CN116008844A - Vehicle lithium battery health state online prediction method adapting to quick charge strategy - Google Patents
Vehicle lithium battery health state online prediction method adapting to quick charge strategy Download PDFInfo
- Publication number
- CN116008844A CN116008844A CN202310020308.1A CN202310020308A CN116008844A CN 116008844 A CN116008844 A CN 116008844A CN 202310020308 A CN202310020308 A CN 202310020308A CN 116008844 A CN116008844 A CN 116008844A
- Authority
- CN
- China
- Prior art keywords
- battery
- model
- super
- online
- lithium battery
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 95
- 230000036541 health Effects 0.000 title claims abstract description 59
- WHXSMMKQMYFTQS-UHFFFAOYSA-N Lithium Chemical compound [Li] WHXSMMKQMYFTQS-UHFFFAOYSA-N 0.000 title claims abstract description 53
- 229910052744 lithium Inorganic materials 0.000 title claims abstract description 53
- 238000012549 training Methods 0.000 claims abstract description 38
- 238000006731 degradation reaction Methods 0.000 claims abstract description 23
- 230000015556 catabolic process Effects 0.000 claims abstract description 22
- 238000013526 transfer learning Methods 0.000 claims abstract description 13
- 238000011161 development Methods 0.000 claims abstract description 12
- 238000005457 optimization Methods 0.000 claims description 30
- 230000004927 fusion Effects 0.000 claims description 24
- 210000004027 cell Anatomy 0.000 claims description 15
- 230000008569 process Effects 0.000 claims description 15
- 238000000605 extraction Methods 0.000 claims description 14
- 238000012795 verification Methods 0.000 claims description 13
- 238000002790 cross-validation Methods 0.000 claims description 10
- 230000015654 memory Effects 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 6
- 238000013508 migration Methods 0.000 claims description 6
- 230000005012 migration Effects 0.000 claims description 6
- 239000000178 monomer Substances 0.000 claims description 6
- 230000003044 adaptive effect Effects 0.000 claims description 5
- 210000002569 neuron Anatomy 0.000 claims description 5
- 238000005259 measurement Methods 0.000 claims description 4
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000012544 monitoring process Methods 0.000 claims description 3
- 230000000306 recurrent effect Effects 0.000 claims description 3
- 230000003862 health status Effects 0.000 claims 2
- 230000032683 aging Effects 0.000 description 15
- 238000012360 testing method Methods 0.000 description 11
- 230000018109 developmental process Effects 0.000 description 8
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 6
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 6
- 238000007599 discharging Methods 0.000 description 6
- 230000006399 behavior Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 238000011156 evaluation Methods 0.000 description 4
- 238000004519 manufacturing process Methods 0.000 description 4
- 238000010200 validation analysis Methods 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- HBBGRARXTFLTSG-UHFFFAOYSA-N Lithium ion Chemical compound [Li+] HBBGRARXTFLTSG-UHFFFAOYSA-N 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 125000004122 cyclic group Chemical group 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 229910001416 lithium ion Inorganic materials 0.000 description 2
- 230000007787 long-term memory Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 230000006403 short-term memory Effects 0.000 description 2
- 101100314406 Saccharomyces cerevisiae (strain ATCC 204508 / S288c) TPS1 gene Proteins 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000013401 experimental design Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- GELKBWJHTRAYNV-UHFFFAOYSA-K lithium iron phosphate Chemical compound [Li+].[Fe+2].[O-]P([O-])([O-])=O GELKBWJHTRAYNV-UHFFFAOYSA-K 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000284 resting effect Effects 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
Images
Landscapes
- Secondary Cells (AREA)
Abstract
The invention relates to the technical field of lithium batteries, in particular to an on-line prediction method of the health state of a vehicle lithium battery adapting to a quick charge strategy. The method consists of two parts, namely offline model development and online application of a multitask model; the off-line model development part comprises super-parameter adjustment and model training, the multi-task model on-line application part comprises model transfer learning, the target application single body is measured on line during the model transfer learning, the voltage characteristics of the private attributes of the battery and the circulation times of the common attributes of the battery are obtained, the voltage characteristics and the circulation times are fused, and then the characteristics are input into the trained off-line model for prediction, and the battery capacity under different degradation modes is predicted on line. The method for online predicting the health state of the lithium battery for the vehicle, which is suitable for various quick charging strategies, can be used for quickly obtaining the lithium battery on line by introducing health features, is suitable for the quick charging strategies, and can be used for online predicting the battery capacity under different degradation modes.
Description
Technical Field
The invention relates to the technical field of lithium batteries, in particular to an on-line prediction method of the health state of a vehicle lithium battery adapting to a quick charge strategy.
Background
At present, in order to meet the fast pace of life and work of people, a new energy automobile generally adopts a fast charging strategy, namely, a battery is charged to a certain available electric quantity in stages with constant high-rate current, for example, the battery is charged to 80% State of Charge (SOC) in 10 minutes.
The current method for predicting the battery capacity under different degradation modes mainly comprises an electrochemical model, an empirical model and a data driving model. Problems with existing methods include the following:
(1) The equivalent circuit model and the experience model of the lithium battery have low prediction precision and poor adaptability, and the electrochemical model is too complex to solve on line;
(2) The data driving method based on machine learning requires a large amount of battery aging data, and the behavior characteristics of the battery aging process are analyzed and obtained from the battery aging data;
(3) The extracted features are poor in practicality and difficult to adapt to actual working conditions, the extracted features in the discharging stage are not matched with the actual working conditions, and the extracted features in the charging stage are not suitable for a rapid charging strategy.
(4) The manufacturing and using conditions and other factors cause the difference among lithium battery monomers, and it is difficult to ensure enough adaptability and generalization capability of the prediction model.
Disclosure of Invention
In view of the above, the invention provides an on-line prediction method of the health state of the vehicle lithium battery adapting to a quick charging strategy, which is used for extracting the characteristics of the lithium ion battery under the condition of quick charging and self-adapting on-line capacity prediction modeling, and improves the adaptability and the accuracy of on-line prediction.
In order to achieve the above purpose, the present invention provides the following technical solutions:
in the first aspect, the invention provides an on-line prediction method of the health state of a vehicle lithium battery adapting to a quick charge strategy, which consists of two parts of off-line model development and on-line application of a multi-task model; the off-line model development part comprises super-parameter adjustment and model training, and the multi-task model on-line application part comprises model transfer learning;
the method for predicting the health state of the lithium battery for the vehicle on line comprises the following steps:
super-parameter adjustment: the super-parameters are adjusted by using high-similarity and low-similarity tasks with different degradation modes in a source domain, and the public area is found through multi-task cross-validation optimization to obtain the optimal super-parameters;
model training: training an offline model by taking complete life cycle data of part of batteries in a target domain as source domain data, and transmitting the trained offline model to an online application program;
Model transfer learning: and carrying out online measurement on the target application monomer to obtain the voltage characteristics of the private attribute of the battery and the cycle times of the common attribute of the battery, carrying out characteristic fusion on the voltage characteristics and the cycle times, and then inputting the fused characteristics into a trained offline model to predict, and carrying out online prediction on the battery capacity under different degradation modes.
As a further scheme of the invention, in the online application program, a long-short-term memory network is adopted to keep sharing attributes, and a full connection layer is adjusted to adapt to the difference between batteries of new battery data.
As a further aspect of the present invention, the method for online predicting a health state of a lithium battery for a vehicle further includes: quantitatively evaluating the model performance by using the root mean square error and the average absolute percentage error;
wherein, root mean square error is:
the average absolute percentage error is:
where n is the total number of cycles of the online prediction.
As a further scheme of the invention, the LSTM-FC network architecture of the long-short-period memory network and the full-connection layer provided in the online application program consists of an input layer, an LSTM layer, an FC layer and an output layer; the long-short-term memory network (LSTM) layer is composed of several recurrent neurons, each of which is judged by forgetting gate, input gate and output gate to determine the importance of the input information.
As a further scheme of the invention, the forgetting door is as follows: f (f) t ＝σ(w f ·[h t-1 ,x t ]+b f )；
Wherein σ is a sigmoid activation function, w f And b f Weight matrix and bias, x, respectively representing forgetting gate f t For input, h t-1 For the hidden layer output state, the subscript t represents the current time step;
the input door is:
i t ＝σ(w i ·[h t-1 ,x t ]+b i )
wherein, tan h is hyperbolic tangent activation function,
is a candidate state, C t Is the current cell state, (:) represents the element multiplication operator;
the output door is:
o t ＝σ(w o ·[h t-1 ,x t ]+b C )
h t ＝o t *tanh(C t )。
as a further aspect of the invention, the FC layer of the LSTM-FC network architecture is also used to learn a specific from predicted cellsCharacteristics, final predicted capacity of time step t
The calculation formula of (2) is as follows:
as a further scheme of the invention, the on-line prediction method of the state of health of the lithium battery for the vehicle further comprises battery feature extraction, wherein the battery feature extraction comprises the following steps:
charging the battery, and recording the remaining capacity Q of the battery at the beginning of charging r And starts to calculate the charge capacity Q based on ampere integral c ，Q c ＝∫ 0 Idt, wherein I is charging current, and t is charging time;
when the residual current Q r And charge capacity Q c Sum Q r +Q c Greater than or equal to the set quick charge capacity requirement Q fc When the charging is stopped briefly and the terminal voltage data V of the battery is collected and recorded in a monitoring mode T ；
When the short stop charging time reaches the set value T r When the battery is charged or works continuously according to the original charging mode;
from collecting recorded terminal voltage data V T In the process, the time just reaching the set time T is obtained r Voltage value V of (2) Tr The V is Tr The value is the extracted characteristic value.
As a further aspect of the present invention, feature fusion is performed on the voltage feature and the cycle number, and the method further includes:
the historically extracted voltage signature v at the current cycle t t-l ,v t-l+1 ,…,v t ]And cycle number Cn t Fusion is carried out through a sliding window;
adjusting the voltage characteristic v by adjusting the window width l+1 t-l ,v t-l+1 ,…,v t ]And cycle number Cn t The proportion of the attribute features changes the feature information input by the model.
As a further scheme of the invention, the optimal super-parameters are obtained by finding a public area through multi-task cross-validation optimization, and the method also comprises model super-parameter closed-loop optimization, wherein the super-parameter closed-loop optimization method comprises the following steps:
distributing cross-validated task sample combinations, in the source domain data sample set { S }, distributing task sample combinations { TSSi, VSi } according to sample similarity j Wherein TSS is the training sample set and VS is the verification sample set;
randomly initializing all super parameters to be optimized in a section to be optimized;
Selecting a group of super parameters to execute gridding search in the interval to be optimized, respectively carrying out cross verification according to the task combination of high similarity and low similarity, and recording the super parameter value corresponding to the minimum verification error;
replacing the corresponding initial value of random initialization with the super-parameter value, and executing gridding search again to obtain the optimal super-parameter in the whole searching range;
after all the super parameters to be optimized are searched at least once, and when all the super parameters are met, stopping the program, and obtaining the optimal super parameter value; otherwise, replacing the corresponding super parameters with the recommended values, selecting the next group to perform the optimization process again, and iterating until all the super parameters are met.
As a further aspect of the present invention, the method further includes model offline training and online prediction before online predicting the battery capacity in different degradation modes, including:
constructing a network model according to the obtained model optimization parameters, using battery data circularly aged to fail in a source domain to train the network model offline, and migrating and integrating the trained network model into a vehicle-mounted battery management system;
extracting voltage characteristics and carrying out characteristic fusion, fine-tuning a migration integrated network model by utilizing the characteristics and capacity data of the existing history records, predicting the capacity health state of the battery by taking the fusion characteristics of the network model after fine-tuning based on the current moment as input, and inputting the model after the sliding window moves forward and fuses the extracted characteristics and the cycle number, so that the health state of the battery can be predicted.
The technical scheme provided by the invention has the following beneficial effects:
according to the on-line prediction method for the state of health of the vehicle lithium battery adapting to the quick charge strategy, a new health feature is introduced from the battery charging stage, namely, the terminal voltage of the battery is quickly charged to 80% of SOC in about 10 minutes, so that the battery capacity is indicated.
The invention provides an on-line prediction method of the state of health of a vehicle lithium battery adapting to a fast charge strategy, and also provides a feature fusion scheme for improving the adaptability of a prediction model, wherein voltage features showing specific battery private properties and cycle times showing common properties of all batteries are fused through a sliding window technology. The proportion of the two attribute features can be adjusted by adjusting the window width, so that the feature information input by the follow-up model is changed.
The invention provides an on-line prediction method of the health state of a vehicle lithium battery adapting to a fast charge strategy, and also provides a self-adaptive multi-task prediction modeling scheme and an optimization strategy based on transfer learning, so as to solve the problems of insufficient model adaptability and generalization capability caused by the difference between a source domain and a target domain. The optimal model is obtained by using a high-similarity and low-similarity task closed-loop optimization method, so that the battery capacity under different degradation modes can be predicted on line.
These and other aspects of the invention will be more readily apparent from the following description of the embodiments. It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention as claimed.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following description will briefly introduce the drawings that are needed in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are only some embodiments of the present invention. In the drawings:
fig. 1 is a flow chart of adaptive online capacity prediction of a lithium battery based on transfer learning in the present invention.
Fig. 2 is a schematic diagram of a voltage characteristic extraction process in an on-line prediction method of a vehicle lithium battery health state according to the present invention.
Fig. 3 is a schematic diagram of sliding window-based feature fusion in an on-line prediction method of a vehicle lithium battery health state adapted to a fast charging strategy.
Fig. 4 is a flowchart of ultra-parameter closed-loop optimization adjustment in an on-line prediction method of the vehicle lithium battery health state, which is adaptive to a fast charging strategy.
Fig. 5 is a schematic diagram of a model development process in an online prediction method of the state of health of a lithium battery for a vehicle, which is adaptive to a fast charging strategy.
Fig. 6-11 are graphs of capacity degradation of lithium battery data sets under different fast charging conditions in an on-line prediction method for a vehicle lithium battery state of health adapted to a fast charging strategy according to the present invention.
Fig. 12 is a graph showing a relationship DV between voltage and discharge capacity and battery aging in an on-line prediction method of a vehicle lithium battery state of health adapted to a fast charge strategy according to the present invention.
Fig. 13 is a graph I C showing the relationship between voltage and discharge capacity and battery aging in an on-line prediction method of the state of health of a lithium battery for a vehicle adapted to a fast charge strategy according to the present invention.
Fig. 14 is a capacity prediction performance chart of case 1 under a first-order fast charge condition in an on-line prediction method of a vehicle lithium battery state of health adapted to a fast charge policy according to the present invention.
Fig. 15 is a capacity prediction performance chart of case 2 under a first-order fast charge condition in an on-line prediction method of a vehicle lithium battery state of health adapted to a fast charge strategy according to the present invention.
Fig. 16 is a capacity prediction performance chart of case 3 under a second-order fast charge condition in an on-line prediction method for a vehicle lithium battery state of health adapted to a fast charge strategy according to the present invention.
Fig. 17 is a capacity prediction performance diagram of case 4 under a second-order fast charge condition in an on-line prediction method for a vehicle lithium battery state of health adapted to a fast charge policy according to the present invention.
Fig. 18 is a capacity prediction performance chart of case 5 under the condition of cross-checking and fast charging in the on-line prediction method of the lithium battery health state for the vehicle adapting to the fast charging strategy.
Fig. 19 is a capacity prediction performance chart of case 6 under the condition of cross-checking and fast charging in the on-line prediction method of the lithium battery health state for the vehicle adapting to the fast charging strategy.
Fig. 20 is a summary diagram of prediction errors of 6 cases in an online prediction method of a vehicle lithium battery health state adapting to a fast charge strategy according to the present invention.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
Today, in order to meet the fast pace of life and work of people, a new energy automobile generally adopts a fast charging strategy, i.e. a battery is charged to a certain available electric quantity in stages with constant high-rate current, for example, a State of Charge (SOC) of 80% in 10 minutes is realized. Based on the background, the invention provides a lithium ion battery characteristic extraction and self-adaptive online capacity prediction modeling method and an optimization scheme under the condition of rapid charge aiming at the remarkable difference of aging courses of battery monomers caused by the conditions of battery manufacturing tolerance, rapid charge, large current discharge and the like.
Technical solutions in exemplary embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in exemplary embodiments of the present invention, and it is apparent that the described exemplary embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to fall within the scope of the invention.
(1) Feature practicality and applicability problems: the health feature extraction of the battery needs to meet various use conditions and current popular fast charging strategies, and the behavior characteristics of the battery in the aging process can be quickly and online obtained and accurately reflected.
The invention introduces a new health feature from the battery charging stage, namely the terminal voltage of the battery when the battery is rapidly charged to 80% SOC for about 10 minutes, so as to indicate the battery capacity.
(2) Characteristic information problem: the characteristic information input by the prediction model is to embody the nonlinear aging behavior of the battery on one hand, and the inconsistent difference among the battery monomers on the other hand, so that the follow-up model is ensured to adapt to the prediction of different batteries.
The invention provides a feature fusion scheme for improving the adaptability of a prediction model. The voltage characteristics showing the private attributes of the specific batteries and the circulation times showing the common attributes of all batteries are fused through a sliding window technology. The proportion of the two attribute features can be adjusted by adjusting the window width, so that the feature information input by the follow-up model is changed.
(3) Model adaptability and generalization capability problems caused by source domain and target domain differences: because the prediction model is trained by using source domain data, the prediction precision of the trained model in a target domain depends on the difference between the source domain and the target domain, and when the difference is overlarge, the prediction performance of the model is difficult to meet.
The invention provides a self-adaptive multitask prediction modeling scheme and a training strategy based on transfer learning. The optimal model is obtained by using a high-similarity and low-similarity task closed-loop optimization method, so that the battery capacity under different degradation modes can be predicted on line.
Specifically, embodiments of the present application are further described below with reference to the accompanying drawings.
Referring to fig. 1, one embodiment of the present invention provides an on-line prediction method for a lithium battery state of health for a vehicle, which is adapted to a fast charging strategy, and comprises two parts, namely an off-line model development and a multi-task model on-line application; the off-line model development part comprises super-parameter adjustment and model training, and the multi-task model on-line application part comprises model transfer learning, and main key technologies of the technical scheme are described in detail below.
The method for predicting the health state of the lithium battery for the vehicle on line comprises the following steps:
super-parameter adjustment: the super-parameters are adjusted by using high-similarity and low-similarity tasks with different degradation modes in a source domain, and the public area is found through multi-task cross-validation optimization to obtain the optimal super-parameters;
model training: training an offline model by taking complete life cycle data of part of batteries in a target domain as source domain data, and transmitting the trained offline model to an online application program;
model transfer learning: and carrying out online measurement on the target application monomer to obtain the voltage characteristics of the private attribute of the battery and the cycle times of the common attribute of the battery, carrying out characteristic fusion on the voltage characteristics and the cycle times, and then inputting the fused characteristics into a trained offline model to predict, and carrying out online prediction on the battery capacity under different degradation modes.
In this embodiment, referring to FIG. 1, the offline model development includes three steps. In step 1, the hyper-parameters are adjusted by using high-similarity and low-similarity tasks with different degradation modes in the source domain, and the optimal hyper-parameters are obtained by finding a common area through multi-task cross-validation optimization. In step 2, the full life cycle data of the partial cells in the target domain is used for offline model training. After good training, the offline model is sent to the online application. In the online application of step 3, a long short-term memory (LSTM) is 'frozen' to maintain shared attributes and a Full Connected (FC) layer is trimmed to accommodate inter-cell differences when new cell data is available.
In the online application program, a long-term and short-term memory network is adopted to keep sharing attributes, and a full connection layer is adjusted to adapt to the difference between batteries of new battery data.
To quantitatively evaluate model performance, the model performance was quantitatively evaluated using root mean square error (root mean square error, RMSE) and mean absolute percentage error (mean absolute percentage error, MAPE);
wherein, root mean square error is:
the average absolute percentage error is:
where n is the total number of cycles of the online prediction.
The LSTM-FC network architecture of the long-term memory network and the full connection layer provided in the online application program of the invention consists of an input layer, an LSTM layer, an FC layer and an output layer. The long-short-term memory network (LSTM) layer is composed of a series of recurrent neurons, each cell being judged by a forgetting gate, an input gate and an output gate as to the importance of the input information.
(1) The forgetting door is as follows: f (f) t ＝σ(w f ·[h t-1 ,x t ]+b f ) (3)；
Wherein σ is a sigmoid activation function, w f And b f Weight matrix and bias, x, respectively representing forgetting gate f t For input, h t-1 For the hidden layer output state, the subscript t represents the current time step;
(2) The input door is:
i t ＝σ(w i ·[h t-1 ,x t ]+b i ) (4)
wherein, tan h is hyperbolic tangent activation function,
is a candidate state, C t Is the current cell state, (:) represents the element multiplication operator;
(3) The output door is:
o t ＝σ(w o ·[h t-1 ,x t ]+b C ) (7)
h t ＝o t *tanh(C t ) (8)。
in order to deal with the cell-to-cell variation, the present invention also contemplates FC layers to learn specific characteristics from predicted cells. Wherein the final predicted capacity of time step t
The calculation formula of (2) is as follows:
during the transfer learning process, the front LSTM layer is 'frozen' and the FC layer is trimmed based on the new cell data. Since the FC layer is a linear model related to input and output, it has low computational complexity and is suitable for on-line prediction.
In some embodiments, the method for online predicting the health state of the lithium battery for the vehicle further comprises battery feature extraction, and in the embodiment, the new feature for predicting the health state of the battery is suitable for different working conditions of the battery, including the current popular 10-minute fast charge strategy.
The battery feature extraction comprises the following steps:
Since the feature extraction method only requires a brief stop of T during charging r The battery discharging and using process has no requirement, so that the battery discharging and using device can meet any working condition and has good adaptability and practicability.
The feature extraction process is described in more detail in the battery data sets published by the university of Stanford and the university of Massachusetts:// data. Matr. Io/1 /). As shown in fig. 2, the proposed feature extraction method is applicable to a battery under any operating condition. Where the charge starts from 0 and charges to 80% with a number of different fast charge strategies, thus Q r =0. The rated capacity of the battery is 1.1Ah, therefore Q fc =0.88 Ah, furthermore, the invention takes T r =10 seconds. Thus, as shown in fig. 2 (a, b, c), the battery under any condition was rapidly charged to 80% capacity for about 10 minutes, and the terminal voltage value V was collected after resting for 10 seconds Tr I.e. is a health feature. Taking 5 batteries with different charging conditions as an example, the characteristics extracted from the whole life cycle are shown in fig. 2 (d), and as the cycle number increases, V Tr Monotonically increases at a rate similar to the rate of capacity decrease. When the failure threshold is reached, V of all cells Tr The eigenvalues are near the upper cutoff voltage and are within the statistical confidence interval. The confidence interval is less than 0.01V, which reflects the degradation process well.
In some embodiments, in combination with the voltage characteristic data extracted in the upper section, the present invention proposes a characteristic fusion scheme for improving adaptability of a prediction model, as shown in fig. 3, wherein the characteristic fusion scheme performs characteristic fusion on the voltage characteristic and the cycle number, and further includes:
the historically extracted voltage signature v at the current cycle t t-l ,v t-l+1 ,…,v t ]And cycle number Cn t Fusion is carried out through a sliding window; i.e. each time the model is input as a fusion feature Φ t . It is noted that due to the voltage characteristic v t-l ,v t-l+1 ,…,v t ]Is directly extracted from specific battery data to embody the private attribute of the battery; and cycle number Cn t Only the number of uses of the battery is reflected, representing the common properties of all batteries.
Adjusting the voltage characteristic v by adjusting the window width l+1 t-l ,v t-l+1 ,…,v t ]And cycle number Cn t The proportion of the attribute features changes the feature information input by the model. The scheme can be suitable for prediction of the inconsistent difference battery, and the prediction precision of the prediction model is improved.
More specifically, the input characteristics and output capacity formulas of the model are as follows:
(1) Training process
Wherein v is i Representing features extracted at the ith cycle, cn i And Q i The number and capacity of the ith cycle after normalization are shown.
(2) Prediction process
Where P (·) represents the predictive model,
the predicted capacity of the t+k cycle is shown.
In some embodiments, to improve the adaptability and generalization ability of the model, in combination with multi-task cross-validation, the present invention proposes a hyper-parametric closed-loop optimization method, as shown in fig. 4.
The method comprises the steps of finding a public area through multitask cross-validation optimization to obtain optimal super-parameters, and further comprises model super-parameter closed-loop optimization, wherein the super-parameter closed-loop optimization method comprises the following steps:
(1) First, a cross-validated task sample combination is assigned, and in the source domain data sample set { S }, the task sample combination { TSS } is assigned according to the similarity of the samples i ,VS i } j Where TSS is the training sample set and VS is the validation sample set, where: if the cell sample aging trend in TSS and VS is the same, { TSS, VS } j Combining tasks with high similarity; otherwise corresponding { TSS, VS } j Combining tasks with low similarity;
(2) Secondly, randomly initializing all super parameters to be optimized in a section to be optimized;
(3) Then, selecting a group of hyper-parameters to execute gridding search in a section to be optimized, respectively carrying out cross verification according to the task combination of high similarity and low similarity, using TSS for training a model, calculating verification error values of VS, and recording hyper-parameter values corresponding to the minimum verification error;
(4) Replacing the corresponding initial value of the step (2) with the super-parameter value obtained in the step (3), and executing the step (3) again to obtain the optimal super-parameter in the whole searching range. After searching at least one time for all the super parameters to be optimized, entering the step (5);
(5) When all the hyper-parameters are satisfied, the process is stopped and the optimal hyper-parameter value is obtained. Otherwise, the corresponding super parameters are replaced by recommended values, and then the next group is selected for similar optimization. These processes iterate until all the hyper-parameters are met.
The feature extraction process is described in more detail by using battery data sets published by the university of Stanford and the university of Massa https://data.matr.io/1/)。
Firstly, the training set samples TSS and the verification samples VS are divided into a source domain data set according to the difference of the battery cell data, and then different task combinations are obtained according to the similarity of the TSS and the VS. Using the condition 5 dataset, seven LIBs are split into four parts, training sample set 1 (TSS 1), training sample set 2 (TSS 2), validation sample 1 (VS 1) and validation sample 2 (VS 2). TSS1 contains batteries 95 and 121, TSS2 contains batteries 88, 102 and 109, VS1 contains battery 97, and VS2 contains battery 117. Four tasks are formed based on the similarity between TSS and VS. High similarity tasks include TSS1-VS1 and TSS2-VS2. The low similarity task consists of TSS1-VS2 and TSS2-VMS 1. At each cross-validation of the superparameter search, a TSS training model is first used and a validation error value of the VS is calculated.
The key hyper-parameters to be optimized of the model of the invention are shown in Table 1, including the number of LSTM neurons and FC units, SW size, learning rate and fine tuning of epochs. The six super parameters are divided into four groups, namely G1, P1 and P4; G2P 2& P5; G3:P3; G4:P6. The specific algorithm is shown in table 2, under the condition of fixing other super parameters, the super parameters are optimized sequentially from G1 to G4 through grid search, and finally the optimal super parameters of the model are obtained.
TABLE 1 super parameters to be optimized
| Structural parameters | P1:LSTM neurons | P2:FC unit | P3:SW size |
| Learning parameters | P4 training learning rate | P5 Fine tuning learning rate | P6 Fine tuning of Epoch |
TABLE 2 super parameter optimization algorithm
In some embodiments, the method further comprises model offline training and online prediction before online predicting the battery capacity in different degradation modes, comprising:
constructing a network model according to the obtained model optimization parameters, using battery data circularly aged to fail in a source domain to train the network model offline, and migrating and integrating the trained network model into a vehicle-mounted battery management system;
extracting voltage characteristics and carrying out characteristic fusion, fine-tuning a migration integrated network model by utilizing the characteristics and capacity data of the existing history records, predicting the capacity health state of the battery by taking the fusion characteristics of the network model after fine-tuning based on the current moment as input, and inputting the model after the sliding window moves forward and fuses the extracted characteristics and the cycle number, so that the health state of the battery can be predicted.
The whole flow is shown in fig. 5, and a network model is constructed by referring to the scheme of the section 1 according to the model optimization parameters obtained in the section 4; the model is then trained offline using battery data in the source domain that is cyclically aged to failure.
The trained model network is then migrated and integrated into an on-board battery management system (Battery management system, BMS). And extracting voltage characteristics according to the method of the section 2, and carrying out characteristic fusion according to the method of the section 3. And along with the working condition use of the battery and the sensor data acquisition record, the migration integration model is finely adjusted by utilizing the characteristics and the capacity data of the existing history record. And the finely tuned model is based on fusion characteristics at the current moment as input, so that the capacity health state prediction of the corresponding battery is realized.
With the cyclic charge and discharge of the battery, the corresponding terminal voltage is measured as an extraction feature only when the battery is charged to a specific capacity value; the sliding window moves forwards, the extracted features are fused with the cycle number and then input into the model, so that the health state of the battery can be predicted, and no requirement is placed on other running conditions of the battery.
The present invention uses battery data sets from the universities of Ottoman and Stanford to verify the benefits (https:// data. Matr. I o/1 /). The dataset consisted of lithium iron phosphate batteries manufactured by a123 Systems. Table 3 lists the key technical indicators of these cells. Experiments were aimed at exploring the degradation of battery capacity under different fast charge conditions due to complex aging mechanisms and manufacturing variability. An incubator was used to set the ambient temperature to 30 ℃. All cells cycled under different fast charge conditions, but the discharge conditions were identical (4C to 2.0V, where 1C is 1.1A). Under fast charge conditions, the battery is first charged to 80% soc by a two-step high CC rate. The average fast charge rate is between 3.6C and 6C, with fast charge times approaching 10 minutes. After 10 seconds of rest, the battery was charged to full state of charge by CC-CV strategy (1C to 3.6V). The capacity degradation curves of these batteries are shown in fig. 6 to 11, and these batteries can be classified into the states shown in conditions 1 to 5 according to the charging strategy. For example, in condition 5, the battery is charged to 19% soc first using 5.6C and then to 80% soc using 4.6C. Although manufacturing defects are amplified by high current rates under the same charging conditions, the battery still exhibits significant degradation variability.
TABLE 3 technical indicators of experimental batteries
In some embodiments, for model hyper-parameter determination, the predictive model hyper-parameters are as shown in table 4, according to the optimization strategy given in fig. 4 and table 2:
TABLE 4 super parameters of predictive model
In some embodiments, the prediction results are analyzed by, first, comparing with the currentThe comparison of the results of the method verifies the validity of the model framework of the invention. Second, the utility of the new feature was analyzed by comparison with the other four features. Thirdly, the adaptability of the method under different fast charging conditions was verified by six experiments. An illustration of training and testing battery sample data for all six cases is shown in table 5. The invention provides a new health characteristic (V Tr Denoted as v80%) and a voltage curve characteristic (V) was extracted based on the DV and IC curves shown in fig. 12 and 13, respectively, and a VT characteristic was composed with the average temperature characteristic (T) of the battery surface
TABLE 5 description of all training and test batteries in case 1 through case 6
In some embodiments, at the time of model frame performance verification evaluation, a comprehensive comparison is made with the results of existing methods to demonstrate the effectiveness of the proposed model frame.
The 91 and 100 cells in condition 2 were used for training and the 124 cell was used for testing. From the extracted VT features, a capacity prediction model is built using gaussian process regression (Gaussian process regression, GPR). The predicted RMSE for this GPR method was 0.6900%. The predicted RMSE of the proposed method was 0.2465%, far lower than the GPR method, keeping the same experimental data set and design. To exclude the effect of features, VT features and LSTM-FC model frameworks proposed in the present invention are added for model performance comparison. It predicts a RMSE of 0.5472% which is 20.7% lower than the predicted RMSE of GPR. This can be considered a great advance. This demonstrates that the proposed LSTM-FC based migration learning model is viable and efficient, since the experimental design and data set are identical.
In some embodiments, in order to further verify the utility and superiority of the proposed features, four additional different features are added, including V80%, cn, VT and vt+cn, for feature performance comparison, when evaluating feature performance verification. Fig. 14-20 show the predicted capacity and error for all six different fast charge cases. The numerical results are summarized in table 6.
For a single v80% feature, RMSE and MAPE were predicted to be lower than VT and Cn features in all six cases. The result shows that the V80% characteristic can better reflect the aging behavior of the LIB, and is more beneficial to establishing the mapping relation with the capacity. Especially in the late degradation stage, as in fig. 14 to 19, the v80% characteristic predicted capacity curve overlaps well with the actual curve. For the integrated v80++cn feature, its capacity prediction RMSE and MAPE are lowest in all six cases compared to the other four features. It demonstrates the beneficial effect of adding cycle number Cn compared to the V80% feature. The Cn characteristic reflects a common characteristic of cyclic aging. Therefore, the integrated feature v80++cn not only utilizes the private attributes of all batteries, but also utilizes the general information, thereby greatly improving the adaptability of the model. It can be shown in fig. 14-19 that an increase in the number of cycles Cn stabilizes the model performance, thereby further reducing the prediction error. It further verifies the utility of the proposed v80% feature compared to the vt+cn feature.
TABLE 6 capacity prediction Performance comparison of five characteristics under six fast Charge conditions
In some embodiments, six comparison cases were performed under different fast charge conditions in order to verify the suitability of the proposed method at the time of the model suitability verification evaluation, as shown in table 6.
In some embodiments, the first order fast charge condition. Fig. 14 and 15 show the predicted capacity and error results for cases 1 and 2 using the first-order quick charge condition. The capacity predictions RMSE for cases 1 and 2 were 0.3350% and 0.2820%, respectively. Good results indicate that the proposed method can be well used for first-order fast charge conditions.
In some embodiments, a second order fast charge condition. Fig. 16 and 17 show the results of cases 3 and 4 under the second-order quick charge condition. In case 3, there is a large difference in capacity degradation curves between the training samples (No. 85, no. 107, and No. 115) and the test unit (No. 119), and the maximum difference between training and test battery life exceeds 1000 cycles. In this case, the predictions RMSE and MAPE with v80++cn as model features were 0.3972% and 0.3190%, respectively, significantly lower than the predictions for the other four features. Similar experimental results appear in case 4. The comparison results show that the proposed method is feasible under second order fast charge conditions.
In some embodiments, the evaluation is cross-validated. Fig. 18 and 19 show the model performance of cases 5 and 6 under the intersecting fast charge condition. In case 5, RMSE and MAPE were predicted to be 0.8600% and 0.7339%, respectively, which is lowest compared to the model performance of the other four features. Although they are slightly larger than in cases 1-4, this is reasonable because the charging conditions of the training and test samples are different. In case 6, RMSE and MAPE were predicted to be 0.1956% and 0.1290%, respectively. This is a good experimental result, since all evaluation criteria are less than 1%.
In summary, the proposed method has a high adaptability. As shown in cases 1-4, it can achieve good capacity predictive performance for the same lot of batteries under the same charging conditions. The experimental results in cases 5-6 also show that the proposed method is adaptive and can effectively handle capacity prediction challenges caused by inter-cell variation.
In some embodiments, model complexity assessment. All experiments were performed on an associative thinkcontre PC520C computer with Intel Xeon W-2175 (2.50 GHz) CPU, 128-GB RAM, nvidia Quadro RTX-4000 graphics card, and Windows 10 (64 bit) system. Data processing and predictive modeling is done on MATLAB 2020 b. Table 7 summarizes the calculation times for the six cases. The experimental results are the average of five runs. Offline training was less than 2 minutes. The online trim time is approximately 1 second, while the predicted time is 0.011 seconds. Thus, online application of the proposed method in an actual BMS is possible.
TABLE 7 numerical results of model complexity assessment
In summary, the method for online predicting the state of health of the vehicle lithium battery adapting to the fast charging strategy introduces a new health feature from the battery charging stage, namely, the terminal voltage when the battery is rapidly charged to 80% of SOC in about 10 minutes, so as to indicate the battery capacity.
The invention provides an on-line prediction method of the state of health of a vehicle lithium battery adapting to a fast charge strategy, and also provides a feature fusion scheme for improving the adaptability of a prediction model, wherein voltage features showing specific battery private properties and cycle times showing common properties of all batteries are fused through a sliding window technology. The proportion of the two attribute features can be adjusted by adjusting the window width, so that the feature information input by the follow-up model is changed.
The invention provides an on-line prediction method of the health state of a vehicle lithium battery adapting to a fast charge strategy, and also provides a self-adaptive multi-task prediction modeling scheme and an optimization strategy based on transfer learning, so as to solve the problems of insufficient model adaptability and generalization capability caused by the difference between a source domain and a target domain. The optimal model is obtained by using a high-similarity and low-similarity task closed-loop optimization method, so that the battery capacity under different degradation modes can be predicted on line.
When the performance state of the battery of the new energy automobile is monitored and evaluated on line, the model provided by the invention firstly develops an optimal model on the basis of the source domain data in an off-line mode, the model is embedded into the vehicle-mounted BMS to realize on-line application, and along with the accumulation of on-line measurement data, the network parameters of the model are continuously finely adjusted to adapt to different aging stages of different batteries.
When the battery product is developed, tested and inspected, the characteristics extracted by the method have larger correlation with the aging behavior characteristics of the battery under different operation conditions, and the internal mechanism of the battery can be reflected to a certain extent, so that the future aging state of the battery can be predicted in early stage according to the characteristics, and the cost is saved for the development of the lithium battery.
In order to adapt to the development trend of the future cloud technology and the Internet of things, the basic part of the model can also share migration, and even if the charging modes of the batteries are different and the discharging modes are different, the model can migrate to realize adaptation, so that an automobile can only need to measure charging and discharging data in the use process of the batteries on line and upload the charging and discharging data to the cloud, all the automobiles share model parameters, and a more accurate prediction effect is realized by virtue of a large amount of working condition data and strong calculation and storage capacity of cloud calculation.
The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, and alternatives falling within the spirit and principles of the invention.
Claims (10)
1. The method is characterized by comprising two parts of off-line model development and multi-task model on-line application; the off-line model development part comprises super-parameter adjustment and model training, and the multi-task model on-line application part comprises model transfer learning;
The method for predicting the health state of the lithium battery for the vehicle on line comprises the following steps:
super-parameter adjustment: the super-parameters are adjusted by using high-similarity and low-similarity tasks with different degradation modes in a source domain, and the public area is found through multi-task cross-validation optimization to obtain the optimal super-parameters;
model training: training an offline model by taking complete life cycle data of part of batteries in a target domain as source domain data, and transmitting the trained offline model to an online application program;
model transfer learning: and carrying out online measurement on the target application monomer to obtain the voltage characteristics of the private attribute of the battery and the cycle times of the common attribute of the battery, carrying out characteristic fusion on the voltage characteristics and the cycle times, and then inputting the fused characteristics into a trained offline model to predict, and carrying out online prediction on the battery capacity under different degradation modes.
2. The method for online predicting the health status of a lithium battery for a vehicle according to claim 1, wherein the online application program uses a long-short-term memory network to maintain the shared attribute, and adjusts the full connection layer to adapt to the difference between the batteries of the new battery data.
3. The on-line prediction method for the health state of a lithium battery for a vehicle, which is adapted to a fast charging strategy, according to claim 2, is characterized in that the on-line prediction method for the health state of a lithium battery for a vehicle further comprises: quantitatively evaluating the model performance by using the root mean square error and the average absolute percentage error;
Wherein, root mean square error is:
the average absolute percentage error is:
where n is the total number of cycles of the online prediction.
4. The method for online predicting the state of health of a lithium battery for a vehicle, which is adaptive to a fast charge strategy according to claim 1, wherein the LSTM-FC network architecture of the long-short-term memory network and the full connection layer proposed in the online application program consists of an input layer, an LSTM layer, an FC layer and an output layer; the long-short-term memory network (LSTM) layer is composed of several recurrent neurons, each of which is judged by forgetting gate, input gate and output gate to determine the importance of the input information.
5. The method for online predicting the health status of a lithium battery for a vehicle adapted to a fast charge strategy according to claim 4, wherein the forgetting gate is: f (f) t ＝σ(w f ·[h t-1 ,x t ]+b f )；
Wherein σ is a sigmoid activation function, w f And b f Weight matrix and bias, x, respectively representing forgetting gate f t For input, h t-1 For the hidden layer output state, the subscript t represents the current time step;
the input door is:
i t ＝σ(w i ·[h t-1 ,x t ]+b i )
wherein, tan h is hyperbolic tangent activation function,
is a candidate state, C t Is the current cell state, (:) represents the element multiplication operator;
the output door is:
o t ＝σ(w o ·[h t-1 ,x t ]+b C )
h t ＝o t *tanh(C t )。
6. the method for online prediction of state of health of a lithium battery for a vehicle adapted to a fast charge strategy according to claim 5, wherein the FC layer of the LSTM-FC network architecture is further adapted to learn a specific characteristic from the predicted battery, the final predicted capacity at time step t 
The calculation formula of (2) is as follows:
7. the on-line prediction method for the state of health of a vehicle lithium battery adapting to a fast charge strategy according to claim 6, further comprising a battery feature extraction, wherein the battery feature extraction comprises the steps of:
charging the battery, and recording the remaining capacity Q of the battery at the beginning of charging r And starts to calculate the charge capacity Q based on ampere integral c ，Q c ＝∫ 0 Idt, wherein I is charging current, and t is charging time;
when the residual current Q r And charge capacity Q c Sum Q r +Q c Greater than or equal to the set quick charge capacity requirement Q fc When the charging is stopped briefly and the terminal voltage data V of the battery is collected and recorded in a monitoring mode T ；
When the short stop charging time reaches the set value T r When the battery is charged or works continuously according to the original charging mode;
from collecting recorded terminal voltage data V T In the process, when the acquisition just reaches the settingM T r Voltage value V of (2) Tr The V is Tr The value is the extracted characteristic value.
8. The method for online predicting the health of a lithium battery for a vehicle, which is adapted to a fast charge strategy according to claim 7, wherein the feature fusion is performed on the voltage feature and the cycle number, and the method further comprises:
the historically extracted voltage signature v at the current cycle t t-l ,v t-l+1 ,…,v t ]And cycle number Cn t Fusion is carried out through a sliding window;
adjusting the voltage characteristic v by adjusting the window width l+1 t-l ,v t-l+1 ,…,v t ]And cycle number Cn t The proportion of the attribute features changes the feature information input by the model.
9. The method for online prediction of the state of health of a lithium battery for a vehicle adapted to a fast charge strategy according to claim 1, wherein the method for online prediction of the state of health of a lithium battery for a vehicle is characterized in that the optimal super-parameters are obtained by finding a common area through a multi-task cross-validation optimization, and further comprises a model super-parameter closed-loop optimization, wherein the super-parameter closed-loop optimization method comprises the following steps:
task sample combinations for cross-validation are assigned, and in the source domain data sample set { S }, task sample combinations { TSS } are assigned by sample similarity i ,VS i } j Wherein TSS is the training sample set and VS is the verification sample set;
randomly initializing all super parameters to be optimized in a section to be optimized;
selecting a group of super parameters to execute gridding search in the interval to be optimized, respectively carrying out cross verification according to the task combination of high similarity and low similarity, and recording the super parameter value corresponding to the minimum verification error;
replacing the corresponding initial value of random initialization with the super-parameter value, and executing gridding search again to obtain the optimal super-parameter in the whole searching range;
After all the super parameters to be optimized are searched at least once, and when all the super parameters are met, stopping the program, and obtaining the optimal super parameter value; otherwise, replacing the corresponding super parameters with the recommended values, selecting the next group to perform the optimization process again, and iterating until all the super parameters are met.
10. The method for online prediction of the state of health of a lithium battery for a vehicle adapted to a fast charge strategy according to claim 1, further comprising model offline training and online prediction before online prediction of the battery capacity in different degradation modes, comprising:
constructing a network model according to the obtained model optimization parameters, using battery data circularly aged to fail in a source domain to train the network model offline, and migrating and integrating the trained network model into a vehicle-mounted battery management system;
extracting voltage characteristics and carrying out characteristic fusion, fine-tuning a migration integrated network model by utilizing the characteristics and capacity data of the existing history records, predicting the capacity health state of the battery by taking the fusion characteristics of the network model after fine-tuning based on the current moment as input, and inputting the model after the sliding window moves forward and fuses the extracted characteristics and the cycle number, so that the health state of the battery can be predicted.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310020308.1A CN116008844A (en) | 2023-01-06 | 2023-01-06 | Vehicle lithium battery health state online prediction method adapting to quick charge strategy |
| DE102024100184.4A DE102024100184A1 (en) | 2023-01-06 | 2024-01-04 | Online prediction method for the health status of lithium batteries in motor vehicles |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310020308.1A CN116008844A (en) | 2023-01-06 | 2023-01-06 | Vehicle lithium battery health state online prediction method adapting to quick charge strategy |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116008844A true CN116008844A (en) | 2023-04-25 |
Family
ID=86029767
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310020308.1A Pending CN116008844A (en) | 2023-01-06 | 2023-01-06 | Vehicle lithium battery health state online prediction method adapting to quick charge strategy |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN116008844A (en) |
| DE (1) | DE102024100184A1 (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117236200A (en) * | 2023-11-16 | 2023-12-15 | 北京航空航天大学 | Method for optimizing quick charge strategy of aerocar battery based on data-driven reduced-order model |
| CN117232097A (en) * | 2023-11-09 | 2023-12-15 | 上海轻环能源科技有限公司 | Central air conditioner refrigerating station optimal control method and system based on self-learning fusion model |
-
2023
- 2023-01-06 CN CN202310020308.1A patent/CN116008844A/en active Pending
-
2024
- 2024-01-04 DE DE102024100184.4A patent/DE102024100184A1/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117232097A (en) * | 2023-11-09 | 2023-12-15 | 上海轻环能源科技有限公司 | Central air conditioner refrigerating station optimal control method and system based on self-learning fusion model |
| CN117232097B (en) * | 2023-11-09 | 2024-02-20 | 上海轻环能源科技有限公司 | Central air conditioner refrigerating station optimal control method and system based on self-learning fusion model |
| CN117236200A (en) * | 2023-11-16 | 2023-12-15 | 北京航空航天大学 | Method for optimizing quick charge strategy of aerocar battery based on data-driven reduced-order model |
| CN117236200B (en) * | 2023-11-16 | 2024-02-02 | 北京航空航天大学 | Method for optimizing quick charge strategy of aerocar battery based on data-driven reduced-order model |
Also Published As
| Publication number | Publication date |
|---|---|
| DE102024100184A1 (en) | 2024-07-11 |
| DE102024100184A8 (en) | 2024-07-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Deng et al. | Data-driven state of charge estimation for lithium-ion battery packs based on Gaussian process regression | |
| Che et al. | Predictive battery health management with transfer learning and online model correction | |
| CN103954913B (en) | Electric automobile power battery life-span prediction method | |
| CN116008844A (en) | Vehicle lithium battery health state online prediction method adapting to quick charge strategy | |
| CN113917337A (en) | Battery health state estimation method based on charging data and LSTM neural network | |
| CN113740736B (en) | Electric vehicle lithium battery SOH estimation method based on deep network self-adaption | |
| Che et al. | Semi-supervised self-learning-based lifetime prediction for batteries | |
| CN113065283A (en) | Battery life prediction method, system, electronic device and storage medium | |
| US20230059529A1 (en) | Characterization of Rechargeable Batteries Using Machine-Learned Algorithms | |
| Zhang et al. | State of health estimation method for lithium-ion batteries using incremental capacity and long short-term memory network | |
| CN111983474A (en) | Lithium ion battery life prediction method and system based on capacity decline model | |
| CN115856678A (en) | Lithium ion battery health state estimation method | |
| CN115201686A (en) | Lithium ion battery health state assessment method under incomplete charging and discharging data | |
| Harippriya et al. | Battery management system to estimate battery aging using deep learning and machine learning algorithms | |
| CN115994441A (en) | Big data cloud platform online battery life prediction method based on mechanism information | |
| CN115308628A (en) | Battery health state monitoring method | |
| CN116298936A (en) | Intelligent lithium ion battery health state prediction method in incomplete voltage range | |
| Chen et al. | A novel battery health indicator and PSO-LSSVR for LiFePO4 battery SOH estimation during constant current charging | |
| CN117169743A (en) | Battery health state estimation method and device based on partial data and model fusion | |
| Hong et al. | Multi-forword-step state of charge prediction for real-world electric vehicles battery systems using a novel LSTM-GRU hybrid neural network | |
| CN117207781A (en) | Fuel cell automobile driving mileage prediction method based on historical real-time energy consumption | |
| CN116540098A (en) | Battery residual value evaluation method, device, equipment and readable storage medium | |
| Cui et al. | An Online State of Health Estimation Method for Lithium-Ion Battery Based on ICA and TPA-LSTM | |
| Zheng et al. | Research on Data-Driven-Based Remaining Useful Life of Lithium-ion Battery | |
| Aliberti et al. | Comparative Analysis of Neural Networks Techniques for Lithium-ion Battery SOH Estimation |
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 |