ER-BILSTM-FM model for estimating state of health of lithium battery

Abstract

The invention discloses an ER-BILSTM-FM model for estimating the state of health of a lithium battery. The ER-BILSTM-FM model comprises a BiLSTM model, a feature mapping layer, a residual fusion layer and a full connection layer, the BiLSTM model is used for acquiring forward correlation characteristics and reverse correlation characteristics at different moments according to an input battery health factor sequence; the feature mapping layer is used for performing feature mapping processing on the forward correlation features and the reverse correlation features to obtain first feature mapping output and second feature mapping output; the residual fusion layer is used for fusing the forward correlation feature and the first feature mapping output to obtain a first fusion output, and fusing the reverse correlation feature and the second feature mapping output to obtain a second fusion output; and the full connection layer is used for obtaining a battery health state estimation result according to the first fusion output and the second fusion output. According to the model, charging data of different dimensions are fully considered, data noise is reduced, and an accurate estimation result is achieved.

Description

Technical Field

[0001] The present invention relates to the field of battery technology, and in particular to an ER-BILSTM-FM model for estimating the health state of a lithium battery. Background Art

[0002] Lithium batteries have the characteristics of low self-discharge rate and high energy density, which enable them to provide longer battery life while providing efficient energy conversion. In addition, lithium batteries can be used in cascades, which improves resource utilization efficiency. Therefore, lithium batteries are highly favored in the field of consumer energy storage components. In the field of electric vehicles, the excellent performance of lithium ions directly affects the safety performance of electric vehicles. That is, as the number of charge and discharge times increases, the aging degree of lithium batteries increases, and the safety of electric vehicles will be negatively affected. When the aging degree of lithium batteries is large, electric vehicles will have accidents such as fire during the charging and discharging process. Therefore, accurate assessment of the aging degree of batteries is an important and challenging research direction.

[0003] The lithium battery charging process is mainly divided into two stages: constant current charging stage and constant voltage charging stage. Traditional research focuses on the battery working information in the constant current charging stage, thereby ignoring the information in the constant voltage charging stage and failing to fully utilize the effective information in the battery charging process. In addition, the existing technology lacks processing for the collected data noise, so the existing technology's estimation of the battery's SOH (battery health state) is often incomplete and inaccurate. Summary of the invention

[0004] The purpose of the present invention is to provide an ER-BILSTM-FM model for estimating the health state of a lithium battery, so as to achieve accurate estimation of the health state of a lithium battery.

[0005] In order to solve the above technical problems, the present invention provides an ER-BILSTM-FM model for estimating the health status of a lithium battery. The ER-BILSTM-FM model includes a BiLSTM model, a feature mapping layer, a residual fusion layer, and a fully connected layer;

[0006] The BiLSTM model is used to obtain forward correlation features and reverse correlation features at different times based on the input battery health factor sequence;

[0007] The feature mapping layer is used to perform feature mapping processing on the forward correlation feature and the reverse correlation feature respectively to obtain a first feature mapping output and a second feature mapping output;

[0008] The residual fusion layer is used to fuse the forward correlation feature with the first feature mapping output to obtain a first fusion output, and to fuse the reverse correlation feature with the second feature mapping output to obtain a second fusion output;

[0009] The fully connected layer is used to obtain a battery health status estimation result according to the first fusion output and the second fusion output.

[0010] According to the above scheme, the battery health factor sequence includes battery health factors arranged over time; the battery health factor is the area enclosed by the battery charging power curve within a certain charging period.

[0011] According to the above scheme, the feature mapping layer performs feature mapping processing on the forward correlation feature according to the set activation function, the first weight matrix, and the first bias to obtain a first feature mapping output;

[0012] The feature mapping layer performs feature mapping processing on the reverse correlation features according to the set activation function, the second weight matrix, and the second bias to obtain a second feature mapping output.

[0013] The present invention also provides a training method for an ER-BILSTM-FM model, comprising:

[0014] S1. Test to obtain a battery health factor sequence and a corresponding battery health status sequence;

[0015] S2. Divide a training set from the battery health factor sequence obtained from the test, and obtain a battery health state sequence corresponding to the training set;

[0016] S3, input the training set into the ER-BILSTM-FM model to obtain the output battery health status estimation set;

[0017] S4, obtaining an error set according to the battery health state estimation set and the battery health state sequence corresponding to the training set;

[0018] S5. Optimize the parameters of the ER-BILSTM-FM model according to the error set;

[0019] S6. When the set conditions are met, the ER-BILSTM-FM model training ends.

[0020] According to the above scheme, in step S1, the battery health factor sequence is obtained by testing under a full-stage charging scenario; the full-stage charging scenario is a charging scenario in which the battery is charged from 0% SOC to 100% SOC.

[0021] According to the above scheme, the method of dividing the battery health factor sequence obtained from the test into a training set includes: dividing the battery health factor sequence in a certain proportion in the time series space to obtain a training set and a test set.

[0022] According to the above scheme, the division ratio of the training set and the test set is 1:1.

[0023] The present invention also provides a lithium battery health state estimation method based on the ER-BILSTM-FM model. The method inputs the battery health factor into the ER-BILSTM-FM model for estimating the health state of the lithium battery described above, and obtains the battery health state estimation result output by the ER-BILSTM-FM model.

[0024] The present invention also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the lithium battery health status estimation method based on the ER-BILSTM-FM model described above when executing the computer program.

[0025] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the lithium battery health state estimation method based on the ER-BILSTM-FM model described above are implemented.

[0026] The beneficial effects of the present invention are as follows: the present invention adds a feature mapping layer to the existing BILSTM model, utilizes feature mapping processing to effectively remove the noise in the input battery health factor sequence, and then fuses the data before and after the feature mapping processing through a further added residual fusion layer, taking into account the respective advantages of the fused data, and effectively improving the accuracy of the battery health status estimation results.

[0027] Furthermore, the present invention effectively integrates the charging voltage and charging current during the charging process to obtain the battery health factor as the input of the ER-BILSTM-FM model, fully considers the working characteristics of the lithium battery in different charging stages (constant current charging stage and constant voltage charging stage), effectively utilizes the effective information in the battery charging process, and further improves the accuracy of the battery health status estimation result output by the ER-BILSTM-FM model. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a schematic diagram of the ER-BILSTM-FM model structure of the first embodiment of the present invention;

[0029] Figure 2 is a flow chart of a training method of the ER-BILSTM-FM model of the second embodiment of the present invention;

[0030] Figure 3 is a schematic diagram of a battery aging platform according to a second embodiment of the present invention;

[0031] Figure 4 is a battery charging power curve obtained in the full-stage charging scenario of the second embodiment of the present invention;

[0032] Figure 5 is a diagram of battery health status estimation results of different estimation models in embodiment 2 of the present invention;

[0033] Figure 6 This is a battery health status estimation result diagram of the ER-BILSTM-FM model based on random fragment charging data according to the second embodiment of the present invention;

[0034] Figure 7 This is a battery health status estimation result diagram of the ER-BILSTM-FM model of the second embodiment of the present invention based on the full-segment charging data of the battery;

[0035] Figure 8 This is a battery health status estimation result diagram of the ER-BILSTM-FM model of Example 2 of the present invention based on random charging data across batteries. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solution and advantages of the embodiments of the present disclosure clearer, the technical solution of the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings of the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, not all of the embodiments. Based on the described embodiments of the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.

[0037] Embodiment 1:

[0038] See also Figure 1 , this embodiment provides an ER-BILSTM-FM model for estimating the health status of a lithium battery, the ER-BILSTM-FM model includes a BiLSTM model, a feature mapping layer, a residual fusion layer, and a fully connected layer;

[0039] The BiLSTM model is used to obtain forward correlation features and reverse correlation features at different times based on the input battery health factor sequence;

[0040] The feature mapping layer is used to perform feature mapping processing on the forward correlation feature and the reverse correlation feature respectively to obtain a first feature mapping output and a second feature mapping output;

[0041] The residual fusion layer is used to fuse the forward correlation feature with the first feature mapping output to obtain a first fusion output, and to fuse the reverse correlation feature with the second feature mapping output to obtain a second fusion output;

[0042] The fully connected layer is used to obtain a battery health status estimation result according to the first fusion output and the second fusion output.

