CN113673176A - Deep learning battery state of charge estimation system and method based on Transformer - Google Patents
Deep learning battery state of charge estimation system and method based on Transformer Download PDFInfo
- Publication number
- CN113673176A CN113673176A CN202111230200.2A CN202111230200A CN113673176A CN 113673176 A CN113673176 A CN 113673176A CN 202111230200 A CN202111230200 A CN 202111230200A CN 113673176 A CN113673176 A CN 113673176A
- Authority
- CN
- China
- Prior art keywords
- battery
- layer
- transformer
- deep learning
- attention mechanism
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/36—Arrangements for testing, measuring or monitoring the electrical condition of accumulators or electric batteries, e.g. capacity or state of charge [SoC]
- G01R31/367—Software therefor, e.g. for battery testing using modelling or look-up tables
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/36—Arrangements for testing, measuring or monitoring the electrical condition of accumulators or electric batteries, e.g. capacity or state of charge [SoC]
- G01R31/382—Arrangements for monitoring battery or accumulator variables, e.g. SoC
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention relates to the technical field of online prediction of SOC (state of charge) of a battery, in particular to a deep learning battery state of charge estimation system and method based on a Transformer. The system comprises: a fully connected neural network for processing and outputting the battery characteristic sequence R and the battery initial state sequence S
(ii) a Transformer neural netA network for processing and outputting the battery charge and discharge process sequence T
(ii) a Linear fusion layer for output
And output
A stitching and weighting calculation is performed to obtain a predicted battery SOC,
(ii) a And an output layer for outputting
. The method is realized based on the system. The invention can better realize the online prediction of the SOC of the battery.
Description
Technical Field
The invention relates to the technical field of online prediction of SOC (state of charge) of a battery, in particular to a deep learning battery state of charge estimation system and method based on a Transformer.
Background
The lithium battery has the advantages of high energy storage density, long service life, low self-discharge rate, light weight, environmental protection and the like, and is widely applied to various daily life scenes such as mobile phones, notebook computers, electric tools, new energy automobiles and the like. Among them, the state of charge (SOC) is a key index in the battery usage process. The accurate estimation of the state of charge has important significance in preventing overcharge and overdischarge, improving the battery energy utilization rate and guaranteeing the safety and stability of a battery system, and provides necessary conditions for subsequent optimization of energy distribution of the whole vehicle. If the state of charge of the battery cannot be accurately measured, the lithium battery may be subjected to overcharge and overdischarge conditions, which may damage the battery, and even cause a poor electrochemical reaction inside the battery, thereby shortening the service life of the battery. In extreme cases, overcharge may cause severe heat generation of the battery, leading to thermal runaway, and causing serious accidents. In addition, in the use process, if the state of charge of the battery cannot be accurately measured, the remaining driving mileage of the electric automobile cannot be predicted, and sudden anchoring in the driving process is easily caused, so that traffic accidents are caused.
Data of the state of charge (SOC) of the battery is defined as follows:
wherein the content of the first and second substances,
indicating the amount of the current remaining charge,
representing the maximum charge at full charge. That is, the state of charge (SOC) of the battery can directly reflect the amount of remaining charge that the battery can provide during use, or the amount of charge that has been charged during charging.
The accurate estimation of the state of charge is a difficult problem in the industry, and the accurate estimation of the state of charge of the battery can effectively avoid the overcharge and the overdischarge of the battery, reduce the damage of the battery and prolong the service life of the battery.
The current methods for researching and predicting the state of charge of the lithium battery mainly comprise four types: ampere-hour integration, open circuit voltage, data-driven, and model-based methods. The ampere-hour integration method is used for estimating the state of charge of the battery by integrating the current in a time dimension, and is simple and easy to implement; but the accumulated error will get larger and larger as time increases. The open-circuit voltage method is to establish the relationship between the battery offline open-account voltage and the SOC, but in actual use, the battery charging and discharging current is constantly changed, so that the open-circuit voltage method is difficult to be applied to online estimation of the SOC of the battery. The data-driven method estimates the SOC of the battery by accumulating a large amount of data and learning based on a machine learning model and a deep learning model. The model-based method comprises an SOC estimation method based on an electrochemical model and an equivalent circuit model, and the electrochemical model is too complex and difficult to solve and use for online estimation of the SOC of the lithium battery.
