A lithium-ion battery capacity attenuation estimation method, electronic device and storage medium

Abstract

A lithium-ion battery capacity decay estimation method, electronic device and storage medium belong to the technical field of lithium-ion battery capacity decay estimation, and solve the problem of how to improve the estimation accuracy of lithium-ion battery capacity decay. The present invention considers the uncertainty of factors such as temperature, discharge depth, and number of cycles to correct the lithium-ion battery empirical degradation model, constructs a lithium-ion battery capacity decay disturbance model, accurately extracts the health characteristics of battery capacity decay by a data-driven method, and establishes a data-driven error compensation model based on a convolutional neural network. Through the fusion of the battery capacity decay disturbance model and the data-driven error compensation model, evaluation is carried out in a model-driven manner, and evaluation errors are suppressed in a data-driven manner. The capacity decay information of the lithium-ion battery is accurately predicted, and the method also has good robustness and applicability for the capacity decay conditions of different batteries of the same type, providing reliable technical support for the evaluation of different types of batteries.

Description

Technical Field

[0001] The present invention belongs to the technical field of lithium-ion battery capacity attenuation estimation, and relates to a lithium-ion battery capacity attenuation estimation method, electronic equipment and storage medium. Background Art

[0002] With the promotion of social energy conservation, emission reduction and sustainable development policies, new energy electric vehicles with lithium-ion batteries as the main power source have gradually emerged. However, with the increasing demand for power lithium-ion batteries, the problem of capacity decay has become increasingly prominent. Affected by the use environment and conditions, the available capacity of power lithium-ion batteries will drop sharply with the increase in the number of cycles, which in turn affects the life and safety of the battery. During use, the capacity decay of lithium-ion batteries cannot be directly measured. Once the capacity degradation of lithium-ion batteries reaches its service life, it may cause major safety accidents. Therefore, accurately evaluating the capacity decay of power lithium-ion batteries is crucial to ensure the power performance and safety of electric vehicles.

[0003] At present, many domestic research institutions have carried out in-depth research on the capacity decay of lithium-ion batteries. However, due to the inherent degradation characteristics of lithium-ion batteries and the differences in actual working environments, the degradation process may be very different even for the same type of batteries. The uncertainty of the internal failure mechanism and operating conditions (including charge and discharge current, voltage, ambient temperature, etc.) of lithium-ion batteries during actual operation cannot be ignored. The impact on the degradation process cannot be ignored. This makes it more difficult to accurately estimate the battery capacity decay.

[0004] In the prior art, the existing invention patent document "A modeling method for a prediction network for estimating the health status of lithium-ion batteries" with an application publication date of June 23, 2023 and an application publication number of CN116298934A, after building a multivariate time series prediction network and training it, takes the charging part in the battery cycle life test as the entry point from the perspective of feature extraction, and extracts features from its voltage curve, divides the voltage at the same interval, and extracts the time required as a feature, and extracts the closed area under the voltage curve corresponding to the corresponding time as a feature. The two features are used as the final features, which solves the problem that the traditional method can only rely on the feature extraction method of the specific value of the voltage, especially the traditional method extracts fewer features, resulting in insufficient accuracy, robustness and poor applicability; the technical solution of the patent document collects discrete data in the cycle test, so it is impossible to accurately calculate the closed area of ​​the charging voltage curve of different cycles, and the prediction performance of the model is limited.

[0005] The existing invention patent document "A method and system for predicting the capacity decay trend of lithium-ion batteries" with an application publication date of May 16, 2023 and an application publication number of CN116125307A creates a normal decay prediction model based on CNN and LSTM, and trains the normal decay prediction model; creates a capacity proliferation prediction model based on DNN, and trains the capacity proliferation prediction model; generates a capacity decay trend prediction model based on the trained normal decay prediction model and the capacity proliferation prediction model; and uses the capacity decay trend prediction model to predict the capacity decay trend of lithium-ion batteries. Although the technical solution of this patent document does not require an accurate understanding of the battery degradation mechanism, it still requires a large amount of valid data to train the model. In actual applications, historical test data often contains multiple harmonic components and cannot be directly used to build a model. In addition, the data processing and model building processes are separated from each other, and the trained model is only valid for the test battery and cannot be applied to other battery data, and its universality is poor.

[0006] In summary, the existing technology has technical problems that rarely take into account the different capacity attenuation characteristics of lithium-ion batteries under different working conditions, the incompatibility of prediction accuracy and speed, and the low applicability of algorithms for different batteries. Summary of the invention

[0007] The technical solution of the present invention is used to solve the problem of how to improve the estimation accuracy of lithium-ion battery capacity decay.