[0043] Further, the battery health factor sequence includes battery health factors arranged over time; the battery health factor is the area enclosed by the battery charging power curve within a certain charging period;

[0044] Specifically, the battery health factor sequence is obtained as follows:

[0045] 1) Collect the charging current [I i1 ,I i2 ,...,I ij ] and charging voltage [U i1 ,U i2 ,...,U ij ];

[0046] 2) Collect the charging current [I' under random segment charging scenarios (charging scenarios in random SOC intervals) i1 ,I' i2 ,...,I' ij ] and charging voltage [U' i1 ,U' i2 ,...,U' ij ];

[0047] Among them, I ij and U ij are the jth charging current and charging voltage samples in the i-th full-stage charging scenario; I' ij and U' ij are the jth charging current and charging voltage samples under the i-th random segment charging scenario, n is the number of working cycles of the battery, and m is the number of samples collected under different charging scenarios;

[0048] 3) Obtain the power signal [P i1 ,P i2 ,...,P ij ] and the power signal [P' i1 ,P' i2 ,...,P' ij ]:

[0049] P ij =I ij *U ij

[0050] P' ij =I' ij *U' ij

[0051] Where P ij and P' ijThe jth charging power sample during the i-th charging process under the full-segment charging scenario and the random-segment charging scenario respectively;

[0052] 4) Use integral calculation to obtain the charging power curve area S of each charging process in the full charging scenario i ;

[0053]

[0054] In the above formula, P i is the number of charging power samples collected during the i-th full-stage charging, T1 and T2 are the starting charging time and the ending charging time of the full-stage charging respectively;

[0055] 5) Use integral operation to obtain the charging power curve area S' of each charging process in the random segment charging scenario i ;

[0056]

[0057] In the above formula, P' i is the number of charging power samples collected during the i-th random segment charging, and t1 and t2 are the starting charging time and the ending charging time of the random segment charging, respectively.

[0058] Further, the feature mapping layer performs feature mapping processing on the forward correlation feature according to the set activation function, the first weight matrix, and the first bias to obtain a first feature mapping output;

[0059] The feature mapping layer performs feature mapping processing on the reverse correlation feature according to the set activation function, the second weight matrix, and the second bias to obtain a second feature mapping output;

[0060] Specifically, the execution process of the feature mapping layer is:

[0061] M 1i =F(W 1 K 1i +b 1 )

[0062] M 2i =F(W 2 K 2i +b 2 )

[0063] In the above formula, K 1i , K 2i are the forward correlation features and reverse correlation features of the battery health factor (i.e., battery health factor sequence) of the BiLSTM model at each moment in the time series space, M 1i and M 2iare the first feature map output and the second feature map output respectively, W1 and W2 are the weight matrices of the forward / reverse temporal feature map parts respectively, b1 and b2 are the biases of the forward / reverse temporal feature map parts respectively, and F() is the activation function.

[0064] Furthermore, the residual fusion layer fuses the output of the BiLSTM model and the output of the feature mapping layer through residual connection. The execution process of the residual fusion layer is:

[0065] O 1i =K 1i +F(W 1 K 1i +b 1 )

[0066] O 2i =K 2i +F(W 2 K 2i +b 2 )

[0067] In the above formula, O 1i and O 2i are the first fusion output and the second fusion output respectively.

[0068] Furthermore, the fully connected layer obtains the battery health status estimation result according to the first fusion output and the second fusion output, which is expressed as i .

[0069] Embodiment 2:

[0070] See also Figure 2 This embodiment provides a training method for an ER-BILSTM-FM model, including:

[0071] S1. Test to obtain a battery health factor sequence and a corresponding battery health status sequence;

[0072] S2. Divide a training set from the battery health factor sequence obtained from the test, and obtain a battery health state sequence corresponding to the training set;

[0073] S3, input the training set into the ER-BILSTM-FM model to obtain the output battery health status estimation set;

[0074] S4, obtaining an error set according to the battery health state estimation set and the battery health state sequence corresponding to the training set;

[0075] S5. Optimize the parameters of the ER-BILSTM-FM model according to the error set;

[0076] S6. When the set conditions are met, the ER-BILSTM-FM model training ends.

[0077] Furthermore, in step S1, a battery health factor sequence is obtained by testing under a full-stage charging scenario; the full-stage charging scenario is a charging scenario in which the battery is charged from 0% SOC to 100% SOC.

[0078] Furthermore, the method of dividing the training set from the battery health factor sequence obtained from the test includes: dividing the battery health factor sequence in a time series space at a ratio of 1:1 to obtain a training set and a test set;

[0079] Specifically, the division process of the training set and the test set is:

[0080] 1) Battery health factors obtained based on the full charging scenario [S1, S2, ..., S i ] and battery SOH values ​​[H1,H2,...,H i ];

[0081] 2) In the time series space, [S1,S2,...,S i ]、[H1,H2,...,H i ] is divided into training set and test set;

[0082] The training set includes [S1,S2,...,S i / 2 ] and [H1,H2,...,H i / 2 ]; the test set includes [S (i / 2)+1 ,S (i / 2)+2 ,...,S i ] and [H (i / 2)+1 ,H (i / 2)+2 ,...,H i ].

[0083] Furthermore, the battery health status estimation set in step S3 is represented as [O1, O2, ..., O i / 2 ]; in step S4, according to the battery health status estimation set [O1, O2, ..., O i / 2 ] and the actual battery SOH value [H1,H2,...,H i ], the error set obtained is expressed as [E1,E2,...,E i / 2 ].

[0084] Furthermore, in step S5, based on the error set represented as [E1, E2, ..., E i / 2 ] and Adam optimization algorithm are used to optimize the parameters of the ER-BILSTM-FM model.

[0085] Furthermore, step S6 includes: repeating the forward propagation, loss calculation, back propagation and parameter update of steps S1 to S5 until the maximum number of iterations M (M is a set value) is reached, and the ER-BILSTM-FM model training is completed.

[0086] This embodiment also provides a verification method after the ER-BILSTM-FM model is trained, including:

[0087] 1) Test set data based on full-stage charging scenario [S (i / 2)+1 ,S (i / 2)+2 ,...,S i ] and [H (i / 2)+1 ,H (i / 2)+2 ,...,H i ], verify the trained ER-BILSTM-FM model;

[0088] 2) The number of tests corresponding to the last 50% of the cycles under random fragment charging conditions [S' (i / 2)+1 ,S' (i / 2)+2 ,...,S' i ] and [H (i / 2)+1 ,H (i / 2)+2 ,...,H i ] to verify the trained ER-BILSTM-FM model.

[0089] To illustrate the estimation process and performance of the ER-BILSTM-FM model proposed in the present invention, an example is provided as follows:

[0090] This example is based on a battery aging platform consisting of a battery capacity tester (Neware BTS-400), a battery constant temperature environmental test chamber (NewareMHW-200) and a battery management system (see Figure 3 ), collect battery working data. In this example, the rated capacity of the battery is 2.5Ah, and the set working conditions are: constant current charging with a charging current of 1250mA and constant voltage charging with a constant voltage of 4.2V, and a multi-stage discharge process with currents of 1250mA and 2500mA. This example is based on the laboratory measured data of the degradation of the discharge capacity of the lithium battery pack during the charge and discharge cycle to achieve accurate evaluation of the battery SOH. This example conducts an estimation experiment on the health status of the lithium battery pack, and the specific operation steps are as follows:

[0091] (1) Collect the battery charging voltage and charging current signals under different charging scenarios, integrate the charging voltage signal and the charging current signal, and obtain the power signal. The power curve obtained under the full charging scenario is as follows: Figure 3 As shown;

[0092] (2) The charging power curve area in different charging scenarios is obtained by using integral operation, and the result is used as the battery health factor; the Pearson correlation coefficient is used to quantify the correlation between the battery health factor extracted from different discharge scenarios and the battery SOH. The final results are all greater than 0.99, indicating that there is a strong positive linear relationship between the battery health factor extracted by the battery health factor extraction method proposed in the present invention and the battery SOH;

[0093] (3) Based on the deep learning network, the ER-BILSTM-FM model is established by introducing the feature mapping layer and the residual connection network structure. The specific parameters of the BiLSTM unit are that the number of units is set to 32, the activation function is tanh, and during model training, the Adam optimizer is selected, the loss function is set to the mean square error, the learning rate is set to 0.001, the maximum number of iterations M is set to 100, and the number of training samples used each time the model parameters are updated is 32.

[0094] (4) The model is trained based on the full charging data of the first 50% of the cycles of a battery as the training set, and then the trained model is verified based on the full charging data of the last 50% of the cycles and the random charging data as the test set. The random charging SOC range is 37%-69%. The final experimental results are as follows Figure 5 and Figure 6 The final model evaluation indicators are shown in Table 1.

[0095] Table 1 Battery SOH evaluation indicators

[0096]

[0097] (5) Based on the trained model and combined with the full charging data and random segment charging data of another single cell, the cross-battery SOH estimation task is realized. The random segment charging SOC range is 41%-76%. The specific working conditions of this single cell are different from those described previously. The final experimental results are shown in the figure below. Figure 7 and Figure 8 The final model evaluation indicators are shown in Table 2.

[0098] Table 2 Cross-battery SOH evaluation indicators

[0099]

[0100] From the comparison chart of the results of the lithium battery pack health state estimation, it can be seen that the estimated value of the lithium battery pack health state estimation method proposed in the present invention is more stable with the true value, and the same conclusion can be drawn from Table 1 and Table 2. The RMSE and MAE of the lithium battery pack health state estimation method proposed in the present invention are lower than those of the comparison model, which shows that the power curve area as a battery health factor can better reflect the degradation of the health state of the lithium battery pack. The battery SOH estimation model ER-BILSTM-FM integrates the useful information in the features before and after feature mapping through residual connection and information fusion, thereby realizing the maximum utilization of effective information in the data. This shows that the lithium battery health state estimation method based on the ER-BILSTM-FM model proposed in the present invention has the advantages of small error and high efficiency.

[0101] Embodiment three:

[0102] This embodiment provides a lithium battery health state estimation method based on the ER-BILSTM-FM model. The method inputs the battery health factor into the ER-BILSTM-FM model for estimating the health state of the lithium battery described above, and obtains the battery health state estimation result output by the ER-BILSTM-FM model.

[0103] Embodiment 4:

[0104] This embodiment also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the lithium battery health status estimation method based on the ER-BILSTM-FM model described in Embodiment 1 are implemented.

[0105] Embodiment five:

[0106] This embodiment also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the lithium battery health status estimation method based on the ER-BILSTM-FM model described in Example 1 are implemented.

[0107] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. An ER-BILSTM-FM model for estimating the health status of a lithium battery, characterized in that: The ER-BILSTM-FM model includes the BiLSTM model, feature mapping layer, residual fusion layer, and fully connected layer; The BiLSTM model is used to obtain forward correlation features and reverse correlation features at different times based on the input battery health factor sequence; The feature mapping layer is used to perform feature mapping processing on the forward correlation feature and the reverse correlation feature respectively to obtain a first feature mapping output and a second feature mapping output; The residual fusion layer is used to fuse the forward correlation feature with the first feature mapping output to obtain a first fusion output, and to fuse the reverse correlation feature with the second feature mapping output to obtain a second fusion output; The fully connected layer is used to obtain a battery health status estimation result according to the first fusion output and the second fusion output.

2. The ER-BILSTM-FM model for estimating the health status of a lithium battery according to claim 1, characterized in that: The battery health factor sequence includes battery health factors arranged over time; the battery health factor is the area enclosed by the battery charging power curve within a certain charging period.

3. The ER-BILSTM-FM model for estimating the health status of a lithium battery according to claim 1, characterized in that: The feature mapping layer performs feature mapping processing on the forward correlation feature according to the set activation function, the first weight matrix, and the first bias to obtain a first feature mapping output; The feature mapping layer performs feature mapping processing on the reverse correlation features according to the set activation function, the second weight matrix, and the second bias to obtain a second feature mapping output.

4. A training method for an ER-BILSTM-FM model, characterized in that: include: S1. Test to obtain a battery health factor sequence and a corresponding battery health status sequence; S2. Divide a training set from the battery health factor sequence obtained from the test, and obtain a battery health state sequence corresponding to the training set; S3, input the training set into the ER-BILSTM-FM model to obtain the output battery health status estimation set; S4, obtaining an error set according to the battery health state estimation set and the battery health state sequence corresponding to the training set; S5. Optimize the parameters of the ER-BILSTM-FM model according to the error set; S6. When the set conditions are met, the ER-BILSTM-FM model training ends.

5. The training method of the ER-BILSTM-FM model according to claim 4, characterized in that: In step S1, a battery health factor sequence is obtained by testing in a full-stage charging scenario; The full charging scenario is a charging scenario in which the battery is charged from 0% SOC to 100% SOC.

6. The training method of the ER-BILSTM-FM model according to claim 4, characterized in that: The method for dividing a training set from a battery health factor sequence obtained from a test includes: dividing the battery health factor sequence in a time series space according to a certain ratio to obtain a training set and a test set.

7. The training method of the ER-BILSTM-FM model according to claim 6, characterized in that: The ratio of training set and test set is 1:

1.

8. A lithium battery health status estimation method based on ER-BILSTM-FM model, characterized in that: The method inputs the battery health factor into the ER-BILSTM-FM model for estimating the health status of a lithium battery as described in any one of claims 1 to 3, and obtains the battery health status estimation result output by the ER-BILSTM-FM model.

9. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the method for estimating the health status of a lithium battery based on the ER-BILSTM-FM model as described in claim 8 are implemented.

10. A computer-readable 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 battery health status estimation method based on the ER-BILSTM-FM model described in claim 8 are implemented.

Images