CN114609538A - Battery aging state estimation method and system based on improved Gaussian process regression - Google Patents

Abstract

The invention relates to a battery aging state estimation method and system based on improved Gaussian process regression, and relates to the technical field of batteries. The method comprises the following steps: acquiring original data sets of the lithium battery in different SOC intervals and different discharge current multiplying powers through experiments; analyzing the original data set to determine a model data set; dividing a model data set into a training set and a test set; establishing an improved Gaussian process regression model; and training and testing the improved Gaussian process regression model by adopting a training set and a testing set to generate the trained improved Gaussian process regression model, and predicting the aging state of the lithium battery. According to the method, the input characteristic value is subjected to coupling processing, and the model estimation value at the last moment is added to serve as the input characteristic value, so that the model dimension is reduced, the training difficulty is reduced, meanwhile, the accuracy of battery aging state estimation is obviously improved, and the uncertainty of a prediction result is greatly reduced.

Description

Battery aging state estimation method and system based on improved Gaussian process regression

Technical Field

The invention relates to the technical field of batteries, in particular to a battery aging state estimation method and system based on improved Gaussian process regression.

Background

Recently, with the use of fossil energy such as petroleum, the ecological environment is seriously damaged, and in order to continue the development of economy and the harmony of human and nature, the development of novel energy which is friendly to the environment is urgently needed. Lithium batteries have been widely used as energy storage elements in many fields due to their advantages of high energy density, high power density, low specific gravity, cyclic charge and discharge, and low environmental pollution. Along with the increase of the cycle charge and discharge frequency of the lithium battery, the aging state of the lithium battery can be deepened continuously, the available capacity of the lithium battery can be greatly reduced, and serious safety accidents can be caused by incorrect use of the lithium battery. Therefore, the method has important practical significance for accurately estimating the aging state of the lithium battery.

At present, research on the aging state estimation of the lithium battery is mainly divided into a mechanism model, an equivalent model and a data model. In the method based on the data model, voltage, current, temperature, SOC (State of Charge) interval and the like of the battery during charging and discharging are generally used as model input values to estimate the aging State of the battery. The Gaussian process regression has strong nonlinear fitting capability, is suitable for the regression problem of high-dimensional, nonlinear and small samples, and is widely applied to the aspects of system identification, control system design, system prediction and the like. Since battery aging is a complex nonlinear process, the gaussian process regression is also more applied to the estimation of lithium battery aging. However, the input characteristic value SOC median, the discharge depth and the discharge rate are fixed values in the whole aging period, and due to the characteristics of the Matern kernel function, the more similar the characteristic values are, the larger the output value of the kernel function is, so that only the equivalent cycle times of the characteristic value are changed in the whole aging period, the constant characteristic value makes the output of the kernel function larger and basically consistent, negative effects are caused on training, the optimization effect of the hyperparameter is poor, and finally the prediction accuracy is poor. In addition, the 95% confidence interval is wider, and the aging prediction process only depends on equivalent degradation rates under other aging stresses for prediction, so that the time attribute among the capacity degradation rates cannot be connected, and the prediction result has larger uncertainty.

Disclosure of Invention

The invention aims to provide a battery aging state estimation method and system based on improved Gaussian process regression so as to greatly reduce uncertainty of a battery aging state prediction result and improve prediction precision.

In order to achieve the purpose, the invention provides the following scheme:

a battery aging state estimation method based on improved Gaussian process regression comprises the following steps:

acquiring original data sets of the lithium battery in different SOC intervals and different discharge current multiplying powers through experiments;

analyzing the original data set to determine a model data set; the model data set comprises a median value of an SOC interval, a discharge depth, a discharge current multiplying power and an equivalent cycle number as input characteristic values, and comprises a lithium battery aging state index as an output value;

dividing the model data set into a training set and a test set;

establishing an improved Gaussian process regression model; the input of the improved Gaussian process regression model is the characteristic value after coupling treatment and the battery aging state at the previous moment, and the output is the battery aging state at the current moment;

training and testing the improved Gaussian process regression model by adopting the training set and the testing set to generate a trained improved Gaussian process regression model;

and predicting the aging state of the lithium battery by adopting the trained improved Gaussian process regression model.

Optionally, the obtaining, through an experiment, an original data set of the lithium battery at different SOC intervals and at different discharge current magnifications specifically includes:

setting a plurality of SOC intervals and a plurality of discharge current multiplying powers of the lithium battery;

performing multiple cyclic charge and discharge experiments on multiple lithium batteries with the same model in different SOC intervals and different discharge current multiplying powers, and recording the SOC interval and the discharge current multiplying power corresponding to each lithium battery in the multiple cyclic charge and discharge experiment processes;

after the repeated cycle charge-discharge experiment, performing a charge-discharge experiment on the lithium battery again under a preset discharge current multiplying power, and recording the discharge capacity in the charge-discharge experiment process as the current available capacity of the lithium battery after repeated charge-discharge cycles;

and generating the original data set according to the SOC interval and the discharge current multiplying power corresponding to each lithium battery in the multi-cycle charge-discharge experimental process and the current available capacity of the lithium battery after the lithium battery is subjected to the multi-cycle charge-discharge.

Optionally, the analyzing the original data set to determine a model data set specifically includes:

calculating corresponding SOC interval median values according to the different SOC intervals;

calculating the depth of discharge according to the different SOC intervals;

calculating the aging state index of the lithium battery after multiple charge and discharge cycles in the nth cycle according to the current available capacity of the lithium battery after multiple charge and discharge cycles;

and taking the SOC interval median, the discharge depth, the corresponding discharge current multiplying power and the equivalent cycle number as input characteristic values, and taking the corresponding aging state index as an output value to form the model data set.

Optionally, the establishing of the improved gaussian process regression model specifically includes:

constructing a Gaussian process regression model adopting different kernel functions;

initializing a hyper-parameter to zero, training the Gaussian process regression models adopting different kernel functions, performing hyper-parameter optimization through a conjugate gradient method, and evaluating errors of the Gaussian process regression models adopting different kernel functions by adopting RMSE (RMSE);

and determining the Gaussian process regression model with the minimum error and adopting a Matern + LIN kernel function as the improved Gaussian process regression model.

Optionally, the training and testing of the improved gaussian process regression model by using the training set and the testing set to generate a trained improved gaussian process regression model specifically includes:

multiplying the equivalent cycle times with the corresponding SOC interval median, the discharge depth and the discharge current multiplying power respectively to obtain a characteristic value after coupling treatment;

taking the characteristic value after the coupling treatment and the battery aging state at the previous moment as input values of the improved Gaussian process regression model, taking the battery aging state at the current moment as an output value of the improved Gaussian process regression model, and training the improved Gaussian process regression model;

in the training process, the improved Gaussian process regression model is tested by adopting the test set, and goodness of fit and root mean square error are used as model evaluation indexes;

and determining an improved Gaussian process regression model meeting the evaluation index of a preset model as the trained improved Gaussian process regression model.

A battery aging state estimation system based on improved gaussian process regression, comprising:

the system comprises an original data acquisition module, a data acquisition module and a data acquisition module, wherein the original data acquisition module is used for acquiring original data sets of the lithium battery in different SOC intervals and different discharge current multiplying powers through experiments;

the original data analysis module is used for analyzing the original data set and determining a model data set; the model data set comprises a median value of an SOC interval, a discharge depth, a discharge current multiplying power and an equivalent cycle number as input characteristic values, and comprises a lithium battery aging state index as an output value;

the model data set dividing module is used for dividing the model data set into a training set and a test set;

the model establishing module is used for establishing an improved Gaussian process regression model; the input of the improved Gaussian process regression model is the characteristic value after coupling treatment and the battery aging state at the previous moment, and the output is the battery aging state at the current moment;

the model training module is used for training and testing the improved Gaussian process regression model by adopting the training set and the testing set to generate a trained improved Gaussian process regression model;

and the battery aging state prediction module is used for predicting the lithium battery aging state by adopting the trained improved Gaussian process regression model.

Optionally, the raw data acquiring module specifically includes:

the SOC interval and multiplying power setting unit is used for setting a plurality of SOC intervals and a plurality of discharging current multiplying powers of the lithium battery;

the cyclic charge-discharge experiment unit is used for performing multiple cyclic charge-discharge experiments on multiple lithium batteries with the same model in different SOC intervals and different discharge current multiplying powers, and recording the SOC interval and the discharge current multiplying power corresponding to each lithium battery in the multiple cyclic charge-discharge experiment process;

the current available capacity experiment unit is used for carrying out a charge-discharge experiment on the lithium battery again under the preset discharge current multiplying power after the repeated cycle charge-discharge experiment, and recording the discharge capacity of the charge-discharge experiment process as the current available capacity of the lithium battery after the repeated charge-discharge cycle;

and the original data set generating unit is used for generating the original data set according to the SOC interval and the discharge current multiplying power corresponding to each lithium battery in the multi-cycle charge-discharge experimental process and the current available capacity of the lithium battery after the lithium battery is subjected to the multi-cycle charge-discharge.

Optionally, the raw data analysis module specifically includes:

the SOC interval median calculating unit is used for calculating corresponding SOC interval median according to the different SOC intervals;

the discharging depth calculating unit is used for calculating the discharging depth according to the different SOC intervals;

the aging state index calculation unit is used for calculating the aging state index of the lithium battery after the nth charge and discharge cycle according to the current available capacity of the lithium battery after the multiple charge and discharge cycles;

and the model data set generating unit is used for forming the model data set by taking the SOC interval median, the discharge depth, the corresponding discharge current multiplying power and the equivalent cycle number as input characteristic values and taking the corresponding aging state index as an output value.

Optionally, the model building module specifically includes:

the Gaussian process regression model construction unit is used for constructing Gaussian process regression models adopting different kernel functions;

the Gaussian process regression model evaluation unit is used for initializing the hyperparameters to zero, training the Gaussian process regression models adopting different kernel functions, carrying out hyperparametric optimization through a conjugate gradient method, and evaluating errors of the Gaussian process regression models adopting different kernel functions by adopting RMSE (RMSE);

and the improved Gaussian process regression model construction unit is used for determining the Gaussian process regression model which has the minimum error and adopts a Matern + LIN kernel function as the improved Gaussian process regression model.

Optionally, the model training module specifically includes:

the characteristic value coupling unit is used for multiplying the equivalent cycle times by the corresponding SOC interval median, the discharge depth and the discharge current multiplying power respectively to obtain a characteristic value after coupling treatment;

the model training unit is used for taking the characteristic value after the coupling processing and the battery aging state at the previous moment as input values of the improved Gaussian process regression model, taking the battery aging state at the current moment as an output value of the improved Gaussian process regression model, and training the improved Gaussian process regression model;

the model testing unit is used for testing the improved Gaussian process regression model by adopting the test set in the training process and taking the goodness of fit and the root mean square error as model evaluation indexes;

and the model training completion unit is used for determining an improved Gaussian process regression model meeting the preset model evaluation index as the trained improved Gaussian process regression model.

According to the specific embodiment provided by the invention, the invention discloses the following technical effects:

the invention provides a battery aging state estimation method and system based on improved Gaussian process regression, wherein the method comprises the following steps: acquiring original data sets of the lithium battery in different SOC intervals and different discharge current multiplying powers through experiments; analyzing the original data set to determine a model data set; the model data set comprises a median value of an SOC interval, a discharge depth, a discharge current multiplying power and an equivalent cycle number as input characteristic values, and comprises a lithium battery aging state index as an output value; dividing the model data set into a training set and a test set; establishing an improved Gaussian process regression model; the input of the improved Gaussian process regression model is the characteristic value after coupling treatment and the battery aging state at the previous moment, and the output is the battery aging state at the current moment; training and testing the improved Gaussian process regression model by adopting the training set and the testing set to generate a trained improved Gaussian process regression model; and predicting the aging state of the lithium battery by adopting the trained improved Gaussian process regression model. According to the invention, the influence of the SOC interval and the discharge current multiplying power on the aging rate of the lithium battery is qualitatively and quantitatively analyzed by researching the cycle discharge of the lithium battery under different SOC intervals and different discharge current multiplying powers; by coupling the input characteristic value and adding the model estimation value at the last moment as the input characteristic value, the dimension of the model is reducedThe training difficulty is reduced, and meanwhile, the accuracy of battery aging state estimation is obviously improved, so that the correlation coefficient R²The RMSE is obviously reduced, the confidence interval width is reduced from 1.25 percent to 0.4 percent, and the uncertainty of a prediction result is greatly reduced.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.

FIG. 1 is a flow chart of a method for estimating a battery aging state based on improved Gaussian process regression according to the present invention;

FIG. 2 is a schematic diagram of a method for estimating a battery aging state based on improved Gaussian process regression according to the present invention;

fig. 3 is a schematic diagram of an improved gaussian process regression model according to an embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

The invention aims to provide a battery aging state estimation method and system based on improved Gaussian process regression, which couple characteristic input values and introduce a model estimation value at the previous moment to enable a correlation coefficient R to be obtained²The method has the advantages that the RMSE is obviously increased, the RMSE is obviously reduced, the width of a confidence interval is reduced from 1.25% to 0.4%, the uncertainty of a prediction result of the battery aging state is greatly reduced, and the prediction precision is improved.

In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.

Fig. 1 is a flowchart of a battery aging state estimation method based on improved gaussian process regression according to the present invention, and fig. 2 is a schematic diagram of a principle of the battery aging state estimation method based on improved gaussian process regression according to the present invention. Referring to fig. 1 and 2, the method for estimating the aging state of a battery based on improved gaussian process regression according to the present invention includes:

step 101: and acquiring original data sets of the lithium battery in different SOC intervals and different discharge current multiplying powers through experiments.

The invention obtains the discharge data of different SOC cycle intervals and different discharge multiplying powers through experiments to form an original data set.

The step 101 of obtaining an original data set of the lithium battery in different SOC intervals and different discharge current multiplying powers through an experiment specifically includes:

step 1.1: setting a plurality of SOC intervals and a plurality of discharge current multiplying powers of the lithium battery;

the invention selects M lithium batteries with the same model to carry out N times of cyclic charge and discharge experiments. Representing the SOC working interval (SOC interval for short) of the lithium battery as SOC_l～SOC_hIn which SOC is_lRepresents the lower limit of the SOC interval, SOC_hRepresents the SOC interval upper limit. In the embodiment of the invention, a plurality of SOC working intervals of the lithium battery are set as SOC_li～SOC_hiWhere i is 4, where SOC_liIndicates the ith SOC interval lower limit, SOC_hiIndicates the ith SOC interval upper limit. The 4 different SOC operating intervals are specifically:

SOC_l1～SOC_h1＝15％～40％

SOC_l2～SOC_h2＝40％～65％

SOC_l3～SOC_h3＝65％～90％

SOC_l4～SOC_h4＝15％～90％

the 4 working intervals are respectively provided with M1 lithium batteries, namely M is 4 × M1.

In the embodiment of the invention, the set charging current multiplying factor (charging multiplying factor for short) is 1C, and the multiple discharging current multiplying factors (discharging multiplying factors for short) comprise three multiplying factors of 2C, 4C and 10C.

Step 1.2: and performing multiple cyclic charge and discharge experiments on multiple lithium batteries with the same model in different SOC intervals and different discharge current multiplying powers, and recording the SOC interval and the discharge current multiplying power corresponding to each lithium battery in the multiple cyclic charge and discharge experiment process.

M lithium batteries with the same model are selected to carry out N times of cyclic charge and discharge experiments, wherein M is equal to 36, and N is equal to 4500 in the embodiment of the invention. The lithium battery in each working interval is charged and discharged in a circulating mode in the corresponding interval, the charging current is 1C multiplying power, and the lithium battery is charged to SOC (state of charge) in a constant current mode_h(ii) a The discharge current is 2C, 4C and 10C multiplying power respectively, and the discharge is performed to SOC by constant current_l. The 3 kinds of discharge multiplying power of the invention are respectively provided with m2 lithium batteries, namely m 1-3 × m 2. And in the experimental process, recording the discharging SOC interval and the discharging multiplying power of the corresponding lithium battery.

Step 1.3: and after the repeated cycle charge-discharge experiment, performing a charge-discharge experiment on the lithium battery again under a preset discharge current multiplying power, and recording the discharge capacity in the charge-discharge experiment process as the current available capacity of the lithium battery after repeated charge-discharge cycles.

In the embodiment of the invention, N is 50. After repeating the cyclic charge-discharge experiment for 50 times, all the lithium batteries are subjected to constant current charging at the rate of 1C until the charge cut-off voltage is reached, and then the lithium batteries are charged at a constant voltage until the charge current is smaller than a preset threshold value, and the charging is stopped; performing constant current discharge on all lithium batteries at a constant rate of 1C until the voltage of the lithium batteries is reduced to a discharge cut-off voltage; the discharge capacity of the entire process was recorded as the current available capacity Q of the lithium battery after 50 charge-discharge cycles_now. The lithium battery aging state index adopted by the invention is defined as follows:

wherein

Defining the aging state index of the lithium battery after repeating 50 times of cycle charge and discharge experiments in the nth cycle, and representing the aging state of the lithium battery after 50 times of cycle charge and discharge experiments in the nth cycle; q₀Defined as the initial rated capacity of the lithium battery,

defined as the current available capacity of the lithium battery after 50 times of n-th cycle.

Step 1.4: and generating the original data set according to the SOC interval and the discharge current multiplying power corresponding to each lithium battery in the multi-cycle charge-discharge experimental process and the current available capacity of the lithium battery after the lithium battery is subjected to the multi-cycle charge-discharge.

Step 102: and analyzing the original data set to determine a model data set.

And analyzing the original data set, and determining characteristic values of the SOC interval median, the discharge depth, the discharge current multiplying power and the cycle number. The model data set comprises a SOC interval median value, a discharge depth, a discharge current multiplying power and an equivalent cycle number as input characteristic values, and comprises a lithium battery aging state index as an output value.

The step 102 of analyzing the original data set to determine a model data set specifically includes:

step 2.1: and calculating corresponding SOC interval median values according to the different SOC intervals.

Calculating the SOC interval to obtain the SOC interval median SOC according to the charge-discharge SOC interval_m(ii) a Median SOC of ith SOC interval_mThe calculation formula of (a) is as follows:

wherein i is 1-4.

Step 2.2: and calculating the depth of discharge according to the different SOC intervals.

Calculating the DOD according to the charge-discharge SOC interval; the calculation formula of the ith depth of discharge DOD is as follows:

DOD_i＝(SOC_hi-SOC_li)

wherein i is 1-4.

Step 2.3: and calculating the aging state index of the lithium battery after the nth charge and discharge cycle according to the current available capacity of the lithium battery after the multiple charge and discharge cycles.

The aging state after the nth cycle is calculated to be

State of aging

The calculation formula of (a) is as follows:

step 2.4: and taking the SOC interval median, the discharge depth, the corresponding discharge current multiplying power and the equivalent cycle number as input characteristic values, and taking the corresponding aging state index as an output value to form the model data set.

In the experimental process, the battery discharge rate corresponding to the ith SOC interval is recorded as C_disiWherein i is 1-4.

In the present invention, for uniform calculation of capacity release, one cycle at 75% depth of discharge was taken as a Standard cycle (C)_s) Thus, 3 cycles of the 25% interval equals 1 standard cycle. Equivalent cycle times (C)_e) The calculation formula of (a) is as follows:

C_s＝3C_e1＝3C_e2＝3C_e3＝C_e4

wherein C is_eiAnd the equivalent cycle number corresponding to the ith SOC interval is shown, wherein i is 1-4.

Calculating the obtained SOC interval median SOC_miSaid depth of discharge DOD_iCorresponding discharge current multiplying power C_disiAnd equivalent cycle number C_e(ii) a And as input characteristic values, using the corresponding aging state indexes as output values to form the model data set.

After the nth cycle of the lithium battery undergoes multiple charge-discharge cycles, the collected model data set comprises:

input values are as follows:

and (3) outputting a value:

wherein

DODn、

And

and respectively representing indexes of a median value, a discharge depth, a discharge current multiplying power, equivalent cycle times and an aging state of the SOC interval after 50 times of repeated cycle charge-discharge experiments of the nth cycle of the lithium battery.

Step 103: the model data set is divided into a training set and a test set.

Dividing a model data set formed by input characteristic data into a training set and a testing set, wherein the relationship between the data and the training set is shown in the following table:

wherein marked with "√" is a training data set and marked with "#" is a test data set.

Step 104: and establishing an improved Gaussian process regression model.

First, a gaussian process regression model using different kernel functions is constructed. Initializing the hyperparameters to zero, training Gaussian process regression models adopting different kernel functions, optimizing the hyperparameters by a conjugate gradient method, evaluating model errors adopting the different kernel functions by using a Root Mean Square Error (RMSE), and finally determining and selecting a matrix + Linear kernel function (LIN) as the kernel function of the improved Gaussian process regression model. Fig. 3 is a schematic diagram of an improved gaussian process regression model according to an embodiment of the present invention, and referring to fig. 3, the input of the improved gaussian process regression model is a coupling processed characteristic value and a battery aging state at a previous time, and the output is a battery aging state at a current time.

Therefore, the step 104 of establishing an improved gaussian process regression model specifically includes:

step 4.1: constructing a Gaussian process regression model adopting different kernel functions;

step 4.2: initializing a hyper-parameter to zero, training the Gaussian process regression models adopting different kernel functions, performing hyper-parameter optimization through a conjugate gradient method, and evaluating errors of the Gaussian process regression models adopting different kernel functions by adopting RMSE (RMSE);

step 4.3: and determining the Gaussian process regression model with the minimum error and adopting a Matern + LIN kernel function as the improved Gaussian process regression model.

Step 105: and training and testing the improved Gaussian process regression model by adopting the training set and the testing set to generate a trained improved Gaussian process regression model.

Training an improved Gaussian process regression model by using a training set, testing the performance of the model by using a test set, and fitting the goodness of fit R²And the root mean square error RMSE as model evaluation indices.

Step 105, training and testing the improved gaussian process regression model by using the training set and the testing set to generate a trained improved gaussian process regression model, which specifically includes:

step 5.1: and multiplying the equivalent cycle times with the corresponding SOC interval median, the discharge depth and the discharge current multiplying power respectively to obtain the characteristic value after coupling treatment.

The equivalent cycle number is multiplied by three fixed input characteristic values (SOC interval median, discharge depth and discharge current multiplying power) respectively to serve as input values of an improved Gaussian process regression model, so that the improved Gaussian process regression model has cycle number information while original characteristic value information is kept. Therefore, the characteristic value obtained by the coupling treatment of the invention comprises

And

step 5.2: and taking the characteristic value after the coupling treatment and the battery aging state at the previous moment as input values of the improved Gaussian process regression model, taking the battery aging state at the current moment as an output value of the improved Gaussian process regression model, and training the improved Gaussian process regression model.

And adding the model estimation value of the previous moment into the Gaussian process regression model, and estimating the aging state of the current moment together. Specifically, input features of a training set are combined

And the model estimate value of the previous time

Leading the model into an improved Gaussian process regression model for training; input features of a test set

And the model estimate value of the previous time

Leading the data into a Gaussian process regression model for testing to obtain output characteristics

As shown in fig. 3.

Step 5.3: in the training process, the improved Gaussian process regression model is tested by adopting the test set, and goodness of fit and root mean square error are used as model evaluation indexes.

