CN115754738A - Battery pack health state estimation method based on small sample learning twin network - Google Patents
Battery pack health state estimation method based on small sample learning twin network Download PDFInfo
- Publication number
- CN115754738A CN115754738A CN202211526900.0A CN202211526900A CN115754738A CN 115754738 A CN115754738 A CN 115754738A CN 202211526900 A CN202211526900 A CN 202211526900A CN 115754738 A CN115754738 A CN 115754738A
- Authority
- CN
- China
- Prior art keywords
- battery pack
- full
- data set
- standing
- small sample
- 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
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/60—Other road transportation technologies with climate change mitigation effect
- Y02T10/70—Energy storage systems for electromobility, e.g. batteries
Landscapes
- Secondary Cells (AREA)
Abstract
The invention discloses a battery pack health state estimation method based on a small sample learning twin network, which comprises the following steps: acquiring a standing-to-full-charge data set D1 of part of vehicles and a historical data distribution data set D2 of all vehicles, and calculating the SOH of the battery pack in the D1; taking SOH as a true label value, and taking a historical data distribution data set X containing the true label value and a historical data distribution data set without the true label valueSimultaneously inputting the two signals into a pre-constructed twin network F, constructing a distance evaluation function on the estimation results obtained by the two signals, and determining the inputAnd (4) analyzing the similarity degree with X, and determining the health state of the battery pack without the real label value according to the analysis result. The invention solves the problem of insufficient real data labels by using a small sample learning mode, and can achieve the network generalization effect and learn the unprecedented data characteristics by mining the bottom layer characteristic information of the small sample through one neural network in the twin network.
Description
Technical Field
The invention relates to the technical field of battery health state detection, in particular to a battery pack health state estimation method based on a small sample learning twin network.
Background
The power battery is used as a core component of the new energy automobile and is directly related to safe and stable running of the automobile. The health state of the battery is very useful for online battery management, used-car assessment and battery echelon utilization, and is particularly important for evaluating the residual value of the battery. In consideration of the non-linear fading characteristics of the battery, the conventional extrapolation method cannot predict the remaining life of the battery well. From the deep learning perspective, the health state of the battery pack is estimated by utilizing small sample training.
Disclosure of Invention
The invention provides a battery pack health state estimation method based on a small sample learning twin network, which aims to effectively estimate the remaining service life of a battery pack and ensure safe and stable running of a new energy vehicle.
The invention realizes the purpose through the following technical scheme:
a battery pack health state estimation method based on a small sample learning twin network comprises the following steps:
s1, acquiring a standing-to-full-charge data set D1 and a historical data distribution data set D2 of all vehicles of a part of vehicles based on battery pack historical data of all vehicles, and calculating the state of health (SOH) of a battery pack in the standing-to-full-charge data set D1;
s2, data with the same vehicle number in the whole vehicle historical data distribution data set D2 and the vehicle number in the data set D1 which is kept to be full are taken to form a historical data distribution data set X containing a real label value, and the rest data in the whole vehicle historical data distribution data set D2 form a historical data distribution data set without the real label valueWherein the real tag value is the state of health SOH of the battery pack;
s3, distributing the historical data X containing the real label value and the historical data X not containing the real label valueSimultaneously inputting the data into a twin network F pre-constructed based on a standing-to-full data set D1, and constructing a distance evaluation function for estimation results obtained by the twoDetermining input by setting distance assessment thresholdDegree of similarity to X;
s4, according to the pairAnalyzing the similarity degree to determine the historical data distribution data set without real label valueThe middle battery pack belongs to the category and the corresponding SOH value of the health condition.
In a further improvement, in step S1, a specific method for acquiring the standing-to-full data set D1 of the partial vehicle is as follows: all pure charging data fragments which are from standing to full charging in battery pack historical data are obtained through a battery standing rule and a full charging rule, the SOC of the battery pack corresponding to standing is searched through an OCV-SOC table, the pure charging data fragments from the standing SOC to the full charging below a preset value are screened, and a standing to full charging data set D1 is obtained.
The further improvement is that the standing rule is as follows: when the time interval between the battery pack power-on time and the previous process is higher than a time threshold value, judging the battery pack power-on time as a standing point; the full charge rule is as follows: and when the highest charging voltage is higher than the full-charging threshold value, judging the charging point to be a full-charging point.
The further improvement is that the specific way of acquiring the pure charging data segment from standing to full charging is as follows: and searching a standing point closest to the full-charge point in the time dimension according to the full-charge point, and determining that the battery pack data between the two points is fully charged as a standing-to-full-charge data segment and acquiring the battery pack data.
A further improvement consists in that said preset value is 30%.
In a further improvement, in the step S1, a specific method for calculating the state of health SOH of the battery pack in the data set D1 from standing to full charge is as follows:
(1) The accumulated electric quantity of the pure charging data which is screened by the ampere-hour integral calculation and then stands to be fully charged is calculatedWherein i is the charging current at the sampling point and Δ t is the sampling time interval;
(2) By fully charging SOC full And a static SOC static Calculating to obtain a state of charge variable delta SOC = SOC full -SOC static Wherein, SOC full Is 100%;
(3) Calculating the current battery capacity Q according to the results of the step (1) and the step (2) now =ΔQ/ΔSOC;
(4) Calculating the health state SOH = Q by an ampere-hour integration method and a capacity estimation method now /Q real Wherein, Q real Indicating the rated capacity of the battery.
In a further improvement, in step S1, a specific method for acquiring the overall vehicle history data distribution data set D2 is as follows: the time and time ratio in various states in battery pack history data of all vehicles is counted.
The further improvement is that in the step S3, a specific method for constructing the twin network F based on the standing-to-full data set D1 is as follows: and taking the standing full-charge data set D1 and the calculated SOH of the battery pack as small sample data used for model training, inputting the small sample data into a neural network for deep learning training, and copying the trained neural network to jointly form a twin network F.
In a further development, in step S3, an input is determinedThe specific method of similarity with X is as follows: taking the estimation result Y of the historical data distribution data set X containing the real label value as a reference, and judging the historical data distribution data set without the real label value through a distance loss functionIs estimated as a result ofThe distance from the estimation result Y of the historical data distribution data set X containing the true tag value, i.e., the degree of similarity between the two.
In a further improvement, in step S3, the distance determination threshold is ≦ 1%.
The invention has the beneficial effects that:
(1) The problem of insufficient real data labels is solved by using a small sample learning mode, the bottom layer characteristic information of the small sample is mined through one neural network in the twin network, the network generalization effect can be achieved, and the unprecedented data characteristics can be learned.
(2) The algorithm is simple to implement, the result accuracy is high, and the method is easy to popularize.
Drawings
FIG. 1 is a network structure diagram of a twin network of the present invention;
FIG. 2 is a diagram illustrating a data structure according to the present invention.
Detailed Description
The present application will now be described in further detail with reference to the drawings, it should be noted that the following detailed description is given for illustrative purposes only and is not to be construed as limiting the scope of the present application, as those skilled in the art will be able to make numerous insubstantial modifications and adaptations to the present application based on the above disclosure.
With reference to fig. 1 and 2, a battery pack state of health estimation method based on a small sample learning twin network includes the steps of:
s1, acquiring a standing-to-full-charge data set D1 and a historical data distribution data set D2 of all vehicles of a part of vehicles based on battery pack historical data of all vehicles, and calculating the state of health (SOH) of a battery pack in the standing-to-full-charge data set D1;
in this step, the specific method for acquiring the standing-to-full data set D1 of a part of vehicles is as follows: all pure charging data segments from standing to full charging in battery pack historical data are obtained through a battery standing rule and a full charging rule, the SOC of the battery pack corresponding to standing is searched through an OCV-SOC table, the pure charging data segments from the standing SOC to the full charging below a preset value are obtained through screening, and a standing-to-full charging data set D1 is obtained.
Wherein the standing rule is as follows: when the time interval between the power-on time of the battery pack and the previous process is higher than a time threshold (the time threshold is different according to different battery models), judging the battery pack as a standing point; the full charge rule is as follows: and when the highest charging voltage is higher than the full charge threshold (the full charge threshold is different according to different battery types), judging the highest charging voltage to be a full charge point. The specific way of acquiring the pure charging data segment from standing to full charging is as follows: and searching a standing point which is closest to the full charge point in the time dimension according to the full charge point (the SOC of the standing point is less than 30% through OCV table lookup), and determining that the battery pack data between the two points is fully charged as a full charge data fragment standing still and acquiring the full charge data fragment.
Preferably, the preset value is 30%, that is, a pure charging data segment from a standing SOC of less than 30% to a full charge is obtained, and since different types of batteries have different plateau periods, it is most reliable to select a table lookup SOC of less than 30%.
In this step, the specific method for obtaining the whole vehicle historical data distribution data set D2 is as follows: the time and time ratio in various states in battery pack history data of all vehicles is counted. For example: and calculating the time ratio of all time at high temperature to the total time, the time ratio of all time at low temperature to the time ratio of all time at high discharge rate and the like in the historical driving process of a certain vehicle, and counting the data to obtain a piece of data serving as the historical data distribution statistics of the vehicle.
In this step, the specific method for calculating the health state SOH is as follows:
(1) The accumulated electric quantity of the pure charging data which is screened by the ampere-hour integral calculation and then stands to be fully charged is calculatedWherein i is the charging current at the sampling point and Δ t is the sampling time interval;
(2) By fully charging SOC full And a static SOC static (the value is obtained by table look-up and is less than 30%) and the state of charge variable delta SOC = SOC is obtained by calculation full -SOC static Wherein, SOC full Is 100%;
(3) Calculating the current battery capacity Q according to the results of the step (1) and the step (2) now =ΔQ/ΔSOC;
(4) Calculating the health state SOH = Q by an ampere-hour integration method and a capacity estimation method now /Q real Wherein Q is real Indicating the rated capacity of the battery.
S2, data with the same vehicle number in the whole vehicle historical data distribution data set D2 and the same vehicle number in the data set D1 after standing to be full are taken to form a historical data distribution data set X containing a real label value, and residual data in the whole vehicle historical data distribution data set D2 form a historical data distribution data set without the real label valueWherein the real tag value is the state of health SOH of the battery pack;
s3, distributing the historical data X with the real label value and the historical data without the real label valueSimultaneously inputting the data into a twin network F pre-constructed based on a standing-to-full data set D1, and constructing a distance evaluation function for estimation results obtained by the twoDetermining input by setting distance assessment thresholdThe similarity degree with X is preferably, the distance judgment threshold is less than or equal to 1%;
in the step, a specific method for constructing the twin network F based on the standing-to-full data set D1 is as follows: and taking the standing full-charge data set D1 and the calculated SOH of the battery pack as small sample data used for model training, inputting the small sample data into a neural network for deep learning training, and copying the trained neural network to jointly form a twin network F in a semi-supervised form.
In this step, an input is determinedThe specific method of similarity with X is as follows: judging the historical data distribution data set without the real label value by using the distance loss function by taking the estimation result Y of the historical data distribution data set X with the real label value as a referenceIs estimated as a result ofThe distance from the estimation result Y of the historical data distribution data set X containing the true tag value, i.e., the degree of similarity between the two.
S4, pairThe degree of similarity is analyzed, where the analysis is preferably: to carry outStatistical classification of similarity degree, and determining historical data distribution data set without real label value according to similarity degree classification resultThe middle battery pack belongs to the category and the corresponding SOH value of the health condition.
The above-mentioned embodiments only express several embodiments of the present invention, and the description thereof is more specific and detailed, but not construed as limiting the scope of the present invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention.
Claims (10)
1. A battery pack state of health estimation method based on a small sample learning twin network is characterized by comprising the following steps:
s1, acquiring a standing-to-full-charge data set D1 and a historical data distribution data set D2 of all vehicles of a part of vehicles based on battery pack historical data of all vehicles, and calculating the state of health (SOH) of a battery pack in the standing-to-full-charge data set D1;
s2, data with the same vehicle number in the whole vehicle historical data distribution data set D2 and the same vehicle number in the data set D1 after standing to be full are taken to form a historical data distribution data set X containing a real label value, and residual data in the whole vehicle historical data distribution data set D2 form a historical data distribution data set without the real label valueWherein the real tag value is the state of health (SOH) of the battery pack;
s3, containing the real label valueAnd a history data distribution data set X not containing a true tag valueSimultaneously inputting the data into a twin network F pre-constructed based on a standing-to-full data set D1, and constructing a distance evaluation function for estimation results obtained by the twoDetermining input by setting distance assessment thresholdDegree of similarity to X;
2. The battery pack state of health estimation method based on small sample learning twin network as claimed in claim 1, wherein in step S1, the specific method for obtaining the still to full charge data set D1 of the partial vehicle is as follows: all pure charging data segments from standing to full charging in battery pack historical data are obtained through a battery standing rule and a full charging rule, the SOC of the battery pack corresponding to standing is searched through an OCV-SOC table, the pure charging data segments from the standing SOC to the full charging below a preset value are obtained through screening, and a standing-to-full charging data set D1 is obtained.
3. The battery pack state of health estimation method based on small sample learning twin network as claimed in claim 2, wherein the standing rule is: when the time interval between the battery pack power-on time and the previous process is higher than a time threshold value, judging the battery pack power-on time as a standing point; the full charging rule is as follows: and when the highest charging voltage is higher than the full-charging threshold value, judging the charging point to be a full-charging point.
4. The battery pack health state estimation method based on the small sample learning twin network as claimed in claim 3, wherein the specific way of obtaining the pure charge data segment from static state to full charge is as follows: and searching a standing point closest to the full-charge point in the time dimension according to the full-charge point, and determining that the battery pack data between the two points is fully charged as a standing-to-full-charge data segment and acquiring the battery pack data.
5. The battery pack state of health estimation method based on small sample learning twin network as claimed in claim 2, wherein the preset value is 30%.
6. The battery pack state of health estimation method based on the small sample learning twin network as claimed in claim 2, wherein in step S1, the specific method for calculating the state of health SOH of the battery pack in the fully charged data set D1 comprises:
(1) The accumulated electric quantity of the pure charging data which is screened by the ampere-hour integral calculation and then stands to be fully charged is calculatedWherein i is the charging current at the sampling point and Δ t is the sampling time interval;
(2) By full charge SOC fill And a static SOC static Calculating to obtain a state of charge variable delta SOC = SOC full -SOC static Wherein, SOC full Is 100%;
(3) Calculating the current battery capacity Q according to the results of the step (1) and the step (2) now =ΔQ/ΔSOC;
(4) Calculating the health state SOH = Q by an ampere-hour integration method and a capacity estimation method now /Q real Wherein Q is real Indicating the rated capacity of the battery.
7. The battery pack state of health estimation method based on small sample learning twin network as claimed in claim 1, wherein in step S1, the specific method for obtaining the whole vehicle history data distribution data set D2 is as follows: and counting the time and time ratio in various states in the battery pack historical data of all vehicles.
8. The battery pack health state estimation method based on the small sample learning twin network as claimed in claim 1, wherein in step S3, the specific method for constructing the twin network F based on the standing-to-full data set D1 is as follows: and taking the standing full-charge data set D1 and the calculated SOH of the battery pack as small sample data used for model training, inputting the small sample data into a neural network for deep learning training, and copying the trained neural network to jointly form a twin network F.
9. The battery pack health state estimation method based on small sample learning twin network as claimed in claim 1, wherein in step S3, input is determinedThe specific method of similarity with X is as follows: judging the historical data distribution data set without the real label value by using the distance loss function by taking the estimation result Y of the historical data distribution data set X with the real label value as a referenceIs estimated as a result ofThe distance from the estimation result Y of the historical data distribution data set X containing the true tag value, i.e., the degree of similarity between the two.
10. The battery pack state of health estimation method based on small sample learning twin network as claimed in claim 1, wherein in step S3, the distance determination threshold is ≦ 1%.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211526900.0A CN115754738B (en) | 2022-12-01 | 2022-12-01 | Battery pack health state estimation method based on small sample learning twin network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211526900.0A CN115754738B (en) | 2022-12-01 | 2022-12-01 | Battery pack health state estimation method based on small sample learning twin network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN115754738A true CN115754738A (en) | 2023-03-07 |
CN115754738B CN115754738B (en) | 2023-07-18 |
Family
ID=85341904
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202211526900.0A Active CN115754738B (en) | 2022-12-01 | 2022-12-01 | Battery pack health state estimation method based on small sample learning twin network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115754738B (en) |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20100063343A (en) * | 2008-12-03 | 2010-06-11 | 기아자동차주식회사 | Estimating apparatus and method of soh of a battery in a hybrid car |
CN113049976A (en) * | 2021-04-27 | 2021-06-29 | 武汉理工大学 | Vehicle battery health state assessment method and system |
CN113111580A (en) * | 2021-04-07 | 2021-07-13 | 东软睿驰汽车技术(沈阳)有限公司 | Method for constructing battery SOH evaluation model and method for evaluating battery SOH value |
CN113917337A (en) * | 2021-10-13 | 2022-01-11 | 国网福建省电力有限公司 | Battery health state estimation method based on charging data and LSTM neural network |
CN114994546A (en) * | 2022-06-15 | 2022-09-02 | 江苏大学 | Battery pack SOH prediction system and method based on block chain and Informer neural network |
US20220283240A1 (en) * | 2021-03-04 | 2022-09-08 | Wuhan University | Method and system for estimating state of health of battery pack |
-
2022
- 2022-12-01 CN CN202211526900.0A patent/CN115754738B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20100063343A (en) * | 2008-12-03 | 2010-06-11 | 기아자동차주식회사 | Estimating apparatus and method of soh of a battery in a hybrid car |
US20220283240A1 (en) * | 2021-03-04 | 2022-09-08 | Wuhan University | Method and system for estimating state of health of battery pack |
CN113111580A (en) * | 2021-04-07 | 2021-07-13 | 东软睿驰汽车技术(沈阳)有限公司 | Method for constructing battery SOH evaluation model and method for evaluating battery SOH value |
CN113049976A (en) * | 2021-04-27 | 2021-06-29 | 武汉理工大学 | Vehicle battery health state assessment method and system |
CN113917337A (en) * | 2021-10-13 | 2022-01-11 | 国网福建省电力有限公司 | Battery health state estimation method based on charging data and LSTM neural network |
CN114994546A (en) * | 2022-06-15 | 2022-09-02 | 江苏大学 | Battery pack SOH prediction system and method based on block chain and Informer neural network |
Non-Patent Citations (1)
Title |
---|
YONGCHAO WANG 等: "Research on health state estimation methods of lithium-ion battery for small sample date", 《ENERGY REPORTS》, vol. 8, pages 2686 - 2698 * |
Also Published As
Publication number | Publication date |
---|---|
CN115754738B (en) | 2023-07-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111707951B (en) | Battery pack consistency evaluation method and system | |
CN107219463B (en) | Battery prediction technique based on battery truck status data | |
CN108919129A (en) | When a kind of under variable working condition power battery life-span prediction method | |
CN109655754B (en) | Battery performance evaluation method based on multi-dimensional grading of charging process | |
EP4019321A1 (en) | Battery managing method and device, and vehicle | |
CN113052464B (en) | Method and system for evaluating reliability of battery energy storage system | |
CN113030761B (en) | Method and system for evaluating battery health state of ultra-large-scale energy storage power station | |
CN113917334B (en) | Battery health state estimation method based on evolution LSTM self-encoder | |
CN109782190A (en) | Method for estimating the remaining life of single battery or single batch of battery | |
CN116609676B (en) | Method and system for monitoring state of hybrid energy storage battery based on big data processing | |
US11835589B2 (en) | Method and apparatus for machine-individual improvement of the lifetime of a battery in a battery-operated machine | |
CN112684363A (en) | Lithium ion battery health state estimation method based on discharge process | |
CN113657360A (en) | Lithium battery health state estimation method, device, equipment and readable storage medium | |
CN115994441A (en) | Big data cloud platform online battery life prediction method based on mechanism information | |
CN114839536A (en) | Lithium ion battery health state estimation method based on multiple health factors | |
CN110542866A (en) | method for estimating residual electric quantity parameter of battery | |
Ben-Marzouk et al. | Generation of a real-life battery usage pattern for electrical vehicle application and aging comparison with the WLTC profile | |
CN112305441A (en) | Power battery health state assessment method under integrated clustering | |
CN114646888A (en) | Assessment method and system for capacity attenuation of power battery | |
Xu et al. | Safety warning analysis for power battery packs in electric vehicles with running data | |
CN114545275A (en) | Indirect prediction method for remaining service life of lithium ion battery | |
CN1879251B (en) | Battery float management | |
CN115754738A (en) | Battery pack health state estimation method based on small sample learning twin network | |
CN116819364A (en) | Battery health determination method, application method, electronic device and storage medium | |
CN116243194A (en) | On-line prediction method for battery health state under mixed working condition |
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 |