Equivalent circuit model parameter identification method, device, equipment and program product

Abstract

The invention relates to the technical field of battery models, and discloses an equivalent circuit model parameter identification method, device, equipment and program product, which can reflect the influence of model parameters on terminal voltage more accurately by randomly selecting any number of model parameter value combinations and constructing a corresponding first observation point set. Further, the observation point set can be continuously focused in an area which is more likely to contain a global optimal solution along with an iteration process, posteriori distribution is estimated based on the first observation point set, and then the first observation point set is updated according to the posteriori distribution and an acquisition function; therefore, the observation point set can be continuously focused in a region which is more likely to contain the globally optimal solution along with the iteration process, the phenomenon that the observation point set is caught in the locally optimal solution too early is avoided, and the probability of finding the globally optimal solution is greatly increased.

Description

Technical Field

[0001] The present invention relates to the technical field of battery models, and particularly relates to a method, device, equipment and program product for identifying parameters of an equivalent circuit model. Background Art

[0002] Parameter identification of an equivalent circuit model (ECM) refers to the process of estimating and optimizing unknown parameters in the ECM through experimental data to improve the prediction accuracy and reliability of the model. This process is crucial for establishing an accurate battery model, especially in the application of lithium-ion batteries. Common methods include: the nonlinear least squares method; genetic algorithms, particle swarm optimization algorithms, simulated annealing algorithms, etc. that search for the global optimal solution based on heuristic rules; the gradient descent method of iterative optimization.

[0003] However, the disadvantages of heuristic algorithms are: they will fall into local optimal solutions in some specific cases and it is difficult to jump out, and finally it is difficult to find the optimal or even approximate optimal solution; some algorithms are prone to premature convergence (prematurely converging) and miss potential global optimal solutions; some algorithms have a prior assumption that the optimization target variable is a continuous variable; when exploring the solution space during the search process, situations violating constraints may be encountered. For example, in a multi-variable optimization problem, it is necessary to control the value ranges of two variables with an order-of-magnitude difference; the limitation of the gradient descent method is that it is necessary to assume that the optimization target is continuous and differentiable, and it is easy to fall into local optimal solutions.

[0004] Therefore, the existing model parameter solving methods are prone to the limitation of local optimal solutions. Summary of the Invention

[0005] In view of this, the present invention provides a method, device, equipment and program product for identifying parameters of an equivalent circuit model to solve the problem that the existing model parameter solving methods are prone to local optimal solutions.

[0006] In a first aspect, the present invention provides a method for identifying parameters of an equivalent circuit model, and the method includes:

[0007] Obtain an initial model parameter sampling point set and a first observation point set for the equivalent circuit model. The first observation point set is composed of any number of model parameter value combinations in the initial model parameter sampling point set and the fitting error value of the terminal voltage corresponding to each group of model parameter value combinations. When the target terminal voltage fitting error value is greater than or equal to the preset terminal voltage fitting error threshold, based on the first observation point set, use the preset probability surrogate model to estimate the posterior distribution between the model parameters of the equivalent circuit model and the terminal voltage fitting error value, and obtain the initial posterior distribution. Among them, the target terminal voltage fitting error value is the smallest terminal voltage fitting error value in the first observation point set. Predict the terminal voltage fitting error of each parameter sampling point in the first parameter sampling point set using the initial posterior distribution, and obtain the predicted terminal voltage fitting error value of each parameter sampling point. The first parameter sampling point set is the other parameter sampling point set in the initial model parameter sampling point set except the first observation point set. Update the first observation point set using the initial posterior distribution, multiple predicted terminal voltage fitting error values, and the acquisition function to obtain the second observation point set. Based on the second observation point set, return to the step of estimating the initial posterior distribution, and iterate repeatedly until the obtained target terminal voltage fitting error value is less than the preset terminal voltage fitting error threshold, and determine the model parameter combination corresponding to the target terminal voltage fitting error value as the global optimal parameter of the equivalent circuit model.

[0008] The equivalent circuit model parameter identification method provided by the present invention randomly selects any number of model parameter value combinations from the initial model parameter sampling point set of the equivalent circuit model to form a first observation point set, which can screen out some data from a large number of potential model parameter combinations for subsequent calculations, reducing the computational amount. Further, the posterior distribution estimation is carried out based on the first observation point set, considering the parameter uncertainty and the complex relationship between data. Then, by estimating the posterior distribution between the model parameters of the equivalent circuit model and the target terminal voltage fitting error, and selecting potential globally optimal parameters to be added to the observation points according to the posterior distribution. Instead of relying solely on local information or a fixed search strategy, when searching for the optimal parameters, all possible situations can be comprehensively considered, avoiding premature convergence to local optimal solutions, and greatly increasing the probability of finding the global optimal solution. At the same time, using the preset probability surrogate model for posterior distribution estimation can better simulate the complex non-linear characteristics of the battery model and avoid direct calculation. Further, when the target terminal voltage fitting error of the first observation point set is greater than or equal to the preset terminal voltage fitting error threshold, the first observation point set is updated using the acquisition function, and parameter combinations with smaller possible errors are found from the estimated global information, so that the observation point set can continuously focus on the region more likely to contain the global optimal solution during the iterative process. At the same time, considering the preset terminal voltage fitting error threshold, the entire parameter identification process has a clear accuracy target and can stop the iteration in time when the acceptable accuracy requirement is met, avoiding excessive calculation, improving the overall computational efficiency while ensuring the model accuracy, and overcoming the problems of waste of computational resources or inability to effectively converge to appropriate parameters that some methods may have due to the lack of a reasonable termination mechanism. Further, by repeatedly iterating and estimating the posterior distribution based on the updated second observation point set, the information contained in the new observation point set can be continuously used to update and improve the probabilistic understanding of the parameter space, enabling the probability surrogate model to more accurately learn the relationship between the parameters and the terminal voltage fitting error, helping to accelerate the convergence speed of the entire iterative process, finding parameter combinations with smaller errors to be added to the observation point set, and then finding the optimal path in the complex parameter space faster to reach the global optimal solution, reducing the number of iterations and improving the efficiency of parameter identification. Finally, by repeatedly iterating and determining the model parameter combination corresponding to the target terminal voltage fitting error value less than the preset terminal voltage fitting error threshold in the observation point set as the global optimal parameters of the equivalent circuit model, it is possible to evaluate the parameter space from a global perspective at a very small computational cost (compared with exhaustive parameter substitution and model calculation), ensuring that the truly global optimal solution is found rather than a local optimal solution, thereby improving the fitting accuracy of the equivalent circuit model to the actual characteristics of the battery.

[0009] In an alternative embodiment, based on the first set of observation points, a preset probability surrogate model is used to estimate the posterior distribution between the model parameters of the equivalent circuit model and the fitting error value of the terminal voltage, and an initial posterior distribution is obtained, including:

[0010] When the target terminal voltage fitting error value is greater than or equal to the preset terminal voltage fitting error threshold and the preset probability surrogate model is a Gaussian process regression model, a kernel function is used to simulate the relationship between each set of model parameter value combinations in the first set of observation points and obtain the covariance matrix of the terminal voltage fitting error value; based on the covariance matrix and Bayes' theorem, the preset probability surrogate model is used to estimate the posterior distribution between the model parameters of the equivalent circuit model and the terminal voltage fitting error value, and an initial posterior distribution is obtained.

[0011] For the equivalent circuit model parameter identification method provided by the present invention, when the preset probability surrogate model is a Gaussian process regression model, the kernel function of the Gaussian process regression model can more accurately simulate the relationship between each set of model parameter value combinations in the first set of observation points and further determine the covariance matrix of the terminal voltage fitting error value. Further, based on the covariance matrix and Bayes' theorem, the preset probability surrogate model is used to estimate the posterior distribution between the model parameters of the equivalent circuit model and the model parameter identification optimization target, making full use of the advantages of Bayesian optimization, comprehensively considering the uncertainty of parameters and the internal relationship between data, and being more likely to find the global optimal posterior distribution situation, thereby increasing the probability of finding the global optimal model parameters, avoiding falling into local optimal solutions, and at the same time overcoming the limitations of the gradient descent method and other methods that rely on the continuity and differentiability of the objective function.

[0012] In an alternative embodiment, the first set of observation points is updated using the initial posterior distribution, multiple predicted terminal voltage fitting error values, and an acquisition function to obtain a second set of observation points, including:

[0013] The acquisition function value of each parameter sampling point in the first set of parameter sampling points is calculated using the initial posterior distribution, multiple predicted terminal voltage fitting error values, and the acquisition function; the third set of observation points is determined using the acquisition function value of each parameter sampling point; the second set of observation points is determined based on the third set of observation points and the first set of observation points.

[0014] For the equivalent circuit model parameter identification method provided by the present invention, by gradually screening and updating the set of observation points, it is possible to targetedly explore more potential parameter regions during the iteration process, avoiding wasting computing resources in some areas that have been explored or are obviously unlikely to have optimal solutions, improving the search efficiency, and also helping to quickly jump out of the trap of local optimal solutions.

[0015] In an alternative embodiment, the method further includes:

[0016] During the iteration process, obtain the error change amount of the target terminal voltage fitting error value, and compare the error change amount with a first preset change threshold to determine the global optimal parameters of the equivalent circuit model; or, during the iteration process, obtain the model parameter change amount in the set of observation points, and compare the model parameter change amount with a second preset change threshold to determine the global optimal parameters of the equivalent circuit model.

[0017] In the equivalent circuit model parameter identification method provided by the present invention, during the iteration process, by comparing the error change amount with the first preset change threshold or by comparing the model parameter change amount with the second preset change threshold, it can be determined whether the iteration converges, and further, by judging whether it converges, the global optimal parameters of the equivalent circuit model can be further determined, which further improves the reliability and controllability of the entire parameter identification method.

[0018] In an optional implementation manner, the method further includes:

[0019] During the iteration process, obtain the error change amount of the target terminal voltage fitting error value and the model parameter change amount in the set of observation points; compare the error change amount with the first preset change threshold, and compare the model parameter change amount with the second preset change threshold to determine the global optimal parameters of the equivalent circuit model.

[0020] In an optional implementation manner, the method further includes:

[0021] When the number of iterations is equal to the preset batch iteration threshold, and the target terminal voltage fitting error value is greater than or equal to the preset terminal voltage fitting error threshold, obtain the current target terminal voltage fitting error value; determine the model parameter combination corresponding to the current target terminal voltage fitting error value as the global optimal parameters of the equivalent circuit model.

[0022] In the equivalent circuit model parameter identification method provided by the present invention, by comparing the preset number of batch iterations with the preset batch iteration threshold to determine the global optimal model parameters, it further improves the reliability and controllability of the entire parameter identification method. Compared with those methods that only rely on a single error judgment or have no clear iteration number limit, by considering both the number of iterations and the error situation simultaneously, it avoids the problems of endless iteration or premature termination of iteration and missing a better solution due to some special situations, enabling the entire algorithm to run more robustly under different data and model conditions, ensuring that the parameter identification can be completed within a reasonable consumption of computing resources and finding the global optimal parameters that can meet the accuracy requirements as much as possible, and improving the applicability and robustness of the present invention in practical applications.

[0023] In a second aspect, the present invention provides an equivalent circuit model parameter identification device, and the device includes:

[0024] A first acquisition module is configured to acquire an initial model parameter sampling point set and a first observation point set of an equivalent circuit model. The first observation point set is composed of any number of model parameter value combinations in the initial model parameter sampling point set and the terminal voltage fitting error values corresponding to each group of model parameter value combinations. An estimation module is configured to, when the target terminal voltage fitting error value is greater than or equal to a preset terminal voltage fitting error threshold, estimate the posterior distribution between the model parameters and the terminal voltage fitting error value of the equivalent circuit model based on the first observation point set by using a preset probability surrogate model to obtain an initial posterior distribution, where the target terminal voltage fitting error value is the smallest terminal voltage fitting error value in the first observation point set. A prediction module is configured to predict the terminal voltage fitting error of each parameter sampling point in the first parameter sampling point set by using the initial posterior distribution to obtain the predicted terminal voltage fitting error value of each parameter sampling point. The first parameter sampling point set is the other parameter sampling point set in the initial model parameter sampling point set except the first observation point set. An update module is configured to update the first observation point set by using the initial posterior distribution, a plurality of predicted terminal voltage fitting error values, and an acquisition function to obtain a second observation point set. An iteration module is configured to, based on the second observation point set, return to the step of estimating the initial posterior distribution, and iterate repeatedly until the obtained target terminal voltage fitting error value is less than the preset terminal voltage fitting error threshold, and determine the model parameter combination corresponding to the target terminal voltage fitting error value as the global optimal parameter of the equivalent circuit model.

[0025] In a third aspect, the present invention provides a computer device, including: a memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to execute the equivalent circuit model parameter identification method according to the first aspect or any corresponding embodiment thereof.

[0026] In a fourth aspect, the present invention provides a computer-readable storage medium, on which computer instructions are stored. The computer instructions are used to cause a computer to execute the equivalent circuit model parameter identification method according to the first aspect or any corresponding embodiment thereof.

[0027] In a fifth aspect, the present invention provides a computer program product, including computer instructions, which are used to cause a computer to execute the equivalent circuit model parameter identification method according to the first aspect or any corresponding embodiment thereof. Description of the Drawings

[0028] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0029] Figure 1 is a schematic flowchart of a method for identifying equivalent circuit model parameters according to an embodiment of the present invention;

[0030] Figure 2 is a schematic flowchart of another method for identifying equivalent circuit model parameters according to an embodiment of the present invention;

[0031] Figure 3 is a schematic flowchart of yet another method for identifying equivalent circuit model parameters according to an embodiment of the present invention;

[0032] Figure 4 is a structural block diagram of an apparatus for identifying equivalent circuit model parameters according to an embodiment of the present invention;

[0033] Figure 5 is a schematic hardware structure diagram of a computer device according to an embodiment of the present invention. Specific Embodiments

[0034] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0035] The limitation of the gradient descent method is that it needs to assume that the optimization objective is continuous and differentiable, and it is easy to fall into a local optimum. Therefore, in practice, the least squares method with lower complexity and faster calculation is generally used to identify the parameters of the ECM. When there is a linear relationship between the multi-dimensional feature X and the target Y, the solution of the least squares method must be the global optimum value because its objective function min||Y - X·θ||2 is a continuous convex function, and the local optimum point where the derivative is 0 is the global optimum point. The optimal solution at this time is θ = (X T X) -1 X T Y (because X may not be a square matrix, it is necessary to multiply both sides by X T to convert it into a square matrix, and then the inverse matrix can be obtained). When X TWhen X is irreversible, there are infinitely many solutions to the target linear equation, all of which point to the global optimal solution of the continuous convex function. Although all these solutions are equivalent mathematically, in practical applications, different solutions may have different interpretability and practicality. For example, one solution may contain fewer non-zero parameters, which may be more advantageous in feature selection and model simplification; another solution may have a more reasonable range in the magnitude of the parameters, avoiding overly large parameter values, which may help maintain the stability of the model. Therefore, it is necessary to select a more reasonable solution according to certain strategies, specifically: Ridge regression, lasso regression, generalized inverse (pseudo-inverse), principal component regression (eliminating some linearly correlated features to make it invertible), etc.

[0036] When X and Y are not linearly related, the least squares method cannot be directly used because a linear equation cannot be formed. Common processing strategies: transform the non-linear problem and use the least squares method for the transformed data. For example, f = 3·x 2 , x and f have a non-linear power relationship, while x 2 and f have a linear relationship, or directly use the non-linear least squares method. Typical algorithms among them include the steepest descent method, Newton's method, etc. However, the disadvantages of these methods are similar to those of the gradient descent method, which requires assuming that the optimization objective function is continuously differentiable and is prone to falling into local optima. Moreover, the least squares method is not flexible enough in implementation. If it is necessary to add linkage constraints between the parameters to be optimized, for example, when parameter A takes values within a certain range, parameter B is a constant, and vice versa, B is a variable. At this time, two optimization objective expressions need to be constructed respectively according to these two situations, and the least squares method is used to solve them respectively. This undoubtedly greatly increases the redundancy and repeated calculation process in the implementation process, seriously wasting computing resources.

[0037] Furthermore, in the problem of ECM parameter identification, there is obviously a non-linear relationship between the RC parameters and the fitting error of the terminal voltage, and it cannot be transformed into a linear relationship. In summary, the existing methods for solving RC parameters all have certain limitations. A method that has a higher probability of finding the global optimal solution, follows constraints, does not rely on the assumption that the objective function is continuous and differentiable, and is robust to the initial value is needed to solve the global optimal RC parameters.

[0038] According to an embodiment of the present invention, an embodiment of an equivalent circuit model parameter identification method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0039] In this embodiment, an equivalent circuit model parameter identification method is provided, which can be used in electronic devices such as computers, mobile phones, and tablet computers.Figure 1 is a flowchart of a method for identifying parameters of an equivalent circuit model according to an embodiment of the present invention. As Figure 1 shown, the process includes the following steps:

[0040] Step S101, obtain an initial model parameter sampling point set and a first observation point set of the equivalent circuit model.

[0041] Among them, the equivalent circuit model (ECM) represents a mathematical model used to simulate and analyze the behavior of a battery. By simplifying the complex physical and chemical processes inside the battery, it can be represented as a combination of a series of circuit elements, facilitating the understanding and analysis of the dynamic characteristics of the battery. Among them, common equivalent circuit models include the first-order RC model, the second-order RC model, the third-order RC model, etc.

[0042] The first observation point set O is composed of any number of model parameter value combinations S in the initial model parameter sampling point set and the fitting error value of the terminal voltage corresponding to each group of model parameter value combinations.

[0043] Specifically, according to factors such as the characteristics of the battery and the research purpose, the type M of the equivalent circuit model can be determined. Specifically, it can be a relatively simple first-order RC model, a more complex second-order RC model, or other higher-order models.

[0044] Furthermore, according to the determined type M of the equivalent circuit model, determine the parameters to be identified of the equivalent circuit model. Taking the equivalent circuit model as a second-order RC model as an example, the RC parameters to be identified include R0, R1, R2, τ1, τ2.

[0045] Set the value range and sampling interval of the RC parameters of the equivalent circuit model. Among them, the value range can be comprehensively set by combining the actual characteristics of the battery, past research experience, and theoretical analysis, etc.; the sampling interval can determine the sampling density within the value range.

[0046] Taking the equivalent circuit model as a second-order RC model as an example above, the time constants τ1 and τ2 respectively represent the fast and slow polarization effects. The two value ranges should differ by an order of magnitude. For example, the value range of τ1 is [1, 10], and the value range of τ2 is [1, 100], thus reflecting the multi-time scale characteristics of the time constants. This not only reduces the parameter space (due to the difference in time scales, the optimal value of τ1 may not appear in [11, 90]), but also does not miss the precise optimal solution. The sampling interval can be set to 0.1. For example, τ1 as [0, 10, 0.1] represents that the possible value set of parameter τ1 is [0, 0.1, 0.2...9.8, 9.9, 10].

[0047] Further, uniform sampling is performed at a sampling interval within the value range of each RC parameter, and all possible combinations of RC parameters are combined to form an initial model parameter sampling point set S.

[0048] Finally, any number of multiple combinations of RC parameters are randomly selected from the initial model parameter sampling point set S. Exemplarily, in the initial model parameter sampling point set S, 0.5% or 1% of multiple combinations of RC parameters are randomly selected.

[0049] Further, the equivalent circuit model parameters are different under different working conditions. Therefore, the equivalent circuit model parameters are identified separately under different working conditions.

[0050] In this embodiment, the battery is charged and discharged under the HPPC working condition at different rates, SOCs, and temperatures, and the experimental process data is recorded and sorted out, including the current value I and voltage value V corresponding to each SOC when the battery is charged and discharged at different temperatures and different rates.

[0051] Further, the selected multiple combinations of RC parameters and the current value I under the current working condition are input into the equivalent circuit model, and the simulated terminal voltage value corresponding to each combination of RC parameters is output

[0052] Further, the simulated terminal voltage value is calculated and the terminal voltage fitting error value between the simulated terminal voltage value and the actual terminal voltage value V to obtain the first observation point set O. The first observation point set O includes the model parameter value combinations (i.e., RC parameter combinations) and the terminal voltage fitting errors corresponding to each group of model parameter value combinations (i.e., the terminal voltage fitting error values corresponding to each combination of RC parameters ).

[0053] Step S102, when the target terminal voltage fitting error value is greater than or equal to the preset terminal voltage fitting error threshold, based on the first observation point set, use the preset probability surrogate model to estimate the posterior distribution between the model parameters of the equivalent circuit model and the terminal voltage fitting error value, and obtain the initial posterior distribution.

[0054] Among them, the target terminal voltage fitting error value is the smallest terminal voltage fitting error value in the first observation point set; the preset probability surrogate model represents a general term for a class of models. This class of models can infer the posterior distribution of the optimization target of the surrogate model in the Bayesian framework, and can be a Gaussian process regression model, a beta - Bernoulli model, a Bayesian linear regression, a generalized linear model, a random forest, etc.

[0055] Specifically, according to the description of step S101, the first observation point set includes multiple data pairs (RC parameter combination, terminal voltage fitting error value).

[0056] When the target terminal voltage fitting error value is greater than or equal to the preset terminal voltage fitting error threshold, it indicates that the RC parameter combination corresponding to the current target terminal voltage fitting error is not ideal enough, the fitting error is still large, and further optimization is required.

[0057] Furthermore, using the preset probability surrogate model to estimate the posterior distribution between the model parameters of the equivalent circuit model and the terminal voltage fitting error value initiates the process of exploring the parameter space from a probability perspective. Different from the deterministic optimization method, it can take into account the parameter uncertainty and the complex relationship between data. Among them, different preset probability surrogate models adopt different estimation methods.

[0058] Furthermore, by estimating the posterior distribution between the RC parameter combination and the terminal voltage fitting error target, it is possible to quantitatively evaluate the possibility of each parameter combination in the entire parameter space, rather than relying solely on local information or a fixed search strategy. Thus, when searching for the optimal RC parameters, all possible situations can be comprehensively considered, avoiding premature convergence to local optimal solutions and greatly increasing the probability of finding the global optimal solution.

[0059] Furthermore, through the preset probability surrogate model, the estimation of the posterior distribution does not depend on the assumptions of the continuity and differentiability of the objective function (the function of the terminal voltage fitting error with respect to the RC parameters), can better adapt to the actual characteristics of the battery model, effectively handle various complex non-linear relationships, and improve the adaptability of the model to different battery types and operating conditions.

[0060] Furthermore, the estimated posterior distribution is equivalent to a probability surrogate of the model optimization objective, and its computational complexity is much smaller than the algorithm that directly simulates and calculates the terminal voltage fitting error value based on the model parameter values. Using it to search for the global optimal parameters will greatly reduce the computational complexity.

[0061] Step S103, use the initial posterior distribution to predict the terminal voltage fitting error of each parameter sampling point in the first parameter sampling point set, and obtain the predicted terminal voltage fitting error value of each parameter sampling point.

[0062] Among them, the first parameter sampling point set S - O is the set of other parameter sampling points in the initial model parameter sampling point set S except the first observation point set O.

[0063] Specifically, according to the above description of the first observation point set O, the terminal voltage fitting error values corresponding to each RC parameter combination in the first observation point set O are known.

[0064] Further, after estimating the posterior distribution of S-O based on O, the terminal voltage fitting error values of each combination of RC parameter values in the S-O set each correspond to a Gaussian distribution, and the mean of this Gaussian distribution is the value with the highest confidence in the entire distribution. Therefore, it is used as the estimated value of the predicted terminal voltage fitting error for this combination of RC parameter values.

[0065] Therefore, by using the initial posterior distribution to predict the terminal voltage fitting error of each parameter sampling point in the first parameter sampling point set, the predicted terminal voltage fitting error value of each parameter sampling point can be obtained.

[0066] Step S104, update the first observation point set by using the initial posterior distribution, multiple predicted terminal voltage fitting error values, and an acquisition function to obtain a second observation point set.

[0067] Among them, the acquisition function represents a function constructed based on the posterior distribution estimated by a probabilistic surrogate model, which is used to balance exploring new parameter regions and exploiting existing better parameter regions, and can guide how to find the next batch of observation points that may achieve the global optimal value according to the posterior distribution, so that in the next iteration batch, the probabilistic surrogate model can perform posterior estimation in the direction of the potential global optimum, and thus the global optimal value can be found faster with fewer iterations.

[0068] Further, the acquisition function can be expected improvement (EI), probability increment, confidence bound strategy, Thompson sampling, entropy search strategy, entropy prediction strategy, etc.

[0069] Further, the parameters of the acquisition function can be determined by maximum likelihood estimation (MLE).

[0070] Specifically, based on the known initial posterior distribution and multiple predicted terminal voltage fitting error values, the acquisition function can guide finding the next batch of observation points that may achieve the global optimal value according to the posterior distribution and updating the first observation point set to form a new second observation point set.

[0071] Step S105, based on the second observation point set, return to the step of estimating the initial posterior distribution, and iterate repeatedly until the obtained target terminal voltage fitting error value is less than the preset terminal voltage fitting error threshold, and determine the model parameter combination corresponding to the target terminal voltage fitting error value as the global optimal parameter of the equivalent circuit model.

[0072] Specifically, according to the obtained second set of observation points, update the first set of observation points to the second set of observation points, return to step S102, and repeat steps S102 to S104, that is, based on the second set of observation points, use the preset probability surrogate model to estimate the posterior distribution between the model parameters of the equivalent circuit model and the fitting error value of the terminal voltage to obtain the initial posterior distribution; then use the initial posterior distribution to predict the fitting error of the terminal voltage of each parameter sampling point in the first set of parameter sampling points to obtain the predicted fitting error value of the terminal voltage of each parameter sampling point. At this time, the first set of parameter sampling points is the set of other parameter sampling points in the initial set of model parameter sampling points except the second set of observation points; until the obtained target fitting error value of the terminal voltage is less than the preset fitting error threshold of the terminal voltage, it means that the RC parameter combination corresponding to the obtained target fitting error value of the terminal voltage at this time is the global optimal parameter of the equivalent circuit model. At this time, the target fitting error value of the terminal voltage is the smallest fitting error value of the terminal voltage in the corresponding set of observation points (i.e., the second set of observation points).

[0073] By continuously using the information contained in the new set of observation points to update and improve the probabilistic understanding of the parameter space, the probability surrogate model can more accurately learn the relationship between the parameters and the fitting error of the terminal voltage, which helps to accelerate the convergence speed of the entire iterative process, and then find the optimal path in the complex parameter space faster to reach the global optimal solution, reducing the number of iterations and improving the efficiency of parameter identification.

[0074] Finally, determining the model parameter combination corresponding to the target fitting error value of the terminal voltage that is less than the preset fitting error threshold of the terminal voltage as the global optimal parameter of the equivalent circuit model can evaluate the parameter space from a global perspective and improve the reliability of the global optimal solution.

[0075] The equivalent circuit model parameter identification method provided in this embodiment randomly selects any number of model parameter value combinations from the initial model parameter sampling point set of the equivalent circuit model to form a first observation point set, and can screen out some data from a large number of potential model parameter combinations for subsequent calculations, thereby reducing the amount of calculation. Furthermore, the posterior distribution estimation is performed based on the first observation point set, taking into account the uncertainty of the parameters and the complex relationship between the data, and then the posterior distribution between the model parameters of the equivalent circuit model and the terminal voltage fitting error target is estimated, and the potential global optimal parameters are selected according to the posterior distribution to be added to the observation points. Instead of relying solely on local information or fixed search strategies, all possible situations can be comprehensively considered when searching for the optimal parameters, avoiding falling into the local optimal solution too early, and greatly increasing the probability of finding the global optimal solution. At the same time, the use of a preset probability proxy model for posterior distribution estimation can better simulate the complex nonlinear characteristics of the battery model and avoid direct calculation. Further, when the target terminal voltage fitting error of the first observation point set is greater than or equal to the preset terminal voltage fitting error threshold, the first observation point set is updated using the acquisition function, and a parameter combination with a smaller error may be found from the estimated global information, so that the observation point set can continuously focus on the area that is more likely to contain the global optimal solution as the iteration process progresses. At the same time, considering the preset terminal voltage fitting error threshold, the entire parameter identification process has a clear accuracy target and can stop the iteration in time when the acceptable accuracy requirement is reached, avoiding over-calculation, while ensuring the accuracy of the model, while improving the overall calculation efficiency, overcoming the problem that some methods may waste computing resources or fail to effectively converge to suitable parameters due to the lack of a reasonable termination mechanism. Further, based on the updated second observation point set, the posterior distribution is repeatedly iterated and estimated, and the information contained in the new observation point set can be continuously used to update and improve the probabilistic understanding of the parameter space, so that the probabilistic proxy model can more accurately learn the relationship between the parameters and the terminal voltage fitting error, which helps to accelerate the convergence speed of the entire iterative process, find a parameter combination with a smaller error to join the observation point set, and then find the optimal path in the complex parameter space more quickly to achieve the global optimal solution, reduce the number of iterations, and improve the efficiency of parameter identification. Finally, through repeated iterations, the model parameter combination corresponding to the target terminal voltage fitting error value that is less than the preset terminal voltage fitting error threshold in the observation point set is determined as the global optimal parameter of the equivalent circuit model. With a very small computational cost (compared to exhaustive parameters and substituting them into model calculations), it is possible to evaluate the parameter space from a global perspective, ensuring that the true global optimal solution is found, rather than the local optimal solution, thereby improving the fitting accuracy of the equivalent circuit model to the actual characteristics of the battery.

[0076] In this embodiment, an equivalent circuit model parameter identification method is provided, which can be used in electronic devices such as computers, mobile phones, tablet computers, etc. Figure 2 It is a flowchart of the equivalent circuit model parameter identification method according to the embodiment of the present invention, as Figure 2 shown. The process includes the following steps:

[0077] Step S201, obtain an initial model parameter sampling point set and a first observation point set of the equivalent circuit model. For details, please refer to Figure 1 step S101 of the embodiment shown herein, which will not be elaborated herein.

[0078] Step S202, when the target terminal voltage fitting error value is greater than or equal to the preset terminal voltage fitting error threshold, based on the first observation point set, use the preset probability surrogate model to estimate the posterior distribution between the model parameters of the equivalent circuit model and the terminal voltage fitting error value, and obtain the initial posterior distribution.

[0079] Specifically, the above step S202 includes:

[0080] Step S2021, when the target terminal voltage fitting error value is greater than or equal to the preset terminal voltage fitting error threshold and the preset probability surrogate model is a Gaussian process regression model, use the kernel function to simulate the relationship between each group of model parameter value combinations in the first observation point set and obtain the covariance matrix of the terminal voltage fitting error value.

[0081] Among them, Gaussian process regression (GPR) is a non-parametric regression method. Using the Gaussian process regression model to describe the data distribution, the similarity between input variables can be measured by defining the kernel function, and the mean and variance of the predicted output can be calculated.

[0082] Specifically, the Gaussian process is a random process composed of an infinite number of random variables defined on a continuous domain (time / space), and each random variable follows a one-dimensional Gaussian distribution. Therefore, the Gaussian process is an infinite-dimensional Gaussian distribution, and the joint distribution of several random variables is still a multi-dimensional (element) Gaussian distribution. The random variables at each point are neither independent nor identically distributed (the same type, different parameters). The covariance between random variables is related to their positions. The closer two random variables are on the continuous domain, the stronger their correlation and the greater the absolute value of the covariance.

[0083] Among them, further, according to Mercer's theorem, the matrix composed of the kernel function values between variables is positive semi-definite.

[0084] Therefore, the kernel function can be used to simulate the covariance between random variables, and its value is related to the positions of the random variables and satisfies the characteristic that the covariance matrix is a positive semi-definite matrix. Among them, common kernel functions include Gaussian kernel, polynomial kernel, sigmoid kernel, etc., and generally the Gaussian kernel function is used.

[0085] Specifically, the combinations of model parameter values and the fitting errors of terminal voltages in the first observation point set O can be denoted as (X, Y). That is, X represents the combination of model parameter values, and Y represents the fitting error of the terminal voltage corresponding to X. Inputting all the (X, Y) values in the first observation point set O into the kernel function, the covariance matrix of X and Y can be obtained. The specific calculation method can be calculated using the existing kernel function formula (such as the Gaussian kernel function), which will not be elaborated in the embodiments of this application.

[0086] It should be noted that the kernel function is essentially a distance function defined based on the difference between two vectors. The smaller the vector difference, the larger the kernel function and the larger the covariance. For the convenience of explanation, X is assumed to be in the position of Y. If X is unfolded in one-dimensional time or space according to the sampling order within and between parameters, within the same parameter dimension of X, the smaller the value, the more forward the position, and the priority of the order between parameters is higher than the order within parameters. For example, there are two parameters P1 and P2, with value ranges of [1, 10] and [5, 20] respectively. Then in the unfolded one-dimensional space, the parameter (2, 10) is before (2, 15), and (1, 10) is before (2, 5).

[0087] Step S2022: Based on the covariance matrix and Bayes' theorem, use a preset probability surrogate model to estimate the posterior distribution between the model parameters and the fitting errors of terminal voltages of the equivalent circuit model, and obtain the initial posterior distribution.

[0088] Specifically, assume that Y1 is the set of current observation points (the fitting errors of terminal voltages corresponding to the combinations of RC parameter values) X1 is the position (the combination of RC parameter values) corresponding to each observation point Y2 and X2 respectively represent the distributions of n unknown random variables to be inferred and their corresponding positions.

[0089] Furthermore, the posterior distribution probability can be calculated based on Bayes' theorem.

[0090] Specifically, Bayes' theorem is shown as the following relational expression (1):

[0091]

[0092] Among them, P(Y2, Y1) represents the joint distribution probability of Y2 and Y1; P(Y1) represents the prior distribution probability of Y1; P(Y2|Y1) represents the probability that Y2 appears under the condition that Y1 is known.

[0093] Therefore, through the above relational expression (1), the unknown fitting error of the terminal voltage Y2 can be predicted using the known fitting error of the terminal voltage Y1.

[0094] Further, assuming that Y1 and Y2 follow a joint Gaussian distribution, their joint probability density function P(Y2, Y1) is shown in the following relational expression (2):

[0095]

[0096] where μ1 represents the mean vector of Y1; μ2 represents the mean vector of Y2; ∑ 11 represents the covariance matrix of Y1; ∑ 22 represents the covariance matrix of T2; ∑ 12 represents the covariance matrix between Y1 and Y2; ∑ 21 represents the covariance matrix between Y2 and Y1, where ∑ 11 / ∑ 22 / ∑ 12 and ∑ 21 are calculated in a similar manner, which will not be elaborated in the embodiments of the present application. Specifically, reference can be made to the calculation method of ∑ 11 .

[0097] Further, the prior distribution probability of Y1 is shown in the following relational expression (3):

[0098] P(Y1) ~ N(μ1, ∑ 11 ) (3)

[0099] Further, by using the method of completing the square to simplify the complex matrix operations (including operations such as block matrix inversion) on the above relational expression (1), the following relational expression (4) is obtained:

[0100]

[0101] Further, when inferring the posterior distribution of Y2, first assume that the mean vectors μ1 and μ2 in Y1 and Y2 are zero vectors, and the value of Y1 is the fitting error value of the terminal voltage of the observation point. Then, substituting this assumption condition into the above relational expression (4), the mean vector and covariance matrix of the posterior distribution can be obtained, thereby obtaining the initial posterior distribution between the model parameters of the equivalent circuit model and the fitting error value of the terminal voltage. Among them, the n unknown random variables represent the combination of RC parameter values in the initial model sampling point set S except for the first observation point set O, that is, S - O.

[0102] Further, if the preset probability surrogate model is other models, corresponding methods can be selected according to the specific type of the preset probability surrogate model for posterior distribution estimation.

[0103] Step S203, use the initial posterior distribution to predict the terminal voltage fitting error of each parameter sampling point in the first parameter sampling point set, and obtain the predicted terminal voltage fitting error value of each parameter sampling point. For details, please refer to Figure 1 Step S103 of the illustrated embodiment, which will not be elaborated here.

[0104] Step S204: Update the first set of observation points by using the initial posterior distribution, multiple predicted terminal voltage fitting error values, and an acquisition function to obtain a second set of observation points. For details, please refer to Figure 1 Step S104 of the embodiment shown, which will not be elaborated here.

[0105] Step S205: Based on the second set of observation points, return to the step of estimating the initial posterior distribution, and iterate repeatedly until the obtained target terminal voltage fitting error value is less than a preset terminal voltage fitting error threshold, and determine the model parameter combination corresponding to the target terminal voltage fitting error value as the global optimal parameters of the equivalent circuit model. For details, please refer to Figure 1 Step S105 of the embodiment shown, which will not be elaborated here.

[0106] For the equivalent circuit model parameter identification method provided in this embodiment, when the preset probability surrogate model is a Gaussian process regression model, the kernel function of the Gaussian process regression model can more accurately simulate the relationship between each set of model parameter value combinations in the first set of observation points and further determine the covariance matrix of the terminal voltage fitting error values. Further, based on the covariance matrix and Bayes' theorem, use the preset probability surrogate model to estimate the posterior distribution between the model parameters of the equivalent circuit model and the model parameter identification optimization objective, making full use of the advantages of Bayesian optimization, comprehensively considering the uncertainty of the parameters and the internal relationship between the data, being more likely to find the global optimal posterior distribution situation, thereby increasing the probability of finding the global optimal model parameters, avoiding falling into local optimal solutions, and also overcoming the limitations of the gradient descent method and other methods that rely on the continuity and differentiability of the objective function. Further, based on the posterior distribution for comprehensive prediction, making full use of the probability information about the relationship between parameters and errors accumulated in the Bayesian optimization process, provides data support for more scientifically and accurately finding the global optimal parameters that minimize the terminal voltage fitting error in the subsequent process.

[0107] In this embodiment, an equivalent circuit model parameter identification method is provided, which can be used in electronic devices such as computers, mobile phones, and tablet computers. Figure 3 is a flowchart of the equivalent circuit model parameter identification method according to an embodiment of the present invention, as Figure 3 shown, and this process includes the following steps:

[0108] Step S301: Obtain an initial model parameter sampling point set and a first set of observation points of the equivalent circuit model. For details, please refer to Figure 1 Step S101 of the embodiment shown, which will not be elaborated here.

[0109] Step S302, when the target terminal voltage fitting error value is greater than or equal to the preset terminal voltage fitting error threshold, based on the first observation point set, use the preset probability surrogate model to estimate the posterior distribution between the model parameters of the equivalent circuit model and the terminal voltage fitting error value, and obtain the initial posterior distribution. For details, please refer to Figure 2 Step S202 of the embodiment shown, which will not be elaborated here.

[0110] Step S303, use the initial posterior distribution to predict the terminal voltage fitting error of each parameter sampling point in the first parameter sampling point set, and obtain the predicted terminal voltage fitting error value of each parameter sampling point. For details, please refer to Figure 1 Step S103 of the embodiment shown, which will not be elaborated here.

[0111] Step S304, use the initial posterior distribution, multiple predicted terminal voltage fitting error values, and the acquisition function to update the first observation point set and obtain the second observation point set.

[0112] Specifically, the above Step S304 includes:

[0113] Step S3041, use the initial posterior distribution, multiple predicted terminal voltage fitting error values, and the acquisition function to calculate the acquisition function value of each parameter sampling point in the first parameter sampling point set.

[0114] Among them, the acquisition function value can reflect the potential gain or improvement that may be brought by selecting the target sampling point as the next observation point under the current knowledge of the initial posterior distribution.

[0115] Specifically, for each target sampling point, substitute its RC parameter combination into the acquisition function, and at the same time combine the information provided by the posterior distribution (such as mean, variance, etc.) to calculate the corresponding acquisition function value.

[0116] In an alternative embodiment, taking the expected improvement acquisition function as an example, the acquisition function value can be calculated by the following relational expression (5)

[0117]

[0118] In the formula: (x i, y i ) represents the data pair of each parameter sampling point in the first parameter sampling point set, that is, x i represents the RC parameter combination of the i-th parameter sampling point, and y i represents the predicted terminal voltage fitting error value corresponding to x i ; y * represents the fitting error threshold, which is a hyperparameter of Bayesian optimization and can be determined in the algorithm initialization stage; N(y i |xi ) represents the initial posterior distribution of y given x i under i the given x.

[0119] Step S3042: Determine the third set of observation points using the acquisition function values of each parameter sampling point.

[0120] Specifically, sort the acquisition function values of each parameter sampling point in descending order. The larger the acquisition function value, the more potential the target sampling point is considered to be an observation point that can further optimize the terminal voltage fitting error under the current posterior distribution.

[0121] Furthermore, select the parameter sampling points with the top preset number of larger acquisition function values from the sorted target sampling points as the new observation points.

[0122] Furthermore, calculate the terminal voltage fitting error values corresponding to the new observation points. The specific calculation process can refer to the description in step S101 above and will not be elaborated here.

[0123] Finally, combine the RC parameter combinations of the new observation points and the terminal voltage fitting error values of the new observation points to form the corresponding third set of observation points.

[0124] Step S3043: Determine the second set of observation points based on the third set of observation points and the first set of observation points.

[0125] Specifically, the obtained third set of observation points can be added to the first set of observation points to form a new second set of observation points.

[0126] Step S305: Based on the second set of observation points, return to the step of estimating the initial posterior distribution and iterate repeatedly until the obtained target terminal voltage fitting error value is less than the preset terminal voltage fitting error threshold, and determine the model parameter combination corresponding to the target terminal voltage fitting error value as the global optimal parameter of the equivalent circuit model. For details, please refer to Figure 1 step S105 of the illustrated embodiment and will not be elaborated here.

[0127] In an alternative embodiment, the above method further includes: during the iteration process, obtaining the error change amount of the target terminal voltage fitting error value, comparing the error change amount with the first preset change threshold, and determining the global optimal parameter of the equivalent circuit model;

[0128] and / or, during the iteration process, obtaining the model parameter change amount in the set of observation points, comparing the model parameter change amount with the second preset change threshold, and determining the global optimal parameter of the equivalent circuit model.

[0129] Among them, the error variation represents the absolute value of the difference between the target terminal voltage fitting error value obtained in the current iteration process and the target terminal voltage fitting error value obtained in the previous iteration process;

[0130] The model parameter variation represents the absolute value of the difference between the number of RC parameter combinations in the set of observation points obtained in the current iteration process and the number of RC parameter combinations in the set of observation points obtained in the previous iteration process.

[0131] Specifically, by comparing the error variation with a preset variation threshold, or by comparing the model parameter variation with a preset variation threshold, it can be determined whether the iteration process converges.

[0132] Furthermore, if it is determined that the iteration process has converged, the model parameter combination corresponding to the target terminal voltage fitting error value obtained in the current iteration process is taken as the global optimal parameter of the equivalent circuit model.

[0133] By judging whether to converge for termination determination and further determining the global optimal parameter of the equivalent circuit model, the reliability and controllability of the entire parameter identification method are further improved.

[0134] In an alternative embodiment, the above method further includes: when the number of iterations is equal to the preset batch iteration threshold, and the target terminal voltage fitting error value is greater than or equal to the preset terminal voltage fitting error threshold, obtaining the current target terminal voltage fitting error value; determining the model parameter combination corresponding to the current target terminal voltage fitting error value as the global optimal parameter of the equivalent circuit model.

[0135] Specifically, comparing the preset number of batch iterations with the preset batch iteration threshold.

[0136] Secondly, if the preset number of batch iterations has reached, that is, equal to the preset batch iteration threshold, it is judged whether the target terminal voltage fitting error value obtained at this time is less than the preset terminal voltage fitting error threshold. Further, if the target terminal voltage fitting error value obtained at this time is still greater than or equal to the preset terminal voltage fitting error threshold, then directly take the model parameter combination corresponding to the current target terminal voltage fitting error value as the global optimal parameter of the equivalent circuit model.

[0137] By comparing the preset number of batch iterations with the preset batch iteration threshold for termination determination and determining the global optimal model parameters, the reliability and controllability of the entire parameter identification method are further improved.

[0138] The equivalent circuit model parameter identification method provided by this embodiment enables targeted exploration of more potential parameter regions during the iterative process by gradually screening and updating the set of observation points. This avoids wasting computational resources in some regions that have already been explored or are obviously unlikely to have an optimal solution, improves the search efficiency, and also helps to quickly jump out of the trap of local optimal solutions. At the same time, by determining whether to converge or comparing the preset number of batch iterations and the preset batch iteration threshold for termination determination and further determining the global optimal parameters of the equivalent circuit model, the reliability and controllability of the entire parameter identification method are further improved. Compared with those methods that only rely on a single error judgment or have no clear iteration number limit, by considering both the number of iterations and the error situation simultaneously, the problem of endless iteration or premature termination of iteration and missing a better solution due to certain special circumstances is avoided, enabling the entire algorithm to run more robustly under different data and model conditions, ensuring that parameter identification can be completed within a reasonable computational resource consumption and finding the global optimal parameters that meet the accuracy requirements as much as possible, thus enhancing the applicability and robustness of the present invention in practical applications.

[0139] In this embodiment, an equivalent circuit model parameter identification device is also provided. This device is used to implement the above-mentioned embodiment and preferred implementation manners, and those that have been described will not be repeated. As used hereinafter, the term "module" can be a combination of software and / or hardware that can achieve a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0140] This embodiment provides an equivalent circuit model parameter identification device, as Figure 4 shown. The device includes:

[0141] A first acquisition module 401, configured to acquire an initial model parameter sampling point set and a first observation point set of the equivalent circuit model. The first observation point set is composed of any number of model parameter value combinations in the initial model parameter sampling point set and the terminal voltage fitting error values corresponding to each group of model parameter value combinations.

[0142] An estimation module 402, configured to, when the target terminal voltage fitting error value is greater than or equal to the preset terminal voltage fitting error threshold, estimate the posterior distribution between the model parameters of the equivalent circuit model and the terminal voltage fitting error value based on the first observation point set by using a preset probabilistic surrogate model to obtain an initial posterior distribution.

[0143] A prediction module 403 is configured to predict the terminal voltage fitting error of each parameter sampling point in the first parameter sampling point set by using the initial posterior distribution, so as to obtain the predicted terminal voltage fitting error value of each parameter sampling point. The first parameter sampling point set is the other parameter sampling point set in the initial model parameter sampling point set except the first observation point set.

[0144] An update module 404 is configured to update the first observation point set by using the initial posterior distribution, multiple predicted terminal voltage fitting error values, and an acquisition function, and obtain a second observation point set.

[0145] An iteration module 405 is configured to, based on the second observation point set, return to the step of estimating the initial posterior distribution, and iterate repeatedly until the obtained target terminal voltage fitting error value is less than a preset terminal voltage fitting error threshold, and determine that the model parameter combination corresponding to the target terminal voltage fitting error value is the global optimal parameter of the equivalent circuit model.

[0146] In an alternative embodiment, the estimation module 402 includes:

[0147] A simulation sub-module is configured to, when the target terminal voltage fitting error value is greater than or equal to the preset terminal voltage fitting error threshold and the preset probability surrogate model is a Gaussian process regression model, simulate the relationship between each set of model parameter value combinations in the first observation point set by using a kernel function, and obtain the covariance matrix of the terminal voltage fitting error values.

[0148] An estimation sub-module is configured to estimate the posterior distribution between the model parameters of the equivalent circuit model and the terminal voltage fitting error values by using the preset probability surrogate model based on the covariance matrix and Bayes' theorem, and obtain the initial posterior distribution.

[0149] In an alternative embodiment, the prediction module 403 includes:

[0150] A prediction sub-module is configured to predict the terminal voltage fitting error of each parameter sampling point in the first parameter sampling point set by using the initial posterior distribution, and obtain multiple initial terminal voltage fitting error values.

[0151] A first determination sub-module is configured to determine the target terminal voltage fitting error value by using the multiple initial terminal voltage fitting error values and the multiple terminal voltage fitting error values in the first observation point set. The target terminal voltage fitting error value is the smallest among the multiple initial terminal voltage fitting error values and the multiple terminal voltage fitting error values in the first observation point set.

[0152] In an alternative embodiment, the update module 404 includes:

[0153] A calculation sub-module, configured to calculate the acquisition function value of each parameter sampling point in the first parameter sampling point set by using the initial posterior distribution, multiple predicted terminal voltage fitting error values, and an acquisition function.

[0154] A second determination sub-module, configured to determine a third observation point set by using the acquisition function value of each parameter sampling point.

[0155] A third determination sub-module, configured to determine a second observation point set according to the third observation point set and the first observation point set.

[0156] In an alternative embodiment, the apparatus further includes:

[0157] A second acquisition module, configured to, during the iteration process, acquire the error change amount of the target terminal voltage fitting error value, compare the error change amount with a first preset change threshold, and determine the global optimal parameters of the equivalent circuit model; or, during the iteration process, acquire the model parameter change amount in the observation point set, compare the model parameter change amount with a second preset change threshold, and determine the global optimal parameters of the equivalent circuit model.

[0158] In an alternative embodiment, the apparatus further includes:

[0159] A third acquisition module, configured to, during the iteration process, acquire the error change amount of the target terminal voltage fitting error value and the model parameter change amount in the observation point set.

[0160] A comparison module, configured to compare the error change amount with the first preset change threshold and compare the model parameter change amount with the second preset change threshold to determine the global optimal parameters of the equivalent circuit model.

[0161] In an alternative embodiment, the apparatus further includes:

[0162] A third acquisition module, configured to, when the number of iterations is equal to a preset batch iteration threshold and the target terminal voltage fitting error value is greater than or equal to a preset terminal voltage fitting error threshold, acquire the current target terminal voltage fitting error value.

[0163] A determination module, configured to determine the model parameter combination corresponding to the current target terminal voltage fitting error value as the global optimal parameters of the equivalent circuit model.

[0164] The further function descriptions of the above-mentioned various modules are the same as those in the corresponding above-mentioned embodiments, and will not be elaborated here.

[0165] The equivalent circuit model parameter identification device in this embodiment is presented in the form of functional units. Here, the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and a memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0166] An embodiment of the present invention further provides a computer device having the above Figure 5 equivalent circuit model parameter identification device shown.

[0167] Please refer to Figure 5 , Figure 5 which is a schematic structural diagram of a computer device provided by an optional embodiment of the present invention. As shown in Figure 5 , the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including a high-speed interface and a low-speed interface. Each component communicates with each other using different buses and can be installed on a common motherboard or installed in other ways as needed. The processor can process instructions executed within the computer device, including instructions stored in the memory or on the memory to display graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a set of blade servers, or a multi-processor system). Figure 5 In

[0168] , a processor 10 is taken as an example.

[0169] The memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiment.

[0170] The memory 20 may include a program storage area and a data storage area. Among them, the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid-state storage devices. In some alternative embodiments, the memory 20 may optionally include a memory remotely disposed relative to the processor 10, and these remote memories may be connected to the computer device through a network. Examples of the above-mentioned network include but are not limited to the Internet, intranet, local area network, mobile communication network, and combinations thereof.

[0171] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid state drive; the memory 20 may further include a combination of the above types of memory.

[0172] The computer device further includes a communication interface 30 for the computer device to communicate with other devices or communication networks.

[0173] The embodiments of the present invention also provide a computer-readable storage medium. The methods according to the embodiments of the present invention may be implemented in hardware, firmware, or be implemented as computer code that can be recorded on a storage medium, or be implemented as computer code originally stored in a remote storage medium or a non-transitory machine-readable storage medium and to be downloaded through a network and stored in a local storage medium, so that the methods described herein can be stored in such software processes on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium may be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk or solid state drive, etc.; further, the storage medium may also include a combination of the above types of memory. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code, and when the software or computer code is accessed and executed by the computer, the processor, or the hardware, the methods shown in the above embodiments are implemented.

[0174] Although the embodiments of the present invention are described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for identifying parameters of an equivalent circuit model, characterized in that, The method includes: Obtaining an initial model parameter sampling point set of an equivalent circuit model and a first observation point set, where the first observation point set is composed of any number of model parameter value combinations in the initial model parameter sampling point set and the fitting error value of the terminal voltage corresponding to each group of model parameter value combinations; When the target terminal voltage fitting error value is greater than or equal to a preset terminal voltage fitting error threshold, based on the first observation point set, using a preset probability surrogate model to estimate the posterior distribution between the model parameters of the equivalent circuit model and the terminal voltage fitting error value, obtaining an initial posterior distribution, where the target terminal voltage fitting error value is the smallest terminal voltage fitting error value in the first observation point set; Predicting the terminal voltage fitting error of each parameter sampling point in the first parameter sampling point set by using the initial posterior distribution, obtaining the predicted terminal voltage fitting error value of each parameter sampling point, where the first parameter sampling point set is the other parameter sampling point set in the initial model parameter sampling point set except the first observation point set; Updating the first observation point set by using the initial posterior distribution, multiple predicted terminal voltage fitting error values and an acquisition function and obtaining a second observation point set; Based on the second observation point set, returning to the step of estimating the initial posterior distribution, iterating repeatedly until the obtained target terminal voltage fitting error value is less than the preset terminal voltage fitting error threshold, and determining the model parameter combination corresponding to the target terminal voltage fitting error value as the global optimal parameter of the equivalent circuit model.

2. The method according to claim 1, wherein When the target terminal voltage fitting error value is greater than or equal to a preset terminal voltage fitting error threshold, based on the first observation point set, using a preset probability surrogate model to estimate the posterior distribution between the model parameters of the equivalent circuit model and the terminal voltage fitting error value, obtaining an initial posterior distribution, including: When the target terminal voltage fitting error value is greater than or equal to a preset terminal voltage fitting error threshold and the preset probability surrogate model is a Gaussian process regression model, using a kernel function to simulate the relationship between each group of model parameter value combinations in the first observation point set and obtaining the covariance matrix of the terminal voltage fitting error value; Based on the covariance matrix and Bayes' theorem, using the preset probability surrogate model to estimate the posterior distribution between the model parameters of the equivalent circuit model and the terminal voltage fitting error value, obtaining the initial posterior distribution.

3. The method according to claim 1, wherein Updating the first observation point set by using the initial posterior distribution, the predicted terminal voltage fitting error value and the acquisition function and obtaining a second observation point set, including: Calculating the acquisition function value of each parameter sampling point in the first parameter sampling point set by using the initial posterior distribution, the predicted terminal voltage fitting error value and the acquisition function; Determining a third observation point set by using the acquisition function value of each parameter sampling point; Determining the second observation point set according to the third observation point set and the first observation point set.

4. The method according to claim 1, wherein The method further includes: During the iteration process, obtain the error change amount of the target terminal voltage fitting error value, and compare the error change amount with a first preset change threshold to determine the global optimal parameter of the equivalent circuit model; or, During the iteration process, obtain the model parameter change amount in the set of observation points, and compare the model parameter change amount with a second preset change threshold to determine the global optimal parameter of the equivalent circuit model.

5. The method according to claim 1, characterized in that, The method further includes: During the iteration process, obtain the error change amount of the target terminal voltage fitting error value and the model parameter change amount in the set of observation points; Compare the error change amount with a first preset change threshold, and compare the model parameter change amount with a second preset change threshold to determine the global optimal parameter of the equivalent circuit model.

6. The method according to claim 1, wherein The method further includes: When the number of iterations is equal to a preset batch iteration threshold and the target terminal voltage fitting error value is greater than or equal to the preset terminal voltage fitting error threshold, obtain the current target terminal voltage fitting error value; Determine the model parameter combination corresponding to the current target terminal voltage fitting error value as the global optimal parameter of the equivalent circuit model.

7. An equivalent circuit model parameter identification device, characterized in that, The device includes: A first acquisition module, configured to acquire an initial model parameter sampling point set of an equivalent circuit model and a first set of observation points, where the first set of observation points is composed of any number of model parameter value combinations in the initial model parameter sampling point set and the terminal voltage fitting error value corresponding to each set of model parameter value combinations; An estimation module, configured to, when the target terminal voltage fitting error value is greater than or equal to a preset terminal voltage fitting error threshold, estimate the posterior distribution between the model parameters and the terminal voltage fitting error value of the equivalent circuit model based on the first set of observation points by using a preset probability surrogate model, to obtain an initial posterior distribution, where the target terminal voltage fitting error value is the smallest terminal voltage fitting error value in the first set of observation points; A prediction module, configured to predict the terminal voltage fitting error of each parameter sampling point in the first parameter sampling point set by using the initial posterior distribution, to obtain the predicted terminal voltage fitting error value of each parameter sampling point, where the first parameter sampling point set is the other parameter sampling point set in the initial model parameter sampling point set except the first set of observation points; An update module, configured to update the first set of observation points by using the initial posterior distribution, multiple predicted terminal voltage fitting error values, and an acquisition function, and determine a second set of observation points; An iteration module, configured to, based on the second set of observation points, return to the step of estimating the initial posterior distribution, and iterate repeatedly until the obtained target terminal voltage fitting error value is less than the preset terminal voltage fitting error threshold, and determine the model parameter combination corresponding to the target terminal voltage fitting error value as the global optimal parameter of the equivalent circuit model.

8. A computer device, characterized in that, Including: A memory and a processor, which are communicatively connected to each other. Computer instructions are stored in the memory, and the processor executes the computer instructions to execute the equivalent circuit model parameter identification method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, Computer instructions are stored on the computer-readable storage medium, and the computer instructions are used to cause a computer to execute the equivalent circuit model parameter identification method according to any one of claims 1 to 6.

10. A computer program product, characterized in that, It includes computer instructions, and the computer instructions are used to cause a computer to execute the equivalent circuit model parameter identification method according to any one of claims 1 to 6.

Images