CN114839539A - Lithium battery SOH estimation method based on multilevel sequence information self-adaptive fusion - Google Patents

Lithium battery SOH estimation method based on multilevel sequence information self-adaptive fusion Download PDF

Info

Publication number
CN114839539A
CN114839539A CN202210463421.2A CN202210463421A CN114839539A CN 114839539 A CN114839539 A CN 114839539A CN 202210463421 A CN202210463421 A CN 202210463421A CN 114839539 A CN114839539 A CN 114839539A
Authority
CN
China
Prior art keywords
layer
data
lithium battery
level
fusion
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
Application number
CN202210463421.2A
Other languages
Chinese (zh)
Inventor
鲍政怡
高明裕
何志伟
董哲康
杨宇翔
林辉品
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hangzhou Dianzi University
Original Assignee
Hangzhou Dianzi University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Hangzhou Dianzi University filed Critical Hangzhou Dianzi University
Priority to CN202210463421.2A priority Critical patent/CN114839539A/en
Publication of CN114839539A publication Critical patent/CN114839539A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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]
    • G01R31/378—Arrangements for testing, measuring or monitoring the electrical condition of accumulators or electric batteries, e.g. capacity or state of charge [SoC] specially adapted for the type of battery or accumulator
    • 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]
    • G01R31/392—Determining battery ageing or deterioration, e.g. state of health
    • 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

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Secondary Cells (AREA)

Abstract

The invention provides a lithium battery SOH estimation method based on multilevel sequence information self-adaptive fusion. The invention builds a brand-new serialization model based on deep learning, wherein the model is built by two cascaded multi-level fusion modules and a bidirectional LSTM layer. Based on the advantage that the extracted model can adaptively extract and fuse multi-level serialized information, the model can solve the problems of too little battery data and insufficient extraction to a certain extent, thereby realizing more accurate on-line estimation of the SOH of the lithium battery. Besides, the model has the advantage of long-term memory, which further improves the online estimation accuracy. The experiment adopts battery degradation data in NASA lithium ion data set to carry out simulation verification on the provided network model, and the result shows that the model can ensure higher robustness and accuracy while completing the SOH on-line estimation task of the lithium battery.

Description

Lithium battery SOH estimation method based on multilevel sequence information self-adaptive fusion
Technical Field
The invention belongs to the technical field of battery management, relates to a lithium battery management system technology, and particularly relates to a lithium battery SOH estimation method based on multi-level sequence information self-adaptive fusion.
Background
Lithium batteries are widely used in Electric Vehicles (EVs) due to their long life, high capacity, and wide operating temperature range. In order to ensure the running safety, reliability and durability of the electric automobile, the method is vital to timely and accurately monitor the battery state of the electric automobile. However, long-term, frequent use of the battery will inevitably shorten its service life. In addition, improper charging and use can accelerate battery aging and even cause safety problems. Therefore, accurately estimating the state of health (SOH) of the lithium battery has become a key element for ensuring safe operation of the electric vehicle.
The traditional lithium battery SOH online estimation method mainly comprises a direct measurement method and a model-based method. The direct measurement method is used for measuring aging indexes such as capacity and internal resistance of the lithium battery off line and calculating a formula to obtain an SOH estimated value. However, the method belongs to an open-loop method, is easily influenced by uncertainty factors, and the estimation precision highly depends on a measuring instrument and a measuring technology. The model-based method mainly comprises an electrochemical model method, an equivalent circuit method and an empirical model method, and the SOH of the battery is predicted by establishing a model by considering the material performance and the degradation mechanism of the lithium battery. However, the currently established battery attenuation model is difficult to accurately reflect the attenuation process of the battery, so that the robustness and the accuracy of prediction are poor.
With the rapid development of deep learning and the benefit of the strong data processing capability of the deep neural network, a large number of lithium battery SOH online estimation methods based on the deep neural network emerge. Different from the traditional method, the method based on the neural network regards the modeling process of the battery data as a black box, directly learns the internal dynamic characteristics of the battery through a large amount of charging and discharging data, and establishes the nonlinear relation between the SOH and the original data such as voltage, current, temperature and the like. The Convolutional Neural Network (CNN) is widely applied to online estimation of the SOH of the lithium battery due to the advantages of the convolutional neural network in the aspect of time series prediction, but a single serial network structure of the convolutional neural network cannot extract rich serialized information, so that the model generalization is poor, and the SOH of the lithium battery cannot be accurately estimated.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a lithium battery SOH estimation method based on multi-level sequence information self-adaptive fusion. A prediction model based on a multi-level fusion module and a bidirectional LSTM layer is built to complete the SOH on-line estimation task of the lithium battery. A large number of experiments are carried out, and results show that the method can complete the SOH on-line estimation task of the lithium battery, and meanwhile can ensure higher robustness and estimation accuracy.
The lithium battery SOH online estimation method based on multilevel sequence information self-adaptive fusion specifically comprises the following steps:
step 1, preprocessing the collected original data by adopting a min-max standardization method in the step 1, uniformly mapping the data between [0 and 1], and summarizing the statistical distribution of the sample data of the uniform lithium battery, so as to accelerate the convergence speed of a network model; the normalization formula is as follows:
Figure BDA0003621226230000021
wherein x' is normalized data, x is original data in the data set, and x min Is the maximum value, x, in the original data sample max Is the minimum value in the original data sample.
Step 2, constructing a brand-new network model based on a multi-level fusion module and a bidirectional LSTM layer, and sequentially comprising two cascaded multi-level fusion modules with the same structure, one bidirectional LSTM layer, one Flatten layer and two full-connection layers; the multi-level fusion module comprises a deep level convolution layer and a shallow level convolution layer, and each layer comprises network layers with different depths;
step 3, setting learning rate, training batch size and training period, inputting the data samples in the training set normalized in the step 1 into the network model in the step 2, and storing the network weight parameters at the moment when training is finished; and (3) inputting the test set data sample normalized in the step (1) into the provided network model to verify the SOH online estimation effect of the lithium battery.
Preferably, the two cascaded multilevel fusion modules perform feature extraction; in each multi-level fusion module, designing two parallel convolution layers with different levels of information extraction; the deep convolutional layer comprises three one-dimensional convolutional modules, and each convolutional module is formed by building a convolutional layer, an active layer and a maximum pooling layer; the shallow convolutional layer only comprises one convolutional layer, which can be beneficial to extracting different levels of serialization information by the model; the specific steps of the multilevel fusion module for feature extraction and data fusion are as follows:
(a) the deep layer convolution layer and the shallow layer convolution layer obtain initial extraction information through respective branches; wherein, the deep layer convolution layer obtains richer semantic information, and the shallow layer convolution layer obtains more original data information;
(b) the deep convolution layer output and the shallow convolution layer output fully reflect the internal correlation state of the battery data by self-adaptively fusing the two information, thereby solving the problem of inaccurate estimation result caused by too small battery data volume and insufficient extraction to a certain extent;
(c) the designed multi-level information fusion module H operates as follows:
H(X input )＝F 1 (X input )+F 2 (X input ) (2)
wherein, X input For the original input information of the network model, F 1 Feature extraction operator, F, representing deep convolutional layers 2 Then the feature extraction operator for the shallow convolutional layer is represented.
Preferably, the bidirectional LSTM network output formula is as follows:
h t ＝f(w 1 X t+ w 2 h t-1 ) (3)
h t '＝f(w 3 X t +w 2 h' t+1 ) (4)
O t ＝g(w 4 h 4 +w 6 h' t ) (5)
wherein h is t For the output of the forward layer at time t, h t-1 Is the output of the forward layer at time t-1, h' t Is output at time t in the reverse layer, h' t+1 Is the output of the inversion layer at time t +1, X t For data input at time t, w i As a weight matrix, i is 1-6, O t For the bi-directional LSTM network layer output at time t, f (-) and g (-) represent MLP operators.
Preferably, the battery data used in step 1 is a NASA lithium battery random use data set, which includes B0005, B0006 and B0018 model battery aging attenuation data samples.
Preferably, the network model learning rate in step 3 is set to 0.001, the training batch size is set to 32, and the training period is set to 200.
The invention has the following effects:
the data characteristics with different semantic information can be acquired by using the multi-level fusion module, so that the problem of inaccurate estimation result caused by insufficient extraction of battery data is solved to a certain extent; using bi-directional LSTM can model the serialized information and effectively learn more time series sample data information. The method combines a multi-level fusion module with the bidirectional LSTM, acquires rich semantic information, has the advantage of long-term memory, and further improves the prediction precision. A large number of simulation experiments prove that the method can realize an end-to-end online estimation task and can be applied to serialized information prediction scenes such as electric automobile battery power estimation and the like.
Drawings
FIG. 1 is a flow chart of an SOH online estimation method for a lithium battery based on multilevel sequence information adaptive fusion;
FIG. 2 is a schematic diagram of a network model structure based on a combination of a multi-level fusion module and a bidirectional LSTM;
the battery state of health estimation in the embodiment of fig. 3.
Detailed Description
The invention is further explained below with reference to the drawings.
The experimental environment used in this example was: CPU Intel (R) core (TM) i5-10600KF CPU @4.10Ghz, GPU RTX 3070, graphics card memory 8GB, Python version 3.7, Cuda version 11.1, the deep learning framework used is TensorFlow-GPU 2.3.0, and the data used is from the battery prediction data set of the United states space agency prediction center.
As shown in fig. 1, the lithium battery SOH online estimation method based on multilevel sequence information adaptive fusion specifically includes the following steps:
step 1, in order to determine the battery health state degradation trend of the lithium battery under different working conditions, in this embodiment, data sets such as B0005, B0006, B0018, and B0029 are selected as a training set, and B0005 is selected as a test set. The model of the battery used in the data set is LG Chem 18650 lithium battery, and the battery can normally work under the conditions that the voltage range is 3.2-4.2V and the rated capacity is 2.1 Ah. The associated description indicates that the end of battery life is reached when the capacitance of the battery falls to 70% (1.4Ah) of the nominal capacitance value (2 Ah).
And 2, constructing a network model based on a multi-level fusion module and a bidirectional LSTM layer. The model structure is shown in fig. 2, and specifically, the model includes two multi-level sequence information fusion modules and a bidirectional LSTM layer. In each multi-level sequence information fusion module, two parallel convolution layers with different levels of information extraction are elaborately designed. The deep convolutional layer comprises three one-dimensional convolution modules, and each convolution module is built by a convolutional layer, an active layer and a maximum pooling layer; in contrast, the shallow convolutional layer contains only one convolutional layer, which can facilitate the model to extract different levels of serialization information. In the module, a deep layer convolution layer acquires richer semantic information, a shallow layer convolution layer acquires more original data information, and the two information are adaptively fused to fully reflect the internal correlation state of battery data, so that the problem of inaccurate estimation result caused by too small battery data volume and insufficient extraction is solved to a certain extent. The designed multi-level information fusion module H operates as follows:
H(X input )＝F 1 (X input )+F 2 (X input ) (2)
wherein, X input For the original input information of the network model, F 1 Feature extraction operator, F, representing deep convolutional layers 2 Then the feature extraction operator for the shallow convolutional layer is represented.
The outputs of the deep and shallow convolutional layers are then adaptively merged into a bi-directional LSTM layer with the hidden layer set to 3. The bidirectional LSTM layer is composed of a forward common LSTM layer and a reverse common LSTM layer, and the output of the bidirectional LSTM layer is determined by the two LSTMs together, so that a more accurate estimation result is obtained while long-term memory is realized. The network output formula of the bidirectional LSTM at the time t is as follows:
h t ＝f(w 1 X t+ w 2 h t-1 ) (3)
h t '＝f(w 3 X t +w 2 h' t+1 ) (4)
O t ＝g(w 4 h 4 +w 6 h' t ) (5)
wherein h is t For the output of the forward layer at time t, h t-1 Is the output of the forward layer at time t-1, h' t Is output at time t in the reverse layer, h' t+1 Is the output of the inversion layer at time t +1, X t For data input at time t, w i (i from 1 to 6) as a weight matrix, O t For the bi-directional LSTM network layer output at time t, f (-) and g (-) represent MLP operators.
And finally, outputting the SOH online estimation result of the lithium battery by the data through a Flatten layer and two full-connection layers.
And 3, in order to verify the validity of the provided network model, the lithium battery data set in the step 1 is processed according to the following steps of 3: 1, the original data of the training collector pool is input into the network model based on the multi-level fusion module and the bidirectional LSTM layer, and the on-line estimation effect of the lithium battery SOH of the model is verified. The whole network model is trained by adopting an Adam optimizer, the training learning rate is fixed to be 0.001, and the training period is set to be 200. In order to fully verify the effectiveness of the proposed model, two different loss functions are adopted as training targets in the experiment, and the calculation formula is as follows:
Figure BDA0003621226230000051
Figure BDA0003621226230000052
wherein N is the total number of the original samples of the lithium battery,
Figure BDA0003621226230000053
and Q i And respectively outputting a predicted value and an actual labeled value for the model of the ith data.
And when the training period reaches 200 of the set period, finishing the training, and storing the network weight parameters at the moment to obtain the battery health estimation model.
Step 4, inputting the test set data into the network model trained in the step 3, and comparing the test result with the CNN + LSTM network model, the test result is shown in the following table:
estimation model MAE MSE Error
CNN-LSTM 0.0790 0.0158 0.07
Method for producing a composite material 0.0231 0.0137 0.025
TABLE 1
Wherein MAE and MSE are average absolute Error and mean square Error loss function respectively, and Error is the Error of lithium battery SOH online estimation. As can be seen from Table 1, the lithium battery SOH online estimation method based on multilevel sequence information adaptive fusion provided by the invention can well realize the task of lithium battery SOH online estimation and can ensure higher accuracy. The test results in the data set are shown in fig. 3.

Claims (6)

1. A lithium battery SOH estimation method based on multilevel sequence information self-adaptive fusion is characterized by comprising the following steps:
step 1, normalizing collected battery data;
step 2, constructing a brand-new network model based on a multi-level fusion module and a bidirectional LSTM layer, and sequentially comprising two cascaded multi-level fusion modules with the same structure, one bidirectional LSTM layer, one Flatten layer and two full-connection layers; the multi-level fusion module comprises a deep level convolution layer and a shallow level convolution layer, and each multilayer convolution layer comprises network layers with different depths;
step 3, setting learning rate, training batch size and training period, inputting the data samples in the training set normalized in the step 1 into the network model in the step 2, and storing the network weight parameters at the moment when training is finished; and (3) inputting the test set data sample normalized in the step (1) into the provided network model to verify the SOH online estimation effect of the lithium battery.
2. The lithium battery SOH estimation method based on multi-level sequence information self-adaptive fusion as claimed in claim 1, characterized in that: preprocessing the collected original data by adopting a min-max standardization method in the step 1, uniformly mapping the data between [0 and 1], and summarizing the statistical distribution of the sample data of the uniform lithium battery; the normalization formula is as follows:
Figure FDA0003621226220000011
wherein x' is normalized data, x is raw data in the NASA lithium ion random use dataset, and x is normalized data min Is the maximum value, x, in the original data sample max Is the minimum value in the original data sample.
3. The lithium battery SOH estimation method based on multi-level sequence information self-adaptive fusion as claimed in claim 1, characterized in that: the two cascaded multilevel fusion modules perform feature extraction; in each multi-level fusion module, designing two parallel convolution layers with different-level information extraction; the deep convolutional layer comprises three one-dimensional convolutional modules, and each convolutional module is formed by building a convolutional layer, an active layer and a maximum pooling layer; the shallow layer convolution layer only comprises one convolution layer, and the specific steps of the multi-layer fusion module for feature extraction and data fusion are as follows:
(a) the deep layer convolution layer and the shallow layer convolution layer obtain initial extraction information through respective branches;
(b) the deep convolution layer and the shallow convolution layer output fully reflect the internal correlation state of the battery data by adaptively fusing the information of the deep convolution layer and the shallow convolution layer;
(c) the designed multi-level information fusion module H operates as follows:
H(X input )＝F 1 (X input )+F 2 (X input ) (2)
wherein, X input For the original input information of the network model, F 1 Feature extraction operator, F, representing deep convolutional layers 2 Then the feature extraction operator for the shallow convolutional layer is represented.
4. The lithium battery SOH estimation method based on multi-level sequence information adaptive fusion as claimed in claim 1, characterized in that: the bidirectional LSTM network output formula is as follows:
h t ＝f(w 1 X t+ w 2 h t-1 ) (3)
h t '＝f(w 3 X t +w 2 h' t+1 ) (4)
O t ＝g(w 4 h 4 +w 6 h' t ) (5)
wherein h is t For the output of the forward layer at time t, h t-1 Is the output of the forward layer at time t-1, h' t Is output at time t in the reverse layer, h' t+1 Is the output of the inversion layer at time t +1, X t For data input at time t, w i As a weight matrix, i is 1-6, O t For the bi-directional LSTM network layer output at time t, f (-) and g (-) represent MLP operators.
5. The lithium battery SOH estimation method based on multi-level sequence information adaptive fusion as claimed in claim 1, characterized in that: the battery data used in the step 1 is a NASA lithium battery random use data set, and comprises B0005, B0006 and B0018 model battery aging attenuation data samples.
6. The lithium battery SOH on-line estimation method based on multi-level sequence information self-adaptive fusion as claimed in claim 1, characterized in that: the learning rate of the network model in step 3 is set to 0.001, the size of the training batch is set to 32, and the training period is set to 200.
CN202210463421.2A 2022-04-28 2022-04-28 Lithium battery SOH estimation method based on multilevel sequence information self-adaptive fusion Pending CN114839539A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210463421.2A CN114839539A (en) 2022-04-28 2022-04-28 Lithium battery SOH estimation method based on multilevel sequence information self-adaptive fusion

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210463421.2A CN114839539A (en) 2022-04-28 2022-04-28 Lithium battery SOH estimation method based on multilevel sequence information self-adaptive fusion

Publications (1)

Publication Number Publication Date
CN114839539A true CN114839539A (en) 2022-08-02

Family

ID=82567396

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210463421.2A Pending CN114839539A (en) 2022-04-28 2022-04-28 Lithium battery SOH estimation method based on multilevel sequence information self-adaptive fusion

Country Status (1)

Country Link
CN (1) CN114839539A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116299005A (en) * 2023-02-07 2023-06-23 江南大学 Power battery health state prediction method based on AAF and deep learning

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116299005A (en) * 2023-02-07 2023-06-23 江南大学 Power battery health state prediction method based on AAF and deep learning
CN116299005B (en) * 2023-02-07 2023-09-05 江南大学 Power battery health state prediction method based on AAF and deep learning

Similar Documents

Publication Publication Date Title
WO2021138925A1 (en) Lithium battery capacity estimation method based on improved convolution-long short term memory neural network
CN109061504B (en) Method and system for predicting remaining service life of lithium ion batteries of same type and different types
CN110457789B (en) Lithium ion battery residual life prediction method
Wang et al. A transferable lithium-ion battery remaining useful life prediction method from cycle-consistency of degradation trend
CN113805064B (en) Lithium ion battery pack health state prediction method based on deep learning
CN107870306A (en) A kind of lithium battery charge state prediction algorithm based under deep neural network
EP4113139A1 (en) Lithium battery soc estimation method and apparatus, and computer-readable storage medium
CN111856287A (en) Lithium battery health state detection method based on stacked residual causal convolutional neural network
CN115201686B (en) Lithium ion battery health state assessment method under incomplete charge and discharge data
Huang et al. A comprehensively optimized lithium-ion battery state-of-health estimator based on Local Coulomb Counting Curve
CN113406503A (en) Lithium battery SOH online estimation method based on deep neural network
CN115718263B (en) Attention-based lithium ion battery calendar aging prediction model and method
CN115409263A (en) Method for predicting remaining life of lithium battery based on gating and attention mechanism
CN111537888A (en) Data-driven echelon battery SOC prediction method
CN112163372A (en) SOC estimation method of power battery
CN116679211A (en) Lithium battery health state prediction method
CN114839539A (en) Lithium battery SOH estimation method based on multilevel sequence information self-adaptive fusion
CN116298936A (en) Intelligent lithium ion battery health state prediction method in incomplete voltage range
CN113791351B (en) Lithium battery life prediction method based on transfer learning and difference probability distribution
CN116008844A (en) Vehicle lithium battery health state online prediction method adapting to quick charge strategy
CN114545275A (en) Indirect prediction method for remaining service life of lithium ion battery
Zhou et al. A light-weight feature extractor for lithium-ion battery health prognosis
Wang et al. Remaining useful life prediction of lithium-ion battery based on cycle-consistency learning
CN116298935B (en) Lithium ion battery health state estimation method based on countermeasure encoder network
CN112014757A (en) Battery SOH estimation method integrating capacity increment analysis and genetic wavelet neural network

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