Lithium ion battery health state estimation method based on label distribution smoothness

Abstract

The invention provides a lithium ion battery health state estimation method based on label distribution smoothness, and the method comprises the steps: collecting a process quantity signal and a state quantity signal of a charging process of a lithium ion battery, and obtaining a health state quantity of the battery; dividing all normalized data into a training set, a verification set and a test set; establishing a convolutional neural network model, taking the health state value as a label, and outputting an estimated health state value by the model; obtaining a final estimation error, and carrying out back propagation on the estimation error to optimize the convolutional neural network model; and verifying the effect of the optimized convolutional neural network model, and outputting a health state estimation result, thereby realizing the estimation of the health state of the lithium ion battery under the condition of data long tail distribution. According to the invention, the estimation error caused by the data imbalance problem in the regression task in the field of the lithium ion battery can be effectively reduced, and the accurate estimation of the health state of the battery is realized.

Description

Technical Field

[0001] The present disclosure belongs to the field of lithium-ion battery health status assessment, and in particular relates to a lithium-ion battery health status estimation method based on label distribution smoothing. Background Art

[0002] In recent years, great progress has been made in the estimation of the health state of lithium-ion batteries based on deep learning. However, due to factors such as the different electrochemical composition of lithium-ion batteries, changes in the use environment, and differences in battery aging experimental methods, the degradation process of some batteries shows a long early degradation time and a short late degradation time, resulting in a long-tail distribution of degradation data. This leads to the problem of data imbalance in the battery health state estimation task, especially when the battery degradation data is limited. This is unacceptable for accurately estimating the health state of lithium-ion batteries, monitoring their real-time usage status, and preventing potential failures, especially in the late stage of degradation. Therefore, a model that can consider the unbalanced regression problem in battery health state estimation is needed.

[0003] The above information disclosed in this Background section is only for enhancement of understanding of the background of the invention and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the invention

[0004] In view of the deficiencies in the prior art, the purpose of the present invention is to provide a method for estimating the health status of lithium-ion battery long-tail distribution data based on label distribution smoothing. This method uses quadratic interpolation and normalization methods to process battery data to ensure the consistency of data features. By performing discrete analysis on continuous problems, the label distribution smoothing method is applied to the convolutional neural network model to analyze and solve the unbalanced regression problem caused by the long-tail distribution of data during the aging process of lithium-ion batteries. Through the convolution processing of the kernel function and the label distribution, the weights corresponding to different labels are calculated, which improves the estimation accuracy.

[0005] To achieve the above objectives, the present disclosure provides the following technical solutions:

[0006] A method for estimating the health status of lithium-ion battery long-tail distribution data based on label distribution smoothing includes the following steps:

[0007] S100: collecting process quantity signals of the lithium-ion battery charging process and state quantity signals representing the health of the battery during the battery aging process, and further obtaining the health state quantity of the battery;

[0008] S200: resampling the process quantity signal, and further normalizing the resampled process quantity signal; and dividing all normalized data into a training set, a validation set, and a test set;

[0009] S300: establishing a convolutional neural network model, taking the health status value as a label, and outputting the model as an estimated health status value;

[0010] S400: Based on the label and the estimated health status value, a discretization analysis and label distribution smoothing method are introduced to obtain a final estimation error, and the estimation error is back-propagated to optimize the convolutional neural network model;

[0011] S500: Use the validation set to verify the effect of the optimized convolutional neural network model, and finally use the model to output the health status estimation result to achieve the estimation of the health status of lithium-ion batteries under the condition of long-tail distribution of data.

[0012] Preferably, in step S100, the process quantity signal includes: the current signal recorded during the charging process , Charging voltage signal and charging temperature signal , and their corresponding lengths are .

[0013] Preferably, in step S100, the state quantity signal includes: the maximum discharge capacity of the battery in each cycle during the aging process is , the battery health status value is expressed as:

[0014]

[0015] in, Indicates the current cycle number. is the nominal capacity of the battery.

