Battery SOC estimation method

Abstract

The invention discloses a battery SOC estimation method. The method comprises the following steps: step 1, obtaining EIS of a battery under different SOCs; step 2, establishing a battery equivalent circuit model for the whole EIS frequency band and performing parameter identification; 3, searching features, and constructing and identifying a simplified ECM model of a feature frequency interval; and 4, constructing a regression model for the features of the simplified model, and performing precision evaluation. Compared with a traditional method for estimating the SOC based on the open-circuit voltage, the method has the advantages that the SOC is comprehensively evaluated through the EIS features, and the accuracy is higher; according to the method, the problem that SOC estimation based on an open-circuit voltage method is difficult due to the fact that lithium iron phosphate and other batteries have platform areas can be effectively solved; according to the method, a simplified ECM model of a specific frequency interval is adopted, and the problems that an original optimization algorithm is multiple in feature and low in speed are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of lithium-ion batteries, and specifically provides a battery SOC estimation method. Background Art

[0002] Traditional means of transportation powered by fossil fuels face problems such as high pollution and shortage of fuel resources. Therefore, more environmentally friendly and energy-saving new energy vehicles are gradually replacing fuel vehicles. Lithium-ion batteries are widely used in energy storage and power devices of new energy vehicles due to their high energy density, long life, and good environmental protection. The state of charge (SOC) of lithium-ion batteries reflects the remaining power of the battery and has a direct impact on the cruising range, power performance, and safety functions of new energy vehicles. With the increasing number of electric vehicles, it is of great significance to quickly and accurately estimate the state of charge of the battery. Therefore, efficient estimation of the state of charge of electric vehicle power batteries has become an important research direction.

[0003] Traditional SOC estimation methods include open circuit voltage method, ampere-hour integration method, Kalman filter and other methods. The open circuit voltage method is based on the SOC-OCV characteristic curve, and estimates the SOC by measuring the terminal voltage of the battery in the open circuit state. This method requires the battery to be stationary for a long time to achieve voltage stability. It is only suitable for fully stationary batteries, and for batteries with platform areas such as lithium iron phosphate, this method is almost impossible to accurately identify (as shown in the attached manual Figure 2 ). The ampere-hour integration method uses the integration of the charge and discharge current over time to calculate the SOC. This method requires accurate current measurement and initial SOC data, and there is a cumulative error over time. The Kalman filter method uses multiple measurements and model predictions to estimate the SOC. It is suitable for dynamic systems, but this method is too affected by the initial value. These methods are all SOC estimation methods based on voltage and current change sequences.

[0004] Electrochemical impedance spectroscopy (EIS) is an important method for characterizing battery dynamics. Since EIS can reflect the impedance characteristics of the battery within a wide frequency range and has the advantages of high precision, wide bandwidth, and non-destructiveness, it is widely used in battery electrochemical dynamics reaction mechanism analysis, battery aging mode analysis, battery health status estimation, and life prediction. The EIS method for estimating SOC often uses the equivalent circuit model method to perform parameter identification on the ECM model and establish a regression equation between parameter characteristics and SOC to achieve the purpose of SOC estimation. However, the commonly used parameter identification adopts a full-band identification method, which has a high feature dimension and high computational complexity, and the features that are finally retained only account for a very small part of all features, wasting a lot of identification time. Therefore, the present invention proposes a method for extracting features from a small number of samples to simplify the ECM model, which greatly improves the recognition speed of the model while ensuring the estimation accuracy. Summary of the invention

[0005] The purpose of the present invention is to provide a battery SOC estimation method to solve the problems existing in the background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a battery SOC estimation method, comprising the following steps:

[0007] Step 1: Obtain EIS of the battery at different SOCs;

[0008] Step 2: Establish a battery equivalent circuit model for the entire EIS frequency band and perform parameter identification;

[0009] Step 3, find features and construct and identify the simplified ECM model of the characteristic frequency interval;

[0010] Step 4: construct a regression model for the features of the simplified model and perform accuracy evaluation;

[0011] In the above step 1, the SOC of the battery is adjusted by a charging and discharging device to obtain different SOCs of the battery, and the EIS under the corresponding different SOCs is obtained by an impedance device;

[0012] In the above step 2, a battery equivalent circuit model that can describe the battery impedance spectrum is constructed, and an impedance-frequency relationship equation satisfied by the current battery equivalent circuit model is established, and an optimization algorithm is used to perform parameter identification on a small number of battery samples;

[0013] In the above step 3, the regularity of each parameter of the battery sample at different SOC is observed, the complex correlation coefficient method is used to find the characteristics, the ECM model is simplified according to the components to which the characteristics belong, and the optimization algorithm is used to perform parameter identification on the EIS data of all batteries;

[0014] In the above step 4, a regression model is constructed for the parameters identified in step 3, the batteries are divided into a training set and a validation set according to the battery numbers, and the goodness of fit R2 of the validation set is used as an indicator to judge the quality of the model.

[0015] Preferably, in step one, the SOC position of the EIS to be tested is 30%, 35%, 40%, 45%, 50%, 55%, and 60%; first, the battery to be tested is discharged to the cut-off voltage, and then charged to 30% SOC at a constant current after standing for 2 hours, and then the impedance test is started after standing for another 2 hours; then, starting from 30% SOC, 5% of the rated capacity is charged at a constant current each time, and the impedance test is started after standing for 2 hours until the SOC reaches 60%, thereby completing the EIS acquisition of the battery at different specified SOCs.

[0016] Preferably, in step 1, the interval of the EIS frequency points to be measured is 100 Hz-0.01 Hz, obeys logarithmic distribution, and the interval [10 k+1 ,10 k ]Hz interval number is 10, with a total of 41 frequency points; the impedance test is performed twice for each SOC using an impedance device, and the arithmetic average of the two impedance spectra is taken as the EIS of the battery under test at the current SOC.

[0017] Preferably, in step 2, the commonly used impedance spectrum equivalent circuit model includes an integer-order equivalent circuit model and a fractional-order equivalent circuit model; when the fractional-order equivalent circuit model is used, the equivalent circuit model circuit used includes the following circuit elements: ohmic resistor R 0 , UHF inductor L 0 , concentration polarization RC link, diffusion link and Weber element Ws; the concentration polarization RC link includes polarization resistor R 1 and polarization CPEQ 1 ; The diffusion link includes the diffusion resistance R 2 and diffusion CPEQ 2 ; Assuming that the angular frequency ω=2πf and j is an imaginary unit, the complex impedances of these components satisfy the following relationship:

[0018] Z R =R(1)

[0019] Z L =jωL(2)

[0020]

[0021] Therefore, the total impedance of the equivalent circuit model circuit is:

[0022]

[0023] Preferably, in step 2, the fractional-order equivalent circuit model needs to identify the following 11 parameters: Ohmic resistance R 0 , UHF inductor L 0 , polarization resistance R 1 , polarization CPE two parameters Q 1P ,Q 1T , diffusion resistance R 2 , diffusion CPE two parameters Q 2P ,Q 2T 、Three parameters of Warburg element Ws R ,Ws T ,Ws P ; Given 11 parameters to be identified, the predicted complex impedance sequence at all frequency points can be calculated Assuming that the real complex impedance sequence collected by the impedance device is Z(ω), the loss function is defined as the root mean square error RMSE of the two complex impedance sequences; the smaller the RMSE, the more accurate the result of parameter identification is.

[0024] Preferably, in the step 2, a small amount of EIS sample data of batteries is selected to use an optimization algorithm for parameter identification; commonly used optimization algorithms include genetic algorithm (GA), particle swarm algorithm (PSO), differential evolution algorithm (DE) and sparrow search algorithm (SSA); when the differential evolution algorithm is used for parameter optimization, the following steps are included:

[0025] The first step is to determine the algorithm population size M and the number of iterations N; then, randomly select M initial solutions X in the feasible domain. k =(R 0,k ,R 1,k ,R 2,k ,τ 1,k ,τ 2,k ,L 0,k ,Ws R,k ,Ws τ,k ,Ws P,k ), as an individual, the set of all individuals is called a population; for each individual X k , its fitness is defined as RMSE(X k ); The individual with the smallest fitness in the population is called the optimal individual of the current population;

[0026] The second step is to iterate the population according to the specified operation. There are three legal operations in differential evolution: mutation, crossover and selection. Each iteration executes the three operations in sequence. First, the mutation operation is executed. For the k+1 generation mutation operation, a candidate individual is generated:

[0027]

[0028] in is a randomly selected individual generated in the kth generation, and F is a mutation factor between 0 and 2; then a crossover operation is performed. For the k+1th generation crossover operation, V i (k) Let d = 9 be the parameter vector to be identified, assuming

[0029]

[0030] CR is a crossover factor between 0 and 1, then

[0031]

[0032]

[0033] Finally, the selection operation is performed. For the k+1th generation selection operation, the candidate individuals generated are selected and retained according to the size of the fitness. or original individual The retained individual is the i-th individual of the k+1 generation population Repeat the three operations until the number of individuals in the k+1 generation population reaches M;

[0034] In the third step, when the number of iterations reaches the maximum value N, the iteration stops; at this time, the optimal individual of the last population is selected as the result of parameter identification.

[0035] Preferably, in step 3, the components are divided into four groups, with an ohmic internal resistance R 0 , high frequency inductive reactance element L 0 , the RC element R in the intermediate frequency polarization process 1 ,Q 1 ,R 2 ,Q 2 and low-frequency diffusion process element Ws; use the complex correlation coefficient method to calculate the complex correlation coefficient of the high-frequency element, medium-frequency element, low-frequency element parameter group and SOC; select a group of elements X with the highest correlation coefficient, and select the frequency interval to construct a simplified version of the high-frequency, medium-frequency or low-frequency ECM model according to the frequency band represented by the element; due to R 0 It is independent of frequency, so the simplified ECM model constructed is R 0 Concatenation with X.

[0036] Preferably, in step three, a simplified ECM model is used to perform parameter identification on the EIS data of all batteries using the differential evolution algorithm in step two.

[0037] Preferably, in the step 4, a regression model is constructed for fitting the SOC based on the features of the simplified ECM model in the step 3; depending on the number of features, the regression model can adopt a traditional generalized linear regression model or a polynomial model, or a more complex Gaussian regression model or a random forest model.

[0038] Preferably, in the step 4, the data of the battery to be tested are divided into a training set and a validation set, the model is trained with the training set, and the accuracy of the model is tested with the validation set; a validation method based on splitting the data set by battery number is adopted, the data set is divided into K parts based on the battery number, 1 battery is selected as the validation set each time, and the remaining K-1 batteries are used as the training set to train the model, and the model with the highest goodness of fit of the validation set is selected as the final identified model.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] Compared with the traditional method of estimating SOC based on open circuit voltage, the present invention uses EIS characteristics to comprehensively evaluate SOC with higher accuracy; the present invention can effectively solve the problem that batteries such as lithium iron phosphate have a platform area, which makes it difficult to estimate SOC based on the open circuit voltage method; the present invention adopts a simplified ECM model in a specific frequency range, which makes up for the problem that the original optimization algorithm has many features and slow speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 is a flow chart of the method of the present invention;

[0042] Figure 2 This is a typical lithium iron phosphate battery OCV characteristic curve;

[0043] Figure 3 The original equivalent circuit model diagram used in the test example of the present invention;

[0044] Figure 4 This is the operation diagram of the differential evolution algorithm used in the test example of the present invention;