[0008] The present invention solves the above technical problems through the following technical solutions:

[0009] A method for estimating capacity attenuation of a lithium-ion battery comprises the following steps:

[0010] Step 1: Based on the capacity decay data of lithium-ion batteries collected by offline testing, an empirical decay model between relative capacity decay of lithium-ion batteries and temperature, charge and discharge rate, discharge depth, and cycle number is constructed through feature analysis;

[0011] Step 2: Considering the deviation caused by different battery environments and instrument measurement errors, the uncertain relationship between temperature, discharge rate, discharge depth and lithium-ion battery capacity decay is established, and a lithium-ion battery capacity decay disturbance model is constructed based on the lithium-ion battery capacity decay empirical model;

[0012] Step 3: Select the time difference between constant current charging to the cut-off voltage and the working voltage, the temperature peak during discharge, and the time to discharge to the peak temperature as health characteristics, and perform correlation analysis to build a data-driven error compensation model based on convolutional neural networks, integrate it with the battery capacity attenuation disturbance model, and evaluate the capacity attenuation of lithium-ion batteries.

[0013] Furthermore, the method for constructing an empirical attenuation model of the relative capacity attenuation degree of a lithium-ion battery and the relationship between the temperature, charge and discharge rate, discharge depth, and cycle number through characteristic analysis described in step 1 is specifically as follows:

[0014] 1) The health status of a lithium-ion battery is defined as the ratio of the current capacity of the lithium-ion battery to the nominal capacity of the lithium-ion battery. The calculation formula is: Where C is the number of charge and discharge cycles, Q C is the maximum available capacity under cycle number C, Q intial is the initial capacity of the battery;

[0015] 2) Define the relative capacity decay of lithium-ion batteries as: Q loss =1-SOH;

[0016] 3) Through characteristic analysis, an empirical attenuation model of the relative capacity attenuation of lithium-ion batteries with respect to temperature, charge and discharge rate, discharge depth, and number of cycles is constructed, as follows:

[0017]

[0018] A total =N c ·D oD ·A Rated

[0019] 4) Establish the unknown factor B and charge / discharge rate C before the index is established rate The nonlinear relationship:

[0020] 5) Preprocess the historical capacity data of the battery by using outlier detection and Newton interpolation method, divide the historical capacity data set of the battery into 20 segments on average, each segment of data satisfies the normal distribution, and regard the data exceeding 3 times the standard deviation in each segment of the data set as outliers, and replace the outliers of each segment of lithium-ion battery capacity historical data by Newton interpolation method; the k-order difference quotient in the Newton interpolation method is defined as:

[0021]

[0022] The polynomial is defined as:

[0023] N n (x)=f(x0)+(x-x0)f(x0,x1)+…+(x-x0)(x-x1)(xx n-1 )f(x0,x1,...,x n )

[0024] Among them, x0,x1,…,x k is the interpolation point, f() is the interpolation function;

[0025] 6) Based on the pre-processed historical capacity data, the least square method is used to identify the unknown parameters in the lithium-ion battery attenuation empirical model. Based on the results of parameter identification, the specific expression of the lithium-ion battery capacity attenuation empirical model is obtained as follows:

[0026]

[0027] Where B is the exponential factor related to the frequency and direction of the internal collision of the battery, k1 and k2 are fitting coefficients, T is the temperature, C rate is the charge and discharge rate, D oD is the discharge depth, N c is the number of cycles, z is the power law factor, R g =8.314 J / (mol·K), is the gas constant, A total is the total discharge capacity of the battery, A Rated is the nominal capacity of the battery.

[0028] Furthermore, the method for establishing the uncertain relationship between temperature, discharge rate, discharge depth and lithium-ion battery capacity decay in step 2 and constructing a lithium-ion battery capacity decay disturbance model based on the lithium-ion battery capacity decay empirical model is as follows:

[0029] 1) The normal distribution is used to model the temperature T measurement error, and the variance is used to reflect the influence of the sensor measurement accuracy and its installation position on the temperature T measurement error, that is, ΔT obeys Normal distribution, where ΔT is the measurement error of temperature and ε1 represents the standard deviation of temperature measurement.

