Equivalent circuit parameter prediction method and device, equipment and storage medium

Abstract

The invention discloses an equivalent circuit parameter prediction method and device, equipment and a storage medium, and relates to the technical field of electrochemical analysis, the method comprises the following steps: processing label-free real impedance spectrum data based on an equivalent circuit parameter prediction model to obtain prediction parameters and reconstruction impedance spectrum data determined based on the prediction parameters; training an equivalent circuit parameter prediction model based on the reconstructed impedance spectrum data and the label-free real impedance spectrum data to obtain a target equivalent circuit parameter prediction model; and based on the target equivalent circuit parameter prediction model, predicting the to-be-predicted impedance spectrum data to obtain target parameters of the equivalent circuit model corresponding to the to-be-predicted impedance spectrum data. By means of the mode, no parameter label is needed, dependence on a real parameter label is avoided, self-supervised training is achieved, and the prediction accuracy and reliability of the model are improved by taking reproduction of a real spectrum measured objectively as a target.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electrochemical analysis, and particularly relates to an equivalent circuit parameter prediction method and device, equipment and a storage medium. BACKGROUND

[0002] Electrochemical impedance spectroscopy (EIS) is a powerful and non-destructive electrochemical characterization technique, which is widely used in battery (such as state of health SOH and state of charge SOC evaluation of lithium-ion batteries), corrosion science, fuel cells, supercapacitors, medical materials and material characterization, etc. For example, in the field of biomedical materials, EIS is often used to evaluate the degradation rate and corrosion mechanism of biologically active metals such as degradable medical zinc alloys, so as to guide the performance optimization and biocompatibility research. EIS obtains the information of the internal dynamics and physical and chemical processes of the system by applying a small-amplitude sinusoidal AC potential (or current) perturbation to the electrochemical system (or a series of small-amplitude sinusoidal AC potential (or current) perturbations) and measuring the impedance response at different frequencies. In order to extract valuable physical and chemical information from complex EIS data, the most commonly used method is to fit it to an equivalent circuit model (ECM).

[0003] Since the traditional fitting scheme is too dependent on artificial experience and low in efficiency, machine learning can usually be used to establish a direct mapping relationship between EIS data and EC parameters, taking impedance spectrum data (such as the real part and the imaginary part of N frequency points) as the input of the neural network, and taking the element values (such as R1, C1, R2) of the equivalent circuit as the output (label). The network is trained by minimizing the mean square error (MSE) and other loss functions between the predicted parameter values and the real parameter values. However, this direct regression method based on “spectrum-to-parameter” lacks physical constraints and relies on biased “real labels”, resulting in unreliable prediction results.

[0004] The above content is only used to assist in understanding the technical solutions of the present application and does not represent the acknowledgement of the above content as prior art. SUMMARY

[0005] The main purpose of the present application is to provide an equivalent circuit parameter prediction method, device, equipment and storage medium, which aims to solve the technical problems that the traditional machine learning scheme in the prior art relies on biased real labels and the prediction results are unreliable.

[0006] To achieve the above purpose, the present application provides an equivalent circuit parameter prediction method, which comprises the following steps: Based on the equivalent circuit parameter prediction model, the real impedance spectrum data without labels is processed to obtain prediction parameters and reconstructed impedance spectrum data determined based on the prediction parameters; Based on the reconstructed impedance spectrum data and the real impedance spectrum data without labels, the equivalent circuit parameter prediction model is trained to obtain a target equivalent circuit parameter prediction model. Based on the target equivalent circuit parameter prediction model, the to-be-predicted impedance spectrum data is predicted to obtain target parameters of the equivalent circuit model corresponding to the to-be-predicted impedance spectrum data.

[0007] In an embodiment, the equivalent circuit parameter prediction model includes a parameter prediction module and an impedance spectrum reconstruction module, and the impedance spectrum reconstruction module is built-in with a complex impedance analytical formula of an equivalent circuit model. The parameter prediction module is configured to predict the real impedance spectrum training data without labels to obtain prediction parameters, and input the prediction parameters into the impedance spectrum reconstruction module. The impedance spectrum reconstruction module is configured to calculate, based on the complex impedance analytical formula, complex impedance data of each angular frequency in a preset frequency list under the prediction parameters, separate the real part and the imaginary part of the complex impedance data, and obtain reconstructed impedance spectrum data.

[0008] In an embodiment, the step of training, based on the reconstructed impedance spectrum data and the real impedance spectrum data without labels, the equivalent circuit parameter prediction model to obtain a target equivalent circuit parameter prediction model includes: Based on the reconstructed impedance spectrum data and the real impedance spectrum data without labels, a prediction total loss is determined. When the prediction total loss does not meet a preset convergence condition, weights of a parameter prediction module in the equivalent circuit parameter prediction model are updated based on the prediction total loss, and the step of processing, based on the equivalent circuit parameter prediction model, the real impedance spectrum data without labels to obtain prediction parameters and reconstructed impedance spectrum data determined based on the prediction parameters is executed until the prediction total loss meets the preset convergence condition. When the prediction total loss meets the preset convergence condition, it is determined that the training of the equivalent circuit parameter prediction model is completed, and the equivalent circuit parameter prediction model is taken as a target equivalent circuit parameter prediction model.

[0009] In an embodiment, the step of determining, based on the reconstructed impedance spectrum data and the real impedance spectrum data without labels, a prediction total loss includes: A first correspondence relationship between the reconstructed impedance spectrum data, the real impedance spectrum data, and an error index is obtained. calculating an error index based on the reconstructed impedance spectrum data, the unlabeled real impedance spectrum data, and the first correspondence relationship; obtaining a second correspondence relationship between the error index and the predicted total loss; determining the predicted total loss based on the error index and the second correspondence relationship.

[0010] In an embodiment, the step of updating the weight of the parameter prediction module in the equivalent circuit parameter prediction model based on the predicted total loss comprises: determining a target gradient between the predicted total loss, the reconstructed impedance spectrum data, and the weight of the parameter prediction module based on the predicted total loss, the reconstructed impedance spectrum data, and the predicted parameter; updating the weight of the parameter prediction module in the equivalent circuit parameter prediction model based on the target gradient.

[0011] In an embodiment, the step of determining the target gradient between the predicted total loss, the reconstructed impedance spectrum data, and the weight of the parameter prediction module based on the predicted total loss, the reconstructed impedance spectrum data, and the predicted parameter comprises: calculating a first gradient between the predicted total loss and the reconstructed impedance spectrum data; calculating a second gradient between the reconstructed impedance spectrum data and the predicted parameter; calculating a third gradient between the predicted parameter and the weight of the parameter prediction module; obtaining a third correspondence relationship between the first gradient, the second gradient, the third gradient, and the target gradient; determining the target gradient based on the first gradient, the second gradient, the third gradient, and the third correspondence relationship.

