CN110501646A - Off-line lithium battery residual capacity estimation method - Google Patents
Off-line lithium battery residual capacity estimation method Download PDFInfo
- Publication number
- CN110501646A CN110501646A CN201910804963.XA CN201910804963A CN110501646A CN 110501646 A CN110501646 A CN 110501646A CN 201910804963 A CN201910804963 A CN 201910804963A CN 110501646 A CN110501646 A CN 110501646A
- Authority
- CN
- China
- Prior art keywords
- remaining capacity
- lithium battery
- value
- status monitoring
- monitoring parameters
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/36—Arrangements for testing, measuring or monitoring the electrical condition of accumulators or electric batteries, e.g. capacity or state of charge [SoC]
- G01R31/382—Arrangements for monitoring battery or accumulator variables, e.g. SoC
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
Abstract
The invention belongs to the field of life prediction, and particularly relates to an off-line lithium battery remaining capacity estimation method which comprises the steps of collecting lithium battery state monitoring parameters, and constructing a training set, a verification set and a test set. And constructing a heterogeneous kernel function, solving heterogeneous kernel coefficients and hyper-parameters of the heterogeneous kernel function, and training a lithium battery residual capacity estimation model. And verifying the precision of the estimation model of the residual electric quantity of the lithium battery obtained by training through error analysis. And inputting the collected battery state monitoring parameters of each charge-discharge cycle of the lithium battery to be detected during working into the trained residual electric quantity estimation model of the lithium battery, so that the residual electric quantity can be estimated. The method well solves the problem that the residual capacity is difficult to accurately estimate in the off-line state, and provides a better scheme for the research of on-line lithium battery product residual capacity estimation, service life estimation and anomaly detection.
Description
Technical field
The invention belongs to lives of product to estimate field, and in particular to one kind passes through lithium battery status monitoring parameter Estimation battery
The method of remaining capacity.
Background technique
Life of product estimation refers to a certain product facility, is carried out to it from now to cannot use completely, or cannot be complete
At the time Estimate until its predetermined function.By implementing life estimation to product, equipment can be allowed to obtain timely and necessary
Maintenance, promoted product using duration and the generation for avoiding serious accident.Lithium battery is remaining electric in use
Amount is constantly degenerated, and is also the crucial intermediate parameters for calculating general health indicator (SOC, SOH).Therefore in the lithium battery remaining longevity
Prediction field is ordered, cell degradation state can also be assessed with lithium battery remaining capacity.Remaining capacity estimation is substantially to electricity
The estimation problem of this product of pond progress remaining life.
Online battery dump energy estimation refers under working, estimates battery dump energy.With reality
The condition for testing room measurement remaining capacity is compared with method, and carrying out On-line Estimation to satellite lithium battery has real-time, discharging condition
The problems such as special and model foundation is difficult, can not directly calculate remaining capacity according to related resolution formula.But battery is working
In the process, a large amount of status monitoring parameters, such as voltage, electric current, temperature can be obtained in the case where not interfering battery to work.Root
According to experience, these parameters have great correlation with remaining capacity.Therefore, according under status monitoring parameter Estimation presence
The remaining capacity of battery becomes current mainstream research method.In the actual operation process, off-line state (such as laboratory is usually used
Degenerate experiment) Imitating work online environment foundation status monitoring parameter and remaining capacity between relationship carry out approximate evaluation and exist
Residual electric quantity under linear state.
Machine learning method is good at excavating that the scale of construction is huge, the hiding information behind the numerous data of structure, is longer than processing not
Determine the relationship between information, intelligence and practicability with height.Therefore, it is established here using Method Using Relevance Vector Machine (RVM)
Relationship under off-line state between status monitoring parameter and remaining capacity.
Data-oriented collectionN is sample size, XiFor input vector, yiFor corresponding to XiObservation to
Amount, f () are the obtained best approximating function that Method Using Relevance Vector Machine learns on data set D.The basic principle of RVM can indicate
For an abstract multi-parameter nonlinear model:
Y=f (X)+εn
In formula, εnObey N (0, σ2), εnIndicate the white Gaussian noise noise output of y.
F=φ ω+ε
Wherein, w=(w0,w1,...,wN)TIt is N+1 right-safeguarding value column vector, k is kernel function, and φ form is as follows:
Known by Bayes' theorem, and p (y | X) obedience N (y | f (X), σ2), then for data set D, possibility predication are as follows:
According to the Regularization Theory in SVM, since above formula lacks regularization parameter, if directly asking very big to ω therein
Necessarily there is over-fitting in possibility predication, result.So in order to make model have good generalization without regard to excessively quasi-
It closes, ω is set to the Gaussian prior that mean value is zero and is distributed, each weight is made to possess an individual hyper parameter αi, i=(1,
2 ..., N), the wherein prior distribution of ω are as follows:
According to Bayesian Estimation, after provable introducing hyper parameter α, the Posterior distrbutionp of weight are as follows:
By calculating, the mean value and variance of Posterior distrbutionp can be obtained are as follows:
μ=σ-2ΣφTy
Σ=(σ-2φTφ+diag(α0,…,αN))-1
Later, need to solve the hyper parameter that variance is less than threshold values.The general sparse pattra leaves using propositions such as Dempester
This algorithm carrys out Optimization Solution process.If giving a new input vector x*, then corresponding point estimate be
y*(x*)=μTφ(x*)
Relevance vector machine for regression can be with the relationship between each battery status monitoring parameters of Fuzzy Processing and remaining capacity, In
Mapping function between the two is established in the case where not knowing explicit mathematic(al) representation, avoid analytic modell analytical model narrow application range,
The low problem of diagnosis performance, substantially increases the robustness of model.But there are the following problems for this method:
(1) different kernel functions has different characteristics, and how academic circles at present is for select kernel function to have no explicitly
Regulation, therefore the kernel function rule of thumb selected has greatly uncertainty, may not necessarily access optimum.