[0030] 2) The normal distribution is used to model the charge and discharge rate measurement error, and the variance is used to reflect the influence of the sensor measurement accuracy and its installation position on the discharge rate measurement error, that is, ΔC rate obey The normal distribution of ΔC rate is the measurement error of the discharge rate, ε2 is the measurement standard deviation of the discharge rate;

[0031] 3) Statistical modeling of discharge depth using Fourier spectrum analysis: The extreme points and boundary points of the SoC-t curve are extracted using the rain flow counting method, and the curve is segmented into process vectors S, where S = [S1, S2, ..., S i ], using Fourier series to transform the process vector S i Convert it into a superposition of a corresponding series of sine waves, and record its amplitude matrix as L i,n , where L i,n =F[S i ]=[L i,1 ,Li,2 ,…,L i,n ], with the amplitude matrix L of each segment i,n Equivalent to the D of the battery SoC history curve oD value, and obtain the equivalent discharge depth in

[0032] Furthermore, the method of selecting the time difference between the constant current charging to the cut-off voltage and the working voltage, the temperature peak during the discharge process, and the time of the discharge to the peak temperature as health characteristics and performing correlation analysis in step 3 is as follows:

[0033] 1) The time difference between constant current charging to the cut-off voltage and the working voltage is selected as the health characteristic, and the calculation formula is: Δt cc =t upper -t normal , where Δt cc The time difference between the battery constant current charging to the cut-off voltage and the working voltage, t upper and t normal are the time to charge to the cut-off voltage and the working voltage respectively;

[0034] 2) The temperature peak during the discharge process and the time from discharge to peak temperature are selected as health characteristics. The calculation formula for the time from discharge to peak temperature is: t PT =[t i |T(t i )=T p ]i=1…n,where T p represents the peak temperature; n represents the number of time series;

[0035] 3) Use the Pearson correlation coefficient to verify the correlation between health features and battery capacity decay, and select health features as the input feature sequence of the data-driven error compensation model based on the analysis results;

[0036] The Pearson correlation coefficient ρ p The calculation formula is as follows:

[0037]

[0038] Among them, X i is the health feature sequence, Y i is the battery remaining capacity sequence, and X i With Y i The average value of .

[0039] 4) Use the Spearman correlation coefficient to test the correlation between the selected health characteristics and battery capacity decay;

[0040] The Spearman correlation coefficient ρ s The calculation formula is as follows:

[0041]

[0042] Among them, d λ is the health feature sequence X i and the battery remaining capacity sequence Y i The position difference after descending order, N is the number of sampled sequences; the Spearman correlation coefficient measures the degree of correlation between factors according to the similarity or dissimilarity of the trends between factors; the output range is [-1,1], 0 means no correlation, negative value means negative correlation, and positive value means positive correlation;

[0043] 5) Fill the selected health features into an input feature matrix X of size n×3, middle.

[0044] Furthermore, the method of constructing a data-driven error compensation model based on a convolutional neural network described in step 3 and integrating it with the battery capacity decay disturbance model to evaluate the capacity decay of a lithium-ion battery is as follows:

[0045] 1) A data-driven error compensation model is established using a one-dimensional convolutional neural network. The model consists of three convolutional layers, one pooling layer, and two fully connected layers. The input data is extracted through the convolutional layers using convolution kernels with different weights. The feature extraction calculation formula for the one-dimensional convolution time series is: in, is the feature vector, σ is the activation function, W i k is the weight matrix, * is the convolution operator symbol, V k-1 is the output vector, is the bias, k is the number of layers, and i is the number of data;

[0046] 2) The pooling layer traverses the feature matrix output by the convolutional layer, uses the pooling kernel to filter and compress data elements and retain the most significant information. The calculation formula is: Among them, V i k (m) is the element of the i-th feature matrix of the m-th layer in the pooling area, is the element of the feature matrix after pooling, D w is the wth pooling coverage area;

[0047] 3) The fully connected layer learns the relationship between the feature quantity output by the pooling layer and the battery capacity, integrates the output of the pooling layer and passes the result to the output layer to obtain the battery capacity attenuation error compensation value ΔQ loss .

[0048] 4) The extracted health feature sequence is taken as vector X, and the corresponding battery capacity attenuation is taken as vector Y. The two vectors are divided into a training set and a validation set. The data-driven error compensation model is trained with the training set, and then the model performance is evaluated using the validation set.

[0049] An electronic device includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the above-mentioned lithium-ion battery capacity decay estimation method, and the processor is configured to execute the program stored in the memory.

[0050] A storage medium stores a computer program, which executes the steps of the above-mentioned lithium-ion battery capacity decay estimation method when executed by a processor.

