CN108872866A - A kind of charge states of lithium ion battery dynamic evaluation and long-acting prediction fusion method - Google Patents
A kind of charge states of lithium ion battery dynamic evaluation and long-acting prediction fusion method Download PDFInfo
- Publication number
- CN108872866A CN108872866A CN201810565223.0A CN201810565223A CN108872866A CN 108872866 A CN108872866 A CN 108872866A CN 201810565223 A CN201810565223 A CN 201810565223A CN 108872866 A CN108872866 A CN 108872866A
- Authority
- CN
- China
- Prior art keywords
- soc
- lithium ion
- ion battery
- battery
- charge
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Abstract
The present invention discloses a kind of charge states of lithium ion battery dynamic evaluation and long-acting prediction fusion method, assesses first with battery charge state of the Extended Kalman filter method to lithium ion battery, obtains charge states of lithium ion battery SOCKEF；Then it is predicted using battery charge state of the Echo State Networks to lithium ion battery, obtains charge states of lithium ion battery SOCESN；Finally to charge states of lithium ion battery SOCKEFWith charge states of lithium ion battery SOCESNIt is weighted fusion, obtains the battery charge state SOC of final lithium ion battery.The present invention improves the adaptability and Evaluation accuracy of existing battery SOC detection method, single method is overcome to carry out the limitation of SOC dynamic evaluation, the fusion method based on model and data-driven is targetedly chosen, the demand of SOC check and evaluation dynamic real-time and long-term long-acting prediction is taken into account.
Description
Technical field
The present invention relates to battery charge state assessment technology fields, and in particular to a kind of charge states of lithium ion battery dynamic
Assessment and long-acting prediction fusion method.
Background technique
Power battery is the critical component of new-energy automobile, and health status and management level are related to the use effect of vehicle
Rate and safe operation.The detection and estimation of battery charge state (SOC) are battery thermal management, balanced management and security reliability pipe
The basis of reason.Accurate SOC estimation can between equilibrium monomer battery difference, optimization charge and discharge strategy, prevent from overheating and overcharge
And over-discharge.Battery status estimation can make battery be used adequately reasonably, and extend battery, be the energy of vehicle
Management provides data foundation, therefore battery status estimation suffers from important meaning for battery management or even vehicle energy management
Justice.It furthers investigate power battery SOC estimation method and realizes that technology develops ev industry and social progress generation is important
It influences.
Lithium ion battery is typical dynamic, time-varying, nonlinear electro-chemical systems, its inner parameter is difficult when on-line operation
Directly to measure, for the state recognition of lithium battery and state estimation, there are still huge challenges at present, while for SOC prediction
For sophisticated method majority still in theory and method level, method system research is still immature.The SOC of the method occurred at present estimates
Although method differs from one another, but generally speaking by the selection of algorithm for estimating initial value, with battery operating condition and use the time
The problems such as constant caused error of model parameter etc. is estimated in variation, and the detection accuracy of existing algorithm and adaptability is caused to reduce, because
This seeks a kind of more accurate SOC appraisal procedure and realization rate is the premise item for realizing the application of lithium ion battery intelligent and safe
Part.
Summary of the invention
The present invention is fixed for existing battery SOC estimation method by model parameter, cannot be with battery actual condition again
The problem of now adjusting, and the detection accuracy of battery charge state and adaptability are reduced, it is charged to provide a kind of lithium ion battery
State dynamic evaluation and long-acting prediction fusion method.
To solve the above problems, the present invention is achieved by the following technical solutions：
A kind of charge states of lithium ion battery dynamic evaluation and long-acting prediction fusion method, specifically include that steps are as follows：
Step 1 is assessed using battery charge state of the Extended Kalman filter method to lithium ion battery, obtain lithium from
Sub- battery charge state SOCKEF；
Step 2 predicted using battery charge state of the Echo State Networks to lithium ion battery, obtain lithium from
Sub- battery charge state SOCESN；
Step 3, charge states of lithium ion battery SOC obtained to step 1KEFWith the obtained lithium ion battery of step 2
State-of-charge SOCESNIt is weighted fusion, obtains the battery charge state SOC of final lithium ion battery, wherein
SOC=SOCKEF×w+SOCESN×(1-w)
Wherein, w is weight, and 0≤w≤1.
Detailed process is as follows for above-mentioned steps 2：
Step 2.1 establishes echo state network model, and determine echo state network outputs and inputs node；
Step 2.2, acquisition M group cell voltage, electric current and temperature data, and corresponding M is obtained to SOC-OCV curve discretization
Group battery charge state data, are divided into K parts of data sets for m group electric current, voltage, electro-temperature and corresponding state-of-charge, each
Contain M/K group data in data set, wherein M and K is setting value；
K parts of data sets are divided into K kind training set and test set, and determine echo state using cross-validation method by step 2.3
The optimized parameter of network, i.e. reserve pool scale N, spectral radius SR, input scaling IS and input displacement IF；It is true using test method(s) simultaneously
Determine degree of rarefication SD；The echo state network determined thus to obtain parameter；
Step 2.4 initializes the obtained echo state network of step 2.3, sets echo state network at random
Input weight matrix and reserve pool internal weights matrix, and remain unchanged；
Step 2.5 selects training set and test set from K parts of data concentrations, and uses the recurrence minimum two with forgetting factor
Multiplication carries out e-learning to echo state network, and real-time update exports weight matrix；
Step 2.6 judges whether e-learning reaches termination condition, if do not reached, return step 2.5 continues to load
Training set and test set are learnt, and are terminated until meeting condition, and obtain trained echo state network；
Cell voltage, electric current and temperature that actual acquisition obtains are input to the obtained echo of step 2.6 by step 2.7
It is predicted in state network, exports charge states of lithium ion battery SOCESN。
In step 2.4, input the element value in weight matrix and reserve pool internal weights matrix between [- 1,1] with
Machine generates.