The invention adopts goodness of fit R²And the root mean square error RMSE as an evaluation index of the algorithm. Wherein R is²The calculation formula of (a) is as follows:

the RMSE calculation is as follows:

wherein y is_iThe real value of the ith aging state;

the model estimation value corresponding to the aging state is the value of the battery aging state index at the current moment predicted by the improved Gaussian process regression model in the invention;

for the state of ageing at all timesAnd calculating the average value, wherein m is the number of the real values or the estimated values of the aging state.

Step 5.4: and determining an improved Gaussian process regression model meeting the preset model evaluation index as the trained improved Gaussian process regression model.

Step 106: and predicting the aging state of the lithium battery by adopting the trained improved Gaussian process regression model.

When the aging state of the lithium battery is predicted, the input characteristics of the lithium battery to be predicted

And the model estimate value of the previous time

Inputting the output characteristics into a trained improved Gaussian process regression model

The output characteristic

The aging state of the lithium battery is represented.

According to the invention, the influence of the SOC working interval and the discharge rate on the aging rate of the lithium battery is qualitatively and quantitatively analyzed by researching the cyclic discharge of the lithium battery under different SOC working intervals and different discharge rates. By carrying out coupling processing on the input characteristic value and adding the model estimation value at the last moment as the input characteristic value, the model dimensionality is reduced, the training difficulty is reduced, meanwhile, the accuracy of battery aging state estimation is obviously improved, and the correlation coefficient R is enabled to be²The RMSE is obviously reduced, the confidence interval width is reduced from 1.25 percent to 0.4 percent, and the uncertainty of a prediction result is greatly reduced.

Based on the method provided by the invention, the invention also provides a battery aging state estimation system based on improved Gaussian process regression, and the system comprises:

the system comprises an original data acquisition module, a data acquisition module and a data acquisition module, wherein the original data acquisition module is used for acquiring original data sets of the lithium battery in different SOC intervals and different discharge current multiplying powers through experiments;

the original data analysis module is used for analyzing the original data set and determining a model data set; the model data set comprises a median value of an SOC interval, a discharge depth, a discharge current multiplying power and an equivalent cycle number as input characteristic values, and comprises a lithium battery aging state index as an output value;

the model data set dividing module is used for dividing the model data set into a training set and a test set;

the model establishing module is used for establishing an improved Gaussian process regression model; the input of the improved Gaussian process regression model is the characteristic value after coupling treatment and the battery aging state at the previous moment, and the output is the battery aging state at the current moment;

the model training module is used for training and testing the improved Gaussian process regression model by adopting the training set and the testing set to generate a trained improved Gaussian process regression model;

and the battery aging state prediction module is used for predicting the aging state of the lithium battery by adopting the trained improved Gaussian process regression model.

The original data acquisition module specifically comprises:

the SOC interval and multiplying power setting unit is used for setting a plurality of SOC intervals and a plurality of discharging current multiplying powers of the lithium battery;

the cyclic charge-discharge experiment unit is used for performing multiple cyclic charge-discharge experiments on multiple lithium batteries with the same model in different SOC intervals and different discharge current multiplying powers, and recording the SOC interval and the discharge current multiplying power corresponding to each lithium battery in the multiple cyclic charge-discharge experiment process;

the current available capacity experiment unit is used for carrying out a charge-discharge experiment on the lithium battery again under the preset discharge current multiplying power after the repeated cycle charge-discharge experiment, and recording the discharge capacity of the charge-discharge experiment process as the current available capacity of the lithium battery after the repeated charge-discharge cycle;

and the original data set generating unit is used for generating the original data set according to the SOC interval and the discharge current multiplying power corresponding to each lithium battery in the multi-cycle charge-discharge experimental process and the current available capacity of the lithium battery after the lithium battery is subjected to the multi-cycle charge-discharge.

The raw data analysis module specifically comprises:

the SOC interval median calculating unit is used for calculating corresponding SOC interval median according to the different SOC intervals;

the discharging depth calculating unit is used for calculating the discharging depth according to the different SOC intervals;

the aging state index calculation unit is used for calculating the aging state index of the lithium battery after the nth charge and discharge cycle according to the current available capacity of the lithium battery after the multiple charge and discharge cycles;

and the model data set generating unit is used for forming the model data set by taking the SOC interval median, the discharge depth, the corresponding discharge current multiplying power and the equivalent cycle number as input characteristic values and taking the corresponding aging state index as an output value.

The model building module specifically comprises:

the Gaussian process regression model construction unit is used for constructing Gaussian process regression models adopting different kernel functions;

the Gaussian process regression model evaluation unit is used for initializing the hyperparameters to zero, training the Gaussian process regression models adopting different kernel functions, carrying out hyperparametric optimization through a conjugate gradient method, and evaluating errors of the Gaussian process regression models adopting different kernel functions by adopting RMSE (RMSE);

and the improved Gaussian process regression model construction unit is used for determining the Gaussian process regression model which has the minimum error and adopts a Matern + LIN kernel function as the improved Gaussian process regression model.

The model training module specifically comprises:

the characteristic value coupling unit is used for multiplying the equivalent cycle times by the corresponding SOC interval median, the discharge depth and the discharge current multiplying power respectively to obtain a characteristic value after coupling treatment;

the model training unit is used for taking the characteristic value after the coupling processing and the battery aging state at the previous moment as input values of the improved Gaussian process regression model, taking the battery aging state at the current moment as an output value of the improved Gaussian process regression model, and training the improved Gaussian process regression model;

the model testing unit is used for testing the improved Gaussian process regression model by adopting the test set in the training process and taking the goodness of fit and the root mean square error as model evaluation indexes;

and the model training completion unit is used for determining an improved Gaussian process regression model meeting the preset model evaluation index as the trained improved Gaussian process regression model.

The invention discloses a battery aging state estimation method and a system of improved Gaussian process regression by using an SOC (system on chip) cycle interval and discharge rate, wherein the method comprises the following steps: the method comprises the steps of obtaining discharge data of different SOC cycle intervals and different discharge multiplying powers through experiments, analyzing a data set, really modeling input values, constructing an improved Gaussian process regression model, and evaluating the improved Gaussian process regression model. The method can qualitatively and quantitatively analyze the influence of different SOC working intervals and different discharge rates on the aging rate of the lithium battery, and performs coupling processing through the characteristic values and adds the model estimation value at the last moment as an input value, so that the technical means reduces the model dimension, reduces the training difficulty and obviously improves the accuracy of the estimation of the aging state of the battery.

The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.

The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.

Claims (10)

1. A battery aging state estimation method based on improved Gaussian process regression is characterized by comprising the following steps:

acquiring original data sets of the lithium battery in different SOC intervals and different discharge current multiplying powers through experiments;

analyzing the original data set to determine a model data set; the model data set comprises a median value of an SOC interval, a discharge depth, a discharge current multiplying power and an equivalent cycle number as input characteristic values, and comprises a lithium battery aging state index as an output value;

dividing the model data set into a training set and a test set;

establishing an improved Gaussian process regression model; the input of the improved Gaussian process regression model is the characteristic value after coupling treatment and the battery aging state at the previous moment, and the output is the battery aging state at the current moment;

training and testing the improved Gaussian process regression model by adopting the training set and the testing set to generate a trained improved Gaussian process regression model;

and predicting the aging state of the lithium battery by adopting the trained improved Gaussian process regression model.

2. The method according to claim 1, wherein the experimentally obtaining the raw data set of the lithium battery at different SOC intervals and different discharge current rates specifically comprises:

setting a plurality of SOC intervals and a plurality of discharge current multiplying powers of the lithium battery;

performing multiple cyclic charge and discharge experiments on multiple lithium batteries with the same model in different SOC intervals and different discharge current multiplying powers, and recording the SOC interval and the discharge current multiplying power corresponding to each lithium battery in the multiple cyclic charge and discharge experiment processes;

after the repeated cycle charge-discharge experiment, performing a charge-discharge experiment on the lithium battery again under a preset discharge current multiplying power, and recording the discharge capacity in the charge-discharge experiment process as the current available capacity of the lithium battery after repeated charge-discharge cycles;

and generating the original data set according to the SOC interval and the discharge current multiplying power corresponding to each lithium battery in the multi-cycle charge-discharge experimental process and the current available capacity of the lithium battery after the lithium battery is subjected to the multi-cycle charge-discharge.

3. The method of claim 2, wherein analyzing the raw data set to determine a model data set comprises:

calculating corresponding SOC interval median values according to the different SOC intervals;

calculating the depth of discharge according to the different SOC intervals;

calculating the aging state index of the lithium battery after multiple charge and discharge cycles in the nth cycle according to the current available capacity of the lithium battery after multiple charge and discharge cycles;

and taking the SOC interval median, the discharge depth, the corresponding discharge current multiplying power and the equivalent cycle number as input characteristic values, and taking the corresponding aging state index as an output value to form the model data set.

4. The method of claim 3, wherein the establishing of the improved Gaussian process regression model specifically comprises:

constructing a Gaussian process regression model adopting different kernel functions;

initializing a hyper-parameter to zero, training the Gaussian process regression models adopting different kernel functions, performing hyper-parameter optimization through a conjugate gradient method, and evaluating errors of the Gaussian process regression models adopting different kernel functions by adopting RMSE (RMSE);

and determining the Gaussian process regression model with the minimum error and adopting a Matern + LIN kernel function as the improved Gaussian process regression model.

5. The method according to claim 4, wherein the training and testing of the improved Gaussian process regression model using the training set and the testing set to generate a trained improved Gaussian process regression model specifically comprises:

multiplying the equivalent cycle times with the corresponding SOC interval median, the discharge depth and the discharge current multiplying power respectively to obtain a characteristic value after coupling treatment;

taking the characteristic value after the coupling treatment and the battery aging state at the previous moment as input values of the improved Gaussian process regression model, taking the battery aging state at the current moment as an output value of the improved Gaussian process regression model, and training the improved Gaussian process regression model;

in the training process, the improved Gaussian process regression model is tested by adopting the test set, and goodness of fit and root mean square error are used as model evaluation indexes;

and determining an improved Gaussian process regression model meeting the preset model evaluation index as the trained improved Gaussian process regression model.

6. A battery state-of-aging estimation system based on improved gaussian process regression, comprising:

the system comprises an original data acquisition module, a data acquisition module and a data acquisition module, wherein the original data acquisition module is used for acquiring original data sets of the lithium battery in different SOC intervals and different discharge current multiplying powers through experiments;

the original data analysis module is used for analyzing the original data set and determining a model data set; the model data set comprises a median value of an SOC interval, a discharge depth, a discharge current multiplying power and an equivalent cycle number as input characteristic values, and comprises a lithium battery aging state index as an output value;

the model data set dividing module is used for dividing the model data set into a training set and a test set;

the model establishing module is used for establishing an improved Gaussian process regression model; the input of the improved Gaussian process regression model is the characteristic value after coupling treatment and the battery aging state at the previous moment, and the output is the battery aging state at the current moment;

the model training module is used for training and testing the improved Gaussian process regression model by adopting the training set and the testing set to generate a trained improved Gaussian process regression model;

and the battery aging state prediction module is used for predicting the lithium battery aging state by adopting the trained improved Gaussian process regression model.

7. The system of claim 6, wherein the raw data acquisition module specifically comprises:

the SOC interval and multiplying power setting unit is used for setting a plurality of SOC intervals and a plurality of discharging current multiplying powers of the lithium battery;

the cyclic charge-discharge experiment unit is used for carrying out cyclic charge-discharge experiments on a plurality of lithium batteries with the same model in different SOC intervals and different discharge current multiplying powers, and recording the SOC interval and the discharge current multiplying power corresponding to each lithium battery in the cyclic charge-discharge experiment process;

the current available capacity experiment unit is used for carrying out a charge-discharge experiment on the lithium battery under a preset discharge current multiplying power after the repeated charge-discharge experiment is carried out, and recording the discharge capacity of the charge-discharge experiment process as the current available capacity of the lithium battery after the repeated charge-discharge circulation;

and the original data set generating unit is used for generating the original data set according to the SOC interval and the discharge current multiplying power corresponding to each lithium battery in the multi-cycle charge-discharge experimental process and the current available capacity of the lithium battery after the lithium battery is subjected to the multi-cycle charge-discharge.

8. The system of claim 7, wherein the raw data analysis module specifically comprises:

the SOC interval median calculating unit is used for calculating corresponding SOC interval median according to the different SOC intervals;

the discharging depth calculating unit is used for calculating the discharging depth according to the different SOC intervals;

the aging state index calculation unit is used for calculating the aging state index of the lithium battery after the nth charge and discharge cycle according to the current available capacity of the lithium battery after the multiple charge and discharge cycles;

and the model data set generating unit is used for forming the model data set by taking the SOC interval median, the discharge depth, the corresponding discharge current multiplying power and the equivalent cycle number as input characteristic values and taking the corresponding aging state index as an output value.

9. The system of claim 8, wherein the model building module specifically comprises:

the Gaussian process regression model construction unit is used for constructing Gaussian process regression models adopting different kernel functions;

the Gaussian process regression model evaluation unit is used for initializing the hyperparameters to zero, training the Gaussian process regression models adopting different kernel functions, carrying out hyperparametric optimization through a conjugate gradient method, and evaluating errors of the Gaussian process regression models adopting different kernel functions by adopting RMSE (RMSE);

and the improved Gaussian process regression model construction unit is used for determining the Gaussian process regression model which has the minimum error and adopts a Matern + LIN kernel function as the improved Gaussian process regression model.

10. The system of claim 9, wherein the model training module specifically comprises:

the characteristic value coupling unit is used for multiplying the equivalent cycle times by the corresponding SOC interval median, the discharge depth and the discharge current multiplying power respectively to obtain a characteristic value after coupling treatment;

the model training unit is used for taking the characteristic value after the coupling processing and the battery aging state at the previous moment as input values of the improved Gaussian process regression model, taking the battery aging state at the current moment as an output value of the improved Gaussian process regression model, and training the improved Gaussian process regression model;

the model testing unit is used for testing the improved Gaussian process regression model by adopting the test set in the training process and taking the goodness of fit and the root mean square error as model evaluation indexes;

and the model training completion unit is used for determining an improved Gaussian process regression model meeting the preset model evaluation index as the trained improved Gaussian process regression model.

Images