CN115994441A - Big data cloud platform online battery life prediction method based on mechanism information - Google Patents
Big data cloud platform online battery life prediction method based on mechanism information Download PDFInfo
- Publication number
- CN115994441A CN115994441A CN202211325322.4A CN202211325322A CN115994441A CN 115994441 A CN115994441 A CN 115994441A CN 202211325322 A CN202211325322 A CN 202211325322A CN 115994441 A CN115994441 A CN 115994441A
- Authority
- CN
- China
- Prior art keywords
- battery
- life
- capacity
- predicted
- cloud platform
- 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
Images
Classifications
-
- 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
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E60/00—Enabling technologies; Technologies with a potential or indirect contribution to GHG emissions mitigation
- Y02E60/10—Energy storage using batteries
Abstract
The invention relates to a big data cloud platform online battery life prediction method based on mechanism information, and belongs to the field of battery performance test. The method comprises the following steps: collecting battery charge-discharge cycle data under various rapid charge cycle protocols, and establishing a database; establishing a capacity-voltage curve through discharge cycle data, and extracting capacity difference value-voltage curves of battery samples at different moments, wherein the capacity difference value-voltage curves comprise aging mechanism information; performing dimension reduction processing on the curve to obtain feature data, taking the feature data as input, and performing regression model training by taking the residual life and capacity fading inflection point of the battery sample as output to obtain a prediction model; and taking the active battery as a battery to be predicted, reading discharge cycle data of the battery to be predicted, predicting the residual life and capacity fading inflection points of the battery through a prediction model, and classifying the life. The invention establishes a mode based on the combination of the characteristics of the mechanism information and the artificial intelligence technology, and accurately predicts the service life and inflection point of the battery on line.
Description
Technical Field
The invention belongs to the field of battery performance test, and relates to a big data cloud platform online battery life prediction method based on mechanism information.
Background
Lithium ion batteries have been developed rapidly in recent years as a novel energy source because of their high operating voltage, large specific energy, high charge and discharge efficiency, low self-discharge rate, no memory effect, long cycle life, and the like. At present, besides being applied to daily equipment such as mobile phones, notebooks and the like, the lithium ion battery is widely applied to important fields such as electric automobiles, aerospace vehicles, artificial satellites and the like. During long-term use of lithium ion batteries, a series of electrochemical reactions and physical changes occur inside the battery, for example: the positive electrode material is dissolved, phase change of the positive electrode material occurs, electrolyte is decomposed and the like. These changes will lead to degradation of battery performance and capacity, which will lead to failure and breakdown of the system that is the main power supply, and in severe cases even to property loss and casualties, such as: the mobile phone explodes or the electric vehicle suddenly runs away. Therefore, the method has great significance in predicting the service life of the battery. The traditional prediction method is to set up an offline prediction model, assume that the future working conditions are similar to the past, and use the early part capacity to predict the decay trajectory until the end of the battery life is reached. For example, after a vehicle power battery is manufactured or designed by a manufacturer, a partial charge-discharge cycle test is performed to evaluate the performance of the battery, including battery life. The supposition principle of the method is unrealistic in practice, updating change is not carried out after the prediction model is established, real-time change of the battery load working condition cannot be recorded, however, factors such as the battery load working condition of the electric automobile can change at any moment, the performance attenuation mechanism of the battery is quite complex, and an accurate prediction result cannot be obtained by an offline prediction model.
With the deep research, the prior art proposes to conduct online prediction of battery life by establishing an online prediction model. The main method at present is to predict the residual service life of the lithium ion battery by establishing a physical model based on a mechanism or adopting a method based on data driving. Because it is difficult to find a model that conforms to the battery aging mechanism in the prior art, it is generally believed that the accuracy of data-driven based methods predictions is higher, such as the chinese patent publication No. CN 109977622B. The method can fully utilize online historical data of a battery body to be predicted and cloud big data containing non-body information to predict the residual service life of the battery, has the advantages of being simple in steps, high in prediction accuracy, capable of greatly improving the utilization rate of available information sources, capable of supporting any number of non-body information as prediction input quantity and the like, and capable of effectively adapting to big data application scenes. However, the data driving-based method generally only uses capacity degradation information, the method lacks a physical information basis, has weak significance on actual representation of the service life of the battery, needs a large amount of training, needs a large amount of data, is easy to generate the problem of data island, takes longer time, and has lower efficiency in predicting the residual service life of the battery.