At present, the ampere-hour integral method is currently adopted by the industry to estimate the state of charge. The method is simple and easy to implement, but has strong limitation. Mainly comprises the following steps: the factor considered in the ampere-hour integral method model is single, the feedback correction capability is lacked, and the estimation precision is greatly reduced along with the increase of the use times of the battery.
Disclosure of Invention
Based on the problem of poor accuracy of the existing estimation of the state of charge of the battery, the invention provides a transform-based deep learning battery state of charge estimation system and a transform-based deep learning battery state of charge estimation method, so that the online high-accuracy prediction of the state of charge of the lithium battery can be better realized.
The deep learning battery state of charge estimation system based on the Transformer comprises the following components:
a fully connected neural network for processing and outputting the battery characteristic sequence R and the battery initial state sequence S
；
A Transformer neural network for processing and outputting the battery charging and discharging process sequence T
；
Linear fusion layer for output
And output
A stitching and weighting calculation is performed to obtain a predicted battery SOC,
(ii) a And
output layer for outputting
。
In the invention, the outputs of the fully-connected neural network and the Transformer neural network can be fused and output through the linear fusion layer, so that the influence of the battery characteristics, the battery initial state and the variation of the parameters of the battery charge-discharge process on the battery SOC on a time axis can be comprehensively considered, and the battery SOC has better robustness.
Preferably, the fully-connected neural network includes a plurality of fully-connected layers, and the formula for the l-th fully-connected layer is:
wherein the content of the first and second substances,
and
respectively representing the weight term and the bias term of the l-th fully-connected layer,
is the input of the l-th fully connected layer,
is the output of the l < th > layer fully-connected layer and serves as the input of the l +1 < th > layer fully-connected layer,
is an activation function. Therefore, the battery characteristic sequence R and the battery initial state sequence S can be preferably processed.
Preferably, the Transformer neural network has an Encoder network and a Decoder network, the Encoder network has an Encoder multi-head attention mechanism layer, and the Decoder network has a Decoder multi-head attention mechanism layer. Therefore, the processing of the battery charging and discharging process sequence T can be preferably realized.
Preferably, the Encode multi-head attention mechanism layer comprises a plurality of Encode self-attention mechanism layers and a full connection layer, and each Encode self-attention mechanism layer is provided with a parameter matrix
、
And
(ii) a For the mth Encode self-attention mechanism layer in the Encode network, the parameter matrixes are respectively
、
And
；
the mth Encoder self-attention mechanism layer is used for acquiring the following matrix:
wherein the content of the first and second substances,
the input of the first Encoder multi-head attention mechanism layer is a battery charging and discharging process sequence T, and the output of the last Encoder multi-head attention mechanism layer is used as the input of the next Encoder multi-head attention mechanism layer;
the mth Encoder self-attention mechanism layer is obtaining
、
And
then, it obtains the output matrix by calculating:
wherein the content of the first and second substances,
is the dimension of the parameter matrix.
Through the above, the processing of the battery charge and discharge process sequence T can be preferably realized.
Preferably, the plurality of encorder self-attention mechanism layers are used for outputting i output matrices, wherein i is the total number of the plurality of self-attention mechanism layers; the full connection layer is used for acquiring the output of the corresponding Encoder multi-head attention mechanism layer according to the following operation
：
Wherein the content of the first and second substances,
is a parameter matrix of the fully-connected layer,
representing a matrix
And (6) splicing.
Therefore, the splicing and weighting calculation of the output results of the fully-connected neural network and the Transformer neural network can be better realized.
The deep learning battery state of charge estimation system method based on the Transformer comprises the following steps:
step S1, constructing the deep learning battery state of charge estimation system;
step S2, training a system;
and step S3, in the process of discharging or charging the battery, acquiring a battery characteristic sequence R, a battery initial state sequence S and a battery charging and discharging process sequence T, predicting the SOC of the battery through the trained deep learning battery state-of-charge estimation system and outputting the SOC.
Through the steps S1-S3, the fully-connected neural network and the Transformer neural network are constructed, so that the SOC of the battery can be predicted by taking various characteristics influencing the SOC of the battery as bases, and the estimation precision of the SOC can be obviously improved, so that the safety and the stability of the battery are guaranteed.