(2) fitting of the hyper parameter strong influence of kernel function Method Using Relevance Vector Machine and Generalization Capability, it is therefore desirable to super
Parameter carries out optimizing solution.
Summary of the invention
In view of the defects existing in the prior art, the purpose of the present invention is to provide a kind of offline lithium battery remaining capacity estimations
Method is selected by establishing the remaining capacity estimation model based on multicore Method Using Relevance Vector Machine, and using quantum genetic algorithm (QGA)
Kernel function coefficient and hyper parameter provide significantly more efficient method to improve remaining capacity estimation precision.
To realize the above-mentioned technical purpose, the technical solution adopted by the present invention is that:
The present invention provides a kind of offline lithium battery method for estimating remaining capacity, comprising the following steps:
The first step, the acquisition of lithium battery data.
The battery status monitoring parameters of each charge and discharge cycles when lithium battery works under off-line state are extracted in seconds,
And obtain calculate each charge and discharge cycles remaining capacity it is true as the corresponding remaining capacity of each group battery status monitoring parameters
Value, building form lithium battery data set by battery status monitoring parameters and corresponding remaining capacity true value.
Second step pre-processes the data in lithium battery data set, and by the lithium battery data set after pretreatment
In data be divided into training set and test set according to a certain percentage (such as 4:1).
Third step constructs isomery kernel function and solves its isomery core coefficient and hyper parameter；
4th step, training lithium battery remaining capacity estimation model；
Isomery core coefficient obtained in third step and hyper parameter are substituted into isomery kernel function, utilize each group electricity in training set
Pond status monitoring parameter is used as input, the corresponding remaining capacity true value conduct of each group battery status monitoring parameters in training set
Output is trained it, the lithium battery remaining capacity estimation model after being trained；
Each group battery status monitoring parameters in test set are input to the lithium battery remaining capacity estimation model after training
In, obtain the corresponding remaining capacity estimation value of each group battery status monitoring parameters in test set；
5th step utilizes the corresponding remaining capacity true value of each group battery status monitoring parameters in test set and remaining electricity
It measures estimated value and calculates error analysis parameter, the lithium battery remaining capacity estimation model accuracy that verifying training obtains；
6th step acquires the battery status monitoring parameters of each charge and discharge cycles when lithium battery to be measured work, is input to instruction
The estimation of remaining capacity can be carried out in the lithium battery remaining capacity estimation model got.
Wherein battery status monitoring parameters include constant-current charge time, constant voltage charging time, constant-current discharge time, chargin level
Equal temperature, electric discharge mean temperature.xi(t), when i=1,2,3,4,5 indicates constant-current charge corresponding in single charge and discharge cycles
Between, constant voltage charging time, the constant-current discharge time, charging mean temperature and electric discharge mean temperature.
The remaining capacity that each charge and discharge cycles are calculated using constant-current discharge formula, that is, formula (1), as each group battery shape
The corresponding remaining capacity true value of state monitoring parameters.
C (t)=Icc(t)(tcc(t)-t0(t)) (1)
Wherein t0(t) and tcc(t) at the beginning of respectively indicating the constant-current charge in the t times charge and discharge cycles and when terminating
Between；Icc(t) the Constant Electric Current flow valuve in the t times charge and discharge cycles is indicated.
Constructed lithium battery data set Xi={ xi(t)~C (t) | t=1 ..., T；I=1,2,3,4,5 }, wherein xi(t),i
In=1,2,3,4,5 the t times charge and discharge cycles of expression when corresponding constant-current charge time, constant voltage charging time, constant-current discharge
Between, charging mean temperature and electric discharge mean temperature.T is charge and discharge cycles number.C (t) indicates battery status monitoring parameters xi(t)
Corresponding remaining capacity true value.
In second step of the invention, pretreatment, which refers to, first carries out at missing values the data outliers in lithium battery data set
Reason, it is ensured that inputoutput data dimension meets the requirements.Then the lithium battery data set after having handled is normalized,
Influence of the dimension to RVM is removed, nothing is carried out to battery status monitoring parameters sequence and the true value sequence of corresponding remaining capacity
Dimensionization processing, so that its element is between 0 to 1.
In second step of the invention, isomery kernel function is constructed based on kernel function commonly used in the art:
Wherein, dmFor m-th of kernel function Km(xi, x) weight coefficient, andxiIt is battery for input
Status monitoring parameter xi(t), i=1,2,3,4,5, x be output i.e. residual electric quantity C (t).
Kernel function commonly used in the art includes but is not limited to following several:
Gaussian kernel function (RBF):
kRBF(xi, x) and=exp (- γ | | x-xi||2) (3)
Polynomial kernel function (Poly):
Sigmoid kernel function:
Chi-Square kernel function:
Kernel function well known in the art is numerous to list herein there are also very much.Different kernel functions has the different overall situations
Or local characteristics, multiple kernel function synthesis can be very good to learn from other's strong points to offset one's weaknesses.The present invention provides a kind of using aforementioned four seed nucleus letter
The isomery kernel function of the constructed multicore Method Using Relevance Vector Machine of number, indicates are as follows:
Wherein: a, b, c, d, e, f, g, h are isomery core coefficient and hyper parameter to be asked, xiIt is battery status prison for input
Survey parameter xi(t), i=1,2,3,4,5, x be the output i.e. residual electric quantity of estimation.
Next, the isomery core system of the isomery kernel function using quantum genetic algorithm (QGA) optimization multicore Method Using Relevance Vector Machine
Several and hyper parameter (a, b, c, d, e, f, g, h), the method is as follows:
1. initializing
The parameter of initialization include Population Size, chromosome number (optimizing number of parameters, as 8), chromosome length, dyeing
Body decimal system maximum value, chromosome decimal system minimum value, the number of iterations and quantum rotation angle；
2. introduce value that quantum coding method is chromosome (isomery core coefficient i.e. to be asked and hyper parameter (a, b, c, d,
E, f, g, h)) binary coding is carried out, it then turns to the decimal system and participates in subsequent step；
3. calculating fitness R2, find current optimal solution；
Training set is extracted a part of sample (taking one every four) as verifying collection, remaining sample by the method based on cross validation
This is as new training set.
(a, b, c, d, e, f, the g, h) obtained according to current search, finds out isomery core, and new training set is substituted into multicore phase
The isomery kernel function training of vector machine is closed, and is verified using verifying collection, the estimated value of the remaining capacity of verifying collection is found out, and
Verifying, which is calculated, by formula (8) concentrates the corresponding remaining capacity estimation value of each group battery status monitoring parameters and remaining capacity true value
Between goodness of fit coefficient of determination R2:
Wherein, CiIndicate that the corresponding remaining capacity true value of i-th group of sample is concentrated in verifying；Indicate multicore Method Using Relevance Vector Machine
The estimated value of the corresponding remaining capacity of i-th group of sample is concentrated to verifying；Indicate that verifying concentrates the remaining capacity of all samples true
The average value of real value, n are verifying collection sample size.
4. judging whether to meet termination condition, termination condition is that verifying concentrates each group battery status monitoring parameters corresponding surplus
Goodness of fit coefficient of determination R between remaining electricity estimation value and remaining capacity true value2Greater than given value (such as 0.999) or repeatedly
Generation number overflows, and such as meets termination condition, then current solution is the optimal solution of (a, b, c, d, e, f, g, h), exports optimal solution；It is no
Then, Quantum rotating gate updates quantum coding, returns to 2..
In 5th step, error analysis parameter includes the fitting between test set remaining capacity true value and remaining capacity estimation value
Goodness coefficient of determination (RT 2), mean square error (RMSE) or/and average relative error (MRE).
Formula (9) is into (11): CTiIndicate the corresponding remaining capacity true value of i-th group of sample in test set；Indicate multicore
Estimated value of the Method Using Relevance Vector Machine to the corresponding remaining capacity of i-th group of sample in test set；Indicate all samples in test set
The average value of remaining capacity true value；NTFor the sample size of test set.
Compared with prior art, the method have the benefit that:
Optimized jointly the invention proposes a kind of based on tetra- kernel functions of RBF, Poly, Sigmoid and Chi-Square
Method Using Relevance Vector Machine estimates the model of the remaining capacity of lithium battery.By quantum genetic algorithm optimization kernel function coefficient and hyper parameter,
Parameters adjusting as reasonable as possible is obtained；Then error-tested is carried out to judge whether to meet condition to its result, it is ensured that can
Estimated accuracy.Remaining capacity is difficult to the problem of accurately estimating under very good solution of the present invention off-line state, is online lithium battery
Research in terms of product remaining capacity estimation, life estimation and abnormality detection provides preferable scheme.
Detailed description of the invention
Fig. 1 is flow chart of the invention.
Specific embodiment
The present invention is described in further details below with reference to Figure of description and specific embodiment.
Referring to Fig.1, the present invention provides a kind of offline lithium battery method for estimating remaining capacity, comprising the following steps:
The first step, the acquisition of lithium battery data.
The battery status monitoring parameters of each charge and discharge cycles when lithium battery works under off-line state are extracted in seconds,
And obtain calculate each charge and discharge cycles remaining capacity it is true as the corresponding remaining capacity of each group battery status monitoring parameters
Value, building form lithium battery data set by battery status monitoring parameters and corresponding remaining capacity true value.
