CN116401770A - Quick charge strategy design method based on battery digital twin model and machine learning - Google Patents
Quick charge strategy design method based on battery digital twin model and machine learning Download PDFInfo
- Publication number
- CN116401770A CN116401770A CN202310549532.XA CN202310549532A CN116401770A CN 116401770 A CN116401770 A CN 116401770A CN 202310549532 A CN202310549532 A CN 202310549532A CN 116401770 A CN116401770 A CN 116401770A
- Authority
- CN
- China
- Prior art keywords
- battery
- charging
- sei
- model
- lithium
- 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
- 238000000034 method Methods 0.000 title claims abstract description 63
- 238000010801 machine learning Methods 0.000 title claims abstract description 18
- 238000013461 design Methods 0.000 title claims abstract description 16
- 238000007600 charging Methods 0.000 claims abstract description 170
- 230000032683 aging Effects 0.000 claims abstract description 39
- 238000005457 optimization Methods 0.000 claims abstract description 39
- 238000000605 extraction Methods 0.000 claims abstract description 4
- 229910052744 lithium Inorganic materials 0.000 claims description 74
- WHXSMMKQMYFTQS-UHFFFAOYSA-N Lithium Chemical compound [Li] WHXSMMKQMYFTQS-UHFFFAOYSA-N 0.000 claims description 68
- 230000006870 function Effects 0.000 claims description 59
- 238000007747 plating Methods 0.000 claims description 24
- 239000002904 solvent Substances 0.000 claims description 22
- 239000002245 particle Substances 0.000 claims description 18
- 239000003792 electrolyte Substances 0.000 claims description 16
- 230000008569 process Effects 0.000 claims description 16
- 230000035882 stress Effects 0.000 claims description 16
- 238000011161 development Methods 0.000 claims description 15
- 230000018109 developmental process Effects 0.000 claims description 15
- 229910001416 lithium ion Inorganic materials 0.000 claims description 14
- HBBGRARXTFLTSG-UHFFFAOYSA-N Lithium ion Chemical compound [Li+] HBBGRARXTFLTSG-UHFFFAOYSA-N 0.000 claims description 13
- 238000009792 diffusion process Methods 0.000 claims description 13
- 239000011149 active material Substances 0.000 claims description 11
- 230000002829 reductive effect Effects 0.000 claims description 11
- 238000010277 constant-current charging Methods 0.000 claims description 10
- 238000005315 distribution function Methods 0.000 claims description 9
- 230000004907 flux Effects 0.000 claims description 9
- 238000006243 chemical reaction Methods 0.000 claims description 8
- 238000002474 experimental method Methods 0.000 claims description 8
- 238000005336 cracking Methods 0.000 claims description 7
- 230000001186 cumulative effect Effects 0.000 claims description 7
- 230000006872 improvement Effects 0.000 claims description 7
- 230000036961 partial effect Effects 0.000 claims description 7
- 238000004088 simulation Methods 0.000 claims description 7
- 239000013543 active substance Substances 0.000 claims description 6
- 238000012937 correction Methods 0.000 claims description 6
- 230000000694 effects Effects 0.000 claims description 6
- 238000009713 electroplating Methods 0.000 claims description 6
- 239000012798 spherical particle Substances 0.000 claims description 6
- 238000012360 testing method Methods 0.000 claims description 6
- 238000010280 constant potential charging Methods 0.000 claims description 4
- 239000008151 electrolyte solution Substances 0.000 claims description 4
- 230000036541 health Effects 0.000 claims description 4
- 230000007246 mechanism Effects 0.000 claims description 4
- 239000000203 mixture Substances 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 230000006399 behavior Effects 0.000 claims description 3
- 238000012512 characterization method Methods 0.000 claims description 3
- 230000000052 comparative effect Effects 0.000 claims description 3
- 150000001875 compounds Chemical class 0.000 claims description 3
- 238000010281 constant-current constant-voltage charging Methods 0.000 claims description 3
- 230000008878 coupling Effects 0.000 claims description 3
- 238000010168 coupling process Methods 0.000 claims description 3
- 238000005859 coupling reaction Methods 0.000 claims description 3
- 238000000354 decomposition reaction Methods 0.000 claims description 3
- 230000003247 decreasing effect Effects 0.000 claims description 3
- 230000008021 deposition Effects 0.000 claims description 3
- 238000001514 detection method Methods 0.000 claims description 3
- 238000006073 displacement reaction Methods 0.000 claims description 3
- 230000014509 gene expression Effects 0.000 claims description 3
- 230000002706 hydrostatic effect Effects 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 229910052751 metal Inorganic materials 0.000 claims description 3
- 239000002184 metal Substances 0.000 claims description 3
- 239000003960 organic solvent Substances 0.000 claims description 3
- 230000010287 polarization Effects 0.000 claims description 3
- 238000001556 precipitation Methods 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- 230000002441 reversible effect Effects 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims description 3
- 238000012546 transfer Methods 0.000 claims description 3
- 239000013598 vector Substances 0.000 claims description 3
- 238000000157 electrochemical-induced impedance spectroscopy Methods 0.000 claims 2
- 238000004422 calculation algorithm Methods 0.000 description 6
- 230000008901 benefit Effects 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 5
- 241000709691 Enterovirus E Species 0.000 description 3
- 239000000243 solution Substances 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 239000000306 component Substances 0.000 description 2
- 238000001453 impedance spectrum Methods 0.000 description 2
- 230000000670 limiting effect Effects 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 208000019901 Anxiety disease Diseases 0.000 description 1
- 230000036506 anxiety Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 239000008358 core component Substances 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 230000002939 deleterious effect Effects 0.000 description 1
- 239000007772 electrode material Substances 0.000 description 1
- 238000009472 formulation Methods 0.000 description 1
- 239000008187 granular material Substances 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000010278 pulse charging Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000007086 side reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/15—Vehicle, aircraft or watercraft design
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N7/00—Computing arrangements based on specific mathematical models
- G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/08—Probabilistic or stochastic CAD
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/04—Ageing analysis or optimisation against ageing
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/60—Other road transportation technologies with climate change mitigation effect
- Y02T10/70—Energy storage systems for electromobility, e.g. batteries
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Geometry (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Computer Hardware Design (AREA)
- Data Mining & Analysis (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Physics (AREA)
- Computational Mathematics (AREA)
- Mathematical Optimization (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Mathematical Analysis (AREA)
- Computing Systems (AREA)
- Automation & Control Theory (AREA)
- Algebra (AREA)
- Probability & Statistics with Applications (AREA)
- Aviation & Aerospace Engineering (AREA)
- Secondary Cells (AREA)
Abstract
The invention relates to a battery digital twin model and machine learning-based fast charge strategy design method, which belongs to the field of battery fast charge optimization and comprises the following steps: when the electric automobile is connected into the charging pile for charging, detecting battery state data of the automobile; inputting the battery state data into a Bayesian optimizer based on a battery digital twin model, and predicting the aging condition of the battery in the current charging under an initial protocol according to the current battery state; transmitting the current charging protocol and the battery aging corresponding to the current charging protocol to a Bayesian optimizer, and recommending a quick charging protocol for next round of extraction; updating the next round of fast charge protocol to the digital twin model of the battery, and predicting the battery state again to obtain the aging condition of the updated battery in the current charge; and adding the updated charging protocol and the battery aging corresponding to the updated charging protocol into the existing data set, transmitting the updated charging protocol to the Bayesian optimizer again for quick charging protocol recommendation, and repeating the steps to obtain the finally recommended charging protocol.