[0051] The advantages of the present invention are:

[0052] The present invention corrects the empirical degradation model of lithium-ion batteries by considering the uncertainties of factors such as temperature, discharge depth, and number of cycles, and constructs a capacity decay disturbance model for lithium-ion batteries; at the same time, a data-driven method is used to accurately extract the health characteristics of battery capacity decay, and a data-driven error compensation model based on a convolutional neural network is established. By integrating the battery capacity decay disturbance model with the data-driven error compensation model, online evaluation is carried out in a model-driven manner, and online evaluation errors are suppressed in a data-driven manner, so that the capacity decay information of lithium-ion batteries is accurately predicted. The method of the present invention can achieve high-precision online estimation of the capacity decay of lithium-ion batteries, and also has good robustness and applicability for the capacity decay conditions of different batteries of the same type, providing reliable technical support for the online evaluation of different types of batteries, and solving the problems in the prior art that the capacity decay characteristics of lithium-ion batteries under different working conditions are very different, the prediction accuracy and speed are incompatible, and the algorithm applicability for different batteries is low. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a flow chart of a lithium-ion battery capacity attenuation estimation method according to Embodiment 1 of the present invention;

[0054] Figure 2 Capacity decay curve of Example 1 of the present invention under measurement errors ε1=0.5 and ε2=0.3;

[0055] Figure 3 Charging voltage curves at different cycle times of Example 1 of the present invention;

[0056] Figure 4 Battery peak temperature variation curve during constant current discharge phase in Example 1 of the present invention;

[0057] Figure 5A diagram of the convolutional neural network architecture of the first embodiment of the present invention;

[0058] Figure 6 A structural diagram of a lithium-ion battery capacity attenuation error compensation model based on a convolutional neural network according to a first embodiment of the present invention;

[0059] Figure 7(a) to Figure 7(d) 7(a) is a graph showing the experimental results of the method of the present invention and other comparative methods, wherein FIG7(b) is a graph showing the experimental results of a B5 battery, FIG7(c) is a graph showing the experimental results of a B7 battery, and FIG7(d) is a graph showing the experimental results of a B18 battery;

[0060] Figure 8(a) to Figure 8(d) 8(a) is a relative error distribution diagram of a B5 battery, FIG8(b) is a relative error distribution diagram of a B6 battery, FIG8(c) is a relative error distribution diagram of a B7 battery, and FIG8(d) is a relative error distribution diagram of a B18 battery;

[0061] FIG9 is a comparison diagram of evaluation indicators of the method of the present invention and other comparative methods, wherein FIG9(a) is a RMSE comparison diagram, and FIG9(b) is a MAE comparison diagram. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0063] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and specific embodiments:

[0064] Embodiment 1

[0065] like Figure 1 As shown, a lithium-ion battery capacity attenuation estimation method designed in an embodiment of the present invention includes the following steps:

[0066] 1. Establish an empirical model for lithium-ion battery capacity attenuation

[0067] 1.1. In view of the different models or different monomers of lithium-ion batteries have different nominal capacities, the SOH (State Of Health) of lithium-ion batteries is defined as the ratio of the current capacity of the lithium-ion battery to the nominal capacity of the lithium-ion battery. The calculation formula is: Where C is the number of charge and discharge cycles, Q C is the maximum available capacity under cycle number C, Q intial is the initial capacity of the battery.

[0068] 1.2. Define the relative capacity attenuation of lithium-ion batteries as: Q loss =1-SOH.

[0069] 1.3. Constructing the relative capacity decay Q of lithium-ion batteries through characteristic analysis loss The empirical attenuation model with temperature, charge and discharge rate, discharge depth and number of cycles is as follows:

[0070]

[0071] A total =N c ·D oD ·A Rated

[0072] Where B is the exponential factor related to the frequency and direction of the internal collision of the battery, k1 and k2 are fitting coefficients, T is the temperature, C rate is the charge and discharge rate, D oD is the discharge depth, N c is the number of cycles, z is the power law factor, R g =8.314 J / (mol·K), is the gas constant, A total is the total discharge capacity of the battery, A Rated is the nominal capacity of the battery.

[0073] 1.4. Establish the unknown factor B and charge / discharge rate C before the index is established rate The nonlinear relationship is:

[0074] 1.5. Based on the original data of battery capacity degradation from NASA PCoE, the historical capacity data of the battery is preprocessed using outlier detection and Newton interpolation method.

[0075] 1.6. Divide the battery's historical capacity data set into 20 segments on average, and each segment of data approximately satisfies the normal distribution.

[0076] 1.7. The data with more than 3 times the standard deviation in each data segment in the data set are regarded as outliers, and the outliers of each segment of lithium-ion battery capacity historical data are replaced by Newton interpolation method;

[0077] The k-order difference quotient in the Newton interpolation method is defined as:

[0078]

[0079] The polynomial is defined as:

[0080] N n (x)=f(x0)+(x-x0)f(x0,x1)+…+(x-x0)(x-x1)(xx n-1 )f(x0,x1,...,x n )

[0081] Among them, x0,x1,…,x k is the interpolation point, and f() is the interpolation function.

[0082] 1.8. Based on the preprocessed historical capacity data, the least squares method is used to identify the unknown parameters in the empirical model of lithium-ion battery degradation.