Wherein battery status monitoring parameters include constant-current charge time, constant voltage charging time, constant-current discharge time, chargin level
Equal temperature, electric discharge mean temperature.xi(t), when i=1,2,3,4,5 indicates constant-current charge corresponding in single charge and discharge cycles
Between, constant voltage charging time, the constant-current discharge time, charging mean temperature and electric discharge mean temperature.
The remaining capacity that each charge and discharge cycles are calculated using constant-current discharge formula, that is, formula (1), as each group battery shape
The corresponding remaining capacity true value of state monitoring parameters.
C (t)=Icc(t)(tcc(t)-t0(t)) (1)
Wherein t0(t) and tcc(t) at the beginning of respectively indicating the constant-current charge in the t times charge and discharge cycles and when terminating
Between；Icc(t) the Constant Electric Current flow valuve in the t times charge and discharge cycles is indicated.
Constructed lithium battery data set Xi={ xi(t)~C (t) | t=1 ..., T；I=1,2,3,4,5 }, wherein xi(t),i
In=1,2,3,4,5 the t times charge and discharge cycles of expression when corresponding constant-current charge time, constant voltage charging time, constant-current discharge
Between, charging mean temperature and electric discharge mean temperature.T is charge and discharge cycles number.C (t) indicates battery status monitoring parameters xi(t)
Corresponding remaining capacity true value.
Second step pre-processes the data in lithium battery data set, and by the lithium battery data set after pretreatment
In data be divided into training set and test set according to a certain percentage.
Pretreatment, which refers to, first carries out missing values processing to the data outliers in lithium battery data set, after having handled
Lithium battery data set is normalized.
Too big wave can should not occur according to specific Long-term change trend in charge and discharge time and temperature when single charge and discharge cycles
It is dynamic.It is contemplated that finding out its variation tendency line, the data that will deviate from 90% confidence interval band are regarded as data outliers, and use trend
Line upper value replaces it.Missing values can also be replaced with Trendline upper value.Final detection parameters and remaining capacity dimension having the same
T × 1 is spent, T is charge and discharge cycles number.
Due to input value (battery status monitoring parameters xi(t)) physical meaning is different, and dimension is generally different.In order to go
Influence except dimension to RVM needs to carry out nondimensionalization processing to monitoring parameters sequence.So that its element is between 0 to 1.
After normalized, all battery status monitoring parameters dimensions are remained unchanged, but scope limitation is [0,1].It is surplus
Remaining charge value does not do normalized, but its value is under marked capacity.
Third step constructs isomery kernel function and solves its isomery core coefficient and hyper parameter.
(1) isomery kernel function is constructed:
Wherein, dmFor m-th of kernel function Km(xi, x) weight coefficient, andxiIt is battery for input
Status monitoring parameter xi(t), i=1,2,3,4,5, x be output i.e. residual electric quantity C (t).
Kernel function commonly used in the art includes but is not limited to following several:
Gaussian kernel function (RBF):
kRBF(xi, x) and=exp (- γ | | x-xi||2) (3)
Polynomial kernel function (Poly):
Sigmoid kernel function:
Chi-Square kernel function:
Kernel function well known in the art is numerous to list herein there are also very much.Different kernel functions has the different overall situations
Or local characteristics, multiple kernel function synthesis can be very good to learn from other's strong points to offset one's weaknesses.The present invention provides a kind of using aforementioned four seed nucleus letter
The isomery kernel function of the constructed multicore Method Using Relevance Vector Machine of number, indicates are as follows:
Wherein: a, b, c, d, e, f, g, h are isomery core coefficient and hyper parameter to be asked, xiIt is battery status prison for input
Survey parameter xi(t), i=1,2,3,4,5, x be the output i.e. residual electric quantity of estimation.
(2) the isomery core coefficient and hyper parameter of the isomery kernel function of multicore Method Using Relevance Vector Machine are solved using quantum genetic algorithm
(a,b,c,d,e,f,g,h)。
1. initializing
The parameter of initialization include Population Size, chromosome number (number of isomery core coefficient and hyper parameter i.e. to be asked),
Chromosome length, chromosome decimal system maximum value, chromosome decimal system minimum value, the number of iterations and quantum rotation angle；
2. introduce quantum coding method be chromosome value (i.e. isomery core coefficient and hyper parameter (a, b, c, d, e, f, g,
H) binary coding) is carried out, the decimal system is then turned to and participates in subsequent step；
1. calculating fitness R2, find current optimal solution；
Training set is extracted into a part of sample (such as " taking one every four ") as verifying collection, remaining sample is as new training
Collection.
(a, b, c, d, e, f, the g, h) obtained according to current search, finds out isomery core, and new training set is substituted into multicore phase
Vector machine training is closed, and is verified using verifying collection, finds out the estimated value of the remaining capacity of verifying collection, and count by formula (8)
Calculate the fitting between the corresponding remaining capacity estimation value of verifying concentration each group battery status monitoring parameters and remaining capacity true value
Goodness coefficient of determination R2:
Wherein, yiIndicate that the corresponding remaining capacity true value of i-th group of sample is concentrated in verifying；Indicate multicore Method Using Relevance Vector Machine
The estimated value that the corresponding remaining capacity of i-th group of sample is concentrated in verifying is collected to verifying；Indicate that the residue of all samples is concentrated in verifying
The average value of electricity true value, n are verifying collection sample size.