Description
Technical Field
The invention belongs to the field of battery fast charge optimization, and relates to a fast charge strategy design method based on a battery digital twin model and machine learning.
Background
With the increasing trend of traffic electrification, automobile manufacturers have made significant advances in the field of electric vehicles for pure electric vehicles (BEVs). The number of BEV models available on the market increases substantially, as well as the performance. At the same time, some challenges have hindered widespread use of electric vehicles, such as long charging times and mileage anxiety problems.
Lithium ion batteries are used as BEVs core components, and have the advantages of high energy density and long service life, occupy the main market of vehicle-mounted batteries, and are paid attention to. The demand of electric vehicles for lithium ion (Li-ion) batteries is increasing, and the demand for new optimal charging methods is accelerated, so as to improve the speed and reliability of the charging process without reducing the performance of the batteries. In addition to developing new materials and new batteries, optimizing charging strategies is also critical to solving the charging time problem. The method of increasing the charging current only to achieve reduced charging time causes temperature rise and deleterious side effects, and the tradeoff between fast charging and battery health should be considered simultaneously.
Generally, existing lithium ion battery charging strategies can be divided into three categories according to internal mathematical models:
the first is a model-free method, including Constant Current (CC), constant Current and Constant Voltage (CCCV), multi-stage constant current and constant voltage, and pulse charging techniques. These methods preset a charging curve without considering the dynamic response of the battery.
The second type of charging strategy utilizes empirical models, such as equivalent circuit based models and neural network models. These models predict battery status and calculate electrical components using past experimental data. The battery state is estimated by different circuit models using a kalman filter, a recursive least square method, a sliding mode observer, and a moving field of view estimation. Meanwhile, methods such as frequency optimization, multi-objective optimization, fuzzy control, linear secondary control and model predictive control are also used for optimizing the charging strategy. However, the empirical model cannot reflect internal changes of the battery, such as internal parameters of the battery and aging states of the battery.
The third class is complex electrochemical model-based charging algorithms. Formulation of closed loop optimization problems can minimize charging time and compensate for model uncertainty and disturbances. The electrochemical model may be coupled to a temperature model, an aging model, to reflect more internal cell characteristics, and the model is more closely related to a real cell. However, the computational complexity of the full-order nonlinear Partial Differential Equations (PDEs) that are difficult to handle limits their further application on real-time charge controllers. More electrochemical models of degradation, such as Single Particle Models (SPM), are used in practice.
The charging curve of the model-free charging algorithm is preset and cannot adapt to the dynamic change of the battery; the charging algorithm based on the empirical model cannot capture the internal state of the battery, such as temperature, lithium plating, SEI film, and change of active materials; the electrochemical model-based charging algorithm has complex model and excessive calculated amount, and cannot be used for real-time charging control. The data-driven optimization algorithm requires a large amount of data for training and can only be applied to the same type of battery as the battery used for training data, and has poor universality. The invention combines the battery twin model with the data driving, adopts the enhanced single particle model as the battery digital twin model, reduces the calculation cost of the model on the premise of ensuring certain precision, and explores the charging protocol space by using a Bayesian optimization method to obtain the optimal charging protocol.
Disclosure of Invention
In view of the above, the present invention aims to provide a fast charge strategy design method based on a battery digital twin model and machine learning, which solves the problem that the existing charge optimization algorithm cannot fully consider the initial state of the battery during charging, and reduces the calculation amount and calculation cost of an electrochemical model by a data driving method. The invention establishes a digital twin model of the battery, fully considers the initial state of the battery, and dynamically determines the optimal charging protocol of the battery in the current state according to different initial states of the battery. And the process of searching the most available protocol is accelerated by using a machine learning method, so that the calculation cost and time are reduced to realize online optimization.
In order to achieve the above purpose, the present invention provides the following technical solutions:
a fast charge strategy design method based on a battery digital twin model and machine learning comprises the following steps:
s1: when an electric automobile is connected into a charging pile for charging, battery state data of the automobile are detected, wherein the battery state data comprise battery temperature, state of charge (SOC), state of health (SOH), ampere-hour throughput, current lithium inventory loss, electrochemical Impedance Spectrum (EIS) detection and the like;
s2: inputting the battery state data into a Bayesian optimizer based on a battery digital twin model, and predicting aging of the battery in the current charging under an initial protocol according to the current battery state through the battery digital twin model; the initial protocol can be any number of protocols, and the corresponding data is used for initializing a Bayesian optimization model;
s3: transmitting an initial charging protocol and battery aging of a battery single cycle under the initial charging protocol as a data set to a Bayesian optimizer, wherein the initial charging protocol is used as input of a Bayesian optimization model, and the battery aging condition of the battery single cycle under the initial charging protocol is used as output; predicting battery aging corresponding to all protocols in a charging protocol space by utilizing Gaussian process regression GPR, and recommending a charging protocol for next round of extraction by utilizing an acquisition function;
s4: based on the battery state in the step S2, the digital twin model of the battery carries out simulation prediction on the aging of the battery under the charging protocol extracted in the next round, so as to obtain the aging of the battery under the charging protocol extracted in the next round;
s5: and adding the new charging protocol and the corresponding battery aging into the existing data set, transmitting the new charging protocol to the Bayesian optimizer again to recommend the next round of extracted quick charging protocol, and repeating the steps S4-S5 until the final recommended charging protocol is obtained, and charging the vehicle battery by the charging protocol.
Further, the initial protocol and the protocol recommended by the bayesian optimizer are both protocols in a charging protocol space; the charging protocol in the charging protocol space aims to charge the battery state of charge (SOC) from 0% to 80% through constant current charging of a plurality of stages within 10 minutes, and in the process, the battery is charged by stages by using j kinds of currents with different magnitudes;
the charging protocol is a multi-stage constant-current constant-voltage charging MSCC, which consists of a plurality of constant-current charging stages and a constant-voltage charging stage which follows the constant-current charging stages;
when j=3, the charging protocol charges the battery state of charge SOC from 0% to 80% in 3 steps within 10 minutes, and parameters of the charging protocol have CC1', CC2', which are also two-dimensional inputs of the black box optimization model; the method comprises the following steps: the charging rate of the 0-26.67% SOC interval is CC1', the charging rate of the 26.67% -53.34% SOC interval is CC2', the charging rate of the 53.34% -80% SOC interval is CC3', the size of the CC3' is determined by CC1', CC2', and the charging rate is expressed as:
when j=4, the charging protocol charges the battery state of charge SOC from 0% to 80% in 10 minutes in 4 steps, and parameters of the charging protocol are CC1, CC2, and CC3, which are also three-dimensional inputs of the black box optimization model; the method comprises the following steps: the charging rate of the 0-20% SOC interval is CC1, the charging rate of the 21-40% SOC interval is CC2, the charging rate of the 41-60% SOC interval is CC3, and the charging rate of the 61-80% SOC interval is CC4, wherein the size of the CC4 is determined by CC1, CC2 and CC3, and the formula is as follows:
when j=5, the charging protocol charges the battery state of charge SOC from 0% to 80% in 5 steps within 10 minutes, with parameters of the charging protocol being CC1 ", CC 2", CC3 ", CC 4", which is also a four-dimensional input of the black box optimization model; the method comprises the following steps: the charging rate of the 0-16% SOC interval is CC1 ", the charging rate of the 16-32% SOC interval is CC 2", the charging rate of the 32-48% SOC interval is CC3 ", the charging rate of the 48-64% SOC interval is CC 4", the charging rate of the 64-80% SOC interval is CC5 ", wherein the size of CC 5" is CC1 ", CC 2", CC3 ", and the formula is as follows:
further, the battery digital twin model is a battery electrochemical model that considers battery aging mechanisms including SEI film growth, lithium plating, particle cracking, and active material loss, integrated into a DFN model (Doyle-Fuller-Newman) to characterize the aging behavior of the battery.
Further, the SEI film growth specifically includes:
the growth of the SEI film is divided into two processes: firstly, polarization of a battery cathode, reduction decomposition of an organic solvent in electrolyte to generate a new compound, and secondly, precipitation of a new chemical substance on the surface of the anode to form a new SEI film; the growth of the SEI film is captured using a diffusion limited model, whose reaction rate is limited by the rate of transport of solvent through the SEI outer layer, assuming that the flux of solvent molecules follows the Phake law:
the boundary conditions are:
wherein c sol Is the concentration of the solvent, c sol,e Is the concentration of the solvent in the electrolyte, D sol (T) is the solvent diffusion coefficient, L is the position of the SEI film, L SEI Is SEI film thickness;
the flux density at the SEI film interface is:
rate of increase of SEI film thickness:
wherein L is SEI Is the SEI film thickness,is the average partial molar volume of the SEI; the growth of the SEI film is affected by temperature and solvent diffusion coefficient, which follows the Arrhenius equation:
wherein T is ref =25℃,E sol Is the activation energy of solvent diffusion;
the growth of the SEI film will result in the following effects:
wherein ε is n Is the porosity of the cathode, a n Is the specific surface area, a for spherical particles n =3ε n /R n ,R n Is the radius eta of the spherical particles of the negative electrode SEI Is the overpotential of SEI film, j tot Is the total current density in conservation of negative charge, σ SEI Is the conductivity of the SEI film.
Further, the lithium plating specifically includes:
when lithium ions are not intercalated into an electrode but lithium metal is formed on the surface of the electrode, lithium plating occurs, which is partially or completely recovered in a subsequent discharge reaction by lithium stripping; lithium electroplating also chemically reacts with the electrolyte solution to form an SEI film, and the metal lithium separated by the SEI film cannot be peeled off, and the part of lithium is called dead lithium; thus, a partially reversible lithium plating model was constructed, according to the butler-furat equation, with the lithium plating/stripping flux at the anode/electrolyte interface in the overpotential correction form being:
wherein k is Li Is the lithium plating/stripping constant, c Li Is the concentration of lithium plating, F is Faraday constant, c e Is Li in electrolyte + Concentration, transfer coefficient alpha a,Li And alpha c,Li Are all set to 0.5; η (eta) Li Is the lithium deposition potential LDP, defined as eta Li =φ s -φ e -η SEI Wherein phi is s And phi e Respectively relative to Li/Li + Electrode potential and electrolyte potential of (a);
by coupling three models of lithium electroplating, lithium stripping and SEI growth, the electroplated lithium will decay over time into SEI film and dead lithium, c Li Is expressed as:
wherein a is - Is the ratio of the surface area to the volume of the cathode, c dl Is the concentration of dead lithium, defined as:
it is assumed that the dead lithium decay rate γ is inversely proportional to the thickness of the SEI film:
wherein gamma is 0 Is a fitting parameter, L SEI,0 Is the initial SEI thickness, L SEI,t Is the SEI thickness at time t.
Further, the particle cracking specifically includes:
radial stress sigma using fatigue crack model based on spherical electrode particles r (r) tangential stress sigma t The expressions of (r) and displacement u are as follows:
where Ω is the partial molar volume, E is the Young's modulus, v is the Poisson's ratio, c avg (R) is a function of average lithium ion concentration and radius, R is particle radius,is the concentration c of lithium ion and the reference concentration c under no stress ref Deviation of (2);
the fatigue crack propagation model follows Paris's law, assuming that all electrode particle surfaces have the same crack length l cr Crack width w cr And crack density ρ cr :
Wherein t is 0 Time of single period, b cr Is a stress intensity correction factor, k cr And m cr Is a constant obtained from the experiment;
the growth model of the SEI film on the new crack is defined as follows:
wherein L is SEI,cr Is the average thickness of SEI film on new crack, L SEI,cr0 =L SEI,0 And/10000 is the thickness of the initial SEI film on the new crack.
Further, the active material loss specifically includes:
loss of active substance epsilon a Simulation by decreasing the volume fraction of active material:
wherein beta, m 2 Sum sigma c Is the coefficient obtained from the experiment, the hydrostatic stress sigma h =(σ r +2σ t )/3,σ h The subscripts max and min of (c) represent the maximum and minimum values, respectively.
Further, the Bayesian optimizer consists of a Gaussian process regression model GPR and an acquisition function, and the optimization target is that the aging of the battery is minimum, wherein the aging of the battery comprises capacity loss, lithium inventory loss, coulombic efficiency, active substance loss and the like;
the GPR model is used for outputting logarithmic variance characteristics of all three-dimensional charging protocols; the acquisition function is used to decide which protocols to test in the next round;
when GPR is applied, the actual output y after taking noise into account is expressed as:
y i =f(x i )+ε i
wherein (x) i ,y i ) Represents the i-th data point, f represents the feature map, f (x i ) Representing the output of the gaussian model; epsilon i Represents observation noise, and
optionally haveSet x= { X of limited input composition 1 ,x 2 ,…,x n } d*n D x n represents n samples, each sample having d features as input, { x 1 ,x 2 ,…,x n } d*n Obeys a joint gaussian probability distribution p (f (x) 1 ),f(x 2 ),…,f(x n ) The mean function and covariance function of f (x) are:
m(x)=E[f(x)]
k(x,x′)=E[(f(x)-m(x))(f(x′)-m(x′))]
since there is no a priori information of the distribution of f (x), m (x) is set to 0, and the distribution of y is written as:
wherein K is ff Is an n-by-n dimensional matrix, and [ K ] ff ] i,j =k(x i ,x j );
For a new inputd x n represents that the new data set has m samples, each sample having d features input thereto; the joint distribution of this and the known observed target values is written as:
Conditional distribution p (f) * |X,y,X * ) Obeys normal distribution:
wherein,,represents the mean of posterior distribution, cov (f * ) The variance of the posterior distribution is expressed as follows:
kernel functions that can be used in the GPR model include the following:
the first kernel function is a gaussian kernel RBF, given by:
wherein II x-x' II 2 Representing the square of the Euclidean distance of the vectors x and x', and sigma represents the standard deviation;
the second kernel function is a Matern kernel, given by:
it has positive parameters v and l, where f (v) is a f function, k v Is a second type of bessel function, d (x, x') represents the euclidean distance.