In step 2.6, the termination condition of e-learning is to reach setting error E rror or step number.
Detailed process is as follows for above-mentioned steps 1：
Step 1.1 establishes simplified GNL equivalent-circuit model；
Step 1.2, using the equivalent electricity of GNL for the simplification established with forgetting factor least square method of recursion to step 1.1
Road model parameter carries out on-line identification, and the battery current that actual acquisition is obtained is as input quantity, with battery terminal voltage and open circuit
Voltage difference starts on-line identification algorithm as observed quantity, determines the number of each equivalence element in simplified GNL equivalent-circuit model
Value is parameter value；
Step 1.3, for obtained by step 1.2 simplify GNL equivalent circuit model parameter value, update spreading kalman filter
Parameter value is corresponded in wave algorithm state space equation formula, then using battery current and temperature as input quantity, is made with battery terminal voltage
For observed quantity, start expanded Kalman filtration algorithm, realizes cell charge state prediction, obtain charge states of lithium ion battery
SOCKEF。
Compared with prior art, the present invention has following features：
1. improving the adaptability and Evaluation accuracy of existing battery SOC detection method, overcomes single method to carry out SOC dynamic and comment
The limitation estimated targetedly chooses the fusion method based on model and data-driven, takes into account SOC check and evaluation dynamic real-time
With the demand of long-term long-acting prediction.
2. establishing the equivalent circuit of the energy storage capacity and battery electrochemical, concentration polarization effect that can preferably describe battery
Model is carried out online Identifying Dynamical Parameters for circuit model parameters, is carried out based on circuit model and be based on spreading kalman
The lithium ion battery complexity dynamic non linear system status assessment of class algorithm, establishes power model parameter and target detection amount SOC
Exact relationship, realize SOC dynamic on-line monitoring and assessment.
3. establish the long-term long-acting algorithm of lithium ion battery SOC assessment prediction based on ESN neural network prediction algorithm, lithium from
Sub- battery terminal voltage, charging and discharging currents, temperature, charge-discharge magnification, charge and discharge number and battery working condition as ESN nerve
The input variable of network is exported using SOC as the prediction of neural network.With lithium ion battery factory charging and discharging curve for most first sample
This training network advanced optimizes network with online acquisition status data, final to realize the ESN neural network prediction based on data
SOC intelligent algorithm.
4. using testing result Weighted Fusion mechanism, the dynamic evaluation of SOC real-time online and long-term long-acting status predication are taken into account
Combination.Realize melting for Kalman filtering algorithm based on circuit model and the neural network prediction algorithm based on data
Conjunction method.Lithium ion battery preliminary work is larger based on the specific gravity that the SOC assessment result of model accounts for final result, that is, weight
Take the larger value；With the passage for using process, ESN algorithm is gradually optimized, and adjustable weight specific gravity keeps ESN neural at this time
Main foundation of the network as assessment result.
5. realizing the combination of lithium ion battery SOC dynamic evaluation and long-acting prediction based on fusion method, play respective
The advantage of method increases adaptation of methods, final raising assessment prediction precision, is the charged shape of lithium ion battery of practical application
State assessment provides etection theory with prediction and supports, the reliability that the accuracy and maintenance for improving assessment are suggested.
Detailed description of the invention
Fig. 1 is the flow chart of charge states of lithium ion battery dynamic evaluation and long-acting prediction fusion method.
Fig. 2 is GNL equivalent-circuit model.
Fig. 3 is simplified GNL equivalent-circuit model.
Fig. 4 is echo state network model.
Fig. 5 is that echo state network predicts flow chart.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention clearer, below in conjunction with specific example, and referring to attached
Figure, the present invention is described in more detail.
Referring to Fig. 1, a kind of charge states of lithium ion battery dynamic evaluation and long-acting prediction fusion method need to undergo following
Process：
(1) the work shape of the batteries such as lithium ion battery end voltage, charging and discharging currents, temperature and charge and discharge number is acquired in real time
State data, the input as SOC assessment and prediction algorithm.
(2) according to the charging and discharging curve of lithium ion battery, it is fitted the relationship of lithium battery open-circuit voltage and SOC, in this, as
Open circuit voltage method detects the foundation of SOC, and when work tables look-up to obtain corresponding SOC value, this hair according to collected open-circuit voltage values
Not using this SOC as the final result of assessment in bright, but using this value as the initial value of current integration method, to improve ampere-hour
The computational accuracy of integration method.
(3) course of work feature based on lithium ion battery establishes the equivalent-circuit model of lithium ion battery, can be accurate
Reflect the state change of battery.On-line identification is carried out to circuit model parameters using the least square method with forgetting factor, to distinguish
Know result dynamic regulating circuit model parameter.
(4) based on lithium ion battery GNL equivalent-circuit model and Identifying Dynamical Parameters result, lithium ion battery is established
The dynamic evaluation algorithm of SOC, in this, as one of the main foundation of the online dynamic evaluation result of lithium ion battery SOC.
(5) with the history charge and discharge data of lithium ion battery and the relationship of SOC training echo status predication neural network
(ESN), the battery parameter acquired in real time with the rise state Processing with Neural Network that training is completed in actual work is carried out operation and obtained
Online SOC value out, in this, as the long-term long-acting main foundation for predicting to dismiss of lithium ion battery SOC, with charge and discharge process
Accumulation this method SOC precision of prediction will be gradually increased to a comparatively ideal stationary value.
(6) predicted value for combining (4) and (5) resulting state-of-charge SOC, is calculated finally using weighting specific gravity appropriate
SOC assessment result.Weight shared by lithium ion battery service life initial stage (4) algorithm evaluation result is larger, using secondary as main
Foundation；With the propulsion of use process, the algorithm that step provides is due to the increase of data volume and the accumulation precision of prediction meeting of experience
It is greatly improved, the weight of (5) assessment result can be properly increased at this time, reaching SOC by method fusion, both dynamic exists
The effect of line assessment and the long-term long-acting prediction of energy.
It is described in detail below for committed step of the invention：
1, the lithium ion battery SOC assessment of Extended Kalman filter method
1.1, equivalent-circuit model is established
Equivalent-circuit model is based on battery working principle, embodies electricity using elements such as resistance, capacitor, constant pressure sources in circuit
The dynamic characteristic in pond.Equivalent-circuit model is suitable for various operating conditions；Related mathematical formulae can be derived according to circuit theory, be convenient for
Simulation analysis；It is easy to consider the influence of temperature in a model.
Battery equivalent circuit model common at present mainly has：Rint model, Thevenin model, PNGV model and GNL
Model.Wherein, PNGV model and GNL model are the direct currents that battery is considered on the basis of Rint model and Thevenin model
Improve and obtain after the features such as characteristic, polarization characteristic, self-discharge characteristics, due to battery in operating condition its inner parameter by these
Characteristic is influenced and is changed, therefore both equivalent-circuit models precision with higher in practical applications, while also phase
To complexity.
General nonlinear (General nonlinear model, GNL) model is a kind of high-order equivalent circuit mould of complexity
Type, the model have used the polarity effect inside two RC link simulated batteries, and precision is higher.As shown in Fig. 2, R in circuit1With
R2It is the activation polarization resistance and concentration polarization resistance of battery, C respectively1And C2Be respectively battery activation polarization capacitor and
Concentration polarization capacitor, RiIt is ohmic internal resistance, CbBe battery plate between capacitor, self discharge internal resistance RsSelf-discharge of battery mistake can be embodied
Kwh loss in journey.The shortcomings that GNL equivalent-circuit model is that parameter is excessive, and identification process is complicated.
Shown in the mathematic(al) representation of the model such as formula (1-1).
The precision of kalman filter method depends on the dynamic model precision for being described system, it is desirable that can use mathematical formulae essence
Relationship between true expression system observed quantity and quantity of state.PNGV model is to increase capacitor on the basis of Thevenin model
Device Cb, the accumulation of electric current in time can be simulated, SOC is caused to change；GNL model is considered on the basis of PNGV model
Self discharge influence factor, and the influence of a RC link simulated battery internal concentration polarization is increased, therefore the model most can
Embody the working characteristics of battery.But the parameter of GNL equivalent-circuit model is more, and structure is complicated, is based on GNL equivalent circuit
The identification algorithm of model and the operand of SOC algorithm for estimating can be relatively large, this can cause very big computational burden to processor,
It is unfavorable for Project Realization.It needs to do GNL equivalent-circuit model certain simplification in engineer application.
The present invention ignores capacitor between self discharge factor and plate on the basis of GNL model, available as shown in Figure 3
Simplified GNL equivalent-circuit model.V in figurebFor battery open circuit voltage；RiFor battery ohmic internal resistance；R1、R2For in battery polarization
Resistance；C1、C2For battery polarization capacitor.R1、C1And R2、C2Link describes the battery electrochemical and concentration polarization effect of battery respectively.
Therefore, tool needs to recognize there are six parameter in simplified GNL equivalent-circuit model.
1.2, lithium battery circuit model parameter on-line identification
Since the equivalence element in battery equivalent circuit model has respective physical significance, and exist with the state of battery
Close relationship.Therefore, the judgement of battery status needs to recognize the parameter in equivalent-circuit model, that is, determines each equivalent
The numerical value of element.There are two types of currently used battery model parameter identification methods：Offline parameter identification and on-line parameter identification.
Off-line parameter identification method is to be fitted voltage retardant curve method by pulse testing can substantially determine battery model
In parameter.However, during electric automobile during traveling, parameter in battery equivalent circuit model can with current discharge multiplying power,
The influence factors such as environment temperature and battery SOC and change, at this point, the determining model parameter of offline parameter identification cannot be real
The actual working state of ground reflection battery.Only online real-time perfoming parameter identification can just obtain the electricity for being best suitable for virtual condition
Pool model parameter improves the precision of the SOC estimation method based on model.
The present invention uses the on-line identification that circuit model parameters are carried out with forgetting factor least square method of recursion (FFRLS),
Specific formula is shown in (1-2)-(1-4).
In formula, forgetting factor λ must select the positive number close to 1, usually in the range of 0.95≤λ≤1.As λ=1,
It is common least square method of recursion that FFRLS algorithm, which is degenerated,.
For the GNL equivalent-circuit model of simplification shown in Fig. 3, the state equation such as formula under frequency domain can be obtained by circuit relationships
Shown in (1-5).
Enable τ1=R1C1, τ2=R2C2, equation both ends are simultaneously multiplied by (τ1s+1)·(τ2S+1 it) obtains：
If a=τ1·τ2, b=τ1+τ2, c=R1+R2+Ri, d=R1τ2+R2τ1+Ri(τ1+τ2) then formula (1-6) can simplify
For：
aVbs2+bVbs+Vb=aRiIs2+dIs+cI+aVs2+bVs+V (1-7)
Enable s=[x (k)-x (k-1)]/T, s2=[x (k) -2x (k-1)+2x (k-2)]/T2, T is the sampling period in formula, will
s、s2It brings the right and left into, calculus of differences is carried out to formula (1-7), arrangement can obtain：
Wherein：
I.e.：
θ=[k1 k2 k3 k4 k5] (1-14)
It, can be directly using its recurrence formula (1-2)-(1-4) to coefficient matrix when carrying out parameter identification using FFRLS algorithm
θ carries out operation, then does further operation by identification result and obtain circuit model parameters R1、R2、C1、C2、Ri.Its process is as follows：
Enable k0=T2+ bT+a, then formula (1-9)-(1-13) can be released：
A=-k0k2 (1-15)
By a=τ1·τ2, b=τ1+τ2, then τ2- b τ+a=0, solves：
By c=R1+R2+Ri, d=R1τ2+R2τ1+Ri(τ1+τ2)：
R2=c-R1-Ri (1-21)
By formula (1-19)-(1-23) it can be concluded that simplify GNL equivalent-circuit model in parameter R1、R2、C1、C2、RiEstimation
Value.
When using FFRLS on-line identification model parameter, it is known that parameter is the V (k) and I (k) at current time, previous moment
V (k-1), I (k-1) and SOC (k-1), V (k-2), I (k-2) and the SOC (k-2) at preceding two moment.
1.3, expanded Kalman filtration algorithm lithium ion battery SOC dynamic assessment method
Expanded Kalman filtration algorithm (Extended Kalman filter, EKF) is proposed for nonlinear system
Method for estimating state is proposed by Stanley Schmidt earliest, for solving the problems, such as the nonlinear navigation of spacecraft.Use this
When algorithm, the nonlinear function of system is made into single order Taylor expansion first, the system equation linearized is recycled to complete
Filter tracking processing.Shown in state space equation and the observational equation such as formula (1-24) of nonlinear system.
In formula, f (xk, uk) be nonlinear system state equation, g (xk, uk) be nonlinear system measurement equation,
Filter calculating process in, v and w it is irrelevant and be Normal Distribution white Gaussian noise.
Using EKF to battery SOC estimate when, usually using the physical quantitys such as the electric current of battery, temperature as the input of system to
U, output Y of the operating voltage as system are measured, and the SOC of battery is included in the quantity of state X of system.
For equivalent-circuit model shown in Fig. 3, can be obtained according to Kirchhoff's law：
Vb(t)=V+RiI(t)+V1(t)+V2(t) (1-27)
State equation, which can be obtained, by formula (1-25), (1-26) and Ah counting method formula is：
Its discretization can be obtained：
Formula (1-29) is arranged can obtain for matrix form：
Δ t is the sampling time in formula, and η is efficiency for charge-discharge, and w (k) is process noise, and it is all side that v (k), which is measurement noise,
The white Gaussian noise that poor known and mean value is zero.
Coefficient matrix is：
It is hereby achieved that the charge states of lithium ion battery assessed value SOC based on spreading kalman algorithmKEF。
2, Echo State Networks predict lithium ion battery SOC
2.1, the foundation of echo state network model
As shown in figure 4, establishing with input unit is K, reserve pool scale is N, and output unit is the echo state network mould of L
Type, wherein input signal is U (n), and output signal is Y (n), and reserve pool intrinsic nerve member output signal is X (n), inputs weight
Matrix W in (N*K), reserve pool internal weights matrix W in (N*N), feedback weight matrix W back (N*L).It is updated inside reserve pool
Shown in state such as formula (2-1), (2-2)：
X (n+1)=f (Win × X (n)+Wres × U (n+1)+Wback × Y (n)) (2-1)
Y (n+1)=fout(X (n+1), U (n+1), Y (n)) (2-2)
Wherein, f (*) is that reserve pool intrinsic nerve member handles function, generally S type function, foutLinear place is exported for network
Function is managed, when netinit, in input signal input echo state network model, reserve pool intrinsic nerve member output signal
X (n) is initialized as X (0)=0, and feedback weight matrix is full null matrix.Input weight matrix, reserve pool internal weights matrix and
Feedback weight matrix is randomly generated, and matrix element value is between -1 to 1, and network training is only with calculating output weight matrix.
Echo state network model has multiple uncertain parameters, including reserve pool scale N, spectral radius SR, input scaling
IS, input displacement IF and degree of rarefication SD, and it is mutually indepedent between parameter, it is independent of each other.The present invention is intersected using a kind of based on K folding
The preferably above unknown parameter of the method for verifying.Voltage, electric current (response curve are obtained by the OCV-SOC curve that battery producer provides
It is to be obtained under certain discharge-rate, represents charging and discharging currents) relation curve of temperature and SOC, electricity is acquired with acquisition device
Real-time current I, voltage V, temperature T when pond works, obtain SOC-OCV curve with open circuit voltage method, curve discretization are obtained
The obtained SOC of data I, V, T and discretization of acquisition is divided into K etc. by the corresponding relationship of voltage V, electric current I, temperature T and SOC
Point, first with K parts for test set, remaining data are training set, then using K-1 parts of data as test set, are left as instruction
Practice collection, and so on, until the 1st part is test set, remaining is training set, adds up to and completes K training and test；Cross validation mistake
Cheng Zhong, input scaling IS and input displacement IF fixed first, appoints the value for taking IS and IF, reserve pool scale N and spectral radius SR are with one
Fixed step-length is changed, and obtains training error and test error, when it is minimum for obtaining the sum of training error and test error,
Corresponding N and SR is theoretically optimized parameter.Can similarly obtain optimized parameter IS and IF, degree of rarefication SD can by the method for exhaustion come
It is chosen, so far echo state network model is completed to establish.To seek obtaining the more accurate optimizing parameter of echo state network, manage
By upper, when K value is bigger, obtained optimizing parameter is more accurate, but K value is excessive will also result in calculating disaster, therefore when selecting K value,
Specifically actual conditions should be combined to carry out value.
