CN116774057A - Method and device for training battery life prediction model and predicting battery life - Google Patents
Method and device for training battery life prediction model and predicting battery life Download PDFInfo
- Publication number
- CN116774057A CN116774057A CN202311041364.XA CN202311041364A CN116774057A CN 116774057 A CN116774057 A CN 116774057A CN 202311041364 A CN202311041364 A CN 202311041364A CN 116774057 A CN116774057 A CN 116774057A
- Authority
- CN
- China
- Prior art keywords
- battery
- life
- battery capacity
- capacity
- prediction model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 60
- 238000012549 training Methods 0.000 title claims abstract description 48
- 230000015556 catabolic process Effects 0.000 claims abstract description 51
- 238000006731 degradation reaction Methods 0.000 claims abstract description 51
- 238000004146 energy storage Methods 0.000 claims abstract description 25
- 238000012545 processing Methods 0.000 claims abstract description 10
- 238000003062 neural network model Methods 0.000 claims description 18
- 238000012360 testing method Methods 0.000 claims description 18
- 238000005562 fading Methods 0.000 claims description 14
- 238000012216 screening Methods 0.000 claims description 10
- 230000002457 bidirectional effect Effects 0.000 claims description 7
- 230000007850 degeneration Effects 0.000 claims description 5
- 238000000605 extraction Methods 0.000 claims description 4
- 238000005259 measurement Methods 0.000 claims description 4
- 125000004122 cyclic group Chemical group 0.000 claims description 3
- 238000012423 maintenance Methods 0.000 claims description 2
- 238000004364 calculation method Methods 0.000 abstract description 6
- 238000011156 evaluation Methods 0.000 abstract description 2
- 239000011159 matrix material Substances 0.000 description 5
- 230000009467 reduction Effects 0.000 description 5
- 238000007599 discharging Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 230000001419 dependent effect Effects 0.000 description 2
- 239000003792 electrolyte Substances 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000014509 gene expression Effects 0.000 description 2
- 230000036541 health Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 239000013598 vector Substances 0.000 description 2
- 230000007423 decrease Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000005055 memory storage Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000004064 recycling Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Landscapes
- Secondary Cells (AREA)
Abstract
The disclosure belongs to the technical field of battery state evaluation, and particularly relates to a method and a device for training a battery life prediction model and a method and a device for predicting battery life. The method for predicting battery life includes: acquiring preset characteristic data of each single battery in the energy storage battery cluster; inputting the characteristic data into a battery life grouping model to group the real-time lives of the single batteries; inputting the characteristic data into a residual battery capacity regression prediction model corresponding to the service life group to predict the residual battery capacity of the single battery; if the ratio of the residual battery capacity to the initial battery capacity of the single battery is smaller than a second preset threshold value, the characteristic data is input into a battery degradation prediction model to predict whether the single battery is degraded or not, and a processing strategy for the single battery is determined based on whether the single battery is degraded or not. The prediction method provided by the disclosure is suitable for each stage of the life cycle of the energy storage battery and has small calculation complexity.
Description
Technical Field
The disclosure belongs to the technical field of battery state evaluation, and particularly relates to a method and a device for training a battery life prediction model and a method and a device for predicting battery life.
Background
With the increase of the number of battery uses and charge-discharge cycles, the battery performance is lowered, the capacity is generally lowered, and the safety is also greatly lowered. Therefore, the service life of the battery, namely the number of times of battery remaining charge and discharge recycling, is an important index for battery health management, and the accurate prediction of the service life of the battery can greatly ensure the safe operation of the energy storage power station. The battery life may also be characterized by a remaining battery capacity, which is the battery capacity at the nth charge-discharge cycle, n being the total number of charge-discharge cycles after the battery has decayed.
Deep learning or machine learning models have been commonly used by the industry to construct energy storage battery life prediction models. However, the impact factors and the decay rates of the energy storage battery at each stage of the life cycle are different, a large amount of battery experimental data is needed for constructing a life prediction model of the energy storage battery, the feature dimensions actually measured and considered are very large, and the problem of model overfitting can be easily caused by directly using a deep learning or machine learning model due to overlarge parameters or the model cannot fall to the ground due to high calculation complexity. Therefore, battery life prediction models and dimension reduction means suitable for specific phases of the life cycle of the energy storage battery are also developed in the industry to reduce the computational complexity. However, the prediction model is severely degraded in performance at other stages of the life cycle of the energy storage battery, and the dimension reduction method is too subjective.
Disclosure of Invention
The embodiment of the disclosure provides a battery life prediction model training and a battery life prediction scheme, so as to solve the problem that the complexity of predicting the battery life by using the existing battery life prediction model is too high or can not be suitable for each stage of the life cycle of an energy storage battery.
A first aspect of an embodiment of the present disclosure provides a method of training a battery life prediction model, comprising:
collecting historical test data of a plurality of battery capacity cycle degenerations, and screening out final characteristics reflecting battery degeneration information from the historical test data;
acquiring the residual battery capacity and the initial battery capacity of the battery corresponding to each record in the final characteristics, grouping the battery life corresponding to the records based on the ratio of the residual battery capacity to the initial battery capacity, taking the records as samples, taking the battery life group corresponding to the records as a label, and training a bidirectional LSTM neural network model to establish a battery life grouping model;
based on each record in the battery life group and the residual battery capacity corresponding to the record, acquiring regression parameters by using a regression model, and establishing a residual battery capacity regression prediction model of the battery life group;
and respectively acquiring records of the battery in a fading state and a non-fading state in the final characteristics, taking the records as samples, taking the fading state of the battery corresponding to the records as a label, and training the GRU neural network model to establish a battery fading prediction model.
In some embodiments of the present disclosure, the collecting historical test data for a plurality of battery capacity cycle degradations includes:
according to historical declining experimental single battery data, obtaining actual measurement values of the battery characteristic variables after the previous k times of charge and discharge according to preset time intervals, wherein k is a natural number not greater than n, and n is the total cycle number of charge and discharge after the battery is declined;
and recording preset events and corresponding severity levels, wherein the preset events are events which can lead to the degradation of the service life of the battery.