[0016] Preferably, in step S200, the process quantity signal is resampled and the length is set to 256; the normalization process adopts the maximum-minimum normalization method, which is expressed as:

[0017]

[0018] in, Represents the process quantity signal after resampling, and Respectively represent the maximum and minimum values ​​of the charging process data, Represents the normalized charging process data; all data are divided into training set, validation set and test set; for the training set ,in Indicates input, Indicates the label, represents the sample size; The set is defined as , is a continuous label space;

[0019] Preferably, step S300 includes the following sub-steps:

[0020] S301: Build a convolutional neural network model. Each layer of the convolutional neural network consists of a one-dimensional convolution layer, a batch normalization layer, and an activation function, for a total of six layers. The one-dimensional convolution layer is used to extract local features of the input data. The batch normalization layer normalizes each batch of input data to reduce the problem of gradient disappearance or explosion and stabilize the training process. The activation function is used to introduce nonlinearity so that the neural network can learn and represent complex nonlinear relationships. After these six layers, a pooling layer is followed to perform a pooling operation on the data to reduce the dimension of the data while retaining important features, reducing the amount of calculation and the risk of overfitting. Finally, a fully connected layer is used for output. Figure 7 shown.

[0021] Step S400 includes the following steps:

[0022] S401: Calculate the mean absolute error (MAE) between the output value and the label. loss);

[0023] S402: Continuous label space Divided into groups, each with equal spacing ; Denote the interval space as ;

[0024] S403: Given degradation data of a battery with a continuous label distribution, a label distribution smoothing method convolves the continuous label distribution with a symmetric kernel function to obtain a smoothed version of the label distribution, referred to as a smoothed label distribution;

[0025] S404: Based on the label distribution smoothing method, each label and its adjacent label interval are considered; for each label , when the tag Belong to the interval When , the corresponding label density for , get each label The corresponding weight After that, the final loss is obtained .

[0026] A computer storage medium includes computer instructions, and when the computer instructions are executed on the computer, the computer executes the above method.

[0027] An electronic device comprises: a memory, a processor and a computer program stored in the memory and executable on the processor; wherein the above method is implemented when the processor executes the program.

[0028] Compared with the prior art, the beneficial effects brought about by the present disclosure are: revealing the unbalanced regression problem in the field of lithium-ion battery health status estimation, and applying discretization analysis and label distribution smoothing methods at the same time, so as to achieve full attention to fewer labels and improve the estimation accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 is a step diagram of a method for estimating the health status of a lithium-ion battery based on label distribution smoothing according to an embodiment of the present disclosure;

[0030] Figures 2 to 5 It is a curve diagram of voltage, current, temperature data of a lithium-ion battery during charging and the maximum discharge capacity of the battery in each cycle during aging according to an embodiment of the present disclosure;

[0031] Figure 6 is a schematic diagram of a discretization analysis of lithium-ion battery degradation data according to an embodiment of the present disclosure;

[0032] Figure 7 is a schematic diagram of a neural network structure used in one embodiment of the present disclosure;

[0033] Figure 8 is a schematic diagram of a label distribution smoothing method according to an embodiment of the present disclosure;

[0034] Fig. 9 It is a full process diagram of a lithium-ion battery health status estimation method based on label distribution smoothing provided by an embodiment of the present disclosure. DETAILED DESCRIPTION

[0035] The following will refer to the attached Figures 1 to 9 Specific embodiments of the present disclosure are described in detail. Although specific embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0036] It should be noted that certain words are used in the specification and claims to refer to specific components. Those skilled in the art should understand that technicians may use different nouns to refer to the same component. This specification and claims do not use the difference in nouns as a way to distinguish components, but use the difference in the functions of the components as the criterion for distinction. As mentioned throughout the specification and claims, "including" or "comprising" is an open term, so it should be interpreted as "including but not limited to". The subsequent description of the specification is a preferred embodiment of the present invention, but the description is based on the general principles of the specification and is not used to limit the scope of the present invention. The scope of protection of the present disclosure shall be determined by the attached claims.

[0037] To facilitate understanding of the embodiments of the present disclosure, further explanation will be given below using specific embodiments as examples in conjunction with the accompanying drawings, and each of the accompanying drawings does not constitute a limitation on the embodiments of the present disclosure.

[0038] In one embodiment, if Figure 1 As shown, the present disclosure provides a method for estimating the health status of a lithium-ion battery based on label distribution smoothing, comprising the following steps:

[0039] S100: Collect process quantity signals of the lithium-ion battery charging process under different working conditions, including voltage, current and temperature, and the maximum discharge capacity of the battery in each cycle during the aging process; and further obtain the health status value of the battery;

[0040] In this step, the process quantity signal includes: the current signal recorded during the charging process , Charging voltage signal and charging temperature signal , and their corresponding lengths are .

[0041] Figures 2 to 5 They are respectively the voltage, current, and temperature data during the charging process of a lithium-ion battery and the curve diagram of the maximum discharge capacity of the battery in each cycle during the aging process.

[0042] use Represents the charging process quantity signal, where Indicates the current cycle number, then x i = [ C V T ] ， x i ∈ ℝ 3 × l The maximum discharge capacity of the battery in each cycle during the aging process is ,Furthermore, the health status value of the battery is expressed as:

[0043]

[0044] in, is the nominal capacity of the battery.

[0045] During the signal acquisition process, the process quantity signals collected in different cycles are different. As the battery ages, the maximum discharge capacity of lithium-ion batteries in different processes will gradually decrease.

[0046] S200: resampling the charging process data to achieve a uniform length, further normalizing the resampled data to eliminate the dimensional differences between features; and dividing all the normalized data into a training set, a validation set, and a test set;

[0047] In this step, the charging process data is resampled and the length is set to 256, that is: x i ' = [ C V T ] ， x ∈ ℝ 3 × 2 5 6 , which is long enough to express the main features of the battery charging curve without increasing too much computational burden; the normalization process adopts the maximum-minimum normalization method, which is expressed as:

[0048]

[0049] in, Represents the charging process data after resampling, and Respectively represent the maximum and minimum values ​​of the charging process data, Represents the normalized charging process data.

[0050] All data are divided into training set, validation set and test set in the ratio of n:1:1, where n depends on the total number of samples; for the training set ,in Indicates input, Indicates the label, represents the sample size; The set is defined as , is a continuous label space.

[0051] S300: Establish a convolutional neural network model, take the processed charging process data as input, take the health state value as a label, and output the model as an estimated health state value;

[0052] S400: Calculate the mean absolute value error between the output value and the label, introduce discretization analysis and label distribution smoothing method, obtain the label weight vector, multiply it with the above error, and then average it to obtain the final estimated error, and back-propagate the estimated error to optimize the convolutional neural network model;

[0053] The maximum number of traversals is set to 200 and the initial learning rate is set to , adjust the learning rate every 25 cycles, reduce the learning rate to 80% of the original, and complete the model training when the training error does not decrease for more than 50 consecutive traversals or the traversal times reach the maximum traversal times;

[0054] S500: The optimized convolutional neural network model is verified using the validation set. When the mean absolute error and root mean square error of the model on the validation set are both lower than 0.05, the model is considered to meet the requirements. Finally, the model is used on the test set to output the health status estimation result, thus realizing the estimation of the health status of lithium-ion batteries under the condition of long-tail data distribution.

[0055] If the verification fails, manually adjust the initial learning rate, learning rate decay strategy, neural network layer input and output and other hyperparameters, and retrain the model until the verification conditions are met.

[0056] In another embodiment, step S300 includes the following steps:

[0057] S301: Build a convolutional neural network model. Each layer of the convolutional neural network consists of a one-dimensional convolution layer, a batch normalization layer, and an activation function, for a total of six layers. The one-dimensional convolution layer is used to extract local features of the input data. The batch normalization layer normalizes each batch of input data to reduce the problem of gradient disappearance or explosion and stabilize the training process. The activation function is used to introduce nonlinearity so that the neural network can learn and represent complex nonlinear relationships. After these six layers, a pooling layer is followed to perform a pooling operation on the data to reduce the dimension of the data while retaining important features, reducing the amount of calculation and the risk of overfitting. Finally, a fully connected layer is used for output. Figure 7 shown.

