CN111880099A - Method and system for predicting service life of battery monomer in energy storage power station - Google Patents

Abstract

The invention discloses a method and a system for predicting the cycle life of a battery cell in an energy storage power station. The technical scheme adopted in the prediction method of the present invention is as follows: collecting a plurality of historical test data of battery capacity cycle degradation, extracting preliminary features reflecting battery degradation information, screening the preliminary features through an elastic network, and extracting the sensitivity to the impact of battery cycle life prediction results. The secondary features with a high degree are used as the final training features to prevent overfitting of training, and then the neural network model is trained by using the selected secondary features, and finally the optimal weight matrix of the neural network model is obtained, and the training is used to train the neural network model. The completed neural network model predicts the future battery life. The invention can ignore the specific type of the battery and directly use its operation data for prediction without considering the specific structure and structure inside the battery. Compared with the battery model prediction method, the invention has better universality and simplicity.

Description

A method and system for predicting the life of battery cells in an energy storage power station

technical field

The invention belongs to the field of battery state evaluation, and in particular relates to a data-driven battery cell life prediction method and system in an energy storage power station station.

Background technique

At present, electrochemical energy storage power stations have been widely used in power systems due to their fast charging and discharging and flexible control. or lack of important role. However, with the cyclic use and frequent charging and discharging of the battery cells in the electrochemical energy storage power station, their availability will be damaged to varying degrees, and the battery life will be attenuated with continuous use. Reducing the use efficiency of electrical equipment will also cause huge damage to the life of the energy storage power station itself. Therefore, how to use the historical operation of the battery to reasonably and effectively evaluate the health status of the battery and accurately predict the future service life of the battery is indispensable for the efficient use, timely maintenance and power replacement of the battery service objects. effect.

Most of the existing mainstream battery life prediction methods are based on battery models to predict their lifespans, which have inaccurate and inaccurate defects. In addition, the establishment of battery models requires a deep understanding of the internal mechanism of the battery and its electrochemical reaction process, which has certain limitations. The complexity of the battery, and different types of batteries have different models, so they are not universal.

With the development of big data and artificial intelligence technology, it has become possible to predict battery life using massive historical battery operation and measurement data.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a data-driven method and system for predicting the life of a battery cell in an energy storage power station, which utilizes historical battery cycle charge and discharge data to predict the future use of a battery cell in an electrochemical energy storage power station. The change of its life with the increase of the number of cycles of charge and discharge.

In order to achieve the above purpose of the invention, the technical solution adopted by the present invention to solve the technical problem is: a battery cell life prediction method in an energy storage power station, which collects a plurality of historical test data of cyclic degradation of battery capacity, and extracts information reflecting battery degradation. The initial features are screened through the elastic network, and the secondary features that have a high degree of influence on the prediction results of the remaining battery life are extracted as the final training features to prevent over-fitting of training, and then use the filtered secondary features to The neural network model is trained, and the optimal weight matrix of the neural network model is finally obtained, and the trained neural network model is used to predict the future life of the battery.

For the first time, the present invention considers the screening of primary features measured during battery operation, and extracts features with a high degree of influence on the prediction results as secondary features, thereby effectively avoiding the dimensionality disaster of data and over-fitting of training, and utilizes the second feature. This method can ignore the specific type of the battery and directly use its operating data to make predictions without considering the specific structure and structure inside the battery. Compared with the battery model prediction method, it has better generality. Suitability and simplicity.

Further, the battery life in the energy storage power station refers to: from the time of the test, before the battery health state reaches the ratio of the minimum allowable capacity of the battery to the maximum capacity of the battery at the factory, the maximum number of cycles of charging and discharging the battery can perform, and the ratio is in In the industry, the general value is 80%;

The extraction is used to input the preliminary features of the elastic network, which is specifically: through the test data of the battery cell during the charging/discharging process of the first n cycles, preliminary extraction of 20 features is used as the input for the elastic network to iterate. initial features;

The screening of the preliminary features through the elastic network is specifically: introducing a regular term P _a (β) into the cost function J (β), and solving the coefficient of the regularized linear regression model that minimizes the cost function, where a is the coefficient of the elastic network, and β is the regularized linear regression coefficient vector;

