A method for early prediction of lithium-ion battery life

Abstract

The present invention discloses a method for early prediction of lithium-ion battery life. The method comprises the following steps: (1) obtaining model training data: obtaining capacity and voltage values; (2) calculating the standard deviation MSE (f n (V)); (3) Early life prediction model training: Use the support vector machine optimized based on the humpback whale algorithm as the prediction algorithm to train the prediction model. During training, the input vector x is used as the input of the model, and the output vector y is used as the output of the model. The model is trained to obtain the prediction model; (4) Early prediction of battery life: Predict the battery cycle life through the prediction model. The present invention can realize early prediction of the life of lithium-ion batteries by calculating and transforming the attenuation curve, and then combining it with the prediction algorithm for training, and can effectively classify or grade the batteries according to their predicted life. The present invention can promote battery management optimization, improve the consistency of battery packs, and improve the output performance and safety of battery packs.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium ion batteries, and more specifically to an early prediction method for the life of a lithium ion battery. Background Art

[0002] As the energy crisis and ecological environment problems become increasingly serious, new energy vehicles have become the development focus of the automotive industry. However, the performance of the battery pack on new energy vehicles directly affects the performance and operation of new energy vehicles. The battery pack used in new energy vehicles is composed of many battery cells connected in parallel and in series. If the life of these battery cells varies greatly, the output performance of the battery pack will decrease after a period of use, and even safety risks will occur. Therefore, it is very important to make early predictions on the life of the battery cells before they are grouped.

[0003] At present, almost all lithium-ion battery negative electrode materials are made of graphite. The loss of active material in the de-lithiation negative electrode of these lithium-ion batteries will cause the discharge voltage to change, but the capacity will not change. This is because the loss of de-lithiation negative electrode material will change the storage potential of lithium ions without changing the total capacity. At a high rate of loss of active material in the de-lithiation negative electrode, the negative electrode capacity will eventually be lower than the remaining lithium ion inventory. At this point, the negative electrode will not have enough sites to accommodate lithium ions during the charging process, resulting in lithium plating. Since lithium plating is irreversible, the capacity loss will be accelerated.

[0004] Currently, there is still a lack of an effective method for early prediction of lithium-ion battery life. Summary of the invention

[0005] The present invention aims at the problem of early prediction of lithium-ion battery life and proposes a method for early prediction of lithium-ion battery life to achieve early prediction of battery life. In the early charge and discharge cycle of lithium-ion batteries, the loss of active material in the de-lithiation negative electrode will change the voltage curve without affecting the capacity decay curve. Therefore, the present invention can achieve early prediction of lithium-ion battery life by utilizing this feature.

[0006] The technical solution provided by the present invention is as follows:

[0007] A method for early prediction of lithium-ion battery life comprises the following steps:

[0008] (1) Obtaining model training data: n lithium-ion battery cells are numbered (distributed as 1, 2, 3, ···, n), and they are charged and discharged using a constant current and constant voltage method, and the charge and discharge cycle is repeated until the effective capacity of the battery decays to 80% of the nominal capacity; the capacity Q collected at equal voltage intervals during each charge and discharge cycle of the nth battery is recorded. n,i and voltage V n,iwhere i is the number of charge and discharge cycles, and each voltage value corresponds to a capacity value; Q n,10 is the capacity data of the nth battery during the 10th discharge process, Q n,80 The capacity data of the nth battery during the 80th discharge process is the capacity data difference Q at the same voltage. n,10-80 =Q n,10 -Q n,80 The discharge voltage V ranges from 2.2V to 3.6V and is the dependent variable. The function graph is plotted and the function formula Q is obtained by polynomial fitting. n,10-80 =f n (V);

[0009] (2) Calculate the standard deviation MSE of the numerical curve (f n (V)) as the input one-dimensional vector x, and the number of charge and discharge cycles at the end of the battery life as the output one-dimensional vector y;

[0010] (3) Early life prediction model training: The support vector machine optimized based on the humpback whale algorithm is used as the prediction algorithm to train the prediction model. During the training, the input vector x is used as the input of the model, and the output vector y is used as the output of the model. The model is trained to obtain the prediction model;

[0011] (4) Early prediction of battery life: At an ambient temperature of 30°C, the battery cells to be predicted are charged with constant current and constant voltage using the charging current value in the intended application scenario. After being fully charged, they are discharged using the discharge current value in the intended application scenario. The MSE (f n (V)) is input as an input value into the prediction model described in step (3) to predict the battery cycle life.

