CN115718263A - Attention-based lithium ion battery calendar aging prediction model and method - Google Patents

Attention-based lithium ion battery calendar aging prediction model and method Download PDF

Info

Publication number
CN115718263A
CN115718263A CN202310023441.2A CN202310023441A CN115718263A CN 115718263 A CN115718263 A CN 115718263A CN 202310023441 A CN202310023441 A CN 202310023441A CN 115718263 A CN115718263 A CN 115718263A
Authority
CN
China
Prior art keywords
attention
model
prediction
module
driven
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
Application number
CN202310023441.2A
Other languages
Chinese (zh)
Other versions
CN115718263B (en
Inventor
胡天宇
王康晟
马惠敏
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.)
University of Science and Technology Beijing USTB
Original Assignee
University of Science and Technology Beijing USTB
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 University of Science and Technology Beijing USTB filed Critical University of Science and Technology Beijing USTB
Priority to CN202310023441.2A priority Critical patent/CN115718263B/en
Publication of CN115718263A publication Critical patent/CN115718263A/en
Application granted granted Critical
Publication of CN115718263B publication Critical patent/CN115718263B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E60/00—Enabling technologies; Technologies with a potential or indirect contribution to GHG emissions mitigation
    • Y02E60/10—Energy storage using batteries

Landscapes

  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses an Attention-based lithium ion battery Calendar Aging prediction model and method (KDACAF), which comprises a semi-experience module (SEM module), a Knowledge-driven Attention module, a Data-driven Attention module and a long-time and short-time memory module (LSTM module). According to the lithium ion battery calendar aging prediction model and method based on attention, the knowledge-driven attention module takes a semi-empirical module based on the Alonenius law as a front end, electrochemical priori knowledge in the battery field is integrated into a data-driven neural network, the attention mechanism is applied to lithium ion battery calendar aging prediction by taking human cognitive decision mechanism as reference, monitoring and management of the health state of the battery are facilitated, and the service life of the battery is prolonged.

Description

基于注意力的锂离子电池日历老化预测模型和方法Attention-based calendar aging prediction model and method for lithium-ion batteries

技术领域technical field

本发明涉及一种电池技术，尤其涉及一种基于注意力的锂离子电池日历老化预测模型和方法。The invention relates to a battery technology, in particular to an attention-based lithium-ion battery calendar aging prediction model and method.

背景技术Background technique

锂离子电池由于其在能量密度和低自放电率方面的优势，已被广泛应用于电动汽车（EV）和智能电网等许多工业电子应用中。然而，有效的健康管理是锂离子电池更广泛应用的一个关键和具有挑战性的问题。在电动汽车等实际应用中，电池会随着日历（calendaraging）和循环模式（cycling）而退化。由于超过70%的汽车电池寿命是在储存条件下度过的，因此迫切需要在日历退化模式下进行有效的电池健康监测和管理解决方案。Lithium-ion batteries have been widely used in many industrial electronics applications such as electric vehicles (EVs) and smart grids due to their advantages in energy density and low self-discharge rate. However, effective health management is a critical and challenging issue for the wider application of Li-ion batteries. In practical applications such as electric vehicles, batteries degrade with calendaraging and cycling. Since more than 70% of automotive battery life is spent in storage conditions, there is an urgent need for effective battery health monitoring and management solutions in calendar degradation mode.

在电池日历模式下，电池容量衰减率会受到一些因素的显著影响，包括存储温度和电池充电状态（State of charge，SoC）。由于电池日历退化是一个高度非线性和强耦合的过程，开发适当的模型来诊断/描述不同存储情况下的电池容量退化行为，同时考虑存储温度和SoC的影响，对于电池健康监测和管理至关重要。目前，电池日历老化预测模型可以分为两类，即知识驱动模型和数据驱动模型。In battery calendar mode, the rate of battery capacity decay is significantly affected by several factors, including storage temperature and battery State of charge (SoC). Since battery calendar degradation is a highly nonlinear and strongly coupled process, developing an appropriate model to diagnose/describe battery capacity degradation behavior under different storage conditions, while considering the effects of storage temperature and SoC, is crucial for battery health monitoring and management. important. At present, battery calendar aging prediction models can be divided into two categories, namely, knowledge-driven models and data-driven models.

对于知识驱动的模型，一些能够反映电池退化机制的知识将被耦合到模型中，以解释电池老化的动态。另一方面，基于一些知识信息，如艾林加速方程或阿伦休斯定律，使得半经验模型成为另一种常用的知识驱动模型来捕捉电池日历老化的动态。For the knowledge-driven model, some knowledge that can reflect the battery degradation mechanism will be coupled into the model to explain the dynamics of battery aging. On the other hand, based on some knowledge information, such as Eileen's acceleration equation or Aron Hughes' law, makes the semi-empirical model another commonly used knowledge-driven model to capture the dynamics of battery calendar aging.

随着机器学习和云计算技术的快速发展，数据驱动模型已经成为另一种常用的工具，并实现了电池健康估计和预测。这种类型的模型可以进一步分为传统的统计模型和基于深度学习的模型。前者的容量相对较小，如支持向量回归（SVR）、高斯过程回归（GPR）等。后者采用了深度结构的大容量神经网络，如卷积神经网络、递归神经网络、深度信念网络，以及一些结合递归神经网络和迁移学习的混合模型。With the rapid development of machine learning and cloud computing technologies, data-driven models have become another commonly used tool and enable battery health estimation and prediction. This type of model can be further divided into traditional statistical models and deep learning-based models. The capacity of the former is relatively small, such as support vector regression (SVR), Gaussian process regression (GPR), etc. The latter uses large-capacity neural networks with deep structures, such as convolutional neural networks, recurrent neural networks, deep belief networks, and some hybrid models that combine recurrent neural networks and transfer learning.

目前，通过数据驱动模型进行电池日历老化预测的一个挑战是：如果没有电池电化学经验知识的指导，纯数据驱动模型主要从训练数据中学习电池老化信息，其泛化性较差。由于电池电化学知识可以支持日历老化建模，因此将电池电化学经验知识合并到数据驱动的模型中，应该在提高预测性能方面具有显著的潜力，特别是在缺乏历史数据的全新运行条件下。然而，现有的数据驱动模型在合并来自不同模态的先验知识和统计规律方面的能力有限，即处理多模态输入的能力有限。最近提出的注意力机制已经部分解决了多模态处理问题。注意力机制是模仿人类的认知过程，即选择性地集中于一件或几件事，而忽略其他事情，如自我注意、全局/软注意、局部/硬注意等。预测领域对注意力机制也进行了多次尝试，如提出了混合注意-长短期记忆（LSTM）模型进行光伏功率预测、结合注意力机制和双向LSTM进行电力负荷预测等。At present, one challenge of battery calendar aging prediction through data-driven models is that without the guidance of empirical knowledge of battery electrochemistry, pure data-driven models mainly learn battery aging information from training data, and their generalization is poor. Since knowledge of battery electrochemistry can support calendar aging modeling, incorporating empirical knowledge of battery electrochemistry into data-driven models should have significant potential to improve predictive performance, especially under novel operating conditions where historical data are scarce. However, existing data-driven models are limited in their ability to incorporate prior knowledge and statistical regularities from different modalities, i.e., limited in their ability to handle multimodal inputs. Recently proposed attention mechanisms have partially addressed the multimodal processing problem. The attention mechanism is to imitate the human cognitive process, that is, to selectively focus on one or a few things while ignoring other things, such as self-attention, global/soft attention, local/hard attention, etc. In the field of forecasting, many attempts have been made to the attention mechanism, such as the hybrid attention-long short-term memory (LSTM) model for photovoltaic power forecasting, and the combination of attention mechanism and bidirectional LSTM for power load forecasting.

即使已经出现了这些基于注意力的预测模型，我们认为仍然可以实现有效改进和性能提升，因为这些模型均基于纯数据驱动模型，而较少考虑将领域知识或专业知识融入模型中。Even though these attention-based predictive models have emerged, we believe that effective improvements and performance gains can still be achieved because these models are based on purely data-driven models with little regard for incorporating domain knowledge or expertise into the models.

发明内容Contents of the invention

针对上述问题，在本申请中设计了基于注意力的锂离子电池日历老化预测模型（KDACAF），其包含两个注意力模块，即知识驱动注意力和数据驱动注意力。知识驱动注意力模块以半经验模型为前端，充分利用电池老化电化学经验知识。由于电化学知识可以引导电池日历老化预测建模，在KDACAF中引入先验知识为模型预测带来了显著的性能改进。且由于所提出的知识-数据驱动注意模型由数据驱动和半经验模块组成，本申请主要将其与其他经典数据驱动模型和半经验模型进行性能对比和评价。Aiming at the above problems, an attention-based lithium-ion battery calendar aging prediction model (KDACAF) is designed in this application, which contains two attention modules, knowledge-driven attention and data-driven attention. The knowledge-driven attention module takes the semi-empirical model as the front end and makes full use of the empirical knowledge of battery aging electrochemistry. Since electrochemical knowledge can guide battery calendar aging prediction modeling, introducing prior knowledge in KDACAF brings significant performance improvement for model prediction. And because the proposed knowledge-data-driven attention model is composed of data-driven and semi-empirical modules, this application mainly compares and evaluates its performance with other classic data-driven models and semi-empirical models.

为实现上述目的，本发明提供了基于注意力的锂离子电池日历老化预测模型，包括半经验模块、知识驱动注意力模块、数据驱动注意力模块和长短时记忆模块；To achieve the above object, the present invention provides an attention-based lithium-ion battery calendar aging prediction model, including a semi-empirical module, a knowledge-driven attention module, a data-driven attention module and a long-short-term memory module;

所述知识驱动注意力模块以半经验模型为前端，所述半经验模型基于阿伦尼乌斯定律。The knowledge-driven attention module is fronted by a semi-empirical model based on the Arrhenius law.

本发明还包括用于测试KDACAF有效性的实验平台，所述实验平台包括用于控制存储电池环境温度的热室、用于保持电池预定存储充电状态（SoC）水平的电池测试设备、用于监测和存储电池老化数据的计算机。The invention also includes an experimental platform for testing the effectiveness of KDACAF, which includes a thermal chamber for controlling the ambient temperature of the storage battery, battery testing equipment for maintaining the battery at a predetermined storage state-of-charge (SoC) level, monitoring and a computer that stores battery aging data.

基于注意力的锂离子电池日历老化预测模型的方法，包括以下步骤：The method of the lithium-ion battery calendar aging prediction model based on attention, comprises the following steps:

S1、数据收集和预处理；S1. Data collection and preprocessing;

S2、建立日历老化预测模型；S2. Establishing a calendar aging prediction model;

S21、建立日历老化预测的问题描述和半经验模型；S21. Establishing a problem description and a semi-empirical model for calendar aging prediction;

S22、KDACAF的结构；S22, the structure of KDACAF;

S23、长短时记忆模块与KDACAF的损失函数；S23, the loss function of the long short-term memory module and KDACAF;

S3、实验和分析S3, experiment and analysis

S31、比较测试；S31, comparison test;

S32、消融测试；S32. Ablation test;

S33、收敛性分析。S33. Convergence analysis.

步骤S1具体包括以下步骤：Step S1 specifically includes the following steps:

数据预处理和评估指标Data Preprocessing and Evaluation Metrics

在训练集上进行三次样条插值，使训练集中的每个电池容量序列都有1小时的分辨率，使每个序列的长度为11521，此外，为了保证KDACAF在训练集、验证集和测试集上的时间分辨率一致性，训练集中的所有容量序列每30天稀疏采样一次，即在KDACAF训练期间，所有

Figure 964306DEST_PATH_IMAGE001
中的任意两次连续的
Figure 899026DEST_PATH_IMAGE002
和
Figure 114107DEST_PATH_IMAGE003
之间的时间间隔仍然是30天，即
Figure 23DEST_PATH_IMAGE004
小时，这与验证集和测试集的时间分辨率相同；采用最大绝对误差（maximum absoluteerror, MAE）和均方根误差（root mean square error, RMSE）进行评价，即：Cubic spline interpolation is performed on the training set, so that each sequence of battery capacity in the training set has a resolution of 1 hour, so that the length of each sequence is 11521. In addition, in order to ensure that KDACAF has the best performance in the training set, validation set and test set Temporal resolution consistency on , all capacity sequences in the training set are sparsely sampled every 30 days, i.e., during KDACAF training, all
Figure 964306DEST_PATH_IMAGE001
Any two consecutive
Figure 899026DEST_PATH_IMAGE002
and
Figure 114107DEST_PATH_IMAGE003
The time interval between is still 30 days, i.e.
Figure 23DEST_PATH_IMAGE004
hours, which is the same as the time resolution of the validation set and the test set; the maximum absolute error (MAE) and the root mean square error (RMSE) are used for evaluation, namely:

Figure 60383DEST_PATH_IMAGE005
Figure 60383DEST_PATH_IMAGE005

Figure 348145DEST_PATH_IMAGE006
。
Figure 348145DEST_PATH_IMAGE006
.

步骤S21具体包括以下步骤：Step S21 specifically includes the following steps:

S211、日历老化预测的问题描述S211. Problem description of calendar aging prediction

锂离子电池的日历老化预测旨在预测它们的容量随着存储时间的变化关系，此任务的解释变量格式化为以下矩阵：Calendar aging prediction of lithium-ion batteries aims to predict their capacity as a function of storage time. The explanatory variables for this task are formatted as the following matrix:

Figure 62023DEST_PATH_IMAGE007
Figure 62023DEST_PATH_IMAGE007

其中，

Figure 662332DEST_PATH_IMAGE008
表示本预测任务所有的解释信息，
Figure 526383DEST_PATH_IMAGE009
表示容量序列的顺序序号，
Figure 668651DEST_PATH_IMAGE010
表示容量序列的第
Figure 225534DEST_PATH_IMAGE011
个测量值，
Figure 86043DEST_PATH_IMAGE012
表示
Figure 19364DEST_PATH_IMAGE013
中的电池存储时间，
Figure 16139DEST_PATH_IMAGE014
表示电池存储的温度，
Figure 743923DEST_PATH_IMAGE015
表示存储SoC，
Figure 327614DEST_PATH_IMAGE016
为滞后间隔；in,
Figure 662332DEST_PATH_IMAGE008
Indicates all the interpretation information of this prediction task,
Figure 526383DEST_PATH_IMAGE009
Indicates the sequence number of the capacity sequence,
Figure 668651DEST_PATH_IMAGE010
Indicates the first of the capacity sequence
Figure 225534DEST_PATH_IMAGE011
measured value,
Figure 86043DEST_PATH_IMAGE012
express
Figure 19364DEST_PATH_IMAGE013
battery storage time in,
Figure 16139DEST_PATH_IMAGE014
Indicates the temperature at which the battery is stored,
Figure 743923DEST_PATH_IMAGE015
Indicates storage SoC,
Figure 327614DEST_PATH_IMAGE016
is the lag interval;

对于单步的容量预测，目标变量为

Figure 533467DEST_PATH_IMAGE017
，因此，日历老化预测任务抽象为以下映射：For a one-step capacity forecast, the target variable is
Figure 533467DEST_PATH_IMAGE017
, therefore, the calendar aging prediction task is abstracted as the following mapping:

Figure 119169DEST_PATH_IMAGE018
Figure 119169DEST_PATH_IMAGE018

其中，

Figure 876910DEST_PATH_IMAGE019
表示数据驱动或知识驱动的预测模型,其中
Figure 180852DEST_PATH_IMAGE020
不一定要使用
Figure 190396DEST_PATH_IMAGE021
中的所有信息。例如：半经验模型只使用
Figure 925878DEST_PATH_IMAGE022
中的存储时间、温度和SoC，而纯数据驱动的模型往往使用
Figure 729886DEST_PATH_IMAGE023
中的所有信息。in,
Figure 876910DEST_PATH_IMAGE019
Represents a data-driven or knowledge-driven predictive model, where
Figure 180852DEST_PATH_IMAGE020
don't have to use
Figure 190396DEST_PATH_IMAGE021
All information in . For example: the semi-empirical model uses only
Figure 925878DEST_PATH_IMAGE022
storage time, temperature, and SoC in , while purely data-driven models often use
Figure 729886DEST_PATH_IMAGE023
All information in .

S212、日历老化预测的半经验模型S212. Semi-empirical model for calendar aging prediction

根据阿伦尼乌斯定律，日历容量的退化最终用特定的半经验形式表示：According to the Arrhenius law, the degradation of the calendar capacity is finally expressed in a specific semi-empirical form:

Figure 52283DEST_PATH_IMAGE024
Figure 52283DEST_PATH_IMAGE024

其中，

Figure 599939DEST_PATH_IMAGE025
和
Figure 691391DEST_PATH_IMAGE026
都是SoC的依赖项，
Figure 931880DEST_PATH_IMAGE027
表示气体常数，
Figure 741573DEST_PATH_IMAGE028
表示持续时间依赖性的低功率参数。in,
Figure 599939DEST_PATH_IMAGE025
and
Figure 691391DEST_PATH_IMAGE026
are all SoC dependencies,
Figure 931880DEST_PATH_IMAGE027
is the gas constant,
Figure 741573DEST_PATH_IMAGE028
Indicates the duration-dependent low-power parameter.

考虑到存储SoC对寄生化学反应的影响会导致电池容量的退化，考虑SoC的线性和指数依赖，即

Figure 827341DEST_PATH_IMAGE029
和
Figure 274765DEST_PATH_IMAGE030
的形式，构建以下半经验模型。Considering the impact of storage SoC on parasitic chemical reactions that can lead to degradation of battery capacity, consider the linear and exponential dependence of SoC, i.e.
Figure 827341DEST_PATH_IMAGE029
and
Figure 274765DEST_PATH_IMAGE030
In the form of , construct the following semi-empirical model.

Figure 686154DEST_PATH_IMAGE031
Figure 686154DEST_PATH_IMAGE031

Figure 451985DEST_PATH_IMAGE032
Figure 451985DEST_PATH_IMAGE032

Figure 200498DEST_PATH_IMAGE033
Figure 200498DEST_PATH_IMAGE033

Figure 141910DEST_PATH_IMAGE034
Figure 141910DEST_PATH_IMAGE034

上述所有SEMs 式子可分别简化表示为

Figure 583255DEST_PATH_IMAGE035
，
Figure 977327DEST_PATH_IMAGE036
，
Figure 565084DEST_PATH_IMAGE037
，
Figure 361002DEST_PATH_IMAGE038
，其中的
Figure 379774DEST_PATH_IMAGE039
分别表示所有需要识别的参数，即：All the above SEMs formulas can be simplified as
Figure 583255DEST_PATH_IMAGE035
,
Figure 977327DEST_PATH_IMAGE036
,
Figure 565084DEST_PATH_IMAGE037
,
Figure 361002DEST_PATH_IMAGE038
,one of them
Figure 379774DEST_PATH_IMAGE039
Respectively represent all the parameters that need to be identified, namely:

Figure 385776DEST_PATH_IMAGE040
Figure 385776DEST_PATH_IMAGE040

Figure 617037DEST_PATH_IMAGE041
Figure 617037DEST_PATH_IMAGE041

Figure 657674DEST_PATH_IMAGE042
Figure 657674DEST_PATH_IMAGE042

Figure 581768DEST_PATH_IMAGE043
。
Figure 581768DEST_PATH_IMAGE043
.

步骤S22 KDACAF的结构Step S22 Structure of KDACAF

KDACAF以半经验模块为基础，从中抽取三个分支，即预测分支、拟合分支和特征分支。预测分支通过上述四个SEM对

Figure 75066DEST_PATH_IMAGE044
进行初步预测，拟合分支用于求解SEM对过去容量序列的回归结果（其作为知识驱动注意力模块的输入），特征分支构建了基于上述四个SEM的特征向量（其作为数据驱动注意力模块的输入）。KDACAF is based on the semi-empirical module, from which three branches are extracted, namely prediction branch, fitting branch and feature branch. The predicted branch passes through the above four SEM pairs
Figure 75066DEST_PATH_IMAGE044
Preliminary prediction is made, the fitting branch is used to solve the regression result of SEM on the past capacity sequence (which is used as the input of the knowledge-driven attention module), and the feature branch constructs the feature vector based on the above four SEMs (which is used as the data-driven attention module input of).

步骤S22、具体包括以下步骤：Step S22 specifically includes the following steps:

S221、半经验模块和预测分支S221, semi-empirical module and prediction branch

半经验模块是S212中阐述的四个SEM的集合。The semi-empirical module is a collection of the four SEMs set forth in S212.

在预测分支中，

Figure 375597DEST_PATH_IMAGE045
被输入到上述半经验模块的四个SEM，分别得到
Figure 880528DEST_PATH_IMAGE046
的回归结果，即初步预测结果如下：In the prediction branch,
Figure 375597DEST_PATH_IMAGE045
The four SEMs that are input into the semi-empirical module above, respectively, get
Figure 880528DEST_PATH_IMAGE046
The regression results, that is, the preliminary prediction results are as follows:

Figure 601622DEST_PATH_IMAGE047
Figure 601622DEST_PATH_IMAGE047

即，对于每个SEM分别有：That is, for each SEM respectively:

Figure 457582DEST_PATH_IMAGE048
；
Figure 457582DEST_PATH_IMAGE048
;

S222、拟合分支和知识驱动注意力模块S222, fitting branch and knowledge-driven attention module

为了拟合过去容量序列的结果，首先，将

Figure 155280DEST_PATH_IMAGE049
解释矩阵分为两部分：To fit the results of past capacity series, first, the
Figure 155280DEST_PATH_IMAGE049
The explanation matrix is divided into two parts:

Figure 780296DEST_PATH_IMAGE050
Figure 780296DEST_PATH_IMAGE050

即：Right now:

Figure 170826DEST_PATH_IMAGE051
Figure 170826DEST_PATH_IMAGE051

式中，

Figure 107558DEST_PATH_IMAGE052
表示近期容量序列，
Figure 15471DEST_PATH_IMAGE053
中的每列表示
Figure 494994DEST_PATH_IMAGE054
中对应元素的影响因素；在拟合分支中
Figure 554961DEST_PATH_IMAGE055
被分别输入的前述四个SEM中，得到
Figure 119934DEST_PATH_IMAGE056
的回归结果：In the formula,
Figure 107558DEST_PATH_IMAGE052
represents the recent capacity sequence,
Figure 15471DEST_PATH_IMAGE053
Each column in
Figure 494994DEST_PATH_IMAGE054
Influencing factors of the corresponding elements in ; in the fitting branch
Figure 554961DEST_PATH_IMAGE055
are input into the aforementioned four SEMs respectively, to obtain
Figure 119934DEST_PATH_IMAGE056
The regression result of:

Figure 159435DEST_PATH_IMAGE057
Figure 159435DEST_PATH_IMAGE057

Figure 24622DEST_PATH_IMAGE058
是
Figure 756955DEST_PATH_IMAGE059
在
Figure 809225DEST_PATH_IMAGE060
的回归结果，如下所示：
Figure 24622DEST_PATH_IMAGE058
yes
Figure 756955DEST_PATH_IMAGE059
exist
Figure 809225DEST_PATH_IMAGE060
The regression result of is as follows:

Figure 652416DEST_PATH_IMAGE061
Figure 652416DEST_PATH_IMAGE061

本知识驱动的注意力模块专用于日历老化预测任务，其中预测目标

Figure 840952DEST_PATH_IMAGE062
对每个SEM的注意（即对于其初步预测结果
Figure 714492DEST_PATH_IMAGE063
的注意）是根据每个SEM模型对过去真实容量的拟合优度设计的，即；This knowledge-driven attention module is dedicated to the calendar aging prediction task, where the predicted target
Figure 840952DEST_PATH_IMAGE062
attention to each SEM (i.e., for its preliminary prediction results
Figure 714492DEST_PATH_IMAGE063
Note) is designed according to the goodness of fit of each SEM model to the past true capacity, namely;

首先，该知识驱动的注意力模型在

Figure 378691DEST_PATH_IMAGE064
和
Figure 900940DEST_PATH_IMAGE065
之间的得分函数定义如下：First, the knowledge-driven attention model is
Figure 378691DEST_PATH_IMAGE064
and
Figure 900940DEST_PATH_IMAGE065
The scoring function between is defined as follows:

Figure 599774DEST_PATH_IMAGE066
Figure 599774DEST_PATH_IMAGE066

Figure 549276DEST_PATH_IMAGE067
表示知识驱动的注意力模型的得分函数；
Figure 549276DEST_PATH_IMAGE067
represents the score function of the knowledge-driven attention model;

然后可以获得

Figure 576138DEST_PATH_IMAGE068
对每个SEM的注意力，记为
Figure 26711DEST_PATH_IMAGE069
Then you can get
Figure 576138DEST_PATH_IMAGE068
The attention to each SEM, denoted as
Figure 26711DEST_PATH_IMAGE069

Figure 455418DEST_PATH_IMAGE070
Figure 455418DEST_PATH_IMAGE070

然后，得到对于

Figure 216568DEST_PATH_IMAGE071
的如下几点精确预测：Then, get for
Figure 216568DEST_PATH_IMAGE071
The following points are accurately predicted:

Figure 730726DEST_PATH_IMAGE072
；
Figure 730726DEST_PATH_IMAGE072
;

S223、特征分支和数据驱动注意力模块S223, feature branch and data-driven attention module

为了扩大KDACAF的模型容量，以便更好地捕捉日历老化过程的多模态，构建了数据驱动注意力，细节如下：To expand the model capacity of KDACAF to better capture the multimodality of the calendar aging process, a data-driven attention is constructed, the details are as follows:

为了从半经验模块中构建特征向量，将

Figure 860356DEST_PATH_IMAGE073
再次输入到四个SEM中，并分别从这些SEM的中间变量中构建四个特征向量，如下：To construct feature vectors from semi-empirical modules, the
Figure 860356DEST_PATH_IMAGE073
Input into the four SEMs again, and construct four feature vectors from the intermediate variables of these SEMs, as follows:

Figure 2624DEST_PATH_IMAGE074
Figure 2624DEST_PATH_IMAGE074

其中，

Figure 825086DEST_PATH_IMAGE075
表示基于
Figure 685595DEST_PATH_IMAGE076
内的中间变量所构建的特征。in,
Figure 825086DEST_PATH_IMAGE075
Indicates based on
Figure 685595DEST_PATH_IMAGE076
The features constructed by the intermediate variables in .

在数据驱动注意力模块中，预测目标

Figure 87758DEST_PATH_IMAGE077
对每个SEM的注意力（即对其初步预测结果
Figure 491057DEST_PATH_IMAGE078
的注意力）都是基于历史容量序列和所构建的特征来设计的，得分函数定义为：In the data-driven attention module, predicting the target
Figure 87758DEST_PATH_IMAGE077
Attention to each SEM (i.e. its initial prediction
Figure 491057DEST_PATH_IMAGE078
attention) are designed based on historical capacity sequences and constructed features, and the scoring function is defined as:

Figure 343475DEST_PATH_IMAGE079
Figure 343475DEST_PATH_IMAGE079

其中

Figure 301067DEST_PATH_IMAGE080
表示该数据驱动的注意力的得分函数，
Figure 664178DEST_PATH_IMAGE081
是第
Figure 390825DEST_PATH_IMAGE082
个SEM对应的可训练矩阵；in
Figure 301067DEST_PATH_IMAGE080
represents the score function for this data-driven attention,
Figure 664178DEST_PATH_IMAGE081
is the first
Figure 390825DEST_PATH_IMAGE082
A trainable matrix corresponding to SEM;

然后，可得到

Figure 148566DEST_PATH_IMAGE083
对每个SEM的注意力数值，记为
Figure 718087DEST_PATH_IMAGE084
，即
Figure 55528DEST_PATH_IMAGE085
Then, you can get
Figure 148566DEST_PATH_IMAGE083
The value of attention for each SEM, denoted as
Figure 718087DEST_PATH_IMAGE084
,Right now
Figure 55528DEST_PATH_IMAGE085

Figure 259851DEST_PATH_IMAGE086
Figure 259851DEST_PATH_IMAGE086

然后，得到对于

Figure 63859DEST_PATH_IMAGE087
的另一个精确的预测：Then, get for
Figure 63859DEST_PATH_IMAGE087
Another exact prediction for :

Figure 386256DEST_PATH_IMAGE088
。
Figure 386256DEST_PATH_IMAGE088
.

步骤S23 长短时记忆模块与KDACAF的损失函数Step S23 Loss function of long short-term memory module and KDACAF

综合两部分注意力机制可以得到

Figure 199491DEST_PATH_IMAGE089
的中间预测：Combining the two parts of the attention mechanism can be obtained
Figure 199491DEST_PATH_IMAGE089
The intermediate prediction of :

Figure 431889DEST_PATH_IMAGE090
Figure 431889DEST_PATH_IMAGE090

其中

Figure 531432DEST_PATH_IMAGE091
表示两个注意力模块对
Figure 75546DEST_PATH_IMAGE092
的中间预测，
Figure 256254DEST_PATH_IMAGE093
和
Figure 936634DEST_PATH_IMAGE094
均为可学习的权重；in
Figure 531432DEST_PATH_IMAGE091
Denotes two attention module pairs
Figure 75546DEST_PATH_IMAGE092
the intermediate forecast of
Figure 256254DEST_PATH_IMAGE093
and
Figure 936634DEST_PATH_IMAGE094
are learnable weights;

然后，将历史容量序列

Figure 675920DEST_PATH_IMAGE095
和
Figure 317117DEST_PATH_IMAGE096
连接为如下矢量：Then, the historical capacity sequence
Figure 675920DEST_PATH_IMAGE095
and
Figure 317117DEST_PATH_IMAGE096
Concatenated as a vector like this:

Figure 596788DEST_PATH_IMAGE097
Figure 596788DEST_PATH_IMAGE097

最后，上述矢量被输入到包含一个LSTM层的长短时记忆模块，而LSTM的输出在被缩放到

Figure 272620DEST_PATH_IMAGE098
之间后作为
Figure 749519DEST_PATH_IMAGE099
的最终预测，表示为
Figure 143591DEST_PATH_IMAGE100
；Finally, the above vectors are input to a long short-term memory module containing an LSTM layer, and the output of the LSTM is scaled to
Figure 272620DEST_PATH_IMAGE098
Between later as
Figure 749519DEST_PATH_IMAGE099
The final prediction of , expressed as
Figure 143591DEST_PATH_IMAGE100
;

Figure 961374DEST_PATH_IMAGE101
Figure 961374DEST_PATH_IMAGE101

KDACAF的损失函数设置如下：The loss function of KDACAF is set as follows:

Figure 22871DEST_PATH_IMAGE102
Figure 22871DEST_PATH_IMAGE102

式中,

Figure 776064DEST_PATH_IMAGE103
为训练样本的数量，采用Adam训练方法训练KDACAF；In the formula,
Figure 776064DEST_PATH_IMAGE103
For the number of training samples, the Adam training method is used to train KDACAF;

步骤S31具体包括以下步骤：Step S31 specifically includes the following steps:

S311、在测试集Set†上的比较；S311, comparison on the test set Set†;

S312、在测试集Set‡上的比较。S312. Comparison on the test set Set‡.

因此，本发明采用上述结构具有以下有益效果：Therefore, the present invention adopts above-mentioned structure to have following beneficial effect:

1、通过将电化学知识作为知识驱动注意力模块的关键基础，KDACAF实现了基于知识-数据双驱动的精准电池日历老化预测。消融实验表明，电化学领域知识的引入显著提高了KDACAF的预测性能。1. By using electrochemical knowledge as the key basis of the knowledge-driven attention module, KDACAF realizes accurate battery calendar aging prediction based on knowledge-data dual drive. Ablation experiments show that the introduction of electrochemical domain knowledge significantly improves the predictive performance of KDACAF.

2、多次比较测试表明，KDACAF优于当前最先进的知识驱动和数据驱动的电池日历老化预测模型。2. Multiple comparison tests show that KDACAF outperforms the current state-of-the-art knowledge-driven and data-driven battery calendar aging prediction models.

下面通过附图和实施例，对本发明的技术方案做进一步的详细描述。The technical solutions of the present invention will be described in further detail below with reference to the accompanying drawings and embodiments.

附图说明Description of drawings

图1为本发明的结构框图；Fig. 1 is a block diagram of the present invention;

图2为本发明的实验平台结构框图；Fig. 2 is a structural block diagram of the experimental platform of the present invention;

图3为本发明的数据处理流程图；Fig. 3 is the data processing flowchart of the present invention;

图4为本发明的长短时记忆模块图；Fig. 4 is a long short-term memory module figure of the present invention;

图5为本发明在测试集set†上的预测结果和相应的预测误差关系图；Fig. 5 is the prediction result of the present invention on the test set† and the corresponding prediction error relationship diagram;

图6为本发明在测试集set‡上的预测结果和相应的预测误差关系图；Fig. 6 is the prediction result of the present invention on the test set set‡ and the corresponding prediction error relationship diagram;

图7为本发明在测试集set†上的收敛性和稳定性分析图；Fig. 7 is the convergence and stability analysis diagram of the present invention on the test set†;

图8为本发明在测试集set‡上的收敛性和稳定性分析图。Fig. 8 is a graph of convergence and stability analysis of the present invention on the test set‡.

具体实施方式Detailed ways

以下将结合附图对本发明作进一步的描述，需要说明的是，本实施例以本技术方案为前提，给出了详细的实施方式和具体的操作过程，但本发明的保护范围并不限于本实施例。The present invention will be further described below in conjunction with the accompanying drawings. It should be noted that this embodiment is based on the technical solution, and provides detailed implementation and specific operation process, but the protection scope of the present invention is not limited to the present invention. Example.

图1为本发明的实施例一种的结构示意图，如图1所示，本发明的结构包括半经验模块、知识驱动注意力模块、数据驱动注意力模块和长短时记忆模块；Fig. 1 is a schematic structural view of an embodiment of the present invention, as shown in Fig. 1, the structure of the present invention includes a semi-empirical module, a knowledge-driven attention module, a data-driven attention module and a long-short-term memory module;

所述知识驱动注意力模块以半经验模型为前端，所述半经验模型基于阿伦尼乌斯定律。The knowledge-driven attention module is fronted by a semi-empirical model based on the Arrhenius law.

本发明还包括用于测试KDACAF有效性的实验平台，所述实验平台包括用于控制存储电池环境温度的热室、用于保持电池预定存储充电状态（SoC）水平的电池测试设备、用于监测和存储电池老化数据的计算机。The invention also includes an experimental platform for testing the effectiveness of KDACAF, which includes a thermal chamber for controlling the ambient temperature of the storage battery, battery testing equipment for maintaining the battery at a predetermined storage state-of-charge (SoC) level, monitoring and a computer that stores battery aging data.

基于注意力的锂离子电池日历老化预测模型的方法，包括以下步骤：The method of the lithium-ion battery calendar aging prediction model based on attention, comprises the following steps:

S1、数据收集和预处理；S1. Data collection and preprocessing;

表1为日历老化数据集表Table 1 is the calendar aging data set table

Figure 782066DEST_PATH_IMAGE104
Figure 782066DEST_PATH_IMAGE104

步骤S1具体包括以下步骤：Step S1 specifically includes the following steps:

数据预处理和评估指标Data Preprocessing and Evaluation Metrics

在训练KDACAF之前，将27个容量序列分为四个子集，即序列13、14、16、17、22、23、25、26在8000小时内的作为训练集，序列15、18、24、27在8000小时内的作为验证集，Con.#5、6、8、9中时间戳超过8000小时的作为测试集set†，Con.#1、2、3、4、7作为测试集set‡。Before training KDACAF, the 27 capacity sequences are divided into four subsets, that is, sequences 13, 14, 16, 17, 22, 23, 25, and 26 are used as training sets within 8000 hours, and sequences 15, 18, 24, and 27 Within 8000 hours as the validation set, Con.#5, 6, 8, and 9 as the test set†, and Con.#1, 2, 3, 4, and 7 as the test set‡.

在训练集上进行三次样条插值，使训练集中的每个电池容量序列都有1小时的分辨率，使每个序列的长度为11521，此外，为了保证KDACAF在训练集、验证集和测试集上的时间分辨率一致性，训练集中的所有容量序列每30天稀疏采样一次，即在KDACAF训练期间，所有

Figure 278906DEST_PATH_IMAGE105
中的任意两次连续的
Figure 53964DEST_PATH_IMAGE106
和
Figure 243637DEST_PATH_IMAGE107
之间的时间间隔仍然是30天，即
Figure 972821DEST_PATH_IMAGE108
小时，这与验证集和测试集的时间分辨率相同；采用最大绝对误差（maximum absoluteerror, MAE）和均方根误差（root mean square error, RMSE）进行评价，即：Cubic spline interpolation is performed on the training set, so that each sequence of battery capacity in the training set has a resolution of 1 hour, so that the length of each sequence is 11521. In addition, in order to ensure that KDACAF has the best performance in the training set, validation set and test set Temporal resolution consistency on , all capacity sequences in the training set are sparsely sampled every 30 days, i.e., during KDACAF training, all
Figure 278906DEST_PATH_IMAGE105
Any two consecutive
Figure 53964DEST_PATH_IMAGE106
and
Figure 243637DEST_PATH_IMAGE107
The time interval between is still 30 days, i.e.
Figure 972821DEST_PATH_IMAGE108
hours, which is the same as the time resolution of the validation set and the test set; the maximum absolute error (MAE) and the root mean square error (RMSE) are used for evaluation, namely:

Figure 7773DEST_PATH_IMAGE109
Figure 7773DEST_PATH_IMAGE109

S2、建立日历老化预测模型；S2. Establishing a calendar aging prediction model;

S21、建立日历老化预测的问题描述和半经验模型；S21. Establishing a problem description and a semi-empirical model for calendar aging prediction;

步骤S21具体包括以下步骤：Step S21 specifically includes the following steps:

S211、日历老化预测的问题描述S211. Problem description of calendar aging prediction

锂离子电池的日历老化预测旨在预测它们的容量随着存储时间的变化关系，此任务的解释变量格式化为以下矩阵：Calendar aging prediction of lithium-ion batteries aims to predict their capacity as a function of storage time. The explanatory variables for this task are formatted as the following matrix:

Figure 371758DEST_PATH_IMAGE110
Figure 371758DEST_PATH_IMAGE110

其中，

Figure 466753DEST_PATH_IMAGE111
表示所有的解释信息，
Figure 181768DEST_PATH_IMAGE112
表示容量序列的顺序序号，
Figure 20411DEST_PATH_IMAGE113
表示容量序列的第
Figure 770061DEST_PATH_IMAGE114
个测量值，
Figure 35958DEST_PATH_IMAGE115
表示
Figure 2383DEST_PATH_IMAGE116
中的电池存储时间，
Figure 379138DEST_PATH_IMAGE117
表示电池存储的温度，
Figure 983295DEST_PATH_IMAGE118
表示存储充电状态（SoC），
Figure 685671DEST_PATH_IMAGE119
为滞后间隔；in,
Figure 466753DEST_PATH_IMAGE111
represents all interpretation information,
Figure 181768DEST_PATH_IMAGE112
Indicates the sequence number of the capacity sequence,
Figure 20411DEST_PATH_IMAGE113
Indicates the first of the capacity sequence
Figure 770061DEST_PATH_IMAGE114
measured value,
Figure 35958DEST_PATH_IMAGE115
express
Figure 2383DEST_PATH_IMAGE116
battery storage time in,
Figure 379138DEST_PATH_IMAGE117
Indicates the temperature at which the battery is stored,
Figure 983295DEST_PATH_IMAGE118
Indicates the storage state of charge (SoC),
Figure 685671DEST_PATH_IMAGE119
is the lag interval;

对于单步的容量预测，目标变量为

Figure 250645DEST_PATH_IMAGE120
，因此，日历老化预测任务抽象为以下映射：For a one-step capacity forecast, the target variable is
Figure 250645DEST_PATH_IMAGE120
, therefore, the calendar aging prediction task is abstracted as the following mapping:

Figure 24566DEST_PATH_IMAGE121
Figure 24566DEST_PATH_IMAGE121

其中，

Figure 217650DEST_PATH_IMAGE122
表示数据驱动或知识驱动的预测模型,其中
Figure 576081DEST_PATH_IMAGE123
不一定要使用
Figure 221826DEST_PATH_IMAGE124
中的所有信息。例如：半经验模型只使用
Figure 205963DEST_PATH_IMAGE125
中的存储时间、温度和SoC，而纯数据驱动的模型往往使用
Figure 277387DEST_PATH_IMAGE126
中的所有信息。in,
Figure 217650DEST_PATH_IMAGE122
Represents a data-driven or knowledge-driven predictive model, where
Figure 576081DEST_PATH_IMAGE123
don't have to use
Figure 221826DEST_PATH_IMAGE124
All information in . For example: the semi-empirical model uses only
Figure 205963DEST_PATH_IMAGE125
storage time, temperature, and SoC in , while purely data-driven models often use
Figure 277387DEST_PATH_IMAGE126
All information in .

S212、日历老化预测的半经验模型S212. Semi-empirical model for calendar aging prediction

根据阿伦尼乌斯定律，日历容量的退化最终用特定的半经验形式表示：According to the Arrhenius law, the degradation of the calendar capacity is finally expressed in a specific semi-empirical form:

Figure 55987DEST_PATH_IMAGE127
Figure 55987DEST_PATH_IMAGE127

其中，

Figure 189029DEST_PATH_IMAGE128
和
Figure 976856DEST_PATH_IMAGE129
都是SoC的依赖项，
Figure 144532DEST_PATH_IMAGE130
表示气体常数，
Figure 953088DEST_PATH_IMAGE131
表示持续时间依赖性的低功率参数。in,
Figure 189029DEST_PATH_IMAGE128
and
Figure 976856DEST_PATH_IMAGE129
are all SoC dependencies,
Figure 144532DEST_PATH_IMAGE130
is the gas constant,
Figure 953088DEST_PATH_IMAGE131
Indicates the duration-dependent low-power parameter.

考虑到存储SoC对寄生化学反应的影响会导致电池容量的退化，考虑SoC的线性和指数依赖，即

Figure 979950DEST_PATH_IMAGE132
和
Figure 666409DEST_PATH_IMAGE133
的形式，构建以下半经验模型。Considering the impact of storage SoC on parasitic chemical reactions that can lead to degradation of battery capacity, consider the linear and exponential dependence of SoC, i.e.
Figure 979950DEST_PATH_IMAGE132
and
Figure 666409DEST_PATH_IMAGE133
In the form of , construct the following semi-empirical model.

Figure 95116DEST_PATH_IMAGE134
Figure 95116DEST_PATH_IMAGE134

Figure 340152DEST_PATH_IMAGE135
Figure 340152DEST_PATH_IMAGE135

Figure 588731DEST_PATH_IMAGE136
Figure 588731DEST_PATH_IMAGE136

Figure 718361DEST_PATH_IMAGE137
Figure 718361DEST_PATH_IMAGE137

上述所有SEMs 式子可分别简化表示为

Figure 391788DEST_PATH_IMAGE138
，
Figure 306261DEST_PATH_IMAGE139
，
Figure 307715DEST_PATH_IMAGE140
，
Figure 100091DEST_PATH_IMAGE141
，其中的
Figure 237811DEST_PATH_IMAGE142
分别表示所有需要识别的参数，即：All the above SEMs formulas can be simplified as
Figure 391788DEST_PATH_IMAGE138
,
Figure 306261DEST_PATH_IMAGE139
,
Figure 307715DEST_PATH_IMAGE140
,
Figure 100091DEST_PATH_IMAGE141
,one of them
Figure 237811DEST_PATH_IMAGE142
Respectively represent all the parameters that need to be identified, namely:

Figure 824650DEST_PATH_IMAGE143
Figure 824650DEST_PATH_IMAGE143

Figure 313400DEST_PATH_IMAGE144
Figure 313400DEST_PATH_IMAGE144

Figure 909467DEST_PATH_IMAGE145
Figure 909467DEST_PATH_IMAGE145

Figure 901693DEST_PATH_IMAGE146
。
Figure 901693DEST_PATH_IMAGE146
.

S22、KDACAF的结构；S22, the structure of KDACAF;

KDACAF以半经验模块为基础，从中抽取三个分支，即预测分支、拟合分支和特征分支。预测分支通过上述四个SEM对

Figure 800379DEST_PATH_IMAGE147
进行初步预测，拟合分支用于求解SEM对过去容量序列的回归结果（其作为知识驱动注意力模块的输入），特征分支构建了基于上述四个SEM的特征向量（其作为数据驱动注意力模块的输入）。KDACAF is based on the semi-empirical module, from which three branches are extracted, namely prediction branch, fitting branch and feature branch. The predicted branch passes through the above four SEM pairs
Figure 800379DEST_PATH_IMAGE147
Preliminary prediction is made, the fitting branch is used to solve the regression result of SEM on the past capacity sequence (which is used as the input of the knowledge-driven attention module), and the feature branch constructs the feature vector based on the above four SEMs (which is used as the data-driven attention module input of).

步骤S22具体包括以下步骤：Step S22 specifically includes the following steps:

S221、半经验模块和预测分支S221, semi-empirical module and prediction branch

半经验模块是S212中阐述的四个SEM的集合。The semi-empirical module is a collection of the four SEMs set forth in S212.

在预测分支中，

Figure 402524DEST_PATH_IMAGE148
被输入到上述半经验模块的四个SEM，分别得到
Figure 412068DEST_PATH_IMAGE149
的回归结果，即初步预测结果如下：In the prediction branch,
Figure 402524DEST_PATH_IMAGE148
The four SEMs that are input into the semi-empirical module above, respectively, get
Figure 412068DEST_PATH_IMAGE149
The regression results, that is, the preliminary prediction results are as follows:

Figure 524381DEST_PATH_IMAGE150
Figure 524381DEST_PATH_IMAGE150

即，对于每个SEM分别有：That is, for each SEM respectively:

Figure 718602DEST_PATH_IMAGE151
；
Figure 718602DEST_PATH_IMAGE151
;

S222、拟合分支和知识驱动注意力模块S222, fitting branch and knowledge-driven attention module

为了拟合过去容量序列的结果，首先，将

Figure 916365DEST_PATH_IMAGE152
解释矩阵分为两部分：To fit the results of past capacity series, first, the
Figure 916365DEST_PATH_IMAGE152
The explanation matrix is divided into two parts:

Figure 854234DEST_PATH_IMAGE153
Figure 854234DEST_PATH_IMAGE153

即：Right now:

Figure 821053DEST_PATH_IMAGE154
Figure 821053DEST_PATH_IMAGE154

式中，

Figure 61541DEST_PATH_IMAGE155
表示近期容量序列，
Figure 297000DEST_PATH_IMAGE156
中的每列表示
Figure 507402DEST_PATH_IMAGE157
中对应元素的影响因素；在拟合分支中
Figure 63148DEST_PATH_IMAGE158
被分别输入的前述四个SEM中，得到
Figure 835057DEST_PATH_IMAGE159
的回归结果：In the formula,
Figure 61541DEST_PATH_IMAGE155
represents the recent capacity sequence,
Figure 297000DEST_PATH_IMAGE156
Each column in
Figure 507402DEST_PATH_IMAGE157
Influencing factors of the corresponding elements in ; in the fitting branch
Figure 63148DEST_PATH_IMAGE158
are input into the aforementioned four SEMs respectively, to obtain
Figure 835057DEST_PATH_IMAGE159
The regression result of:

Figure 866467DEST_PATH_IMAGE160
Figure 866467DEST_PATH_IMAGE160

Figure 287084DEST_PATH_IMAGE161
是
Figure 962916DEST_PATH_IMAGE162
在
Figure 935420DEST_PATH_IMAGE163
的回归结果，如下所示：
Figure 287084DEST_PATH_IMAGE161
yes
Figure 962916DEST_PATH_IMAGE162
exist
Figure 935420DEST_PATH_IMAGE163
The regression result of is as follows:

Figure 329493DEST_PATH_IMAGE164
Figure 329493DEST_PATH_IMAGE164

本知识驱动的注意力模块专用于日历老化预测任务，其中预测目标

Figure 288221DEST_PATH_IMAGE165
对每个SEM的注意（即对于其初步预测结果
Figure 208773DEST_PATH_IMAGE166
的注意）是根据每个SEM模型对过去真实容量的拟合优度设计的，即；This knowledge-driven attention module is dedicated to the calendar aging prediction task, where the predicted target
Figure 288221DEST_PATH_IMAGE165
attention to each SEM (i.e., for its preliminary prediction results
Figure 208773DEST_PATH_IMAGE166
Note) is designed according to the goodness of fit of each SEM model to the past true capacity, namely;

首先，该知识驱动的注意力模型在

Figure 696386DEST_PATH_IMAGE167
和
Figure 200923DEST_PATH_IMAGE168
之间的得分函数定义如下：First, the knowledge-driven attention model is
Figure 696386DEST_PATH_IMAGE167
and
Figure 200923DEST_PATH_IMAGE168
The scoring function between is defined as follows:

Figure 963343DEST_PATH_IMAGE169
Figure 963343DEST_PATH_IMAGE169

Figure 941663DEST_PATH_IMAGE170
表示知识驱动的注意力模型的得分函数；
Figure 941663DEST_PATH_IMAGE170
represents the score function of the knowledge-driven attention model;

然后可以获得

Figure 255970DEST_PATH_IMAGE171
对每个SEM的注意力，记为
Figure 359055DEST_PATH_IMAGE172
Then you can get
Figure 255970DEST_PATH_IMAGE171
The attention to each SEM, denoted as
Figure 359055DEST_PATH_IMAGE172

Figure 518641DEST_PATH_IMAGE173
Figure 518641DEST_PATH_IMAGE173

然后，得到对于

Figure 554730DEST_PATH_IMAGE174
的如下几点精确预测：Then, get for
Figure 554730DEST_PATH_IMAGE174
The following points are accurately predicted:

Figure 649725DEST_PATH_IMAGE175
；
Figure 649725DEST_PATH_IMAGE175
;

S223、特征分支和数据驱动注意力模块S223, feature branch and data-driven attention module

为了扩大KDACAF的模型容量，以便更好地捕捉日历老化过程的多模态，构建了数据驱动注意力，细节如下：To expand the model capacity of KDACAF to better capture the multimodality of the calendar aging process, a data-driven attention is constructed, the details are as follows:

为了从半经验模块中构建特征向量，将

Figure 131784DEST_PATH_IMAGE176
再次输入到四个SEM中，并分别从这些SEM的中间变量中构建四个特征向量，如下：To construct feature vectors from semi-empirical modules, the
Figure 131784DEST_PATH_IMAGE176
Input into the four SEMs again, and construct four feature vectors from the intermediate variables of these SEMs, as follows:

Figure 970427DEST_PATH_IMAGE177
Figure 970427DEST_PATH_IMAGE177

其中，

Figure 861023DEST_PATH_IMAGE178
表示基于
Figure 251553DEST_PATH_IMAGE179
内的中间变量所构建的特征。in,
Figure 861023DEST_PATH_IMAGE178
Indicates based on
Figure 251553DEST_PATH_IMAGE179
The features constructed by the intermediate variables in .

在数据驱动注意力模块中，预测目标

Figure 329230DEST_PATH_IMAGE180
对每个SEM的注意力（即对其初步预测结果
Figure 96198DEST_PATH_IMAGE181
的注意力）都是基于历史容量序列和所构建的特征来设计的，得分函数定义为：In the data-driven attention module, predicting the target
Figure 329230DEST_PATH_IMAGE180
Attention to each SEM (i.e. its initial prediction
Figure 96198DEST_PATH_IMAGE181
attention) are designed based on historical capacity sequences and constructed features, and the scoring function is defined as:

Figure 575721DEST_PATH_IMAGE079
Figure 575721DEST_PATH_IMAGE079

其中

Figure 278098DEST_PATH_IMAGE182
表示该数据驱动的注意力的得分函数，
Figure 967705DEST_PATH_IMAGE183
是第
Figure 413730DEST_PATH_IMAGE184
个SEM对应的可训练矩阵；in
Figure 278098DEST_PATH_IMAGE182
represents the score function for this data-driven attention,
Figure 967705DEST_PATH_IMAGE183
is the first
Figure 413730DEST_PATH_IMAGE184
A trainable matrix corresponding to SEM;

然后，可得到

Figure 201006DEST_PATH_IMAGE185
对每个SEM的注意力数值，记为
Figure 808705DEST_PATH_IMAGE186
，即
Figure 126554DEST_PATH_IMAGE187
Then, you can get
Figure 201006DEST_PATH_IMAGE185
The value of attention for each SEM, denoted as
Figure 808705DEST_PATH_IMAGE186
,Right now
Figure 126554DEST_PATH_IMAGE187

Figure 235324DEST_PATH_IMAGE188
Figure 235324DEST_PATH_IMAGE188

然后，得到对于

Figure 689439DEST_PATH_IMAGE189
的另一个精确的预测：Then, get for
Figure 689439DEST_PATH_IMAGE189
Another exact prediction for :

Figure 592673DEST_PATH_IMAGE190
Figure 592673DEST_PATH_IMAGE190

S23、长短时记忆模块与KDACAF的损失函数；S23, the loss function of the long short-term memory module and KDACAF;

综合两部分注意力机制可以得到

Figure 397818DEST_PATH_IMAGE191
的中间预测：Combining the two parts of the attention mechanism can be obtained
Figure 397818DEST_PATH_IMAGE191
The intermediate prediction of :

Figure 920067DEST_PATH_IMAGE192
Figure 920067DEST_PATH_IMAGE192

其中

Figure 353322DEST_PATH_IMAGE193
表示两个注意力模块对
Figure 568403DEST_PATH_IMAGE194
的中间预测，
Figure 221363DEST_PATH_IMAGE195
和
Figure 281723DEST_PATH_IMAGE196
均为可学习的权重；in
Figure 353322DEST_PATH_IMAGE193
Denotes two attention module pairs
Figure 568403DEST_PATH_IMAGE194
the intermediate forecast of
Figure 221363DEST_PATH_IMAGE195
and
Figure 281723DEST_PATH_IMAGE196
are learnable weights;

然后，将历史容量序列

Figure 303906DEST_PATH_IMAGE197
和
Figure 955467DEST_PATH_IMAGE198
连接为如下矢量：Then, the historical capacity sequence
Figure 303906DEST_PATH_IMAGE197
and
Figure 955467DEST_PATH_IMAGE198
Concatenated as a vector like this:

Figure 328679DEST_PATH_IMAGE199
Figure 328679DEST_PATH_IMAGE199

最后，上述矢量被输入到包含一个LSTM层的长短时记忆模块，而LSTM的输出在被缩放到

Figure 192730DEST_PATH_IMAGE200
之间后作为
Figure 741523DEST_PATH_IMAGE201
的最终预测，表示为
Figure 157461DEST_PATH_IMAGE202
；Finally, the above vectors are input to a long short-term memory module containing an LSTM layer, and the output of the LSTM is scaled to
Figure 192730DEST_PATH_IMAGE200
Between later as
Figure 741523DEST_PATH_IMAGE201
The final prediction of , expressed as
Figure 157461DEST_PATH_IMAGE202
;

Figure 158915DEST_PATH_IMAGE203
Figure 158915DEST_PATH_IMAGE203

KDACAF的损失函数设置如下：The loss function of KDACAF is set as follows:

Figure 449826DEST_PATH_IMAGE204
Figure 449826DEST_PATH_IMAGE204

式中,

Figure 321967DEST_PATH_IMAGE205
为训练样本的数量，采用Adam训练方法训练KDACAF；In the formula,
Figure 321967DEST_PATH_IMAGE205
For the number of training samples, the Adam training method is used to train KDACAF;

S3、实验和分析S3, experiment and analysis

S31、比较测试；S31, comparison test;

步骤S31具体包括以下步骤：Step S31 specifically includes the following steps:

S311、在测试集Set†上的比较；S311, comparison on the test set Set†;

S312、在测试集Set‡上的比较。S312. Comparison on the test set Set‡.