The quadratic feature with high sensitivity to the impact of the battery cycle life prediction result is extracted as the final training feature, which is specifically: screening the quadratic feature X corresponding to the coefficient that is not 0 in the obtained regression coefficient vector β , used for training the neural network model;

The training of the neural network model by using the selected secondary features is specifically as follows: the secondary feature X of the first n cycles of the battery is used as the input of the model, the life span Y is used as the output of the model, and the Bayesian iterative method is adopted. method, using (X, Y) to solve the optimal weight matrix W of the neural network model;

The described use of the trained neural network model to predict the future life of the battery is specifically: in the same way of extracting the secondary feature X of the battery used for training the neural network model, the battery is charged from the first n cycles of the battery to be predicted. Extract secondary features X _p from the data of the discharge process, that is, first extract 20 initial features from the data of the first n cycles of cyclic charge/discharge process data of the battery to be predicted, and then filter out the secondary features X _p through the same elastic network, As the input of the neural network model, the neural network model will output the prediction result of the future life of the battery.

Further, the expression of the ratio of the minimum allowable discharge capacity of the battery to the initial maximum discharge capacity of the battery is:

Among them, Qi _,min is the minimum discharge capacity allowed by the battery, Qi _,0 is the initial maximum discharge capacity of the battery, and i is the battery serial number.

Further, the preliminary features described are:

Feature 1, the initial maximum discharge capacity Q _i,0 of the battery;

Feature 2, the maximum discharge capacity Q _i,n of the nth cycle of the battery;

Feature 3, the difference Q _i,n -Q _i,0 between the maximum discharge capacity of the battery in the nth cycle and the initial maximum discharge capacity;

Feature 4, the maximum capacity change Q _i,1 -Q _i,0 of the battery after the battery is charged and discharged in the first cycle;

Feature 5, the first-order coefficient of the least squares linear fitting during the first n cycles of the battery;

Feature 6, the constant term coefficient of the least squares linear fitting during the first n cycles of the battery;

Feature 7, the first-order coefficient of the least squares linear fitting during the first 90 to the nth cycle of the battery;

Feature 8, the constant term coefficient of the least squares linear fitting during the first 90 to the nth cycle of the battery;

Feature 9, the minimum temperature T _imin during the first n cycles of the battery;

Feature 10, the maximum temperature T _imax during the first n cycles of the battery;

其中，t_i为第i次循环所耗时间，T_i为第i次循环过程中的温度；Feature 11, the average value of temperature × time during the first n cycles of the battery:

Wherein, t _i is the time spent in the i-th cycle, and T _i is the temperature during the i-th cycle;

Feature 12, the average time to discharge the battery during the first n cycles

Feature 13, the total time during the first n cycles of the battery

Feature 14, the minimum value of the capacity change of the battery under different voltages;

Feature 15, the average value of the capacity change of the battery under different voltages;

Feature 16, the divergence value of the capacity change of the battery under different voltages;

Feature 17, the kurtosis coefficient value of the capacity change of the battery under different voltages;

Feature 18, the initial internal resistance value of the battery

Feature 19, the internal resistance value of the nth cycle of the battery

Feature 20, the difference between the maximum value and the minimum value of the internal resistance in the first n cycles of the battery.

Further, the expressions of the cost function J(β) and the regular term P _α (β) are:

Among them, y _i is the life of the battery set used for training in the i-th cycle, x _i ^T is the feature initially extracted in the i-th cycle, β ₀ is the initial value of the regression coefficient vector, and β is the regression coefficient to be solved. vector, λ is the regular term coefficient, α is the coefficient of the elastic network, β ₁ , β ₂ are the 1-norm and 2-norm of β, respectively; using the above expression, set reasonable λ and α, to find the J (β) The smallest regression coefficient vector β.

Further, the neural network model is simplified as:

W=g(X,Y),

Among them, g represents the training process of the BP neural network model, and the model is iteratively solved by the Bayesian method.

Another technical solution adopted by the present invention is: a battery cell life prediction system in an energy storage power station, which includes:

Historical test data collection unit: collects historical test data of cyclic degradation of battery capacity;

Preliminary feature extraction unit: extracts preliminary features reflecting battery degradation information;

Secondary feature extraction unit: Screen the preliminary features through the elastic network, and extract the secondary features that have a high degree of influence on the prediction results of the remaining battery life as the final training features;

Neural network model training unit: then use the selected secondary features to train the neural network model, and finally obtain the optimal weight matrix of the neural network model;

Future battery life prediction unit: Use the trained neural network model to predict the future battery life.

The present invention has the beneficial effects of screening the primary features measured during battery operation for the first time, and extracting features with high sensitivity to the battery cycle life prediction result as secondary features, thereby effectively avoiding the dimensionality of the data Disaster and overfitting of training, and use the secondary feature to train the neural network model, and use the trained neural network model to predict the future life of the battery. The invention can ignore the specific type of the battery and directly use its operation data for prediction without considering the specific structure and structure inside the battery. Compared with the battery model prediction method, the invention has better universality and simplicity.

Description of drawings

Fig. 1 is the flow chart of the battery cell life prediction method in the energy storage power station of the present invention;

Fig. 2 is the comparison chart of the life expectancy result of lithium iron phosphate battery and the real life life in the application example of the present invention (Fig. 2a is the contrast chart of the life expectancy and the real life expectancy using the training sample prediction, Fig. 2b is the life expectancy using the test sample prediction and the real life span Comparison chart, Figure 2c is a comparison chart between the predicted lifespan and the real lifespan using all samples);

FIG. 3 is a graph showing the distribution of life prediction errors of lithium iron phosphate batteries in an application example of the present invention;

FIG. 4 is a structural block diagram of a battery cell life prediction system in an energy storage power station according to the present invention.

Detailed ways

The present invention will be further elaborated and described below with reference to the accompanying drawings and embodiments. The technical features of the various embodiments of the present invention can be combined correspondingly on the premise that there is no conflict with each other.

Example 1

This embodiment provides a data-driven method for predicting the life of a battery cell in an energy storage power station. Preliminary features are screened, and secondary features that are highly sensitive to the impact of battery cycle life prediction results are extracted as final training features to prevent overfitting of training, and then the selected secondary features are used to train the neural network model, and finally The optimal weight matrix of the neural network model is obtained, and the trained neural network model is used to predict the future life of the battery.

The specific steps of the above prediction method are as follows:

First, the ratio of the minimum discharge capacity allowed by the battery to the initial maximum discharge capacity of the battery is specified as:

Among them, Qi _,min is the minimum discharge capacity allowed by the battery, Qi _,0 is the initial maximum discharge capacity of the battery, and i is the battery serial number. The life of the battery is the number of cycles of charge and discharge when the battery's health status reaches SOH _i,min from the time of the test.

According to the test data of the battery cell during the charging/discharging process of the first n cycles, 20 features are initially extracted as the initial features for inputting the elastic network for iteration. The initial features are as follows:

其中i为电池序号；Feature 1, the initial maximum discharge capacity of the battery

where i is the battery serial number;

Feature 2, the maximum discharge capacity of the nth cycle of the battery

Feature 3, the difference between the maximum discharge capacity of the nth cycle of the battery and the initial maximum discharge capacity

Feature 4, the maximum capacity change of the battery after the battery is charged and discharged in the first cycle;

Feature 5, the first-order coefficient of the least squares linear fitting during the first n cycles of the battery;

Feature 6, the constant term coefficient of the least squares linear fitting during the first n cycles of the battery;

Feature 7, the first-order coefficient of the least squares linear fitting during the first 90 to the nth cycle of the battery;

Feature 8, the constant term coefficient of the least squares linear fitting during the first 90 to the nth cycle of the battery;

Feature 9, the minimum temperature T _imin during the first n cycles of the battery;

Feature 10, the maximum temperature T _imax during the first n cycles of the battery;

其中，t_i为第i次循环所耗时间；Feature 11, the average value of temperature × time during the first n cycles of the battery

Among them, t _i is the time consumed by the i-th cycle;

Feature 12, the average time to discharge the battery during the first n cycles

Feature 13: Total time during the first n cycles of the battery

Feature 14: The minimum value of the capacity change of the battery under different voltages;

Feature 15: The average value of battery capacity changes under different voltages;

Feature 16: Divergence value of battery capacity change under different voltages;

Feature 17: The kurtosis coefficient value of the battery capacity change under different voltages;

Feature 18: The initial internal resistance value of the battery

Feature 19: Internal resistance value of the nth cycle of the battery

Feature 20: The difference between the maximum value and the minimum value of the internal resistance in the first n cycles of the battery.

Next, according to the above features, construct the cost function J(β):

Among them, y _i is the life of the battery set used for training in the ith cycle, x _i ^T is the feature initially extracted in the ith cycle, β ₀ is the initial value of the regression coefficient, β is the regression coefficient to be solved, λ is the regular term coefficient, P _α (β) is the regular term, and its expression is:

Among them, α is the coefficient of the elastic network, and β ₁ and β ₂ are the 1-norm and 2-norm of β, respectively. Using the above expressions, select reasonable λ and α, and solve the β that minimizes J(β).

Next, the features corresponding to the coefficients that are not 0 in the obtained regression coefficient vector β are selected as the secondary features X used for model training. Use the selected secondary features to train the neural network model, take the secondary features X of the first n cycles of the battery as the input of the model, and the life Y as the output of the model, using the Bayesian iteration method, using (X, Y ) to solve the optimal weight matrix W of the neural network model;

W=g(X,Y).

Finally, the trained neural network model is used to predict the lifespan of battery cells in the same type of energy storage power station. Extract the secondary feature Xp from the first _n cycle charge/discharge process data of the battery cell to be predicted in the same way as the secondary feature X of the battery used for training the neural network model, that is, firstly from the battery to be predicted life 20 initial features are extracted from the first n cycles of cyclic charge/discharge process data, and then the secondary features X _p are screened out through the same elastic network as the input of the neural network model, which will output the battery's future life prediction results.

Application example

The method of the present invention is specifically described below by taking the public data set of battery cycle life test of Stanford University as an example.

The test set includes 124 lithium iron phosphate battery cell cycle life test data. The data set uses lithium iron phosphate battery cells with a rated voltage of 3.3V and a capacity of 1.1Ah to conduct cycle life tests at different currents. The battery life ranges from 150 to 2300 cycles of charge and discharge cycles. Data such as voltage, current, capacity, temperature, time, internal resistance, etc. during cyclic charge/discharge.

Let λ=0.00369, α=0.9, through elastic network iteration, 10 main eigenvalues that have a great influence on the result are screened out from 20 eigenvalues, which are features 1, 3, 4, 5, 10, 11, 13, 15, 19, 20, among which feature 15 has the highest influence on battery life measurement, and the above features are selected as the secondary features of this experiment.

After obtaining the above secondary features, the neural network model is trained using the lithium iron phosphate battery training data set. After the model training is completed, the secondary features of the historical data sets of the lithium iron phosphate battery used for training and testing are respectively used. As input, the lifetimes of 124 batteries are predicted and compared with the real lifetimes. The predictions are shown in Figure 2. For the life prediction of the battery cells used for training, the fitting degree of the training sample is R=0.99224 (see Figure 2a), and the fitting degree of the test sample is R=0.96536>0.95 (see Figure 2b), with high accuracy The fitting degree of all samples is R=0.98995 (see Figure 2c).

The sample prediction error is shown in the figure. It can be seen from Figure 3 that the prediction error of the training sample is mostly concentrated in the interval of [-60,60], the maximum prediction is not more than 127, and the error of the test sample is mostly concentrated. In the interval [-14,64], the maximum error does not exceed 115.

Example 2

This embodiment provides a battery cell life prediction system in an energy storage power station, which includes:

Historical test data collection unit: collects historical test data of cyclic degradation of battery capacity;

Preliminary feature extraction unit: extracts preliminary features reflecting battery degradation information;

Secondary feature extraction unit: Screen the preliminary features through the elastic network, and extract the secondary features that have a high degree of influence on the prediction results of the remaining battery life as the final training features;

Neural network model training unit: then use the selected secondary features to train the neural network model, and finally obtain the optimal weight matrix of the neural network model;

Future battery life prediction unit: Use the trained neural network model to predict the future battery life.

The battery life in the energy storage power station refers to: from the time of the test, before the battery health state reaches the ratio of the minimum allowable capacity of the battery to the maximum capacity of the battery at the factory, the maximum number of cycles of charge and discharge that the battery can perform;

The extraction of preliminary features reflecting the battery degradation information is specifically as follows: through the test data of the battery cell during the first n cycles of cyclic charge/discharge, 20 features are initially extracted as the initial features for inputting the elastic network;

The screening of the preliminary features through the elastic network is specifically: introducing a regular term P _a (β) into the cost function J (β), and solving the coefficient of the regularized linear regression when the cost function is minimized, where a is Coefficient of elastic network, β is a vector of regularized linear regression coefficients;

The extraction of the secondary features with a high degree of influence on the prediction result of the remaining service life of the battery is used as the final training feature. For training of neural network models;

The training of the neural network model by using the selected secondary features is specifically as follows: the secondary feature X of the first n cycles of the battery is used as the input of the model, the life span Y is used as the output of the model, and the Bayesian iterative method is adopted. method, using (X, Y) to solve the optimal weight matrix W of the neural network model;

The described use of the trained neural network model to predict the future life of the battery is specifically: in the same way of extracting the secondary feature X of the battery used for training the neural network model, the battery is charged from the first n cycles of the battery to be predicted. Extract secondary features X _p from the data of the discharge process, that is, first extract 20 initial features from the data of the first n cycles of cyclic charge/discharge process data of the battery to be predicted, and then filter out the secondary features X _p through the same elastic network, As the input of the neural network model, the neural network model will output the prediction result of the future life of the battery.

The expression of the ratio of the minimum allowable discharge capacity of the battery to the initial maximum discharge capacity of the battery is:

Among them, Qi _,min is the minimum discharge capacity allowed by the battery, Qi _,0 is the initial maximum discharge capacity of the battery, and i is the battery serial number.

The preliminary features described are:

Feature 1, the initial maximum discharge capacity Q _i,0 of the battery;

Feature 2, the maximum discharge capacity Q _i,n of the nth cycle of the battery;

Feature 3, the difference Q _i,n -Q _i,0 between the maximum discharge capacity of the battery in the nth cycle and the initial maximum discharge capacity;

Feature 4, the maximum capacity change Q _i,1 -Q _i,0 of the battery after the battery is charged and discharged in the first cycle;

Feature 5, the first-order coefficient of the least squares linear fitting during the first n cycles of the battery;

Feature 6, the constant term coefficient of the least squares linear fitting during the first n cycles of the battery;

Feature 7, the first-order coefficient of the least squares linear fitting during the first 90 to the nth cycle of the battery;

Feature 8, the constant term coefficient of the least squares linear fitting during the first 90 to the nth cycle of the battery;

Feature 9, the minimum temperature T _imin during the first n cycles of the battery;

Feature 10, the maximum temperature T _imax during the first n cycles of the battery;

其中，t_i为第i次循环所耗时间，T_i为第i次循环过程中的温度；Feature 11, the average value of temperature × time during the first n cycles of the battery:

Wherein, t _i is the time spent in the i-th cycle, and T _i is the temperature during the i-th cycle;

Feature 12, the average time to discharge the battery during the first n cycles

Feature 13, the total time during the first n cycles of the battery

Feature 14, the minimum value of the capacity change of the battery under different voltages;

Feature 15, the average value of the capacity change of the battery under different voltages;

Feature 16, the divergence value of the capacity change of the battery under different voltages;

Feature 17, the kurtosis coefficient value of the capacity change of the battery under different voltages;

Feature 18, the initial internal resistance value of the battery

Feature 19, the internal resistance value of the nth cycle of the battery

Feature 20, the difference between the maximum value and the minimum value of the internal resistance in the first n cycles of the battery.

The expression of the cost function J(β) and the regular term P _α (β) is:

Among them, y _i is the life of the battery set used for training in the i-th cycle, x _i ^T is the feature initially extracted in the i-th cycle, β ₀ is the initial value of the regression coefficient vector, and β is the regression coefficient to be solved. vector, λ is the regular term coefficient, α is the coefficient of the elastic network, β ₁ , β ₂ are the 1-norm and 2-norm of β, respectively; using the above expression, set reasonable λ and α, to find the J (β) The smallest regression coefficient vector β.

The neural network model is simplified as:

W=g(X,Y),

Among them, g represents the training process of the BP neural network model, and the model is iteratively solved by the Bayesian method.

While the content of the present invention has been described in detail by way of the above preferred embodiments, it should be appreciated that the above description should not be construed as limiting the present invention. Various modifications and alternatives to the present invention will be apparent to those skilled in the art upon reading the foregoing. Accordingly, the scope of protection of the present invention should be defined by the appended claims.

Claims (10)

1. A method for predicting the life of battery cells in an energy storage power station, characterized in that, collecting a plurality of historical test data of cyclic degradation of battery capacity, extracting preliminary features reflecting battery degradation information, screening the preliminary features through an elastic network, and extracting The secondary features that have a high degree of influence on the prediction results of the remaining service life of the battery are used as the final training features, and then the selected secondary features are used to train the neural network model, and finally the optimal weight matrix of the neural network model is obtained. The completed neural network model predicts the future battery life. 2. The method for predicting the life of a battery cell in an energy storage power station according to claim 1, wherein, The battery life in the energy storage power station refers to: from the time of the test, before the battery health state reaches the ratio of the minimum allowable capacity of the battery to the maximum capacity of the battery at the factory, the maximum number of cycles of charge and discharge that the battery can perform; The extraction of the preliminary features reflecting the battery degradation information is specifically as follows: through the test data of the battery cell during the first n cycles of cyclic charge/discharge, preliminary extraction of 20 features is used as the initial features for inputting the elastic network; The screening of the preliminary features through the elastic network is specifically: introducing a regular term P _a (β) into the cost function J (β), and solving the coefficient of the regularized linear regression when the cost function is minimized, where a is Coefficient of elastic network, β is a vector of regularized linear regression coefficients; The extraction of the secondary features with a high degree of influence on the prediction result of the remaining service life of the battery is used as the final training feature. For training of neural network models; The training of the neural network model by using the selected secondary features is specifically as follows: the secondary feature X of the first n cycles of the battery is used as the input of the model, the life span Y is used as the output of the model, and the Bayesian iterative method is adopted. method, using (X, Y) to solve the optimal weight matrix W of the neural network model; The described use of the trained neural network model to predict the future life of the battery is specifically: in the same way of extracting the secondary feature X of the battery used for training the neural network model, the battery is charged from the first n cycles of the battery to be predicted. Extract secondary features X _p from the data of the discharge process, that is, first extract 20 initial features from the data of the first n cycles of cyclic charge/discharge process data of the battery to be predicted, and then filter out the secondary features X _p through the same elastic network, As the input of the neural network model, the neural network model will output the prediction result of the future life of the battery. 3. The method for predicting the life of a battery cell in an energy storage power station according to claim 2, wherein the expression of the ratio of the minimum allowable discharge capacity of the battery to the initial maximum discharge capacity of the battery is:

Among them, Qi _,min is the minimum discharge capacity allowed by the battery, Qi _,0 is the initial maximum discharge capacity of the battery, and i is the battery serial number. 4. The method for predicting the life of battery cells in an energy storage power station station according to claim 2, wherein the preliminary features are: Feature 1, the initial maximum discharge capacity Q _i,0 of the battery; Feature 2, the maximum discharge capacity Q _i,n of the nth cycle of the battery; Feature 3, the difference Q _i,n -Q _i,0 between the maximum discharge capacity of the battery in the nth cycle and the initial maximum discharge capacity; Feature 4, the maximum capacity change Q _i,1 -Q _i,0 of the battery after the battery is charged and discharged in the first cycle; Feature 5, the first-order coefficient of the least squares linear fitting during the first n cycles of the battery; Feature 6, the constant term coefficient of the least squares linear fitting during the first n cycles of the battery; Feature 7, the first-order coefficient of the least squares linear fitting during the first 90 to the nth cycle of the battery; Feature 8, the constant term coefficient of the least squares linear fitting during the first 90 to the nth cycle of the battery; Feature 9, the minimum temperature T _imin during the first n cycles of the battery; Feature 10, the maximum temperature T _imax during the first n cycles of the battery; Feature 11, the average value of temperature × time during the first n cycles of the battery:

Wherein, t _i is the time spent in the i-th cycle, and T _i is the temperature during the i-th cycle; Feature 12, the average time to discharge the battery during the first n cycles

Feature 13, the total time during the first n cycles of the battery

Feature 14, the minimum value of the capacity change of the battery under different voltages; Feature 15, the average value of the capacity change of the battery under different voltages; Feature 16, the divergence value of the capacity change of the battery under different voltages; Feature 17, the kurtosis coefficient value of the capacity change of the battery under different voltages; Feature 18, the initial internal resistance value of the battery

