CN116224118A - Battery remaining cycle life prediction method based on TCN - Google Patents

Battery remaining cycle life prediction method based on TCN Download PDF

Info

Publication number
CN116224118A
CN116224118A CN202310075013.4A CN202310075013A CN116224118A CN 116224118 A CN116224118 A CN 116224118A CN 202310075013 A CN202310075013 A CN 202310075013A CN 116224118 A CN116224118 A CN 116224118A
Authority
CN
China
Prior art keywords
input
battery
tcn
prediction
convolution
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
CN202310075013.4A
Other languages
Chinese (zh)
Inventor
孙玉树
彭大健
李宁宁
张国伟
唐西胜
裴玮
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.)
Institute of Electrical Engineering of CAS
Original Assignee
Institute of Electrical Engineering of CAS
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 Institute of Electrical Engineering of CAS filed Critical Institute of Electrical Engineering of CAS
Priority to CN202310075013.4A priority Critical patent/CN116224118A/en
Publication of CN116224118A publication Critical patent/CN116224118A/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]
    • G01R31/392—Determining battery ageing or deterioration, e.g. state of health
    • 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

Landscapes

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

Abstract

The invention provides a battery residual cycle life prediction method based on TCN, which applies five characteristic factors strongly related to the health state: the method comprises the steps of taking cycle time, voltage average value, voltage sample entropy, temperature sample entropy and current value as input characteristic factors, reducing the dimension of the input characteristic factors by using a kernel principal component analysis method, and selecting principal components with larger contribution rate based on a time convolution neural network to conduct battery SOH prediction; because the battery has capacity regeneration phenomenon, the capacity is combined on the basis of five characteristic factors, and the prediction of the residual service life is performed, so that the accuracy and the reliability of a prediction result are improved.

Description

Battery remaining cycle life prediction method based on TCN
Technical Field
The invention belongs to the field of intelligent operation and maintenance of battery energy storage systems, and particularly relates to a battery residual cycle life prediction method based on TCN (time convolution neural network).
Background
Because of the characteristics of high power density, environmental protection, long service life and the like, the battery energy storage is widely applied to various fields of renewable energy power generation and the like, and has important roles in promoting the consumption of renewable energy sources, improving the efficiency, economy and the like of a power system and a regional energy system. Battery energy storage is a dynamic, constantly changing electrochemical system with nonlinear behavior and complex internal mechanisms, whose performance inevitably deteriorates with prolonged use. The Remaining Useful Life (RUL) may be considered the normal usable time before the battery reaches the failure threshold during service. When the battery capacity falls below 80% of its rated capacity, the battery stability decreases, possibly resulting in a decrease in charge-discharge performance and even a catastrophic accident. Accurate prediction of the RUL can guide health management, battery replacement, and system maintenance of the battery, preventing significant loss due to battery failure or premature replacement, to achieve predictive maintenance of the battery.
Traditional battery health state analysis mainly focuses on capacity prediction, but in the actual use process, the battery has capacity regeneration phenomenon, namely the phenomenon that the batteries have the same capacity in different time periods; furthermore, the prediction input is mainly basic parameters such as voltage, current, temperature and the like, the data size is small, and the prediction precision is required to be improved; the conventional neural network has the defects of easy explosion/disappearance of gradient, difficult processing of long sequence data, poor generalization capability and the like in predictive modeling.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a battery residual cycle life prediction method based on TCN (time convolution neural network). Five characteristic factors that are strongly related to the application and health status: the method comprises the steps of taking cycle time, voltage average value, voltage sample entropy, temperature sample entropy and current value as input characteristic factors, reducing the dimension of the input characteristic factors by using a kernel principal component analysis method, and selecting principal components with larger contribution rate based on a time convolutional neural network (TCN) to predict SOH of a battery; because the battery has capacity regeneration phenomenon, the capacity is combined on the basis of five characteristic factors, and RUL prediction is performed, so that the accuracy and the reliability of a prediction result are improved.
In order to achieve the above purpose, the invention adopts the following technical scheme:
a battery remaining cycle life prediction method based on TCN applies five characteristic factors strongly related to health status: the method comprises the steps of taking cycle time, voltage average value, voltage sample entropy, temperature sample entropy and current value as input characteristic factors, reducing the dimension of the input characteristic factors by using a kernel principal component analysis method, and selecting principal components with larger contribution rate based on a time convolution neural network to conduct battery SOH prediction; because the battery has capacity regeneration phenomenon, the capacity is combined on the basis of five characteristic factors, and the prediction of the residual service life is performed, so that the accuracy and the reliability of a prediction result are improved.
Further, the core principal component analysis method is adopted to separate and optimize the overlapping information and redundant information among the input characteristic factors so as to improve the calculation efficiency and accuracy, and the method specifically comprises the following steps:
for n input samples x k (k=1, 2, … n) whose covariance matrix is expressed as:
Figure BDA0004065866220000021
/>
the kernel principal component analysis method introduces a linear mapping function to input a sample x k (k=1, 2, … n) into the feature space:
Figure BDA0004065866220000022
the corresponding covariance matrix is:
Figure BDA0004065866220000023
solving equations in feature space using principal component analysis
Figure BDA0004065866220000024
Where λ is the eigenvalue and v is the eigenvector:
Figure BDA0004065866220000025
the feature vector v is linearized by feature space data:
Figure BDA0004065866220000026
wherein a is i Linear coefficients for each feature space data;
an n-order kernel matrix K is defined,
Figure BDA0004065866220000027
then equation (3) converts to:
nλa＝Ka (5)
where a is a coefficient matrix of the feature space data.
Solving to obtain a characteristic value lambda and a characteristic vector v;
by projection V of an input sample onto a feature space k The method comprises the steps of obtaining a projection characteristic value and a projection characteristic vector with large contribution rate as main components:
Figure BDA0004065866220000028
further, the time convolution neural network is used for establishing a mapping relation between an input sequence and an output sequence to enable the actual output y 0 ,…,y T And predicted output y' 0 ,…,y′ T The error loss between the two is minimum, and the construction comprises the following steps:
(1) TCN modeling:
let the given input sequence be x 0 ,…,x T The predicted output is expected to be y 0 ,…,y T ；
The relation between predicted output and input sequence is:
Figure BDA0004065866220000031
wherein y 'is' t Input sequence x just before time t 0 ,…,x t In connection with future input x t ,…,x T Is irrelevant;
(2) The expansion causal convolution is a convolution operation for performing a step-up operation on an input sequence, and specifically comprises the following steps:
Figure BDA0004065866220000032
wherein F (i) is a para sequence x 0 ,…,x T The convolution result of the ith element; h (j) is a filter, also known as a convolution kernel; d is an expansion factor, a standard causal convolution when d=1;
(3) Adding a residual block:
assuming that the input of the residual block is x and the output is o, the equation (9) is obtained by linear variation and mapping by the activation function:
o＝Activation(x+Γ(x)) (9)
wherein, the Activation (·) is an Activation function, Γ (x) is x Is a function of the relationship of (2).
The beneficial effects are that:
because of the unavoidable crossing and overlapping of the information among the influence factors of the battery health state, the invention adopts a nuclear principal component analysis method to separate and optimize the overlapping information among the influence factors from redundant information, thereby improving the efficiency and accuracy of subsequent calculation; the TCN has the characteristics of calculation parallelism, flexible receptive field, low memory requirement during training and the like, so that the longer short-time memory neural network and the convolutional neural network have more advantages during data prediction; the combined prediction of SOH and RUL can eliminate the influence of capacity regeneration phenomenon and provide more accurate battery health status.
Drawings
FIG. 1 is a schematic diagram of a TCN expanded causal convolution structure;
FIG. 2 is a schematic capacity diagram;
FIG. 3 is a schematic diagram of each discharge cycle time;
FIG. 4 is a schematic diagram of voltage averages;
FIG. 5 is a schematic view of voltage sample entropy;
FIG. 6 is a schematic diagram of temperature sample entropy;
FIG. 7 is a schematic diagram of current;
FIG. 8 is a schematic diagram of remaining useful life;
FIG. 9 is a schematic diagram of the principal components 1 and 2;
FIG. 10 is a schematic diagram of TCN capacity prediction results;
FIG. 11 is a diagram showing the result of predicting the number of remaining cycles.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention. In addition, the technical features of the embodiments of the present invention described below may be combined with each other as long as they do not collide with each other.
The invention relates to a battery residual cycle life prediction method based on TCN, which applies five characteristic factors strongly related to the health state: the method comprises the steps of taking cycle time, voltage average value, voltage sample entropy, temperature sample entropy and current value as input characteristic factors, reducing the dimension of the input characteristic factors by using a kernel principal component analysis method, and selecting principal components with larger contribution rate based on a time convolutional neural network (TCN) to predict SOH of a battery; because the battery has capacity regeneration phenomenon, the capacity is combined on the basis of five characteristic factors, and RUL prediction is performed, so that the accuracy and the reliability of a prediction result are improved.
The nuclear principal component analysis method specifically comprises the following steps:
the invention adopts a kernel principal component analysis method to separate and optimize the overlapping information and redundant information among the input characteristic factors so as to improve the calculation efficiency and accuracy.
The kernel principal component analysis is a principal component analysis in a nonlinear extended form. Since the principal component analysis is a linear method, it is difficult to effectively extract a nonlinear structure in data, and there is a complex nonlinear relationship between battery life influencing factors, so that the effect of reducing the dimension of battery data by using the principal component analysis method is not ideal. And the kernel principal component analysis uses kernel functions to map the low-dimensional variable space to the high-dimensional feature space, so that the problem of nonlinearity in the low-dimensional space is converted into the problem of linearity in the high-dimensional space, and the effect of reducing the dimension by applying principal component analysis in the high-dimensional space is achieved.
For n input samples x k (k=1, 2, … n), the covariance matrix of which can be expressed as:
Figure BDA0004065866220000041
the principal component analysis method takes eigenvalues and eigenvectors with larger contribution rates in covariance matrixes as principal components. The kernel principal component analysis method introduces a linear mapping function to input a sample x k (k=1, 2, … n) into the feature space:
Figure BDA0004065866220000042
the corresponding covariance matrix is:
Figure BDA0004065866220000051
solving equations in feature space using principal component analysis
Figure BDA0004065866220000052
Where λ is the eigenvalue and v is the eigenvector:
Figure BDA0004065866220000053
the feature vector v is linearized by feature space data:
Figure BDA0004065866220000054
wherein a is i Linear coefficients for each feature space data. An n-order kernel matrix K is defined,
Figure BDA0004065866220000055
then equation (3) converts to:
nλa＝Ka (5)
where a is a coefficient matrix of the feature space data.
And solving to obtain the eigenvalue lambda and the eigenvector v.
By projection V of an input sample onto a feature space k The method comprises the steps of obtaining a projection characteristic value and a projection characteristic vector with large contribution rate as main components:
Figure BDA0004065866220000056
the time convolution neural network (Temporal Convolution Network, TCN) adopts structures such as an enlarged causal convolution block and a residual block, so that the time convolution neural network has the capability of extracting characteristics from a large sample time sequence and realizing prediction, and can effectively solve the problem of performance degradation of a deep network in the network training process, and the construction comprises the following steps:
(1) TCN modeling
Let the given input sequence be x 0 ,…,x T The predicted output is expected to be y 0 ,…,y T . The relation between predicted output and input sequence is that
(y′ 0 ,…,y′ T )＝f(x 0 ,…,x T ) (7)
Wherein y 'is' t Input sequence x just before time t 0 ,…,x t In connection with future input x t ,…,x T Irrespective of the fact that the first and second parts are. TCN modeling is to build a mapping relationship between an input sequence and an output sequence, with the goal of letting the actual output y 0 ,…,y T And predicted output y' 0 ,…,y′ T The error loss between them is minimal.
(2) Enlarging causal convolution:
compared with a recurrent neural network, the model using causal convolution does not use circular connection, so that time sequence data can be input in parallel, faster network training speed can be obtained, and the model has unique advantages in the training aspect of a large sample time sequence. However, standard causal convolution requires increasing the receptive field of neurons in a neural network by stacking many network layers or using large convolution kernels when processing large sample time series. For this reason, TCN employs an extended causal convolution (Dilated Causal Convolution, DCC) technique to boost receptive fields without significantly increasing computational costs. DCC is a convolution operation for performing a skip operation on an input sequence, specifically:
Figure BDA0004065866220000061
wherein F (i) is a para sequence x 0 ,…,x T The convolution result of the ith element; h (j) is a filter, also known as a convolution kernel; d is an expansion factor, a standard causal convolution when d=1.
The DCC structure is shown in fig. 1, which allows the output y to be correlated as much as possible to the input x for the same number of network layers as compared to the standard causal convolution; the multi-layer stacking principle of DCC can make the deep learning network reach a very large receptive field with fewer network layers; the convolution kernel is capable of performing a sliding operation on the input data so that the TCN can handle variable length inputs, and new predictions can be continuously calculated and output as the model input data is updated.
(3) Adding a residual block:
the residual block is mainly applied to solve the degradation problem of the deep learning network, and the core idea is to introduce a 'jump connection' operation for skipping one or more layers. Assuming that the input of the residual block is x and the output is o, equation (9) is obtained by linear variation and mapping by the activation function. Since the residual error is not 0 in practice, the stacked layers in the deep learning network can always learn new features, i.e. the learning performance of the deep network is not degraded.
o＝Activation(x+Γ(x)) (9)
In the formula, the Activation (·) is an Activation function. Γ (x) is a relational function of x.
In conclusion, when TCN modeling is performed, the network structure combining the residual block and DCC can be used, so that the characteristic learning capacity and robustness of the TCN model can be effectively improved.
The present invention will be described in detail with reference to specific examples.
The capacity of the battery can simply and clearly indicate the aging degree of the battery, but in actual operation, the capacity is not easily measured directly. According to the invention, a certain lithium battery is adopted, and is charged at a constant current of 1.5A at room temperature, and when the voltage reaches 4.2V, the constant voltage is converted to constant voltage to continue charging until the charging current is reduced to 20mA; and then discharging with 2A constant current until the voltage is reduced to 2.5V; the repeated charge and discharge cycles accelerate battery aging, and the acquired battery capacity data are shown in fig. 2. Then, from the time, voltage, current and temperature which are easy to measure, five characteristic factors of cycle time, voltage average value, voltage sample entropy, temperature sample entropy and current value are extracted, as shown in fig. 3-7.
The principal component contribution rates of five characteristic factors are calculated by a kernel principal component analysis method as shown in table 1, and the contribution rates of both principal components 1 and 2 account for 97.97%, so that the principal component 1 and the principal component 2 are mainly used as input parameters of the neural network, and as shown in fig. 9, prediction of the battery capacity degradation is performed.
TABLE 1 contribution rates of different principal Components
Figure BDA0004065866220000071
The capacity of the battery was predicted based on TCN, and the prediction result is shown in fig. 10. From the above, it can be seen that TCN can better track the capacity fading trend, and can also well predict the capacity fluctuation in the aging process, so that more accurate capacity prediction is obtained under a smaller data size, and the prediction error RMSE is 0.0205.
The evaluation of the battery capacity can reflect the state of health SOH of the battery, and the evaluation of the remaining cycle times of the battery can reflect the remaining service life RUL of the battery, and the battery can be characterized in different forms on the health degree of the battery. However, in the practical use process, the battery has capacity regeneration phenomenon, namely, the battery has the same capacity in different time periods. As shown in fig. 2, the capacities at 45 th and 59 th cycles are equivalent, both being around 1.59Ah, but differ by 14 cycles; the capacities at the 100 th cycle and 126 th cycle are equivalent and are both around 1.37Ah, but they differ by 26 cycles, so that the equivalent capacities occur at different times and do not represent the equivalent SOH of the battery at both times.
In order to solve the problems, the invention further analyzes the residual cycle times of the battery on the basis of the capacity, namely, predicts the residual cycle times by using the cycle time, the voltage average value, the voltage sample entropy, the temperature sample entropy and the current value and adding the capacity value. First, the six characteristic factors were subjected to dimension reduction, and the contribution ratio of each principal component was obtained by using a kernel principal component analysis method, as shown in table 2. Wherein the sum of the contribution rates of the main components 1 and 2 accounts for 98.17%, so that the main component 1 and the main component 2 are selected as inputs to predict the remaining cycle times of the battery. As shown in fig. 11, the TCN can well predict the RUL of the battery, and the error RMSE is 0.0193. Therefore, based on TCN, the capacity is predicted by using the cycle time, the voltage average value, the voltage sample entropy, the temperature sample entropy and the current value, and then the residual cycle times are evaluated, so that the health degree of the battery can be more comprehensively grasped.
TABLE 2 contribution rates of different principal Components 2
Figure BDA0004065866220000072
Figure BDA0004065866220000081
It will be readily appreciated by those skilled in the art that the foregoing description is merely a preferred embodiment of the invention and is not intended to limit the invention, but any modifications, equivalents, improvements or alternatives falling within the spirit and principles of the invention are intended to be included within the scope of the invention.

Claims (3)

1. A battery residual cycle life prediction method based on TCN is characterized in that five characteristic factors which are strongly related to the health state are applied: the method comprises the steps of taking cycle time, voltage average value, voltage sample entropy, temperature sample entropy and current value as input characteristic factors, reducing the dimension of the input characteristic factors by using a kernel principal component analysis method, and selecting principal components with larger contribution rate based on a time convolution neural network to conduct battery SOH prediction; because the battery has capacity regeneration phenomenon, the capacity is combined on the basis of five characteristic factors, and the prediction of the residual service life is performed, so that the accuracy and the reliability of a prediction result are improved.
2. The TCN-based battery remaining cycle life prediction method according to claim 1, wherein the core principal component analysis method is adopted to separate and optimize overlapping information between input feature factors from redundant information, so as to improve calculation efficiency and accuracy, and specifically comprises:
for n input samples x k (k=1, 2, … n) whose covariance matrix is expressed as:
Figure FDA0004065866200000011
the kernel principal component analysis method introduces a linear mapping function to input a sample x k (k=1, 2, … n) into the feature space:
Figure FDA0004065866200000012
the corresponding covariance matrix is:
Figure FDA0004065866200000013
solving equations in feature space using principal component analysis
Figure FDA0004065866200000014
Where λ is the eigenvalue and v is the eigenvector:
Figure FDA0004065866200000015
the feature vector v is linearized by feature space data:
Figure FDA0004065866200000016
wherein a is i Linear coefficients for each feature space data;
an n-order kernel matrix K is defined,
Figure FDA0004065866200000017
then equation (3) converts to:
nλa＝Ka (5)
where a is a coefficient matrix of the feature space data.
Solving to obtain a characteristic value lambda and a characteristic vector v;
by projection V of an input sample onto a feature space k The method comprises the steps of obtaining a projection characteristic value and a projection characteristic vector with large contribution rate as main components:
Figure FDA0004065866200000021
3. the TCN-based battery remaining cycle life prediction method according to claim 1, wherein the time convolutional neural network is configured toEstablishing a mapping relation between an input sequence and an output sequence to ensure that the actual output y 0 ,…,y T And predicted output y' 0 ,…,y′ T The error loss between the two is minimum, and the construction comprises the following steps:
(1) TCN modeling:
let the given input sequence be x 0 ,…,x T The predicted output is expected to be y 0 ,…,y T ；
The relation between predicted output and input sequence is:
(y′ 0 ,…,y′ T )＝f(x 0 ,…,x T ) (7)
wherein y 'is' t Input sequence x just before time t 0 ,…,x t In connection with future input x t ,…,x T Is irrelevant;
(2) The expansion causal convolution is a convolution operation for performing a step-up operation on an input sequence, and specifically comprises the following steps:
Figure FDA0004065866200000022
wherein F (i) is a para sequence x 0 ,…,x T The convolution result of the ith element; h (j) is a filter, also known as a convolution kernel; d is an expansion factor, a standard causal convolution when d=1;
(3) Adding a residual block:
assuming that the input of the residual block is x and the output is o, the equation (9) is obtained by linear variation and mapping by the activation function:
o＝Activation(x+Γ(x)) (9)
in the formula, the Activation (·) is an Activation function, and Γ (x) is a relation function of x.
CN202310075013.4A 2023-01-14 2023-01-14 Battery remaining cycle life prediction method based on TCN Pending CN116224118A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310075013.4A CN116224118A (en) 2023-01-14 2023-01-14 Battery remaining cycle life prediction method based on TCN

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310075013.4A CN116224118A (en) 2023-01-14 2023-01-14 Battery remaining cycle life prediction method based on TCN

Publications (1)

Publication Number Publication Date
CN116224118A true CN116224118A (en) 2023-06-06

Family

ID=86574324

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310075013.4A Pending CN116224118A (en) 2023-01-14 2023-01-14 Battery remaining cycle life prediction method based on TCN

Country Status (1)

Country Link
CN (1) CN116224118A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117872167A (en) * 2024-03-12 2024-04-12 深圳市杰维工业设备有限公司 Battery performance influence factor analysis method

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117872167A (en) * 2024-03-12 2024-04-12 深圳市杰维工业设备有限公司 Battery performance influence factor analysis method
CN117872167B (en) * 2024-03-12 2024-05-14 深圳市杰维工业设备有限公司 Battery performance influence factor analysis method

Similar Documents

Publication Publication Date Title
CN109061504B (en) Method and system for predicting remaining service life of lithium ion batteries of same type and different types
Jinlei et al. A novel active equalization method for series-connected battery packs based on clustering analysis with genetic algorithm
Khalid et al. Prediction of li-ion battery state of charge using multilayer perceptron and long short-term memory models
CN111736084B (en) Valve-regulated lead-acid storage battery health state prediction method based on improved LSTM neural network
CN113064093A (en) Energy storage battery state of charge and state of health joint estimation method and system
Fernández-Serantes et al. Hybrid intelligent model to predict the SOC of a LFP power cell type
CN116224118A (en) Battery remaining cycle life prediction method based on TCN
CN112684363A (en) Lithium ion battery health state estimation method based on discharge process
CN116449218B (en) Lithium battery health state estimation method
De Sousa et al. Comparison of different approaches for lead acid battery state of health estimation based on artificial neural networks algorithms
CN115963420A (en) Battery SOH influence factor analysis method
CN113361692A (en) Lithium battery residual life combined prediction method
WO2022237661A1 (en) Lithium battery online life span prediction method based on two-point life span characteristics
Hasan et al. Performance comparison of machine learning methods with distinct features to estimate battery SOC
CN117031310A (en) Method for predicting residual service life of power battery of electric automobile
Cai et al. Remaining useful life prediction method of EV power battery for DC fast charging condition
Rastegarpanah et al. Predicting the remaining life of lithium-ion batteries using a CNN-LSTM model
Yang et al. Lithium-ion battery life cycle prediction with deep learning regression model
CN113884936B (en) ISSA coupling DELM-based lithium ion battery health state prediction method
Li et al. Remaining useful life prediction of lithium battery based on ACNN-Mogrifier LSTM-MMD
Hantono et al. LSTM for state of charge estimation of lithium polymer battery on Jetson nano
Fangdan et al. Capacity estimation of large-scale retired li-ion batteries for second use based on support vector machine
Zhang et al. State‐of‐Charge Prediction of Lithium‐Ion Batteries Based on Sparse Autoencoder and Gated Recurrent Unit Neural Network
Fan et al. SOH estimation method of lithium-ion battery based on multilayer feedforward neural network
Wang et al. State of health estimation for Lithium-ion battery based on fusion features

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