2.2, the training and prediction of echo state network
To make echo state network show optimal precision of prediction and generalization ability, respectively to collect data
50%, 60%, 70%, 80%, 90% is training set, remaining is test set.To choose optimal training set and test set, receive respectively
Training and test error after collecting network training and test, when training error and test error all meet set error model
It encloses, and when the sum of the two error is minimum, corresponding is then optimal training set and test set.By optimal training set and test
Collection carries out the study of network using the recurrent least square method with forgetting factor, to make network output and teacher's supervisory signals y
(n) guarantee that error is minimum, then has formula (2-3) between, acquire output weight matrix Wout and make E (k) minimum:
Forgetting factor λ is introduced, then has J (n) satisfaction：
Meet the input weight matrix of formula (2-3) by asking local derviation to calculate：
To avoid network training from falling into local optimum, when meeting formula (2-5), setting constraint error E rror, if instruction
When practicing error less than constraint error, weight matrix is exported, therefore when error function J (n) acquirement extreme value, has reached set mistake
It is the output weight matrix for meeting network optimal conditions that difference constraint, which can be approximately considered the corresponding Wout of gained, by band forgetting factor
Recurrent least square method can obtain the renewal equation of ESN output weight matrix such as (2-6), wherein QkFor weight gain matrix.
Wout k+1=Wout k+QkE(k) (2-6)
It is as follows referring to Fig. 5, echo state network model prediction SOC implementation steps：
Step 1：Data acquisition, determination output and input node, establish echo state network model；
Step 2：Initial cells data are divided into K kind training set and test set, cross validation determine optimized parameter N, SR,
IS, IF, test method(s) determine SD；
Step 3：Netinit, random setting input weight matrix, reserve pool internal weights matrix, matrix element value
It is randomly generated, and remains unchanged between [- 1,1]；
Step 4：It is preferred that training set and test set, carry out e-learning using the recurrent least square method with forgetting factor,
Real-time update exports weight matrix；
Step 5：Judge whether e-learning reaches termination condition (reaching setting error E rror or step number), if do not reached
It arrives, return step 4, continues to load training set and test set is learnt, terminated until meeting condition；
Step 6：After the completion of network training, the cell voltage V that is obtained according to actual acquisition, electric current I, temperature T are as network
Input is predicted that SOC is predicted in outputESNValue.
3, charge states of lithium ion battery fusion forecasting
It will be above based on obtained by the two kinds of algorithm evaluation predictions of expanded Kalman filtration algorithm and Echo State Networks
Charge states of lithium ion battery SOC assessment prediction numerical value SOCKEFAnd SOCESNThe pre- of final result is carried out using weighting algorithm
Survey output, i.e. SOC=SOCKEF×w+SOCESN× (1-w), wherein w value range is 0-1.
As w=0, the reproduction prediction result of final SOC is SOC=SOCESN, i.e., ESN neural network algorithm is used completely
Assessment result；As w=1, SOC=SOCKEF, i.e., the assessment result of spreading kalman algorithm is used completely.
When carrying out the assessment of lithium ion battery SOC on-line prediction, w can be set in be set as being closer between 0-1 1 numerical value,
It is larger to can be the specific gravity that expanded Kalman filtration algorithm accounts in assessment result using initial stage in battery, when with using
Between extension ESN network evaluation precision be gradually increased, can be gradually reduced w increase ESN algorithm specific gravity.
In conclusion a kind of charge states of lithium ion battery dynamic evaluation proposed by the invention and the long-acting prediction side of fusion
Method, including steps are as follows：
Step 1 is assessed using battery charge state of the Extended Kalman filter method to lithium ion battery, obtain lithium from
Sub- battery charge state SOCKEF.The specific steps are that：
Step 1.1 establishes simplified GNL equivalent-circuit model；
Step 1.2, using the equivalent electricity of GNL for the simplification established with forgetting factor least square method of recursion to step 1.1
Road model parameter carries out on-line identification, and the battery current that actual acquisition is obtained is as input quantity, with battery terminal voltage and open circuit
Voltage difference starts on-line identification algorithm as observed quantity, determines the number of each equivalence element in simplified GNL equivalent-circuit model
Value is parameter value；
Step 1.3, for obtained by step 1.2 simplify GNL equivalent circuit model parameter value, update spreading kalman filter
Parameter value is corresponded in wave algorithm state space equation formula (1-30), then using battery current and temperature as input quantity, with battery-end
Voltage starts expanded Kalman filtration algorithm as observed quantity, realizes cell charge state prediction, it is charged to obtain lithium ion battery
State SOCKEF。
Step 2 predicted using battery charge state of the Echo State Networks to lithium ion battery, obtain lithium from
Sub- battery charge state SOCESN.The specific steps are that：
Step 2.1 establishes echo state network model, and determine echo state network outputs and inputs node；
Step 2.2, acquisition M group cell voltage, electric current and temperature data, and corresponding M is obtained to SOC-OCV curve discretization
Group battery charge state data, are divided into K parts of data sets for m group electric current, voltage, electro-temperature and corresponding state-of-charge, each
Contain M/K group data in data set, wherein M and K is setting value；
K parts of data sets are divided into K kind training set and test set, and determine echo state using cross-validation method by step 2.3
The optimized parameter of network, i.e. reserve pool scale N, spectral radius SR, input scaling IS and input displacement IF；It is true using test method(s) simultaneously
Determine degree of rarefication SD；The echo state network determined thus to obtain parameter；
Step 2.4 initializes the obtained echo state network of step 2.3, sets echo state network at random
Input weight matrix and reserve pool internal weights matrix, the element value in matrix is randomly generated between [- 1,1], and protects
It holds constant；
Step 2.5 selects training set and test set from K parts of data concentrations, and uses the recurrence minimum two with forgetting factor
Multiplication carries out e-learning to echo state network, and real-time update exports weight matrix；
Step 2.6 judges e-learning whether to reach termination condition that (termination condition of e-learning is to reach setting error
Error or step number), if do not reached, return step 2.5 continues to load training set and test set is learnt, until meeting item
Part terminates, and obtains trained echo state network；
Cell voltage, electric current and temperature that actual acquisition obtains are input to the obtained echo of step 2.6 by step 2.7
It is predicted in state network, exports charge states of lithium ion battery SOCESN。
Step 3, charge states of lithium ion battery SOC obtained to step 1KEFWith the obtained lithium ion battery of step 2
State-of-charge SOCESNIt is weighted fusion, obtains the battery charge state SOC of final lithium ion battery, wherein
SOC=SOCKEF×w+SOCESN×(1-w)
Wherein, w is weight, and 0≤w≤1.