Feature 19, the internal resistance value of the nth cycle of the battery

Feature 20, the difference between the maximum value and the minimum value of the internal resistance in the first n cycles of the battery. 5. The method for predicting the life of a battery cell in an energy storage power station according to claim 2, wherein the expression of the cost function J(β) and the regular term P _α (β) is:

Among them, y _i is the life of the battery set used for training in the i-th cycle, x _i ^T is the feature initially extracted in the i-th cycle, β ₀ is the initial value of the regression coefficient vector, and β is the regression coefficient to be solved. vector, λ is the regular term coefficient, α is the coefficient of the elastic network, ||β|| ₁ and ||β|| ₂ are the 1-norm and 2-norm of β respectively; using the above expression, set a reasonable λ and α are used to obtain the regression coefficient vector β that minimizes J(β). 6. The method for predicting the life of a battery cell in an energy storage power station according to claim 2, wherein the neural network model is simplified as: W=g(X,Y), Among them, g represents the training process of the DNN neural network model of the BP algorithm, and the model is iteratively solved by the Bayesian method. 7. A battery cell life prediction system in an energy storage power station, characterized in that it comprises: Historical test data collection unit: collects historical test data of cyclic degradation of battery capacity; Preliminary feature extraction unit: extracts preliminary features reflecting battery degradation information; Secondary feature extraction unit: Screen the preliminary features through the elastic network, and extract the secondary features that have a high degree of influence on the prediction results of the remaining battery life as the final training features; Neural network model training unit: then use the selected secondary features to train the neural network model, and finally obtain the optimal weight matrix of the neural network model; Future battery life prediction unit: Use the trained neural network model to predict the future battery life. 8 . The battery cell life prediction system in an energy storage power station according to claim 7 , wherein, The battery life in the energy storage power station refers to: from the time of the test, before the battery health state reaches the ratio of the minimum allowable capacity of the battery to the maximum capacity of the battery at the factory, the maximum number of cycles of charge and discharge that the battery can perform; The extraction of the preliminary features reflecting the battery degradation information is specifically as follows: through the test data of the battery cell during the first n cycles of cyclic charge/discharge, preliminary extraction of 20 features is used as the initial features for inputting the elastic network; The screening of the preliminary features through the elastic network is specifically: introducing a regular term P _a (β) into the cost function J (β), and solving the coefficient of the regularized linear regression when the cost function is minimized, where a is Coefficient of elastic network, β is a vector of regularized linear regression coefficients; The extraction of the secondary features with a high degree of influence on the prediction result of the remaining service life of the battery is used as the final training feature. For training of neural network models; The training of the neural network model by using the selected secondary features is specifically as follows: the secondary feature X of the first n cycles of the battery is used as the input of the model, the life span Y is used as the output of the model, and the Bayesian iterative method is adopted. method, using (X, Y) to solve the optimal weight matrix W of the neural network model; The described use of the trained neural network model to predict the future life of the battery is specifically: in the same way of extracting the secondary feature X of the battery used for training the neural network model, the battery is charged from the first n cycles of the battery to be predicted. Extract secondary features X _p from the data of the discharge process, that is, first extract 20 initial features from the data of the first n cycles of cyclic charge/discharge process data of the battery to be predicted, and then filter out the secondary features X _p through the same elastic network, As the input of the neural network model, the neural network model will output the prediction result of the future life of the battery. 9. The battery cell life prediction system in an energy storage power station according to claim 8, wherein the expression of the ratio of the minimum allowable discharge capacity of the battery to the initial maximum discharge capacity of the battery is:

Among them, Qi _,min is the minimum discharge capacity allowed by the battery, Qi _,0 is the initial maximum discharge capacity of the battery, and i is the battery serial number. 10. The battery cell life prediction system in an energy storage power station according to claim 8, wherein the expression of the cost function J(β) and the regular term P _α (β) is:

Among them, y _i is the life of the battery set used for training in the i-th cycle, x _i ^T is the feature initially extracted in the i-th cycle, β ₀ is the initial value of the regression coefficient vector, and β is the regression coefficient to be solved. vector, λ is the regular term coefficient, α is the coefficient of the elastic network, ||β|| ₁ and ||β|| ₂ are the 1-norm and 2-norm of β respectively; using the above expression, set a reasonable λ and α are used to obtain the regression coefficient vector β that minimizes J(β).

Images