Further, the acquisition function is a confidence interval upper bound function UCB, an improvement probability function PI, or an expected delta function EI, wherein:
the UCB acquisition function is as follows:
UCB(x)=μ(x)+βσ(x)
wherein μ (x) and σ (x) are the mean and standard deviation of the posterior distribution, respectively, β is a hyper-parameter to be optimized for balanced exploration and development, said exploration being to explore untested parameter spaces with high uncertainty, said development being to test promising areas according to completed experimental results; as the number of steps increases, the focus is changed from exploration to development, and beta is gradually reduced, so that the following steps are obtained:
UCB(x)=μ(x)+β 0 ε k σ(x)
wherein beta is 0 ,ε k Are super parameters, k represents the optimized times, epsilon is (0, 1) in the value range, epsilon k Characterization reduces the weight of σ (x) to turn emphasis towards development;
the PI acquisition function is as follows:
the PI utilizes posterior model to estimate that the function value is larger than the maximum value mu in the actual observed value at present + To find the next round of sampling points, PI (x) represents the probability, and the formula is as follows:
wherein phi represents a normal cumulative distribution function, mu (x) and sigma (x) respectively represent a Gaussian average value and a standard deviation of posterior prediction, and xi is more than or equal to 0 and is an adjustable parameter, wherein the aim is to find the sum mu + The point at which the probability of a comparative boost ζ is greatest;
the EI acquisition function is as follows:
the weight Φ (Z) is a normal cumulative distribution function, and ζ represents the sum μ + The degree of improvement in the phase ratio, phi (Z), represents the normal probability distribution function, and the objective of EI is to find the point (μ (x) - μ) with the greatest improvement expectations + ζ) Φ (Z) and σ (x) Φ (Z) are used for development and exploration, respectively.
The invention has the beneficial effects that: model-based bayesian optimization can greatly save time: the states of the battery are different when the vehicle is charged each time, the combination of the states such as temperature, SOC, SOH and the like is changed, the charging protocol space of the battery is large, each protocol in the current battery state is simulated by using a battery model, and the calculation amount is huge. The diversity of battery states during charging also makes it difficult to determine battery protocols using look-up tables. The battery digital twin model is combined with Bayesian optimization, so that the running time and cost of the model can be greatly reduced, and quick optimization is realized.
Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention. The objects and other advantages of the invention may be realized and obtained by means of the instrumentalities and combinations particularly pointed out in the specification.
Drawings
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in the following preferred detail with reference to the accompanying drawings, in which:
fig. 1 is a flowchart of a method for designing a fast charge strategy based on a battery digital twin model and machine learning according to the present invention.
FIG. 2 is a graph comparing the effect of a method according to the invention with a charging method without optimized constant current;
FIG. 3 is a graph showing the results of Bayesian parallel optimization, wherein (a) - (d) are graphs showing the results of optimization for the 1 st to 4 th charges, respectively.
Detailed Description
Other advantages and effects of the present invention will become apparent to those skilled in the art from the following disclosure, which describes the embodiments of the present invention with reference to specific examples. The invention may be practiced or carried out in other embodiments that depart from the specific details, and the details of the present description may be modified or varied from the spirit and scope of the present invention. It should be noted that the illustrations provided in the following embodiments merely illustrate the basic idea of the present invention by way of illustration, and the following embodiments and features in the embodiments may be combined with each other without conflict.
Wherein the drawings are for illustrative purposes only and are shown in schematic, non-physical, and not intended to limit the invention; for the purpose of better illustrating embodiments of the invention, certain elements of the drawings may be omitted, enlarged or reduced and do not represent the size of the actual product; it will be appreciated by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.
The same or similar reference numbers in the drawings of embodiments of the invention correspond to the same or similar components; in the description of the present invention, it should be understood that, if there are terms such as "upper", "lower", "left", "right", "front", "rear", etc., that indicate an azimuth or a positional relationship based on the azimuth or the positional relationship shown in the drawings, it is only for convenience of describing the present invention and simplifying the description, but not for indicating or suggesting that the referred device or element must have a specific azimuth, be constructed and operated in a specific azimuth, so that the terms describing the positional relationship in the drawings are merely for exemplary illustration and should not be construed as limiting the present invention, and that the specific meaning of the above terms may be understood by those of ordinary skill in the art according to the specific circumstances.
Referring to fig. 1, a fast charge strategy design method based on a battery digital twin model and machine learning includes the following steps:
s1: when an electric automobile is connected into a charging pile for charging, battery state data of the automobile are detected, including battery temperature, state of charge (SOC), state of health (SOH), ampere-hour throughput, current lithium inventory loss, electrochemical Impedance Spectrum (EIS) detection and the like;
s2: inputting the battery state data into a Bayesian optimizer based on a battery digital twin model, and predicting the aging condition of the battery in the current charging under an initial protocol according to the current battery state through the battery digital twin model; in this embodiment, the initial protocol is to randomly select 10 charging protocols from the charging protocol space; the charging protocol in the charging protocol space aims to charge the state of charge (SOC) of the battery from 0% to 80% by constant current charging of a plurality of stages within 10 minutes, and in the process, the battery is charged by stages by using j kinds of currents with different magnitudes;
the charging protocol is a multi-stage constant-current constant-voltage charging MSCC, which consists of a plurality of constant-current charging stages and a constant-voltage charging stage which follows the constant-current charging stages;
when j=3, the charging protocol charges the battery state of charge SOC from 0% to 80% in 3 steps within 10 minutes, and parameters of the charging protocol have CC1', CC2', which are also two-dimensional inputs of the black box optimization model; the method comprises the following steps: the charging rate of the 0-26.67% SOC interval is CC1', the charging rate of the 26.67% -53.34% SOC interval is CC2', the charging rate of the 53.34% -80% SOC interval is CC3', the size of the CC3' is determined by CC1', CC2', and the charging rate is expressed as:
when j=4, the charging protocol charges the battery state of charge SOC from 0% to 80% in 10 minutes in 4 steps, and parameters of the charging protocol are CC1, CC2, and CC3, which are also three-dimensional inputs of the black box optimization model; the method comprises the following steps: the charging rate of the 0-20% SOC interval is CC1, the charging rate of the 21-40% SOC interval is CC2, the charging rate of the 41-60% SOC interval is CC3, and the charging rate of the 61-80% SOC interval is CC4, wherein the size of the CC4 is determined by CC1, CC2 and CC3, and the formula is as follows:
when j=5, the charging protocol charges the battery state of charge SOC from 0% to 80% in 5 steps within 10 minutes, with parameters of the charging protocol being CC1 ", CC 2", CC3 ", CC 4", which is also a four-dimensional input of the black box optimization model; the method comprises the following steps: the charging rate of the 0-16% SOC interval is CC1 ", the charging rate of the 16-32% SOC interval is CC 2", the charging rate of the 32-48% SOC interval is CC3 ", the charging rate of the 48-64% SOC interval is CC 4", the charging rate of the 64-80% SOC interval is CC5 ", wherein the size of CC 5" is CC1 ", CC 2", CC3 ", and the formula is as follows:
the battery digital twin model is a battery electrochemical model that considers battery aging mechanisms including SEI film growth, lithium plating, particle cracking, and active material loss, integrated into a DFN model (Doyle-Fuller-Newman) to characterize the aging behavior of the battery.
SEI film growth, specifically including:
the growth of the SEI film is divided into two processes: firstly, polarization of a battery cathode, reduction decomposition of an organic solvent in electrolyte to generate a new compound, and secondly, precipitation of a new chemical substance on the surface of the anode to form a new SEI film; the growth of the SEI film is captured using a diffusion limited model, whose reaction rate is limited by the rate of transport of solvent through the SEI outer layer, assuming that the flux of solvent molecules follows the Phake law:
the boundary conditions are:
wherein c sol Is the concentration of the solvent, c sol,e Is the concentration of the solvent in the electrolyte, D sol (T) is the solvent diffusion coefficient, L is the position of the SEI film, L SEI Is SEI film thickness;
the flux density at the SEI film interface is:
rate of increase of SEI film thickness:
wherein L is SEI Is the SEI film thickness,is the average partial molar volume of the SEI; the growth of the SEI film is affected by temperature and solvent diffusion coefficient, which follows the Arrhenius equation:
wherein T is ref =25℃,E sol Is the activation energy of solvent diffusion;
the growth of the SEI film will result in the following effects:
wherein ε is n Is the porosity of the cathode, a n Is the specific surface area, a for spherical particles n =3ε n /R n ,R n Is the radius eta of the spherical particles of the negative electrode SEI Is the overpotential of SEI film, j tot Is the total current density in conservation of negative charge, σ SEI Is the conductivity of the SEI film.
The lithium plating specifically includes:
when lithium ions are not inserted into an electrode but lithium metal is formed on the surface of the electrode, lithium plating occurs, which is partially or completely recovered in a subsequent discharge reaction by stripping, the lithium plating chemically reacts with an electrolyte solution to form an SEI film, and the metal lithium separated by the SEI film cannot be stripped, which is called "dead lithium"; thus, a partially reversible lithium plating model was constructed, according to the butler-furat equation, with the lithium plating/stripping flux at the anode/electrolyte interface in the overpotential correction form being:
wherein k is Li Is the lithium plating/stripping constant, c Li Is the concentration of lithium plating, F is Faraday constant, c e Is Li in electrolyte + Concentration, transfer coefficient alpha a,Li And alpha c,Li Are all set to 0.5; η (eta) Li Is the lithium deposition potential LDP, defined as eta Li =φ s -φ e -η SEI Wherein phi is s And φ e is respectively relative to Li/Li + Electrode potential and electrolyte potential of (a); the electrode potential in the anode is close to 0, and the internal resistance of the SEI film is small at 25 ℃. Thus, LDP is primarily affected by electrolyte potential.
By coupling three models of lithium electroplating, lithium stripping and SEI growth, the electroplated lithium will decay over time into SEI film and dead lithium, c Li Is expressed as:
wherein a is - Is the ratio of the surface area to the volume of the cathode, c dl Is the concentration of dead lithium, defined as:
since dead lithium is formed by the reaction of lithium electroplating with an electrolyte solution, the reaction rate is affected by the solvent diffusion rate, as is SEI film formation. It is assumed that the dead lithium decay rate γ is inversely proportional to the thickness of the SEI film:
wherein gamma is 0 Is a fitting parameter, L SEI,0 Is the initial SEI thickness, L SEI,t Is the SEI thickness at time t.
The granule cracking specifically includes:
the electrode material expands and contracts greatly as lithium ions intercalate and deintercalate on the electrode. Alternating deformation of the electrode volume can lead to alternating stresses, resulting in fracture propagation on the active particles and the creation of new surfaces, which in turn affect side reactions such as SEI and lithium plating. Radial stress sigma using fatigue crack model based on spherical electrode particles r (r) tangential stress sigma t The expressions of (r) and displacement u are as follows:
wherein Ω is the partial molar volume, E is the Young's modulus, v is the Poisson's ratio, c avg (R) is a function of average lithium ion concentration and radius, R is particle radius,is the concentration c of lithium ion and the reference concentration c under no stress ref Deviation of (2);
the fatigue crack propagation model follows Paris's law, assuming that all electrode particle surfaces have the same crack length l cr Crack width w cr And crack density ρ cr :
Wherein t is 0 Time of single period, b cr Is a stress intensity correction factor, k cr And m cr Is a constant obtained from the experiment;
the growth model of the SEI film on the new crack is defined as follows:
wherein L is SEI,cr Is the average thickness of SEI film on new crack, L SEI,cr0 =L SEI,0 And/10000 is the thickness of the initial SEI film on the new crack.
Particle cracking also results in Loss of Active Material (LAM). The basic principle of the physical phenomenon is the same, so the above-described mechanism-based model can also be used to calculate the active substance loss. Loss of active substance epsilon a Simulation by decreasing the volume fraction of active material:
wherein beta, m 2 Sum sigma c Is the coefficient obtained from the experiment, the hydrostatic stress sigma h =(σ r +2σ t )/3,σ h The subscripts max and min of (c) represent the maximum and minimum values, respectively.
S3: transmitting an initial charging protocol and battery aging of a battery single cycle under the initial charging protocol as a data set to a Bayesian optimizer, wherein the initial charging protocol is used as input of a Bayesian optimization model, and the battery aging condition of the battery single cycle under the initial charging protocol is used as output; predicting battery aging corresponding to all protocols in a charging protocol space by utilizing Gaussian process regression GPR, and recommending a charging protocol for next round of extraction by utilizing an acquisition function; the Bayesian optimizer consists of a Gaussian process regression model GPR and an acquisition function; the optimization targets are that the battery has minimum aging, including but not limited to capacity loss, lithium inventory loss, coulombic efficiency, active material loss, etc.; the GPR model is used for outputting the logarithmic variance characteristics of all three-dimensional charging protocols; the acquisition function is used to decide which protocols to test in the next round;
when GPR is applied, the actual output y after taking noise into account is expressed as:
y i =f(x i )+ε i
wherein (x) i ,y i ) Represents the i-th data point, f represents the feature map, f (x i ) Representing the output of the gaussian model; epsilon i Represents observation noise, and
set x= { X of arbitrary finite input composition 1 ,x 2 ,…,x n } d*n D x n represents n samples, each sample having d features as input, { x 1 ,x 2 ,…,x n } d*n Are all subject to a joint gaussian probability distribution p (f (x) 1 ),f(x 2 ),…,f(x n ) The mean function and covariance function of f (x) are:
m(x)=E[f(x)]
k(x,x′)=E[(f(x)-m(x))(f(x′)-m(x′))]
since there is no a priori information of the distribution of f (x), m (x) is set to 0, and the distribution of y is written as:
wherein K is ff Is an n-by-n dimensional matrix, and [ K ] ff ] i,j =k(x i ,x j );
For a new inputd x n represents that the new data set has m samples, each sample having d features input thereto; the joint distribution of this and the known observed target values is written as:
Conditional distribution p (f) * |X,y,X * ) Obeys normal distribution:
wherein,,represents the mean of posterior distribution, cov (f * ) The variance of the posterior distribution is expressed as follows:
kernel functions that can be used in the GPR model include the following:
the first kernel function is a gaussian kernel RBF, given by:
wherein II x-x' II 2 Representing the square of the Euclidean distance of the vectors x and x', and sigma represents the standard deviation;
the second kernel function is a Matern kernel, given by:
it has positive parameters v and l, where f (v) is a f function, K v Is a second type of bessel function, d (x, x') represents the euclidean distance.
The acquisition function is a confidence interval upper bound function UCB, an improved probability function PI or an expected delta function EI, wherein:
the UCB acquisition function is as follows:
UCB(x)=μ(x)+βσ(x)
wherein μ (x) and σ (x) are the mean and standard deviation of the posterior distribution, respectively, β is a hyper-parameter to be optimized for balanced exploration and development, said exploration being to explore untested parameter spaces with high uncertainty, said development being to test promising areas according to completed experimental results; as the number of steps increases, the focus is changed from exploration to development, and beta is gradually reduced, so that the following steps are obtained:
UCB(x)=μ(x)+β 0 ε k σ(x)
wherein beta is 0 ,ε k Are super parameters, k represents the optimized times, epsilon is (0, 1) in the value range, epsilon k Characterization reduces the weight of σ (x) to turn emphasis towards development;
the PI acquisition function is as follows:
the PI utilizes posterior model to estimate that the function value is larger than the maximum value mu in the actual observed value at present + To find the next round of sampling points, PI (x) represents the probability, and the formula is as follows:
wherein phi represents a normal cumulative distribution function, mu (x) and sigma (x) respectively represent a Gaussian average value and a standard deviation of posterior prediction, and xi is more than or equal to 0 and is an adjustable parameter, wherein the aim is to find the sum mu + The point at which the probability of a comparative boost ζ is greatest;
the EI acquisition function is as follows:
the weight Φ (Z) is a normal cumulative distribution function, and ζ represents the sum μ + The degree of improvement in the phase ratio, phi (Z), represents the normal probability distribution function, and the objective of EI is to find the point (μ (x) - μ) with the greatest improvement expectations + ζ) Φ (Z) and σ (x) Φ (Z) are used for development and exploration, respectively.
S4: based on the battery state in the step S2, the digital twin model of the battery carries out simulation prediction on the aging of the battery under the charging protocol extracted in the next round, so as to obtain the aging of the battery under the charging protocol extracted in the next round;
s5: and adding the new charging protocol and the corresponding battery aging into the existing data set, transmitting the new charging protocol to the Bayesian optimizer again to recommend the next round of extracted quick charging protocol, and repeating the steps S4-S5 until the final recommended charging protocol is obtained, and charging the vehicle battery by the charging protocol.
Taking the single cycle lithium inventory loss as an example of an optimization objective, as shown in fig. 2, the SPMe model as a digital twin model of the battery, and the DFN model as an actual vehicle, the cumulative lithium inventory loss is shown in fig. 2. When 100 loops are continuously optimized, an optimal charging protocol is obtained for each loop, 100 loops correspond to 100 charging protocols, 100 loops are continuously simulated by an SPMe model and a DFN model, and the results are respectively shown as a simulation result-SPMe curve and an experimental result-DFN curve. To verify the optimized performance, 100 cycles were simulated in a DFN model with constant current charging, as shown by the "constant current experimental results-DFN". By comparison, the method provided by the invention is used for optimizing, so that the loss of lithium inventory can be effectively reduced. In this experiment, a charging protocol from 0 to 80% SOC for 30 minutes was used, and the "constant current experiment result-DFN" corresponds to a charging rate of 2.4C.
The bayesian optimization can adopt a parallel optimization method, the idea is that a plurality of charging protocols are recommended for each time by an acquisition function, namely, suboptimal protocols are recommended in addition to optimal protocols, then the recommended protocols are simulated in parallel by a battery digital twin model, the recommended protocols and corresponding battery aging are added into a dataset to obtain a new dataset, gaussian process regression is performed based on the new dataset, the charging protocols are recommended by the acquisition function, and the processes are repeated until the optimal charging protocols are obtained. As shown in (a) - (d) in fig. 3, the optimization results of the 10 th, 30 th, 50 th and 80 th charging of the vehicle are selected and displayed, the initial 3+parallel 3 represents that the initial data set contains three samples, and the acquisition function recommends three charging protocols each time during parallel optimization. It can be seen that the greater the number of protocols for initializing the data set and for parallel optimization, the better the result of the optimization and the faster the convergence rate of the optimization.
Finally, it is noted that the above embodiments are only for illustrating the technical solution of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications and equivalents may be made thereto without departing from the spirit and scope of the present invention, which is intended to be covered by the claims of the present invention.
Claims (9)
1. A fast charge strategy design method based on a battery digital twin model and machine learning is characterized in that: the method comprises the following steps:
s1: when an electric automobile is connected into a charging pile for charging, battery state data of the automobile are detected, wherein the battery state data comprise battery temperature, state of charge (SOC), state of health (SOH), ampere-hour throughput, current lithium inventory loss and Electrochemical Impedance Spectroscopy (EIS) detection;
s2: inputting the battery state data into a Bayesian optimizer based on a battery digital twin model, and predicting the aging condition of the battery in the current charging under an initial protocol according to the current battery state through the battery digital twin model; the initial protocol is any number of protocols, and corresponding data are used for initializing a Bayesian optimization model;
s3: transmitting an initial charging protocol and battery aging of a battery single cycle under the initial charging protocol as a data set to a Bayesian optimizer, wherein the initial charging protocol is used as an input of a Bayesian optimization model, and the battery aging of the battery single cycle under the initial charging protocol is used as an output; predicting battery aging corresponding to all protocols in a charging protocol space by utilizing Gaussian process regression GPR, and recommending a charging protocol for next round of extraction by utilizing an acquisition function;
s4: based on the battery state in the step S2, the digital twin model of the battery carries out simulation prediction on the aging of the battery under the charging protocol extracted in the next round, and the aging condition of the battery under the charging protocol extracted in the next round is obtained;
s5: and adding the new charging protocol and the corresponding battery aging into the existing data set, transmitting the new charging protocol to the Bayesian optimizer again to recommend the next round of extracted quick charging protocol, and repeating the steps S4-S5 until the final recommended charging protocol is obtained, and charging the vehicle battery by the charging protocol.
2. The battery digital twin model and machine learning based fast charge strategy design method of claim 1, wherein the method comprises the following steps: the initial protocol and the protocol recommended by the Bayesian optimizer are both protocols in a charging protocol space; the charging protocol in the charging protocol space aims to charge the battery state of charge (SOC) from 0% to 80% through constant current charging of a plurality of stages within 10 minutes, and in the process, the battery is charged by stages by using j kinds of currents with different magnitudes;
the charging protocol is a multi-stage constant-current constant-voltage charging MSCC, which consists of a plurality of constant-current charging stages and a constant-voltage charging stage which follows the constant-current charging stages;
when j=3, the charging protocol charges the battery state of charge SOC from 0% to 80% in 3 steps within 10 minutes, and parameters of the charging protocol have CC1', CC2', which are also two-dimensional inputs of the black box optimization model; the method comprises the following steps: the charging rate of the 0-26.67% SOC interval is CC1', the charging rate of the 26.67% -53.34% SOC interval is CC2', the charging rate of the 53.34% -80% SOC interval is CC3', the size of the CC3' is determined by CC1', CC2', and the charging rate is expressed as:
when j=4, the charging protocol charges the battery state of charge SOC from 0% to 80% in 10 minutes in 4 steps, and parameters of the charging protocol are CC1, CC2, and CC3, which are also three-dimensional inputs of the black box optimization model; the method comprises the following steps: the charging rate of the 0-20% SOC interval is CC1, the charging rate of the 21-40% SOC interval is CC2, the charging rate of the 41-60% SOC interval is CC3, and the charging rate of the 61-80% SOC interval is CC4, wherein the size of the CC4 is determined by CC1, CC2 and CC3, and the formula is as follows:
when j=5, the charging protocol charges the battery state of charge SOC from 0% to 80% in 5 steps within 10 minutes, with parameters of the charging protocol being CC1 ", CC 2", CC3 ", CC 4", which is also a four-dimensional input of the black box optimization model; the method comprises the following steps: the charging rate of the 0-16% SOC interval is CC1 ", the charging rate of the 16-32% SOC interval is CC 2", the charging rate of the 32-48% SOC interval is CC3 ", the charging rate of the 48-64% SOC interval is CC 4", the charging rate of the 64-80% SOC interval is CC5 ", wherein the size of CC 5" is CC1 ", CC 2", CC3 ", and the formula is as follows:
3. the battery digital twin model and machine learning based fast charge strategy design method of claim 1, wherein the method comprises the following steps: the battery digital twin model is a battery electrochemical model that considers battery aging mechanisms including SEI film growth, lithium plating, particle cracking, and active material loss, integrated into a DFN model (Doyle-Fuller-Newman) to characterize the aging behavior of the battery.
4. The battery digital twin model and machine learning based fast charge strategy design method according to claim 3, wherein: the SEI film growth specifically comprises:
the growth of the SEI film is divided into two processes: firstly, polarization of a battery cathode, reduction decomposition of an organic solvent in electrolyte to generate a new compound, and secondly, precipitation of a new chemical substance on the surface of the anode to form a new SEI film; the growth of the SEI film is captured using a diffusion limited model, whose reaction rate is limited by the rate of transport of solvent through the SEI outer layer, assuming that the flux of solvent molecules follows the Phake law:
the boundary conditions are:
wherein c sol Is the concentration of the solvent, c sol,e Is the concentration of the solvent in the electrolyte, D sol (T) is the solvent diffusion coefficient, L is the position of the SEI film, L SEI Is SEI film thickness;
the flux density at the SEI film interface is:
rate of increase of SEI film thickness:
wherein L is SEI Is SThe thickness of the EI film is set,is the average partial molar volume of the SEI; the growth of the SEI film is affected by temperature and solvent diffusion coefficient, which follows the Arrhenius equation:
wherein T is ref =25℃,E sol Is the activation energy of solvent diffusion;
the growth of the SEI film will result in the following effects:
wherein ε is n Is the porosity of the cathode, a n Is the specific surface area, a for spherical particles n =3ε n /R n ,R n Is the radius eta of the spherical particles of the negative electrode SEI Is the overpotential of SEI film, j tot Is the total current density in conservation of negative charge, σ SEI Is the conductivity of the SEI film.
5. The battery digital twin model and machine learning based fast charge strategy design method according to claim 3, wherein: the lithium plating specifically includes:
when lithium ions are not intercalated into an electrode but lithium metal is formed on the surface of the electrode, lithium plating occurs, which is partially or completely recovered in a subsequent discharge reaction by lithium stripping; lithium electroplating also chemically reacts with the electrolyte solution to form an SEI film, and the metal lithium separated by the SEI film cannot be peeled off, and the part of lithium is called dead lithium; thus, a partially reversible lithium plating model was constructed, according to the butler-furat equation, with the lithium plating/stripping flux at the anode/electrolyte interface in the overpotential correction form being:
wherein k is Li Is the lithium plating/stripping constant, c Li Is the concentration of lithium plating, F is Faraday constant, c e Is Li in electrolyte + Concentration, transfer coefficient alpha a,Li And alpha c,Li Are all set to 0.5; η (eta) Li Is the lithium deposition potential LDP, defined as eta Li =φ s -φ e -η SEI Wherein phi is s And phi e Respectively relative to Li/Li + Electrode potential and electrolyte potential of (a);
by coupling three models of lithium electroplating, lithium stripping and SEI growth, the electroplated lithium will decay over time into SEI film and dead lithium, c Li Is expressed as:
wherein a is - Is the ratio of the surface area to the volume of the cathode, c dl Is the concentration of dead lithium, defined as:
it is assumed that the dead lithium decay rate γ is inversely proportional to the thickness of the SEI film:
wherein gamma is 0 Is a fitting parameter, L SEI,0 Is the initial SEI thickness, L SEI,t Is the SEI thickness at time t.
6. The battery digital twin model and machine learning based fast charge strategy design method according to claim 3, wherein: the particle cracking specifically includes:
radial stress sigma using fatigue crack model based on spherical electrode particles r (r) tangential stress sigma t The expressions of (r) and displacement u are as follows:
where Ω is the partial molar volume, E is the Young's modulus, v is the Poisson's ratio, c avg (R) is a function of average lithium ion concentration and radius, R is particle radius,is the concentration c of lithium ion and the reference concentration c under no stress ref Deviation of (2);
the fatigue crack propagation model follows Paris's law, assuming that all electrode particle surfaces have the same crack length l cr Crack width w cr And crack density ρ cr :
Wherein t is 0 Time of single period, b cr Is a stress intensity correction factor, k cr And m cr Is a constant obtained from the experiment;
the growth model of the SEI film on the new crack is defined as follows:
wherein L is SEI,cr Is the average thickness of SEI film on new crack, L SEI,cr0 =L SEI,0 And/10000 is the thickness of the initial SEI film on the new crack.
7. The battery digital twin model and machine learning based fast charge strategy design method according to claim 3, wherein: the active material loss specifically includes:
loss of active substance epsilon a Simulation by decreasing the volume fraction of active material:
wherein beta, m 2 Sum sigma c Is the coefficient obtained from the experiment, the hydrostatic stress sigma h =(σ r +2σ t )/3,σ h The subscripts max and min of (c) represent the maximum and minimum values, respectively.
8. The battery digital twin model and machine learning based fast charge strategy design method of claim 1, wherein the method comprises the following steps: the Bayesian optimizer consists of a Gaussian process regression model GPR and an acquisition function, and the optimization target is that the aging of the battery is minimum, wherein the aging of the battery comprises capacity loss, lithium inventory loss, coulomb efficiency and active substance loss;
the GPR model is used for outputting logarithmic variance characteristics of all three-dimensional charging protocols; the acquisition function is used to decide which protocols to test in the next round;
when GPR is applied, the actual output y after taking noise into account is expressed as:
y i =f(x i )+ε i
wherein (x) i ,y i ) Represents the i-th data point, f represents the feature map, f (x i ) Representing the output of the gaussian model; epsilon i Represents observation noise, and
set x= { X of arbitrary finite input composition 1 ,x 2 ,...,x n } d*n D x n represents n samples, each sample having d features as input, { x 1 ,x 2 ,...,x n } d*n Obeys a joint gaussian probability distribution p (f (x) 1 ),f(x 2 ),...,f(x n ) The mean function and covariance function of f (x) are:
m(x)=E[f(x)]
k(x,x′)=E[(f(x)-m(x))(f(x′)-m(x′))]
since there is no a priori information of the distribution of f (x), m (x) is set to 0, and the distribution of y is written as:
wherein K is ff Is an n-by-n dimensional matrix, and [ K ] ff ] i,j =k(x i ,x j );
For a new inputd x n represents that the new data set has m samples, each sample having d features input thereto; the joint distribution of this and the known observed target values is written as:
Conditional distribution p (f) * |X,y,X * ) Obeys normal distribution:
wherein,,represents the mean of posterior distribution, cov (f * ) The variance of the posterior distribution is expressed as follows:
kernel functions that can be used in the GPR model include the following:
the first kernel function is a gaussian kernel RBF, given by:
wherein II x-x' II 2 Representing the square of the Euclidean distance of the vectors x and x', and sigma represents the standard deviation;
the second kernel function is a Matern kernel, given by:
having positive parameters v and l, wherein v is a f functionNumber, K ν Is a second type of bessel function, d (x, x') represents the euclidean distance.
9. The battery digital twin model and machine learning based fast charge strategy design method of claim 8, wherein the method comprises the following steps: the acquisition function is a confidence interval upper bound function UCB, an improved probability function PI or an expected increment function EI, wherein:
the UCB acquisition function is as follows:
UCB(x)=μ(x)+βσ(x)
wherein μ (x) and σ (x) are the mean and standard deviation of the posterior distribution, respectively, β is a hyper-parameter to be optimized for balanced exploration and development, said exploration being to explore untested parameter spaces with high uncertainty, said development being to test promising areas according to completed experimental results; as the number of steps increases, the focus is changed from exploration to development, and beta is gradually reduced, so that the following steps are obtained:
UCB(x)=μ(x)+β 0 ε k σ(x)
wherein beta is 0 ,ε k Are super parameters, k represents the optimized times, epsilon is (0, 1) in the value range, epsilon k Characterization reduces the weight of σ (x) to turn emphasis towards development;
the PI acquisition function is as follows:
the PI utilizes posterior model to estimate that the function value is larger than the maximum value mu in the actual observed value at present + To find the next round of sampling points, PI (x) represents the probability, and the formula is as follows:
wherein phi represents a normal cumulative distribution function, mu (x) and sigma (x) respectively represent a Gaussian average value and a standard deviation of posterior prediction, and xi is more than or equal to 0 and is an adjustable parameter, wherein the aim is to find the sum mu + The point at which the probability of a comparative boost ζ is greatest;
the EI acquisition function is as follows:
the weight Φ (Z) is a normal cumulative distribution function, and ζ represents the sum μ + The degree of improvement in the phase ratio, σ (Z), represents a normal probability distribution function, and the objective of EI is to find the point (μ (x) - μ) with the greatest improvement expectations + ζ) Φ (Z) and σ (x) Φ (Z) are used for development and exploration, respectively.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310549532.XA CN116401770A (en) | 2023-05-16 | 2023-05-16 | Quick charge strategy design method based on battery digital twin model and machine learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310549532.XA CN116401770A (en) | 2023-05-16 | 2023-05-16 | Quick charge strategy design method based on battery digital twin model and machine learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116401770A true CN116401770A (en) | 2023-07-07 |
Family
ID=87007758
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310549532.XA Pending CN116401770A (en) | 2023-05-16 | 2023-05-16 | Quick charge strategy design method based on battery digital twin model and machine learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116401770A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116577680A (en) * | 2023-07-13 | 2023-08-11 | 无锡大派电子有限公司 | Lithium battery lithium precipitation detection method and device |
CN117236200A (en) * | 2023-11-16 | 2023-12-15 | 北京航空航天大学 | Method for optimizing quick charge strategy of aerocar battery based on data-driven reduced-order model |
-
2023
- 2023-05-16 CN CN202310549532.XA patent/CN116401770A/en active Pending
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116577680A (en) * | 2023-07-13 | 2023-08-11 | 无锡大派电子有限公司 | Lithium battery lithium precipitation detection method and device |
CN117236200A (en) * | 2023-11-16 | 2023-12-15 | 北京航空航天大学 | Method for optimizing quick charge strategy of aerocar battery based on data-driven reduced-order model |
CN117236200B (en) * | 2023-11-16 | 2024-02-02 | 北京航空航天大学 | Method for optimizing quick charge strategy of aerocar battery based on data-driven reduced-order model |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Xiao et al. | Accurate state-of-charge estimation approach for lithium-ion batteries by gated recurrent unit with ensemble optimizer | |
Yang et al. | State-of-charge estimation of lithium-ion batteries using LSTM and UKF | |
Zhang et al. | A reliable data-driven state-of-health estimation model for lithium-ion batteries in electric vehicles | |
Vennam et al. | A survey on lithium-ion battery internal and external degradation modeling and state of health estimation | |
CN116401770A (en) | Quick charge strategy design method based on battery digital twin model and machine learning | |
CN107741568B (en) | Lithium battery SOC estimation method based on state transition optimization RBF neural network | |
CN111398832A (en) | Bus battery SOC prediction method based on ANFIS model | |
Yu et al. | CPSO-based parameter-identification method for the fractional-order modeling of lithium-ion batteries | |
Li et al. | A novel state estimation approach based on adaptive unscented Kalman filter for electric vehicles | |
Huang et al. | A comprehensively optimized lithium-ion battery state-of-health estimator based on Local Coulomb Counting Curve | |
Liu et al. | State of charge estimation for lithium-ion batteries based on improved barnacle mating optimizer and support vector machine | |
Chung et al. | State-of-charge estimation of lithium-ion batteries using LSTM deep learning method | |
Takyi-Aninakwa et al. | A hybrid probabilistic correction model for the state of charge estimation of lithium-ion batteries considering dynamic currents and temperatures | |
Zhang et al. | State of health estimation method for lithium-ion batteries using incremental capacity and long short-term memory network | |
Lin et al. | Lithium-ion batteries SOH estimation with multimodal multilinear feature fusion | |
CN116643196A (en) | Battery health state estimation method integrating mechanism and data driving model | |
Savargaonkar et al. | Uncorrelated sparse autoencoder with long short-term memory for state-of-charge estimations in lithium-ion battery cells | |
Huang et al. | State of charge estimation of li-ion batteries based on the noise-adaptive interacting multiple model | |
Takyi-Aninakwa et al. | An ASTSEKF optimizer with nonlinear condition adaptability for accurate SOC estimation of lithium-ion batteries | |
CN114839537A (en) | Battery state of charge prediction method and device | |
Alsuwian et al. | A Review of Expert Hybrid and Co-Estimation Techniques for SOH and RUL Estimation in Battery Management System with Electric Vehicle Application | |
Wang et al. | A novel capacity estimation method for li-ion battery cell by applying ensemble learning to extremely sparse significant points | |
Lin et al. | Physics-informed deep learning for lithium-ion battery diagnostics using electrochemical impedance spectroscopy | |
Tang et al. | Model parameter identification for lithium-ion batteries using adaptive multi-context cooperatively co-evolutionary parallel differential evolution algorithm | |
Qian et al. | A CNN-SAM-LSTM hybrid neural network for multi-state estimation of lithium-ion batteries under dynamical operating conditions |
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 |