The physical model based on the mechanism mainly obtains the evolution process of the performance degradation of the lithium ion battery by quantifying the influencing factors of the battery performance and according to the chemical reaction and the specific physical characteristics of the chemical reaction in the charging and discharging processes of the lithium ion battery, and the biggest challenge faced at present is that the evolution process of the battery performance has a plurality of factors to influence each other, so that the performance is reduced, and the complex dynamics system of the service life aging of the lithium ion battery is difficult to be reliably simulated. For example, chinese patent publication No. CN 114814631A discloses an online life prediction method for a vehicle lithium battery based on cloud computing and feature selection, which collects data of the life of a brand new lithium battery in the same constant-current and constant-voltage charging stage as a preset charging and discharging cycle interval and stores the data in a cloud; respectively obtaining the number distribution feature set of the constant-current constant-voltage charging mode stage of the current lithium battery by utilizing cloud computing; repeating the steps to obtain the service life and the number distribution characteristic set of each lithium battery; the number distribution feature sets are fused and then feature selection is carried out, so that a feature optimization training set is obtained; training and obtaining a trained lithium battery life prediction regression model; and during online prediction, acquiring and obtaining a feature optimization selection set of the lithium battery to be predicted, and predicting the cycle life of the current lithium battery to be predicted by using a trained lithium battery life prediction regression model. The method solves the problem that the online life prediction of the lithium battery in practical application depends on a specific discharge mode and accurate capacity measurement. However, this method starts to collect data from within a preset charge-discharge cycle interval of the battery, and also has high time cost, high data quality dependence, and does not consider prediction of the remaining life of the battery at any stage of battery use and at different cycle interval lengths.
Therefore, how to improve the efficiency of predicting the remaining service life of the battery and realize online battery life prediction, and simultaneously ensure the accuracy of prediction is a problem to be solved urgently.
Disclosure of Invention
In view of the above, the present invention aims to establish a mode based on the combination of feature extraction and artificial intelligence technology including battery aging physical information on a cloud platform, so as to realize accurate prediction of battery performance.
In order to achieve the above purpose, the present invention provides the following technical solutions:
the big data cloud platform online battery life prediction method based on the mechanism information comprises the following steps:
s1: collecting battery charge-discharge cycle data under various rapid charge cycle protocols, and establishing a battery online cloud platform database; the battery charge-discharge cycle data comprise battery temperature, current, voltage and charge-discharge capacity;
s2: establishing a capacity-voltage curve through discharge cycle data, and extracting capacity difference value-voltage curves of battery samples at different moments from the capacity-voltage curve, wherein the capacity difference value-voltage curve contains sample aging mechanism information, and the calculation method of the capacity difference value-voltage curve is to subtract the capacity of a moving window in the mth period from the capacity of the corresponding voltage in the nth period capacity-voltage curve;
s3: performing dimension reduction processing on the data, setting a voltage sampling interval to be 80mv, and taking characteristic data delta Q_V of 20 points from the capacity difference value-voltage curve; taking delta Q_V as input, taking the residual life and capacity fading inflection point of a battery sample as output, and carrying out regression model training under the frame of a data driving method to obtain an online life prediction model and an inflection point prediction model of the battery; the regression model adopts a Gaussian process regression algorithm or a partial least squares regression algorithm;
s4: taking the active battery as a battery to be predicted, reading discharge cycle data of the battery to be predicted in the service period, calculating a capacity-voltage curve of the battery to be predicted, and extracting characteristic data delta Q_V from the capacity-voltage curve of the battery to be predicted;
s5: substituting the delta Q_V obtained from the S4 into the online life prediction model to predict the residual life of the battery to be predicted, and predicting the capacity fading inflection point of the predicted battery through the inflection point prediction model.
Further, in the step S3, when the regression model is a gaussian process regression algorithm, probability distribution on a function is defined by gaussian process regression, which is defined as:
f(x)～GP(m(x)，k(x，x′))
wherein m (x) is a mean function, and k (x, x) is a covariance function, and the formula is as follows:
m(x)＝E[f(x)]
k(x，x′)＝E[(f(x)-m(x))(f(x′)-m(x′)) T ]
for any finite set of input data, as x= (X) 1 ，x 2 ，...，x p ) n×p There is a probability distribution, i.e. p (f (x 1 )，f(x 2 )，...f(x p ) Wherein the mean function m (x) and the covariance function K (x, x') are determined by a kernel function K i，j ＝κ(x i ，x j ) Is given;
using a Matern kernel function:
wherein v=5/2, r v Is a modified Bessel (Bessel) function, the average function being defined as m (x) =0;
there is now a training set d= { (x) containing inputs and outputs i ，y i ) X where prediction is required }, then * The condition distribution p (y) * |X * X, y), the gaussian distribution is given by:
p(y * |X * ，X，y)＝N(y * |m * ，σ * )
according to the conditional distribution properties of the multidimensional gaussian distribution, there are:
m * ＝K(X，X * ) T K(X，X) -1 y
σ * ＝K(X * ，X * )-K(X，X * ) T K(X，X) -1 K(X，X * )。
further, in the step S3, when the regression model is a partial least squares regression algorithm, principal components in the input and the output are extracted to maximize correlation between principal components extracted from the input and the output, respectively, that is:
wherein E is 0 And F 0 Respectively, data sets after input and output normalization processing, t 1 Is E 0 U 1 Is F 0 Is defined as the first component of: t is t 1 ＝E 0 ω 1 ，u 1 ＝F 0 c 1 。
Further, in the S3, the capacity fade inflection point is obtained by a knee algorithm; the knee algorithm calculates a line distance from the beginning of the life of the battery to the end of the life in the aging track, and selects a point with the maximum line distance.
Further, in the step S3, life classifier training is further included, where the life classifier training includes: classifying battery samples into two groups of long service life and short service life for classifier training to obtain a battery online service life classifier; and S6, grouping the batteries to be predicted through the life classifier.
Further, when the lifetime classifier training method is a support vector machine (Support Vector Machine, SVM) classifier, the method specifically includes:
solving a hyperplane of a feature space, wherein the hyperplane is defined as:
ω T x+b＝0
the corresponding classification function is:
wherein α= (α) 1 ，α 2 ，...α n ) Is the Lagrange multiplier, K is a kernel function operation as follows:
K(x 1 ，x 2 )＝(<x 1 ，x 2 >+R) d
wherein r=1, d=2; the objective function of the quadratic SVM is:
further, obtaining a residual recyclable period through the residual life of the battery, and classifying the life of the battery according to the residual recyclable period and a life classification threshold; the remaining recyclable period being longer than the life classification threshold and the remaining recyclable period being shorter than the life classification threshold being shorter.
Further, the lifetime classification threshold is a difference between an initial classification threshold and a cycled period, the initial classification threshold being 550 cycles. The cycled period is included in battery charge-discharge cycle data for a period of service of the battery to be predicted.
The invention has the beneficial effects that:
according to the method, the mode based on the combination of the characteristics including the battery aging mechanism information and the artificial intelligence technology is established on the cloud platform, so that the online life prediction and the accurate prediction of capacity fading inflection point prediction of the battery are realized, and the battery is classified according to the life. The method can play an important role in online management of the battery, and even if the battery is used for many years, the residual service life and capacity fading inflection point of the battery can be accurately predicted. Knowing the remaining life and inflection point of the battery not only provides clear guidance for predictive maintenance of the battery system, but also reveals the battery residuals required by warranty companies or buyer second-hand commerce. In addition, the on-line classification of the 'good (long service life)/bad (short service life)' batteries is carried out only according to the information of a few times of circulation, so that the sequencing/recombination process of the retired batteries can be greatly accelerated for secondary use, the improvement of the energy utilization rate is facilitated, and the development of new energy is promoted.
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 flow chart of a big data cloud platform online battery life prediction method based on mechanism information;
FIG. 2 is a graph of ΔQ (V) for an embodiment of the present method;
FIG. 3 shows the ΔQ extracted in the present method embodiment (100-10) A signature graph of (V);
FIG. 4 is a schematic diagram of a method for inflection point identification in an embodiment;
FIG. 5 is a graphical representation of inflection point results for some cells identified by a knee algorithm;
FIG. 6 is a schematic diagram showing cycle life prediction with a cycle interval of 120 times for different cycle periods;
wherein, (a) windows 130-10, (b) windows 240-120, (c) windows 370-250, (d) windows 490-370, (e) windows 600-480, and (f) windows 720-600, the result being based on the GPR model;
FIG. 7 is root mean square error for training and testing datasets for Gaussian process regression and partial least squares regression under different windows;
wherein (a) and (b) represent root mean square errors of training and testing data sets based on a gaussian process regression method, respectively, and (c) and (d) represent root mean square errors of training and testing data sets based on a partial least squares regression method;
FIG. 8 is an average percent error of training and testing data sets for Gaussian process regression and partial least squares regression models under different windows in life prediction;
wherein the different colored multi-segment lines represent different windows, (a) and (b) represent average percent errors of training and testing data sets based on a gaussian process regression method, respectively, and (c) and (d) represent average percent errors of training and testing data sets based on a partial least squares regression method;
FIG. 9 is a diagram of corner prediction for a different window with a window length of 120;
wherein the window lengths are (a) 130-10, (b) 200-80, (c) 270-150, (d) 340-220, (e) 400-280, (f) 480-360, respectively;
fig. 10 is a schematic diagram showing the results of classification prediction, showing classification accuracy under different moving windows in 5 cycle intervals, (a) and (b) show the classification accuracy of training and test data sets based on support vector machine classifier, respectively, and (c) and (d) show the classification accuracy of training and test data sets based on logistic regression classifier.
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 to 5, the method for predicting the service life of the big data cloud platform on line based on the mechanism information includes the following steps:
s1: collecting battery charge-discharge cycle data under various rapid charge cycle protocols, and establishing a battery online cloud platform database; the battery charge-discharge cycle data comprise battery temperature, current, voltage and charge-discharge capacity;
s2: establishing a capacity-voltage curve through discharge cycle data, and extracting capacity difference value-voltage curves of battery samples at different moments from the capacity-voltage curve, wherein the capacity difference value-voltage curve contains sample aging mechanism information, and the calculation method of the capacity difference value-voltage curve is to subtract the capacity of a moving window in the mth period from the capacity of the corresponding voltage in the nth period capacity-voltage curve;
s3: performing dimension reduction processing on the data, setting a voltage sampling interval to be 80mv, and taking characteristic data delta Q_V of 20 points from the capacity difference value-voltage curve; taking delta Q_V as input, taking the residual life and capacity fading inflection point of a battery sample as output, and carrying out regression model training under the frame of a data driving method to obtain an online life prediction model and an inflection point prediction model of the battery; the regression model adopts a Gaussian process regression algorithm or a partial least squares regression algorithm;
s4: taking the active battery as a battery to be predicted, reading discharge cycle data of the battery to be predicted in the service period, calculating a capacity-voltage curve of the battery to be predicted, and extracting characteristic data delta Q_V from the capacity-voltage curve of the battery to be predicted;
s5: substituting the delta Q_V obtained from the S4 into the online life prediction model to predict the residual life of the battery to be predicted, and predicting the capacity fading inflection point of the predicted battery through the inflection point prediction model.
Specifically, in S1, the published "a123" dataset is employed. The data set contains 124 commercial lithium iron phosphate (LFP)/graphite cells, and these cell samples are cycled under various fast charge conditions until the cells reach the end of life (EOL) of the cell, in this method, EOL is defined as the number of cycles required for each cell to reach 80% of the initial capacity. Typically, in the dataset, the battery is cycled under rapid charge conditions varying from 3.6C to 6C and then discharged at 4C until the lower cutoff voltage is reached. It contains a total of 72 charging protocols (almost all batteries have a charging time between 9 and 15 minutes) and the cycle life of the batteries in the dataset ranges from 150 to 2300 times (average cycle life 806, standard deviation 377). The data set includes the temperature, current, voltage, charge capacity, discharge capacity, etc. of each battery.
In this scenario, the data set is divided into a training data set for training the model and a test data set for evaluating the model. The number of data sets in different moving windows is different, and the ratio of the training data set to the test data set is approximately 3:2 as a whole.
The capacity-voltage curve contains rich information about the aging characteristics of the battery. As shown in fig. 2, the capacity-voltage curve moves with increasing number of cycles, and the area in the middle of adjacent voltage curves, which can characterize the discharge energy of the battery, decreases as the battery ages. Thus, according to the extracted DeltaQ (m-n) (V), e.g. DeltaQ (100-10) (V) (capacity difference)Value-voltage curve), Δq (100-10) (V) represents the energy change of the battery during different cycles to represent the battery life. ΔQ (100-10) (V) is the voltage subtraction of the 100 th cycle to 10 th cycle capacity-voltage curve. A typical shape of this feature is shown in fig. 3.
ΔQ of different m and n values (m-n) (V) also has a strong correlation with battery life, indicating that this feature within the moving window (defined by m and n) can be a robust representation of the remaining battery life (RUL).
To reduce the input dimension of the data, the dimension is reduced by varying the different sampling intervals without affecting the characteristic curve. For example, FIG. 3 shows ΔQ for different sample points (100-10) (V) -effect of voltage sampling results. It can be clearly observed that the basic shape of the curve is unchanged when the sampling interval is 80 mV. That is, in this example, only 20 points are needed to capture ΔQ (100-10) The main characteristic of the (V) curve greatly reduces the dimension and storage space of the input data. Therefore, only 20 points in the ΔQ (m-n) (V) curve need be sampled and input into a Machine Learning (ML) model to predict RUL. The reduced dimensionality of the input data facilitates efficient online battery life prediction without affecting the physical information contained.
Furthermore, since the number of ultra-long life batteries in the dataset is small (over 1500 cycles), the difference in discharge capacity between the current charge-discharge cycle and the initial discharge cycle is chosen as a calibration feature in order to better predict the ultra-long life batteries. The discharge capacity of the 10 th cycle was used as an initial value, and the additional feature was designated as DeltaQ m-10 。
A knee algorithm (knee algorithm) is employed in S3 to identify the inflection point. The inflection point of the battery degradation trajectory is a point at which the battery degradation rate is changed from slow to fast, and as shown in fig. 4, the inflection point is calculated as a point in the aging trajectory at which the line distance from the beginning of the battery life to the end of the life is maximum.
Fig. 5 shows the inflection point identification results of some cells in the dataset by using the knee algorithm.
The method extracts features containing battery aging mechanism information from discharge voltage data to indicate the state of health of the battery. These features are constructed based on data within a moving window, with boundaries defined by two different aging states in the battery decay trajectory. With the aging of the battery, the moving window can capture the latest aging and running information of the battery, and reasonably predict the aging track of the battery according to the latest aging and running information.
The machine learning in S3 specifically includes:
regression model:
gaussian process regression defines the probability distribution over the function, defined as:
f(x)～GP(m(x)，k(x，x′))
wherein m (x) is a mean function, and k (x, x') is a covariance function, and the formula is as follows:
m(x)＝E[f(x)]
k(x，x′)＝E[(f(x)-m(x))(f(x′)-m(x′)) T ]
for any finite set of input data, as x= (X) 1 ，x 2 ，...，x p ) n×p There is a probability distribution, i.e. p (f (x 1 )，f(x 2 )，...f(x p ) Wherein the mean function m (x) and the covariance function K (x, x') are determined by a kernel function K i，j ＝κ(x i ，x j ) Is given;
in the method, a Matern kernel function is used:
wherein v=5/2, r v Is a modified Bessel function, the average function is typically defined as m (x) =0, and the method follows this convention.
There is now a training set d= { (x) containing inputs and outputs i ，y i ) X where prediction is required }, then * The condition distribution p (y) * |X * X, y), the gaussian distribution is given by:
p(y * |X * ，X，y)＝N(y * |m * ，σ * )
according to the conditional distribution properties of the multidimensional gaussian distribution, there are:
m * ＝K(X，X * ) T K(X，X) -1 y
σ * ＝K(X * ，X * )-K(X，X * ) T K(X，X) -1 K(X，X * )
the value of covariance super-parameter θ can be determined by minimizing the negative log-marginal likelihood (defined as
Minimizing NLML automatically trades off between model complexity and data fit, improving over-fitting to the data. Given the expression of NLML and its derivatives with respect to θ, gradient-based optimization can be used to estimate θ. MatlabRegression learner are used to implement these algorithms.
In another embodiment, the regression model may also select the PLSR algorithm. PLSR is a multiple linear regression algorithm that considers both the extraction of principal components in Y and X as much as possible and the maximization of the correlation between principal components extracted from X and Y, respectively, namely:
wherein X is input (DeltaQ_V), Y is output (remaining life and capacity fade inflection point of battery), E 0 And F 0 Respectively, data sets after input and output normalization processing, t 1 Is E 0 U 1 Is F 0 Is defined as the first component of:
t 1 ＝E 0 ω 1 ，u 1 ＝F 0 c 1 。
to illustrate the prediction accuracy of the method, the test is performed by an evaluation model, which adopts a test data set, and the specific prediction model evaluation index uses Root Mean Square Error (RMSE) and average percent error (% err), namely the evaluation model is:
in addition to the basic approach described above, in practical applications, the online life classification may provide the necessary ordering reference for battery secondary life applications. In order to improve the consistency of the battery pack, it is necessary to recalibrate and reorganize the retired batteries for reliable and safe use during their second battery life phase.
The battery capacity is a direct indicator of battery health and is critical in the screening and reconstitution process. However, battery capacity is not the only important factor in determining battery performance. Even in the case where the initial capacities are similar, there may be a large difference in the battery degradation trajectories. Therefore, other characteristics such as internal resistance, electrochemical Impedance Spectroscopy (EIS), etc. are proposed as important supplements for improving the consistency of battery classification. Although sorting performance is improved, at least two problems arise. Firstly, the relationship between the additional sorting criteria and battery deterioration and EOL is unclear, and secondly, testing of these health characteristics (EIS etc.) is often very costly and time-intensive.
In order to solve the important problems, the life classifier training is performed simultaneously with the training of the prediction model, namely in S3, the characteristics containing the aging information of the battery sample are taken as input, and the battery sample is divided into two groups of long life and short life for the classifier training, so that the online life classifier of the battery is obtained; and when the residual life of the battery to be predicted is predicted, grouping the battery to be predicted through the life classifier.
The life of the battery is divided into long life and short life by taking the life classification threshold as a boundary, the value of the residual recyclable period of the battery is longer than the life classification threshold and is longer than the life classification threshold, and the value of the residual recyclable period of the battery is smaller than the life classification threshold and is shorter than the life classification threshold. The life classification threshold is the difference between the initial classification threshold and the value of the cycled cycle, the cycled cycle is contained in the battery charge-discharge cycle data of the service period of the battery to be predicted, and the initial classification threshold is 550 cycles.
Training of the life classifier specifically includes:
in the classification test, a method of selecting an SVM, when solving a two-classification problem, the SVM aims at solving a hyperplane of a feature space,
hyperplane is defined as:
ω T x +b＝0
the corresponding classification function is:
wherein α= (α) 1 ，α 2 ，...α n ) Is a lagrange multiplier, K is a kernel function operation,
the following are provided:
K(x 1 ，x 2 )＝(<x 1 ，x 2 >+R) d
wherein r=1, d=2; the objective function of the quadratic SVM is:
in another embodiment, a Logistic regression classification algorithm (Logistic) may be chosen, where the Logistic regression aims at learning a 0/1 classification model from the features, and this model uses linear combinations of features as arguments, using Logistic functions (or sigmoid functions) to map the arguments to (0, 1), where the mapped values are considered to be probabilities of belonging to the 0/1 classification, and similar effects can be achieved.
The prediction effect of the method is as follows:
1. life prediction
The RUL of the battery was predicted using Gaussian Process Regression (GPR) and Partial Least Squares Regression (PLSR) methods, moving windows starting from 10 th, 120 th, 250 th, 370 th, 480 th and 600 th cycles (6 moving windows total), respectively, with window lengths set to 30, 60, 90 and 120 cycles, respectively. If the design life of the battery is 10 years, the life before EOL is 1000 times, 30 times and 120 times respectively, which corresponds to the life of the battery being one season and one year.
As shown in fig. 6, (a), (b), (c), (d), (e) and (f) are training and test dataset predictions using the GPR method at 120 cycle window lengths, which indicate that the RUL values of the battery can be accurately predicted online at different aging stages, which is critical to reliably evaluate the performance of the battery during operation.
TABLE 1 residual life prediction results for different machine learning models
Table 1 gives the prediction results of RMSE and average percent error results for the remaining battery life for the GPR algorithm and PLSR algorithm at a 120 cycle window length. For both algorithms, RMSE and range error (MPE) decrease with forward movement of the window, indicating that the extracted features can more accurately capture the aging characteristics of the battery. This ever increasing prediction accuracy is favored in practice because an accurate battery RUL can be predicted even if the battery has been put into use for many years. As shown in Table 1, when GPR predicts RUL, RMSE and MPE are mostly less than 100 times and 10%, respectively, and when the battery approaches EOL, the errors are reduced to 55 times and 3.55%, respectively. PLSR performs slightly worse, most RMSE and MPE are larger than GPR.
As shown in fig. 7 and 8, RMSE and MPE of training and test datasets for online RUL prediction using GPR and PLSR models, respectively, are shown, with different colored polylines representing different windows. The results demonstrate that GPR and PLSR predict battery RUL with similar accuracy, which suggests that efficient mechanistic information features are extracted for reliable battery RUL prediction. However, GPR-based test results indicate that both RMSE and MPE decrease with increasing window length. In addition, RMSE and MPE decrease when GPR predicts RUL due to the increase in window size (the more operation information is acquired). Both the RMSE and MPE trends are advantageous in practical applications for aging batteries, but are not apparent in the RUL predictions based on the PLSR approach.
In addition, the PLSR model requires an artificial search for an optimal principal component number, the value of which varies with experimental data within each moving window. This variation increases the difficulty and impossibility of PLSR algorithms in practical applications. Furthermore, the GPR model has an important advantage over PLSR in that it can provide predictions of uncertainty, which makes online battery life predictions more reliable. Therefore, in practical applications, GPR versus PLSR is the first potential tool for accurately predicting battery RUL.
2. Inflection point prediction
On the basis of the RUL prediction, the inflection points are also predicted on line using the GPR model and the PLSR model, moving windows start from 10 th, 80 th, 150 th, 220 th, 280 th and 360 th cycles, respectively, and window lengths are set to 30, 60, 90 and 120 cycles, respectively.
FIG. 9 shows the prediction results of the inflection point of the GPR model at a window length of 120 cycles. The cell data set selected here is located before the inflection point. In the whole, the inflection point prediction is similar to the RUL prediction mode, the correlation between the inflection point and the service life is proved from the side, and the feasibility of the inflection point online prediction is also demonstrated.
TABLE 2 inflection point prediction results for different machine learning models
As can be seen from Table 2, the GPR-based inflection point predictions RMSE and MPE are within 80 cycles and 8%, respectively, and this accuracy can be used for online prediction. The prediction results for PLSR are also shown in Table 2, with RMSE and MPE being mostly greater than GPR. Therefore, compared with PLSR, the GPR model can more accurately predict RUL and inflection points of a battery under the condition of uncertainty, and is more suitable and feasible for online prediction scenes.
In summary, gaussian Process Regression (GPR) based on different moving windows predicts battery RUL and inflection points more accurately than Partial Least Squares Regression (PLSR), with Root Mean Square Error (RMSE) and Mean Percent Error (MPE) of mostly less than 100 times and 8%, respectively. As the window moves, both RUL and inflection point prediction accuracy is improved, and RMSE and MPE at 120 cycles window length are reduced below 85 cycles and below 5% respectively, near battery EOL or inflection point. The result shows that the RUL and the inflection point of the battery can still be accurately predicted after the battery is put into use for a plurality of years, and the high reliability and the feasibility of the method are verified.
3. Life classification
In the life classification, 4 moving windows are set, which are located in the 100 th cycle, the 200 th cycle, the 300 th cycle, and the 400 th cycle, respectively, using features including battery aging mechanism information, and different window lengths are considered, from the length of 5 cycles to the length of only 1 cycle. The battery cells were divided into two groups, a long life group and a short life group, and the life classification threshold was set to 550 cycles. Two typical classification methods include support vector machine classifiers and logistic regression classifiers are used for online classification.
As shown in fig. 10, the experimental result shows that the support vector machine can accurately classify the service life of the battery by only relying on data of one cycle, and the classification accuracy is mainly above 85%. As the single cycle window moves forward, the accuracy further rises to around 90%, again indicating the high performance of the online battery life classification. This breakthrough not only makes it possible to greatly improve the consistency of classification by covering battery life assessment, but also can significantly reduce the cost and time required for traditional classification by utilizing only one cycle of online data.
TABLE 3 Classification results for different machine learning models
The classification accuracy of the SVM classifier and the logistic regression classifier at a window length of 1 cycle is given in Table 3. The battery can be classified in good/bad mode only by one period of data, the classification accuracy of the SVM classifier is more than 85%, and the effect of the logistic regression classifier is slightly poorer.
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 (8)
1. The big data cloud platform online battery life prediction method based on the mechanism information is characterized by comprising the following steps of:
and (S1) S: collecting battery charge-discharge cycle data under various rapid charge cycle protocols, and establishing a battery online cloud platform database; the battery charge-discharge cycle data comprise battery temperature, current, voltage and charge-discharge capacity;
s2: establishing a capacity-voltage curve through discharge cycle data, and extracting capacity difference value-voltage curves of battery samples at different moments from the capacity-voltage curve, wherein the capacity difference value-voltage curve contains sample aging mechanism information, and the calculation method of the capacity difference value-voltage curve is to subtract the capacity of a moving window in the mth period from the capacity of the corresponding voltage in the nth period capacity-voltage curve;
s3: performing dimension reduction processing on the data, setting a voltage sampling interval to be 80mv, and taking characteristic data delta Q_V of 20 points from the capacity difference value-voltage curve; taking delta Q_V as input, taking the residual life and capacity fading inflection point of a battery sample as output, and carrying out regression model training under the frame of a data driving method to obtain an online life prediction model and an inflection point prediction model of the battery; the regression model adopts a Gaussian process regression algorithm or a partial least squares regression algorithm;
s4: taking the active battery as a battery to be predicted, reading discharge cycle data of the battery to be predicted in the service period, calculating a capacity-voltage curve of the battery to be predicted, and extracting characteristic data delta Q_V from the capacity-voltage curve of the battery to be predicted;
s5: substituting the delta Q_V obtained from the S4 into the online life prediction model to predict the residual life of the battery to be predicted, and predicting the capacity fading inflection point of the predicted battery through the inflection point prediction model.
2. The mechanism information-based big data cloud platform online battery life prediction method according to claim 1, wherein in the step S3, when the regression model is a gaussian process regression algorithm, probability distribution on a function is defined by gaussian process regression, which is defined as:
f(x)～GP(m(x)，k(x，x′))
wherein m (x) is a mean function, and k (x, x') is a covariance function, and the formula is as follows:
m(x)＝E[f(x)]
k(x，x′)＝E[(f(x)-m(x))(f(x′)-m(x′)) T ]
for any finite set of input data, as x= (X) 1 ,x 2 ,...，x p ) n×p There is a probability distribution, i.e. p (f (x 1 )，f(x 2 )，…f(x p ) Wherein the mean function m (x) and the covariance function K (x, x') are determined by a kernel function K i，j ＝κ(x i ，x j ) Is given;
using a Matern kernel function:
wherein v=5/2, r v Is a modified bessel function, the average function being defined as m (x) =0;
now have the include inputs andoutput training set d= { (x) i ,y i ) X where prediction is required }, then * The condition distribution p (y) * |X * X, y), the gaussian distribution is given by:
p(y * |X * ,X,y)＝N(y * |m * ,σ * )
conditional distribution properties according to a multidimensional gaussian distribution, wherein:
m * ＝K(X,X * ) T K(X,X) -1 y
σ * ＝K(X * ,X * )-K(X,X * ) T K(X,X) -1 K(X,X * )。
3. the mechanism information-based big data cloud platform online battery life prediction method according to claim 1, wherein in the step S3, when the regression model is a partial least squares regression algorithm, principal components in the input and the output are extracted to maximize correlation between principal components extracted from the input and the output, respectively, namely:
4. The mechanism information-based big data cloud platform online battery life prediction method according to claim 1, wherein in the S3, the capacity fade inflection point is obtained by a knee algorithm; the knee algorithm calculates a line distance from the beginning of the life of the battery to the end of the life in the aging track, and selects a point with the maximum line distance.
5. The mechanism information-based big data cloud platform online battery life prediction method according to claim 1, further comprising life classifier training in the S3, the life classifier training comprising: classifying battery samples into two groups of long service life and short service life for classifier training to obtain a battery online service life classifier; and S6, grouping the batteries to be predicted through the life classifier.
6. The method for predicting the service life of the big data cloud platform on line based on the mechanism information according to claim 1, wherein when the service life classifier training method is a support vector machine classifier, the method specifically comprises:
solving a hyperplane of a feature space, wherein the hyperplane is defined as:
ω T x+b＝0
the corresponding classification function is:
wherein α= (α) 1 ,α 2 ,…α n ) Is the Lagrangian multiplier, K is a kernel function operation as follows:
K(x 1 ,x 2 )＝(<x 1 ,x 2 >+R) d
wherein r=1, d=2; the objective function of the quadratic SVM is:
7. the big data cloud platform online battery life prediction method based on the mechanism information according to claim 1, wherein a remaining recyclable period is obtained through the remaining life of the battery, and the battery life is classified according to the remaining recyclable period and a life classification threshold; the remaining recyclable period being longer than the life classification threshold and the remaining recyclable period being shorter than the life classification threshold being shorter.
8. The mechanism information-based big data cloud platform online battery life prediction method of claim 1, wherein the life classification threshold is a difference between an initial classification threshold and a cycled period, and the initial classification threshold is 550 periods.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211325322.4A CN115994441A (en) | 2022-10-27 | 2022-10-27 | Big data cloud platform online battery life prediction method based on mechanism information |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211325322.4A CN115994441A (en) | 2022-10-27 | 2022-10-27 | Big data cloud platform online battery life prediction method based on mechanism information |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115994441A true CN115994441A (en) | 2023-04-21 |
Family
ID=85991021
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211325322.4A Pending CN115994441A (en) | 2022-10-27 | 2022-10-27 | Big data cloud platform online battery life prediction method based on mechanism information |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115994441A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116413609A (en) * | 2023-06-08 | 2023-07-11 | 江苏正力新能电池技术有限公司 | Battery diving identification method and device, electronic equipment and storage medium |
| CN116879753A (en) * | 2023-06-21 | 2023-10-13 | 重庆邮电大学 | Big data-based battery life prediction method |
| CN117233630A (en) * | 2023-11-16 | 2023-12-15 | 深圳屹艮科技有限公司 | Method and device for predicting service life of lithium ion battery and computer equipment |
-
2022
- 2022-10-27 CN CN202211325322.4A patent/CN115994441A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116413609A (en) * | 2023-06-08 | 2023-07-11 | 江苏正力新能电池技术有限公司 | Battery diving identification method and device, electronic equipment and storage medium |
| CN116413609B (en) * | 2023-06-08 | 2023-08-29 | 江苏正力新能电池技术有限公司 | Battery diving identification method and device, electronic equipment and storage medium |
| CN116879753A (en) * | 2023-06-21 | 2023-10-13 | 重庆邮电大学 | Big data-based battery life prediction method |
| CN117233630A (en) * | 2023-11-16 | 2023-12-15 | 深圳屹艮科技有限公司 | Method and device for predicting service life of lithium ion battery and computer equipment |
| CN117233630B (en) * | 2023-11-16 | 2024-03-15 | 深圳屹艮科技有限公司 | Method and device for predicting service life of lithium ion battery and computer equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108896914B (en) | Gradient lifting tree modeling and prediction method for health condition of lithium battery | |
| CN109031153B (en) | Online health state estimation method for lithium ion battery | |
| CN107290679B (en) | The Intelligentized battery method for detecting health status of charging pile is shared for electric car | |
| CN110058178A (en) | A kind of lithium battery method for detecting health status and system | |
| Zhang et al. | Cloud-based in-situ battery life prediction and classification using machine learning | |
| CN115994441A (en) | Big data cloud platform online battery life prediction method based on mechanism information | |
| US20210132151A1 (en) | Method and apparatus for estimating state of health of battery | |
| Cui et al. | A dynamic spatial-temporal attention-based GRU model with healthy features for state-of-health estimation of lithium-ion batteries | |
| CN113740736B (en) | Electric vehicle lithium battery SOH estimation method based on deep network self-adaption | |
| CN113917334B (en) | Battery health state estimation method based on evolution LSTM self-encoder | |
| Xiong et al. | A data-driven method for extracting aging features to accurately predict the battery health | |
| CN109782190A (en) | Method for estimating the remaining life of single battery or single batch of battery | |
| CN113484774B (en) | Lithium battery pack capacity estimation method based on temperature calibration and neural network | |
| CN111983474A (en) | Lithium ion battery life prediction method and system based on capacity decline model | |
| Che et al. | Battery health prognostic with sensor-free differential temperature voltammetry reconstruction and capacity estimation based on multi-domain adaptation | |
| CN113311337A (en) | Lithium ion battery residual life detection method based on relevance vector regression | |
| CN115201686B (en) | Lithium ion battery health state assessment method under incomplete charge and discharge data | |
| CN115308628A (en) | Battery health state monitoring method | |
| Fei et al. | Early-stage lifetime prediction for lithium-ion batteries: A deep learning framework jointly considering machine-learned and handcrafted data features | |
| Huang et al. | State of health estimation of lithium-ion batteries based on fine-tuning or rebuilding transfer learning strategies combined with new features mining | |
| Peng et al. | Data-driven battery capacity estimation based on partial discharging capacity curve for lithium-ion batteries | |
| Li et al. | Degradation Pattern Recognition and Features Extrapolation for Battery Capacity Trajectory Prediction | |
| CN116243194A (en) | On-line prediction method for battery health state under mixed working condition | |
| Burzyński | Useful energy prediction model of a Lithium-ion cell operating on various duty cycles | |
| Ganesh et al. | Prediction of Residual Energy in Batteries using CNN-BiGRU and Attention Mechanism Model |
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 |