S32、消融测试；S32. Ablation test;

S33、收敛性分析。S33. Convergence analysis.

基于以下模型分别测试知识驱动的方法和数据驱动的方法；The knowledge-driven approach and the data-driven approach were tested separately based on the following models;

参与对比的模型：四个SEM（其参数由生物地理学优化算法BBO确定）、SVR、GPR、通过叠加多个LSTM层构建的Deep LSTM（DLSTM）、LSTM+全连接层+迁移学习（LSTM-FC-TL）混合模型。其中，SVR、GPR和DLSTM是数据驱动的，所有SEM都是知识驱动的，KDACAF是知识-数据联合驱动的。Models participating in the comparison: four SEMs (whose parameters are determined by the biogeographical optimization algorithm BBO), SVR, GPR, Deep LSTM (DLSTM) constructed by superimposing multiple LSTM layers, LSTM+full connection layer+transfer learning (LSTM-FC -TL) mixed model. Among them, SVR, GPR, and DLSTM are data-driven, all SEM are knowledge-driven, and KDACAF is knowledge-data joint-driven.

表2为在测试集Set†上的表现结果表Table 2 shows the performance results on the test set Set†

Figure 908806DEST_PATH_IMAGE206
Figure 908806DEST_PATH_IMAGE206

由表2可知：(1)总体而言，四个SEM的表现比其他的差，表明知识驱动的SEM相对较小的容量限制了它们的近似能力和回归优度。(2)每种SEM在四种情况下的表现都不同，

Figure 397556DEST_PATH_IMAGE207
在Con.#5和#6中的表现要优于Con. #8和#9；
Figure 728043DEST_PATH_IMAGE208
和
Figure 985849DEST_PATH_IMAGE209
在Con.#5和#8中表现得比在Con.#6和#9更好；
Figure 884535DEST_PATH_IMAGE210
在Con.#8和#9中表现得比在Con. #5 and #6中更好。这一现象表明，知识驱动的SEM在处理不同条件下的容量预测任务的多模态方面能力有限，即没有一种SEM能够同时在四种测试条件下均表现良好。相比之下，SVR、DLSTM、GPR、LSTM-FC和KDACAF在四种条件下的性能表现相对一致。(3)LSTM-FC的表现优于SVR，但略差于DLSTM，这是由于DLSTM比LSTM-FC具有更深的结构（也有更大的模型容量）。(4)KDACAF在MAE和RMSE方面都具有最好的性能，这意味着先验知识和数据统计在某种程度上是互补的，并且将它们合并到一个模型（即KDACAF）中可以获得比数据驱动和知识驱动模型更显著的性能改进。It can be seen from Table 2 that: (1) Overall, the performance of the four SEMs is worse than the others, indicating that the relatively small capacity of the knowledge-driven SEMs limits their approximation ability and regression goodness. (2) Each SEM behaves differently in the four cases,
Figure 397556DEST_PATH_IMAGE207
Outperformed Con. #8 and #9 in Con. #5 and #6;
Figure 728043DEST_PATH_IMAGE208
and
Figure 985849DEST_PATH_IMAGE209
Performed better in Con.#5 and #8 than in Con.#6 and #9;
Figure 884535DEST_PATH_IMAGE210
Performed better at Con. #8 and #9 than at Con. #5 and #6. This phenomenon indicates that the knowledge-driven SEM has limited capability in handling the multimodality of the capacity prediction task under different conditions, i.e., no single SEM can perform well under all four test conditions at the same time. In contrast, SVR, DLSTM, GPR, LSTM-FC and KDACAF perform relatively consistently across the four conditions. (3) LSTM-FC performs better than SVR, but slightly worse than DLSTM, which is due to DLSTM has a deeper structure (and also has a larger model capacity) than LSTM-FC. (4) KDACAF has the best performance in terms of both MAE and RMSE, which means that prior knowledge and data statistics are complementary to some extent, and combining them into one model (i.e., KDACAF) can obtain better performance than data More significant performance improvements for both knowledge-driven and knowledge-driven models.

此外，图5为本发明在测试集set†上的预测结果和相应的预测误差关系图，图6为本发明在测试集set‡上的预测结果和相应的预测误差关系图。In addition, Fig. 5 is the relationship diagram of the prediction results of the present invention on the test set† and the corresponding prediction errors, and Fig. 6 is the relationship diagram of the prediction results of the present invention on the test set set‡ and the corresponding prediction errors.

在测试集Set‡上的比较Comparison on the test set Set‡

表3为所有模型在测试集Set‡上的性能表Table 3 is the performance table of all models on the test set Set‡

Figure 985215DEST_PATH_IMAGE211
Figure 985215DEST_PATH_IMAGE211

由表3可知，KDACAF具有最令人满意的性能，即在所有条件下都具有最低的MAE和RMSE。每个模型在测试集set†中的平均性能也列在表3中最后一列用于比较和分析。结果为：(1)虽然所有模型在测试集set‡上的性能都比在测试集set†上的性能差，但KDACAF在测试集set†与测试集set‡之间的性能差距最小，显示出最高的泛化性和通用性。(2)由于模型捕获多模态的能力有限，SEM的表现再次比其他模型更差。(3) LSTM-FC-TL的性能是所有模型中第二好的，这是因为迁移学习机制使LSTM-FC-TL在新条件下有很好的泛化能力。 (4)对于SVR、DLSTM、GPR、LSTM-FC-TL、KDACAF五个最好模型它们在Con.#1上的表现最差，这是因为Con.#1与#5、#6、#8的相似度最低。此外，这些模型在Con.#4中的表现优于Con.#2，这意味着将温度从25◦C降至10◦C是比在电池日历老化期间将SoC从50%降低到20%更重要的影响因素，因此，将模型应用于Con.#2时存在比Con.#4更大的模式失配。消融测试：As can be seen from Table 3, KDACAF has the most satisfactory performance, i.e., it has the lowest MAE and RMSE under all conditions. The average performance of each model on the test set† is also listed in the last column of Table 3 for comparison and analysis. The results are: (1) While all models perform worse on the test set‡ than on the test set†, KDACAF has the smallest performance gap between the test set† and the test set‡, showing that Highest generalization and versatility. (2) SEM again performs worse than other models due to the model's limited ability to capture multimodality. (3) The performance of LSTM-FC-TL is the second best among all models, because the transfer learning mechanism makes LSTM-FC-TL have good generalization ability under new conditions. (4) For the five best models of SVR, DLSTM, GPR, LSTM-FC-TL, and KDACAF, they performed the worst on Con.#1, because Con.#1 and #5, #6, #8 the lowest similarity. Furthermore, these models outperformed Con.#2 in Con.#4, which means that reducing the temperature from 25◦C to 10◦C is more efficient than reducing the SoC from 50% to 20% during battery calendar aging. Significant influencing factors, therefore, there is a larger schema mismatch when applying the model to Con.#2 than Con.#4. Ablation Test:

表4为消融测试结果表Table 4 is the table of ablation test results

Figure 729180DEST_PATH_IMAGE212
Figure 729180DEST_PATH_IMAGE212

为了验证每个模块的有效性和必要性，我们进行了消融测试，结果列在表4中，表4中的K、D、L和S分别表示知识驱动注意力模块、数据驱动注意力模块、长短时记忆模块和半经验模块。表4结果表明：(1)这两个注意力模块对预测性能的提高至关重要，即 "D+L+S "和 "K+L+S "都获得了比 "L+S "低得多的MAE和RMSE。(2) 知识驱动的注意力和数据驱动注意力模块相互补充，即"K+D+S "和 "K+D+L+S "的表现比 "D+L+S "和 "K+L+S "好得多。(3) 知识驱动的注意力比数据驱动的注意力贡献更大，也就是说，"K+L+S "的表现比 "D+L+S "略好一些。(4) 将长短时记忆模块加入到现有的框架工作中可以带来进一步的改善，即 "L+S "和 "K+D+L+S "分别比 "S "和 "K+D+S "表现更好。In order to verify the effectiveness and necessity of each module, we conducted an ablation test, and the results are listed in Table 4. K, D, L and S in Table 4 represent the knowledge-driven attention module, data-driven attention module, Long short-term memory module and semi-empirical module. The results in Table 4 show that: (1) These two attention modules are crucial to the improvement of the prediction performance, that is, both "D+L+S" and "K+L+S" achieved lower results than "L+S". Multiple MAE and RMSE. (2) Knowledge-driven attention and data-driven attention modules complement each other, i.e. "K+D+S" and "K+D+L+S" outperform "D+L+S" and "K+L +S "Much better. (3) Knowledge-driven attention contributes more than data-driven attention, that is, "K+L+S" performs slightly better than "D+L+S". (4) Adding long-short-term memory modules to the existing framework work can lead to further improvements, namely, "L+S" and "K+D+L+S" are better than "S" and "K+D+ S" performed better.

表5为SEM的消融测试结果表Table 5 is the ablation test result table of SEM

Figure 467591DEST_PATH_IMAGE213
Figure 467591DEST_PATH_IMAGE213

为了验证每个SEM的贡献和必要性，我们进行了另一项消融测试，结果列在表中5中，由表5可知：(1)所有SEM对测试集set†性能的贡献顺序如下：

Figure 477791DEST_PATH_IMAGE214
。(2)所有SEM对测试集set‡性能的贡献遵循以下顺序：
Figure 941134DEST_PATH_IMAGE215
。(3)比较测试集set†和测试集set‡，发现它们的贡献的波动是显著的。这种现象可能是由于SEM相对较小的容量使它们对每种条件的适应度差异很大，因此每个SEM不能在所有条件（即模式）上都表现良好。即每个SEM都在一定程度上偏向于某些特定条件。因此，将它们合并为一个单一的模型，即KDACAF，可以被认为是对这四个SEM的比较理想的组合，KDACAF中所有SEM的贡献和必要性是通过注意力模块来实现的。In order to verify the contribution and necessity of each SEM, we conducted another ablation test, and the results are listed in Table 5. From Table 5, we can see that: (1) The order of contribution of all SEMs to the performance of the test set† is as follows:
Figure 477791DEST_PATH_IMAGE214
. (2) The contribution of all SEMs to the performance of the test set‡ follows the following order:
Figure 941134DEST_PATH_IMAGE215
. (3) Comparing the test set† and the test set‡, it is found that the fluctuation of their contributions is significant. This phenomenon may be due to the fact that the relatively small capacity of SEMs makes their fitness to each condition very different, so each SEM cannot perform well on all conditions (i.e., modes). That is, each SEM is biased towards some specific conditions to some extent. Therefore, merging them into a single model, namely KDACAF, can be considered as an ideal combination of these four SEMs, and the contribution and necessity of all SEMs in KDACAF are realized through the attention module.

KDACAF的收敛性分析Convergence Analysis of KDACAF

将图3中所示的KDACAF实例的训练再执行1000次，在测试集set†和测试集set‡上的最终MAE和RMSE为分别通过图7和图8中的箱线图进行演示，证明：KDACAF具有较好的收敛性和稳定性，即最终MAE和RMSE的波动方差非常低，表明KDACAF已经训练到较优的阶段。收敛性分析表明，知识和数据联合驱动KDACAF兼具有大容量和小容量模型的优点，即具有良好的预测性能和高训练稳定性。Executing the training of the KDACAF instance shown in Figure 3 for another 1000 times, the final MAE and RMSE on the test set† and test set‡ are respectively demonstrated by the boxplots in Figures 7 and 8, proving that: KDACAF has better convergence and stability, that is, the fluctuation variance of the final MAE and RMSE is very low, indicating that KDACAF has been trained to a better stage. Convergence analysis shows that knowledge- and data-driven KDACAF combines the advantages of large-capacity and small-capacity models, namely, good predictive performance and high training stability.

本发明采用上述结构的基于注意力的锂离子电池日历老化预测模型，将注意力机制应用于锂离子电池日历老化预测，即KDACAF，有利于电池的监测和管理。KDACAF以电池电化学经验知识作为其关键基础，即知识驱动注意力模块，实现了领域知识和数据的有效互补。消融测试表明，领域知识的引入显著提高了KDACAF的预测性能。在KDACAF中，来自SEM的先验知识比数据集中包含的数据统计规律发挥了更重要的作用。对实际日历老化数据的案例测试表明，KDACAF在预测和推广到未见过的（全新的）测试条件方面具有优越性。与测试集set‡上的SEM相比，MAE和RMSE分别降低了5.78%和3.57%，表明所设计的KDACAF性能优越。对于电池健康预测，一种方法能达到的预测误差率越低，这种方法的预测性能就越好。在实际应用中，可接受的标准误差率会根据不同的要求而变化。例如，一些汽车公司建议2%，而一些能源系统公司建议为3%。由于电池的健康状况对于保证电池的效率和安全至关重要，因此值得探索更低的可接受误差率以扩大电池的应用范围。The present invention adopts the attention-based lithium-ion battery calendar aging prediction model of the above structure, and applies the attention mechanism to the lithium-ion battery calendar aging prediction, that is, KDACAF, which is beneficial to battery monitoring and management. KDACAF takes the empirical knowledge of battery electrochemistry as its key foundation, that is, the knowledge-driven attention module, which realizes the effective complementarity of domain knowledge and data. Ablation tests show that the introduction of domain knowledge significantly improves the prediction performance of KDACAF. In KDACAF, the prior knowledge from SEM plays a more important role than the statistical regularity of the data contained in the dataset. Case testing on real calendar aging data demonstrates the superiority of KDACAF in predicting and generalizing to unseen (brand new) test conditions. Compared with SEM on the test set‡, the MAE and RMSE are reduced by 5.78% and 3.57%, respectively, indicating the superior performance of the designed KDACAF. For battery health prediction, the lower the prediction error rate a method can achieve, the better the prediction performance of this method. In practical applications, the acceptable standard error rate will vary according to different requirements. For example, some automotive companies recommend 2%, while some energy systems companies recommend 3%. Since the health of the battery is crucial to ensure the efficiency and safety of the battery, it is worth exploring a lower acceptable error rate to expand the application range of the battery.

最后应说明的是：以上实施例仅用以说明本发明的技术方案而非对其进行限制，尽管参照较佳实施例对本发明进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对本发明的技术方案进行修改或者等同替换，而这些修改或者等同替换亦不能使修改后的技术方案脱离本发明技术方案的精神和范围。Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that: it still Modifications or equivalent replacements can be made to the technical solutions of the present invention, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present invention.

Claims (8)

1. A lithium ion battery calendar aging prediction model based on knowledge-data joint driving attention is characterized in that: the system comprises a semi-experience module, a knowledge driving attention module, a data driving attention module and a long-time and short-time memory module;
the knowledge-driven attention module takes a semi-empirical model as a front end, and the semi-empirical model is based on the arrhenius law.
2. The lithium ion battery calendar aging prediction model based on knowledge-data joint driving attention of claim 1, characterized in that: also included is an experimental platform for testing the effectiveness of KDACAF, said experimental platform comprising a hotroom for controlling the ambient temperature of the storage battery, battery test equipment for maintaining a predetermined storage state of charge level for the battery, a computer for monitoring and storing battery aging data.
3. A method of attention-based lithium ion battery calendar aging prediction model comprises the following steps:
s1, collecting and preprocessing data;
s2, establishing a calendar aging prediction model;
s21, establishing a problem description and a semi-empirical model of calendar aging prediction;
s22, structure of KDACAF;
s23, loss functions of the long-time and short-time memory module and the KDACAF;
s3, experiments and analysis
S31, comparing and testing;
s32, ablation testing;
and S33, convergence analysis.
4. The method of claim 3, wherein the model is based on a prediction model of calendar aging of lithium ion battery of attention:
the step S1 specifically includes the following steps:
data preprocessing and evaluation index
Carrying out cubic spline interpolation on a training set to ensure that each battery capacity sequence in the training set has 1 hour resolution, the length of each sequence is 11521, and in addition, in order to ensure the time resolution consistency of KDACAF on the training set, the verification set and the test set, all capacity sequences in the training set are sparsely sampled once every 30 days, namely all capacity sequences in the KDACAF training period
Figure DEST_PATH_IMAGE001
Any two consecutive times of
Figure DEST_PATH_IMAGE002
And
Figure DEST_PATH_IMAGE003
the time interval between is still 30 days, i.e.
Figure DEST_PATH_IMAGE004
Hours, which is the same as the time resolution of the validation set and test set; the maximum absolute error and the root mean square error are used for evaluation, namely:
Figure DEST_PATH_IMAGE005
Figure DEST_PATH_IMAGE006
。
5. the method of claim 3, wherein the model is based on a prediction model of calendar aging of lithium ion battery of attention:
step S21 specifically includes the following steps:
s211, problem description of calendar aging prediction
Calendar aging prediction of lithium ion batteries aims at predicting the variation of their capacity with storage time, the interpretation variables for this task are formatted as the following matrix:
Figure DEST_PATH_IMAGE007
wherein,
Figure DEST_PATH_IMAGE008
it is meant that all of the interpretation information,
Figure DEST_PATH_IMAGE009
the sequential order number representing the capacity sequence,
Figure DEST_PATH_IMAGE010
indicating a sequence of capacities
Figure DEST_PATH_IMAGE011
The measured value of the number of the first measurement,
Figure DEST_PATH_IMAGE012
represent
Figure DEST_PATH_IMAGE013
The storage time of the battery in (1),
Figure DEST_PATH_IMAGE014
which is indicative of the temperature at which the battery is stored,
Figure DEST_PATH_IMAGE015
indicating storage chargeIn the electrical state of the electric motor, the electric motor is in a closed state,
Figure DEST_PATH_IMAGE016
is a lag interval;
for capacity prediction for a single step, the target variable is
Figure DEST_PATH_IMAGE017
Thus, the calendar aging prediction task is abstracted as the following mapping:
Figure DEST_PATH_IMAGE018
wherein,
Figure DEST_PATH_IMAGE019
a predictive model representing data-driven or knowledge-driven;
s212, semi-empirical model of calendar aging prediction
According to arrhenius' law, the degradation of calendar capacity is ultimately expressed in a particular semi-empirical form:
Figure DEST_PATH_IMAGE020
wherein,
Figure DEST_PATH_IMAGE021
and
Figure DEST_PATH_IMAGE022
are all the dependent items of the SoC,
Figure DEST_PATH_IMAGE023
which represents the constant of the gas,
Figure DEST_PATH_IMAGE024
a low power parameter representing a duration dependency;
consider a memory SoC pair registerThe effects of biochemical reactions can lead to degradation of battery capacity, taking into account the linear and exponential dependence of SoC, i.e.
Figure DEST_PATH_IMAGE025
And
Figure DEST_PATH_IMAGE026
in the form of (a), the following semi-empirical model is constructed:
Figure DEST_PATH_IMAGE027
all the SEMs mentioned above can be respectively simplified and expressed as
Figure DEST_PATH_IMAGE028
，
Figure DEST_PATH_IMAGE029
，
Figure DEST_PATH_IMAGE030
，
Figure DEST_PATH_IMAGE031
Wherein
Figure DEST_PATH_IMAGE032
Respectively representing all parameters to be identified, namely:
Figure DEST_PATH_IMAGE033
Figure DEST_PATH_IMAGE034
Figure DEST_PATH_IMAGE035
Figure DEST_PATH_IMAGE036
。
6. the method of claim 3, wherein the model is based on a prediction model of calendar aging of lithium ion battery of attention:
the structure of the step S22 KDACAF specifically comprises the following steps:
s221, semi-empirical module and prediction branch
The semi-empirical module is a set of four SEMs set forth in S212;
in the case of a predicted branch, the branch is predicted,
Figure DEST_PATH_IMAGE037
the four SEM's input to the semi-empirical module were obtained separately
Figure DEST_PATH_IMAGE038
The regression results, i.e., the preliminary prediction results of (1), are as follows:
Figure DEST_PATH_IMAGE039
that is, for each SEM there are:
Figure DEST_PATH_IMAGE040
；
s222, fitting branch and knowledge-driven attention module
To fit the results of past capacity sequences, first, one will fit
Figure DEST_PATH_IMAGE041
The interpretation matrix is divided into two parts:
Figure DEST_PATH_IMAGE042
namely:
Figure DEST_PATH_IMAGE043
in the formula,
Figure DEST_PATH_IMAGE044
a sequence of recent capacities is indicated,
Figure DEST_PATH_IMAGE045
each column in (1) represents
Figure DEST_PATH_IMAGE046
Influence factors of corresponding elements in the Chinese character; in the fitting branch
Figure DEST_PATH_IMAGE047
Respectively input into the four SEM to obtain
Figure DEST_PATH_IMAGE048
The regression results of (1):
Figure DEST_PATH_IMAGE049
Figure DEST_PATH_IMAGE050
is that
Figure DEST_PATH_IMAGE051
In that
Figure DEST_PATH_IMAGE052
The regression results of (a) were as follows:
Figure DEST_PATH_IMAGE053
the present knowledge-driven attention module is dedicated to calendar aging prediction tasks, where targets are predicted
Figure DEST_PATH_IMAGE054
The attention to each SEM was designed based on the goodness of fit of each SEM model to the past real volume;
first, the knowledge-driven attention model is
Figure DEST_PATH_IMAGE055
And
Figure DEST_PATH_IMAGE056
the score function between is defined as follows:
Figure DEST_PATH_IMAGE057
wherein,
Figure DEST_PATH_IMAGE058
a scoring function representing a knowledge-driven attention model;
can then obtain
Figure DEST_PATH_IMAGE059
Attention to each SEM, note
Figure DEST_PATH_IMAGE060
Figure DEST_PATH_IMAGE061
Then, get to
Figure DEST_PATH_IMAGE062
The following points of (1) are accurately predicted:
Figure DEST_PATH_IMAGE063
；
s223, feature branching and data-driven attention module
In order to expand the model capacity of KDACAF to better capture the multimodalities of the calendar aging process, data-driven attention was constructed as follows:
to construct feature vectors from semi-empirical modules, we will
Figure DEST_PATH_IMAGE064
Again, the four SEMs were entered and four eigenvectors were constructed from the intermediate variables of these SEMs, respectively, as follows:
Figure DEST_PATH_IMAGE065
wherein,
Figure DEST_PATH_IMAGE066
the representation is based on
Figure DEST_PATH_IMAGE067
The constructed features of intermediate variables within;
predicting a target in a data-driven attention module
Figure DEST_PATH_IMAGE068
The attention to each SEM was designed based on the historical capacity sequence and the constructed features, and the score function was defined as:
Figure DEST_PATH_IMAGE069
wherein
Figure DEST_PATH_IMAGE070
A scoring function representing the data-driven attention,
Figure DEST_PATH_IMAGE071
is the first
Figure DEST_PATH_IMAGE072
A trainable matrix corresponding to the SEM;
then, can be obtained
Figure DEST_PATH_IMAGE073
Attention number to each SEM, note
Figure DEST_PATH_IMAGE074
I.e. by
Figure DEST_PATH_IMAGE075
Figure DEST_PATH_IMAGE076
Then, get to
Figure DEST_PATH_IMAGE077
Another accurate prediction of:
Figure DEST_PATH_IMAGE078
。
7. the method of claim 3, wherein the model is based on a prediction model of calendar aging of lithium ion battery of attention:
step S23 is a loss function of the long-time and short-time memory module and the KDACAF, and specifically includes the following steps:
the comprehensive two-part attention mechanism can be obtained
Figure DEST_PATH_IMAGE079
The intermediate prediction of (2):
Figure DEST_PATH_IMAGE080
wherein
Figure DEST_PATH_IMAGE081
Representing two attention module pairs
Figure DEST_PATH_IMAGE082
The inter-prediction of (2) is performed,
Figure DEST_PATH_IMAGE083
and
Figure DEST_PATH_IMAGE084
all are learnable weights;
then, the historical capacity is sequenced
Figure DEST_PATH_IMAGE085
And
Figure DEST_PATH_IMAGE086
the concatenation is as follows vector:
Figure DEST_PATH_IMAGE087
finally, the vector is input to a long-short-term memory module comprising an LSTM layer, and the output of the LSTM is scaled to
Figure DEST_PATH_IMAGE088
In front of and behind
Figure DEST_PATH_IMAGE089
Is expressed as
Figure DEST_PATH_IMAGE090
；
Figure DEST_PATH_IMAGE091
The loss function of KDACAF is set as follows:
Figure DEST_PATH_IMAGE092
in the formula,
Figure DEST_PATH_IMAGE093
to train the number of samples, the Adam training method was used to train KDACAF.
8. The method of claim 3, wherein the model is based on a prediction model of calendar aging of lithium ion battery of attention:
step S31 specifically includes the following steps:
s311, comparison in a test Set of \822458;
s312, comparison in test Set \8225j.
CN202310023441.2A 2023-01-09 2023-01-09 Attention-based calendar aging prediction model and method for lithium-ion batteries Active CN115718263B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310023441.2A CN115718263B (en) 2023-01-09 2023-01-09 Attention-based calendar aging prediction model and method for lithium-ion batteries

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310023441.2A CN115718263B (en) 2023-01-09 2023-01-09 Attention-based calendar aging prediction model and method for lithium-ion batteries

Publications (2)

Publication Number Publication Date
CN115718263A true CN115718263A (en) 2023-02-28
CN115718263B CN115718263B (en) 2023-04-07

Family

ID=85257877

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310023441.2A Active CN115718263B (en) 2023-01-09 2023-01-09 Attention-based calendar aging prediction model and method for lithium-ion batteries

Country Status (1)

Country Link
CN (1) CN115718263B (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117269805A (en) * 2023-11-23 2023-12-22 中国人民解放军国防科技大学 Vehicle lithium battery pack health status assessment model training, prediction method and device
CN117630683A (en) * 2024-01-25 2024-03-01 北京科技大学 Multi-step prediction method and system for automotive battery SOC based on multi-scale fusion GRU network
CN118884236A (en) * 2024-07-18 2024-11-01 吉林大学 A method, system, device and storage medium for predicting internal aging mode of lithium battery
CN119167799A (en) * 2024-11-20 2024-12-20 北京科技大学 Battery calendar aging capacity prediction method based on generative adversarial network model
CN120044406A (en) * 2025-04-08 2025-05-27 宁德时代新能源科技股份有限公司 Method, device, equipment and storage medium for predicting battery capacity inflection point

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140214348A1 (en) * 2013-01-29 2014-07-31 Mitsubishi Electric Research Laboratories, Inc. Method for Estimating a State of Charge of Batteries
CN111710376A (en) * 2020-05-13 2020-09-25 中国科学院计算机网络信息中心 Load balancing method and system for block computing in macromolecular and cluster systems
CN112098874A (en) * 2020-08-21 2020-12-18 杭州电子科技大学 Lithium ion battery electric quantity prediction method considering aging condition
US20210049460A1 (en) * 2019-08-15 2021-02-18 Noodle Analytics, Inc. Deep probabilistic decision machines
CN114325450A (en) * 2022-01-11 2022-04-12 青岛科技大学 Lithium ion battery health state prediction method based on CNN-BilSTM-AT hybrid model
CN114596010A (en) * 2022-05-09 2022-06-07 西南石油大学 BiGRU network drilling condition identification method combined with attention mechanism
CN114743600A (en) * 2022-04-15 2022-07-12 大连理工大学 A Deep Learning Prediction Method for Target-Ligand Binding Affinity Based on Gated Attention Mechanism
CN115047350A (en) * 2022-06-24 2022-09-13 哈尔滨工业大学 Digital-analog linkage based lithium ion battery remaining service life prediction method
CN115409263A (en) * 2022-08-30 2022-11-29 北京航空航天大学 Method for predicting remaining life of lithium battery based on gating and attention mechanism

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140214348A1 (en) * 2013-01-29 2014-07-31 Mitsubishi Electric Research Laboratories, Inc. Method for Estimating a State of Charge of Batteries
US20210049460A1 (en) * 2019-08-15 2021-02-18 Noodle Analytics, Inc. Deep probabilistic decision machines
CN111710376A (en) * 2020-05-13 2020-09-25 中国科学院计算机网络信息中心 Load balancing method and system for block computing in macromolecular and cluster systems
CN112098874A (en) * 2020-08-21 2020-12-18 杭州电子科技大学 Lithium ion battery electric quantity prediction method considering aging condition
CN114325450A (en) * 2022-01-11 2022-04-12 青岛科技大学 Lithium ion battery health state prediction method based on CNN-BilSTM-AT hybrid model
CN114743600A (en) * 2022-04-15 2022-07-12 大连理工大学 A Deep Learning Prediction Method for Target-Ligand Binding Affinity Based on Gated Attention Mechanism
CN114596010A (en) * 2022-05-09 2022-06-07 西南石油大学 BiGRU network drilling condition identification method combined with attention mechanism
CN115047350A (en) * 2022-06-24 2022-09-13 哈尔滨工业大学 Digital-analog linkage based lithium ion battery remaining service life prediction method
CN115409263A (en) * 2022-08-30 2022-11-29 北京航空航天大学 Method for predicting remaining life of lithium battery based on gating and attention mechanism

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
赵珈卉: "锂离子电池状态估计与剩余寿命预测方法综述", 《发电技术》 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117269805A (en) * 2023-11-23 2023-12-22 中国人民解放军国防科技大学 Vehicle lithium battery pack health status assessment model training, prediction method and device
CN117630683A (en) * 2024-01-25 2024-03-01 北京科技大学 Multi-step prediction method and system for automotive battery SOC based on multi-scale fusion GRU network
CN117630683B (en) * 2024-01-25 2024-03-29 北京科技大学 Multi-step prediction method and system for automotive battery SOC based on multi-scale fusion GRU network
CN118884236A (en) * 2024-07-18 2024-11-01 吉林大学 A method, system, device and storage medium for predicting internal aging mode of lithium battery
CN119167799A (en) * 2024-11-20 2024-12-20 北京科技大学 Battery calendar aging capacity prediction method based on generative adversarial network model
CN120044406A (en) * 2025-04-08 2025-05-27 宁德时代新能源科技股份有限公司 Method, device, equipment and storage medium for predicting battery capacity inflection point

Also Published As

Publication number Publication date
CN115718263B (en) 2023-04-07

Similar Documents

Publication Publication Date Title
Wang et al. A data-driven method with mode decomposition mechanism for remaining useful life prediction of lithium-ion batteries
CN116609668B (en) Lithium ion battery health state and residual life prediction method
Zhao et al. A compact methodology via a recurrent neural network for accurate equivalent circuit type modeling of lithium-ion batteries
Che et al. Semi-supervised self-learning-based lifetime prediction for batteries
Cui et al. A dynamic spatial-temporal attention-based GRU model with healthy features for state-of-health estimation of lithium-ion batteries
CN115718263A (en) Attention-based lithium ion battery calendar aging prediction model and method
US20230029810A1 (en) Battery model construction method and battery degradation prediction device
Wang et al. A method for state-of-charge estimation of Li-ion batteries based on multi-model switching strategy
Mamo et al. Long short-term memory with attention mechanism for state of charge estimation of lithium-ion batteries
CN108896914A (en) A kind of modeling of gradient boosted tree and prediction technique of lithium battery health status
Li et al. A recurrent neural network with long short-term memory for state of charge estimation of lithium-ion batteries
Zhang et al. A state-of-charge estimation method based on bidirectional LSTM networks for lithium-ion batteries
Lu et al. Deep learning to predict battery voltage behavior after uncertain cycling-induced degradation
Ezzouhri et al. A data-driven-based framework for battery remaining useful life prediction
Channegowda et al. An attention‐based synthetic battery data augmentation technique to overcome limited dataset challenges
Fang et al. Fusion model based RUL prediction method of lithium-ion battery under working conditions
Ren et al. The LiFePO4 battery lifespan SoC estimation using Ham-Informer and internal pressure
Wang et al. A double broad learning approach based on variational modal decomposition for Lithium-Ion battery prognostics
Lin et al. An Ensemble Convolutional Neural Network Based Transfer Learning Framework for Achievable Lithium-Ion Battery Health Perception in Flexible Charging Procedure
CN117169743A (en) Battery health state estimation method and device based on partial data and model fusion
CN119623274A (en) A lithium battery life prediction method, device, terminal equipment and medium
CN118859008A (en) Method, system, device and medium for predicting remaining service life of lithium-ion power battery
Ni et al. Battery state of charge estimation using long short-term memory network and extended Kalman filter
Ma et al. Artificial Intelligence Based SOC Technology for Electric Vehicle Batteries
CN119959764A (en) Battery health status prediction method and device

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