CN109061508A

CN109061508A - A kind of estimation method of electric automobile lithium battery SOH

Info

Publication number: CN109061508A
Application number: CN201811056160.2A
Authority: CN
Inventors: 夏飞; 施恩威; 彭道刚; 孟娟; 钱玉良
Original assignee: Shanghai University of Electric Power
Current assignee: Shanghai University of Electric Power
Priority date: 2018-09-11
Filing date: 2018-09-11
Publication date: 2018-12-21

Abstract

The invention relates to a method for estimating the SOH of an electric vehicle lithium battery, which is characterized in that it comprises the following steps: 1) constructing a second-order RC equivalent model of a power lithium battery; 2) parameter identification of the second-order RC equivalent model, including ohm Internal resistance R ₀ and second-order RC link parameters; 3) Calculate the ohmic internal resistance R ₀ by using the improved UPF method; 4) Calculate the SOH value of the power lithium battery. Compared with the prior art, the present invention has the advantages of strong tracking and prediction ability, high estimation precision and the like.

Description

A kind of estimation method of electric automobile lithium battery SOH

Technical field

The present invention relates to dynamic lithium battery technical fields, more particularly, to the estimation side of electric automobile lithium battery SOH a kind of Method.

Background technique

Global Oil resource is increasingly depleted, and the popularize caused problem of environmental pollution and energy crisis of fuel-engined vehicle cause The highest attention of people.Electric car promotes countries in the world to throw as a kind of environmentally friendly, resource-conserving vehicles Enter the research and development that a large amount of manpower and material resources are engaged in electric car.Core component of the power battery as electric car, ecology potential must Key effect is played to the performance of electric car.Currently, application prospect of the lithium ion battery on electric car is best, to its property The research of energy also becomes hot spot in recent years, and the state parameter that characterize battery performance mainly has state-of-charge (State of Charge, SOC) and health status (State of Health, SOH).State-of-charge SOC characterizes the energy storage capacity of battery, shadow Performances, the health status SOH such as the course continuation mileage of electric car is rung then to characterize the service life cycle of battery, directly affect electricity The reliability and safety of electrical automobile.Research compared to SOC, domestic and foreign scholars are less to the research of SOH, and SOH is that battery is ground The weak link studied carefully, accurately monitors SOH and estimation helps to improve battery management system (Battery Management System, BMS), SOC estimated accuracy is improved, overcharge, overdischarge is avoided, facilitates fault diagnosis and safety Early warning is of great significance to electric car general safety stable operation.

Currently, the main approaches in terms of lithium battery SOH estimation have: defining method (" Intelligent prognostics for battery health monitoring based on sample entropy."Expert Systems with Applications ", " State-of-Health estimation of Lithium-ion battery Based on interval capacity "), also known as volumetric method, direct electric discharge, be substantially by the capacity of battery, electricity, The proportionate relationship of the Feature changes such as power carries out SOH assessment.Internal resistance method (" grind by power battery pack health status evaluation method Study carefully ", " State-of-health monitoring of lithium-ion batteries in electric vehicles By on-board internal resistance estimation "), mainly by establishing the pass between internal resistance and SOH System is to estimate SOH.Neural network (Neural Network, NN) (" A Neural Network Based State-of- Health Estimation of Lithium-ion Battery in Electric Vehicles ", " grey neural network Model estimation on line lithium ion battery SOH "), SOH is obtained by obtaining a large amount of experimental data come training pattern.Kalman's filter Wave method (Kalman Filter, KF) (" the 18650 lithium ion battery health estimations based on UKF ", " based on the lithium of AUKF from Sub- cell health state estimation "), battery-based state-space model passes through the SOH of recursion iterative estimate battery.Particle filter (" a kind of lithium ion battery health forecast algorithm based on particle filter " " is based on SVR- to wave method (Particle Filter, PF) The unmanned plane cell health state diagnosis research of PF ", " State-of-health estimation of lithium-ion battery packs in electric vehicles based on genetic resampling particle Filter "), basic thought is with one group of sample come the Posterior probability distribution of approximate representation system, then approximate using this Indicate the state to estimate nonlinear system.Definition method is the effective ways for the evaluation cell health state that industry is generally acknowledged, but needs By battery off-line measurement, difficulty is realized for vehicle mounted dynamic battery group, and special test load is needed to carry out battery Repeated charge, experimental period is long, and practical value is not high.It is related to its internal resistance that internal resistance method establishes cell degradation degree Property, its health status is estimated by accurately measuring internal resistance of cell value.Internal resistance method can be realized power battery health status On-line Estimation, but the internal resistance of cell belongs to small signal, it is relatively difficult to accurately measure, so this method obtains not yet at present To a large amount of application.

The fundamental characteristics of neural network is non-linear, and has very strong generalization ability and parallel processing capability, is suitble to mould Quasi- electric automobile power battery working condition, versatility are stronger.But it is to need to carry out battery the shortcomings that neural network The precision that a large amount of experiment is trained model with obtaining comprehensive sample data, and estimates is largely by training number According to influence and be difficult to realize on-line training.Kalman filtering algorithm is essentially recursive feedback, be substantially carried out the time update and Measuring state updates.Since Kalman filtering is only applicable to linear system, at present using it is more be all its innovatory algorithm, such as expand Open up Kalman filtering (Extended Kalman Filter, EKF) and Unscented kalman filtering (Unscented Kalman Filter, UKF) etc., it is aided with other improvement means, to adapt to the nonlinear system of battery.But EKF and UKF are only applicable to Non-linear Gaussian system effectively overcomes though UKF has abandoned the traditional method that EKF linearizes nonlinear function The defect that EKF estimated accuracy is low, stability is poor.But UKF is still the posterior probability density with Gaussian Profile come approximation system state Function, if error is larger in the case where non-gaussian.

Particle filter is able to solve nonlinear and non-Gaussian problem, and nonlinear and non-Gaussian system as research battery is come It says, effective method of can yet be regarded as, and application of the PF algorithm in terms of lithium battery SOH estimation be not also extensive, has larger Research space.In recent years, " a kind of lithium ion battery health forecast algorithm based on particle filter " is proposed with lumped parameter mould Particle filter algorithm based on type predicts battery SOH." the unmanned plane cell health state diagnosis research based on SVR-PF " Propose it is a kind of based on support vector regression particle filter (Support Vector Machine Particle Filter, SVR-PF cell health state diagnostic method).In order to solve the degenerate problem of particle filter algorithm, " State-of-health estimation of lithium-ion battery packs in electric vehicles based on genetic Resampling particle filter " genetic algorithm is combined with particle filter algorithm, most using forgetting factor recursion Small square law carries out parameter identification, so that the estimation for carrying out battery pack internal resistance is tested to evaluate SOH.Generally speaking, particle filter Algorithm has certain superiority for the solution of nonlinear and non-Gaussian system, but often will appear particle during realization The problems such as degeneration, particle Loss of diversity, and calculation amount is larger, needs to make improvements.

Summary of the invention

It is an object of the present invention to overcome the above-mentioned drawbacks of the prior art and provide a kind of electric car lithium electricity The estimation method of pond SOH.

The purpose of the present invention can be achieved through the following technical solutions:

A kind of estimation method of electric automobile lithium battery SOH, comprising the following steps:

1) the Order RC equivalent model of construction force lithium battery；

2) parameter identification of Order RC equivalent model, including ohmic internal resistance R₀And Order RC link parameter；

3) ohmic internal resistance R is calculated using improvement UPF method₀；

4) SOH value of dynamic lithium battery is calculated.

In the step 1), Order RC equivalent model is that polarization resistance in parallel is added based on Thevenin model R_piWith polarization capacity C_piOrder RC equivalent-circuit model is constructed, and by the U of two RC links_p1And U_p2, ohmic internal resistance R₀With SOC constructs observational equation as observed quantity as state variable x, the end voltage U of battery.

The circuit equation of the Order RC equivalent model are as follows:

U=U_OCV(SOC)-IR₀-U_p1-U_p2

Wherein, R₀For ohmic internal resistance, U_OCVIt (SOC) is the open-circuit voltage under SOC, I is the charging and discharging currents of battery, and U is end Voltage, U_p1And U_p2For the voltage of two RC links, U_piFor polarizing voltage.

The observational equation are as follows:

Wherein, T is sampling time, Q_NFor battery rated capacity, I_kFor charging and discharging currents, w_1,k、w_2,k、w_3,k、w₄,_kTo be System state-noise, ν_kFor systematic observation noise, C_p1、C_p2The capacitor of respectively two RC links, R_p1、R_p2Respectively two RC rings The resistance of section, subscript k indicate sampling instant.

In the step 2), the parameter in second-order model is recognized using HPPC pulse charge-discharge test, specifically:

21) it is tested by volume test, demarcates the rated capacity Q of lithium battery_N；

22) relational expression between low current charge-discharge test fitting acquisition SOC and open-circuit voltage OCV is carried out；

23) parameter in Order RC equivalent-circuit model is recognized by HPPC pulse charge-discharge test, obtains charge and discharge The ohmic internal resistance of each SOC point on direction, and be fitted to obtain the relationship expression of SOC and ohmic internal resistance on charge and discharge direction Formula.

The step 3) specifically includes the following steps:

31) sampling instant k=0 is enabled, total number of particles N, and init state variable x are set₀, covariance matrix P and just Beginning weightI=1,2 ..., N；

32) sampling instant k=k+1 is enabled, particle is predicted using UKF algorithm, complete the estimation of weighting particle and is adopted Sample improves UPF algorithm by UKF and generates the importance density function of PF algorithm to carry out forecast updating to particle；

33) weight of all N number of particles is calculated separatelyAnd it is normalized to obtain normalization weight

34) number of effective particle is countedGiven threshold N_th=2N/3, if N_eff> N_th, Step 35) is then carried out, if N_eff≤N_th, then to particle collectionCarry out resampling；

35) optimal estimation and the corresponding variance matrix of each particle of k sample moment state are obtained；

36) terminate when sampling number reaches setting value, otherwise return step 32).

In the step 33), i-th of particle weightCalculating formula are as follows:

Wherein, S is the standard deviation that observational equation introduces noise, U_{pred_ukf}For the mean value of observation prediction.

The resampling of the step 34) specifically:

341) to all normalization weightsIt is ranked up from small to large, and will be with normalization weightCorresponding grain Sub- estimated valueWith true valueDifferenceIt is ranked up；

342) weight will be normalizedWith differenceQuotientIt is ranked up from small to large, if existing in quotient Two groups of equal data then delete more new sort after lesser one group of weight of normalization；

343) setting survival of the fittest threshold value ω_thFor quotientThe 2/5 of average value, ifThen retain correspondence Particle weight and particle estimated value otherwise delete corresponding particle weight and particle estimated value, and more new particle collection.

In the step 4), the calculating formula of the SOH value of dynamic lithium battery are as follows:

Wherein, R_EOLIt (SOC) is the internal resistance value at the end of battery life, R_NOWIt (SOC) is the current internal resistance value of battery, i.e., it is optimal Particle estimated valueIn ohmic internal resistance, R_NEW(SOC) internal resistance value when dispatching from the factory for battery.

Compared with prior art, the invention has the following advantages that

The present invention has initially set up dynamic lithium battery Order RC equivalent-circuit model, experimental data is then based on, using most Small two, which multiply curve-fitting method, picks out parameter in model.It proposes according to the R in charge and discharge process at difference SOC point₀Value is estimated SOH is counted, and using improvement UPF algorithm come tracking prediction R₀.For the accuracy of the mentioned model and method of the verifying present invention, using changing R into UPF algorithm to battery in constant current charge-discharge₀Carry out real-time tracking prediction, and the reality obtained with second-order model Value compares, the results showed that improves UPF algorithm to R₀Tracking prediction ability be better than UKF algorithm, by comparing same loop The estimated accuracy of SOH, again demonstrates side proposed by the invention under number difference SOC point and identical SOC point difference cycle-index The superiority of method.

Detailed description of the invention

Fig. 1 is dynamic lithium battery Order RC equivalent-circuit model.

Relational graph of the Fig. 2 between SOC and open-circuit voltage.

Fig. 3 is that SOC is 30%~40% charging pulse experimental result.

Fig. 4 is that SOC is 40%~30% discharge pulse experimental result.

Fig. 5 is HPPC charging experiment SOC-R₀Curve matching.

Fig. 6 is HPPC discharge test SOC-R₀Curve matching.

Specific embodiment

The present invention is described in detail with specific embodiment below in conjunction with the accompanying drawings.

In order to improve the calculating simplicity and estimated accuracy of electric automobile lithium battery (dynamic lithium battery) SOH, root of the present invention According to ohmic internal resistance R in battery charge and discharge process₀SOH is estimated with the situation of change by stages of remaining capacity SOC, and using improvement Carry out tracking prediction R without mark particle filter (Unscented Particle Filter, UPF) algorithm₀, i.e., on the basis of UPF algorithm On, discard tradition method for resampling, carries out resampling using the particle selection algorithm of the weight sequence survival of the fittest.Through the invention The sample degeneracy problem that UPF algorithm can be improved, can effectively improve the estimated accuracy of dynamic lithium battery SOH.

Step of the invention is as follows:

Step 1: establish dynamic lithium battery Order RC equivalent model:

Equivalent-circuit model common at present mainly has RINT model, PNGV model, Thevenin model etc..Wherein, RINT model belongs to ideal model, and there is no consider influence of the extraneous factor to battery；The internal resistance parameter and battery of PNGV model DC internal resistance is more accurate, but the performance under different temperatures environment is unsatisfactory；Thevenin equivalent-circuit model energy ratio The static and dynamic performance of preferable reaction cell, but original Thevenin model is limited to the responding ability of dynamic current.

The present invention is constructed on the basis of considering Thevenin model by increasing polarization resistance and polarization capacity in parallel Order RC equivalent-circuit model, as shown in Figure 1.

Step 2: the parameter identification of Order RC battery model:

The constant current charge-discharge for carrying out at least 3 times to completely new lithium battery first is tested, if discharge capacity difference is 2% every time Within, using this discharge capacity as battery rated capacity.

Then carry out 0.05C low current charge-discharge test fitting SOC and open-circuit voltage (Open Circuit Voltage, OCV the relationship between) chooses 5%~95% SOC point (being spaced 5% between each SOC point) and its correspondence in experimentation Charge and discharge stage voltage value, the charging/discharging voltage of each SOC point is averaged as the open-circuit voltage at each SOC point Value.

It is distinguished followed by HPPC (Hybrid Pulse Power Characterization) pulse charge-discharge test Know the parameter in second-order model, is 5%, 10%, 20% ... 90% in SOC, carries out HPPC charging respectively on 95% 11 points Pulse test and the experiment of HPPC discharge pulse.Charging pulse is 0.2C electric current charging 30min, stands 90min；Discharge pulse is 0.2C current discharge 30min stands 90min, to the parameter identification of R0 and RC link, needs to combine in pulse charge-discharge test Battery each SOC point response process, it is hereby achieved that HPPC charge and discharge Process Model Identification parameter.

Step 3: it is calculated based on the R0 for improving UPF method:

Calculating process is specific as follows:

(1) enabling k=0, population N is 50, init state variable x₀=[U_p1U_p2R₀SOC]^TWith covariance matrix P and set Initial weightI=1,2 ..., N, wherein U_p1And U_p2For the polarizing voltage of polarizing voltage RC link, ohmic internal resistance R₀In addition state variable of the SOC as system.

(2) weight prediction, the sampling of particle: k=k+1 carries out forecast updating to particle using UKF algorithm and calculates Sigma point,To suggest that distribution function isExtract particle

(3) N number of particle weight is calculatedAnd carry out weight normalization

(4) judge whether resampling: calculating effective particle numberGiven threshold N_th=2N/3, if N_eff > N_th, (5) are gone to, otherwise to particle collectionResampling.

First particle weights during resamplingAccording to sequence sequence from small to large, while handle corresponds Particle estimated valueWith true valueDifferenceIt is ranked up.ThenDivided byIt obtainsIt willAccording to sequence sequence from small to large, if the result being divided byIn have two groups of equal data, then retain grain Sub- weightBiggish one group.?The 2/5 of average value is used as threshold value ω_thIt selects the superior and eliminates the inferior, if a certain momentGreater than ω_th, then corresponding particle weightsWith particle estimated valueIt remains.Grain after resampling Subset isWherein

(5) optimal estimation of k moment state is calculatedAnd the corresponding variance matrix of each particle

(6) if k reaches the number of iterations of setting, algorithm terminates, and otherwise goes to (2).

The present invention realizes SOC and R using UPF algorithm is improved₀State estimation, because of R₀Change with the variation of SOC, estimates Internal resistance R can be constantly corrected while counting SOC₀Value improves UPF algorithm to R to further improve₀Tracking prediction effect. Estimation is participated in by improving the small particle of UPF algorithm picks difference comparsion, improves sample degeneracy and Loss of diversity problem, from And make internal resistance R₀Estimation it is more accurate.

Step 4: dynamic lithium battery SOH is calculated:

The health status of lithium battery refers to the current capacity of lithium battery, i.e., under certain condition, lithium battery can fill Enter or release the percentage of electricity Yu battery nominal capacity.For pure electric automobile, when power battery SOH drops to 80% When, it is believed that the end-of-life of battery.Since internal resistance can increase as battery SOH is reduced in the use process of battery, so The health status of battery can be defined from the angle that the internal resistance of cell changes, it may be assumed that

R_NOWIt (SOC) is the current internal resistance value of battery, R_NEW(SOC) be battery factory when internal resistance value, R_EOLIt (SOC) is battery Internal resistance value when end-of-life, R_NOW(SOC) estimated to obtain by step 3), that is, optimal particle estimated value In R₀。R_NEW(SOC) and R_EOL(SOC) it is solved and is obtained by linear equation in two unknowns group, because after carrying out cycle charge-discharge experiment The practical charge/discharge capacity of battery can be obtained, so that practical SOH is obtained, the practical R during cycle charge-discharge₀It again can be from SOC and R₀It is obtained in the matched curve of relationship, according to the calculating formula of SOH value, with R_NEW(SOC) and R_EOL(SOC) unknown for two Number, only it is to be understood that practical SOH and practical R under different charge and discharge number₀Value (corresponding R_NOW(SOC)), so that it may which it is primary that column write binary Equation group obtains R_NEW(SOC) and R_EOL(SOC), the current internal resistance value R of battery is finally estimated that by step 3)_NOW(SOC), And then cell health state (SOH) can be estimated according to formula (1).

Embodiment:

The embodiment that the present embodiment provides summary of the invention is divided into four steps.

1, model is established

As shown in Figure 1, according to Kirchoff s voltage, current law, the circuit equation of model are as follows:

U=U_OCV(SOC)-IR₀-U_p1-U_p2 (2)

Wherein R₀For ohmic internal resistance, U_OCVIt (SOC) is open-circuit voltage, there are non-linear relation, R with SOC_pi(i=1,2) it is Polarization resistance, C_pi(i=1,2) is polarization capacity, U_pi(i=1,2) is polarizing voltage, and I is the charging and discharging currents of battery, and U is end Voltage.

Choose the voltage U of two RC links_p1And U_p2, ohmic internal resistance R₀In addition state variable of the SOC as system chooses electricity The end voltage in pond is as observed quantity, shown in structural regime equation such as formula (4), shown in observational equation such as formula (5).

Wherein, T is the sampling time；Q_NFor battery rated capacity；I_kFor charging and discharging currents；w_i,k~(0, Q_i) it is system mode Noise, wherein i=1,2,3,4；v_k~(0, R) is systematic observation noise.

2, parameter identification

It is tested first by volume test, the rated capacity for demarcating lithium battery is 3.2Ah.Secondly low current charge and discharge are carried out Relationship between experimental fit SOC and open-circuit voltage (Open Circuit Voltage, OCV), as shown in Figure 2.

The polynomial fitting of SOC and OCV relationship can be obtained by testing above are as follows:

Then HPPC (Hybrid Pulse Power Characterization) pulse charge-discharge test is carried out to recognize Parameter in second-order model.To R₀And the parameter identification of RC link, it needs in conjunction with battery in pulse charge-discharge test each The response process of SOC point.

It is individually below 30% charging 30 minutes to SOC with SOC is that be 40% electric discharge 30 minutes be 40%, SOC to SOC For 30%, internal resistance R is sketched₀In the identification process in HPPC charge and discharge stage.

As shown in figure 3, the voltage that this stage A → B, which is SOC, to be mutated when being 30%, C → D and E → F are SOC when being 40% The voltage of mutation, B → C are 30 minutes constant-current charge virtual voltage curves, and the C moment starts to stand, and D → E is to stand 90 minutes in fact Border voltage curve.

As shown in figure 4, the voltage that this stage A → B, which is SOC, to be mutated when being 40%, C → D and E → F are SOC when being 30% The voltage of mutation, B → C are 30 minutes constant-current discharge virtual voltage curves, and the C moment starts to stand, and D → E is to stand 90 minutes in fact Border voltage curve.

The section charge and discharge stage B → C is zero state response:

The section charge and discharge stage D → E is zero input response:

It takes battery charging and discharging to terminate moment and stands the average value of end transient voltage mutation to calculate R₀, calculation formula It is as follows:

According to the above method, successively identification obtains the ohmic internal resistance R of each SOC point in charge and discharge direction₀, in the R of identification₀Base On plinth, SOC-R is obtained by polynomial fitting method₀Relation curve, Fig. 5 are HPPC charging experiment SOC-R₀Matched curve, Fig. 6 are HPPC discharge test SOC-R₀Matched curve.

HPPC charging experiment SOC-R₀Six rank multinomial fitting formulas are as follows:

R₀=0.0571 × SOC⁶-0.3695×SOC⁵+1.3369×SOC⁴-2.1067×SOC³

+1.5349×SOC²-0.5171×SOC+0.1244 (10)

HPPC discharge test SOC-R₀Six rank multinomial fitting formulas are as follows:

R₀=6.2941 × SOC⁶-21.1941×SOC⁵+28.2674×SOC⁴-18.9771×SOC³

+6.7523×SOC²-1.2194×SOC+0.1467 (11)

3, it is calculated based on the R0 for improving UPF method

Summary improves UPF algorithm calculating process by taking the 20th constant-current charge process as an example:

(1) enabling k=0, population N is 50, init state variable x₀=[0 0 0.1262 0.04]^TAnd covariance matrix P=10^-4×[1 0 0 0；0 1 00；0 0 10；000 1], if initial weight isI=1,2 ..., N.It adopts Process noise matrix is Q=[10^-9 10^-10 10^-11 10^-8]^T, observation noise matrix is R=0.001, sampling period T= 1s。

(2) weight prediction, the sampling of particle: k=k+1 carries out forecast updating to particle using UKF algorithm and calculates 9 Sigma point set and its weight coefficient.

(3) according to formulaN number of particle weight is calculated, and carries out weight normalizationS is the standard deviation that observational equation introduces noise.

(6) if k reaches the sampling number 9400 of setting, algorithm terminates, and otherwise goes to (2).

Finally obtain state variable matrixThe practical SOH difference of 20th cycle charging and the 40th cycle charging experiment For 0.9635 and 0.9549, when SOC is 0.5, the practical R of the 20th cycle charging and the experiment of the 40th cycle charging₀Respectively For 0.058 Ω and 0.0622 Ω, column write linear equation in two unknowns group:

R can be obtained_NEWIt (0.5) is 0.0402 Ω, R_EOL(0.5) it is 0.528 Ω, when SOC is 0.5, improves the of UPF estimation The R of 20 cycle chargings₀Value is 0.0589 Ω, to can estimate the SOH value of the 20th cycle charging according to formula (1).

4, dynamic lithium battery SOH is calculated

SOH=(0.528-0.0589)/(0.528-0.0402)=0.9617 can be obtained according to formula (1).

Illustrate effect of the invention to further choose, has chosen the different SOC points of the 20th constant current charge-discharge process To compare the method for the present invention (improving UPF method) and not improve UKF method for the estimated accuracy of SOH, as shown in table 1.

The corresponding SOH value of the 20th constant current charge-discharge process difference SOC point of table 1

As shown in Table 1, either charging process still during discharge, this hair is estimated for the SOH of different SOC points The method of bright proposition is superior to unmodified UPF method, to illustrate the superiority of improvement UPF algorithm.

Claims

1. an estimation method of electric vehicle lithium battery SOH, is characterized in that, comprises the following steps:

1) Construct the second-order RC equivalent model of the power lithium battery;

2) Parameter identification of the second-order RC equivalent model, including ohmic internal resistance R ₀ and second-order RC link parameters;

3) The improved UPF method is used to calculate the ohmic internal resistance R ₀ ;

4) Calculate the SOH value of the power lithium battery.

2. the estimation method of a kind of electric vehicle lithium battery SOH according to claim 1, is characterized in that, in described step 1), second-order RC equivalent model is based on Thevenin model and adds the polarization resistance of parallel connection R _pi and polarized capacitance C _pi construct a second-order RC equivalent circuit model, and U _p1 and U _p2 of the two RC links, ohmic internal resistance R ₀ and SOC are used as the state variable x, and the terminal voltage U of the battery is used as the observation , to construct the observation equation.

3. the estimation method of a kind of electric vehicle lithium battery SOH according to claim 2 is characterized in that, the circuit equation of described second-order RC equivalent model is:

U=U _OCV (SOC)-IR ₀ -U _p1 -U _p2

Among them, R ₀ is the ohmic internal resistance, U _OCV (SOC) is the open circuit voltage under SOC, I is the charging and discharging current of the battery, U is the terminal voltage, U _p1 and U _p2 are the voltages of the two RC links, and U _pi is Polarization voltage.

4. the estimation method of a kind of electric vehicle lithium battery SOH according to claim 3, is characterized in that, described observation equation is:

Among them, T is the sampling time, Q _N is the rated capacity of the battery, I _k is the charge and discharge current, w _1,k , w _2,k , w _3,k , w ₄ , _k are the system state noise, ν _k is the system observation Noise, C _p1 and C _p2 are the capacitances of the two RC links respectively, R _p1 and R _p2 are the resistances of the two RC links respectively, and the subscript k indicates the sampling time.

5. the estimation method of a kind of electric vehicle lithium battery SOH according to claim 1, is characterized in that, in described step 2), adopts HPPC pulse charge and discharge experiment to identify the parameter in the second-order model, specifically:

21) Through the capacity test experiment, the rated capacity Q _N of the lithium battery is calibrated;

22) Carry out small current charging and discharging experiment fitting to obtain the relationship expression between SOC and open circuit voltage OCV;

23) Identify the parameters in the second-order RC equivalent circuit model through HPPC pulse charge and discharge experiments, obtain the ohmic internal resistance of each SOC point in the charge and discharge direction, and perform fitting to obtain the relationship between SOC and ohmic internal resistance in the charge and discharge direction expression.

6. the estimation method of a kind of electric vehicle lithium battery SOH according to claim 4, is characterized in that, described step 3) specifically comprises the following steps:

31) Let the sampling time k=0, set the total number of particles N, and initialize the state variable x ₀ , covariance matrix P and initial weights

32) Make the sampling time k=k+1, use the UKF algorithm to predict the particles, and complete the estimation and sampling of the weighted particles;

33) Calculate the weights of all N particles separately And perform normalization processing to obtain normalized weights

34) Count the number of effective particles Set the threshold N _th =2N/3, if N _eff >N _th , go to step 35), if N _eff ≤ _{N th} , for the particle set perform resampling;

35) Obtain the optimal estimate of the state at k sampling time and the variance matrix corresponding to each particle;

36) End when the number of sampling times reaches the set value, otherwise return to step 32).

7. The method for estimating the lithium battery SOH of a kind of electric vehicle according to claim 6, characterized in that, in the described step 33), the i-th particle weight The calculation formula is:

Among them, S is the standard deviation of the noise introduced by the observation equation, and _{Upred_ukf} is _the mean value of the observation prediction.

8. The estimation method of a kind of electric vehicle lithium battery SOH according to claim 7, is characterized in that, the resampling of described step 34) is specifically:

341) For all normalized weights Sort from small to large, and use the normalized weight Corresponding particle estimates with the true value difference put in order;

342) Normalize the weights and difference Quotient Sorting from small to large, if there are two sets of equal data in the quotient, delete the set with the smaller normalized weight and update the sorting;

343) Set the survival of the fittest threshold ω _th as the quotient 2/5 of the mean, if Then keep the corresponding particle weight and particle estimated value, otherwise, delete the corresponding particle weight and particle estimated value, and update the particle set.

9. the estimation method of a kind of electric vehicle lithium battery SOH according to claim 1, is characterized in that, in described step 4), the computing formula of the SOH value of power lithium battery is:

Among them, _REOL (SOC) is the internal resistance value at the end of the battery life, R _NOW (SOC) is the current internal resistance value of the battery, that is, the optimal particle estimation value The ohmic internal resistance in R _NEW (SOC) is the internal resistance value of the battery when it leaves the factory.