In some embodiments of the disclosure, the screening the historical test data for final features reflecting battery degradation information includes:
normalizing the measured values of the battery characteristic variables, calculating the correlation between the characteristic variables, reserving the characteristic variables with larger characteristic information values between the two characteristic variables with the correlation larger than a preset threshold value so as to obtain a first effective characteristic variable, and deleting multiple co-linear variables from the first effective characteristic variable so as to obtain a second effective characteristic variable;
and extracting final characteristics reflecting battery degradation information from the second effective characteristic variables by using a stepwise regression model.
In some embodiments of the disclosure, the grouping the battery life corresponding to the record based on the ratio of the remaining battery capacity and the initial battery capacity includes:
if the ratio of the remaining battery capacity of the battery corresponding to the record to the initial battery capacity is not less than a first preset threshold value, grouping the battery life corresponding to the record into a reliable life group;
if the ratio of the remaining battery capacity of the battery corresponding to the record to the initial battery capacity is smaller than the first preset threshold value and not smaller than a second preset threshold value, grouping the battery life corresponding to the record into a median life group;
and if the ratio of the remaining battery capacity of the battery corresponding to the record to the initial battery capacity is smaller than the second preset threshold value, grouping the battery life corresponding to the record into a low-order life group.
In some embodiments of the disclosure, the separately obtaining records of the battery in the final feature in the degraded state and in the non-degraded state includes:
acquiring the electric corresponding to the final characteristicFirst and second times of poolRecording after secondary charge and discharge, wherein n is the total cycle number of charge and discharge after battery decay;
and acquiring a record corresponding to the nth charge and discharge of the battery in the final characteristic.
In some embodiments of the present disclosure, the method further comprises:
and when the number of the degraded batteries exceeds a preset threshold, updating the final characteristics based on the characteristic data of the degraded batteries, and updating the battery life grouping model, the residual battery capacity regression prediction model and the battery degradation prediction model based on the updated final characteristics.
A second aspect of an embodiment of the present disclosure provides an apparatus for training a battery life prediction model, comprising:
the characteristic extraction module is used for collecting historical test data of the cyclic degradation of the capacity of the plurality of batteries and screening out final characteristics reflecting the degradation information of the batteries from the historical test data;
the battery life grouping model training module is used for acquiring the residual battery capacity and the initial battery capacity of the battery corresponding to each record in the final characteristics, grouping the battery life corresponding to the record based on the ratio of the residual battery capacity to the initial battery capacity, taking the record as a sample, taking the battery life group corresponding to the record as a label, and training a bidirectional LSTM neural network model to establish a battery life grouping model;
the residual battery capacity regression prediction model building module is used for obtaining regression parameters by using a regression model based on each record in the battery life group and the residual battery capacity corresponding to the record, and building the residual battery capacity regression prediction model of the battery life group;
and the battery degradation prediction model training module is used for respectively acquiring records of the battery in a degradation state and a non-degradation state in the final characteristics, taking the records as samples, taking the degradation state of the battery corresponding to the records as a label, and training the GRU neural network model to establish the battery degradation prediction model.
A third aspect of an embodiment of the present disclosure provides a method of predicting battery life, including:
acquiring preset characteristic data of each single battery in an energy storage battery cluster, inputting the characteristic data into a battery life grouping model trained according to the method of the first aspect of the disclosure, and grouping the single batteries in real time life;
inputting the characteristic data into a residual battery capacity regression prediction model corresponding to the life grouping to predict the residual battery capacity of the single battery, wherein the residual battery capacity regression prediction model is established according to the method of the first aspect of the disclosure;
if the ratio of the remaining battery capacity to the initial battery capacity of the single battery is smaller than the second preset threshold, inputting the characteristic data into a battery degradation prediction model trained by the method according to the first aspect of the disclosure to predict whether the single battery is degraded, and determining a processing strategy for the single battery based on whether the single battery is degraded.
In some embodiments of the disclosure, the determining a processing policy for the battery based on whether the battery is degraded comprises:
if the ratio of the remaining battery capacity to the initial battery capacity of the single battery is smaller than the second preset threshold value and the single battery is degenerated, triggering a low-life red early warning and a single battery replacement operation corresponding to the red early warning;
and if the ratio of the residual battery capacity to the initial battery capacity of the single battery is smaller than the second preset threshold value and the single battery is not degenerated yet, triggering a low-life yellow early warning and a single battery maintenance operation corresponding to the yellow early warning.
A fourth aspect of embodiments of the present disclosure provides an apparatus for predicting battery life, comprising:
the battery life grouping module is used for acquiring preset characteristic data of each single battery in the energy storage battery cluster, inputting the characteristic data into the battery life grouping model trained according to the method of the first aspect of the disclosure, and grouping the single batteries in real time;
a remaining battery capacity prediction module configured to input the feature data into a remaining battery capacity regression prediction model corresponding to the lifetime group to predict a remaining battery capacity of the unit battery, where the remaining battery capacity regression prediction model is established according to the method of the first aspect of the present disclosure;
and the degradation prediction module is used for inputting the characteristic data into a battery degradation prediction model trained according to the method of the first aspect of the disclosure to predict whether the single battery is degraded or not if the ratio of the residual battery capacity to the initial battery capacity of the single battery is smaller than the second preset threshold value, and determining a processing strategy for the single battery based on whether the single battery is degraded or not.
In summary, according to the method for training the battery life prediction model, the method for predicting the battery life, the device for training the battery life prediction model and the device for predicting the battery life provided by the embodiments of the present disclosure, the regression model and the deep learning model are fused, the optimal characteristic variable is screened out objectively and effectively by using the co-linearity screening and stepwise regression, then the stage of the energy storage battery in the life cycle is estimated by using the battery life grouping model based on the optimal characteristic variable, the regression model corresponding to the stage is selected to predict the remaining battery capacity of the energy storage battery, finally the degradation prediction model is used to estimate whether the remaining battery capacity is close to the threshold value, and the accurate and precise prediction of the life of the energy storage battery in each stage of the life cycle of the energy storage battery can be realized by the combination of 3 prediction models. Meanwhile, the regression model is respectively built based on different stages of the life cycle, and feature dimensions required by training of the battery life grouping model and the regression prediction model are greatly reduced, so that the calculation complexity is greatly reduced.
Drawings
The features and advantages of the present disclosure will be more clearly understood by reference to the accompanying drawings, which are schematic and should not be construed as limiting the disclosure in any way, in which:
FIG. 1 is a flow chart of a method of training a battery life prediction model, shown in accordance with some embodiments of the present disclosure;
FIG. 2 is a schematic illustration of an apparatus for training a battery life prediction model, according to some embodiments of the present disclosure;
FIG. 3 is a flow chart of a method of predicting battery life shown in accordance with some embodiments of the present disclosure;
fig. 4 is a schematic diagram of an apparatus for predicting battery life, according to some embodiments of the present disclosure.
Detailed Description
In the following detailed description, numerous specific details are set forth by way of examples in order to provide a thorough understanding of the relevant disclosure. However, it will be apparent to one of ordinary skill in the art that the present disclosure may be practiced without these specific details. It should be appreciated that the use of "system," "apparatus," "unit," and/or "module" terms in this disclosure is one method for distinguishing between different parts, elements, portions, or components at different levels in a sequential arrangement. However, these terms may be replaced with other expressions if the other expressions can achieve the same purpose.
It will be understood that when a device, unit, or module is referred to as being "on," "connected to," or "coupled to" another device, unit, or module, it can be directly on, connected to, or coupled to, or in communication with the other device, unit, or module, or intervening devices, units, or modules may be present unless the context clearly indicates an exception. For example, the term "and/or" as used in this disclosure includes any and all combinations of one or more of the associated listed items.
The terminology used in the present disclosure is for the purpose of describing particular embodiments only and is not intended to limit the scope of the present disclosure. As used in the specification and the claims, the terms "a," "an," "the," and/or "the" are not specific to a singular, but may include a plurality, unless the context clearly dictates otherwise. In general, the terms "comprises" and "comprising" are intended to cover only those features, integers, steps, operations, elements, and/or components that are explicitly identified, but do not constitute an exclusive list, as other features, integers, steps, operations, elements, and/or components may be included.
These and other features and characteristics of the present disclosure, as well as the methods of operation, functions of the related elements of structure, combinations of parts and economies of manufacture, may be better understood with reference to the following description and the accompanying drawings, all of which form a part of this specification. It is to be expressly understood, however, that the drawings are for the purpose of illustration and description only and are not intended as a definition of the limits of the disclosure. It will be understood that the figures are not drawn to scale.
Various block diagrams are used in the present disclosure to illustrate various modifications of the embodiments according to the present disclosure. It should be understood that the foregoing or following structures are not intended to limit the present disclosure. The protection scope of the present disclosure is subject to the claims.
An embodiment of the present disclosure provides a method for training a battery life prediction model, as shown in fig. 1, the method including:
s110, collecting historical test data of a plurality of battery capacity cycle degradations, and extracting final characteristics reflecting battery degradation information from the historical test data.
Specifically, step 1 is performed first.
Step 1: according to historical declining experimental single battery data, obtaining related characteristic variable actual measurement values after the previous k times of charge and discharge, wherein k=1, … and n. n is the final life of the battery (the total number of cycles of charge and discharge after the battery has decayed).
The data acquisition comprises the following steps:
real-time measurement data: initial battery capacity, initial internal resistance, minimum voltage, maximum voltage, average voltage, depth of discharge, highest battery temperature, lowest temperature, average temperature, ambient temperature in the energy storage power station, full charge, maximum current, minimum current, average current, maximum internal resistance, minimum internal resistance, average value of temperature x time, total duration of discharge, average duration of discharge; the internal resistance value, electrolyte concentration, battery capacity and SOH value after the kth charge and discharge are stored in a residual life value table.
Recording an event: specific events and severity levels, such as short circuits, leakage, over/under voltage, overload, over power, over/under frequency, etc., which may produce a certain decay in life are recorded and stored in a record event record table.
The final acquired feature variables include, but are not limited to, the following feature variables:
feature 1, initial Battery Capacity
Feature 2, initial internal resistance of battery
Feature 3, maximum voltage of kth charge-discharge, k=1, …, n
Characteristic 4, minimum voltage of kth charge-discharge, k=1, …, n
Characteristic 5, average voltage of kth charge-discharge, k=1, …, n
Characteristic 6, maximum current of kth charge-discharge, k=1, …, n
Feature 7, minimum current for kth charge-discharge, k=1, …, n
Characteristic 8, average current of kth charge-discharge, k=1, …, n
Feature 9, maximum temperature of kth charge-discharge, k=1, …, n
Characteristic 10, minimum temperature of kth charge-discharge, k=1, …, n
Characteristic 11, average temperature of kth charge-discharge, k=1, …, n
Characteristic 12, ambient temperature for kth charge-discharge, k=1, …, n
Feature 13, maximum depth of discharge for the first k discharges, k=1, …, n
Feature 14, maximum charge fullness for the first k charges, k=1, …, n
Feature 15, average value of temperature x time during the previous k charge and discharge, k=1, …, n
Feature 16, total duration of the first k discharges, k=1, …, n
Feature 17, average duration of the first k discharges, k=1, …, n
Characteristic 18, internal resistance of battery after kth charge-discharge, k=1, …, n
Characteristic 19, battery capacity value after kth charge-discharge, k=1, …, n
Characteristic 20, SOH value of battery after kth charge-discharge, k=1, …, n
Characteristic 21, the internal resistance of the battery in the kth charge and discharge is extremely poor, k=1, …, n
Characteristic 22, the battery capacity of the kth charge and discharge is extremely poor, k=1, …, n
Feature 23, electrolyte concentration in cell after kth charge-discharge, k=1, …, n
Feature 24, specific event severity level value
Step 2 is then performed.
Step 2: and (3) screening and obtaining effective feature variables by utilizing a correlation coefficient and variable goodness algorithm according to all the features in the step (1). The method comprises the following steps:
normalizing all characteristic data by using a min-max normalization function, and recording the normalized matrix as
，
Where n is the total number of samples and m1 is the total number of features.
And calculating the degree of correlation between the features by using pearson correlation coefficients, wherein the range of the correlation is [ -1,1], wherein positive values represent positive correlations, negative values represent negative correlations, and the larger the absolute value is, the stronger the correlation between the variables is.
Setting a correlation threshold Cr, when the absolute value of a correlation coefficient is larger than Cr, considering that the correlation between two variables is extremely strong, then calculating the merits of the two variables by utilizing a characteristic information value IV, deleting the variable with smaller IV value to reserve the more optimal variable, wherein the calculation formula of the IV is as follows:
，
wherein the method comprises the steps ofThe number of positive samples being the jth feature variable is +.>Is the j thThe number of negative examples of the individual feature variables is the ratio.
Step 3 is then carried out:
step 3: and (3) deleting multiple collinearity variables by continuously utilizing a variance expansion factor algorithm according to the effective characteristics obtained in the step (2). The method comprises the following steps:
assuming that m2 effective features are extracted through the step 2, the corresponding matrix is:
the collinearity between the variables is calculated using the variance-expansion factor VIF, as follows:
note the variance expansion factor of the j-th feature variable:。
wherein the method comprises the steps ofIs the regression determinable coefficient: by characteristic variables->The other independent variables are dependent variables, and the other independent variables are characteristic variables. The calculation formula is as follows: />。
Wherein the method comprises the steps ofIs a variable->Sum of squares of residuals>Is->Is the sum of the total squares of (c). />Is the predicted value of the ith sample of the jth feature (taking the jth feature as the dependent variable and the other features as the independent variable),/th feature>Is the sample average of the j-th feature.
A collinearity threshold variable Vmax is set. The VIF of all features is calculated first, and when the VIF > =vmax of a certain feature(s) exists, it is indicated that there is a serious multiple collinearity, and the feature in which the VIF is the largest is directly deleted. And then continuing to calculate the VIFs of all the features by using the residual features, and if the VIFs of some feature(s) =Vmax, continuing to delete the feature with the largest VIF. The loop is then stopped until the VIF of all the remaining features is less than Vmax (i.e., the remaining features are considered to have no multicollinearity).
Step 4 is then performed.
Step 4: and (3) continuously utilizing the stepwise regression model according to the effective characteristics obtained in the step (3) to obtain valuable characteristic variables, and simultaneously obtaining a residual battery capacity prediction model M1. The method comprises the following steps:
assuming that m3 effective features are extracted after deleting in the step 3, the matrix is:
the regression equation between the remaining battery capacity and all features is recorded as:
where Ck is the battery capacity after the kth charge and discharge,is characterized by->Beta is the decay rate and C0 is the initial battery capacity.
Step 4.1, first, each is individually processed by forward methodAdding the features into a regression model according to the linear regression equation, calculating the AIC value of the model after the addition, and setting an AIC change threshold valueWhen the AIC reduction value is greater than +.>The feature is added.
And 4.2, deleting the features of the model once in a backward way every time the number of newly added features reaches 3 and the total number of features exceeds 5. I.e. setting a significance thresholdAnd a change threshold->When the feature significance value is less than +.>Or AIC reduction value is less than +.>And eliminating the features that maximize the AIC value after feature reduction.
Step 4.3, loop 4.1 and 4.2 until no new features can be added and no features can be screened out.
And 4.4, obtaining the residual valuable features and a regression model M1 according to 4.1 to 4.3, and recording the extracted value feature matrix for measuring the battery health index as follows:
，
where m4 is the number of value features remaining after screening.
Step 4.5, obtaining the predicted value of the residual capacity of the battery by using the model M1 and storing the predicted value into a residual life value table, and obtaining the difference value of the residual capacity of the battery according to the predicted value and the actual value, and recording the difference value asThe model M1 is adjusted by using the difference, specifically as follows:
obtaining maximum variance modulus of each feature using battery capacity difference，
Wherein the method comprises the steps ofIs the difference between the predicted value and the actual value of the remaining capacity of the battery in the kth line record, +.>Is->Mean value of->Is->Is a variance of (c). Setting a reasonable data threshold interval: />The method comprises the steps of carrying out a first treatment on the surface of the Then traversing all data records (i.e. whether scanning each row of records is within a reasonable interval of data) if +.>And directly eliminating the data record if the data record is not in the corresponding reasonable interval.
S120, acquiring the residual battery capacity and the initial battery capacity of the battery corresponding to each record in the final characteristics, grouping the battery life corresponding to the records based on the ratio of the residual battery capacity to the initial battery capacity, taking the records as samples, taking the battery life group corresponding to the records as labels, and training a bidirectional LSTM neural network model to establish a battery life grouping model.
Specifically, training and obtaining a life grouping model M2 by continuously utilizing the bidirectional LSTM neural network model according to the final characteristic data obtained in the step 4, wherein the prediction label is 2: reliable life group, 1: median lifetime group and 0: low lifetime group.
Let n2 effective records extracted after deleting in step 4, the matrix is:
the batteries are first divided into reliable life groups (2 SOH according to the residual capacity of the batteries>=90%), median lifetime group (noted 1, 80%<=SOH<90%) and a low lifetime group (noted 0, SOH<80%) where soh=remaining battery capacity/initial battery capacity x100%, then the battery is classified according to the stage in which the battery history data is located, and stored in the remaining life value table. Based on the associated feature vectorsTraining tests were performed using a two-way LSTM neural network model: setting an expected accuracy target as P1, stopping training when the model accuracy exceeds a preset threshold P1, taking the obtained model as a life grouping model M2, if the model accuracy does not exceed the preset threshold, adding a training sample or adjusting training parameters of a neural network model, and retraining the model until the model accuracy exceeds the preset threshold.
S130, acquiring regression parameters by using a regression model based on each record in the battery life group and the residual battery capacity corresponding to the record, and establishing a residual battery capacity regression prediction model of the battery life group.
In particular, using feature vectorsAnd calculating by using a regression model according to the service life group to obtain regression parameters. The optimized residual battery capacity regression prediction models are built, namely M11 (reliable life group), M12 (median life group) and M13 (low life group).
And S140, respectively acquiring records of the battery in a fading state and a non-fading state in the final characteristics, taking the records as samples, taking the fading state of the battery corresponding to the records as a label, and training the GRU neural network model to establish a battery fading prediction model.
Specifically, the final feature obtained in the step 4 is continuously utilized to train a battery degradation prediction model M3 by using a GRU neural network model, and the prediction labels are as follows: 1: decay, 0: is not degenerated.
Preserving historical decline battery related featuresLine 1, ->(i.e. n/2 rounded up) and nth rows (n is the number of charge and discharge cycles, i.e. lifetime), (k=1,)>N), obtain: />
Wherein, line 1A record of behaving as if the battery were in a non-degraded state, a record of behaving as if the battery were in a degraded state,
according toTraining tests were performed using the GRU neural network model: firstly setting an expected accuracy target as P2, stopping training when the model accuracy exceeds a preset threshold P2, taking the obtained model as a battery degradation prediction model M3, if the model accuracy does not exceed the preset threshold, adding a training sample or adjusting training parameters of a neural network model, and retraining the model until the model accuracy exceeds the preset threshold.
In some embodiments of the present disclosure, a number threshold T for automatic model update is set, and when the number of degraded batteries exceeds T, new data is added to the original training set and retrained to obtain iteratively optimized models M1, M2, and M3.
Fig. 2 is a schematic illustration of an apparatus for training a battery life prediction model, according to some embodiments of the present disclosure. As shown in fig. 2, the apparatus 200 for training a battery life prediction model includes a feature extraction module 210, a battery life grouping model training module 220, a residual battery capacity regression prediction model building module 230, and a battery degradation prediction model training module 240. Wherein:
a feature extraction module 210, configured to collect historical test data of a plurality of battery capacity cyclic degradation, and extract final features reflecting battery degradation information from the historical test data;
a battery life grouping model training module 220, configured to obtain a remaining battery capacity and an initial battery capacity of the battery corresponding to each record in the final feature, group battery lives corresponding to the records based on a ratio of the remaining battery capacity to the initial battery capacity, use the records as samples, use the battery life group corresponding to the records as a label, and train a bidirectional LSTM neural network model to establish a battery life grouping model;
a remaining battery capacity regression prediction model establishing module 230, configured to obtain regression parameters with a regression model based on each record in the battery life group and the remaining battery capacity corresponding to the record, and establish a remaining battery capacity regression prediction model of the battery life group;
and the battery degradation prediction model training module 240 is configured to respectively obtain records of the battery in a degradation state and a non-degradation state in the final feature, train the GRU neural network model with the records as samples and the degradation state of the battery corresponding to the records as a label, so as to establish a battery degradation prediction model.
An embodiment of the present disclosure provides a method of predicting battery life, as shown in fig. 3, the method including:
s310, obtaining preset characteristic data of each single battery in the energy storage battery cluster, inputting the characteristic data into a battery life grouping model trained according to the method of S120, and grouping the single batteries in real time life.
And S320, inputting the characteristic data into a residual battery capacity regression prediction model corresponding to the service life group to predict the residual battery capacity of the single battery, wherein the residual battery capacity regression prediction model is established according to the method of S130.
And S330, if the ratio of the residual battery capacity to the initial battery capacity of the single battery is smaller than the second preset threshold, inputting the characteristic data into a battery degradation prediction model trained according to the method of S140 to predict whether the single battery is degraded, and determining a processing strategy for the single battery based on whether the single battery is degraded.
Specifically, for the energy storage battery cluster, the above life grouping model M2 is used to obtain real-time life groupings of the unit batteries, and then the regression prediction model M13 (assuming that the life groupings are low-life groups) under the corresponding groupings is used to predict the real-time remaining battery capacity of each unit battery of the battery cluster. If the remaining battery capacity exceeds 80%, no operation is performed. If the residual battery capacity is smaller than 80%, predicting whether the single battery is degenerated by using a battery degeneration prediction model M3, if the predicted result is degenerated, considering that the battery cannot be used continuously, automatically carrying out low-life red early warning by the system, and informing a worker of changing the battery immediately by a short message; if the predicted result is not declined, the system automatically performs yellow early warning and informs the staff of timely paying attention to or replacing the battery by a short message. After the battery is replaced, the battery is not directly abandoned, but is continuously charged and discharged in an experimental environment, a battery degradation capacity threshold C0 (which is determined according to practical conditions and is set to be 70% of the minimum battery capacity of the battery in normal operation) is preset, until a battery capacity predicted value is smaller than the threshold C0, the battery degradation is determined, and the final service life of the battery is L times of charging and discharging (if the last charging and discharging is the L times).
The fading prediction mainly considers that the capacity space of the battery is large when the battery is in a low-service-life group, and the residual battery capacity prediction is inaccurate, so that the fading prediction is supplemented to reduce the false alarm rate of low electric quantity, thereby fully utilizing the battery capacity and using the battery to the maximum extent.
In some embodiments of the present disclosure, to enable the remaining cycle life length of the entire battery cluster to be extended, we achieve equalization control by equalizing the discharge of the cells of the battery cluster. I.e. balanced scheduling according to SOH for each cell of the battery pack: for a battery with a larger SOH, discharging with a larger current to accelerate SOH decay), a battery with a smaller SOH, discharging with a smaller current to slow down SOH decay).
Fig. 4 is a schematic diagram of an apparatus for predicting battery life, according to some embodiments of the present disclosure. As shown in fig. 4, the apparatus 400 for predicting battery life includes a battery life grouping module 410, a remaining battery capacity prediction module 420, and a degradation prediction module 430. Wherein:
the battery life grouping module 410 is configured to obtain preset feature data of each unit battery in the energy storage battery cluster, input the feature data into a battery life grouping model trained according to the method of S120, and perform real-time life grouping on the unit batteries;
a remaining battery capacity prediction module 420, configured to input the feature data into a remaining battery capacity regression prediction model corresponding to the lifetime group to predict a remaining battery capacity of the unit battery, where the remaining battery capacity regression prediction model is established according to the method of S130;
and the degradation prediction module 430 is configured to, if the ratio of the remaining battery capacity and the initial battery capacity of the single battery is smaller than the second preset threshold, input the feature data into a battery degradation prediction model trained according to the method of S140 to predict whether the single battery is degraded, and determine a processing policy for the single battery based on whether the single battery is degraded.
In summary, according to the method for training the battery life prediction model, the method for predicting the battery life, the device for training the battery life prediction model and the device for predicting the battery life provided by the embodiments of the present disclosure, the regression model and the deep learning model are fused, the optimal characteristic variable is screened out objectively and effectively by using the co-linearity screening and stepwise regression, then the stage of the energy storage battery in the life cycle is estimated by using the battery life grouping model based on the optimal characteristic variable, the regression model corresponding to the stage is selected to predict the remaining battery capacity of the energy storage battery, finally the degradation prediction model is used to estimate whether the remaining battery capacity is close to the threshold value, and the accurate and precise prediction of the life of the energy storage battery in each stage of the life cycle of the energy storage battery can be realized by the combination of 3 prediction models. Meanwhile, the regression model is respectively built based on different stages of the life cycle, and feature dimensions required by training of the battery life grouping model and the regression prediction model are greatly reduced, so that the calculation complexity is greatly reduced.
It will be clear to those skilled in the art that, for convenience and brevity of description, the specific operation of the apparatus and modules described above may refer to the corresponding description in the foregoing apparatus embodiments, which is not repeated here.
While the subject matter described herein is provided in the general context of operating systems and application programs that execute in conjunction with the execution of a computer system, those skilled in the art will recognize that other implementations may also be performed in combination with other types of program modules. Generally, program modules include routines, programs, components, data structures, and other types of structures that perform particular tasks or implement particular abstract data types. Those skilled in the art will appreciate that the subject matter described herein may be practiced with other computer system configurations, including hand-held devices, multiprocessor systems, microprocessor-based or programmable consumer electronics, minicomputers, mainframe computers, and the like, as well as distributed computing environments that have tasks performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote memory storage devices.
Those of ordinary skill in the art will appreciate that the elements and method steps of the examples described in connection with the embodiments disclosed herein can be implemented as electronic hardware, or as a combination of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present disclosure.
It is to be understood that the above-described embodiments of the present disclosure are merely illustrative or explanatory of the principles of the disclosure and are not restrictive of the disclosure. Accordingly, any modifications, equivalent substitutions, improvements, or the like, which do not depart from the spirit and scope of the present disclosure, are intended to be included within the scope of the present disclosure. Furthermore, the appended claims of this disclosure are intended to cover all such changes and modifications that fall within the scope and boundary of the appended claims, or the equivalents of such scope and boundary.
Claims (10)
1. A method of training a battery life prediction model, comprising:
collecting historical test data of a plurality of battery capacity cycle degenerations, and screening out final characteristics reflecting battery degeneration information from the historical test data;
acquiring the residual battery capacity and the initial battery capacity of the battery corresponding to each record in the final characteristics, grouping the battery life corresponding to the records based on the ratio of the residual battery capacity to the initial battery capacity, taking the records as samples, taking the battery life group corresponding to the records as a label, and training a bidirectional LSTM neural network model to establish a battery life grouping model;
based on each record in the battery life group and the residual battery capacity corresponding to the record, acquiring regression parameters by using a regression model, and establishing a residual battery capacity regression prediction model of the battery life group;
and respectively acquiring records of the battery in a fading state and a non-fading state in the final characteristics, taking the records as samples, taking the fading state of the battery corresponding to the records as a label, and training the GRU neural network model to establish a battery fading prediction model.
2. The method of claim 1, wherein collecting historical test data for a plurality of battery capacity cycle degradations comprises:
according to historical declining experimental single battery data, obtaining actual measurement values of the battery characteristic variables after the previous k times of charge and discharge according to preset time intervals, wherein k is a natural number not greater than n, and n is the total cycle number of charge and discharge after the battery is declined;
and recording preset events and corresponding severity levels, wherein the preset events are events which can lead to the degradation of the service life of the battery.
3. The method of claim 1, wherein the screening out final features reflecting battery degradation information in the historical test data comprises:
normalizing the measured values of the battery characteristic variables, calculating the correlation between the characteristic variables, reserving the characteristic variables with larger characteristic information values between the two characteristic variables with the correlation larger than a preset threshold value so as to obtain a first effective characteristic variable, and deleting multiple co-linear variables from the first effective characteristic variable so as to obtain a second effective characteristic variable;
and extracting final characteristics reflecting battery degradation information from the second effective characteristic variables by using a stepwise regression model.
4. The method of claim 1, wherein the grouping the battery life corresponding to the record based on the ratio of the remaining battery capacity and the initial battery capacity comprises:
if the ratio of the remaining battery capacity of the battery corresponding to the record to the initial battery capacity is not less than a first preset threshold value, grouping the battery life corresponding to the record into a reliable life group;
if the ratio of the remaining battery capacity of the battery corresponding to the record to the initial battery capacity is smaller than the first preset threshold value and not smaller than a second preset threshold value, grouping the battery life corresponding to the record into a median life group;
and if the ratio of the remaining battery capacity of the battery corresponding to the record to the initial battery capacity is smaller than the second preset threshold value, grouping the battery life corresponding to the record into a low-order life group.
5. The method of claim 1, wherein the separately obtaining records of battery in the final feature in a degraded state and in a non-degraded state comprises:
acquiring the first and second time corresponding to the battery in the final featureRecording after secondary charge and discharge, wherein n is the total cycle number of charge and discharge after battery decay;
and acquiring a record corresponding to the nth charge and discharge of the battery in the final characteristic.
6. The method according to claim 1, wherein the method further comprises:
and when the number of the degraded batteries exceeds a preset threshold, updating the final characteristics based on the characteristic data of the degraded batteries, and updating the battery life grouping model, the residual battery capacity regression prediction model and the battery degradation prediction model based on the updated final characteristics.
7. An apparatus for training a battery life prediction model, comprising:
the characteristic extraction module is used for collecting historical test data of the cyclic degradation of the capacity of the plurality of batteries and screening out final characteristics reflecting the degradation information of the batteries from the historical test data;
the battery life grouping model training module is used for acquiring the residual battery capacity and the initial battery capacity of the battery corresponding to each record in the final characteristics, grouping the battery life corresponding to the record based on the ratio of the residual battery capacity to the initial battery capacity, taking the record as a sample, taking the battery life group corresponding to the record as a label, and training a bidirectional LSTM neural network model to establish a battery life grouping model;
the residual battery capacity regression prediction model building module is used for obtaining regression parameters by using a regression model based on each record in the battery life group and the residual battery capacity corresponding to the record, and building the residual battery capacity regression prediction model of the battery life group;
and the battery degradation prediction model training module is used for respectively acquiring records of the battery in a degradation state and a non-degradation state in the final characteristics, taking the records as samples, taking the degradation state of the battery corresponding to the records as a label, and training the GRU neural network model to establish the battery degradation prediction model.
8. A method of predicting battery life comprising:
acquiring preset characteristic data of each single battery in an energy storage battery cluster, inputting the characteristic data into a battery life grouping model trained according to the method of claim 1, and grouping the single batteries in real time life;
inputting the characteristic data into a residual battery capacity regression prediction model corresponding to the life grouping to predict the residual battery capacity of the single battery, wherein the residual battery capacity regression prediction model is established according to the method of claim 1;
if the ratio of the remaining battery capacity and the initial battery capacity of the single battery is smaller than the second preset threshold value, inputting the characteristic data into a battery degradation prediction model trained according to the method of claim 1 to predict whether the single battery is degraded or not, and determining a processing strategy for the single battery based on whether the single battery is degraded or not.
9. The method of claim 8, wherein the determining a processing policy for the battery based on whether the battery is degraded comprises:
if the ratio of the remaining battery capacity to the initial battery capacity of the single battery is smaller than the second preset threshold value and the single battery is degenerated, triggering a low-life red early warning and a single battery replacement operation corresponding to the red early warning;
and if the ratio of the residual battery capacity to the initial battery capacity of the single battery is smaller than the second preset threshold value and the single battery is not degenerated yet, triggering a low-life yellow early warning and a single battery maintenance operation corresponding to the yellow early warning.
10. An apparatus for predicting battery life, comprising:
the battery life grouping module is used for acquiring preset characteristic data of each single battery in the energy storage battery cluster, inputting the characteristic data into the battery life grouping model trained according to the method of claim 1, and grouping the single batteries in real time life;
a remaining battery capacity prediction module for inputting the feature data into a remaining battery capacity regression prediction model corresponding to the lifetime group to predict the remaining battery capacity of the single battery, wherein the remaining battery capacity regression prediction model is established according to the method of claim 1;
and the degradation prediction module is used for inputting the characteristic data into a battery degradation prediction model trained according to the method of claim 1 to predict whether the single battery is degraded or not if the ratio of the residual battery capacity to the initial battery capacity of the single battery is smaller than the second preset threshold value, and determining a processing strategy for the single battery based on whether the single battery is degraded or not.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311041364.XA CN116774057B (en) | 2023-08-18 | 2023-08-18 | Method and device for training battery life prediction model and predicting battery life |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311041364.XA CN116774057B (en) | 2023-08-18 | 2023-08-18 | Method and device for training battery life prediction model and predicting battery life |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116774057A true CN116774057A (en) | 2023-09-19 |
| CN116774057B CN116774057B (en) | 2023-11-14 |
Family
ID=88011927
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311041364.XA Active CN116774057B (en) | 2023-08-18 | 2023-08-18 | Method and device for training battery life prediction model and predicting battery life |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116774057B (en) |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130297244A1 (en) * | 2011-02-28 | 2013-11-07 | Mitsubishi Heavy Industries, Ltd. | Secondary battery lifetime prediction apparatus, battery system and secondary battery lifetime prediction method |
| CN104656021A (en) * | 2013-11-19 | 2015-05-27 | 艾默生网络能源有限公司 | Method and device for predicting remaining capacity of battery |
| CN110161425A (en) * | 2019-05-20 | 2019-08-23 | 华中科技大学 | A kind of prediction technique of the remaining life divided based on lithium battery catagen phase |
| CN111880099A (en) * | 2020-06-30 | 2020-11-03 | 国网浙江省电力有限公司电力科学研究院 | Method and system for predicting service life of battery monomer in energy storage power station |
| CN112084459A (en) * | 2019-06-12 | 2020-12-15 | 电计贸易(上海)有限公司 | Method and device for predicting battery charge-discharge cycle life, electronic terminal and storage medium |
| CN112180258A (en) * | 2019-07-01 | 2021-01-05 | 电计贸易(上海)有限公司 | Method, device, medium, terminal and system for measuring average coulomb efficiency of battery |
| US20210190877A1 (en) * | 2017-11-16 | 2021-06-24 | Semiconductor Energy Laboratory Co., Ltd. | Lifetime estimation device, lifetime estimation method, and abnormality detection method of secondary battery |
| CN114861527A (en) * | 2022-04-15 | 2022-08-05 | 南京工业大学 | Lithium battery life prediction method based on time series characteristics |
| US11527786B1 (en) * | 2022-03-28 | 2022-12-13 | Eatron Technologies Ltd. | Systems and methods for predicting remaining useful life in batteries and assets |
| US20230029810A1 (en) * | 2021-07-27 | 2023-02-02 | Honda Motor Co., Ltd. | Battery model construction method and battery degradation prediction device |
| CN115856678A (en) * | 2022-11-10 | 2023-03-28 | 盐城工学院 | Lithium ion battery health state estimation method |
| CN116466236A (en) * | 2023-02-27 | 2023-07-21 | 北京海博思创科技股份有限公司 | Battery remaining life prediction method, device, equipment and readable storage medium |
-
2023
- 2023-08-18 CN CN202311041364.XA patent/CN116774057B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130297244A1 (en) * | 2011-02-28 | 2013-11-07 | Mitsubishi Heavy Industries, Ltd. | Secondary battery lifetime prediction apparatus, battery system and secondary battery lifetime prediction method |
| CN104656021A (en) * | 2013-11-19 | 2015-05-27 | 艾默生网络能源有限公司 | Method and device for predicting remaining capacity of battery |
| US20210190877A1 (en) * | 2017-11-16 | 2021-06-24 | Semiconductor Energy Laboratory Co., Ltd. | Lifetime estimation device, lifetime estimation method, and abnormality detection method of secondary battery |
| CN110161425A (en) * | 2019-05-20 | 2019-08-23 | 华中科技大学 | A kind of prediction technique of the remaining life divided based on lithium battery catagen phase |
| CN112084459A (en) * | 2019-06-12 | 2020-12-15 | 电计贸易(上海)有限公司 | Method and device for predicting battery charge-discharge cycle life, electronic terminal and storage medium |
| CN112180258A (en) * | 2019-07-01 | 2021-01-05 | 电计贸易(上海)有限公司 | Method, device, medium, terminal and system for measuring average coulomb efficiency of battery |
| CN111880099A (en) * | 2020-06-30 | 2020-11-03 | 国网浙江省电力有限公司电力科学研究院 | Method and system for predicting service life of battery monomer in energy storage power station |
| US20230029810A1 (en) * | 2021-07-27 | 2023-02-02 | Honda Motor Co., Ltd. | Battery model construction method and battery degradation prediction device |
| US11527786B1 (en) * | 2022-03-28 | 2022-12-13 | Eatron Technologies Ltd. | Systems and methods for predicting remaining useful life in batteries and assets |
| CN114861527A (en) * | 2022-04-15 | 2022-08-05 | 南京工业大学 | Lithium battery life prediction method based on time series characteristics |
| CN115856678A (en) * | 2022-11-10 | 2023-03-28 | 盐城工学院 | Lithium ion battery health state estimation method |
| CN116466236A (en) * | 2023-02-27 | 2023-07-21 | 北京海博思创科技股份有限公司 | Battery remaining life prediction method, device, equipment and readable storage medium |
Non-Patent Citations (3)
| Title |
|---|
| LINXING XIE SCHOOL OF ELECTRICAL ENGINEERING AND INFORMATION, SOUTHWEST PETROLEUM UNIVERSITY, CHENGDU, CHINA: "Remaining Useful Life Prediction of Lithium Batteries Based on CNN-GRU Model", 2022 IEEE 6TH CONFERENCE ON ENERGY INTERNET AND ENERGY SYSTEM INTEGRATION (EI2) * |
| 苏来锁: "多应力作用下能量型锂离子电池的老化行为研究", 中国优秀硕士学位论文全文数据库 (工程科技Ⅱ辑), pages 042 - 1223 * |
| 顾泰宇: "基于数据扩展的月度光伏电量组合预测研究", 中国优秀硕士学位论文全文数据库 (工程科技Ⅱ辑), pages 042 - 1113 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116774057B (en) | 2023-11-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111584952B (en) | Method and system for online evaluation of electrochemical cells of energy storage power station | |
| CN110161425B (en) | Method for predicting remaining service life based on lithium battery degradation stage division | |
| JP5203394B2 (en) | Expert diagnostic method for analyzing storage battery performance. | |
| CN112731159A (en) | Method for pre-judging and positioning battery fault of battery compartment of energy storage power station | |
| CN112798960B (en) | Battery pack residual life prediction method based on migration deep learning | |
| CN111680848A (en) | Battery life prediction method based on prediction model fusion and storage medium | |
| CN110535159B (en) | Method and system for early warning of faults of operation units of large-scale energy storage power station | |
| CN109406929B (en) | On-line monitoring alarm device for open circuit in storage battery pack of transformer substation | |
| CN112255559A (en) | Method for predicting residual life of lithium battery energy storage power station based on multiple linear regression | |
| CN114035098A (en) | Lithium battery health state prediction method integrating future working condition information and historical state information | |
| CN116113961A (en) | Battery life prediction using machine learning model | |
| CN112098876A (en) | Method for detecting abnormality of single battery in storage battery | |
| CN116401585B (en) | Energy storage battery failure risk assessment method based on big data | |
| CN104316879A (en) | Method for forecasting service life of lead-acid storage battery | |
| CN113687255A (en) | Method and device for diagnosing state of battery cell and storage medium | |
| CN111669123A (en) | Method and device for fault diagnosis of photovoltaic string | |
| CN111460656B (en) | Method and system for evaluating operation life of communication power supply of electric power machine room | |
| EP4235198A1 (en) | Apparatus and method for managing battery | |
| US20230273262A1 (en) | Battery capacity estimation apparatus and method | |
| US20220099752A1 (en) | Method and Apparatus for Machine-Individual Improvement of the Lifetime of a Battery in a Battery-Operated Machine | |
| CN116488219B (en) | Distributed energy storage acquisition strategy control method based on multi-level architecture | |
| CN116774057B (en) | Method and device for training battery life prediction model and predicting battery life | |
| CN117310538A (en) | Energy storage battery electric quantity monitoring system capable of automatically detecting electric quantity residual conversion efficiency | |
| CN116736166A (en) | Battery cell abnormality detection method and device of battery pack and battery pack | |
| CN116754967A (en) | Method and system for online evaluation of electrochemical cells of an energy storage power station |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |