CN114839536A - A method for estimating the state of health of lithium-ion batteries based on multiple health factors - Google Patents

Abstract

The invention discloses a lithium ion battery health state estimation method based on multiple health factors, belongs to the technical field of battery management, and mainly solves the problem that the estimation accuracy of the battery health state is not high under the condition of quick charging. Based on voltage and current test data in a battery rapid charge-discharge cycle experiment, health factors are extracted from a constant-current charging process to form a characteristic vector, and the characteristic vector comprises charging time, charging energy and information entropy in a local voltage interval in the charging process. And establishing a Gaussian process regression prediction model by taking the feature vector as input and the battery SOH as output, and training the Gaussian process regression model by using experimental data. And in an online state, acquiring an input feature vector, inputting the input feature vector into a trained Gaussian process regression model, and predicting the SOH of the battery. The method does not need to establish a complex battery physical model, can realize the online evaluation of the SOH of the battery by a data driving method, and has very high accuracy and better universality.

Description

A method for estimating the state of health of lithium-ion batteries based on multiple health factors

technical field

The invention relates to a method for estimating the state of health of a lithium ion battery based on multiple health factors, which belongs to the technical field of power battery management and is used for estimating the state of health of a vehicle powered lithium ion battery under the condition of fast charging.

Background technique

In the field of electric vehicles, lithium-ion battery packs are the main energy storage device. With the prolongation of use time, the performance of the battery will decline, affecting its usable capacity and output power, etc., and may also cause safety problems under certain conditions. The battery industry generally uses the battery state of health (SOH) to characterize the degree of performance degradation of the battery, and stipulates that the SOH of the new power battery is 1 or 100%; when the battery performance cannot meet the use requirements, the battery is considered to be invalid and the service life is over. The SOH value of battery failure defined in different applications is different. For pure electric vehicles mainly based on energy demand, it is considered that when the power battery SOH=70%, the battery cannot meet the normal requirements, that is, the battery life is over, and the power For demand-based hybrid vehicles, the power battery SOH=80% is often used as the termination condition for the power battery. Therefore, proper management and monitoring of lithium-ion batteries, and accurate assessment of battery health status can ensure the performance of electric vehicles, effectively prevent battery abuse, and avoid safety accidents.

Fast charging technology is the entry point for electric vehicles to gain more recognition. Compared with ordinary charging methods, high-current fast charging can save charging time, which can be compared with the traditional car refueling time. However, fast charging is more likely to cause internal side reactions and overheating of the battery, which has potential damage to the battery electrode structure and may lead to rapid changes in the battery's health status.

Estimation of the state of health of lithium-ion batteries based on the conventional charging process is usually based on the health factors extracted from the battery charging experimental data to estimate the battery state of health, such as constant current charging time, constant voltage charging time, etc. The extraction of these factors relies on the conventional constant current and constant voltage charging mode. Under the condition of high-current charging, the charging current is often distributed in steps, with a very large current at first, and after reaching the limit voltage, the current decreases, and continues to charge to the limit voltage, and the process is repeated. Therefore, charging is segmented, and the length of charging time is closely related to the amount of current used. Under such conditions, just choosing the constant current charging time can no longer effectively estimate the state of health of the battery.

In the practice of battery state of health estimation, due to the inconsistency within the battery, even the same batch of batteries, the degree of performance degradation is not uniform, and even there is a large difference. Using multiple health factors to estimate the state of health of the battery can effectively solve the limitation of a single health factor and estimate the state of health of the battery more accurately.

This patent proposes an estimation method based on multiple health factors, which is used to estimate the state of health of batteries under the current fast charging conditions, and designs a method for extracting health factors based on partial voltage intervals. The battery health can also be achieved for batteries that do not use conventional charging methods. Accurate estimation of state.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a method for estimating the state of health of a lithium-ion battery under the condition of rapid charging: extracting three health factors of charging time, charging energy and information entropy in the local voltage interval during the rapid charging of the battery, using Gaussian Process regression (GPR) model to estimate battery SOH. This method can realize the battery state of health estimation of the battery under the high-current fast charging strategy, and the estimation accuracy is greatly improved compared with a single health factor.