It should be noted that although the above embodiment of the present invention be it is illustrative, this be not be to the present invention
Limitation, therefore the invention is not limited in above-mentioned specific embodiment.Without departing from the principles of the present invention, all
The other embodiment that those skilled in the art obtain under the inspiration of the present invention is accordingly to be regarded as within protection of the invention.
Claims (5)
1. a kind of charge states of lithium ion battery dynamic evaluation and long-acting prediction fusion method, characterized in that specifically include step
It is as follows：
Step 1 is assessed using battery charge state of the Extended Kalman filter method to lithium ion battery, obtains lithium-ion electric
Pond state-of-charge SOCKEF；
Step 2 is predicted using battery charge state of the Echo State Networks to lithium ion battery, obtains lithium-ion electric
Pond state-of-charge SOCESN；
Step 3, charge states of lithium ion battery SOC obtained to step 1KEFIt is charged with the obtained lithium ion battery of step 2
State SOCESNIt is weighted fusion, obtains the battery charge state SOC of final lithium ion battery, wherein
SOC=SOCKEF×w+SOCESN×(1-w)
Wherein, w is weight, and 0≤w≤1.
2. a kind of charge states of lithium ion battery dynamic evaluation according to claim 1 and long-acting prediction fusion method,
It is characterized in, detailed process is as follows for step 2：
Step 2.1 establishes echo state network model, and determine echo state network outputs and inputs node；
Step 2.2, acquisition M group cell voltage, electric current and temperature data, and corresponding M group electricity is obtained to SOC-OCV curve discretization
M group electric current, voltage, electro-temperature and corresponding state-of-charge are divided into K parts of data sets, each data by pond state-of-charge data
It concentrates and contains M/K group data, wherein M and K is setting value；
K parts of data sets are divided into K kind training set and test set, and determine echo state network using cross-validation method by step 2.3
Optimized parameter, i.e. reserve pool scale N, spectral radius SR, input scaling IS and input displacement IF；It is determined simultaneously using test method(s) dilute
Dredge degree SD；The echo state network determined thus to obtain parameter；
Step 2.4 initializes the obtained echo state network of step 2.3, random to set the defeated of echo state network
Enter weight matrix and reserve pool internal weights matrix, and remains unchanged；
Step 2.5 selects training set and test set from K parts of data concentrations, and uses the recurrent least square method with forgetting factor
E-learning is carried out to echo state network, real-time update exports weight matrix；
Step 2.6 judges whether e-learning reaches termination condition, if do not reached, return step 2.5 continues load training
Collection and test set are learnt, and are terminated until meeting condition, and obtain trained echo state network；
Cell voltage, electric current and temperature that actual acquisition obtains are input to the obtained echo state of step 2.6 by step 2.7
It is predicted in network, exports charge states of lithium ion battery SOCESN。
3. a kind of charge states of lithium ion battery dynamic evaluation according to claim 2 and long-acting prediction fusion method,
It is characterized in, in step 2.4, it is random between [- 1,1] inputs the element value in weight matrix and reserve pool internal weights matrix
It generates.
4. a kind of charge states of lithium ion battery dynamic evaluation according to claim 2 and long-acting prediction fusion method,
It is characterized in, in step 2.6, the termination condition of e-learning is to reach setting error E rror or step number.
5. a kind of charge states of lithium ion battery dynamic evaluation according to claim 1 and long-acting prediction fusion method, step
Rapid 1 detailed process is as follows：
Step 1.1 establishes simplified GNL equivalent-circuit model；
Step 1.2, using the GNL equivalent circuit mould for the simplification established with forgetting factor least square method of recursion to step 1.1
Shape parameter carries out on-line identification, and the battery current that actual acquisition is obtained is as input quantity, with battery terminal voltage and open-circuit voltage
Difference starts on-line identification algorithm as observed quantity, determines that the numerical value of each equivalence element in simplified GNL equivalent-circuit model is
Parameter value；
Step 1.3, for obtained by step 1.2 simplify GNL equivalent circuit model parameter value, update Extended Kalman filter calculate
Parameter value is corresponded in method state space equation formula, then using battery current and temperature as input quantity, using battery terminal voltage as sight
Measurement starts expanded Kalman filtration algorithm, realizes cell charge state prediction, obtains charge states of lithium ion battery SOCKEF。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810565223.0A CN108872866B (en) | 2018-06-04 | 2018-06-04 | Dynamic evaluation and long-acting prediction fusion method for charge state of lithium ion battery |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810565223.0A CN108872866B (en) | 2018-06-04 | 2018-06-04 | Dynamic evaluation and long-acting prediction fusion method for charge state of lithium ion battery |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108872866A true CN108872866A (en) | 2018-11-23 |
| CN108872866B CN108872866B (en) | 2021-02-05 |
Family
ID=64335055