[0058] Step S400 includes the following steps:

[0059] S401: Calculate the mean absolute error (MAE) between the output value and the label. loss), The loss is calculated as:

[0060]

[0061] in, is the true value, is an estimate;

[0062] S402: Continuous label space Divided into groups, each with equal spacing ; Denote the interval space as , in this way, each label in the original continuous space can be Classify into their corresponding groups;

[0063] In the field of battery health state estimation, the main focus is on the stage before the health state drops to 80%, that is, the interval (1, 0.8], so for each label , which belongs to a related interval , expressed as:

[0064] t j ≜ ( R o u n d ( y i ) + 1 − 8 0 % N , R o u n d ( y i ) ]

[0065] in, The function ensures that the specified number of decimal places is retained in the label; therefore, the label space is divided into multiple intervals within (1, 0.8], that is: ( 1 , 1 − Δ n ] , ( 1 − Δ n , 1 − 2 Δ n ] , ⋯ , ( 0 . 8 + Δ n , 0 . 8 ] ;

[0066] S403: Given degradation data of a battery with a continuous label distribution, a label distribution smoothing method convolves the continuous label distribution with a symmetric kernel function to obtain a smoothed version of the label distribution, referred to as a smoothed label distribution;

[0067] The label distribution smoothing method fully exploits the similarity of adjacent labels within the label interval, thereby improving the performance of the model in unbalanced regression tasks; the smooth label distribution intuitively reflects the continuous relationship and similarity between adjacent labels;

[0068] Based on the above discretization process, the calculation formula of the label distribution smoothing method is expressed as:

[0069]

[0070] in, represents the symmetric kernel function, Indicates the interval number, Representation interval The number of labels in Representation interval The density of labels in ;

[0071] S404: Based on the label distribution smoothing method, each label and its adjacent label interval are considered; for each label , when the tag Belong to the interval When , the corresponding label density for , further, get each label The corresponding weight , the calculation formula is expressed as:

[0072]

[0073] Then obtain a The smoothing weight set , after normalization, Loss Weighted to get the final loss , the normalization method is the maximum normalization method, expressed as:

[0074]

[0075] Final loss The calculation formula is:

[0076]

[0077] For the above steps, combined with the content mentioned above, the degradation process of the battery shows that the early degradation time is longer and the later degradation time is shorter, resulting in a long-tail distribution of the degradation data, which leads to the problem of data imbalance in the battery health status estimation task.

[0078] By adopting the training method of this embodiment, the estimation error caused by the data imbalance problem in the regression task in the field of lithium-ion batteries can be reduced, and accurate estimation of the battery health status can be achieved.

[0079] Figure 6 This is a schematic diagram of the discretization analysis of lithium-ion battery degradation data. Figure 6 As shown, since discrete data can be more effectively represented and visualized through charts and graphs, it provides a more intuitive means to observe and analyze data trends and characteristics. In order to simplify the data processing and analysis procedures, the discretization analysis is incorporated into the battery degradation curve by performing interval division. The degradation process of the battery from the initial state (100% healthy state) to the end of life state (80% healthy state) is divided into multiple different intervals, and different intervals contain different numbers of cycles. Through discretization analysis, the long-tail distribution of data during the battery aging process can be more intuitively observed.

[0080] Figure 8 is a schematic diagram of the label distribution smoothing method, which is Figure 6 Further processing based on Figure 8 As shown in the figure, by dividing the continuous labels into different intervals, the label density of each interval is different, so the weight of each label is different, so the weight vector can be obtained. After convolution operation with the symmetric kernel function, a smooth vector can be obtained, and then a smoothed label density set can be obtained. Through further reciprocal operation and normalization, the weight corresponding to each label can be obtained, and the final error can be obtained by multiplying it with the standard mean absolute error or root mean square error, and it can be used for model optimization.

[0081] Fig. 9It is a full process diagram of the lithium-ion battery health state estimation method based on label distribution smoothing, which is a summary of the above process.

[0082] Below, by conducting a comparative experiment between a convolutional neural network model that incorporates discretization analysis and label distribution smoothing methods and an ordinary convolutional neural network model without the above methods, the beneficial effects of the technical solution of the present disclosure are further demonstrated.

[0083] Specifically, the main evaluation indicators used in this comparative experiment are root mean square error (RMSE) and mean absolute error (MAE).

[0084] The data used in the experiment are data collected from two lithium battery data sets under different charging and discharging strategies (operating conditions). In data set 1, the battery takes one operating condition, with a total of 3 batteries; in data set 2, the battery takes three operating conditions, with 3 batteries in each operating condition, with a total of 9 batteries, as shown in Table 1.

[0085] Table 1

[0086]

[0087] In the table, 7C-4.8C-4.8C-3.652C means charging from 0 to 20% at 7C, charging from 20% to 40% at 4.8C, charging from 40% to 60% at 4.8C, and charging from 60% to 80% at 3.652C. Each charging strategy ensures that the total time from 0 to 80% is 10 minutes. Finally, the constant voltage and constant current mode is used to charge from 80% to 100%.

[0088] To ensure the fairness of the experiment, all methods used the same hyperparameter settings. The prediction accuracy of all methods is shown in Table 2.

[0089] Table 2

[0090]

[0091] In the table, all loss data are magnified for easy observation and comparison. times, that is, the actual loss should be the loss in the table Bold and ↓ indicate that the loss is lower than that of the ordinary convolutional neural network model.

[0092] It can be seen that after applying the label distribution smoothing (LDS) method, the accuracy of the original convolutional neural network (CNN) model in estimating the battery health status has been generally improved.

[0093] The present invention also provides a computer storage medium, which includes computer instructions. When the computer instructions are executed on the computer, the computer executes the above method.

[0094] An electronic device comprises: a memory, a processor and a computer program stored in the memory and executable on the processor; wherein the above method is implemented when the processor executes the program.

[0095] The above specific embodiments are merely illustrative and instructive, and are not restrictive. A person skilled in the art may make many different forms based on the guidance of this specification and without departing from the scope of protection of the claims of this disclosure, all of which are protected by this disclosure.

Claims

1. A method for estimating the health status of lithium-ion batteries based on label distribution smoothing, wherein: The method comprises the following steps: S100: collecting process quantity signals of the lithium-ion battery charging process and state quantity signals representing the health of the battery during the battery aging process, and further obtaining the health state value of the battery; S200: resampling the process quantity signal, and further normalizing the resampled process quantity signal; and dividing all normalized data into a training set, a validation set, and a test set; S300: establishing a convolutional neural network model, taking the health status value as a label, and the model outputting an estimated health status value; S400: Based on the label and the estimated health status value, a discretization analysis and label distribution smoothing method are introduced to obtain a final estimation error, and the estimation error is back-propagated to optimize the convolutional neural network model; S500: Use the validation set to verify the effect of the optimized convolutional neural network model, and finally use the model to output the health status estimation result to achieve the estimation of the health status of the lithium-ion battery under the condition of long-tail distribution of data.

2. The method according to claim 1, wherein: Preferably, in step S100, the process quantity signal includes: the current signal recorded during the charging process , Charging voltage signal and charging temperature signal , and their corresponding lengths are .

3. The method according to claim 1, wherein: In step S100, the state quantity signal includes: the maximum discharge capacity of the battery in each cycle during the aging process , the battery health status value is expressed as: ； in, Indicates the current cycle number. is the nominal capacity of the battery.

4. The method according to claim 1, wherein: In step S200, the process quantity signal is resampled and the length is set to 256.

5. The method according to claim 1, wherein: Step S400 includes: S401: Calculate the mean absolute error (MAE) between the output value and the label; S402: Continuous label space Divided into groups, each with equal spacing ; The interval space is represented by ; S403: Given degradation data of a battery with a continuous label distribution, a label distribution smoothing method convolves the continuous label distribution with a symmetric kernel function to obtain a smoothed version of the label distribution, referred to as a smoothed label distribution; S404: Based on the label distribution smoothing method, each label and its adjacent label interval are considered; for each label , when the tag Belong to the interval When , the corresponding label density for , get each label The corresponding weight After that, the final loss is obtained .

6. A computer storage medium, wherein: The storage medium includes computer instructions, which, when executed on a computer, enable the computer to execute the method according to any one of claims 1 to 5.

7. An electronic device, wherein: The electronic device comprises: a memory, a processor and a computer program stored in the memory and executable on the processor; wherein the processor implements any one of the methods of claims 1-5 when executing the program.

Images