[0012] In an embodiment, the step of processing the unlabeled real impedance spectrum data based on the equivalent circuit parameter prediction model to obtain the predicted parameter and the reconstructed impedance spectrum data determined based on the predicted parameter further comprises: obtaining original impedance spectrum data; normalizing the original impedance spectrum data based on a preset interval range to obtain the unlabeled real impedance spectrum data.

[0013] In addition, to achieve the above-mentioned purpose, the present application further provides an equivalent circuit parameter prediction device, which comprises: a training module configured to process the unlabeled real impedance spectrum data based on the equivalent circuit parameter prediction model to obtain the predicted parameter and the reconstructed impedance spectrum data determined based on the predicted parameter; The training module is further configured to train the equivalent circuit parameter prediction model based on the reconstructed impedance spectrum data and the unlabeled real impedance spectrum data, to obtain a target equivalent circuit parameter prediction model. The prediction module is configured to predict the to-be-predicted impedance spectrum data based on the target equivalent circuit parameter prediction model, to obtain target parameters of an equivalent circuit model corresponding to the to-be-predicted impedance spectrum data.

[0014] In addition, to achieve the above-mentioned purpose, the present application further provides an equivalent circuit parameter prediction device, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the computer program is configured to implement the steps of the equivalent circuit parameter prediction method as described above.

[0015] In addition, to achieve the above-mentioned purpose, the present application further provides a storage medium, which is a computer-readable storage medium, and the storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the equivalent circuit parameter prediction method as described above.

[0016] In addition, to achieve the above-mentioned purpose, the present application further provides a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the steps of the equivalent circuit parameter prediction method as described above.

[0017] The present application provides an equivalent circuit parameter prediction method, which processes unlabeled real impedance spectrum data based on an equivalent circuit parameter prediction model to obtain predicted parameters and reconstructed impedance spectrum data determined based on the predicted parameters; trains the equivalent circuit parameter prediction model based on the reconstructed impedance spectrum data and the unlabeled real impedance spectrum data to obtain a target equivalent circuit parameter prediction model; and predicts to-be-predicted impedance spectrum data based on the target equivalent circuit parameter prediction model to obtain target parameters of an equivalent circuit model corresponding to the to-be-predicted impedance spectrum data. The present application compares the reconstructed spectrum with the input spectrum, the training signal completely comes from the input data itself, and does not require any parameter label, which can eliminate the dependence on real parameter labels, realize self-supervised training, improve the reliability of the prediction result, and at the same time, takes the real spectrum of the objective measurement as the target, instead of imitating a fitting value that may not be accurate, so as to improve the accuracy of the prediction result, and solve the technical problem that the traditional machine learning scheme depends on biased real labels and the prediction result is unreliable. BRIEF DESCRIPTION OF DRAWINGS

[0018] The accompanying drawings, which are incorporated into and form a part of the specification, illustrate one embodiment consistent with the present application and, together with the description, serve to explain the principles of the application.

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows. Obviously, for those skilled in the art, other drawings can also be obtained based on these drawings without any creative effort.

[0020] Figure 1 Flowchart of the equivalent circuit parameter prediction method according to an embodiment of the present application; Figure 2 Structure diagram of the prediction model of the equivalent circuit parameter prediction method according to the first embodiment of the present application; Figure 3 Flowchart of the equivalent circuit parameter prediction method according to an embodiment of the present application; Figure 4 Structure diagram of the parameter prediction module of the equivalent circuit parameter prediction method according to the second embodiment of the present application; Figure 5 Traditional model prediction result diagram of the equivalent circuit parameter prediction method according to the second embodiment of the present application; Figure 6 Prediction result diagram of the equivalent circuit parameter prediction method according to the second embodiment of the present application; Figure 7 Module structure diagram of the equivalent circuit parameter prediction device according to the embodiment of the present application; Figure 8 Device structure diagram of the hardware running environment involved in the equivalent circuit parameter prediction method according to the embodiment of the present application.

[0021] The implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0022] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application, and are not used to limit the present application.

[0023] In order to better understand the technical solutions of the present application, the following will be described in detail with reference to the drawings in the specification and the specific embodiments.

[0024] The main solution of the embodiment of the present application is: based on the equivalent circuit parameter prediction model, the unlabeled real impedance spectrum data is processed to obtain the predicted parameters and the reconstructed impedance spectrum data determined based on the predicted parameters; based on the reconstructed impedance spectrum data and the unlabeled real impedance spectrum data, the equivalent circuit parameter prediction model is trained to obtain a target equivalent circuit parameter prediction model; based on the target equivalent circuit parameter prediction model, the to-be-predicted impedance spectrum data is predicted to obtain the target parameters of the equivalent circuit model corresponding to the to-be-predicted impedance spectrum data.

[0025] Currently, the realization of EIS data to the equivalent circuit parameters is mainly dependent on traditional fitting schemes and machine learning schemes improved on the basis of traditional fitting schemes. The traditional fitting scheme mainly relies on complex nonlinear least squares (CNLS) for fitting, which is highly dependent on manual operation, and its shortcomings are as follows: Initial value sensitivity. The fitting algorithm (such as Levenberg-Marquardt method) is very sensitive to the setting of the initial value of the parameter, and improper initial value can easily lead to the fitting process falling into a local optimal solution and obtaining incorrect parameter results.

[0026] Low efficiency. This process requires manual intervention and repeated debugging, which is time-consuming and labor-intensive, and cannot meet the needs of high-throughput experimental data processing or real-time online monitoring (such as battery management system BMS).

[0027] The traditional machine learning scheme usually takes the impedance spectrum data (such as the real and imaginary parts of N frequency points) as input and the element values of the equivalent circuit (such as R1, C1, R2) as output (label). The network is trained by minimizing the mean square error (MSE) between the predicted parameter value and the true parameter value. However, this direct regression method of "spectrum-to-parameter" has some fundamental defects: Dependence on high-quality "true labels". This scheme requires a large labeled training set, i.e. each impedance spectrum must correspond to a set of known and accurate "true" equivalent circuit parameter values.

[0028] Paradox of obtaining "true labels". These "true" parameter labels are usually obtained by the conventional CNLS fitting scheme with defects, which means that the "ceiling" of the neural network training is limited by the accuracy of the traditional fitting scheme, and the network is just "imitating" a slow and possibly inaccurate fitting process.