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810565223.0A Active CN108872866B (en) | 2018-06-04 | 2018-06-04 | Dynamic evaluation and long-acting prediction fusion method for charge state of lithium ion battery |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108872866B (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109581225A (en) * | 2018-12-28 | 2019-04-05 | 深圳市超思维电子股份有限公司 | The energy state evaluation method and battery management system of battery on-line parameter identification |
| CN109598449A (en) * | 2018-12-14 | 2019-04-09 | 北京匠芯电池科技有限公司 | Lithium ion battery packet performance estimating method based on charge data |
| CN109685207A (en) * | 2018-11-27 | 2019-04-26 | 电子科技大学 | A kind of multi core chip thermal management algorithm based on Recognition with Recurrent Neural Network |
| CN110501652A (en) * | 2019-09-05 | 2019-11-26 | 上海毅信环保科技有限公司 | A kind of retired lithium battery active volume fast evaluation method and assessment device |
| CN110531281A (en) * | 2019-09-09 | 2019-12-03 | 合肥工业大学 | The method and system of health status for estimated driving force secondary battery unit |
| CN110888059A (en) * | 2019-12-03 | 2020-03-17 | 西安科技大学 | Algorithm based on improved random forest combined cubature Kalman power battery state of charge estimation |
| CN111062170A (en) * | 2019-12-03 | 2020-04-24 | 广东电网有限责任公司 | Transformer top layer oil temperature prediction method |
| CN112018465A (en) * | 2020-10-13 | 2020-12-01 | 北京理工大学 | Multi-physical-field-constrained intelligent quick charging method for lithium ion battery |
| CN112798967A (en) * | 2020-12-04 | 2021-05-14 | 电子科技大学 | Long-term and short-term prediction method for solid oxide fuel cell |
| CN113406500A (en) * | 2021-06-29 | 2021-09-17 | 同济大学 | Method for estimating residual electric quantity of power lithium battery |
| CN113495213A (en) * | 2020-03-20 | 2021-10-12 | 北京京东乾石科技有限公司 | Electric quantity prediction model construction method, electric quantity prediction model carrying equipment deployment method and electric quantity prediction model carrying equipment deployment device |
| CN113795760A (en) * | 2019-10-18 | 2021-12-14 | 株式会社Lg新能源 | SOC estimation apparatus and method |
| US20220203028A1 (en) * | 2020-12-30 | 2022-06-30 | Baxter International Inc. | System and method for generating battery alarms in infusion devices |
| CN114705990A (en) * | 2022-03-31 | 2022-07-05 | 上海玫克生储能科技有限公司 | Battery cluster state of charge estimation method and system, electronic equipment and storage medium |
| CN115453376A (en) * | 2022-10-26 | 2022-12-09 | 国网浙江省电力有限公司宁波市北仑区供电公司 | State monitoring method and system for power battery of electric automobile and electric automobile |
| CN113795760B (en) * | 2019-10-18 | 2024-05-17 | 株式会社Lg新能源 | SOC estimation apparatus and method |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102930173A (en) * | 2012-11-16 | 2013-02-13 | 重庆长安汽车股份有限公司 | SOC(state of charge) online estimation method for lithium ion battery |
| CN103295075A (en) * | 2013-04-01 | 2013-09-11 | 沈阳航空航天大学 | Ultra-short-term power load forecasting and early warning method |
| CN103293487A (en) * | 2013-06-28 | 2013-09-11 | 哈尔滨工业大学 | Lithium ion battery service life forecasting method based on integrated model |
| CN103336913A (en) * | 2013-07-25 | 2013-10-02 | 哈尔滨工业大学 | Method for monitoring state and predicting cut-off voltage of spatial lithium ion battery |
| CN105548896A (en) * | 2015-12-25 | 2016-05-04 | 南京航空航天大学 | Power-cell SOC online closed-loop estimation method based on N-2RC model |
| CN107402353A (en) * | 2017-06-30 | 2017-11-28 | 中国电力科学研究院 | A kind of state-of-charge to lithium ion battery is filtered the method and system of estimation |
| CN107748336A (en) * | 2017-11-06 | 2018-03-02 | 清华大学 | The state-of-charge On-line Estimation method and system of lithium ion battery |
-
2018
- 2018-06-04 CN CN201810565223.0A patent/CN108872866B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102930173A (en) * | 2012-11-16 | 2013-02-13 | 重庆长安汽车股份有限公司 | SOC(state of charge) online estimation method for lithium ion battery |
| CN103295075A (en) * | 2013-04-01 | 2013-09-11 | 沈阳航空航天大学 | Ultra-short-term power load forecasting and early warning method |
| CN103293487A (en) * | 2013-06-28 | 2013-09-11 | 哈尔滨工业大学 | Lithium ion battery service life forecasting method based on integrated model |
| CN103336913A (en) * | 2013-07-25 | 2013-10-02 | 哈尔滨工业大学 | Method for monitoring state and predicting cut-off voltage of spatial lithium ion battery |
| CN105548896A (en) * | 2015-12-25 | 2016-05-04 | 南京航空航天大学 | Power-cell SOC online closed-loop estimation method based on N-2RC model |
| CN107402353A (en) * | 2017-06-30 | 2017-11-28 | 中国电力科学研究院 | A kind of state-of-charge to lithium ion battery is filtered the method and system of estimation |
| CN107748336A (en) * | 2017-11-06 | 2018-03-02 | 清华大学 | The state-of-charge On-line Estimation method and system of lithium ion battery |
Non-Patent Citations (1)
| Title |
|---|
| 唐燕青: "基于KF-ESN算法的储能电站磷酸铁锂电池SOC估算的研究", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109685207A (en) * | 2018-11-27 | 2019-04-26 | 电子科技大学 | A kind of multi core chip thermal management algorithm based on Recognition with Recurrent Neural Network |
| CN109598449A (en) * | 2018-12-14 | 2019-04-09 | 北京匠芯电池科技有限公司 | Lithium ion battery packet performance estimating method based on charge data |
| CN109598449B (en) * | 2018-12-14 | 2023-09-12 | 蓝谷智慧(北京)能源科技有限公司 | Lithium ion battery pack performance evaluation method based on charging data |
| CN109581225A (en) * | 2018-12-28 | 2019-04-05 | 深圳市超思维电子股份有限公司 | The energy state evaluation method and battery management system of battery on-line parameter identification |