[0045] Figure 5 is the complex correlation coefficient between each feature of the test example of the present invention and SOC;

[0046] Figure 6 This is a simplified equivalent circuit model diagram used in the test example of the present invention. DETAILED DESCRIPTION

[0047] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0048] Please see attached Figure 1-6 The present invention provides a battery SOC estimation method, comprising the following steps:

[0049] Step 1: Obtain EIS of the battery at different SOCs;

[0050] Step 2: Establish a battery equivalent circuit model for the entire EIS frequency band and perform parameter identification;

[0051] Step 3, find features and construct and identify the simplified ECM model of the characteristic frequency interval;

[0052] Step 4: construct a regression model for the features of the simplified model and perform accuracy evaluation;

[0053] In the above step 1, the SOC of the battery is adjusted by the charging and discharging equipment to obtain different SOCs of the battery, and the EIS under the corresponding SOC is obtained by the impedance equipment; the SOC position of the EIS to be tested is 30%, 35%, 40%, 45%, 50%, 55%, 60%; first, the battery to be tested is discharged to the cut-off voltage, and then charged to 30% SOC at a constant current after standing for 2 hours, and then the impedance test is started after standing for another 2 hours; then, starting from 30% SOC, 5% of the rated capacity is charged at a constant current each time, and the impedance test is started after standing for 2 hours until the SOC reaches 60%, completing the EIS acquisition of the battery under the specified different SOCs; the frequency point of the EIS to be tested is in the range of 100Hz-0.01Hz, which obeys the logarithmic distribution, and the interval [10 k+1 ,10 k ]Hz interval is 10, with a total of 41 frequency points; impedance equipment is used to perform two impedance tests on each SOC, and the arithmetic average of the two impedance spectra is taken as the EIS of the battery under test at the current SOC;

[0054] In the above step 2, a battery equivalent circuit model that can describe the battery impedance spectrum is constructed, and the impedance-frequency relationship equation satisfied by the current battery equivalent circuit model is established, and an optimization algorithm is used to perform parameter identification on a small number of battery samples; commonly used impedance spectrum equivalent circuit models include integer-order equivalent circuit models and fractional-order equivalent circuit models; when a fractional-order equivalent circuit model is used, the equivalent circuit model circuit used includes the following circuit elements: ohmic resistor R 0 , UHF inductor L 0, concentration polarization RC link, diffusion link and Weber element Ws; the concentration polarization RC link includes polarization resistor R 1 and polarization CPEQ 1 ; The diffusion link includes the diffusion resistance R 2 and diffusion CPEQ 2 ; Assuming that the angular frequency ω=2πf and j is an imaginary unit, the complex impedances of these components satisfy the following relationship:

[0055] Z R =R(1)

[0056] Z L =jωL(2)

[0057]

[0058] Therefore, the total impedance of the equivalent circuit model circuit is:

[0059]

[0060] The fractional-order equivalent circuit model requires the identification of the following 11 parameters: Ohmic resistance R 0 , UHF inductor L 0 , polarization resistance R 1 , polarization CPE two parameters Q 1P ,Q 1T , diffusion resistance R 2 , diffusion CPE two parameters Q 2P ,Q 2T 、Three parameters of Warburg element Ws R ,Ws T ,Ws P ; Given 11 parameters to be identified, the predicted complex impedance sequence at all frequency points can be calculated Assuming that the real complex impedance sequence collected by the impedance device is Z(ω), the loss function is defined as the root mean square error RMSE of the two complex impedance sequences; the smaller the RMSE, the more accurate the result of parameter identification is.

