Lithium-ion battery health state estimation method and system based on multi-source fragment information

Abstract

The present invention discloses a lithium ion battery health state estimation method and system based on multi-source fragment information, belonging to the technical field of lithium ion batteries. The method comprises the following steps: selecting a random fragment SOC interval in a constant current charging stage of each charge and discharge cycle of the lithium ion battery, obtaining information of the fragment SOC interval, performing incremental capacity analysis, and taking the variance and mean difference of the charging capacity of the corresponding SOC interval as electrical aging characteristics; randomly selecting several stages in all charge and discharge cycles of the lithium ion battery, and performing EIS test to obtain impedance information; using DRT test to analyze the impedance information, and taking the relaxation function peak value within a set time period as the impedance aging characteristic; constructing a HAMN health state estimation model, and using the HAMN health state estimation model to estimate the battery health state of an unknown lithium ion battery; the present invention can avoid information loss, thereby improving the accuracy of health state assessment of lithium ion batteries.

Description

Technical Field

[0001] The present invention belongs to the technical field of lithium-ion batteries, and in particular relates to a method and system for estimating the health status of lithium-ion batteries based on multi-source fragment information. Background Art

[0002] As the main energy source for many mobile devices and electric vehicles today, accurate assessment of the health status of lithium-ion batteries is critical to ensuring system safety, performance reliability, and service life. Using fragment data as battery aging data is a mainstream solution for estimating the health status of lithium-ion batteries in a data-driven way to deeply simulate the charging and discharging conditions of electric vehicles.

[0003] However, the existing fragmented data estimation schemes on the market weaken the expression of electrical aging information due to their fragmented acquisition method and have no means to suppress noise, which reduces the estimation accuracy of the existing model; in addition, the data in the constant current charging stage is always incomplete, because the battery SOC range is not fixed during actual charging, and the charging range endpoints are usually not at 0% or 100%. Therefore, the charging data measured by the electric vehicle battery management system in the prior art is always a partial charging fragment, but the use of fragmented information will lead to information missing, thereby affecting the accuracy of health status assessment. Summary of the invention

[0004] The purpose of the present invention is to overcome the deficiencies in the prior art and to provide a method and system for estimating the health status of a lithium-ion battery based on multi-source fragment information. The present invention can avoid information loss and thus improve the accuracy of evaluating the health status of a lithium-ion battery.

[0005] The present invention provides the following technical solutions:

[0006] In a first aspect, a method for estimating the health status of a lithium-ion battery based on multi-source fragment information is provided, comprising:

[0007] Based on the actual working conditions of the simulated lithium-ion battery, a random segment SOC interval is selected in the constant current charging stage of each charge and discharge cycle of the lithium-ion battery to obtain the voltage, capacity and health status information of the segment SOC interval;

[0008] According to the voltage, capacity and health status information of the segmented SOC interval, the incremental capacity analysis is performed on each segmented SOC interval, and the variance and average difference of the charging capacity in the corresponding SOC interval are used as electrical aging characteristics;

[0009] Based on the simulated actual working conditions of lithium-ion batteries, several stages are randomly selected in all charge and discharge cycles of lithium-ion batteries and EIS tests are performed to obtain impedance information and the frequency of EIS tests;

[0010] The acquired impedance information is analyzed using DRT test, and the peak value of relaxation function within the set time period is used as the impedance aging feature;

[0011] Based on electrical aging characteristics, impedance aging characteristics and health status information, a HAMN health status estimation model is constructed, wherein the HAMN health status estimation model is based on a deep neural network and introduces a hybrid attention mechanism;

[0012] Based on the electrical aging characteristics and impedance aging characteristics of unknown lithium-ion batteries, the battery health state estimation model of unknown lithium-ion batteries is used to estimate the battery health state.

[0013] Optionally, the method of analyzing the acquired impedance information by using the DRT test and taking the peak value of the relaxation function within a set time period as the impedance aging feature includes:

[0014] Based on the acquired impedance information, Fourier transform is performed on the complex impedance to obtain the frequency domain of the complex impedance; the impedance information is the complex impedance;

[0015] Based on the frequency of the EIS test, construct the Fourier frequency;

[0016] Construct the Tikhonov regularization matrix according to the Fourier frequency and the frequency of the EIS test;

[0017] Perform inverse Fourier transform on the Tikhonov regularization matrix and the frequency domain of the complex impedance to determine the relaxation time and relaxation distribution function;

[0018] The peak value of the relaxation distribution function within 0.01s-10s is taken as the impedance aging feature.

[0019] Optionally, the relaxation time τ is:

[0020]

[0021] Among them, G m,f is the frequency domain of complex impedance, Tikhonov is the Tikhonov regularization matrix, N is the impedance spectrum sequence length, j is the imaginary unit, n is the time domain sequence variable, f is the frequency domain sequence variable; Im(.) is the imaginary part;

[0022] The relaxation distribution function γ(τ) is:

[0023]

[0024] Among them, Rm(.) is the real part;

[0025] The impedance aging characteristic peak m for:

[0026]

[0027] Wherein, m is the number of EIS tests performed.

[0028] Optionally, before constructing the HAMN health state estimation model based on the electrical aging features, the impedance aging features and the health state information, the impedance aging features need to be copied according to the ratio of the feature numbers so as to align the electrical aging features with the impedance aging features.

[0029] Optionally, the replicating the impedance aging characteristics according to the ratio of the characteristic numbers includes:

[0030] Calculate the number of feature reuses num:

[0031]

[0032] Where total is the total number of charge and discharge cycles, and T is the sampling number of the EIS test;

[0033] To perform feature replication on the impedance aging feature:

[0034]

[0035] Wherein, peak is the impedance aging characteristic after feature alignment; peak m Impedance aging characteristics extracted from the mth EIS test.

[0036] Optionally, based on the acquired voltage, capacity and health status information, an incremental capacity analysis is performed on each SOC interval, and the variance and average difference of the charging capacity in the corresponding SOC interval are used as electrical aging features.

[0037] The charge capacity variance Var of the SOC interval of the i-th charge and discharge cycle i for:

[0038]

[0039] The average difference of charging capacity MDA in the SOC interval of the i-th charge and discharge cycle i for:

[0040]

[0041] Among them, u is the sampling point number of the segment SOC interval, q is the sampling time when the segment SOC interval ends, k is the sampling time when the segment SOC interval starts, IC i,u is the IC function value of the u-th sampling point in the i-th charge-discharge cycle, IC i,mean is the mean value of the IC curve of the i-th charge and discharge cycle.

[0042] Optionally, a HAMN health status estimation model is constructed based on electrical aging characteristics, impedance aging characteristics and health status information; the HAMN health status estimation model is based on a deep neural network and introduces a hybrid attention mechanism, wherein the hybrid attention mechanism includes a local attention mechanism and a global attention mechanism; the deep neural network on which the HAMN health status estimation model is based is a fully connected neural network.

[0043] Optionally, the HAMN health state estimation model is constructed based on the electrical aging characteristics, the impedance aging characteristics and the health state information; the HAMN health state estimation model is based on a deep neural network and introduces a hybrid attention mechanism.

[0044] The HAMN health status estimation model also introduces residual connection and position encoding;

[0045] The calculation method of the residual connection is:

[0046] F(input)=f(input)+input

[0047]

[0048] Among them, input is the input, f(·) is the structure of the deep network structure layer; F(input) is the output after the deep network structure layer; F out (input) is the final output of the residual connection, μ F(input) is the mean of the residual connection output data, σ F(input) The standard deviation of the residual connection output data.

[0049] Optionally, the position code is calculated as follows:

[0050] PE pso,2×index =sin(pso / 10000 2×index / d )

[0051] PE pso,2×index+1 =cos(pso / 10000 2×index+1 / d )

[0052] Among them, pso is the position of the data in the sequence, index is the number of all positions in the sequence, and d is the dimension of the encoding vector; PE pso,2×index is the element of even dimension 2×index in the position encoding matrix, PE pso,2×index+1 It is the element of odd dimension 2×index+1 in the position encoding matrix.

[0053] In a second aspect, a lithium-ion battery health status estimation system based on multi-source fragment information is provided, comprising:

[0054] The first acquisition module is used to select a random segment SOC interval in the constant current charging stage of each charge and discharge cycle of the lithium-ion battery based on the simulated actual working condition of the lithium-ion battery, and obtain the voltage, capacity and health status information of the segment SOC interval;

[0055] The electrical aging feature extraction module is used to perform incremental capacity analysis on each SOC interval according to the voltage, capacity and health status information of the SOC interval, and use the variance and mean difference of the charging capacity of the corresponding SOC interval as the electrical aging feature;

[0056] The second acquisition module is used to randomly select several stages in all charge and discharge cycles of the lithium-ion battery based on the simulated actual working conditions of the lithium-ion battery, and perform EIS tests to obtain impedance information and the frequency of the EIS test;

[0057] The impedance aging feature extraction module is used to analyze the acquired impedance information using the DRT test and use the relaxation function peak value within a set time period as the impedance aging feature;

[0058] A construction module is used to construct a HAMN health state estimation model based on electrical aging characteristics, impedance aging characteristics and health state information, wherein the HAMN health state estimation model is based on a deep neural network and introduces a hybrid attention mechanism;

[0059] The estimation module is used to estimate the battery health state of the unknown lithium-ion battery based on the electrical aging characteristics and impedance aging characteristics of the unknown lithium-ion battery using the HAMN health state estimation model.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] (1) The actual working conditions of lithium-ion batteries are simulated, and the ICA analysis method is used to obtain the electrical aging information of random charging segments. At the same time, random stages are selected for EIS testing, and impedance aging information is obtained in the set relaxation time interval of DRT analysis. The addition of impedance aging information makes up for the problem of missing information in electrical characteristics caused by using traditional feature extraction methods for segment information. At the same time, the variance and mean difference of the charging capacity in the SOC interval are proposed as electrical aging features, and the peak value of the relaxation function within a set time period is proposed as the impedance aging feature to comprehensively evaluate the health status of the battery. These features are integrated into the model, making the evaluation of the battery health status more practical and diversified.

[0062] (2) The local attention mechanism is used to strengthen the aging characteristics of information from each source and suppress the influence of noise. The global attention mechanism is used to globally perceive the aging information to facilitate the construction of mapping relationships, thereby alleviating the problem of performance degradation of traditional models when facing noisy aging information. The HAMN health status estimation model of the present invention is suitable for the evaluation of multi-source aging information. The use of parallel computing improves the computing efficiency while establishing a more accurate mapping relationship between features and health status. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 is a flowchart of the steps of the lithium-ion battery health state estimation method based on multi-source fragment information of the present invention;

[0064] Figure 2 is a network structure diagram of the HAMN health status estimation model of the present invention;

[0065] Figure 3 It is the SOC range of each cycle of the aging test of the B0005 battery of the present invention;

[0066] Figure 4 is an IC curve graph obtained by the incremental capacity analysis of the segment SOC interval data of the present invention;

[0067] Figure 5 The electrical aging characteristics of the variance and mean difference extracted from the IC curve on the B0005 battery fragment data of the present invention;

[0068] Figure 6 This is the relaxation time distribution diagram of the B0005 battery after DRT analysis of the present invention;

[0069] Figure 7 It is the distributed peak impedance characteristic within 0.01 to 10 seconds of the relaxation time distribution diagram of the present invention;

[0070] Figure 8 It is a characteristic graph after impedance characteristic multiplexing and alignment of the present invention;

[0071] Fig. 9 The following are the prediction results and 99.7% confidence interval diagrams of the HAMN model when training and testing with 50% of the data respectively;

[0072] Fig.10 It is the fitting scatter plot of the HAMN model prediction value and training value and the regression evaluation index identification;

[0073] Fig.11 This is a comparison chart of the health status estimation effects of each model and the HAMN model;

[0074] Fig.12 This is an enlarged graph of the 50% cycle number after comparing the health status estimation effects of each model with the HAMN model. DETAILED DESCRIPTION

[0075] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0076] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0077] Embodiment 1

[0078] like Figure 1 As shown, a method for estimating the health status of a lithium-ion battery based on multi-source fragment information is provided, comprising the following steps:

[0079] S1: Based on the actual working condition of the simulated lithium-ion battery, a random segment SOC interval is selected in the constant current charging stage of each charge and discharge cycle of the lithium-ion battery to obtain the voltage, capacity and health status information of the segment SOC interval.

[0080] Specifically, the health status data of the lithium-ion battery in each charge and discharge cycle is a sequence of H1, H2, ..., H total , Among them, H i is the battery health status of the i-th (i=1, 2, ..., n) charge and discharge cycle, and total is the total number of charge and discharge cycles.

[0081] C i is the maximum discharge capacity of the lithium-ion battery in the ith charge and discharge cycle, and C is the rated capacity of the lithium-ion battery.

[0082] In the randomly selected SOC interval of the i-th charge and discharge cycle:

[0083] The charging voltage data is a sequence of U i,k×t ,U i,(k+1)×t ,...,U i,q×t , U i,k×t is the initial sampling time voltage of the ith charge and discharge cycle, U i,q×tis the voltage at the end sampling time of the i-th charge and discharge cycle;

[0084] Charging capacity data is sequence C i,k×t ,C i,(k+1)×t ,...,C i,q×t , C i,k×t is the initial sampling time capacity of the ith charge and discharge cycle, C i,q×t is the termination sampling time capacity of the ith charge and discharge cycle, t is the data sampling interval during charging, k and q are the initial and termination sampling moments respectively, and (qk)×t is the total sampling time.

[0085] S2: Based on the voltage, capacity and health status information of the segmented SOC interval, an incremental capacity analysis is performed on each segmented SOC interval, and the variance and mean difference of the charging capacity in the corresponding SOC interval are used as electrical aging characteristics.

[0086] Specifically, the IC function value IC obtained by the incremental capacity analysis of the u-th sampling point is i,u for:

[0087]

[0088] Wherein, u=k, k+1, ..., q, and u is the number of the sampling point.

[0089] Using variance and mean difference for electrical aging feature extraction can be better suitable for fragment information extraction. The variance characterizes the fluctuation of the IC curve with the increase of the number of charge and discharge cycles, and the mean difference reflects the absolute discrete degree of the IC curve from the mean. Both reflect that the degree of fluctuation of the IC curve during the battery aging process is strongly correlated with the health status.

[0090] The charge capacity variance Var of the SOC interval in the i-th charge and discharge cycle i for:

[0091]

[0092] The average difference of charging capacity MDA in the SOC interval of the i-th charge and discharge cycle i for:

[0093]

[0094] Among them, u is the sampling point number of the segment SOC interval, q is the sampling time when the segment SOC interval ends, k is the sampling time when the segment SOC interval starts, IC i,u is the IC function value of the u-th sampling point in the i-th charge-discharge cycle, IC i,mean is the mean value of the IC curve of the i-th charge and discharge cycle.

[0095] The mean IC value of the IC curve for the i-th charge and discharge cycle i,mean for:

[0096]

[0097] S3: Based on the simulated actual working conditions of the lithium-ion battery, several stages are randomly selected in all charge and discharge cycles of the lithium-ion battery, and EIS tests are performed to obtain impedance information and the frequency of the EIS test.

[0098] Specifically, the EIS test is an electrochemical impedance spectroscopy test, and an EIS test is performed on each randomly selected stage.

[0099] The frequency sequence of the EIS test is F m,0 ,F m,1×r ,...,F m,T×r , the real impedance sequence is Z m,0 ,Z m,1×r ,...,Z m,T×r , the imaginary impedance sequence is Z′ m,0 ,Z′ m,1×r ,...,Z′ m,T×r , m (1≤m≤n) is the number of EIS tests, T is the number of EIS test sampling times, T≤qk; r is the impedance test sampling interval, and T×r is the total EIS test time.

[0100] S4: Use DRT testing to analyze the acquired impedance information, and use the peak value of the relaxation function within a set time period as the impedance aging feature.

[0101] Specifically, it includes: S41: Based on the acquired impedance information, Fourier transform is performed on the complex impedance to obtain the frequency domain of the complex impedance; the impedance information is the complex impedance, and the complex impedance includes the real impedance and the imaginary impedance.

[0102] The frequency domain of the complex impedance G m,f for:

[0103]

[0104] Where N is the impedance spectrum sequence length, Z m is the real impedance of the mth EIS test, Z′ m is the imaginary impedance of the mth EIS test, i is the imaginary unit, j is the imaginary unit, n is the time domain sequence variable, and f is the frequency domain sequence variable.

[0105] S42: construct Fourier frequency based on the frequency of EIS test;

[0106] When constructing the Fourier frequency based on the frequencies obtained from the EIS test, the frequency resolution needs to be determined;

[0107] The frequency resolution dfreq is:

[0108]

[0109] Where, df is the sampling interval frequency;

[0110] The Fourier frequency freq is:

[0111] freq=[0,dfreq,...,(N-1)×dfreq].

[0112] S43: Construct the Tikhonov regularization matrix according to the Fourier frequency and the frequency of the EIS test.

[0113]

[0114] Among them, λ is the Tikhonov regularization parameter.

[0115] S44: Perform an inverse Fourier transform on the Tikhonov regularization matrix and the frequency domain of the complex impedance to determine the relaxation time and the relaxation distribution function.

[0116] S45: The peak value of the relaxation distribution function within 0.01s-10s is taken as the impedance aging feature.

[0117] The relaxation time τ is:

[0118]

[0119] Among them, G m,f is the frequency domain of complex impedance, Tikhonov is the Tikhonov regularization matrix, N is the impedance spectrum sequence length, j is the imaginary unit, n is the time domain sequence variable, f is the frequency domain sequence variable; Im(.) is the imaginary part;

[0120] The relaxation distribution function γ(τ) is:

[0121]

[0122] Among them, Rm(.) is the real part;

[0123] The impedance aging characteristic peak m for:

[0124]

[0125] Wherein, m is the number of EIS tests performed.