[0012] Furthermore, in the step (1), n ​​is 200 to 500. Preferably, n is 300.

[0013] Furthermore, in the step (1), the current A in the constant current charging stage is set to A=4C-n / 100C, and the discharge is performed in a 3C constant current mode.

[0014] Furthermore, in the step (1), i is 50 to 100.

[0015] Furthermore, in step (2), MSE(f n ) is calculated as follows:

[0016]

[0017] Among them, f n (V) is the capacity data difference fitting curve function of the nth battery, f n(V), k is the number of values ​​in the voltage interval.

[0018] Furthermore, in the step (2), k is 100-800.

[0019] Furthermore, in step (2), when the effective capacity of the battery decays to 80% of the nominal capacity, the battery life ends.

[0020] Furthermore, the training method of using the support vector machine optimized based on the humpback whale algorithm as the prediction algorithm to train the prediction model includes the following steps:

[0021] Step ①: Read data: read vector x, vector y;

[0022] Step ②: Initialize the support vector machine SVM: call the svmtrain function in the Matlab7.10.0 function library to train the support vector machine model;

[0023] Step 3: Initialize the loss function ε: The loss function ε determines the expected regression error. Its value will affect the number of support vectors and generalization ability of the corresponding model. The larger the value of ε, the smaller the support vector of the corresponding model and the lower the prediction accuracy, and vice versa.

[0024] Step ④: Initialize the penalty factor parameter C and kernel function parameter σ:

[0025] The kernel function parameter σ is obtained from 1 to 300 by the humpback whale optimization algorithm:

[0026] A=2a·rand1-a

[0027] S = 2 rand2

[0028] D=|S·σ q，t -σ i，t |

[0029] D′=|σ q，t -σ i，t |

[0030]

[0031] Where: A and S are calculation coefficients, D is the distance between the current whale individual and the current optimal solution, D′ is the distance between the i-th whale and the prey; σ q，t is the current optimal parameter, preferably 150; σ i，t is a random parameter with a value range of 1 to 300; σ i，t+1are the optimized parameters; rand and rand2 are random numbers between [0, 1]; a=2-2t / T is the convergence factor, which decreases linearly from 2 to 0 as the iteration proceeds, t is the current number of iterations, T is the maximum number of iterations, and the preferred T is 300; p is a random value, which is 0 or 1; l is a random number between -1 and 1; b is a constant, which is 5.

[0032] The penalty factor parameter C is mainly used to balance the approximation error and model complexity. The larger the value of C, the smaller the corresponding model fitting error and the higher the degree of fitting to the data, but the greater the model complexity. C is generally taken as (1 to 1000); the parameter σ is obtained from 1 to 1000 through the humpback whale optimization algorithm:

[0033] A=2a·rand1-a

[0034] S = 2 rand2

[0035] D=|S·C q，t -C i，t |

[0036] D′=|C q，t -C i，t |

[0037]

[0038] Where: C q，t is the current optimal parameter, preferably 500; C i，t is a random parameter with a value range of 1 to 1000; C i，t+1 is the optimized parameter; rand and rand2 are random numbers between [0, 1]; a = 2-2t / T is the convergence factor, which decreases linearly from 2 to 0 as the iteration proceeds, t is the current iteration number, T is the maximum iteration number, and the preferred T is 300; p is a random value, which is 0 or 1; l is a random number between -1 and 1; b is a constant, which is 5;

[0039] Step ⑤: Optimized WOA-SVM prediction model: Compare the calculated predicted life with the measured life value, and continuously modify the kernel function parameter σ and the penalty factor parameter C to optimize the support vector machine and obtain the prediction model.

[0040] Furthermore, in step ③ of the training method, ε is 0.0001 to 0.01. Preferably, ε is 0.001.

[0041] Furthermore, in step ④ of the training method, the kernel function includes a polynomial kernel function, a radial basis function (RBF) kernel function and a sigmoid kernel function. Preferably, the kernel function is a radial basis function (RBF) kernel function.

[0042] Beneficial effects of the present invention

[0043] (1) The present invention can achieve early prediction of the life of lithium-ion batteries by calculating and transforming the attenuation curve and combining it with the prediction algorithm for training, and can effectively classify or grade the batteries according to their predicted life;

[0044] (2) The present invention can realize early prediction of battery life, providing an effective basis for battery research and development and improvement of manufacturing processes;

[0045] (3) The present invention can achieve effective management of the battery life and provide a basis for the maintenance and replacement of the battery;

[0046] (4) The present invention can promote battery management optimization, improve the consistency of battery packs, and enhance the output performance and safety of battery packs. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A schematic diagram of the process flow for early prediction of battery life;

[0048] Figure 2 is the discharge curve of the discharge process; Figure 2 (a) shows three discharge curves of a battery with a cycle life of 534 weeks. The solid line on the right is the discharge curve of the 10th discharge of the battery, the solid line in the middle is the discharge curve of the 80th discharge, and the dotted line on the left is the discharge curve of the 534th discharge; Figure 2 (b) shows three discharge curves of a battery with a cycle life of 788 weeks. The solid line on the right is the discharge curve of the 10th discharge of the battery, the solid line in the middle is the discharge curve of the 80th discharge, and the dotted line on the left is the discharge curve of the 788th discharge; Figure 2 (c) shows three discharge curves of a battery with a cycle life of 1054 weeks. The solid line on the right is the discharge curve of the 10th discharge of the battery, the solid line in the middle is the discharge curve of the 80th discharge, and the dotted line on the left is the discharge curve of the 1054th discharge;

[0049] Figure 3 The discharge voltage V-capacity difference Q 10-80 Numerical curves;

[0050] Figure 4 This is the cycle life diagram of No. 1, No. 2, No. 3, and No. 4 batteries. DETAILED DESCRIPTION

[0051] The present invention is further described below in conjunction with specific embodiments, but the content of the present invention is not limited thereto at all.

[0052] Example

[0053] Figure 1The life prediction flow chart is shown, and the steps are as follows:

[0054] (1) Model training data acquisition: 300 lithium-ion battery cells are numbered from 1 to 300. At an ambient temperature of 30°C, the 300 battery cells are charged using a constant current and constant voltage method. The current A in the constant current charging stage is set to A=4C-N / 100C; the discharge is discharged using a 3C constant current method until the battery discharge cut-off voltage is reached and the discharge stops. When the effective capacity of the battery decays to 80% of the nominal capacity, it represents the end of life, and the battery cell stops charging and discharging tests.

[0055] Record the capacity Q collected at equal voltage intervals during each charge and discharge cycle of the nth battery n,i and voltage V n,i where i is the number of charge and discharge cycles, and each voltage value corresponds to a capacity value; Q n,10 is the capacity data of the nth battery during the 10th discharge process, Q n,80 The capacity data of the nth battery during the 80th discharge process is the capacity data difference Q at the same voltage. n,10-80 =Q n,10 -Q n,80 The discharge voltage V ranges from 2.2V to 3.6V and is the dependent variable. The function graph is plotted and the function formula Q is obtained by polynomial fitting. n,10-80 =f n (V). Figure 1 As shown in the figure, the left one of the two solid lines is the 10th discharge curve, and the right one is the 80th discharge curve. A horizontal line is drawn with the discharge voltage of 2.5V, and the difference in capacity values ​​at the intersection of the two solid lines is the capacity data difference of 2.5V. In the discharge voltage V range of 2.2V to 3.6V, the real-time voltage value and capacity value are obtained every 0.01V. Figure 2 It is a curve diagram of discharge capacity difference-voltage.

[0056] Take 4 batteries randomly selected as the test data set, where the 4 batteries are numbered 1, 2, 3, and 4 respectively, and the remaining 296 batteries are used as the prediction model training data set as an example.

[0057] (2) Calculate the standard deviation MSE of the numerical curve (f n (V)) as the input one-dimensional vector x, and the number of charge and discharge cycles at the end of the battery life as the output one-dimensional vector y;

[0058] MSE(f n ) is calculated as follows:

[0059]

[0060] Among them, f n (V) is the capacity data difference fitting curve function of the nth battery, f n The average value of (V), f n (V) divided by k. k is the number of voltage interval values, which is 100-800. Preferably, k is 500.

[0061] (3) Early life prediction model training: A support vector machine optimized based on the humpback whale algorithm is used as the prediction algorithm to train the prediction model. 296 batteries are randomly selected during training. The above method is used to obtain the input vector x as the input of the model, and the output vector y as the output of the model. The model is trained to obtain the prediction model.

[0062] Use the training samples to train the support vector machine model. The steps are as follows:

[0063] Step ①: Read data: read vector x, vector y;

[0064] Step ②: Initialize the support vector machine SVM: call the svmtrain function in the Matlab7.10.0 function library to train the support vector machine model;

[0065] Step 3: Initialize the loss function ε: The loss function ε determines the expected regression error. Its value will affect the number of support vectors and generalization ability of the corresponding model. The larger the value of ε, the smaller the support vector of the corresponding model and the lower the prediction accuracy. Vice versa, ε is generally taken as (0.0001~0.01), and 0.001 is preferred;

[0066] Step ④: Initialize the penalty factor parameter C and kernel function parameter σ:

[0067] Common kernel functions of support vector machines include polynomial kernel function, radial basis function (RBF) kernel function, and sigmoid kernel function. This embodiment uses the RBF function, and the kernel function parameter σ is obtained from 1 to 300 through the humpback whale optimization algorithm:

[0068] A=2a·rand1-a

[0069] S = 2 rand2

[0070] D=|S·σ q，t -σ i，t |

[0071] D′=|σ q，t -σ i，t |

[0072]

[0073] Where: A and S are calculation coefficients, D is the distance between the current whale individual and the current optimal solution, D′ is the distance between the i-th whale and the prey; σ q，t is the current optimal parameter, preferably 150; σ i，t is a random parameter with a value range of 1 to 300; σ i，t+1 are the optimized parameters; rand and rand2 are random numbers between [0, 1]; a=2-2t / T is the convergence factor, which decreases linearly from 2 to 0 as the iteration proceeds, t is the current number of iterations, T is the maximum number of iterations, and the preferred T is 300; p is a random value, which is 0 or 1; l is a random number between -1 and 1; b is a constant, which is 5.

[0074] The penalty factor parameter C is mainly used to balance the approximation error and model complexity. The larger the value of C, the smaller the corresponding model fitting error and the higher the degree of fitting to the data, but the greater the model complexity. C is generally taken as (1 to 1000); the parameter σ is obtained from 1 to 1000 through the humpback whale optimization algorithm:

[0075] A=2a·rand1-a

[0076] S = 2 rand2

[0077] D=|S·C q，t -C i，t |

[0078] D′=|C q，t -C i，t |

[0079]

[0080] Where: C q，t is the current optimal parameter, preferably 500; C i，t is a random parameter with a value range of 1 to 1000; C i，t+1 is the optimized parameter; rand and rand2 are random numbers between [0, 1]; a = 2-2t / T is the convergence factor, which decreases linearly from 2 to 0 as the iteration proceeds, t is the current iteration number, T is the maximum iteration number, and the preferred T is 300; p is a random value, which is 0 or 1; l is a random number between -1 and 1; b is a constant, which is 5;

[0081] Step ⑤: Optimized WOA-SVM prediction model: Compare the calculated predicted life with the measured life value, and continuously modify the kernel function parameter σ and the penalty factor parameter C to optimize the support vector machine and obtain the prediction model.

[0082] (4) Early prediction of battery life: The remaining 4 batteries are used to test the prediction effect of the model. The input vector x is obtained using the above method as the input of the model. The output of the model is the predicted cycle life of the 4 batteries. The capacity change value Q of the 4 batteries during the discharge process is extracted. 10-80 like Figure 3 The cycle life prediction results are shown in Figure 4 As shown, the early prediction results of the battery life of the present invention are very close to the experimental results, and the errors of the prediction results of the four batteries are all less than 9%, indicating that the present invention can use the early charge and discharge cycle data of the first 80 weeks of the battery to effectively predict the battery cycle life.

[0083] For batteries with the same material formula, effective prediction can be achieved: the battery to be predicted is placed at an ambient temperature of 30°C, and the battery monomer to be predicted is charged with constant current and constant voltage using the charging current value in the intended application scenario. After full charge, it is discharged using the discharge current value in the intended application scenario. After 80 cycles of charge and discharge, the experiment is stopped. Extract Q during discharge 10 , Q 80 , calculate MSE(f n ) is input into the prediction model as an input value, and the battery cycle life predicted by the model can be obtained.

[0084] The above description is only a preferred specific implementation manner of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention should be included in the scope of protection of the invention.

Claims

1. A method for early prediction of lithium-ion battery life, characterized in that: The following steps are involved: (1) Obtaining model training data: n lithium-ion battery cells are numbered 1, 2, 3, ···, n, and charged using a constant current and constant voltage method, and discharged using a constant current method, and the charge and discharge cycle is repeated until the effective capacity of the battery decays to 80% of the nominal capacity; Record the capacity Q collected at equal voltage intervals during each charge and discharge cycle of the nth battery n,i and voltage V n,i where i is the number of charge and discharge cycles, and each voltage value corresponds to a capacity value; Q n,10 is the capacity data of the nth battery during the 10th discharge process, Q n,80 The capacity data of the nth battery during the 80th discharge process is the capacity data difference Q at the same voltage. n,10-80 =Q n,10 -Q n,80 As the dependent variable, the discharge voltage V range of 2.2V~3.6V is the independent variable, the function graph is drawn, and the function formula Q is obtained by polynomial fitting. n,10-80 = f n (V); (2) Calculate the standard deviation MSE of the numerical curve ( f n (V)) as the input one-dimensional vector x , and the number of charge and discharge cycles at the end of the battery life is used as the output vector y ; MSE( f n ) is calculated as follows: in, Fitting curve function for the capacity data difference of the nth battery, for The average value of The number of voltage interval values; (3) Early life prediction model training: The support vector machine optimized based on the humpback whale algorithm is used as the prediction algorithm to train the prediction model. The input vector x As input to the model, the output vector y As the output of the model, the model is trained to obtain a prediction model; (4) Early prediction of battery life: At an ambient temperature of 30°C, the battery cell to be predicted is charged with constant current and constant voltage using the charging current value in the intended application scenario. After being fully charged, it is discharged using the discharge current value in the intended application scenario, and the MSE during the discharge process is extracted ( f n (V)) is input as an input value into the prediction model described in step (3) to predict the battery cycle life.

2. The method according to claim 1, characterized in that: In the step (1), n ​​is 200-500.

3. The method according to claim 1, characterized in that: In the step (1), the current A in the constant current charging stage is set to A=4C-N / 100C, and the discharge is performed in a 3C constant current mode.

4. The method according to claim 1, characterized in that: In the step (1), i is 50-100.

5. The method according to claim 1, characterized in that: In the step (2), It ranges from 100 to 800.

6. The method according to claim 1, characterized in that: In step (2), when the effective capacity of the battery decays to 80% of the nominal capacity, the battery life ends.

7. The method according to claim 1, characterized in that: The training method of using the support vector machine optimized based on the humpback whale algorithm as the prediction algorithm to train the prediction model comprises the following steps: Step ①: Read data: read vector x, vector y; Step ②: Initialize the support vector machine SVM: call the svmtrain function in the Matlab7.10.0 function library to train the support vector machine model; Step 3: Initialize the loss function ε: The loss function ε determines the expected regression error. Its value will affect the number of support vectors and generalization ability of the corresponding model. The larger the value of ε, the smaller the support vector of the corresponding model and the lower the prediction accuracy, and vice versa. Step ④: Initialize the penalty factor parameter C and kernel function parameter σ: The kernel function parameter σ is obtained from 1 to 300 by the humpback whale optimization algorithm: Where: A and S are calculation coefficients, D is the distance between the current whale individual and the current optimal solution, is the distance between the i-th whale and its prey; is the current optimal parameter; is a random parameter with a value range of 1 to 300; are the optimized parameters; rand and rand2 are random numbers between [0, 1]; is the convergence factor, which decreases linearly from 2 to 0 as the iteration proceeds, t is the current iteration number, and T is the maximum iteration number; is a random number, whose value is 0 or 1; is a random number between -1 and 1; is a constant, with a value of 5; The penalty factor parameter C is mainly used to balance the approximation error and model complexity. The larger the value of C, the smaller the corresponding model fitting error and the higher the degree of fitting to the data, but the greater the model complexity. C is 1 to 1000; the parameter σ is obtained from 1 to 1000 through the humpback whale optimization algorithm: Where: is the current optimal parameter; is a random parameter with a value range of 1 to 1000; are the optimized parameters; rand and rand2 are random numbers between [0, 1]; is the convergence factor, which decreases linearly from 2 to 0 as the iteration proceeds, t is the current iteration number, and T is the maximum iteration number; is a random number, whose value is 0 or 1; is a random number between -1 and 1; is a constant, with a value of 5; Step ⑤: Optimized WOA-SVM prediction model: Compare the calculated predicted life with the measured life value, and continuously modify the kernel function parameter σ and the penalty factor parameter C to optimize the support vector machine and obtain the prediction model.

8. The method according to claim 7, characterized in that: In step ③ of the training method, ε is 0.0001 to 0.

01.

9. The method according to claim 7, characterized in that: In step ④ of the training method, the kernel function includes a polynomial kernel function, a radial basis function (RBF) kernel function and a sigmoid kernel function.

Images