The specific implementation steps are as follows:

S1. Extract the charging time in the local voltage interval [V _a , V _b ] during the high-current charging process as the first health factor, which can be obtained by using formula (1).

t=t _b -t _a (1)

Among them, t _a is the corresponding charging time when the charging voltage rises to V _a , t _b is the corresponding charging time when the charging voltage rises to V _b , and t is the corresponding charging time in the [V _a , V _b ] voltage interval. The voltage range can be selected according to the battery type, wherein the corresponding voltage of the lithium iron phosphate battery is Va ₌ 3.15V, Vb = _3.55V .

S2 extracts the charging energy in the local voltage interval [V _a , V _b ] during the high-current charging process as the second health factor, which can be obtained by formula (2)

Among them, t _a is the corresponding charging time when the charging voltage rises to V _a , t _b is the corresponding charging time when the charging voltage rises to V _b , I is the charging current, and V(t) is the time-varying voltage.

S3 extracts the information entropy in the partial voltage interval [V _a , V _b ] as the third health factor in the process of high-current charging, and can be obtained by formula (3)

Among them, the entropy index E _k is represented by the distribution of the charging voltage in _{a certain voltage range, and the voltage range is determined by the minimum voltage value Va and the maximum voltage value V b .} The voltage range is divided into a fixed number of intervals, that is, the voltage range [V _a , V _b ] is divided into a small interval every 0.1V, M is the number of small intervals, p(i) indicates that the voltage measurement value is in each small interval. The frequency of occurrence of the voltage interval.

S4 uses the three health factors extracted above as input and the health status of the battery as output to build a training data set and a prediction data set

为步骤(1)、 (2)、(3)中所提取的健康因子，ε_i为服从高斯分布的均值为0，方差为σ_n的高斯噪声，表示为公式(4)，y_i为i时刻电池的健康状态。f(HF_i)为关于健康因子的函数，属于高斯过程，表示为

该过程由均值函数和协方差函数决定，分别表示为公式(5)和(6)S5 uses the Gaussian process regression algorithm to establish the model, that is, the Gaussian process regression model y _i =f(HF _i )+ε _i is established with HF _i and y _i as the input and output respectively, where,

is the health factor extracted in steps (1), (2) and (3), ε _i is Gaussian noise with a Gaussian distribution with a mean value of 0 and a variance of σ _n , expressed as formula (4), y _i is i The health status of the battery at all times. f(HF _i ) is a function of the health factor, belonging to a Gaussian process, expressed as

The process is determined by the mean function and covariance function, which are expressed as equations (5) and (6), respectively

ε _i ～N(0,σ _n ² ) (4)

_mHF = E(f(HF)) (5)

The selected mean function and covariance function are 0 and Matern5/2, respectively, where Matern5/2 is expressed as formula (7)

σ_f和σ_l为协方差函数的超参数。in,

σ _f and σ _l are hyperparameters of the covariance function.

S6. Import the training data set into the Gaussian process regression model for training, and obtain and optimize the hyperparameters of the model. In the model established in step 5, there are hyperparameters Θ=[σ _n , σ _l , σ _f ], and the training data is used to optimize the hyperparameters of the model to obtain optimal results. The hyperparameters are optimized by maximizing the log-marginal likelihood function, as shown in Equation (8):

The function consists of three parts, the first part is the data fitting term, which represents the fitting degree of the hyperparameters; the second part is the complexity penalty term, which is used to prevent overfitting; the third part is the constant term. The gradient ascent method is used to optimize the hyperparameters, and the partial derivative of formula (8) is obtained:

where β=(K _{HF,HF +} σ _n ² In ) ^-1 y.

After the optimized hyperparameters are obtained through formulas (8) and (9), a new input HF' is given to the model, and the predicted value y' of the battery state of health is output.

S7 imports the prediction data set into the trained model for verification, and judges the accuracy of the model by root mean square error and mean absolute error.

When S8 is online, the extracted three health factors are used as the input vector of the Gaussian process regression model, and the model outputs the health status of the battery.

Description of drawings

Figure 1 shows the predicted SOH of the No. 1 battery based on the GPR model of a single health factor.

Figure 2 shows the predicted SOH of the No. 1 battery based on the multi-health factor GPR model.

Figure 3 shows the predicted SOH of the No. 2 battery based on the GPR model of a single health factor.

Figure 4 shows the SOH predicted by the GPR model of No. 2 battery based on multiple health factors.

Figure 5 shows the SOH predicted by the GPR model based on a single health factor for the No. 3 battery.

Figure 6 shows the SOH predicted by the GPR model of No. 3 battery based on multiple health factors.

Figure 7 shows the predicted SOH of battery 4 based on the GPR model of a single health factor.

Figure 8 shows the SOH predicted by the GPR model based on multiple health factors for the No. 4 battery.

Figure 9 shows the SOH predicted by the GPR model of the AA battery based on a single health factor.

Figure 10 shows the SOH predicted by the GPR model of the 5th battery based on multiple health factors.

Figure 11 shows the SOH predicted by the GPR model of the AA battery based on a single health factor.

Figure 12 shows the SOH predicted by the GPR model of the AA battery based on multiple health factors.

FIG. 13 is a schematic flowchart of the implementation of the method of the present invention.

Detailed ways

The technical solutions of the present invention will be described in detail below with reference to the accompanying drawings and implementation cases.

According to the above method, using the cycle charge and discharge data of laboratory lithium iron phosphate batteries, the SOH was estimated and calculated for 6 batteries under 3 different charging strategies.

Table 1 Description of the charging strategy and discharge test conditions of the test battery

Implementation case:

Batteries 1-6 were charged and discharged under the conditions shown in Table 1, respectively, and the data set was divided into training data and test data based on the voltage, current and capacity data information obtained from the charge and discharge cycles. In the training data, t, E, 1/E _k are extracted as the input of the multivariate GPR model, and SOH is used as the output to train the GPR model. In the test data set, the multi-health factor SOH estimation results of 6 batteries were obtained by using the established model. In addition, in order to compare the superiority of the multi-health factor estimation method, the battery state of health was estimated by taking the charging time t of the battery in the same voltage range as the health factor, and compared with the estimation method with the multi-health factor as the input. Since the mean absolute error (MAE) can better reflect the actual situation of the predicted value error, the root mean square error (RMSE) is very sensitive to very large or very small errors in a set of measurements, so it can reflect the prediction well. accuracy. Therefore, the mean absolute error (MAE) and the root mean square error (RMSE) are used to measure the estimation accuracy of the two methods, as shown in Table 2.

Table 2 SOH estimation errors of single health factor and multiple health factors

From Table 2, the estimation accuracy of multiple health factors is very high, the MAE is in the range of [0.0039, 0.0058], and the RMSE is in the range of [0.00480.0076]. For cell 2, the MAE error decreased from 0.0193 to 0.0039, and the RMSE decreased from 0.0244 to 0.0048. For the No. 6 battery, the MAE error decreased from 0.0149 to 0.0053, and the RMSE decreased from 0.0187 to 0.0061. The SOH estimation accuracy of the multi-health factor for the 6 batteries was improved by at least 37% compared with the estimation accuracy of the single health factor, and except for the 1 and 1 and Except for No. 5 battery, the estimation accuracy of the remaining batteries is improved by more than 50%, and the error analysis shows that the estimation accuracy of multiple health factors has been significantly improved compared with the single health factor.

From the above applications, the single-feature estimation method can realize the SOH estimation of the battery, but due to the limitation of a single health feature in reflecting the health status of the battery, the SOH estimation accuracy of the battery is low, which is different from the single-feature estimation. The difference between the methods is that the estimation method of multiple health features can make up for the defect of the limitation of a single feature and greatly improve the estimation accuracy. In addition, since the health features are extracted based on partial voltage intervals, it is possible to estimate the state of health of the battery under the condition of incomplete charging.

The above examples can effectively prove the superiority of the method of the present invention: the use of fast charging experimental data to extract the health factor realizes the estimation of the battery SOH under the condition of fast charging, and can accurately estimate the state of health of the battery under the condition of incomplete charging of the battery, And compared with the single factor estimation method, this method can make up for the deficiency of the single feature estimation method and greatly improve the estimation accuracy. In addition, the method of the present invention has been verified under 6 different batteries and 3 different high-current charging conditions, and has good generality.

Claims (1)

1. A lithium-ion battery state-of-health estimation method based on multiple health factors, characterized in that: the method takes the charging time t, the charging energy E and the information entropy E _k in the local voltage interval of the fast charging process as the health factors, and uses Gaussian The process regression algorithm establishes the battery capacity degradation model, and finally determines the state of health SOH of the battery; The specific implementation steps are as follows: Step (1): Extract the charging time in the local voltage interval [V _a , V _b ] in the process of high current charging as the first health factor, and use formula (1) to get: t=t _b -t _a (1) Among them, t _a is the corresponding charging time when the charging voltage rises to V _a , t _b is the corresponding charging time when the charging voltage rises to V _b , and t is the corresponding charging time in the [V _a , V _b ] voltage interval; Step (2): Extract the charging energy in the local voltage interval [V _a , V _b ] during the high-current charging process as the second health factor, and use formula (2) to obtain:

Among them, t _a is the corresponding charging time when the charging voltage rises to V _a , t _b is the corresponding charging time when the charging voltage rises to V _b , I is the charging current, and V(t) is the time-varying voltage; Step (3): Extract the information entropy in the partial voltage interval [V _a , V _b ] as the third health factor in the process of high-current charging, and use formula (3) to obtain:

Among them, the entropy index E _k is represented by the distribution of charging voltage in a certain voltage range, and the voltage range is determined by the minimum voltage value Va and the maximum voltage value V _b ; the voltage range is divided into _a fixed number of intervals, that is, the voltage range [V _a , V _b ] is divided into a small voltage interval every 0.1V, M is the number of small voltage intervals, p(ii) represents the frequency of the voltage measurement value appearing in each small voltage interval, ii is the serial number of the voltage interval number; Step (4): using the extracted three health factors as input and the health status of the battery as output, establish a training data set and a prediction data set; Step (5): use the Gaussian process regression algorithm to establish a model, that is, use HF _i and y _i as input and output respectively to establish a Gaussian process regression model y _i =f(HF _i )+ε _i , wherein,

is the health factor extracted in steps (1), (2) and (3), ε _i is Gaussian noise with a Gaussian distribution with a mean of 0 and a variance of σ _n , expressed as formula (4), y _i is i The health state of the battery at the moment; f(HF _i ) is a function of the health factor, which belongs to a Gaussian process and is expressed as

The Gaussian process is determined by the mean function and the covariance function, which are expressed as formulas (5) and (6), respectively ε _i ~N(0,σ _n ² ) (4) _mHF = E(f(HF)) (5)

The selected mean function and covariance function are 0 and Matem5/2, respectively, where Matern5/2 is expressed as formula (7)

in,

σ _f and σ _l are the hyperparameters of the covariance function; Step (6): import the training data set into the Gaussian process regression model for training, and obtain and optimize the hyperparameters of the model; in the Gaussian process regression model established in step (5), there are hyperparameters Θ=[σ _n , σ _l , σ _f ], use the training data to optimize the hyperparameters of the Gaussian process regression model to obtain the optimal results; use the maximizing logarithmic marginal likelihood function to optimize the hyperparameters, as shown in formula (8):

The maximum log-marginal likelihood function consists of three parts. The first part is the data fitting term, which represents the degree of fitting of the hyperparameters; the second part is the complexity penalty term, which is used to prevent overfitting; the third part is the Constant term; the hyperparameters are optimized by gradient ascent method, and the partial derivative of formula (8) is obtained:

Wherein, β=(K _{HF, HF} +σ _n ² I _n ) ^-1 y; After the optimized hyperparameters are obtained by formulas (8) and (9), a new input HF' is given to the model, and the predicted value y' of the battery state of health is output; Step (7): import the prediction data set into the trained model for verification, and judge the accuracy of the model with root mean square error and mean absolute error; Step (8): In the online case, using the extracted three health factors as the input vector of the Gaussian process regression model, the model can output the health state of the battery.

Images