network model is a randomly initialized scalar, and the output is a vector composed of a weight vector of a radial basis function and an impedance estimate of a lithium battery, wherein the radial basis function is used to discretize a continuous relaxation time distribution, and the weight of the radial basis function is a discrete point of the relaxation time distribution, and the loss function of the initial Bayesian neural network model is based on an error function between an impedance value of an electrochemical impedance spectrum of the lithium battery and the impedance estimate; A training module, used for training the initial Bayesian neural network model based on the electrochemical impedance spectrum so that the loss value of the loss function satisfies a preset condition to obtain a trained Bayesian neural network model; The determination module is used to determine the relaxation time distribution of the lithium battery based on the weight vector of the radial basis function output by the trained Bayesian neural network model.

10. An electronic device, characterized in that: include: Memory and processor; The memory is used to store computer programs; The processor is configured to execute a computer program stored in the memory, and when the computer program is executed, the processor is enabled to execute the method according to any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor executes the method according to any one of claims 1 to 8.

Images