[0029] Lack of physical constraints. During training, the network only tries to mathematically minimize the L2 loss of the parameters, and it does not understand the physical meaning of these parameter combinations, which can lead the network to predict a set of parameters that are close to the "label values" in numerical terms, but when these parameters are substituted into the equivalent circuit physical formula, the resulting impedance spectrum deviates greatly from the true spectrum. In other words, the network lacks "cognition" of the electrochemical mechanism (i.e. the equivalent circuit topology).

[0030] Limitations of synthetic data. To avoid label acquisition, "synthetic data" (i.e. set parameters and then generate spectra by formula) can be used for training, but this leads to poor generalization of the model to real experimental noise, instrument artifacts and non-ideal responses (such as CPE).

[0031] It can be seen that the traditional scheme is either overly dependent on human experience and inefficient (traditional fitting), or in the application of machine learning, the prediction result is unreliable due to the lack of physical constraints and the dependence on biased "true labels".

[0032] The present application provides a solution, by comparing the reconstructed spectrum with the input spectrum, the training signal is completely from the input data itself, without any parameter label, can get rid of the dependence on the true parameter label, realize self supervised training, improve the reliability of the prediction result, at the same time, with the goal of reproducing the true spectrum of objective measurement, instead of imitating a fitting value which may not be accurate, so as to improve the accuracy of the prediction result, solve the technical problem that the traditional machine learning scheme depends on the biased true label and the prediction result is unreliable.

[0033] It should be noted that the execution subject of the present embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, an equivalent circuit parameter prediction device, etc., and the present embodiment does not make specific limitation thereto. In the following, the equivalent circuit parameter prediction device is taken as an example to explain the present embodiment and each of the following embodiments.

[0034] The present application provides an equivalent circuit parameter prediction method, which refers to Figure 1 , Figure 1 The present application provides an equivalent circuit parameter prediction method, which refers to

[0035] In the present embodiment, the equivalent circuit parameter prediction method comprises steps S10-S30: Step S10, based on the equivalent circuit parameter prediction model, processing the unlabelled true impedance spectrum data to obtain the predicted parameters and the reconstructed impedance spectrum data determined based on the predicted parameters; It should be noted that in order to extract valuable physical and chemical information from complex EIS data, neural network can be used to fit it into an equivalent circuit model (ECM). The equivalent circuit model is usually composed of resistance (R), capacitance (C), inductance (L) and constant phase angle element (CPE), Warburg impedance (W) and other elements, which are used to simulate the actual process in the electrochemical system, such as solution resistance, charge transfer resistance, double layer capacitance and diffusion process, etc. The impedance spectrum data measured by experiment is fitted to the selected equivalent circuit, and the parameter values of each element can be obtained. These parameter values (such as charge transfer resistance) have clear physical meaning and are the key to quantitative analysis and understanding of the internal state of the system.

[0036] In addition, it should be noted that, since the traditional fitting scheme relies too much on artificial experience and is inefficient, the present embodiment uses a neural network to establish a direct mapping relationship between the EIS data and the EC parameters. Since the traditional machine learning scheme relies on biased true labels, the prediction result is unreliable, and therefore the present embodiment improves the traditional prediction model.

[0037] It can be understood that the equivalent circuit parameter prediction model, i.e. the model designed by the present embodiment for predicting the equivalent circuit (model) element parameters, in the present embodiment, refers to Figure 2 , the equivalent circuit parameter prediction model includes a parameter prediction module and an impedance spectrum reconstruction module, and the impedance spectrum reconstruction module is built-in with a complex impedance analytical formula of the equivalent circuit model.

[0038] In specific implementation, the parameter prediction module is used to generate predicted parameters based on the unlabeled real impedance spectrum training data, and the predicted parameters are input into the impedance spectrum reconstruction module, and the impedance spectrum reconstruction module is used to generate reconstructed impedance spectrum data based on the predicted parameters, a preset frequency list and the complex impedance analytical formula. Specifically, the impedance spectrum reconstruction module is used to calculate the complex impedance data of each angular frequency in the preset frequency list under the predicted parameters based on the complex impedance analytical formula, separate the real part and the imaginary part of the complex impedance data, and obtain the reconstructed impedance spectrum data.

[0039] It should be noted that the real impedance spectrum data, i.e. the actually obtained impedance spectrum data, is used for model training, and in the present embodiment, the data input into the equivalent circuit parameter prediction model are all unlabeled data. The predicted parameters, i.e. the element parameters predicted by the equivalent circuit parameter prediction model, are usually in the form of vectors. The reconstructed impedance spectrum data, i.e. the impedance spectrum data obtained by inversion according to the complex impedance analytical formula based on the predicted parameters and the preset frequency list.

[0040] In addition, it should be noted that the preset frequency list, i.e. the pre-set fixed frequency list, contains multiple angular frequencies, i.e. The preset frequency list is not a training variable of the neural network, nor is it an output of the parameter prediction module, but an external input constant tensor. Its content is determined according to the sampling frequency of the original experimental data in the data preprocessing stage, and is passed to the impedance spectrum reconstruction module as a configuration parameter when the model is initialized. It is a key dimensional anchor point for the impedance spectrum reconstruction module to calculate correctly, so as to ensure that the reconstructed impedance spectrum data can be calculated at the same frequency points as the real impedance spectrum data , ensuring that the subsequent loss function can perform accurate point-by-point error comparison.

[0041] It can be understood that the first part of the model is a trainable parameter prediction module (from the input layer to the second last layer), which is a standard neural network such as MLP or CNN, responsible for receiving the pre-processed real impedance spectrum vector as input, and output a K-dimensional predicted parameter vector , where K is the number of equivalent circuit parameters to be predicted. The second part of the model is the impedance spectrum reconstruction module (i.e. the last layer), which is a non-trainable, prior knowledge-based fixed calculation layer, which has built-in precise physical mathematical formulas of the equivalent circuit, such as the complex impedance analytical formula , where is the angular frequency and the element parameter vector . The impedance spectrum reconstruction module receives the predicted parameter vector from the first part and a fixed frequency list (i.e. a preset frequency list), and strictly according to the built-in physical formula inversion calculation to obtain the corresponding reconstructed impedance spectrum data , , which has exactly the same data structure as the input .

[0042] Further, in a possible implementation, step S10 can include: obtaining original impedance spectrum data; based on a preset interval range, performing normalization processing on the original impedance spectrum data to obtain unlabeled real impedance spectrum data.

[0043] It should be noted that the original impedance spectrum data, i.e. the initially obtained impedance spectrum data without preprocessing, is a set of complex impedances measured in a predetermined frequency range , which can be used as input to the neural network after preprocessing. The preset interval range is a pre-set normalization range, for example: [0, 1]. In this embodiment, the preprocessing at least includes linear normalization processing.

