CN115047350A - Digital-analog linkage based lithium ion battery remaining service life prediction method - Google Patents

Digital-analog linkage based lithium ion battery remaining service life prediction method Download PDF

Info

Publication number
CN115047350A
CN115047350A CN202210729394.9A CN202210729394A CN115047350A CN 115047350 A CN115047350 A CN 115047350A CN 202210729394 A CN202210729394 A CN 202210729394A CN 115047350 A CN115047350 A CN 115047350A
Authority
CN
China
Prior art keywords
battery
attention mechanism
service life
representing
network
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
CN202210729394.9A
Other languages
Chinese (zh)
Other versions
CN115047350B (en
Inventor
张九思
李翔
罗浩
田纪伦
李明磊
蒋宇辰
尹珅
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.)
Harbin Institute of Technology
Original Assignee
Harbin Institute of Technology
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 Harbin Institute of Technology filed Critical Harbin Institute of Technology
Priority to CN202210729394.9A priority Critical patent/CN115047350B/en
Publication of CN115047350A publication Critical patent/CN115047350A/en
Application granted granted Critical
Publication of CN115047350B publication Critical patent/CN115047350B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/36—Arrangements for testing, measuring or monitoring the electrical condition of accumulators or electric batteries, e.g. capacity or state of charge [SoC]
    • G01R31/367—Software therefor, e.g. for battery testing using modelling or look-up tables
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/36—Arrangements for testing, measuring or monitoring the electrical condition of accumulators or electric batteries, e.g. capacity or state of charge [SoC]
    • G01R31/378—Arrangements for testing, measuring or monitoring the electrical condition of accumulators or electric batteries, e.g. capacity or state of charge [SoC] specially adapted for the type of battery or accumulator
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/36—Arrangements for testing, measuring or monitoring the electrical condition of accumulators or electric batteries, e.g. capacity or state of charge [SoC]
    • G01R31/392—Determining battery ageing or deterioration, e.g. state of health
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • 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

Abstract

The invention discloses a method for predicting the remaining service life of a lithium ion battery based on digital-analog linkage, and relates to a method for predicting the remaining service life of the lithium ion battery. The invention aims to solve the problems that a model and data-driven residual service life prediction method is difficult to combine, the traditional data-driven method is difficult to measure the uncertainty of the residual service life, and the importance degree of different moments in time window data is difficult to reflect. The process is as follows: step one, building a bidirectional gating cycle unit network model based on a time attention mechanism; the process is as follows: the model sequentially comprises three parts of an attention mechanism network, a bidirectional gating circulation unit network and a full connection layer; step two, training a bidirectional gating cycle unit network model based on a time attention mechanism; step three, constructing a battery degradation model based on particle filtering; and fourthly, predicting the residual service life on line. The method is suitable for the field of battery service life prediction.

Description

Digital-analog linkage based lithium ion battery remaining service life prediction method
Technical Field
The invention relates to the interdisciplinary field of deep learning, state space theory and battery remaining service life prediction, in particular to a lithium ion battery remaining service life prediction method.
Background
With the rapid development of new energy technologies, lithium ion batteries are widely used due to their advantages of high energy density, wide working temperature range, long cycle life, portability, and the like. However, during the cycle of charge and discharge of the lithium ion battery, the performance of the battery is gradually degraded due to the formation of an internal passivation film, the decomposition of an electrolyte, and the irreversible dissolution of an electrode active material, which greatly reduces the reliability and safety of electric devices. Therefore, how to accurately predict the Remaining service life (RUL) of the lithium ion battery in time is of great significance to the energy supply system.
The remaining useful life of a lithium-ion battery refers to the number of charge and discharge cycles that the state of health of the battery can undergo before degrading to a point where the device will not continue to operate or reach (failure threshold) under cyclic charge and discharge conditions. The existing lithium ion battery remaining service life prediction methods can be generally divided into two types: model-based methods and data-driven methods. The model-based method comprises an equivalent circuit model, an electrochemical model and a state space modeling method. However, due to the complexity of the electrochemical reactions inside the cell, there are significant limitations to accurate electrochemical modeling. Accordingly, state space modeling methods, such as Kalman Filter (KF), Extended Kalman Filter (EKF), and Particle Filter (PF), can implement effective remaining service life prediction in the framework of a probabilistic model. Different from a model-based residual service life method, the data-driven method only excavates the degradation rule of the battery from historical data without considering the internal chemical reaction and the fault reason of the battery, so that the residual service life prediction of the battery is realized. Common data-driven methods include machine learning methods such as Random Forest Regression (RFR), Support vector machine regression (SVR), Extreme gradient boosting (XGBoost), and deep learning methods such as a Convolutional Neural Network (CNN), a Recurrent Neural Network (RNN), and a variant Long-Term Memory Network (LSTM).