| CN110501652A (en) * | 2019-09-05 | 2019-11-26 | 上海毅信环保科技有限公司 | A kind of retired lithium battery active volume fast evaluation method and assessment device |
| CN110501652B (en) * | 2019-09-05 | 2021-11-19 | 上海毅信环保科技有限公司 | Rapid assessment method and device for available capacity of retired lithium battery |
| CN110531281B (en) * | 2019-09-09 | 2021-08-13 | 合肥工业大学 | Method and system for estimating state of health of power storage battery unit |
| CN110531281A (en) * | 2019-09-09 | 2019-12-03 | 合肥工业大学 | The method and system of health status for estimated driving force secondary battery unit |
| CN113795760B (en) * | 2019-10-18 | 2024-05-17 | 株式会社Lg新能源 | SOC estimation apparatus and method |
| CN113795760A (en) * | 2019-10-18 | 2021-12-14 | 株式会社Lg新能源 | SOC estimation apparatus and method |
| CN110888059A (en) * | 2019-12-03 | 2020-03-17 | 西安科技大学 | Algorithm based on improved random forest combined cubature Kalman power battery state of charge estimation |
| CN110888059B (en) * | 2019-12-03 | 2022-04-19 | 西安科技大学 | Charge state estimation algorithm based on improved random forest combined volume Kalman |
| CN111062170A (en) * | 2019-12-03 | 2020-04-24 | 广东电网有限责任公司 | Transformer top layer oil temperature prediction method |
| CN113495213A (en) * | 2020-03-20 | 2021-10-12 | 北京京东乾石科技有限公司 | Electric quantity prediction model construction method, electric quantity prediction model carrying equipment deployment method and electric quantity prediction model carrying equipment deployment device |
| CN112018465B (en) * | 2020-10-13 | 2021-01-29 | 北京理工大学 | Multi-physical-field-constrained intelligent quick charging method for lithium ion battery |
| CN112018465A (en) * | 2020-10-13 | 2020-12-01 | 北京理工大学 | Multi-physical-field-constrained intelligent quick charging method for lithium ion battery |
| CN112798967A (en) * | 2020-12-04 | 2021-05-14 | 电子科技大学 | Long-term and short-term prediction method for solid oxide fuel cell |
| US20220203028A1 (en) * | 2020-12-30 | 2022-06-30 | Baxter International Inc. | System and method for generating battery alarms in infusion devices |
| CN113406500A (en) * | 2021-06-29 | 2021-09-17 | 同济大学 | Method for estimating residual electric quantity of power lithium battery |
| CN114705990A (en) * | 2022-03-31 | 2022-07-05 | 上海玫克生储能科技有限公司 | Battery cluster state of charge estimation method and system, electronic equipment and storage medium |
| CN114705990B (en) * | 2022-03-31 | 2023-10-20 | 上海玫克生储能科技有限公司 | Method and system for estimating state of charge of battery cluster, electronic device and storage medium |
| CN115453376A (en) * | 2022-10-26 | 2022-12-09 | 国网浙江省电力有限公司宁波市北仑区供电公司 | State monitoring method and system for power battery of electric automobile and electric automobile |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108872866B (en) | 2021-02-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108872866A (en) | A kind of charge states of lithium ion battery dynamic evaluation and long-acting prediction fusion method | |
| Chang et al. | Lithium-ion battery state of health estimation using the incremental capacity and wavelet neural networks with genetic algorithm | |
| CN106443474B (en) | A kind of electrokinetic cell system service life Decline traits quickly know method for distinguishing | |
| CN104505894B (en) | Power management system and state estimation method based on mining lithium ion batteries | |
| CN107576919A (en) | Power battery charged state estimating system and method based on ARMAX models | |
| CN104678316B (en) | Charge states of lithium ion battery evaluation method and device | |
| CN107390127A (en) | A kind of SOC estimation method | |
| CN103020445B (en) | A kind of SOC and SOH Forecasting Methodology of electric-vehicle-mounted ferric phosphate lithium cell | |
| CN105717460B (en) | A kind of power battery SOC methods of estimation and system based on nonlinear observer | |
| CN104849671B (en) | A kind of battery capacity detecting system based on combination neural net | |
| CN103744026A (en) | Storage battery state of charge estimation method based on self-adaptive unscented Kalman filtering | |
| CN102565710A (en) | Method and apparatus for assessing battery state of health | |
| CN102468521A (en) | Method and apparatus for assessing battery state of health | |
| CN105319515A (en) | A combined estimation method for the state of charge and the state of health of lithium ion batteries | |
| Boujoudar et al. | Lithium-ion batteries modeling and state of charge estimation using artificial neural network | |
| CN113253116A (en) | Lithium ion battery state of charge estimation method and storage medium | |
| CN110515011A (en) | A kind of Accurate Estimation Method of lithium-ion-power cell SOC | |
| CN104698382A (en) | Method for predicting the SOC and SOH of battery pack | |
| Sheikhan et al. | State of charge neural computational models for high energy density batteries in electric vehicles | |
| CN104680024A (en) | Method for predicting remaining useful life of lithium ion battery based on GA (Genetic Algorithms) and ARMA (Auto Regressive and Moving Average) models | |
| Feng et al. | Intelligent battery management for electric and hybrid electric vehicles: A survey | |
| Rui-hao et al. | Battery state of charge estimation for electric vehicle based on neural network | |
| Hu et al. | Study on SOC estimation of lithium battery based on improved BP neural network | |
| CN106372272A (en) | Lithium battery capacity and service life prediction method based on generalized degradation model and multiscale analysis | |
| Zhao et al. | Digital twin-driven estimation of state of charge for Li-ion battery |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |