A method for estimating battery state of health, storage medium and electronic device

Abstract

The application relates to the technical field of battery management, and particularly provides a battery health state estimation method, a storage medium and an electronic device. The method can comprise the following steps: inputting a charging feature vector in a battery charging process into a mapping network, obtaining an initial prediction value of a battery health state; determining a prediction value of the battery health state based on the initial prediction value, the charging feature vector and a dynamics network; the mapping network and the dynamics network constitute a fusion neural network model; a loss value between the prediction value and an actual value of the battery health state is solved by using a target function; model parameters of the fusion neural network model are optimized based on the loss value, and a battery health state evaluation model is obtained; and the battery health state evaluation model is used for predicting a real-time value of the battery health state. The application can realize accurate estimation of the battery health state.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of battery management, in particular to a battery state of health estimation method, a storage medium and an electronic device. BACKGROUND

[0002] The state of health (SOH) of a battery is an important indicator for evaluating the performance of the battery. It reflects the ability of the battery to store electrical energy relative to a new battery and is usually expressed in percentage form.

[0003] There are various methods for evaluating the SOH of a battery pack, such as direct measurement and indirect measurement. The direct measurement method mainly determines the internal resistance, capacity and other parameters of the battery through techniques such as electrochemical impedance spectroscopy, thereby judging the state of health of the battery. The indirect measurement method calculates the SOH of the battery by analyzing the discharge curve, self-discharge rate and other data of the battery. However, the direct measurement method is easily affected by external environments and affects the accuracy of the results; the indirect measurement method relies on full discharge curves, which are difficult to obtain in actual applications.

[0004] Therefore, how to provide a technical solution for a more efficient and accurate battery state of health estimation method becomes a technical problem to be solved. SUMMARY

[0005] Some embodiments of the present application aim to provide a battery state of health estimation method, a storage medium and an electronic device. The technical solution of the embodiments of the present application can realize accurate estimation of the state of health of the battery, has high estimation efficiency and good universality.

[0006] In a first aspect, some embodiments of the present application provide a battery state of health estimation method, comprising: inputting a charging feature vector in a charging process of a battery into a mapping network to obtain an initial prediction value of a state of health of the battery; determining a prediction value of the state of health of the battery based on the initial prediction value, the charging feature vector and a kinetic network; wherein the charging feature vector comprises multi-dimensional charging features and charging cycle features; wherein the mapping network and the kinetic network constitute a fusion neural network model; solving a loss value between the prediction value and a true value of the state of health of the battery by using an objective function; wherein the objective function comprises a data fitting loss of the state of health of the battery, a physical residual loss and a monotonicity change trend loss of the state of health of the battery; optimizing model parameters of the fusion neural network model based on the loss value to obtain a battery state of health evaluation model; wherein the battery state of health evaluation model is used to predict a real-time value of the state of health of the battery.

[0007] Some embodiments of the present application input the charging feature vector extracted in the charging process into the fusion neural network model constructed by the present application to output a predicted value; then use a target function containing multiple losses to determine the loss value between the predicted value and the true value, so as to optimize the fusion neural network model and obtain a battery health state evaluation model. The present application can realize accurate estimation of the battery health state through the extracted basic charging feature vector in the charging process, is suitable for various battery health state estimation scenarios, has good universality, and has high estimation efficiency.

[0008] In some embodiments, before the charging feature vector in the battery charging process is input into the mapping network, the method further comprises: selecting a charging section for feature extraction from the battery charging process; performing feature extraction on the voltage change curve and the current change curve in the charging section respectively to obtain multi-dimensional original charging features; and performing normalization processing on the multi-dimensional original charging features and the charging cycle number to obtain the multi-dimensional charging features and the charging cycle features.

[0009] Some embodiments of the present application extract features in the selected charging section to obtain multi-dimensional original charging features, and then combine the charging cycle number to perform normalization processing to obtain multi-dimensional charging features and charging cycle features, so as to realize standardization processing of data and provide a data basis for subsequent model training.

[0010] In some embodiments, the method further comprises: based on the initial predicted value, the charging feature vector and the kinetic network, determining a predicted value of the battery health state, comprising: constructing a network input vector based on at least the initial predicted value and the charging feature vector; inputting the network input vector into the kinetic network to obtain the predicted value.

[0011] Some embodiments of the present application process the charging feature vector through the mapping network and the kinetic network to obtain the predicted value, so as to realize accurate prediction of the battery health state.

[0012] In some embodiments, the method further comprises: based on the initial predicted value, the charging feature vector and the kinetic network, determining a predicted value of the battery health state, comprising: fitting a differential equation corresponding to the battery health state in the battery charging process; performing differential extraction on the differential equation to obtain a first differential result and a second differential result; and splicing the charging feature vector, the initial predicted value, the first differential result and the second differential result to obtain the network input vector.

[0013] Some embodiments of the present application construct the network input vector of the kinetic network through the initial predicted value and the differential result related to the battery health state, so as to facilitate accurate prediction of the battery health state.

[0014] In some embodiments, the mapping network is connected with a decay rate network; the method further comprises: inputting the initial prediction value into the decay rate network to obtain a change trend of the battery health state; wherein the decay rate network is a full connection network, a residual prediction network or a inflection point detection network.

[0015] Some embodiments of the present application determine the change value of the battery health state through the initial prediction value and the decay rate network, and realize the prediction of the change trend of the battery health state.

[0016] In some embodiments, the target function is constructed by: setting a first coefficient and a second coefficient corresponding to the physical residual loss and the monotonic change trend loss respectively; and constructing the target function based on the physical residual loss, the monotonic change trend, the first coefficient, the second coefficient and the data fitting loss.

[0017] Some embodiments of the present application construct the target function through multiple losses, which can realize the accurate optimization of the model and improve the accuracy and robustness of the final battery health state evaluation model.

[0018] In some embodiments, the data fitting loss represents the mean square error value between the prediction value and the true value; the physical residual loss represents the differential residual value; and the monotonic change trend loss represents the consistency between the prediction trend of the prediction value and the true trend of the true value.

[0019] In the second aspect, some embodiments of the present application provide a method for estimating battery health state, comprising: obtaining a real-time charging feature vector in a current charging process; inputting the charging feature vector into a battery health state evaluation model to obtain a real-time value of the battery health state; wherein the battery health state evaluation model is obtained by executing any method embodiment in the first aspect.

[0020] In some embodiments, the method further comprises: confirming that the real-time value is less than a set threshold, triggering a battery maintenance prompt to inform relevant personnel.

[0021] In a third aspect, some embodiments of the present application provide a device for estimating battery state of health, comprising: a training module configured to input a charging feature vector in a battery charging process into a mapping network to obtain an initial predicted value of a battery state of health; determine a predicted value of the battery state of health based on the initial predicted value, the charging feature vector and a dynamics network; wherein the charging feature vector comprises a multi-dimensional charging feature and a charging cycle feature; a loss calculation module configured to solve a loss value between the predicted value and a true value of the battery state of health by using an objective function; wherein the objective function comprises a data fitting loss of the battery state of health, a physical residual loss and a monotonicity change trend loss of the battery state of health; and an optimization module configured to optimize model parameters of the fusion neural network model based on the loss value to obtain a battery state of health evaluation model; wherein the battery state of health evaluation model is configured to predict a real-time value of the battery state of health.

[0022] In a fourth aspect, some embodiments of the present application provide a device for estimating battery state of health, comprising: an acquisition module configured to acquire a real-time charging feature vector in a current charging process; and an estimation module configured to input the charging feature vector into a battery state of health evaluation model to obtain a real-time value of a battery state of health; wherein the battery state of health evaluation model is obtained by executing any method embodiment in the first aspect.

[0023] In a fifth aspect, some embodiments of the present application provide a computer readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, can implement the method according to any embodiment of the first aspect and the second aspect.

[0024] In a sixth aspect, some embodiments of the present application provide an electronic device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, can implement the method according to any embodiment of the first aspect and the second aspect.

[0025] In a seventh aspect, some embodiments of the present application provide a computer program product, comprising a computer program, wherein the computer program, when executed by a processor, can implement the method according to any embodiment of the first aspect and the second aspect. BRIEF DESCRIPTION OF DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of some embodiments of the present application, the following will briefly introduce the drawings needed to be used in some embodiments of the present application. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope, and for those skilled in the art, other related drawings can also be obtained without creative labor.

[0027] Figure 1 System diagram of battery state of health estimation provided for some embodiments of the present application;

[0028] Figure 2 Method flow diagram of obtaining battery state of health estimation model provided for some embodiments of the present application;

[0029] Figure 3 Method flow diagram of battery state of health estimation provided for some embodiments of the present application;

[0030] Figure 4 Device composition block diagram of battery state of health estimation provided for some embodiments of the present application;

[0031] Figure 5 Device composition block diagram of battery state of health estimation provided for some embodiments of the present application;

[0032] Figure 6 Electronic device schematic diagram provided for some embodiments of the present application. DETAILED DESCRIPTION

[0033] The technical solutions in some embodiments of the present application will be described below with reference to the drawings in some embodiments of the present application.

[0034] It should be noted that similar reference numerals and letters indicate similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. Meanwhile, in the description of the present application, the terms “first”, “second”, etc. are only used for differentiation in description, and cannot be understood as indicating or implying relative importance.

[0035] In the related art, in the field of lithium electronic battery management system (BMS), the SOH is usually estimated by using a physical model method and a data-driven method. Specifically, the physical model method mainly realizes SOH estimation based on equivalent circuit model (ECM), Pseudo-2D electrochemical model (P2D), etc.; this method needs to establish a large number of chemical and thermodynamic parameters, the calculation amount is large, and the parameter identification is difficult, and needs to be repeatedly calibrated for different chemical compositions. The other data-driven method mainly uses regression, support vector machine, Gaussian process, deep neural network, etc., directly extracts features from complete charge-discharge curves to regress SOH; however, this method has high requirements for feature universality, needs to design special features for different data sets or working conditions, and has poor generalization ability; and, it often relies on full discharge curves, which is difficult to obtain in actual electric vehicles and energy storage systems.

[0036] From the above related technologies, it can be seen that the existing SOH estimation method has the problems of poor stability, high calibration cost, and difficulty in balancing generality, estimation accuracy and calculation efficiency.

[0037] In view of this, some embodiments of the present application provide a battery health state estimation method, which can extract a charging feature vector from a certain period of the charging process, input it into the mapping network and the dynamics network designed by the present application, and obtain the predicted value of the battery health state; then, the loss value between the predicted value and the true value is calculated using the target function containing multiple losses, and the fusion neural network model composed of the mapping network and the dynamics network is optimized based on this to obtain the battery health state evaluation model. Through the battery health state evaluation model, the prediction of the battery health state can be realized. The battery health state estimation method provided by the embodiments of the present application uses the charging feature vector which is easy to extract to train the model and obtain the battery health state evaluation model. The model has good stability and can balance generality, estimation accuracy and estimation efficiency, and has high practicability. It can be seen that the embodiments of the present application do not need full discharge curves and can be suitable for multiple battery types and different charge-discharge protocols; at the same time, through optimization of model parameters in other scenarios, rapid adaptation between batteries of different chemical compositions can also be realized.

[0038] The following will be described in conjunction with the accompanying drawings Figure 1 Exemplary of the overall structure of the battery health state estimation system provided by some embodiments of the present application.

[0039] As Figure 1 shown, some embodiments of the present application provide a battery health state estimation system diagram, which can include: a device terminal containing a lithium battery pack and a data processing end 200. Wherein, the data processing end 200 can obtain a charging period of the device terminal 100 in the charging process; the data processing end 200 extracts multi-dimensional charging features in the charging period, inputs the real-time charging feature vector composed of the charging cycle features into the battery health state evaluation model pre-trained and deployed internally, and obtains the real-time predicted value of SOH.

[0040] In order to realize accurate estimation of the lithium ion battery health state, it is necessary to first obtain the battery health state evaluation model through training.

[0041] The following will be described in conjunction with the accompanying drawings Figure 2 Exemplary of the implementation process of obtaining the battery health state evaluation model by the data processing end 200 provided by some embodiments of the present application.

[0042] Please refer to the accompanying drawings Figure 2 , Figure 2 A method flow chart for obtaining a battery health state evaluation model is provided for some embodiments of the present application.

[0043] In some embodiments of the present application, before performing the method of obtaining the battery health state evaluation model, it is necessary to first prepare a training data set containing charging feature vectors in different cycles and their corresponding true values of SOH.

[0044] Specifically, in some embodiments of the present application, the charging feature vector is obtained by the following method: selecting a charging section for feature extraction from the battery charging process; performing feature extraction on the voltage change curve and the current change curve in the charging section to obtain multi-dimensional original charging features; and normalizing the multi-dimensional original charging features and the charging cycle number to obtain the multi-dimensional charging features and the charging cycle features.

[0045] For example, in specific embodiments of the present application, the constant current-constant voltage (CC-CV) charging end section (as a specific example of the charging section) is selected from the battery charging process; wherein the voltage interval is 4.0V~4.2V (CC later section), and the current interval is 0.2A~1.0A (CV initial section). This section of data is relatively stable and widely exists in different charging strategies, with good universality and availability. The following 8 statistical features are extracted from the voltage-time curve (as a specific example of the voltage change curve) and the current-time curve (as a specific example of the current change curve) of the charging end section, respectively, for a total of 16 features (as a specific example of the multi-dimensional original charging features); and the current cycle number d (as a specific example of the charging cycle number) is added, thereby forming a 17-dimensional input feature vector together with the 16 features.

[0046] Specifically, the 8 statistical features mainly include: mean, standard deviation, skewness and kurtosis (used to measure the data distribution form), curve slope, cumulative charge, entropy (Shannon entropy measures the curve information complexity), and charging duration (i.e. the time span of the charging end).

[0047] The calculation of the mean and the standard deviation uses the conventional calculation formula. The calculation method of the curve slope slope is as follows:

[0048]

[0049] wherein, t i and x i are known parameters in the curve; is the mean of the parameter t is the mean of the parameter is the mean of the parameter x is the mean of the parameter

[0050] Cumulative electric quantity Q: .

[0051] In the formula, I i is the current at the i-th moment, is the time-varying value.

[0052] After that, all the input features described above are scaled using the Min-Max normalization method, so that the 17-dimensional features are mapped to the range of [-1, 1], thereby ensuring the stability of the model training and the balance of the contribution of each feature.

[0053] The normalization formula of the Min-Max normalization method is:

[0054]

[0055] Among them, denotes the normalized result of the i-th dimensional feature (the set of all features is denoted as j), denotes the i-th dimensional feature.

[0056] After normalization processing by the above formula, multi-dimensional charging features and charging cycle features can be obtained.

[0057] The process of obtaining the battery health state evaluation model is exemplarily described below. It should be understood that the battery health state evaluation model is obtained by training the fusion neural network model composed of the mapping network and the dynamics network.

[0058] In some embodiments of the present application, the method for obtaining the battery health state evaluation model comprises:

[0059] S210, input the charging feature vector in the battery charging process to the mapping network to obtain the initial prediction value of the battery health state; wherein the charging feature vector includes multi-dimensional charging features and charging cycle features.

[0060] For example, in the specific embodiments of the present application, the core network structure is a physical information neural network structure (PINN) (as a specific example of the fusion neural network model), which is composed of a mapping network F and a dynamics network G, and a decay rate network is cascaded after F to supplement the nonlinear trend change characteristics. By inputting the 17-dimensional features after the above normalization processing into the physical information neural network structure, the prediction of SOH in the training process is realized, and the prediction value is obtained.

[0061] Specifically, the input variable F(d', j ; ); wherein d' is the cycle number characteristic of the current battery (as a specific example of the charge cycle characteristic, i.e., the result of the d normalization processing); such as the 1st cycle or the 200th cycle, representing the time dimension of the battery aging process, which reflects the long-term trend on the SOH decline curve. j is a 16-dimensional statistical feature vector extracted from the end of the CC-CV charging and normalized, including features such as mean, slope, and entropy of the voltage and current curves, which have good representativeness and availability; is a set of trainable parameters of F (such as neural network weights W and bias b), which is continuously optimized through gradient descent in the training process. Correspondingly, F can output the predicted value of the SOH of the current cycle, denoted as: = F (d', j ). Specifically, the network structure of F contains multiple layers of MLP (Multilayer Perceptron), each layer uses torch.sin() as the activation function, the weight is initialized using Xavier Normal, and the Dropout probability is 0.2. In order to improve the stability of model training, the periodic activation function is used to realize feature mapping under physical constraints in the embodiments of the present application, which improves the smoothness of the SOH curve and the frequency domain response ability, and is more consistent with the nonlinear oscillation trend existing in battery degradation; for small sample training, sin activation is more stable than ReLU or GELU in fitting the SOH differential evolution trend. This method uses 0.2 probability Dropout, not only to prevent overfitting, but also to suppress the problem of strong coupling between statistical features and degradation trends. When there is only one battery data and the dimension is high (17 dimensions), it is easy to appear the overfitting of the model "remembering each dimension"; the introduction of Dropout makes the network train in multiple suboptimal paths, thereby enhancing the generalization ability and supporting the transfer learning goal.

[0062] In some embodiments of the present application, the mapping network is connected with a decay rate network; the method further comprises: inputting the initial predicted value into the decay rate network to obtain the change trend of the battery health state; wherein the decay rate network is a fully connected network, a residual prediction network, or a inflection point detection network.

[0063] For example, in the specific embodiments of the present application, after the F network output, a small fully connected network (as a specific example of the decay rate network) is accessed to learn the rate of battery SOH decline (i.e., ASOH, as a specific example of the change trend), to assist in improving the prediction sensitivity of the model at the aging inflection point. ASOH represents the degradation rate of a period, which is non-stationary, nonlinear, and change sudden; after introducing the ASOH branch, the model has the ability to model the "change trend" in structure, especially in the acceleration stage. In addition to using ASOH, the fully connected network of the ASOH branch can be replaced by a residual prediction network, which learns ASOH (ASOH k = SOH k -SOH k-1 ) as a supervised signal; in this way, F has a trend modeling function module, which can learn the nonlinear mapping of the first derivative term, and in the specific implementation, a two-layer MLP can be used, the input includes d', j and SOH k-1 . In addition, a change point detection subnetwork (Change-point-aware Subnet) can also be introduced, which judges whether the current period is in the "degradation acceleration stage" by training a small network; output a binary or soft score as a dynamic weighting factor for ASOH output; SOH curve usually has obvious S-type degradation trend, the network can judge through the change rate of change trend (such as the third derivative).

[0064] S220, determining the predicted value of the battery health state based on the initial predicted value, the charging feature vector, and the dynamics network.

[0065] In some embodiments of the present application, S220 can include:

[0066] S221, constructing a network input vector based at least on the initial predicted value and the charging feature vector.

[0067] For example, in the specific embodiments of the present application, the F output and other feature data are used as the network input vector of G.

[0068] In some embodiments of the present application, S221 can include: fitting a differential equation corresponding to the battery health state during the battery charging process; differentiating the differential equation to obtain a first differential result and a second differential result; and splicing the charging feature vector, the initial predicted value, the first differential result, and the second differential result to obtain the network input vector.

[0069] For example, in the specific embodiments of the present application, first fit the implicit degradation differential equation of SOH:

[0070]

[0071] After automatic differentiation extraction, the first differential result and the second differential result are obtained. The extraction formula of the first differential result is:

[0072]

[0073] The extraction formula of the second differential result is:

[0074]

[0075] After the relevant data are spliced, the network input vector [d', d'', d''', d'''', d''''' ] of G is obtained, which has a total of 35 dimensions. j , , ,

[0076] S222, inputting the network input vector into the dynamics network to obtain the prediction value.

[0077] For example, in the specific embodiments of the present application, the 35-dimensional network input vector is input into G to obtain the prediction value .

[0078] In order to realize the accurate optimization of the physical information neural network, some embodiments of the present application further construct a target function containing multiple losses; wherein the target function includes a data fitting loss of the battery state of health, a physical residual loss and a monotonicity change trend loss of the battery state of health. The data fitting loss represents the mean square error value between the prediction value and the true value; the physical residual loss represents the differential residual value; and the monotonicity change trend loss represents the consistency of the prediction trend of the prediction value and the true trend of the true value.

[0079] Specifically, the target function is constructed by the following method: setting the first coefficient and the second coefficient corresponding to the physical residual loss and the monotonicity change trend loss respectively; and constructing the target function based on the physical residual loss, the monotonicity change trend, the first coefficient, the second coefficient and the data fitting loss.

[0080] For example, in the specific embodiments of the present application, the result of weighted summation of the physical residual loss, the monotonicity change trend, the first coefficient and the second coefficient is added to the data fitting loss to obtain the target function L. The formula of the target function L is:

[0081]

[0082] wherein,​α is the first coefficient, β is the second coefficient, both are empirical parameters.

[0083] L data is the data fitting loss, used to ensure that the predicted value fits the true label:

[0084]

[0085] wherein, is the predicted value of the i-th sample predicted by the model, is the SOH true value of the i-th sample; MSE is the mean square error.

[0086] L PDE is the physical residual loss, which is a differential residual based on automatic differentiation and dynamic network, used to simulate the implicit dynamic law of the degradation process; the difference between the predicted value and the autograd value constitutes the PDE residual:

[0087]

[0088] wherein, is the partial derivative of the model predicted value with respect to time, indicating the degradation rate of SOH with respect to cycle time. f is the difference between the two, i.e. the difference between the actual derivative and the model predicted derivative, f 1 and f 2 may be the residual corresponding to the data of two different samples or batches.

[0089] It should be noted that the MLP in the embodiment is not trained in isolation, but is jointly backpropagated with the PDE constraint term. The continuous differentiability of the sin activation function makes the autograd calculation , more stable. The network model in the application must satisfy the existence and smoothness of the gradient under automatic differentiation.

[0090] L mono is the monotonicity change trend loss (i.e. monotonicity constraint), by introducing monotonicity physical prior, a sign consistency product loss function is designed to suppress unreasonable prediction rebound and strengthen SOH degradation trend modeling. Specifically:

[0091]

[0092] wherein, is the change of SOH predicted by the physical information neural network model, For the real SOH change. If the model prediction trend is inconsistent with the real trend, the product is positive, ReLU has output, and is punished; if the trends are consistent, the product is negative, ReLU = 0, and is not punished.

[0093] S230, using the objective function to solve the loss value between the predicted value and the real value of the battery health state.

[0094] For example, in the specific embodiments of the present application, the loss value between the predicted value and the real SOH is solved by the above-mentioned constructed objective function L.

[0095] S240, optimizing the model parameters of the fusion neural network model based on the loss value to obtain a battery health state evaluation model; wherein the battery health state evaluation model is used to predict the real-time value of the battery health state.

[0096] For example, in the specific embodiments of the present application, the loss value obtained above is used to optimize the model parameters of the physical information neural network to obtain a required battery health state evaluation model. For example, through optimization, a battery health state evaluation model with a precision rate reaching a set value can be obtained; or after optimization, the number of iterations of the trained model is reached, at which time the battery health state evaluation model can be output.

[0097] In the process of training the physical information neural network described above, in the data set construction stage, a self-made battery can be collected, and two hundred cycles of complete charging and discharging data can be used to construct two hundred charging feature vectors and real value training samples, of which one hundred and fifty are training sets, thirty are validation sets, and twenty are test sets. Then the hyperparameters of the model are set, such as batch size, total epoch number, warm-UP epoch, warm-UP learning rate, initial learning rate, final learning rate, and learning rate strategy. Multiple independent experiments (such as 10 times) can be performed, and the results are stable. The training end condition can be that the average MAPE is at 0.6%~1.2%, and the RMSE<0.04.

[0098] The model is trained and exported in ONNX format; a lightweight inference engine is deployed in the MCU of the embedded BMS to complete online SOH inference; the battery health state evaluation model has low inference delay, small calculation resource occupation, and strong adaptability.

[0099] Some embodiments of the present application are exemplarily described below with reference to the accompanying drawings. Figure 3 The specific process of battery health state estimation provided by some embodiments of the present application is exemplarily described below.

[0100] Please refer to the accompanying drawings Figure 3 , Figure 3 ​A method flowchart of battery state of health estimation is provided for some embodiments of the present application. The method of battery state of health estimation can comprise: S310, obtaining a real-time charging feature vector in a current charging process; S320, inputting the charging feature vector into a battery state of health evaluation model to obtain a real-time value of the battery state of health.

[0101] For example, in specific embodiments of the present application, the real-time charging feature vector composed of the above-mentioned 17-dimensional features can be obtained by collecting signals during the battery charging process and then performing feature extraction and normalization processing. Load the battery state of health evaluation model trained by the above-mentioned embodiments, input the 17-dimensional features into it, and output the current SOH value (as a specific example of a real-time value) and the decay rate ΔSOH.

[0102] In some embodiments of the present application, the method of battery state of health estimation further comprises: confirming that the real-time value is less than a set threshold, triggering a battery maintenance prompt to inform the relevant personnel.

[0103] For example, in specific embodiments of the present application, if the current SOH value is less than 0.8 (as a specific example of a set threshold), a maintenance prompt is triggered to remind the relevant personnel to maintain the battery. Through real-time prediction of SOH, battery-related strategy customization and battery life prediction can be supported.

[0104] As can be seen from the above some embodiments of the present application, the present application only needs complete charging and discharging data of a single battery, without the need for additional collection of different chemical systems or working conditions, which can reduce the calibration cost and realize convenient acquisition of the training data set; the full connection network has low inference delay and is suitable for embedded real-time operation; the present application does not need complex physical models, but only relies on CC-CV charging statistical features and cascade networks, and the feature extraction and network structure are universal and can be seamlessly migrated to other battery types or larger scale data sets. Therefore, the present application provides a lithium ion battery SOH estimation method with high precision, high robustness and low calibration cost, which is very suitable for fields such as electric vehicles, power grid energy storage, aerospace, etc. that have strict requirements on battery health management, and has high practicality.

[0105] Please refer to Figure 4 , Figure 4 A composition block diagram of a battery state of health estimation device provided by some embodiments of the present application is shown. It should be understood that the battery state of health estimation device corresponds to the above-mentioned method embodiments and can perform each step involved in the above-mentioned method embodiments. The specific functions of the battery state of health estimation device can be referred to the description in the above, and the detailed description is appropriately omitted here to avoid repetition.

[0106] Figure 4The battery health state estimation device includes at least one software function module stored in the form of software or firmware in the memory or solidified in the battery health state estimation device, and the battery health state estimation device includes: a training module 410 configured to input a charging feature vector in a battery charging process into a mapping network, and obtain an initial prediction value of a battery health state; determine a prediction value of the battery health state based on the initial prediction value, the charging feature vector and a dynamics network; wherein the charging feature vector includes a multi-dimensional charging feature and a charging cycle feature; wherein the mapping network and the dynamics network constitute a fusion neural network model; a loss calculation module 420 configured to solve a loss value between the prediction value and an actual value of the battery health state by using an objective function; wherein the objective function includes a data fitting loss of the battery health state, a physical residual loss and a monotonicity change trend loss of the battery health state; and an optimization module 430 configured to optimize model parameters of the fusion neural network model based on the loss value, and obtain a battery health state evaluation model; wherein the battery health state evaluation model is used to predict a real-time value of the battery health state.

[0107] Reference is made to Figure 5 , Figure 5 A component block diagram of the battery health state estimation device provided by some embodiments of the present application is shown. It should be understood that the battery health state estimation device corresponds to the above-mentioned method embodiments, and can perform each step involved in the above-mentioned method embodiments. The specific functions of the battery health state estimation device can be referred to the description in the foregoing, and the detailed description is appropriately omitted here to avoid repetition.

[0108] Figure 5 The battery health state estimation device includes at least one software function module stored in the form of software or firmware in the memory or solidified in the battery health state estimation device, and the battery health state estimation device includes: an acquisition module 510 configured to acquire a real-time charging feature vector in a current charging process; and an estimation module 520 configured to input the charging feature vector into a battery health state evaluation model to obtain a real-time value of a battery health state.

[0109] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described device can refer to the corresponding process in the foregoing method, and will not be described in more detail here.

[0110] Some embodiments of the present application also provide a computer readable storage medium having a computer program stored thereon, and the program is executed by a processor to implement the operations of the method corresponding to any of the above-mentioned embodiments provided by the above-mentioned embodiments.

[0111] Some embodiments of the present application further provide a computer program product, which comprises a computer program, wherein the computer program is used to perform the operations of the method according to any of the above embodiments when executed by a processor.

[0112] As shown in Figure 6 Some embodiments of the present application provide an electronic device 600, which comprises a memory 610, a processor 620 and a computer program stored in the memory 610 and executable on the processor 620, wherein the processor 620 reads the program from the memory 610 through a bus 630 and performs the program to implement the method according to any of the above embodiments.

[0113] The processor 620 can process digital signals and can comprise various computing structures, such as a complex instruction set computer structure, a reduced instruction set computer structure or a structure implementing a combination of multiple instruction sets. In some examples, the processor 620 can be a microprocessor.

[0114] The memory 610 can be used to store instructions executed by the processor 620 or data related to the execution of the instructions. The instructions and / or data can include code for implementing some or all of the functions of one or more modules described in the embodiments of the present application. The processor 620 of the embodiments of the present disclosure can be used to execute the instructions in the memory 610 to implement the method shown above. The memory 610 comprises a dynamic random access memory, a static random access memory, a flash memory, an optical memory or other memories well known to those skilled in the art.

[0115] The above only describes the embodiments of the present application and is not used to limit the protection scope of the present application. For those skilled in the art, the present application can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0116] The above only describes the embodiments of the present application and is not used to limit the protection scope of the present application. For those skilled in the art, the present application can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0117] It is to be noted that, as used in this specification and the appended claims, the singular forms "a," "an," and "the" include plural referents unless the context clearly dictates otherwise. Thus, for example, reference to "a component" can include a combination of two or more components. Additionally, the terms "comprise," "comprises," and "comprising," or any variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements is not necessarily limited to those elements, but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. Furthermore, unless otherwise indicated herein, the terms "first," "second," "third," etc., are used herein merely as labels, and are not intended to impose ordinal import.

Claims

1. A method for estimating the state of health of a battery, characterized in that, include: The charging feature vector during the battery charging process is input into the mapping network to obtain the initial predicted value of the battery health status; wherein, the charging feature vector includes multi-dimensional charging features and charging cycle features; Based on the initial predicted value, the charging feature vector, and the dynamics network, the predicted value of the battery health state is determined; wherein, the mapping network and the dynamics network constitute a fusion neural network model; The loss between the predicted value and the true value of the battery health state is calculated using an objective function; wherein, the objective function includes the data fitting loss of the battery health state, the physical residual loss, and the monotonicity trend loss of the battery health state; The model parameters of the fusion neural network model are optimized based on the loss value to obtain a battery health status assessment model; wherein, the battery health status assessment model is used to predict the real-time value of the battery health status.

2. The method as described in claim 1, characterized in that, Before inputting the charging feature vector during the battery charging process into the mapping network, the method further includes: Select the charging segment used for feature extraction during the battery charging process; Feature extraction is performed on the voltage change curve and current change curve within the charging segment to obtain multidimensional original charging features; The multidimensional original charging characteristics and the number of charging cycles are normalized to obtain the multidimensional charging characteristics and the charging cycle characteristics.

3. The method as described in claim 1 or 2, characterized in that, The step of determining the predicted value of the battery health state based on the initial predicted value, the charging feature vector, and the dynamic network includes: The network input vector is constructed based at least on the initial predicted value and the charging feature vector; The network input vector is input into the dynamic network to obtain the predicted value.

4. The method as described in claim 3, characterized in that, The construction of the network input vector based at least on the initial predicted value and the charging feature vector includes: Fit the differential equation corresponding to the battery's health state during the battery charging process; Differential extraction is performed on the differential equation to obtain the first differential result and the second differential result; The charging feature vector, the initial predicted value, the first differential result, and the second differential result are concatenated to obtain the network input vector.

5. The method as described in claim 3, characterized in that, The mapping network is connected to an attenuation rate network; the method further includes: The initial predicted value is input into the degradation rate network to obtain the changing trend of the battery health status; wherein, the degradation rate network is a fully connected network, a residual prediction network, or an inflection point detection network.

6. The method according to any one of claims 1-2 and 4-5, characterized in that, The objective function is constructed as follows: Set a first coefficient and a second coefficient corresponding to the physical residual loss and the monotonicity trend loss, respectively; The objective function is constructed based on the physical residual loss, the monotonicity trend, the first coefficient, the second coefficient, and the data fitting loss.

7. The method according to any one of claims 1-2 and 4-5, characterized in that, The data fitting loss characterizes the mean square error between the predicted value and the true value; the physical residual loss characterizes the differential residual value. The monotonicity trend loss characterizes the consistency between the predicted trend of the predicted value and the true trend of the true value.

8. A method for estimating the state of health of a battery, characterized in that, include: Obtain the real-time charging feature vector during the current charging process; The charging feature vector is input into the battery health status assessment model to obtain the real-time value of the battery health status; wherein the battery health status assessment model is obtained by performing the method of any one of claims 1-7.

9. The method as described in claim 8, characterized in that, The method further includes: If the real-time value is confirmed to be less than the set threshold, a battery maintenance prompt will be triggered to notify relevant personnel.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein the computer program, when executed by a processor, performs the method as described in any one of claims 1-9.

11. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the computer program is executed by the processor to perform the method as claimed in any one of claims 1-9.

Images