[0126] S5: Based on the electrical aging characteristics, impedance aging characteristics and health status information, a HAMN health status estimation model is constructed. The HAMN health status estimation model is based on a deep neural network and introduces a hybrid attention mechanism.

[0127] Specifically, before constructing the HAMN health state estimation model, the impedance aging features need to be copied according to the ratio of the feature numbers so that the electrical aging features and the impedance aging features can be aligned; the aligned electrical aging features and impedance aging features are used as the reused input of the HAMN health state estimation model, aiming to utilize the impedance aging features to reuse and align the data dimensions of the electrical aging features, so as to achieve the effect of completing the alignment of network input features.

[0128] The step of replicating the impedance aging characteristics according to the ratio of the characteristic numbers includes (step a and step b):

[0129] a. Calculate the number of feature reuses num:

[0130]

[0131] Where total is the total number of charge and discharge cycles, and T is the sampling number of the EIS test;

[0132] b. Perform feature replication on impedance aging features:

[0133]

[0134] Wherein, peak is the impedance aging characteristic after feature alignment; peak m Impedance aging characteristics extracted from the mth EIS test.

[0135] As a further example, Figure 2 As shown, the HAMN (Hybrid attention mechanism network) health status estimation model is based on a deep neural network and introduces a local attention mechanism and a global attention mechanism.

[0136] Specifically, the HAMN health status estimation model of the present application is based on a fully connected neural network. First, the effective parts of the electrical aging information and the EIS aging information are strengthened and noise is suppressed through the local attention mechanism. Then, the information from the two sources is fused through the cross-attention mechanism to achieve the fusion perception of aging information. Finally, the global receptive field of the aging information is obtained through the global attention mechanism and the mapping relationship between the aging information and SOH is constructed in conjunction with the fully connected neural network.

[0137] In some other embodiments, the HAMN health state estimation model may construct a mapping relationship between aging information and SOH based on a gated recurrent unit network or a long short-term memory network.

[0138] The local attention mechanism uses a local window mechanism to focus on important information and ignore noise information. The construction of the local attention mechanism includes:

[0139] Step 1-1: Select the window area to focus on key K and value V;

[0140]

[0141] Step 1-2: Construct trainable weights for queries Q, keys K, and values ​​V;

[0142]

[0143] Step 1-3: Calculate the attention score of the window area;

[0144]

[0145] Step 1-4: Mask the attention scores outside the window;

[0146]

[0147] score d×w =score d×w +M d×w

[0148] Step 1-5: Calculate the attention weight and output of the window area;

[0149]

[0150]

[0151] The construction of the global attention mechanism includes:

[0152] Step 2-1: Construct the weights of query Q, key K, and value V;

[0153]

[0154]

[0155]

[0156] Step 2-2: Calculate the global attention score;

[0157]

[0158] Step 2-3: Normalize the global attention score and output.

[0159]

[0160]

[0161] Where l is the sequence length of the data K, Q, V, d is the sequence dimension of K, Q, V, and w is the size of the local attention window. dim=x means operating on the xth dimension of the data. softmax(·) is the normalization function.

[0162] Furthermore, the HAMN health status estimation model also introduces residual connections and position encoding.

[0163] Residual connection is a method used to solve the vanishing gradient problem of deep networks. Assume that the input is input, and the deep network structure after the attention mechanism is f(·).

[0164] The calculation method of the residual connection is:

[0165] F(input)=f(input)+input

[0166]

[0167] Among them, input is the input, f(·) is the structure of the deep network structure layer; F(input) is the output after the deep network structure layer; F out (input) is the final output of the residual connection, μ F(input) is the mean of the residual connection output data, σ F(input) The standard deviation of the residual connection output data.

[0168] Position encoding is a means to overcome the loss of position information caused by parallel input. The position encoding is calculated as follows:

[0169] PE pso,2×index =sin(pso / 10000 2×index / d )

[0170] PE pso,2×index+1 =cos(pso / 10000 2×index+1 / d )

[0171] Among them, pso is the position of the data in the sequence, index is the number of all positions in the sequence, and d is the dimension of the encoding vector; PE pso,2×index is the element of even dimension 2×index in the position encoding matrix, PE pso,2×index+1 It is the element of odd dimension 2×index+1 in the position encoding matrix.

[0172] The overfitting problem of the model is alleviated by adding regularization methods and residual connections in each module.

[0173] By setting the mean square error loss function, the back propagation algorithm is used to optimize the model hyperparameters on the training set. The specific process is as follows:

[0174] Take the first p sample values ​​of the input sequence and health status sequence and As a training set, input sample values ​​into the HAMN health status estimation model to obtain

[0175] Optimize the mean square error loss function through backpropagation: Make the network parameter parm reach the best combination.

[0176] The test set Input the HAMN network with trained parameters to get the predicted value

[0177] The local attention mechanism is used to strengthen the aging characteristics of each source of information and suppress the influence of noise. The global attention mechanism is used to globally perceive the aging information to facilitate the construction of mapping relationships. This alleviates the problem of performance degradation of traditional models when facing noisy fragments of aging information. The model is suitable for the evaluation of multi-source aging information. The use of parallel computing improves the computing efficiency and establishes a more accurate mapping relationship between features and health status.

[0178] S6: Based on the electrical aging characteristics and impedance aging characteristics of the unknown lithium-ion battery, the battery health state estimation model is used to estimate the battery health state of the unknown lithium-ion battery.

[0179] The actual working conditions of lithium-ion batteries are simulated, and the ICA (Incremental Capacity Analysis) method is used to obtain the electrical aging information of random charging segments. At the same time, random stages are selected for EIS testing, and impedance aging information is obtained in the set relaxation time interval of DRT analysis as a supplement to the missing electrical information. The addition of impedance aging features makes up for the missing information on electrical features of segment information using traditional feature extraction methods.

[0180] At the same time, the variance and mean difference of the charging capacity in the SOC interval are proposed as electrical aging characteristics, and the peak value of the relaxation function within a set time period is proposed as the impedance aging characteristic to comprehensively evaluate the health status of the battery. These are integrated into the model to make the evaluation of the battery health status more practical and diversified.

[0181] Embodiment 2

[0182] Take the B0005 battery in the NASA dataset as an example. Figure 3 is the SOC range of B0005 battery at different cycle times, such as Figure 4 To perform incremental capacity analysis in the segment SOC interval, Figure 5 The calculated variance and mean difference are used as electrical aging characteristics.

[0183] For the B0005 battery, an electrochemical impedance spectroscopy test was performed from the 21st cycle to the 147th cycle before charging. The results of the DRT analysis are as follows: Figure 6 As shown in Figure 2, the peak impedance characteristics extracted within 0.01s-10s are as follows Figure 7 shown.

[0184] like Figure 8 Shown are the impedance spectrum characteristics after alignment.

[0185] The present invention takes 50% of the data as a training set and 50% of the data as a test set as an example to conduct a health status estimation experiment. Fig. 9 The model test results of B0005 battery and its stability confidence interval of 100 tests at 99.7% confidence level. Fig.10 It is the fitting degree judgment and its evaluation index. Fig. 9 and 10 It can be seen that HAMN has excellent lithium-ion battery health status estimation capabilities.

[0186] In order to verify the performance of the HAMN health status estimation model, the present invention compares multiple methods applied to the regression model: the machine learning model Support Vector Regression (SVR), the sequence prediction model Long-Short-Term Memory neural networks (LSTM), the Gated Recurrent Units (GRU) and the Transformer. Fig.11 The prediction effects of various models on B0005 battery. The regression evaluation index and model consumption time are shown in Table 1.

[0187] Table 1 Regression evaluation indicators and time consumption of each model

[0188] Model MAE(%) RMSE(%) MAPE(%) <![CDATA[R 2 ]]> Time(s) HAMN 0.4985 0.7226 0.6637 0.9584 37.4936 Transformer 0.0087 0.9737 0.9349 0.9533 55.2073 LSTM 0.8804 1.0700 1.2091 0.9088 24.6668 GRU 0.5642 0.7888 0.7449 0.9504 21.6817 SVR 0.9128 1.1524 1.2077 0.8942 7.5491

[0189] It can be concluded from Table 1 that R 2It is 0.9584, which is the highest among the five models, indicating that the HAMN model has the best fitting shape; the MAE, RMSE and MAPE of the HAMN health status estimation model of the present invention are 0.4985%, 0.7226% and 0.6637% respectively, which are the lowest among all models, indicating that the HAMN health status estimation model constructed by the present invention has good stability.

[0190] Embodiment 3

[0191] A lithium-ion battery health state estimation system based on multi-source fragment information is provided, comprising:

[0192] The first acquisition module is used to select a random segment SOC interval in the constant current charging stage of each charge and discharge cycle of the lithium-ion battery based on the simulated actual working condition of the lithium-ion battery, and obtain the voltage, capacity and health status information of the segment SOC interval;

[0193] The electrical aging feature extraction module is used to perform incremental capacity analysis on each SOC interval according to the voltage, capacity and health status information of the SOC interval, and use the variance and mean difference of the charging capacity of the corresponding SOC interval as the electrical aging feature;

[0194] The second acquisition module is used to randomly select several stages in all charge and discharge cycles of the lithium-ion battery based on the simulated actual working conditions of the lithium-ion battery, and perform EIS tests to obtain impedance information and the frequency of the EIS test;

[0195] The impedance aging feature extraction module is used to analyze the acquired impedance information using the DRT test and use the relaxation function peak value within a set time period as the impedance aging feature;

[0196] A construction module is used to construct a HAMN health state estimation model based on electrical aging characteristics, impedance aging characteristics and health state information, wherein the HAMN health state estimation model is based on a deep neural network and introduces a hybrid attention mechanism;

[0197] The estimation module is used to estimate the battery health state of the unknown lithium-ion battery based on the electrical aging characteristics and impedance aging characteristics of the unknown lithium-ion battery using the HAMN health state estimation model.

[0198] For more specific processes of the above method, please refer to the corresponding contents disclosed in the aforementioned embodiments, which will not be repeated here.

[0199] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0200] Those skilled in the art can clearly understand that the technology in the embodiments of the present invention can be implemented by means of software plus a necessary general hardware platform. Based on this understanding, the technical solution in the embodiments of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a disk, an optical disk, etc., and includes a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention or some parts of the embodiments.

[0201] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.

Claims

1. A method for estimating the health status of a lithium-ion battery based on multi-source fragment information, characterized in that: include: Based on the actual working conditions of the simulated lithium-ion battery, a random segment SOC interval is selected in the constant current charging stage of each charge and discharge cycle of the lithium-ion battery to obtain the voltage, capacity and health status information of the segment SOC interval; According to the voltage, capacity and health status information of the segmented SOC interval, the incremental capacity analysis is performed on each segmented SOC interval, and the variance and average difference of the charging capacity in the corresponding SOC interval are used as electrical aging characteristics; Based on the simulated actual working conditions of lithium-ion batteries, several stages are randomly selected in all charge and discharge cycles of lithium-ion batteries and EIS tests are performed to obtain impedance information and the frequency of EIS tests; The acquired impedance information is analyzed using DRT test, and the peak value of relaxation function within the set time period is used as the impedance aging feature; Based on electrical aging characteristics, impedance aging characteristics and health status information, a HAMN health status estimation model is constructed; the HAMN health status estimation model is based on a deep neural network and introduces a hybrid attention mechanism; Based on the electrical aging characteristics and impedance aging characteristics of unknown lithium-ion batteries, the HAMN health state estimation model is used to estimate the battery health state of unknown lithium-ion batteries; The method of using the DRT test to analyze the acquired impedance information and taking the peak value of the relaxation function within a set time period as the impedance aging feature includes: Based on the acquired impedance information, Fourier transform is performed on the complex impedance to obtain the frequency domain of the complex impedance; the impedance information is the complex impedance; Based on the frequency of the EIS test, construct the Fourier frequency; Construct the Tikhonov regularization matrix according to the Fourier frequency and the frequency of the EIS test; Perform inverse Fourier transform on the Tikhonov regularization matrix and the frequency domain of the complex impedance to determine the relaxation time and relaxation distribution function; The peak value of the relaxation distribution function within 0.01s-10s is taken as the impedance aging feature.

2. The method for estimating the health status of a lithium-ion battery based on multi-source fragment information according to claim 1, characterized in that: The relaxation time τ is: Among them, G m,f is the frequency domain of complex impedance, Tikhonov is the Tikhonov regularization matrix, N is the impedance spectrum sequence length, j is the imaginary unit, n is the time domain sequence variable, f is the frequency domain sequence variable; Im(.) is the imaginary part; The relaxation distribution function γ(τ) is: Among them, Rm(.) is the real part; The impedance aging characteristic peak m for: Wherein, m is the number of EIS tests performed.

3. The method for estimating the health status of a lithium-ion battery based on multi-source fragment information according to claim 1, characterized in that: Before constructing the HAMN health state estimation model based on the electrical aging characteristics, the impedance aging characteristics and the health state information, the impedance aging characteristics need to be copied according to the ratio of the characteristic numbers so as to align the electrical aging characteristics with the impedance aging characteristics.

4. The method for estimating the health status of a lithium-ion battery based on multi-source fragment information according to claim 3, characterized in that: The step of replicating the impedance aging characteristics according to the ratio of the characteristic numbers includes: Calculate the number of feature reuses num: Where total is the total number of charge and discharge cycles, and T is the sampling number of the EIS test; To perform feature replication on the impedance aging feature: Wherein, peak is the impedance aging characteristic after feature alignment; peak m Impedance aging characteristics extracted from the mth EIS test.

5. The method for estimating the health status of a lithium-ion battery based on multi-source fragment information according to claim 1, characterized in that: According to the voltage, capacity and health status information of the segment SOC interval, the incremental capacity analysis is performed on each segment SOC interval, and the variance and average difference of the charging capacity of the corresponding SOC interval are used as the electrical aging characteristics. The charge capacity variance Var of the SOC interval of the i-th charge and discharge cycle i for: The average difference of charging capacity MDA in the SOC interval of the i-th charge and discharge cycle i for: Where u is the sampling point number of the segment SOC interval, q is the sampling time when the segment SOC interval ends, k is the sampling time when the segment SOC interval starts, IC i,u is the IC function value of the u-th sampling point in the i-th charge-discharge cycle, IC i,mean is the mean value of the IC curve of the i-th charge and discharge cycle.

6. The method for estimating the health status of a lithium-ion battery based on multi-source fragment information according to claim 1, characterized in that: The HAMN health status estimation model is constructed based on the electrical aging characteristics, impedance aging characteristics and health status information; the HAMN health status estimation model is based on a deep neural network and introduces a hybrid attention mechanism. The hybrid attention mechanism includes a local attention mechanism and a global attention mechanism; The deep neural network on which the HAMN health status estimation model is based is a fully connected neural network.

7. The method for estimating the health status of a lithium-ion battery based on multi-source fragment information according to claim 1, characterized in that: The HAMN health status estimation model is constructed based on the electrical aging characteristics, impedance aging characteristics and health status information; the HAMN health status estimation model is based on a deep neural network and introduces a hybrid attention mechanism. The HAMN health status estimation model also introduces residual connection and position encoding; The calculation method of the residual connection is: F(input)=f(input)+input Among them, input is the input, f(·) is the structure of the deep network structure layer; F(input) is the output after the deep network structure layer; F out (input) is the final output of the residual connection, μ F(input) is the mean of the residual connection output data, σ F(input) The standard deviation of the residual connection output data.

8. The method for estimating the health status of a lithium-ion battery based on multi-source fragment information according to claim 7, characterized in that: The position code is calculated as follows: PE pso,2×index =sin(pso / 10000 2×index / d ) ON pso,2×index+1 =cos(pso / 10000 2×index+1 / d ) Among them, pso is the position of the data in the sequence, index is the number of all positions in the sequence, and d is the dimension of the encoding vector; PE pso,2×index is the element of even dimension 2×index in the position encoding matrix, PE pso,2×index+1 It is the element of odd dimension 2×index+1 in the position encoding matrix.

9. A lithium-ion battery health status estimation system based on multi-source fragment information, characterized in that: include: The first acquisition module is used to select a random segment SOC interval in the constant current charging stage of each charge and discharge cycle of the lithium-ion battery based on the simulated actual working condition of the lithium-ion battery, and obtain the voltage, capacity and health status information of the segment SOC interval; The electrical aging feature extraction module is used to perform incremental capacity analysis on each SOC interval according to the voltage, capacity and health status information of the SOC interval, and use the variance and mean difference of the charging capacity of the corresponding SOC interval as the electrical aging feature; The second acquisition module is used to randomly select several stages in all charge and discharge cycles of the lithium-ion battery based on the simulated actual working conditions of the lithium-ion battery, and perform EIS tests to obtain impedance information and the frequency of the EIS test; The impedance aging feature extraction module is used to analyze the acquired impedance information using the DRT test and use the relaxation function peak value within a set time period as the impedance aging feature; A construction module is used to construct a HAMN health state estimation model based on electrical aging characteristics, impedance aging characteristics and health state information, wherein the HAMN health state estimation model is based on a deep neural network and introduces a hybrid attention mechanism; An estimation module, used for estimating the battery health state of an unknown lithium-ion battery using a HAMN health state estimation model based on the electrical aging characteristics and impedance aging characteristics of the unknown lithium-ion battery; The method of using the DRT test to analyze the acquired impedance information and taking the peak value of the relaxation function within a set time period as the impedance aging feature includes: Based on the acquired impedance information, Fourier transform is performed on the complex impedance to obtain the frequency domain of the complex impedance; the impedance information is the complex impedance; Based on the frequency of the EIS test, construct the Fourier frequency; Construct the Tikhonov regularization matrix according to the Fourier frequency and the frequency of the EIS test; Perform inverse Fourier transform on the Tikhonov regularization matrix and the frequency domain of the complex impedance to determine the relaxation time and relaxation distribution function; The peak value of the relaxation distribution function within 0.01s-10s is taken as the impedance aging feature.

Images