[0044] It can be understood that before the data is input into the model, preprocessing is required. In a specific implementation, linear normalization processing is performed on the original impedance spectrum data to map it to the range of [0, 1] to obtain unlabeled real impedance spectrum data. Subsequently, the unlabeled real impedance spectrum data obtained after preprocessing is input into the model.

[0045] Step S20, based on the reconstructed impedance spectrum data and the unlabeled real impedance spectrum data, training the equivalent circuit parameter prediction model to obtain a target equivalent circuit parameter prediction model; It should be noted that the target equivalent circuit parameter prediction model is the model obtained after the equivalent circuit parameter prediction model is trained and can be applied.

[0046] In an implementation, step S20 can include steps S201-S203: Step S201, based on the reconstructed impedance spectrum data and the unlabeled real impedance spectrum data, determine a predicted total loss; It should be noted that when training the model, the embodiment does not need to set the real label of the equivalent circuit parameter. In the training process, a piece of unlabeled real impedance spectrum After forward propagation, the predicted parameter vector is generated in sequence through the parameter prediction module , and the reconstructed impedance spectrum is generated through the impedance spectrum reconstruction module .

[0047] It can be understood that the predicted total loss is the loss function obtained in the model training process. In the embodiment, the key loss function (Loss Function) is not to compare the predicted parameter with the real label, but to compare the difference between the reconstructed spectrum and the original input spectrum .

[0048] In an implementation, step S201 can include: obtaining a first correspondence relationship between the reconstructed impedance spectrum data, the real impedance spectrum data and the error indicator; based on the reconstructed impedance spectrum data, the unlabeled real impedance spectrum data and the first correspondence relationship, calculating the error indicator; obtaining a second correspondence relationship between the error indicator and the predicted total loss; based on the error indicator and the second correspondence relationship, determining the predicted total loss.

[0049] It should be noted that the error indicator is used to represent the difference between the reconstructed impedance spectrum and the original input real impedance spectrum , such as mean square error, mean absolute error, complex field weighted loss, etc., which is not limited in the embodiment.

[0050] In addition, it should be noted that the first correspondence relationship between the reconstructed impedance spectrum data, the real impedance spectrum data and the error indicator is the calculation relationship of the error indicator. For example, when the error indicator is the mean square error, the first correspondence relationship is the calculation relationship of the mean square error between the reconstructed impedance spectrum and the original input real impedance spectrum , that is .

[0051] It can be understood that the second correspondence relationship between the error indicator and the predicted total loss is the calculation relationship of the predicted total loss. For example, predicted total loss = error indicator 1 + error indicator 2, that is , predicted total loss = error indicator, that is The embodiment is not limited in this regard.

[0052] It should be understood that the corresponding error indicators are calculated according to the difference between the reconstructed impedance spectrum data and the real impedance spectrum data, and then the final prediction total loss is calculated according to the error indicators.

[0053] In step S202, when the prediction total loss does not meet the preset convergence condition, the weights of the parameter prediction module in the equivalent circuit parameter prediction model are updated based on the prediction total loss, and the step of processing the untagged real impedance spectrum data based on the equivalent circuit parameter prediction model to obtain the predicted parameters and the reconstructed impedance spectrum data determined based on the predicted parameters is returned until the prediction total loss meets the preset convergence condition. In a possible implementation, the step of updating the weights of the parameter prediction module in the equivalent circuit parameter prediction model based on the prediction total loss includes: determining a target gradient between the prediction total loss and the weights of the parameter prediction module based on the prediction total loss, the reconstructed impedance spectrum data and the predicted parameters; and updating the weights of the parameter prediction module in the equivalent circuit parameter prediction model based on the target gradient.

[0054] It can be understood that, in order to update the weights of the parameter prediction module , it is necessary to calculate the gradient of the loss prediction total loss to the weights , that is, the target gradient. The optimizer calculates the final target gradient according to the “spectrum-spectrum” difference (the difference between the reconstructed impedance spectrum and the original input real impedance spectrum ).

[0055] In a possible implementation, the step of determining the target gradient between the prediction total loss and the weights of the parameter prediction module based on the prediction total loss, the reconstructed impedance spectrum data and the predicted parameters includes: calculating a first gradient between the prediction total loss and the reconstructed impedance spectrum data; calculating a second gradient between the reconstructed impedance spectrum data and the predicted parameters; calculating a third gradient between the predicted parameters and the weights of the parameter prediction module; obtaining a third correspondence relationship between the first gradient, the second gradient, the third gradient and the target gradient; and determining the target gradient based on the first gradient, the second gradient, the third gradient and the third correspondence relationship.

[0056] It can be understood that the calculation of the target gradient follows the back propagation and the chain rule, and according to the chain rule, the target gradient can be decomposed into three main parts, that is, the first gradient, the second gradient and the third gradient. The third correspondence between the first gradient, the second gradient, the third gradient and the target gradient is the calculation relationship of the target gradient, as follows:

[0057] In the formula, is the first gradient, is the second gradient, is the third gradient, is the target gradient.

[0058] It should be noted that the first gradient is the gradient between the predicted total loss and the reconstructed impedance spectrum data, that is, the gradient of the predicted total loss with respect to the reconstructed impedance spectrum data, which is directly derived from the selected loss function; the second gradient is the gradient between the reconstructed impedance spectrum data and the predicted parameters, that is, the gradient of the reconstructed impedance spectrum data with respect to the predicted parameters, which represents how the physical formula of the equivalent circuit changes with the change of the predicted parameters , and this derivative tensor can essentially characterize which direction each parameter should be slightly changed in order to reduce the loss of the neural network; the third gradient is the gradient between the predicted parameters and the weights of the parameter prediction module, that is, the gradient of the predicted parameters with respect to the weights of the parameter prediction module, which can be back propagated to each layer of the parameter prediction module, and finally to the weights .

[0059] It can be understood that the whole training process uses the physical formula as a strong constraint to force the parameter prediction module to learn how to output the "correct" parameters that can reproduce the real spectrum.

[0060] Step S203: When the predicted total loss meets the preset convergence condition, it is determined that the training of the equivalent circuit parameter prediction model is completed, and the equivalent circuit parameter prediction model is taken as a target equivalent circuit parameter prediction model.

[0061] It should be noted that the preset convergence condition is the condition that the model convergence needs to reach, for example: the decrease amplitude of the predicted total loss is less than 10 -6 in the next 10 consecutive training rounds, which can be flexibly adjusted according to actual needs, and is not limited in this regard.

[0062] It can be understood that when the prediction total loss in the training process meets the preset convergence condition, it can be considered that the model converges, that is, the equivalent circuit parameter prediction model at this time has been trained and can be used as a final target model for inference prediction. When the prediction total loss in the training process does not meet the preset convergence condition, it is considered that the model has not converged, that is, the weights of the parameter prediction module in the model need to be further updated, the target gradient is calculated according to the prediction total loss, and the weights are updated according to the target gradient.

[0063] In step S30, the target equivalent circuit parameter prediction model is used to predict the to-be-predicted impedance spectrum data, so as to obtain the target parameters of the equivalent circuit model corresponding to the to-be-predicted impedance spectrum data.

[0064] It should be noted that the to-be-predicted impedance spectrum data is the impedance spectrum data for which the equivalent circuit element parameters need to be determined at present. The target parameters are the equivalent circuit element parameters of the to-be-predicted impedance spectrum data finally predicted, which are usually in the form of a vector.

[0065] It can be understood that after the model training is completed, the target equivalent circuit parameter prediction model is obtained, and the target equivalent circuit parameter prediction model can be used for inference prediction. At this time, a new real impedance spectrum is input into the trained model to perform a forward propagation. Only the output vector of the parameter prediction module is extracted, and this vector is the final prediction value of the equivalent circuit parameters corresponding to In the inference stage, the output of the impedance spectrum reconstruction module and the loss function are no longer used.

[0066] The embodiment provides an equivalent circuit parameter prediction method. Based on an equivalent circuit parameter prediction model, unlabeled real impedance spectrum data is processed to obtain predicted parameters and reconstructed impedance spectrum data determined based on the predicted parameters. The equivalent circuit parameter prediction model is trained based on the reconstructed impedance spectrum data and the unlabeled real impedance spectrum data to obtain a target equivalent circuit parameter prediction model. The target equivalent circuit parameter prediction model is used to predict to-be-predicted impedance spectrum data to obtain target parameters of an equivalent circuit model corresponding to the to-be-predicted impedance spectrum data. In the embodiment, the reconstructed spectrum is compared with the input spectrum, the training signal completely comes from the input data itself, and no parameter label is needed, so that the dependence on a real parameter label can be eliminated, self-supervised training is realized, the reliability of a prediction result is improved, the real spectrum of a reproduced objective measurement is used as a target instead of a fitting value that may be inaccurate, so that the accuracy of the prediction result can be improved. In addition, the physical constraints of the equivalent circuit can be effectively used, the data preparation cost is significantly reduced, and the model generalization capability and prediction precision are improved.

[0067] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as the above embodiment one can be referred to the above introduction, and the subsequent will not be described. On this basis, please refer to Figure 3 , step S10 can include steps S101-S102: Step S101, based on the parameter prediction module, the real impedance spectrum training data without label is predicted, and the predicted parameter is obtained; It should be noted that, referring to Figure 4 , the parameter prediction module can include convolution layer (Convolutional), pooling layer (Pooling) and multi-layer fully connected layer (Dense), wherein the activation function can use ReLU (Linear rectification function, linear rectification function), and the convolution layer is used for processing the local correlation of impedance spectrum data. The parameter prediction module is responsible for receiving the preprocessed real impedance spectrum vector as input (Input), and outputting a K-dimensional predicted parameter vector .

[0068] It can be understood that if Randles circuit is selected as the equivalent circuit model, it covers series resistance, charge transfer process, double layer effect and Warburg diffusion process, and has typical electrochemical representation, then the parameter vector to be predicted by the neural network includes , , , , , a total of five key element values, wherein is the solution resistance, is the charge transfer resistance, is the amplitude parameter of the constant phase angle element (CPE), is the phase index of CPE, is the Warburg coefficient.

[0069] Step S102, based on the impedance spectrum reconstruction module, the predicted parameter, the complex impedance analytical formula and the preset frequency list are inverted to obtain the reconstructed impedance spectrum data.

[0070] In one possible implementation, step S102 can include: based on the impedance spectrum reconstruction module, calculating the complex impedance data of each angular frequency in the preset frequency list under the predicted parameter; based on the impedance spectrum reconstruction module, separating the real part and the imaginary part of the complex impedance data to obtain the reconstructed impedance spectrum data.

[0071] It should be noted that the primary condition that the complex impedance analytical formula must satisfy is differentiability, because the training of the model relies on the gradient being able to smoothly pass through this layer, propagating from the loss function back to the parameter prediction module to update the network weights, which means that all mathematical operations in the formula must be continuous and differentiable, and the impedance formulas of common equivalent circuit elements (R, L, C, CPE, Warburg) are continuous and differentiable, which enables the gradient to smoothly pass through this layer. The output of this formula must be a complex impedance matching the dimension of the input impedance spectrum, and before loss calculation, it must be separated into real part and imaginary part , so as to calculate the mean square error (MSE) point by point with the real part and the imaginary part of the original input impedance spectrum.

[0072] It can be understood that, for example, assuming that the Randles circuit is selected as the equivalent circuit model, the impedance spectrum reconstruction module will receive the predicted parameters output by the parameter prediction module and the preset fixed frequency list at each forward propagation. For each angular frequency in the list , the module will strictly follow the built-in Randles circuit complex impedance analytical formula to calculate as follows:

[0073] wherein, is the series impedance, is the parallel impedance of charge transfer resistance and CPE, is the Warburg impedance.

[0074] It should be understood that, since is a complex number, the impedance spectrum reconstruction module needs to separate it into real part and imaginary part using complex algebra. The separation of the real part and the imaginary part is performed for all frequency points in the list, and finally the reconstructed impedance spectrum vector is obtained.

[0075] It can be understood that the conventional NN scheme can be considered as a "black box", which can predict a set of parameter combinations that are mathematically close to the label but physically meaningless. The present embodiment ensures that any set of parameters predicted by the network is "self-consistent" on the physical model, i.e. they combine to indeed (in the view of the network) optimally reproduce the original spectrum, by embedding the physical formula of the equivalent circuit as a strong constraint in the last layer. Like the conventional NN scheme, the training process of the present embodiment can be time-consuming, but once the training is completed, its inference process is millisecond-level, without time-consuming and laborious debugging, and is very suitable for real-time state monitoring of battery management system (BMS) or high-throughput material screening scenarios.

[0076] Assuming that the conventional neural network model used by the conventional machine learning scheme is the same as the parameter prediction module used by the present embodiment, both include convolutional layers, pooling layers and multiple layers of fully connected layers. The conventional neural network model adopts the conventional "spectrum-value" supervised learning paradigm, and the output dimension of the last fully connected layer is 5, which directly corresponds to the predicted parameters . When training the model, a large number of parameter real labels obtained by fitting through the conventional complex nonlinear least squares method are needed in advance. Its loss function focuses on minimizing the Euclidean distance or mean square error (MSE) between the predicted parameters and the parameter real labels . The effectiveness of this training mode is limited by the accuracy and robustness of the parameter real labels used.

[0077] Using the equivalent circuit parameter prediction model of the present embodiment, the output dimension of the parameter prediction module is still 5, generating a predicted parameter vector . The layer uses ReLU or Softplus activation function to ensure that the output physical parameters are positive. The impedance spectrum reconstruction module (physical information layer) is a non-trainable, hard-coded calculation module of the Randles circuit complex impedance calculation formula , which receives and a fixed frequency list to invert the reconstructed impedance spectrum , The dimension of the reconstructed impedance spectrum is the same as the input . The self-supervised loss function: the model is trained by minimizing . This "spectrum-spectrum" comparison mechanism enables the gradient signal to pass through the physical formula layer, forcing the neural network to learn a set of parameters that must be physically able to optimally reconstruct the original input spectrum.

[0078] By training the traditional neural network model and the equivalent circuit parameter prediction model and performing prediction performance evaluation on the same test set, there is a significant difference between the two. Reference Figure 5 , Figure 5 The horizontal coordinate is the impedance real part (Zre), and the vertical coordinate is the negative impedance imaginary part (-Zimg). The units of the horizontal and vertical coordinates are ohms (Ω). From Figure 5 It can be seen that the traditional neural network model has obvious generalization defects in the test set, especially in the equivalent circuit element value, especially the charge transfer resistance is in the high value interval, the predicted atlas and the actual atlas have a visible deviation and inconsistency, which proves that the traditional model performs well in the parameter space of the training set, but the learned parameter mapping relationship lacks physical constraints. Once the data enters the sparse or high value area, the predicted parameter combination produces significant errors when substituted into the physical formula, showing very low physical self-consistency. In contrast, the equivalent circuit parameter prediction model of the present embodiment exhibits excellent robustness and accuracy. Reference Figure 6 , Figure 6 The horizontal coordinate is the impedance real part (Zre), and the vertical coordinate is the negative impedance imaginary part (-Zimg). The units of the horizontal and vertical coordinates are ohms (Ω). From Figure 6 It can be seen that in the same high value area and the entire parameter range, the predicted atlas can perfectly fit the actual atlas, which powerfully proves the role of the impedance spectrum reconstruction module: by taking the "spectrum-spectrum" physical consistency as the only optimization target, the network is forced to learn the inherent and accurate physical relationship between the parameters and the spectrum, which means that the predicted parameters are not only mathematically reasonable, but also highly self-consistent in physics, thereby overcoming the fundamental defects of traditional models in prediction accuracy and generalization ability, and solving the problem of increasing model error with element value. It can be trained efficiently using massive unlabeled data and is superior to traditional supervised learning-based neural network methods in prediction accuracy and physical reasonableness.

[0079] The embodiment provides an equivalent circuit parameter prediction method, which comprises the following steps: based on a parameter prediction module, predicting real impedance spectrum training data without labels to obtain predicted parameters; and based on an impedance spectrum reconstruction module, inverting the predicted parameters to obtain reconstructed impedance spectrum data. In the embodiment, the reconstructed spectrum is compared with the input spectrum, the training signal is completely from the input data itself, and no parameter label is needed, so that the dependence on the real parameter label is eliminated, self-supervised training is realized, the reliability of the prediction result is improved, and the accuracy of the prediction result is improved by taking the real spectrum of the objective measurement as the target instead of imitating a fitting value which may be inaccurate. In addition, the physical constraint of the equivalent circuit can be effectively utilized, the data preparation cost is significantly reduced, and the model generalization capability and the prediction precision are improved.

[0080] It should be noted that the above examples are only used for understanding the present application and do not constitute a limitation on the equivalent circuit parameter prediction method of the present application. More forms of simple transformation based on the technical concept are within the protection scope of the present application.

[0081] The present application also provides an equivalent circuit parameter prediction device, please refer to Figure 7 The equivalent circuit parameter prediction device comprises: a training module 10 configured to process real impedance spectrum data without labels based on an equivalent circuit parameter prediction model to obtain predicted parameters and reconstructed impedance spectrum data determined based on the predicted parameters; The training module 10 is further configured to train the equivalent circuit parameter prediction model based on the reconstructed impedance spectrum data and the real impedance spectrum data without labels to obtain a target equivalent circuit parameter prediction model; a prediction module 20 configured to predict impedance spectrum data to be predicted based on the target equivalent circuit parameter prediction model to obtain target parameters of an equivalent circuit model corresponding to the impedance spectrum data to be predicted.

[0082] In a feasible implementation, the equivalent circuit parameter prediction model comprises a parameter prediction module and an impedance spectrum reconstruction module, and the impedance spectrum reconstruction module is built-in with a complex impedance analytical formula of an equivalent circuit model. The parameter prediction module is configured to predict real impedance spectrum training data without labels to obtain predicted parameters, and input the predicted parameters into the impedance spectrum reconstruction module. The impedance spectrum reconstruction module is configured to calculate, based on the complex impedance analytical formula, complex impedance data of each angular frequency in a preset frequency list under the predicted parameters, separate the real part and the imaginary part of the complex impedance data, and obtain reconstructed impedance spectrum data.

[0083] In an implementable embodiment, the prediction module 20 is further configured to determine a prediction total loss based on the reconstructed impedance spectrum data and the unlabeled real impedance spectrum data. When the prediction total loss does not meet the preset convergence condition, weights of the parameter prediction module in the equivalent circuit parameter prediction model are updated based on the prediction total loss, and the step of processing the unlabeled real impedance spectrum data based on the equivalent circuit parameter prediction model to obtain the predicted parameters and the reconstructed impedance spectrum data determined based on the predicted parameters is performed until the prediction total loss meets the preset convergence condition. When the prediction total loss meets the preset convergence condition, it is determined that the training of the equivalent circuit parameter prediction model is completed, and the equivalent circuit parameter prediction model is taken as a target equivalent circuit parameter prediction model.

[0084] In an implementable embodiment, the prediction module 20 is further configured to obtain a first correspondence relationship between the reconstructed impedance spectrum data, the real impedance spectrum data and an error index. Based on the reconstructed impedance spectrum data, the unlabeled real impedance spectrum data and the first correspondence relationship, the error index is calculated. A second correspondence relationship between the error index and the prediction total loss is obtained. Based on the error index and the second correspondence relationship, the prediction total loss is determined.

[0085] In an implementable embodiment, the prediction module 20 is further configured to determine a target gradient between the prediction total loss, the reconstructed impedance spectrum data and the predicted parameters based on the prediction total loss, the reconstructed impedance spectrum data and the predicted parameters. Based on the target gradient, the weights of the parameter prediction module in the equivalent circuit parameter prediction model are updated.

[0086] In an implementable embodiment, the prediction module 20 is further configured to calculate a first gradient between the prediction total loss and the reconstructed impedance spectrum data. A second gradient between the reconstructed impedance spectrum data and the predicted parameters is calculated. A third gradient between the predicted parameters and the weights of the parameter prediction module is calculated. A third correspondence relationship between the first gradient, the second gradient, the third gradient and the target gradient is obtained. Based on the first gradient, the second gradient, the third gradient and the third correspondence relationship, the target gradient is determined.

[0087] In an implementable embodiment, the training module 10 is further configured to obtain original impedance spectrum data. Based on a preset interval range, the original impedance spectrum data is normalized to obtain real impedance spectrum data without labels.

[0088] The equivalent circuit parameter prediction device provided in the application adopts the equivalent circuit parameter prediction method in the above embodiments, and can solve the technical problem that the traditional machine learning scheme relies on biased real labels and the prediction result is unreliable. Compared with the prior art, the equivalent circuit parameter prediction device provided in the application has the same beneficial effects as the equivalent circuit parameter prediction method provided in the above embodiments, and other technical features in the equivalent circuit parameter prediction device are the same as the features disclosed in the above embodiments, and thus will not be described herein.

[0089] The application provides an equivalent circuit parameter prediction device, which comprises at least one processor and a memory connected with the at least one processor in communication; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the equivalent circuit parameter prediction method in the above embodiment one.

[0090] Reference will be made to the following Figure 8 which shows a structural schematic diagram of an equivalent circuit parameter prediction device suitable for being used to implement the embodiments of the application. The equivalent circuit parameter prediction device in the embodiments of the application can include but is not limited to mobile terminals such as mobile phones, notebook computers, digital broadcast receivers, PDAs (Personal Digital Assistant), PADs (Portable Application Description), PMPs (Portable Media Player), vehicle-mounted terminals (for example, vehicle-mounted navigation terminals) and the like, and fixed terminals such as digital TVs, desktop computers and the like. Figure 8 The equivalent circuit parameter prediction device shown is only an example, and should not bring any limitation to the functions and use range of the embodiments of the application.

[0091] As Figure 8As shown, the equivalent circuit parameter prediction device can include a processing device 1001 (e.g., a central processing unit, a graphics processing unit, etc.) that can perform various appropriate actions and processes according to a program stored in a ROM (Read Only Memory) 1002 or a program loaded from a storage device 1003 into a RAM (Random Access Memory) 1004. In the RAM 1004, various programs and data required for the equivalent circuit parameter prediction device to operate are also stored. The processing device 1001, the ROM 1002, and the RAM 1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; the storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the equivalent circuit parameter prediction device to communicate with other devices wirelessly or by wire to exchange data. Although the equivalent circuit parameter prediction device having various systems is shown in the figure, it should be understood that all the shown systems are not required to be implemented or possessed. More or less systems can be alternatively implemented or possessed.

[0092] In particular, according to embodiments of the present disclosure, the processes described above with reference to the flowcharts can be implemented as a computer software program. For example, embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network by a communication device, or installed from the storage device 1003, or installed from the ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the methods of the embodiments of the present disclosure are performed.

[0093] The equivalent circuit parameter prediction device provided by the present disclosure adopts the equivalent circuit parameter prediction method in the above embodiments, and can solve the technical problem that the conventional machine learning scheme relies on biased true labels and the prediction result is unreliable. Compared with the prior art, the equivalent circuit parameter prediction device provided by the present disclosure has the same beneficial effects as the equivalent circuit parameter prediction method provided by the above embodiments, and other technical features in the equivalent circuit parameter prediction device are the same as the features disclosed in the previous embodiment method, which will not be repeated here.

[0094] It should be understood that various aspects of the disclosure can be implemented in hardware, software, firmware, or combinations thereof, to achieve the various aspects of the disclosure. In the description above, specific features, structures, materials or characteristics can be combined in any suitable manner without necessarily being limited to one or more embodiments or examples.

[0095] The above merely provides a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0096] The present application provides a computer readable storage medium having stored thereon computer readable program instructions (i.e. computer programs) for performing the equivalent circuit parameter prediction method in the above-described embodiments.

[0097] The computer readable storage medium provided by the present application may, for example, be a U disk, but is not limited to an electric, magnetic, optical, electromagnetic, infrared, or semiconductor system, system or device, or any combination of the above. More specific examples of the computer readable storage medium can include, but are not limited to, an electric connection with one or more conductive wires, a portable computer disk, a hard disk, a random access memory (RAM), a read only memory (ROM), an erasable programmable read only memory (EPROM or flash memory), an optical fiber, a portable compact disk read only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present embodiment, the computer readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system or device. The program code contained on the computer readable storage medium can be transmitted by any suitable medium, including but not limited to an electric wire, an optical cable, an RF (Radio Frequency), etc., or any suitable combination of the above.

[0098] The above computer readable storage medium can be contained in the equivalent circuit parameter prediction device; or can exist separately without being assembled into the equivalent circuit parameter prediction device.

[0099] The computer readable storage medium described above carries one or more programs, when the one or more programs are executed by the equivalent circuit parameter prediction device, cause the equivalent circuit parameter prediction device to: based on the equivalent circuit parameter prediction model, process the unlabeled real impedance spectrum data to obtain a predicted parameter and reconstructed impedance spectrum data determined based on the predicted parameter; based on the reconstructed impedance spectrum data and the unlabeled real impedance spectrum data, train the equivalent circuit parameter prediction model to obtain a target equivalent circuit parameter prediction model; and based on the target equivalent circuit parameter prediction model, predict the to-be-predicted impedance spectrum data to obtain a target parameter of an equivalent circuit model corresponding to the to-be-predicted impedance spectrum data.

[0100] Computer program code for carrying out operations of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0101] The flow diagrams and the block diagrams in the drawings are illustrations of architectures, functionalities, and operations of possible implementations of systems, methods, and computer program products according to various embodiments of present application. In this regard, each block in the flow diagrams or block diagrams can represent a module, a procedure, or a part of code, which comprises one or more executable instructions for implementing the specified functions. It should also be noted that, in some alternative implementations, the functions noted in the blocks can occur in a different order than that noted in the figures. For example, two blocks noted in succession can in fact be executed substantially concurrently or in the opposite order, depending on the functionality involved. It should also be noted that each block in the block diagrams and / or flow diagrams, and combinations of blocks in the block diagrams and / or flow diagrams, can be implemented by dedicated hardware-based systems that perform the specified functions or operations, or can be implemented by a combination of dedicated hardware-based systems and computer instructions.

[0102] The modules described in the embodiments of the present application can be implemented in the form of software or in the form of hardware. In some cases, the name of the module does not constitute a limitation on the module itself.

[0103] The readable storage medium provided by the present application is a computer readable storage medium, which stores computer readable program instructions (i.e., a computer program) for executing the equivalent circuit parameter prediction method described above, and can solve the technical problem that the traditional machine learning scheme relies on biased real labels and the prediction result is unreliable. Compared with the prior art, the computer readable storage medium provided by the present application has the same beneficial effects as the equivalent circuit parameter prediction method provided by the above embodiments, and will not be described here.

[0104] The present application also provides a computer program product comprising a computer program, which, when executed by a processor, implements the steps of the equivalent circuit parameter prediction method as described above.

[0105] The computer program product provided by the present application can solve the technical problem that the traditional machine learning scheme relies on biased real labels and the prediction result is unreliable. Compared with the prior art, the computer program product provided by the present application has the same beneficial effects as the equivalent circuit parameter prediction method provided by the above embodiments, and will not be described here.

[0106] The above is only some embodiments of the present application, and does not limit the patent scope of the present application. Any equivalent structural transformation made by using the content of the present application specification and drawings, or direct / indirect application in other related technical fields is included in the patent protection scope of the present application.

Claims

1. An equivalent circuit parameter prediction method characterized by, The method comprises: processing the unlabeled real impedance spectrum data based on an equivalent circuit parameter prediction model to obtain a predicted parameter and reconstructed impedance spectrum data determined based on the predicted parameter; training the equivalent circuit parameter prediction model based on the reconstructed impedance spectrum data and the unlabeled real impedance spectrum data to obtain a target equivalent circuit parameter prediction model; predicting the to-be-predicted impedance spectrum data based on the target equivalent circuit parameter prediction model to obtain a target parameter of an equivalent circuit model corresponding to the to-be-predicted impedance spectrum data.

2. The method of claim 1, wherein, The equivalent circuit parameter prediction model comprises a parameter prediction module and an impedance spectrum reconstruction module, and the impedance spectrum reconstruction module is internally provided with a complex impedance analytical formula of an equivalent circuit model; The parameter prediction module is configured to predict the unlabeled real impedance spectrum training data to obtain a predicted parameter, and input the predicted parameter into the impedance spectrum reconstruction module; The impedance spectrum reconstruction module is configured to calculate, based on the complex impedance analytical formula, complex impedance data of each angular frequency in a preset frequency list under the predicted parameter, separate the real part and the imaginary part of the complex impedance data to obtain reconstructed impedance spectrum data.

3. The method of claim 1, wherein, The step of training the equivalent circuit parameter prediction model based on the reconstructed impedance spectrum data and the unlabeled real impedance spectrum data to obtain a target equivalent circuit parameter prediction model comprises: determining a total prediction loss based on the reconstructed impedance spectrum data and the unlabeled real impedance spectrum data; when the total prediction loss does not meet a preset convergence condition, updating the weight of the parameter prediction module in the equivalent circuit parameter prediction model based on the total prediction loss, and returning to execute the step of processing the unlabeled real impedance spectrum data based on the equivalent circuit parameter prediction model to obtain a predicted parameter and reconstructed impedance spectrum data determined based on the predicted parameter until the total prediction loss meets the preset convergence condition; when the total prediction loss meets the preset convergence condition, determining that the training of the equivalent circuit parameter prediction model is completed, and taking the equivalent circuit parameter prediction model as a target equivalent circuit parameter prediction model.

4. The method of claim 3, wherein, The step of determining a total prediction loss based on the reconstructed impedance spectrum data and the unlabeled real impedance spectrum data comprises: obtaining a first correspondence relationship between the reconstructed impedance spectrum data, the real impedance spectrum data and an error index; calculating the error index based on the reconstructed impedance spectrum data, the unlabeled real impedance spectrum data and the first correspondence relationship; obtaining a second correspondence relationship between the error index and the total prediction loss; determining the total prediction loss based on the error index and the second correspondence relationship.

5. The method of claim 3, wherein, The step of updating the weight of the parameter prediction module in the equivalent circuit parameter prediction model based on the total prediction loss comprises: determining a target gradient between the total prediction loss, the reconstructed impedance spectrum data and the predicted parameter based on the total prediction loss, the reconstructed impedance spectrum data and the predicted parameter; updating the weight of the parameter prediction module in the equivalent circuit parameter prediction model based on the target gradient.

6. The method of claim 5, wherein, The step of determining a target gradient between the predicted total loss and the weight of the parameter prediction module based on the predicted total loss, the reconstructed impedance spectrum data and the predicted parameter comprises: calculating a first gradient between the predicted total loss and the reconstructed impedance spectrum data; calculating a second gradient between the reconstructed impedance spectrum data and the predicted parameter; calculating a third gradient between the predicted parameter and the weight of the parameter prediction module; obtaining a third correspondence relationship between the first gradient, the second gradient, the third gradient and the target gradient; determining the target gradient based on the first gradient, the second gradient, the third gradient and the third correspondence relationship.

7. The method of any one of claims 1 to 6, wherein, The step of processing the unlabeled real impedance spectrum data based on the equivalent circuit parameter prediction model to obtain the predicted parameter and the reconstructed impedance spectrum data determined based on the predicted parameter further comprises: obtaining original impedance spectrum data; normalizing the original impedance spectrum data based on a preset interval range to obtain the unlabeled real impedance spectrum data.

8. An equivalent circuit parameter prediction device characterized by comprising: The device comprises: a training module configured to process the unlabeled real impedance spectrum data based on the equivalent circuit parameter prediction model to obtain the predicted parameter and the reconstructed impedance spectrum data determined based on the predicted parameter; the training module is further configured to train the equivalent circuit parameter prediction model based on the reconstructed impedance spectrum data and the unlabeled real impedance spectrum data to obtain a target equivalent circuit parameter prediction model; a prediction module configured to predict the to-be-predicted impedance spectrum data based on the target equivalent circuit parameter prediction model to obtain target parameters of an equivalent circuit model corresponding to the to-be-predicted impedance spectrum data.

9. An equivalent circuit parameter prediction device characterized by comprising: The device comprises a memory, a processor and a computer program stored on the memory and executable on the processor, and the computer program is configured to implement the steps of the equivalent circuit parameter prediction method according to any one of claims 1 to 7.

10. A storage medium, characterized by The storage medium is a computer readable storage medium, and the storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the equivalent circuit parameter prediction method according to any one of claims 1 to 7.

Images