[0061] Select a small amount of battery EIS sample data and use optimization algorithms for parameter identification; commonly used optimization algorithms include genetic algorithm (GA), particle swarm algorithm (PSO), differential evolution algorithm (DE) and sparrow search algorithm (SSA); when using differential evolution algorithm for parameter optimization, the following steps are included:

[0062] The first step is to determine the algorithm population size M and the number of iterations N; then, randomly select M initial solutions X in the feasible domain. k =(R 0,k ,R 1,k ,R 2,k ,τ1,k ,τ 2,k ,L 0,k ,Ws R,k ,Ws τ,k ,Ws P,k ), as an individual, the set of all individuals is called a population; for each individual X k , its fitness is defined as RMSE(X k ); The individual with the smallest fitness in the population is called the optimal individual of the current population;

[0063] The second step is to iterate the population according to the specified operation. There are three legal operations in differential evolution: mutation, crossover and selection. Each iteration executes the three operations in sequence. First, the mutation operation is executed. For the k+1 generation mutation operation, a candidate individual is generated:

[0064]

[0065] in is a randomly selected individual generated in the kth generation, and F is a mutation factor between 0 and 2; then a crossover operation is performed. For the k+1th generation crossover operation, V i (k) Let d = 9 be the parameter vector to be identified, assuming

[0066]

[0067] CR is a crossover factor between 0 and 1, then

[0068]

[0069]

[0070] Finally, the selection operation is performed. For the k+1th generation selection operation, the candidate individuals generated are selected and retained according to the size of the fitness. or original individual The retained individual is the i-th individual of the k+1 generation population Repeat the three operations until the number of individuals in the k+1 generation population reaches M;

[0071] The third step is that when the number of iterations reaches the maximum value N, the iteration stops; at this time, the optimal individual of the last population is selected as the result of parameter identification;

[0072] In the above step 3, the rules of various parameters of battery samples at different SOCs are observed, the complex correlation coefficient method is used to find the characteristics, the ECM model is simplified according to the components to which the characteristics belong, and the optimization algorithm is used to perform parameter identification on the EIS data of all batteries; the components are divided into four groups, the ohmic internal resistance R 0, high frequency inductive reactance element L 0 , the RC element R in the intermediate frequency polarization process 1 ,Q 1 ,R 2 ,Q 2 and low-frequency diffusion process element Ws; use the complex correlation coefficient method to calculate the complex correlation coefficient of the high-frequency element, medium-frequency element, low-frequency element parameter group and SOC; select a group of elements X with the highest correlation coefficient, and select the frequency interval to construct a simplified version of the high-frequency, medium-frequency or low-frequency ECM model according to the frequency band represented by the element; due to R 0 It is independent of frequency, so the simplified ECM model constructed is R 0 and X in series; using a simplified ECM model, the differential evolution algorithm in step 2 is used for all battery EIS data to complete parameter identification;

[0073] In the above step 4, a regression model is constructed for the parameters identified in the step 3, and the batteries are divided into a training set and a validation set according to the battery numbers, and the goodness of fit R2 of the validation set is used as an indicator for judging the quality of the model; for the features of the simplified ECM model in the step 3, a regression model is constructed to fit the SOC; according to the number of features, the regression model can adopt a traditional generalized linear regression model or a polynomial model, or a more complex Gaussian regression model or a random forest model;

[0074] The data of the battery to be tested are divided into a training set and a validation set. The model is trained with the training set, and the accuracy of the model is tested with the validation set. The validation method based on the battery number segmentation data set is adopted. The data set is divided into K parts based on the battery number. One battery is selected as the validation set each time, and the remaining K-1 batteries are used as the training set to train the model. The model with the highest goodness of fit of the validation set is selected as the final identification model.

[0075] Compared with the prior art, the present invention has the following beneficial effects:

[0076] Compared with the traditional method of estimating SOC based on open circuit voltage, the present invention uses EIS characteristics to comprehensively evaluate SOC with higher accuracy; the present invention can effectively solve the problem that batteries such as lithium iron phosphate have a platform area, which makes it difficult to estimate SOC based on the open circuit voltage method; the present invention adopts a simplified ECM model in a specific frequency range, which makes up for the problem that the original optimization algorithm has many features and slow speed.

[0077] Test example:

[0078] Using the method proposed in the above embodiment, two 280Ah lithium iron phosphate energy storage batteries were selected as experimental objects, and the battery numbers were 1 and 2 respectively. According to the method of step one of the embodiment, the battery was discharged at a constant current of 50A to a cut-off voltage of 2.7V. After standing for 2 hours, 84Ah was charged at a constant current of 50A and the capacity was adjusted to 30SOC. After standing for 2 hours, the impedance test was performed twice for each of the 41 frequency points in the range of 100Hz-0.01Hz, and the average of the two impedances was taken as the impedance value of each frequency point. After that, 14Ah of capacity was charged each time, and the impedance test was performed after standing for 2 hours. The experimental part was ended until the SOC of the battery reached 60.

[0079] Then, create an attached Figure 3 The full-band equivalent circuit model in is used to construct the relationship equation between complex impedance and frequency. Based on some characteristics of the impedance spectrum itself, the feasible domain of the 11 parameters to be identified is determined. For this type of battery, the feasible domains of the 11 parameters to be identified are selected as follows:

[0080] L 0 ∈[0,10 -7 ]

[0081] R 0 ∈[0,0.01]

[0082] R 1 ∈[0,0.01]

[0083] Q 1T ∈[0,10]

[0084] Q 1P ∈[0,1]

[0085] R 2 ∈[0,0.01]

[0086] Q 2T ∈[0,1000]

[0087] Q 2P ∈[0,1]

[0088] W R ∈[0,0.01]

[0089] W T ∈[0,150]

[0090] W P ∈[0,1]

[0091] The Cartesian product of the above 11 intervals is recorded as D. The next step is to randomly select 100 individuals from D to form the initial population of the differential evolution algorithm, using the formula:

[0092]

[0093] The RMSE of each individual in the population can be calculated. According to the operation rules of the differential evolution algorithm, each iteration will generate a new population of 100 individuals. The iteration stops when the number of iterations reaches 300, and the individual with the smallest RMSE is selected from the last population generated in sequence as the parameter identification result of this differential evolution algorithm. In order to avoid the influence of contingency and local optimal solutions, the same differential evolution algorithm needs to be repeated 10 times, and the weighted average of the identification parameters according to RMSE is taken as the final output parameter identification result.

[0094] Next, apply the formula:

[0095]

[0096] The complex correlation coefficient of each feature with respect to SOC is calculated. The complex correlation coefficients of these features are shown in Figure 5 As can be seen from the figure, Ws P The complex correlation coefficient of is the highest, so the identification of the Weber element Ws can also reflect the characteristics of this type of battery. In this test example, Ws reflects the low-frequency characteristics below 0.1 Hz, so the present invention constructs Figure 6 R 0 The low-frequency simplified ECM model connected in series with Ws is used for parameter identification in the range of 0.1 Hz-0.01 Hz to significantly speed up the calculation.

[0097] The simplified ECM model only contains R 0 and Ws, the number of parameters to be identified is reduced to 4, and their feasible regions are determined as:

[0098] R 0 ∈[0,0.01]Ws R ∈[0,0.01]Ws T ∈[0,150]Ws P ∈[0,1]

[0099] The differential evolution algorithm in the test case is used to identify the above four parameters of all batteries.

[0100] Next, we identify the R 0 The parameters such as SOC and SOC are used as features and the real SOC is used as a label to establish a generalized linear regression model. In actual calculation, one battery is used as a training set and the other battery is used as a validation set to establish a generalized linear regression model for calculation. The model with the highest goodness of fit R2 is used as the final model for identifying SOC.

[0101] After calculation, in the range of 30SOC-60SOC, the maximum deviation of SOC identified by the validation set does not exceed 3%, and the model fit goodness R2 reaches 0.995. Using the simplified model, each EIS data parameter identification step takes less than 3 seconds, which is more than 90% less than the 30 seconds of the traditional global model.

[0102] Based on the above, the advantages of the present invention are that compared with the traditional method of estimating SOC based on open circuit voltage, the present invention uses EIS characteristics to comprehensively evaluate SOC with higher accuracy; the present invention can effectively solve the problem that batteries such as lithium iron phosphate have a platform area, which makes it difficult to estimate SOC based on the open circuit voltage method; the present invention adopts a simplified ECM model in a specific frequency range, which makes up for the problem of the original optimization algorithm having many features and slow speed.

[0103] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A battery SOC estimation method, characterized in that: The following steps are involved: Step 1: Obtain EIS of the battery at different SOCs; Step 2: Establish a battery equivalent circuit model for the entire EIS frequency band and perform parameter identification; Step 3, find features and construct and identify the simplified ECM model of the characteristic frequency interval; Step 4: construct a regression model for the features of the simplified model and perform accuracy evaluation; In the above step 1, the SOC of the battery is adjusted by a charging and discharging device to obtain different SOCs of the battery, and the EIS under the corresponding different SOCs is obtained by an impedance device; In the above step 2, a battery equivalent circuit model that can describe the battery impedance spectrum is constructed, and an impedance-frequency relationship equation satisfied by the current battery equivalent circuit model is established, and an optimization algorithm is used to perform parameter identification on a small number of battery samples; In the above step 3, the regularity of each parameter of the battery sample at different SOC is observed, the complex correlation coefficient method is used to find the characteristics, the ECM model is simplified according to the components to which the characteristics belong, and the optimization algorithm is used to perform parameter identification on the EIS data of all batteries; In the above step 4, a regression model is constructed for the parameters identified in step 3, the batteries are divided into a training set and a validation set according to the battery numbers, and the goodness of fit R2 of the validation set is used as an indicator to judge the quality of the model.

2. A battery SOC estimation method according to claim 1, characterized in that: In the step 1, the SOC position of the EIS to be tested is 30%, 35%, 40%, 45%, 50%, 55%, and 60%; first, the battery to be tested is discharged to the cut-off voltage, and then charged to 30% SOC at a constant current after standing for 2 hours, and then the impedance test is started after standing for another 2 hours; then, starting from 30% SOC, 5% of the rated capacity is charged at a constant current each time, and the impedance test is started after standing for 2 hours until the SOC reaches 60%, thereby completing the EIS acquisition of the battery at different specified SOCs.

3. A battery SOC estimation method according to claim 1, characterized in that: In the step 1, the interval of the EIS frequency points to be measured is 100 Hz-0.01 Hz, obeys logarithmic distribution, and the interval [10 k+1 ,10 k ]Hz interval number is 10, with a total of 41 frequency points; the impedance test is performed twice for each SOC using an impedance device, and the arithmetic average of the two impedance spectra is taken as the EIS of the battery under test at the current SOC.

4. A battery SOC estimation method according to claim 1, characterized in that: In the step 2, commonly used impedance spectrum equivalent circuit models include integer-order equivalent circuit models and fractional-order equivalent circuit models; when the fractional-order equivalent circuit model is used, the equivalent circuit model circuit used includes the following circuit elements: ohmic resistor R0, ultra-high frequency inductor L0, concentration polarization RC link, diffusion link and Weber element Ws; the concentration polarization RC link includes polarization resistor R1 and polarization CPEQ1; the diffusion link includes diffusion resistor R2 and diffusion CPEQ2; assuming that the angular frequency ω=2πf, j is an imaginary unit, the complex impedance of these elements satisfies the following relationship: Z R =R (1) Z L =jωL (2) Therefore, the total impedance of the equivalent circuit model circuit is:

5. A battery SOC estimation method according to claim 4, characterized in that: In step 2, the fractional-order equivalent circuit model needs to identify the following 11 parameters: ohmic resistance R0, ultra-high frequency inductor L0, polarization resistance R1, polarization CPE two parameters Q 1P ,Q 1T , diffusion resistance R2, diffusion CPE two parameters Q 2P ,Q 2T 、Three parameters of Warburg components Given 11 parameters to be identified, the predicted complex impedance sequence at all frequency points can be calculated. Assuming that the real complex impedance sequence collected by the impedance device is Z(ω), the loss function is defined as the root mean square error RMSE of the two complex impedance sequences; the smaller the RMSE, the more accurate the result of parameter identification is.

6. A battery SOC estimation method according to claim 5, characterized in that: In the step 2, a small amount of EIS sample data of batteries is selected and an optimization algorithm is used to identify parameters; commonly used optimization algorithms include genetic algorithm, particle swarm algorithm, differential evolution algorithm and sparrow search algorithm; when the differential evolution algorithm is used for parameter optimization, the following steps are included: The first step is to determine the algorithm population size M and the number of iterations N; then, randomly select M initial solutions X in the feasible domain. k =(R 0,k ,R 1,k ,R 2,k ,τ 1,k ,τ 2,k ,L 0,k ,Ws R,k ,Ws τ,k ,Ws P,k ), as an individual, the set of all individuals is called a population; for each individual X k , its fitness is defined as RMSE(X k ); The individual with the smallest fitness in the population is called the optimal individual of the current population; The second step is to iterate the population according to the specified operation. There are three legal operations in differential evolution: mutation, crossover and selection. Each iteration executes the three operations in sequence. First, the mutation operation is executed. For the k+1 generation mutation operation, a candidate individual is generated: in is a randomly selected individual of the kth generation, and F is a mutation factor between 0 and 2; then a crossover operation is performed. For the k+1th generation crossover operation, Let d = 9 be the parameter vector to be identified, assuming CR is a crossover factor between 0 and 1, then Finally, the selection operation is performed. For the k+1th generation selection operation, the candidate individuals generated are selected and retained according to the size of the fitness. or original individual The retained individual is the i-th individual X of the k+1 generation population i (k+1) ; Repeat the three operations until the number of individuals in the k+1 generation population reaches M; In the third step, when the number of iterations reaches the maximum value N, the iteration stops; at this time, the optimal individual of the last population is selected as the result of parameter identification.

7. A battery SOC estimation method according to claim 6, characterized in that: In step 3, the components are divided into four groups, namely, ohmic internal resistance R0, high frequency inductive reactance component L0, medium frequency polarization process RC components R1, Q1, R2, Q2 and low frequency diffusion process component Ws; the complex correlation coefficient method is used to calculate the complex correlation coefficient of the high frequency component, medium frequency component, low frequency component parameter group and SOC; A group of components X with the highest correlation coefficient are selected, and according to the frequency band represented by the components, a frequency interval is selected to construct a simplified version of the ECM model of high frequency, medium frequency or low frequency; since R0 is independent of frequency, the simplified version of the ECM model constructed is the series connection of R0 and X.

8. A battery SOC estimation method according to claim 6, characterized in that: In step three, the simplified ECM model is used to apply the differential evolution algorithm in step two to the EIS data of the entire battery to complete parameter identification.

9. A battery SOC estimation method according to claim 1, characterized in that: In the step 4, a regression model is constructed to fit the SOC based on the features of the simplified ECM model in the step 3; depending on the number of features, the regression model can adopt a traditional generalized linear regression model or a polynomial model, or a more complex Gaussian regression model or a random forest model.

10. A battery SOC estimation method according to claim 1, characterized in that: In the step 4, the data of the battery to be tested are divided into a training set and a validation set, the model is trained with the training set, and the accuracy of the model is tested with the validation set; a validation method based on the battery number segmentation data set is adopted, the data set is divided into K parts based on the battery number, one battery is selected as the validation set each time, and the remaining K-1 batteries are used as the training set to train the model, and the model with the highest goodness of fit of the validation set is selected as the final identified model.

Images