Preferably, step S2 includes the steps of,
step SA, using Gaussian distribution
Randomly initializing the hyper-parameters of each network layer;
step SB, constructing a training sample set;
and SC, constructing a Loss function, and continuously updating network parameters through random Gradient descent (Gradient parameter) to enable the Loss of the model to be minimum.
Therefore, the training of the deep learning battery state of charge estimation system can be better realized.
Preferably, in the step SB, simulating the charging and discharging process of the lithium battery under the laboratory condition, grouping according to the material of the battery cell, the production date, the number of the battery cells, the BMS model, the initial SOC, the initial cycle number and the initial SOH, simultaneously collecting the voltage, the current, the temperature and the pressure difference in the charging and discharging process, forming a time sequence, and simultaneously obtaining the real SOC of the current battery by a technical detection means during each sampling; and then, constructing a training sample set by taking the real SOC as a label and taking the rest data as features. A training sample set can be preferably acquired.
Preferably, in step SC, the constructed loss function is an L2 loss function. Therefore, the evaluation of the training effect can be preferably realized.
Drawings
FIG. 1 is a schematic diagram of a transform-based deep learning battery state of charge estimation system according to embodiment 1;
fig. 2 is a schematic diagram of an Encoder network and a Decoder network in embodiment 1;
fig. 3 is a schematic training flow diagram of a transform-based deep learning battery state of charge estimation system in embodiment 1.
Detailed Description
For a further understanding of the invention, reference should be made to the following detailed description taken in conjunction with the accompanying drawings and examples. It is to be understood that the examples are illustrative of the invention and not limiting.
Example 1
Referring to fig. 1, the present embodiment provides a transform-based deep learning battery state of charge estimation system, which includes:
the fully-connected neural network is used for processing and outputting the battery characteristic sequence R and the battery initial state sequence S;
the Transformer neural network is used for processing and outputting the battery charging and discharging process sequence T;
linear fusion layer for output
And output
A stitching and weighting calculation is performed to obtain a predicted battery SOC,
(ii) a And
output layerFor outputting of
。
In this embodiment, the battery characteristic sequence R, the battery initial state sequence S, and the battery charge-discharge process sequence T can be simultaneously constructed.
The characteristic of the battery characteristic sequence R is the inherent characteristics of the battery, such as the cell material, the production date, the cell number, the BMS model and the like, so that different inherent characteristics of the battery can be preferably used as the prediction basis of the SOC, and the characteristics influencing the SOC of the battery can be preferably and fully considered.
The initial state sequence S of the battery is characterized by an initial state when the battery starts to be charged or discharged, such as an initial SOC, an initial cycle number, an initial SOH and the like, so that initial parameters of the battery charging and discharging process can be preferably introduced into the prediction of the SOC, and the characteristics influencing the SOC of the battery can be preferably and fully considered.
The battery charging and discharging process sequence T represents a time sequence of parameters in the battery charging and discharging process, such as voltage, current, differential pressure, temperature and the like, so that the prediction of the SOC according to data on a time axis can be better realized, and the prediction result is more scientific and reasonable.
In addition, in the embodiment, the battery charging and discharging process sequence T can be processed through the Transformer neural network, so that the battery charging and discharging process sequence T has excellent nonlinear fitting capability, and has a faster processing speed compared with a cyclic neural network such as RNN because instructions can be executed in parallel.
In addition, in this embodiment, the battery feature sequence R and the battery initial state sequence S can be processed by the fully-connected neural network, so that the influence of different types and different initial states on the battery SOC can be fully considered, and the deep learning battery state of charge estimation system in this embodiment can have stronger adaptability.
In this embodiment, the outputs of the fully-connected neural network and the transform neural network can be fused and output through the linear fusion layer, so that the influence of the battery characteristics, the battery initial state, and the variation of the parameters of the battery charge and discharge process on the time axis on the SOC of the battery can be considered comprehensively, and the battery can have better robustness.
In this embodiment, the fully-connected neural network can have a plurality of fully-connected layers, and for the l-th fully-connected layer, the formula is:
wherein the content of the first and second substances,
and
respectively representing a weight item and a bias item of the l-th layer full connection layer, which are obtained through training; wherein the content of the first and second substances,
is the input of the l-th fully connected layer,
the output of the l < th > layer full connection layer is used as the input of the l < th > and 1 < th > layer full connection layer;
for activating the function, in the present embodiment, the Relu function is used as the activating function, i.e. the activating function is a function
。
In this embodiment, the characteristics of the battery characteristic sequence R include the battery cell material, the production date, the battery cell number, and the BMS model, and the battery characteristic sequence R is preferably converted into a language convenient for machine processing by performing Embedding operation on the type characteristics; therefore, the collection of the inherent characteristics of the battery can be preferably realized.
In this embodiment, the characteristics in the battery initial state sequence S include an initial SOC, an initial cycle number, and an initial SOH. It is possible to preferably collect values such as initial SOC, initial cycle number, initial SOH, etc. at the start of charging or discharging of the battery.
In this embodiment, by constructing the battery characteristic sequence R and the battery initial state sequence S, the inherent characteristics and the initial state of the battery can be preferably introduced into the determination of the SOC of the battery as influencing factors, so that the determination result is more reliable.
Wherein the final output of the fully-connected neural network is
。
Through the fully-connected neural network constructed in the embodiment, the battery characteristic sequence R and the battery initial state sequence S can be better processed.
As shown in fig. 2, the transform neural network in the present embodiment includes an Encoder network and a Decoder network, the Encoder network includes an Encoder multi-head attention mechanism layer, and the Decoder network includes a Decoder multi-head attention mechanism layer.
In the Encode network, the Encode multi-head attention mechanism layer comprises a plurality of Encode self-attention mechanism layers and a full connection layer, and each Encode self-attention mechanism layer is provided with a parameter matrix
、
And
(ii) a For the mth Encode self-attention mechanism layer in the Encode network, the parameter matrixes are respectively
、
And
。
the mth Encoder self-attention mechanism layer is used for acquiring the following matrix:
wherein the content of the first and second substances,
for the input of the encor multi-head attention mechanism layer, it can be understood that the encor multi-head attention mechanism layer can have a plurality of layers, the input of the first encor multi-head attention mechanism layer is the battery charging and discharging process sequence T, and the output of the previous encor multi-head attention mechanism layer is used as the input of the next encor multi-head attention mechanism layer.
The mth Encoder self-attention mechanism layer is obtaining
、
And
then, it obtains the output matrix by calculating:
wherein the content of the first and second substances,
is a parameter matrix (i.e.
、
And
) Of (c) is calculated.
It is understood that the plurality of Encoder self-attention mechanism layers can output i output matrices, i being the total number of the plurality of self-attention mechanism layers.
The full connection layer can obtain the output of the corresponding Encoder multi-head attention mechanism layer according to the following operation
：
Wherein the content of the first and second substances,
is a parameter matrix of the fully-connected layer,
representing a matrix
And (6) splicing.
In the Decoder network, the Decoder multi-head attention mechanism layer is provided with a plurality of Decoder self-attention mechanism layers, an Encoder-Decoder attention mechanism layer and a full connection layer.
Similar to the Encode self-attention mechanism layer in the Encode network, each Decode self-attention mechanism layer and Encode-Decode self-attention mechanism layer also have parameter matrixes
、
And
(ii) a And in the Encode-Decoder attention mechanism layer, the output of the Encode network is used as the parameter matrix thereof
And
the output of the corresponding Decoder multi-head attention mechanism layer is used as a parameter matrix
。
In the Decoder network, the operation method is the same as that of the Encoder network, and therefore, the description thereof is omitted.
The final output of the Transformer neural network can be obtained through the Decoder network
。
Obtaining the final output of the fully-connected neural network
And final output of the Transformer neural network
Then, the linear fusion layer can obtain the final output according to the following formula
：
Wherein the content of the first and second substances,
is a weight matrix of the linear fusion layer,
is the bias of the linear fused layer and,
for a splicing operation.
Wherein the content of the first and second substances,
i.e. the final predicted battery SOC.
In this embodiment, the constructed battery charging and discharging process sequence T is obtained by sampling N cells during the battery charging or discharging process, and current voltage, current, temperature, and differential pressure data are obtained during each sampling, so the battery charging and discharging process sequence T can be expressed as:
in the embodiment, after each sampling is completed, the battery charging and discharging process sequence T is updated, and after each updating, the prediction can be performed once, so that the predicted SOC of the current battery can be obtained in real time.
The processing of the battery charging and discharging process sequence T can be preferably realized through the above steps.
Based on the above, the embodiment further provides a deep learning battery state of charge estimation system method based on a Transformer, which includes the following steps:
step S1, constructing a deep learning battery state of charge estimation system;
step S2, training a system;
and step S3, in the process of discharging or charging the battery, acquiring a battery characteristic sequence R, a battery initial state sequence S and a battery charging and discharging process sequence T, predicting the SOC of the battery through the trained deep learning battery state-of-charge estimation system and outputting the SOC.
Through the above steps S1-S3, the prediction of the battery SOC can be preferably achieved. The method in the embodiment belongs to a data driving method, and can be used for predicting the SOC of the battery by constructing the full-connection neural network and the Transformer neural network and simultaneously taking various characteristics influencing the SOC of the battery as bases, so that the estimation precision of the SOC can be obviously improved, and the safety and the stability of the battery can be guaranteed.
In addition, after the deep learning battery state of charge estimation system of the embodiment is built, training needs to be performed on the deep learning battery state of charge estimation system, and as shown in fig. 3, the deep learning battery state of charge estimation system specifically includes the following steps:
step SA, using Gaussian distribution
Randomly initializing the hyper-parameters of each network layer;
step SB, constructing a training sample set;
simulating the charging and discharging process of the lithium battery under the condition of a laboratory, grouping according to the material of a battery core, the production date, the number of the battery cores, the BMS model, the initial SOC, the initial cycle number and the initial SOH, simultaneously collecting voltage, current, temperature and pressure difference in the charging and discharging process, forming a time sequence, and simultaneously acquiring the real SOC of the current battery by a technical detection means during each sampling; then, a training sample set is constructed by taking the real SOC as a label and taking the rest data as features;
and SC, constructing a Loss function, and continuously updating network parameters through random Gradient descent (Gradient parameter) to enable the Loss of the model to be minimum.
In step SC, the constructed loss function is an L2 loss function.
By the above, training of the model can be preferably realized.
The present invention and its embodiments have been described above schematically, without limitation, and what is shown in the drawings is only one of the embodiments of the present invention, and the actual structure is not limited thereto. Therefore, if the person skilled in the art receives the teaching, without departing from the spirit of the invention, the person skilled in the art shall not inventively design the similar structural modes and embodiments to the technical solution, but shall fall within the scope of the invention.
Claims (9)
1. Deep learning battery state of charge estimation system based on Transformer, its characterized in that: comprises the steps of (a) preparing a mixture of a plurality of raw materials,
a fully connected neural network for processing and outputting the battery characteristic sequence R and the battery initial state sequence S
；
A Transformer neural network for processing and outputting the battery charging and discharging process sequence T
；
Linear fusion layer for output
And output
A stitching and weighting calculation is performed to obtain a predicted battery SOC,
(ii) a And
output layer for outputting
。
2. The Transformer-based deep learning battery state of charge estimation system of claim 1, wherein: is totally connected withThe neural network is provided with a plurality of full connection layers, and for the l-th layer full connection layer, the formula is as follows:
wherein the content of the first and second substances,
and
respectively representing the weight term and the bias term of the l-th fully-connected layer,
is the input of the l-th fully connected layer,
is the output of the l < th > layer fully-connected layer and serves as the input of the l +1 < th > layer fully-connected layer,
is an activation function.
3. The Transformer-based deep learning battery state of charge estimation system of claim 1, wherein: the Transformer neural network is provided with an Encoder network and a Decoder network, the Encoder network is provided with an Encoder multi-head attention mechanism layer, and the Decoder network is provided with a Decoder multi-head attention mechanism layer.
4. The Transformer-based deep learning battery state of charge estimation system of claim 3, wherein: the Encoder multi-head attention machine layer comprises a plurality of Encoder self-attention machine layers and a full-connection layer, and each Encoder self-attention machine layer is provided with a parameter matrix
、
And
(ii) a For the mth Encode self-attention mechanism layer in the Encode network, the parameter matrixes are respectively
、
And
；
the mth Encoder self-attention mechanism layer is used for acquiring the following matrix:
wherein the content of the first and second substances,
the input of the first Encoder multi-head attention mechanism layer is a battery charging and discharging process sequence T, and the output of the last Encoder multi-head attention mechanism layer is used as the input of the next Encoder multi-head attention mechanism layer;
the first mentionedm Encoder self-attention mechanism layers are obtaining
、
And
then, it obtains the output matrix by calculating:
wherein the content of the first and second substances,
is the dimension of the parameter matrix.
5. The Transformer-based deep learning battery state of charge estimation system of claim 4, wherein: the plurality of Encoder self-attention mechanism layers are used for outputting i output matrixes, wherein i is the total number of the plurality of self-attention mechanism layers; the full connection layer is used for acquiring the output of the corresponding Encoder multi-head attention mechanism layer according to the following operation
：
Wherein the content of the first and second substances,
is a parameter matrix of the fully-connected layer,
representing a matrix
And (6) splicing.
6. The deep learning battery state of charge estimation system method based on the Transformer comprises the following steps:
step S1, constructing the deep learning battery state of charge estimation system according to any one of claims 1 to 5;
step S2, training a system;
and step S3, in the process of discharging or charging the battery, acquiring a battery characteristic sequence R, a battery initial state sequence S and a battery charging and discharging process sequence T, predicting the SOC of the battery through the trained deep learning battery state-of-charge estimation system and outputting the SOC.
7. The Transformer-based deep learning battery state of charge estimation system method of claim 6, wherein: the step S2 includes the steps of,
step SA, using Gaussian distribution
Randomly initializing the hyper-parameters of each network layer;
step SB, constructing a training sample set;
and SC, constructing a Loss function, and continuously updating network parameters through random Gradient descent (Gradient parameter) to enable the Loss of the model to be minimum.
8. The Transformer-based deep learning battery state of charge estimation system method of claim 7, wherein: in the step SB, simulating the charging and discharging process of the lithium battery under the laboratory condition, grouping according to the material of the battery core, the production date, the number of the battery cores, the BMS model, the initial SOC, the initial cycle number and the initial SOH, simultaneously collecting the voltage, the current, the temperature and the pressure difference in the charging and discharging process, forming a time sequence, and simultaneously obtaining the real SOC of the current battery by a technical detection means during each sampling; and then, constructing a training sample set by taking the real SOC as a label and taking the rest data as features.
9. The Transformer-based deep learning battery state of charge estimation system method of claim 7, wherein: in step SC, the constructed loss function is an L2 loss function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111230200.2A CN113673176B (en) | 2021-10-22 | 2021-10-22 | Deep learning battery state of charge estimation system and method based on Transformer |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111230200.2A CN113673176B (en) | 2021-10-22 | 2021-10-22 | Deep learning battery state of charge estimation system and method based on Transformer |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113673176A true CN113673176A (en) | 2021-11-19 |
| CN113673176B CN113673176B (en) | 2022-02-22 |
Family
ID=78550818
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111230200.2A Active CN113673176B (en) | 2021-10-22 | 2021-10-22 | Deep learning battery state of charge estimation system and method based on Transformer |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113673176B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116879761A (en) * | 2023-09-06 | 2023-10-13 | 杭州宇谷科技股份有限公司 | Multi-mode-based battery internal short circuit detection method, system, device and medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107870306A (en) * | 2017-12-11 | 2018-04-03 | 重庆邮电大学 | A kind of lithium battery charge state prediction algorithm based under deep neural network |
| CN110412470A (en) * | 2019-04-22 | 2019-11-05 | 上海博强微电子有限公司 | Electric automobile power battery SOC estimation method |
| CN110488202A (en) * | 2019-07-24 | 2019-11-22 | 北京航空航天大学 | Vehicular battery state-of-charge estimation method based on deep neural network |
| CN112946484A (en) * | 2021-02-07 | 2021-06-11 | 中南大学 | SOC estimation method and system based on BP neural network, terminal equipment and readable storage medium |
| CN113253116A (en) * | 2021-05-18 | 2021-08-13 | 齐鲁工业大学 | Lithium ion battery state of charge estimation method and storage medium |
-
2021
- 2021-10-22 CN CN202111230200.2A patent/CN113673176B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107870306A (en) * | 2017-12-11 | 2018-04-03 | 重庆邮电大学 | A kind of lithium battery charge state prediction algorithm based under deep neural network |
| CN110412470A (en) * | 2019-04-22 | 2019-11-05 | 上海博强微电子有限公司 | Electric automobile power battery SOC estimation method |
| CN110488202A (en) * | 2019-07-24 | 2019-11-22 | 北京航空航天大学 | Vehicular battery state-of-charge estimation method based on deep neural network |
| CN112946484A (en) * | 2021-02-07 | 2021-06-11 | 中南大学 | SOC estimation method and system based on BP neural network, terminal equipment and readable storage medium |
| CN113253116A (en) * | 2021-05-18 | 2021-08-13 | 齐鲁工业大学 | Lithium ion battery state of charge estimation method and storage medium |
Non-Patent Citations (1)
| Title |
|---|
| 倪水平等: "基于一维卷积神经网络与长短期记忆网络结合的电池荷电状态预测方法", 《计算机应用》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116879761A (en) * | 2023-09-06 | 2023-10-13 | 杭州宇谷科技股份有限公司 | Multi-mode-based battery internal short circuit detection method, system, device and medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113673176B (en) | 2022-02-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Tian et al. | A review of the state of health for lithium-ion batteries: Research status and suggestions | |
| Qiao et al. | Toward safe carbon–neutral transportation: Battery internal short circuit diagnosis based on cloud data for electric vehicles | |
| CN109655754B (en) | Battery performance evaluation method based on multi-dimensional grading of charging process | |
| CN111007401A (en) | Electric vehicle battery fault diagnosis method and device based on artificial intelligence | |
| Mamo et al. | Long short-term memory with attention mechanism for state of charge estimation of lithium-ion batteries | |
| CN105866700B (en) | A kind of method that lithium ion battery quickly screens | |
| CN108009585B (en) | Lead-acid battery health state prediction method based on local information fusion | |
| CN110850298B (en) | Lithium battery SOH estimation method and system based on data driving | |
| CN111812515A (en) | XGboost model-based lithium ion battery state of charge estimation | |
| CN112686380A (en) | Neural network-based echelon power cell consistency evaluation method and system | |
| CN113673176B (en) | Deep learning battery state of charge estimation system and method based on Transformer | |
| CN112782594A (en) | Method for estimating SOC (state of charge) of lithium battery by data-driven algorithm considering internal resistance | |
| CN112462282A (en) | Method for determining real-time state of charge of battery pack based on mechanism model | |
| Zhou et al. | Practical State of Health Estimation for LiFePO 4 Batteries Based on Gaussian Mixture Regression and Incremental Capacity Analysis | |
| Xu et al. | State-of-charge estimation and health prognosis for lithium-ion batteries based on temperature-compensated Bi-LSTM network and integrated attention mechanism | |
| Wu et al. | State of health estimation for lithium-ion batteries in real-world electric vehicles | |
| CN113721151B (en) | Battery capacity estimation model and method based on double-tower deep learning network | |
| CN116736172A (en) | Lithium battery pack health state evaluation method based on generation type countermeasure network | |
| CN114325449B (en) | SOH prediction method for lithium ion battery | |
| CN115986229A (en) | Method for judging health degree of battery based on charging segment data | |
| Vatani et al. | Cycling lifetime prediction model for lithium-ion batteries based on artificial neural networks | |
| CN115327389A (en) | Lithium battery SOC estimation method based on genetic algorithm improved double-Kalman filtering | |
| CN114545279A (en) | Lithium battery health state estimation method based on neural network ordinary differential equation | |
| CN112946480A (en) | Lithium battery circuit model simplification method for improving SOC estimation real-time performance | |
| Wang et al. | A hybrid approach on estimating state of charge of lithium-ion batteries based on data driven Model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CP01 | Change in the name or title of a patent holder |
Address after: Room 603, 6th floor, building 1, No.1, Qixian Road, Liangzhu street, Yuhang District, Hangzhou City, Zhejiang Province Patentee after: Hangzhou Yugu Technology Co.,Ltd. Address before: Room 603, 6th floor, building 1, No.1, Qixian Road, Liangzhu street, Yuhang District, Hangzhou City, Zhejiang Province Patentee before: HANGZHOU YUGU TECHNOLOGY Co.,Ltd. |
|
| CP01 | Change in the name or title of a patent holder |