[0083] 1.9. Based on the results of parameter identification, the specific expression of the empirical model of lithium-ion battery capacity attenuation is as follows:

[0084]

[0085] 2. Constructing a capacity decay disturbance model for lithium-ion batteries

[0086] In view of the low applicability and limited predictive ability of the empirical degradation model of power lithium-ion battery capacity decay in complex practical situations, the uncertainty modeling method is introduced to establish the uncertain relationship between temperature, discharge rate, discharge depth and lithium-ion battery capacity decay, and the dynamic mathematical modeling of various stress factors affecting lithium-ion battery capacity decay is carried out. Considering the comprehensive effects of multiple factors such as temperature, discharge rate, charge and discharge depth, the influence of these factors on the lithium-ion battery capacity decay model is accurately described, and the lithium-ion battery capacity decay disturbance model is constructed based on this, which more accurately describes the dynamic change process of the battery capacity decay process, and provides a more reliable basis for the optimization of the battery management system.

[0087] 2.1. Considering that the battery surface temperature will change during the discharge process of lithium-ion batteries and there are disturbances in sensor measurement, the normal distribution is used to model the temperature T measurement error, and the variance is used to reflect the influence of the sensor measurement accuracy and its installation position on the temperature T measurement error, that is, ΔT obeys Normal distribution, where ΔT is the measurement error of temperature and ε1 represents the standard deviation of temperature measurement.

[0088] 2.2. Considering the measurement disturbance of discharge rate and the large drift of sensor measurement value in long-term observation, the normal distribution is used to model the measurement error of charge and discharge rate, and the variance is used to reflect the influence of sensor measurement accuracy and its installation position on the measurement error of discharge rate, that is, ΔC rate obey The normal distribution of ΔC rateis the measurement error of the discharge rate, and ε2 is the measurement standard deviation of the discharge rate.

[0089] 2.3. According to the principle of battery material fatigue failure, Fourier spectrum analysis is used to statistically model the discharge depth; the extreme points and boundary points of the SoC-t curve are extracted by rain flow counting method, and it is segmented into a history vector S, where S = [S1, S2, ..., S i ], using Fourier series to transform the process vector S i Convert it into a superposition of a corresponding series of sine waves, and record its amplitude matrix as L i,n , where L i,n =F[S i ]=[L i,1 ,L i,2 ,…,L i,n ], with the amplitude matrix L of each segment i,n Equivalent to the D of the battery SoC history curve oD value, and obtain the equivalent discharge depth in

[0090] like Figure 2 As shown, the influence of sensor measurement and environmental factors on the results of the empirical degradation model is considered, the influencing factors of the model are quantitatively analyzed, and the errors caused by local capacity regeneration and sensor measurement are dynamically compensated, so as to realize the correction of the empirical degradation model and improve the stability and accuracy of the model-based method.

[0091] 3. Construct a data-driven error compensation model based on convolutional neural network

[0092] 3.1. The characteristic curve obtained from the charge and discharge history data of each battery is used to extract the time difference between the cut-off voltage and the working voltage, the temperature peak, and the time required to reach the temperature peak as the health characteristics of the battery. Figure 3 As shown in the figure, as the battery ages, the time for the charging voltage to reach the cut-off voltage changes accordingly. The time difference between constant current charging to the cut-off voltage and the working voltage is selected as the health feature, and the calculation formula is: Δt cc =t upper -t normal , where Δt cc The time difference between the battery constant current charging to the cut-off voltage and the working voltage, t upper and t normal are the time to charge to the cut-off voltage and the working voltage respectively. Figure 4 As shown in the figure, as the number of cycles increases, the battery aging degree increases, the battery peak temperature continues to rise, and the time to reach the peak temperature gradually shortens. The temperature peak during the discharge process and the time to discharge to the peak temperature are selected as health characteristics. The calculation formula for the time to discharge to the peak temperature is: tPT =[t i |T(t i )=T p ]i=1…n,where T p represents the peak temperature; n represents the number of time series.

[0093] 3.2. Use the Pearson Correlation Coefficient (PCC) to verify the correlation between the selected features and the battery capacity decay, and select the features as the input feature sequence of the data-driven error compensation model based on the analysis results;

[0094] The Pearson correlation coefficient ρ p The calculation formula is as follows:

[0095]

[0096] Among them, X i is the health feature sequence, Y i is the battery remaining capacity sequence, and X i With Y i The average value of .

[0097] 3.3. Use the Spearman Correlation Coefficient (SCC) to test the correlation between the selected features and battery capacity decay;

[0098] The Spearman correlation coefficient ρ s The calculation formula is as follows:

[0099]

[0100] Among them, d λ is the health feature sequence X i and the battery remaining capacity sequence Y i The position difference after descending order, N is the number of sampled sequences; the Spearman correlation coefficient measures the degree of correlation between factors according to the similarity or dissimilarity of the trends between factors; the output range is [-1,1], 0 means no correlation, negative value means negative correlation, and positive value means positive correlation.

[0101] 3.4. Analyze the correlation coefficients of the extracted health characteristics and determine the health characteristics

[0102] Table 1 Correlation between health characteristics and battery capacity

[0103] <![CDATA[Δt cc ]]> <![CDATA[T p ]]> <![CDATA[t PT ]]> Spearman correlation coefficient 0.9255 0.8901 0.9995 Pearson correlation coefficient 0.8628 0.9353 0.9998

[0104] It can be seen from Table 1 that the correlation coefficients of the extracted health features are all higher than 0.85, indicating that they have a strong correlation with the battery capacity and can be effectively used for online estimation of lithium-ion battery capacity decay.

[0105] 3.5. Fill the selected health features into an input feature matrix X of size n×3. For specific operations, refer to equation

[0106] 3.6, such as Figure 5 and Figure 6 As shown in the figure, a one-dimensional convolutional neural network is used to establish a lithium-ion battery capacity attenuation error compensation model. The model consists of three convolutional layers, one pooling layer, and two fully connected layers. The input data is extracted through the convolutional layer using convolution kernels with different weights to evaluate the matching between different positions in the data and the corresponding features. The feature extraction calculation formula of the one-dimensional convolution time series is: in, is the feature vector, σ is the activation function, W i k is the weight matrix, * is the convolution operator symbol, V k-1 is the output vector, is the bias, k is the number of layers, and i is the number of data.

[0107] 3.7. The pooling layer traverses the feature matrix output by the convolutional layer, uses the pooling kernel to filter and compress data elements and retain the most significant information. The calculation formula is: Among them, V i k (m) is the element of the i-th feature matrix of the m-th layer in the pooling area, is the element of the feature matrix after pooling, D w is the w-th pooling coverage area.

[0108] 3.8. The fully connected layer learns the relationship between the feature quantity output by the pooling layer and the battery capacity, integrates the output of the pooling layer and passes the result to the output layer to obtain the battery capacity attenuation error compensation value ΔQ loss .

[0109] 3.9. The extracted health feature sequence is taken as vector X, and the corresponding battery capacity attenuation is taken as vector Y. The two vectors are divided into a training set and a validation set. The data-driven error compensation model is trained with the training set, and then the model performance is evaluated with the validation set.

[0110] Through in-depth analysis of the inherent laws in the battery charge and discharge history data, three key health characteristics were extracted: the time difference between the cut-off voltage and the working voltage, the temperature peak, and the time to reach the temperature peak. The Pearson correlation coefficient and the Spearman correlation coefficient were used to verify the correlation between the characteristics and the battery capacity decay. Based on this, a data-driven error compensation model based on a convolutional neural network was constructed, which provides strong support for the high-precision online evaluation of battery capacity decay. Under the fusion framework based on model and data drive, the advantages of the capacity decay perturbation model and the data-driven error compensation model are combined to form a closed-loop system, which effectively improves the prediction accuracy and system stability of lithium-ion battery capacity decay, and realizes the technical application of online estimation of lithium-ion battery capacity decay.

[0111] 4. Experimental verification

[0112] The method of the present invention is experimentally verified using the NASA PCoE 18650 lithium-ion battery aging experimental data set, and its evaluation results on different batteries are analyzed.

[0113] The data sets with battery numbers B5, B6, B7 and B18 were selected as the original data, and the data of the first 100 cycles of the B5, B6 and B7 battery packs and the first 70 cycles of the B18 battery pack were selected as training sets, and the remaining data were used as test sets for training the data-driven error compensation model.

[0114] The error modeling distribution is determined through the battery thermal gradient experimental data and the current measurement instrument error data. Combined with the error range provided during the sample data test, the model parameters of the battery capacity attenuation disturbance model are determined.

[0115] The mean absolute error (MAE), mean absolute percentage error (MAPE) and root mean square error (RMSE) were used as performance evaluation indicators.

[0116]

[0117]

[0118] Where N is the total number of prediction samples; and i They are the estimated value and actual value of battery capacity attenuation rate respectively.

[0119] The method (EMR+CNN) of the present invention is experimentally verified by verifying it on the data sets of B5, B6, B7 and B18, and comparing it with the empirical degradation model (EM), the capacity disturbance model (EMR) and the fusion model based on the empirical degradation model and the convolutional neural network (EM+CNN).

[0120] To verify the accuracy and stability of the method of the present invention, the prediction curves of the method of the present invention and the comparative method in the data sets of B5, B6, B7 and B18 are as follows: Figure 7(a) to Figure 7(d) As shown. Taking the B5 data set shown in Figure 7(a) as an example, it can be seen that although the estimation results output by the four models are generally similar to the decay trend of the true value of the battery capacity, the model error effects and stability are quite different. Compared with EDM, EMR, and EM+CNN, the EMR+CNN estimation value well depicts the changing trend of the true capacity both in general and in local details, with a smaller prediction error and a more stable fluctuation range. This is because EMR+CNN takes into account instrument measurement deviations, complex model simplification errors, and model parameter drift, and can dynamically compensate for the prediction results, thereby effectively improving the estimation accuracy of battery capacity decay.

[0121] The relative error distributions of the method of the present invention and the comparative method in the data sets of B5, B6, B7 and B18 are shown in the following figure: Figure 8(a) to Figure 8(d) As shown. Taking the B5 data set shown in Figure 8 (a) as an example, it can be seen that the relative error of the method of the present invention is mostly 4% in the prediction period. In the later stage of prediction, the relative error will increase slightly, but the maximum value is only 6.4515%. Compared with the error distribution curves of single models such as EM and EMR, the data-driven model can dynamically correct the model-driven estimation value by analyzing the inherent laws of historical data. Therefore, the error curve of the fusion model based on model and data drive shows a stable trend within a certain range. Further analysis shows that although the relative error of EM+CNN can also show relatively stable fluctuations in short time series, the relative error increases significantly in the later stage of prediction, and the fluctuation trend is more violent. The error peak can reach 16.8673%, which is much higher than EMR+CNN and can hardly play a compensatory effect. The EMR+CNN curve fluctuates more stably, and the MAPE of the full-cycle prediction is 1.9532%, which is 62.723%, 57.803% and 35.904% lower than that of EM, EMR and EM+CNN models, respectively, further illustrating its superior robustness and accuracy.

[0122] In order to further evaluate the prediction performance of each model, RMSE and MAE are used to compare and analyze the prediction results. The comparison results of the evaluation indicators are as follows: Figure 9(a) to Figure 9(b)As shown, it can be seen from the error values ​​corresponding to the evaluation indicators in the figure that compared with the estimation of a single model, which is easily affected by the inconsistency of battery cells and has unstable estimation performance, the method of the present invention has lower error values ​​than the comparison method on all data sets, and the maximum MAE value does not exceed 0.82939; the maximum RMSE does not exceed 1.0376, and the accuracy of the model estimation is very high. In summary, the online estimation method and system for lithium-ion battery capacity decay based on model and data-driven fusion proposed in the present invention takes into account the complexity of multi-factor coupling in the process of battery capacity decay, and can well describe the influence of actual operating conditions on the capacity decay process. Its prediction effect is close to the actual degradation level of the battery, which fundamentally improves the estimation accuracy of battery capacity decay.

[0123] Embodiment 2

[0124] An electronic device includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the lithium-ion battery capacity decay estimation method in embodiment 1, and the processor is configured to execute the program stored in the memory.

[0125] Embodiment 3

[0126] A storage medium stores a computer program, which, when executed by a processor, executes the steps of the lithium-ion battery capacity attenuation estimation method in embodiment 1.

[0127] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for estimating capacity attenuation of a lithium-ion battery, characterized in that: The following steps are involved: Step 1: Based on the capacity decay data of lithium-ion batteries collected by offline testing, an empirical decay model between relative capacity decay of lithium-ion batteries and temperature, charge and discharge rate, discharge depth, and cycle number is constructed through feature analysis; The attenuation empirical model is specifically as follows: in, B is the exponential factor related to the frequency and direction of internal collisions in the battery, C rate Indicates the charge and discharge rate, k 1 and k 2 is the fitting coefficient, R g is the gas constant, A total is the total discharge capacity of the battery, A Rated is the nominal capacity of the battery, D oD is the discharge depth, N c is the number of cycles, z is the power law factor, T is temperature; Step 2: Considering the deviation caused by different battery environments and instrument measurement errors, the uncertain relationship between temperature, discharge rate, discharge depth and lithium-ion battery capacity decay is established, and a lithium-ion battery capacity decay disturbance model is constructed based on the lithium-ion battery capacity decay empirical model. The specific method is as follows: 1) Use normal distribution to calculate temperature T The measurement error is modeled and the variance is used to reflect the effect of the sensor measurement accuracy and its installation position on the temperature. T The degree of influence of measurement error, that is obey The normal distribution of T is the temperature measurement error, represents the standard deviation of the temperature measurement; 2) Normal distribution is used to model the charge and discharge rate measurement error, and the variance is used to reflect the influence of the sensor measurement accuracy and its installation position on the discharge rate measurement error, that is, obey The normal distribution of C rate is the measurement error of the discharge rate, ε 2 is the measurement standard deviation of the discharge rate; 3) Statistical modeling of discharge depth using Fourier spectrum analysis: Extraction using rain flow counting method SoC - t The extreme points and boundary points of the curve are divided into process vectors S ,in, , using Fourier series to transform the process vector S i Converted into a superposition of a corresponding series of sine waves, the amplitude matrix is ​​recorded as L i,n ,in , with the amplitude matrix of each segment L i,n Equivalent to a battery SoC The course curve D oD value, and obtain the equivalent discharge depth ,in ; Step 3: Select the time difference between constant current charging to the cut-off voltage and the working voltage, the temperature peak during discharge, and the time of discharge to the peak temperature as health characteristics, and perform correlation analysis to build a data-driven error compensation model based on a convolutional neural network, and integrate it with the battery capacity decay disturbance model to form a closed-loop system to evaluate the capacity decay of lithium-ion batteries; The method for constructing a data-driven error compensation model based on a convolutional neural network is specifically as follows: 1) A data-driven error compensation model is established using a one-dimensional convolutional neural network. The model consists of three convolutional layers, one pooling layer, and two fully connected layers. The input data is extracted through the convolutional layers using convolution kernels with different weights. The feature extraction calculation formula for the one-dimensional convolutional time series is: ;in, is the feature vector, σ is the activation function, is the weight matrix, * is the convolution operator symbol, is the output vector, is the bias, k is the number of layers, i is the number of data; 2) The pooling layer traverses the feature matrix output by the convolutional layer, uses the pooling kernel to filter and compress data elements and retain the most significant information. The calculation formula is: ;in, is the first m Layer i The elements of the feature matrix, is the element of the feature matrix after pooling, D w It is w Pooling coverage area; 3) The fully connected layer learns the relationship between the feature quantity output by the pooling layer and the battery capacity, integrates the output of the pooling layer and passes the result to the output layer to obtain the battery capacity attenuation error compensation value .

2. The lithium-ion battery capacity attenuation estimation method according to claim 1, characterized in that: The method of selecting the time difference between constant current charging to the cut-off voltage and the working voltage, the temperature peak during the discharge process, and the time of discharge to the peak temperature as health characteristics and performing correlation analysis described in step 3 is as follows: 1) The time difference between constant current charging to the cut-off voltage and the working voltage is selected as the health characteristic, and the calculation formula is: , where Δ t cc The time difference between the battery constant current charging to the cut-off voltage and the working voltage, t upper and t normal are the time to charge to the cut-off voltage and the working voltage respectively; 2) The temperature peak during the discharge process and the time from discharge to peak temperature are selected as health characteristics. The calculation formula for the time from discharge to peak temperature is: ,in, T p represents the peak temperature; n Indicates the number of time series; 3) Use the Pearson correlation coefficient to verify the correlation between health features and battery capacity decay, and select health features as the input feature sequence of the data-driven error compensation model based on the analysis results; 4) Use the Spearman correlation coefficient to test the correlation between the selected health characteristics and battery capacity decay; 5) Fill the selected health characteristics into a scale of The input feature matrix X， middle.

3. The lithium-ion battery capacity attenuation estimation method according to claim 2, characterized in that: The Pearson correlation coefficient described in step 3 The calculation formula is as follows: in, X i is the sequence of health characteristics, Y i is the battery remaining capacity sequence, and They are X i and Y i The average value of .

4. The lithium-ion battery capacity attenuation estimation method according to claim 2, characterized in that: The Spearman correlation coefficient described in step 3 The calculation formula is as follows: in, d λ Health feature sequence X i Sequence with remaining battery capacity Y i According to the position difference after descending order, N is the number of sampled sequences; the Spearman correlation coefficient measures the degree of correlation between factors based on the similarity or dissimilarity of the trends between factors; the output range is [-1,1], 0 means no correlation, negative values ​​mean negative correlation, and positive values ​​mean positive correlation.

5. The lithium-ion battery capacity attenuation estimation method according to claim 1, characterized in that: The method for evaluating the capacity decay of lithium-ion batteries described in step 3 is as follows: The extracted health feature sequence is used as a vector X , the corresponding battery capacity attenuation is taken as the vector Y , the two vectors are divided into training set and validation set, the data-driven error compensation model is trained by the training set, and then the model performance is evaluated by the validation set.

6. An electronic device, comprising a memory and a processor, characterized in that: The memory is used to store a program that supports the processor to execute the lithium-ion battery capacity decay estimation method according to any one of claims 1 to 5, and the processor is configured to execute the program stored in the memory.

7. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the lithium-ion battery capacity decay estimation method according to any one of claims 1 to 5 are executed.

Images