4. judging whether to meet termination condition, termination condition is that verifying concentrates each group battery status monitoring parameters corresponding surplus
Goodness of fit coefficient of determination R between remaining electricity estimation value and remaining capacity true value2Greater than given value (such as 0.999) or repeatedly
Generation number overflows, and such as meets termination condition, then current solution is the optimal solution of (a, b, c, d, e, f, g, h), exports optimal solution；It is no
Then, Quantum rotating gate updates quantum coding, returns to 2..
In quantum genetic algorithm, the operation object of intersection is the bit angle of all chromosome values in population.This
Outside, quantum variation is realized by Quantum rotating gate, essence be by change population in all chromosome values each
The quantum angle of quantum bit, so that chromosome value is drawn close to better chromosome.It therefore, there is no need to calculate as traditional heredity
Method specializes crossover probability like that, and mutation probability does not have to select specific variation rule from various mutation operators naturally.
Quantum genetic algorithm (QGA) introduces the value that quantum coding method is chromosome and is encoded.Chromosome each
Binary digit is referred to as quantum bit, any intermediate state of 0 or 1,0 and 1 not determining centre of quantum bit, so measure
Son can indicate that biggish solution space, superposition state form are as follows with seldom number of individuals:
In formula, | α |2With | β |2The probability of 0 and 1 appearance is represented, and is met | α |2+|β|2=1.When initial chromosome value,
Random selection one is located at the angle of θ ∈ (0,2 π), and enables α=sin (θ), β=cos (θ).After successive ignition, alternative one becomes
In 0, i.e. quantum collapsing reaches determining state, and quantum uncertainty disappears.By k iteration, dyed for the υ articles in chromosome population
Body can indicate are as follows:
Wherein: υ is the quantum digit that every chromosome includes；V is chromosome number, i.e. population number.
In QGA, quantum door for realizing each generation quantum chromosomes variation, and according to the probability amplitude of current chromosome into
Row rotation, makes model preferably be intended to optimal solution.Most common quantum rotation door operation is as follows:
In formula, U (Δ θ) is spin matrix；Δ θ is rotation angle, control algolithm convergence rate.
It is as follows that quantum coding switchs to binary-coded operation rules: random number is randomly generated in each quantum bit, if should
Random number is less than | α |2, then otherwise it is 0 that it is 1 that the bit, which is value,.Binary coding is converted into 10 systems and defers to common genetic algorithm
Rule.
4th step, training lithium battery remaining capacity estimation model.
Isomery core coefficient obtained in third step and hyper parameter are substituted into isomery kernel function, utilize each group electricity in training set
Pond status monitoring parameter is used as input, the corresponding remaining capacity true value conduct of each group battery status monitoring parameters in training set
Output is trained it, the lithium battery remaining capacity estimation model after being trained；
Each group battery status monitoring parameters in test set are input to the lithium battery remaining capacity estimation model after training
In, obtain the corresponding remaining capacity estimation value of each group battery status monitoring parameters in test set
5th step, error analysis.
Utilize the corresponding remaining capacity true value C of each group battery status monitoring parameters and remaining capacity estimation in test set
ValueCalculate error analysis parameter, the lithium battery remaining capacity estimation model accuracy that verifying training obtains.
Error analysis parameter includes the goodness of fit coefficient of determination between remaining capacity true value and remaining capacity estimation value
(RT 2), mean square error (RMSE) or/and average relative error (MRE).
Formula (9) is into (11): CTiIndicate the corresponding remaining capacity true value of i-th group of sample in test set；Indicate multicore
Estimated value of the Method Using Relevance Vector Machine to the corresponding remaining capacity of i-th group of sample in test set；Indicate all samples in test set
The average value of remaining capacity true value；NTFor the sample size of test set.
6th step, the battery status monitoring parameters of each charge and discharge cycles when acquisition lithium battery work are input to trained
To lithium battery remaining capacity estimation model in carry out remaining capacity estimation.
The foregoing is merely a preferred embodiment of the present invention, are not intended to restrict the invention, for this field
For technical staff, the invention may be variously modified and varied.All within the spirits and principles of the present invention, made any
Modification, equivalent replacement, improvement etc., should all be included in the protection scope of the present invention.
Claims (8)
1. a kind of offline lithium battery method for estimating remaining capacity, it is characterised in that: the following steps are included:
The first step, the acquisition of lithium battery data；
The battery status monitoring parameters of each charge and discharge cycles when lithium battery works under off-line state are extracted in seconds, and are obtained
Take the remaining capacity for calculating each charge and discharge cycles as the corresponding remaining capacity true value of each group battery status monitoring parameters, structure
It builds and forms lithium battery data set by battery status monitoring parameters and corresponding remaining capacity true value；
Second step pre-processes the data in lithium battery data set, and will be in the lithium battery data set after pretreatment
Data are divided into training set and test set according to a certain percentage；
Third step constructs isomery kernel function and solves its isomery core coefficient and hyper parameter；
4th step, training lithium battery remaining capacity estimation model；
Isomery core coefficient obtained in third step and hyper parameter are substituted into isomery kernel function, utilize each group battery shape in training set
State monitoring parameters are as inputting, and the corresponding remaining capacity true value of each group battery status monitoring parameters in training set is as output
It is trained, the lithium battery remaining capacity estimation model after being trained；
Each group battery status monitoring parameters in test set are input in the lithium battery remaining capacity estimation model after training, are obtained
The corresponding remaining capacity estimation value of each group battery status monitoring parameters into test set；
5th step is estimated using the corresponding remaining capacity true value of each group battery status monitoring parameters and remaining capacity in test set
Evaluation calculates error analysis parameter, the lithium battery remaining capacity estimation model accuracy that verifying training obtains；
6th step, the battery status monitoring parameters of each charge and discharge cycles when acquiring lithium battery to be measured work are input to trained
To lithium battery remaining capacity estimation model in can carry out the estimation of remaining capacity.
2. offline lithium battery method for estimating remaining capacity according to claim 1, it is characterised in that: in the first step, battery
Status monitoring parameter includes the constant-current charge time, constant voltage charging time, the constant-current discharge time, charging mean temperature, discharges averagely
Temperature.
3. offline lithium battery method for estimating remaining capacity according to claim 2, it is characterised in that: in the first step, use
Constant-current discharge formula, that is, formula (1) calculates the remaining capacity of each charge and discharge cycles, as each group battery status monitoring parameters pair
The remaining capacity true value answered:
C (t)=Icc(t)(tcc(t)-t0(t)) (1)
Wherein t0(t) and tcc(t) respectively indicate at the beginning of the constant-current charge in the t times charge and discharge cycles and terminate the time；
Icc(t) the Constant Electric Current flow valuve in the t times charge and discharge cycles is indicated；
Constructed lithium battery data set Xi={ xi(t)~C (t) | t=1 ..., T；I=1,2,3,4,5 }, wherein xi(t), i=1,
2,3,4,5 expression the t time charge and discharge cycles in corresponding to the constant-current charge time, constant voltage charging time, the constant-current discharge time, fill
Electric mean temperature and electric discharge mean temperature；T is charge and discharge cycles number；C (t) indicates battery status monitoring parameters xi(t) corresponding
Remaining capacity true value.
4. offline lithium battery method for estimating remaining capacity according to claim 2, it is characterised in that: pre- to locate in second step
Reason, which refers to, first carries out missing values processing to the data outliers in lithium battery data set, to the lithium battery data set after having handled
It is normalized.
5. offline lithium battery method for estimating remaining capacity according to claim 1, it is characterised in that: in third step, building
Isomery kernel function:
Wherein, dmFor m-th of kernel function Km(xi, x) weight coefficient, anddm≥0；xiIt is battery status prison for input
Survey parameter xi(t), i=1,2,3,4,5, x be the output i.e. residual electric quantity of estimation.
6. offline lithium battery method for estimating remaining capacity according to claim 5, it is characterised in that: in third step,
Wherein: a, b, c, d, e, f, g, h are isomery core coefficient and hyper parameter to be asked, xiIt is battery status monitoring ginseng for input
Number xi(t), i=1,2,3,4,5, x be the output i.e. residual electric quantity of estimation.
7. offline lithium battery method for estimating remaining capacity according to claim 6, it is characterised in that: calculated using quantum genetic
Method optimizes the isomery core coefficient and hyper parameter (a, b, c, d, e, f, g, h) of the isomery kernel function of multicore Method Using Relevance Vector Machine, and method is such as
Under:
1. initializing；
The parameter of initialization includes Population Size, chromosome number, chromosome length, chromosome decimal system maximum value, chromosome ten
System minimum value, the number of iterations and quantum rotation angle；Wherein chromosome number is of isomery core coefficient and hyper parameter to be asked
Number, chromosome number 8；
2. introduce the value that quantum coding method is chromosome be isomery core coefficient to be asked and hyper parameter (a, b, c, d, e, f, g,
H) binary coding is carried out, the decimal system is then turned to and participates in subsequent step；
3. calculating fitness R2, find current optimal solution；
Training set is extracted a part of sample as verifying to collect, remaining sample is as new training set；
(a, b, c, d, e, f, the g, h) obtained according to current search, finds out isomery core, by new training set substitute into multicore correlation to
The training of amount machine, and verified using verifying collection, the estimated value of the remaining capacity of verifying collection is found out, and calculate and test by formula (8)
Card concentrates the goodness of fit between the corresponding remaining capacity estimation value of each group battery status monitoring parameters and remaining capacity true value
Coefficient of determination R2:
Wherein, CiIndicate that the corresponding remaining capacity true value of i-th group of sample is concentrated in verifying；Indicate multicore Method Using Relevance Vector Machine to testing
Card concentrates the estimated value of the corresponding remaining capacity of i-th group of sample；Indicate that the remaining capacity true value of all samples is concentrated in verifying
Average value, n be verifying collection sample size；
4. judging whether to meet termination condition, termination condition is that the corresponding remaining electricity of each group battery status monitoring parameters is concentrated in verifying
Measure the goodness of fit coefficient of determination R between estimated value and remaining capacity true value2It is overflowed greater than given value or the number of iterations, such as
Meet termination condition, then current solution is the optimal solution of (a, b, c, d, e, f, g, h), exports optimal solution；Otherwise, Quantum rotating gate
Quantum coding is updated, is returned to 2..
8. offline lithium battery method for estimating remaining capacity according to claim 6, it is characterised in that: error analysis parameter packet
Include the goodness of fit coefficient of determination R between remaining capacity true value and remaining capacity estimation valueT 2, mean square error εRMSEOr/and it is average
Relative error εMRE:
Formula (9) is into (11): CTiIndicate the corresponding remaining capacity true value of i-th group of sample in test set；Indicate that multicore is related
Estimated value of the vector machine to the corresponding remaining capacity of i-th group of sample in test set；Indicate the residue of all samples in test set
The average value of electricity true value；NTFor the sample size of test set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910804963.XA CN110501646A (en) | 2019-08-29 | 2019-08-29 | Off-line lithium battery residual capacity estimation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910804963.XA CN110501646A (en) | 2019-08-29 | 2019-08-29 | Off-line lithium battery residual capacity estimation method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110501646A true CN110501646A (en) | 2019-11-26 |
Family
ID=68590203
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910804963.XA Pending CN110501646A (en) | 2019-08-29 | 2019-08-29 | Off-line lithium battery residual capacity estimation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110501646A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111025161A (en) * | 2020-01-10 | 2020-04-17 | 济南大学 | Lithium battery constant-current constant-voltage charging amount estimation method and terminal |
| CN111090047A (en) * | 2019-12-09 | 2020-05-01 | 泉州装备制造研究所 | Lithium battery health state estimation method based on multi-model fusion |
| CN111123127A (en) * | 2019-12-03 | 2020-05-08 | 浙江捷创智能技术有限公司 | Function fitting and electric quantity prediction method for lithium battery charging electric quantity |
| CN111832221A (en) * | 2020-06-23 | 2020-10-27 | 华中科技大学无锡研究院 | Lithium battery life prediction method based on feature screening |
| CN111856298A (en) * | 2020-07-23 | 2020-10-30 | 上海空间电源研究所 | On-orbit residual capacity prediction method for lithium ion storage battery for spacecraft |
| CN112485689A (en) * | 2020-10-26 | 2021-03-12 | 江苏慧智能源工程技术创新研究院有限公司 | Method for predicting residual cycle life of lithium battery in energy storage system based on Xgboost model |
| CN112834927A (en) * | 2021-01-06 | 2021-05-25 | 合肥工业大学 | Lithium battery residual life prediction method, system, device and medium |
| CN113093012A (en) * | 2021-03-23 | 2021-07-09 | 浙江吉利控股集团有限公司 | Battery energy state detection method, battery energy state detection equipment, storage medium and device |
| CN113267733A (en) * | 2021-04-13 | 2021-08-17 | 西安理工大学 | Automatic configuration method for lithium battery health state estimation based on Gaussian process regression |
| 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 |
| CN113567863A (en) * | 2021-06-11 | 2021-10-29 | 北京航空航天大学 | Abnormal degraded lithium battery capacity prediction method based on quantum assimilation and data filling |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107015165A (en) * | 2017-06-14 | 2017-08-04 | 安庆师范大学 | Lithium battery method for predicting residual useful life based on sparse coefficient multinuclear Method Using Relevance Vector Machine |
| CN107132490A (en) * | 2017-07-05 | 2017-09-05 | 福州大学 | A kind of method for realizing the estimation of lithium battery group state-of-charge |
| CN108805217A (en) * | 2018-06-20 | 2018-11-13 | 山东大学 | A kind of health state of lithium ion battery method of estimation and system based on support vector machines |
-
2019
- 2019-08-29 CN CN201910804963.XA patent/CN110501646A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107015165A (en) * | 2017-06-14 | 2017-08-04 | 安庆师范大学 | Lithium battery method for predicting residual useful life based on sparse coefficient multinuclear Method Using Relevance Vector Machine |
| CN107132490A (en) * | 2017-07-05 | 2017-09-05 | 福州大学 | A kind of method for realizing the estimation of lithium battery group state-of-charge |
| CN108805217A (en) * | 2018-06-20 | 2018-11-13 | 山东大学 | A kind of health state of lithium ion battery method of estimation and system based on support vector machines |
Non-Patent Citations (3)
| Title |
|---|
| TAO WANG等: "Transformer Health Management Based on Self-Powered RFID Sensor and Multiple Kernel RVM", 《IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT》 * |
| YANG ZHANG等: "Online Capacity Estimation of Lithium-Ion Batteries Based on Novel Feature Extraction and Adaptive Multi-Kernel Relevance Vector Machine", 《ENERGIES》 * |
| 马春辉等: "基于量子遗传算法与多输出混合核相关向量机的堆石坝材料参数自适应反演研究", 《岩土力学》 * |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111123127A (en) * | 2019-12-03 | 2020-05-08 | 浙江捷创智能技术有限公司 | Function fitting and electric quantity prediction method for lithium battery charging electric quantity |
| CN111090047A (en) * | 2019-12-09 | 2020-05-01 | 泉州装备制造研究所 | Lithium battery health state estimation method based on multi-model fusion |
| CN111090047B (en) * | 2019-12-09 | 2022-01-28 | 泉州装备制造研究所 | Lithium battery health state estimation method based on multi-model fusion |
| CN111025161A (en) * | 2020-01-10 | 2020-04-17 | 济南大学 | Lithium battery constant-current constant-voltage charging amount estimation method and terminal |
| CN111025161B (en) * | 2020-01-10 | 2022-02-22 | 济南大学 | Lithium battery constant-current constant-voltage charging amount estimation method and terminal |
| 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 |
| CN111832221A (en) * | 2020-06-23 | 2020-10-27 | 华中科技大学无锡研究院 | Lithium battery life prediction method based on feature screening |
| CN111832221B (en) * | 2020-06-23 | 2023-09-26 | 华中科技大学无锡研究院 | Lithium battery life prediction method based on feature screening |
| CN111856298A (en) * | 2020-07-23 | 2020-10-30 | 上海空间电源研究所 | On-orbit residual capacity prediction method for lithium ion storage battery for spacecraft |
| CN112485689A (en) * | 2020-10-26 | 2021-03-12 | 江苏慧智能源工程技术创新研究院有限公司 | Method for predicting residual cycle life of lithium battery in energy storage system based on Xgboost model |
| CN112485689B (en) * | 2020-10-26 | 2024-02-06 | 沃太能源股份有限公司 | Method for predicting residual cycle life of lithium battery in energy storage system based on Xgboost model |
| CN112834927A (en) * | 2021-01-06 | 2021-05-25 | 合肥工业大学 | Lithium battery residual life prediction method, system, device and medium |
| CN113093012A (en) * | 2021-03-23 | 2021-07-09 | 浙江吉利控股集团有限公司 | Battery energy state detection method, battery energy state detection equipment, storage medium and device |
| CN113267733A (en) * | 2021-04-13 | 2021-08-17 | 西安理工大学 | Automatic configuration method for lithium battery health state estimation based on Gaussian process regression |
| CN113267733B (en) * | 2021-04-13 | 2023-11-17 | 西安理工大学 | Automatic configuration method for lithium battery health state estimation based on Gaussian process regression |
| CN113567863A (en) * | 2021-06-11 | 2021-10-29 | 北京航空航天大学 | Abnormal degraded lithium battery capacity prediction method based on quantum assimilation and data filling |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110501646A (en) | Off-line lithium battery residual capacity estimation method | |
| CN110824364B (en) | Lithium battery SOH estimation and RUL prediction method based on AST-LSTM neural network | |
| Liu et al. | Towards long lifetime battery: AI-based manufacturing and management | |
| Fan et al. | A novel machine learning method based approach for Li-ion battery prognostic and health management | |
| CN106055775B (en) | A kind of service life of secondary cell prediction technique that particle filter is combined with mechanism model | |
| CN114325450A (en) | Lithium ion battery health state prediction method based on CNN-BilSTM-AT hybrid model | |
| Song et al. | Remaining life prediction of lithium-ion batteries based on health management: A review | |
| Ye et al. | State-of-health estimation for lithium-ion batteries using domain adversarial transfer learning | |
| CN110542819A (en) | transformer fault type diagnosis method based on semi-supervised DBNC | |
| CN113917334A (en) | Battery health state estimation method based on evolution LSTM self-encoder | |
| CN112834927A (en) | Lithium battery residual life prediction method, system, device and medium | |
| Xue et al. | Estimating state of health of lithium-ion batteries based on generalized regression neural network and quantum genetic algorithm | |
| Chen et al. | A deep learning feature fusion based health index construction method for prognostics using multiobjective optimization | |
| Ge et al. | An improved PF remaining useful life prediction method based on quantum genetics and LSTM | |
| Li et al. | A hybrid framework for predicting the remaining useful life of battery using Gaussian process regression | |
| CN115219906A (en) | Multi-model fusion battery state of charge prediction method and system based on GA-PSO optimization | |
| CN113203953B (en) | Lithium battery residual service life prediction method based on improved extreme learning machine | |
| Song et al. | Capacity estimation method of lithium-ion batteries based on deep convolution neural network | |
| CN116736133A (en) | Early prediction method for capacity degradation track of lithium ion battery in full life cycle | |
| CN116720743A (en) | Carbon emission measuring and calculating method based on data clustering and machine learning | |
| Liang et al. | Identification of slope stability based on the contrast of BP neural network and SVM | |
| Zhao et al. | Bayesian network based imprecise probability estimation method for wind power ramp events | |
| CN113269356B (en) | Missing data-oriented equipment residual life prediction method and system | |
| Zhu et al. | State of health estimation of lithium-ion battery using time convolution memory neural network | |
| Zhang et al. | Online state of health estimation for lithium-ion batteries based on gene expression programming |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191126 |
|
| RJ01 | Rejection of invention patent application after publication |