The method for predicting the remaining service life of the lithium ion battery based on deep learning, such as a long-time memory network, a gating cycle unit and the like, which are rising in the recent days, achieves the purpose of prediction by combining data into a time window form. Although these methods can learn the dependency of the battery capacity data on the time level through the gating structure, it is difficult to reflect the degree of importance of different times in the time window data, and there is no description of the uncertainty of the remaining service life of the lithium ion battery in the degradation process. Meanwhile, when the data-driven deep learning method is used for carrying out iterative prediction on the battery capacity in a time window mode, feedback and correction of data are lacked. Furthermore, when only a model-based method is used to predict the battery capacity, only trend extrapolation can be performed on the basis of a set prediction starting point, and model parameters are not updated in the process.
Disclosure of Invention
The invention aims to solve the problems that a model and data driving-based residual service life prediction method is difficult to combine, the traditional data driving method is difficult to measure the uncertainty of the residual service life and the importance degree of different moments in time window data is difficult to reflect, and provides a digital-analog linkage-based lithium ion battery residual service life prediction method.
A method for predicting the remaining service life of a lithium ion battery based on digital-analog linkage comprises the following specific processes:
step one, building a bidirectional gating cycle unit network model based on a time attention mechanism; the specific process is as follows:
the bidirectional gating circulation unit network model based on the time attention mechanism sequentially comprises an attention mechanism network, a bidirectional gating circulation unit network and a full connection layer;
step two, training a bidirectional gating cycle unit network model based on a time attention mechanism; the specific process is as follows:
training the model by taking 80% of historical data of the battery as a training data set, and taking the rest 20% of the historical data as a verification data set to check the prediction effect of the model;
inputting the training data set into the bidirectional gating cycle unit network model which is built in the first step and is based on the time attention mechanism, and building a mapping relation between input past battery capacity data and input future capacity data;
selecting and storing a bidirectional gating cycle unit network model with the best prediction effect on the verification data set based on a time attention mechanism for online battery capacity prediction;
step three, constructing a battery degradation model based on particle filtering;
and fourthly, predicting the residual service life on line.
The invention has the beneficial effects that:
the invention aims to solve the problem that a model-based residual service life prediction method and a data-driven residual service life prediction method are difficult to combine, and provides a digital-analog linkage battery residual service life prediction method based on a bidirectional gate control cycle unit (PF-TAM-BiGRU) of a particle filter-time attention machine system.
(1) Compared with the traditional data-driven residual service life prediction method, the time attention mechanism layer can consider the importance degree of different moments in time window data and allocate different weights to each time step in the time window data.
(2) In order to combine the advantages of a model-based method and a data driving method, the digital-analog linkage method combines a particle filter algorithm and a time attention mechanism-bidirectional gate control circulation unit network, and enables the two methods to be mutually corrected in the prediction process.
(3) According to the invention, by introducing a model-based particle filtering algorithm into the data-driven residual service life method, the uncertainty of the residual service life can be well described in the battery degradation process according to the distribution of particles.
Drawings
FIG. 1 is a work flow diagram of the present invention;
FIG. 2 is a schematic diagram of the time attention mechanism of the present invention;
FIG. 3 is a schematic diagram of the structure of a gated loop unit of the present invention;
FIG. 4 is a diagram of a deep neural network architecture for a time attention mechanism bi-directional gated loop unit, including details of portions of the network; wherein BiGRU represents a bidirectional gating circulating unit, and FC represents a full connection layer;
FIG. 5a is a graph showing the predicted and true values of battery capacity for B0005 battery at 60 th duty cycle according to the method for predicting remaining useful life of the present invention;
FIG. 5B is a graph showing the predicted and true values of battery capacity at 60 th duty cycle for the remaining useful life prediction method of the present invention for a B0006 battery;
FIG. 5c is a graph showing the predicted and true values of battery capacity at 60 th duty cycle for the remaining useful life prediction method of the present invention for a B0007 battery;
fig. 5d is a graph showing the predicted and true values of the battery capacity at the 60 th duty cycle for the B0018 battery in the method for predicting remaining useful life of the present invention.
Detailed Description
The first embodiment is as follows: the method for predicting the remaining service life of the lithium ion battery based on digital-analog linkage comprises the following specific processes:
step one, building a bidirectional gating cycle unit network model based on a time attention mechanism; the specific process is as follows:
the bidirectional gating circulation unit network model based on the time attention mechanism sequentially comprises an attention mechanism network, a bidirectional gating circulation unit network and a full connection layer;
the main functions are as follows:
attention mechanism network to time: an attention mechanism is adopted to give greater weight to important moments in an important time window;
bidirectional gated cyclic cell network: extracting the implicit time dependence in the time sequence data to obtain high-dimensional characteristics;
full connection of a single layer: mapping of the high-dimensional features to the remaining service life is achieved.
Step two, training a bidirectional gating cycle unit network model based on a time attention mechanism; the specific process is as follows:
training the model by taking 80% of historical data of the battery as a training data set, and taking the rest 20% of the historical data as a verification data set to check the prediction effect of the model;
inputting a training data set into the bidirectional gating cycle unit network model which is built in the step one and is based on the time attention mechanism, and constructing a mapping relation between input past battery capacity data and input future capacity data;
selecting and storing a bidirectional gating cycle unit network model with the best prediction effect on the verification data set based on a time attention mechanism for online battery capacity prediction;
step three, constructing a battery degradation model based on particle filtering;
a dynamic degradation model based on particle filtering is constructed to describe the degradation process of the battery through online data of the battery capacity, and the battery degradation model is constructed through the steps of particle initialization, importance sampling, weight updating and normalization and state variable updating, so that the battery degradation model is used for online residual service life prediction.
And fourthly, predicting the residual service life on line.
The particle filter algorithm and a bidirectional gating circulation unit of a time attention mechanism are combined, so that the two methods are mutually corrected in the prediction process. Specifically, starting from a set residual service life prediction starting point, updating parameters of a particle filter model by taking a predicted value of the residual service life of a bidirectional gating circulation unit based on a time attention mechanism as a posterior value of the particle filter, and inputting a predicted value of the battery capacity of the particle filter model into a time window for predicting the future battery capacity;
the above process is repeated until the predicted battery capacity is less than the battery threshold, and the number of battery cycles that are passed in the middle is the remaining service life of the battery.
Wherein, the first step to the second step belong to an off-line network training stage, and the third step to the fourth step belong to an on-line network prediction stage.
Evaluating the residual service life prediction effect:
and measuring the residual service life prediction effect of the lithium ion battery based on digital-analog linkage by adopting an Absolute Error (AE).
The second embodiment is as follows: the first step is to build a bidirectional gating cycle unit network model based on a time attention mechanism;
the bidirectional gating circulation unit network model based on the time attention mechanism sequentially comprises an attention mechanism network, a bidirectional gating circulation unit network and a full connection layer;
the specific process is as follows:
in deep learning, the more network model parameters are, the larger the amount of stored information is, the stronger the ability of approximating a nonlinear relationship is, but the problem of information overload is brought along. Attention mechanisms stem from the complex cognitive functions essential to humans, and can focus limited computing resources on information critical to the current task, while reducing attention to garbage. Therefore, the attention mechanism is introduced into the prediction problem of the lithium ion battery capacity, on the premise of not needing any prior knowledge of battery degradation, the importance degree of the capacities of different working cycles in a time window to the battery capacity of a future cycle is determined by using the attention neural network layer, a larger weight is given to an important moment, a smaller weight is given to a secondary moment, and the purpose of improving the prediction effect of the battery capacity is achieved.
The time window data input to the attention mechanism network may be expressed as:
Q＝[q 1 ,q 2 ,…,q j ,…,q L ]
wherein L is the length of the time window; q. q.s j Is the battery capacity at the jth moment;
a schematic drawing of the attention mechanism is shown in fig. 2. The core of the attention mechanism proposed by the invention is to construct the battery capacity q at the jth moment in a time window j And its importance s j The mapping relationship between the two can be expressed as shown in formula (1):
Figure BDA0003712408280000051
wherein s is j Is the importance degree of the jth moment, e is the natural logarithm, T is the transposition operation of the matrix, sigma is the sigmoid activation function, W j And b j Weights and offsets representing respective time instants;
the weights of the attention mechanism network can be written as W s ＝[W 1 ,…,W j ,…,W L ]，b s ＝[b 1 ,…,b j ,…,b L ]；
The importance degree is normalized by a softmax function, and the expression is shown as (2):
Figure BDA0003712408280000052
wherein alpha is j Representing the degree of importance of the respective moment;
on the basis, the output of the attention mechanism network can be obtained
Figure BDA0003712408280000053
The expression is shown in formula (3):
Figure BDA0003712408280000054
wherein the content of the first and second substances,
Figure BDA0003712408280000055
the output of the network layer is controlled for attention at the jth moment;
after the attention mechanism network has assigned weights to the battery capacities at different times in the time window, it will assign weights to the battery capacities at the different times in the time window
Figure BDA0003712408280000056
Inputting the data into a Bidirectional Gated Recurrent Unit (BiGRU);
the specific process of the bidirectional gating circulation unit network is as follows:
a gated cyclic unit is an improved recurrent neural network. A typical gated loop unit consists of an update gate for controlling the extent to which the current state utilizes past time information, and a reset gate that controls the extent to which new input information is combined with past memory. The schematic diagram of the gating cycle unit is shown in fig. 3, and the calculation formula of the gating cycle unit is shown as formula (4) -formula (7):
Figure BDA0003712408280000057
Figure BDA0003712408280000061
Figure BDA0003712408280000062
Figure BDA0003712408280000063
wherein the content of the first and second substances,
Figure BDA0003712408280000064
representing the input vector of the cell at time t, h t And h t-1 Representing the outputs of the network elements of the unit at time t and time t-1, respectively, z t 、r t 、c t Respectively representing the outputs of the refresh gate, reset gate and memory cell, W z 、W r 、W c Connection matrix, U, representing the input information of the refresh gate, reset gate and memory cell, respectively z And b z Respectively representing the weight and offset vector, U, of the update gate r And b r Respectively representing the weight and offset vector, U, of the reset gate c And b c Respectively representing the weight and the offset vector of the memory unit, sigma representing sigmoid activation function, tanh representing hyperbolic tangent function,
Figure BDA0003712408280000065
representing a dot product operation;
in order to utilize the reverse battery capacity time sequence, the invention adopts a bidirectional GRU network, adopts circulation layers in the forward direction and the reverse direction to respectively obtain the states of a hidden layer, and then obtains the output of the hidden layer through splicing. On this basis, the input to the forward GRU is
Figure BDA0003712408280000066
Through forward operation (contents of equations (4) to (7)), a forward output sequence of the hidden layer is obtained as shown in equation (8):
Figure BDA0003712408280000067
wherein the content of the first and second substances,
Figure BDA0003712408280000068
representing the mapping relation of forward GRU units;
accordingly, the inputs to the GRU are
Figure BDA0003712408280000069
The input sequence is opposite to the forward GRU, and the reverse output sequence of the hidden layer is obtained through reverse operation (the contents of the formulas (4) to (7)) as shown in the formula (9):
Figure BDA00037124082800000610
wherein the content of the first and second substances,
Figure BDA00037124082800000611
representing the mapping relation of backward GRU units;
the hidden layer output at the current moment is obtained as shown in formula (10):
Figure BDA00037124082800000612
the forward and reverse battery capacity sequence information is effectively combined in the splicing mode, the use efficiency of the information is increased, and therefore the battery capacity prediction effect of the traditional unidirectional GRU is improved.
Inputting the output of a Bidirectional Gated recirculation Unit (BiGRU) network layer into a full connection layer;
the specific process of the full connection layer is as follows:
after obtaining the output of the hidden layer, the TAM-BiGRU network will complete the conversion from the hidden layer to the future battery capacity through the full connection layer, i.e. equation (11)
Figure BDA00037124082800000613
The mapping relationship of (1):
Figure BDA0003712408280000071
wherein
Figure BDA0003712408280000072
A battery capacity value at the L +1 th moment predicted by a bidirectional gate-controlled cyclic unit network model (TAM-BiGRU) based on a time attention mechanism, wherein eta (eta) represents an entire all-connected layer mapping function, xi (eta)) represents an activation function of an all-connected layer, and W (eta) (. eta.)) represents an activation function of the all-connected layer 1 And b 1 Weight matrix and offset vector, W, representing the 1 st fully-connected layer, respectively u And b u Respectively representing the weight matrix and the offset vector of the u-th fully-connected layer.
Other steps and parameters are the same as those in the first embodiment.
The third concrete implementation mode: the second step is to train a bidirectional gating cycle unit network model based on a time attention mechanism; the specific process is as follows:
training the model by taking 80% of historical data of the battery as a training data set, and taking the rest 20% of the historical data as a verification data set to check the prediction effect of the model;
inputting a training data set into the bidirectional gating cycle unit network model which is built in the step one and is based on the time attention mechanism, and constructing a mapping relation between input past battery capacity data and input future capacity data;
selecting and storing a bidirectional gating cycle unit network model with the best prediction effect on the verification data set based on a time attention mechanism for online battery capacity prediction;
the specific process is as follows:
to minimize the difference between the input vector and the output vector, the network parameters of the two-way gated-round cell network model (TAM-BiGRU) based on the time attention mechanism will be parametrically updated by the mean square error loss function shown in equation (12):
Figure BDA0003712408280000073
where T' is the number of training data samples, a is the sample number, and W and b are the set of weight matrix and offset vector W ═ W, respectively s ,W z ,W r ,W c ,U z ,U r ,U c ,W 1 ,…,W u }，b＝{b s ,b z ,b r ,b c ,b 1 ,…,b u }；q a Is the actual battery capacity of the a-th sample,
Figure BDA0003712408280000074
for the predicted battery capacity of the a-th sample,
Figure BDA0003712408280000075
the square operation of two norms is carried out;
the loss function of the network model training is a mean square error loss function, the optimization algorithm is an Adam optimization algorithm, the learning rate is 0.001, and the network model training process is carried out in the hardware environment of 1 GPU (GTX 1660Ti display card).
Other steps and parameters are the same as those in the first or second embodiment.
The fourth concrete implementation mode: the present embodiment is different from the first to the third embodiments in that a battery degradation model based on particle filtering is constructed in the third step;
a dynamic degradation model based on particle filtering is constructed to describe the degradation process of the battery through online data of the battery capacity, and the battery degradation model is constructed through the steps of particle initialization, importance sampling, weight updating and normalization and state variable updating, so that the battery degradation model is used for online residual service life prediction.
The specific process is as follows:
compared with a Kalman filtering algorithm and an extended Kalman filtering algorithm, the particle filtering algorithm has more remarkable advantages in processing the parameter identification problem of a nonlinear non-Gaussian system. The particle filter based battery degradation model may be expressed in the form of a state space in the form shown in equation (13):
Figure BDA0003712408280000081
wherein x is k For the state variable at the kth duty cycle (previously offline, which is equivalent to online, k is the duty cycle number from start of commissioning to failed online), x k ＝[a k ,b k ,c k ,d k ] T ，a k For the first component of the state variable at the kth duty cycle (first term of the dual-exponential function in equation (14))
Figure BDA0003712408280000082
Coefficient of (b), b k For the second component of the state variable at the kth duty cycle (first term of the dual-exponential function in equation (14))
Figure BDA0003712408280000083
Index of (c)), c) k Is the third component of the state variable at the kth duty cycle (the second term of the dual-exponential function in equation (14))
Figure BDA0003712408280000084
Coefficient of (d), d k For the fourth component of the state variable at the kth duty cycle (second term of the dual-exponential function in equation (14))
Figure BDA0003712408280000085
Index of (d); f (x) k ) Is the state variable at the kth duty cycle, f (x) k )＝x k ；u k ＝[u a ,u b ,u c ,u d ] T For the noise term of the state transition equation, u a Is the noise term, u, of the first component of the state variable at the k-th duty cycle b As a noise term, u, of the second component of the state variable at the k-th duty cycle c Noise term, u, being the third component of the state variable at the k-th duty cycle d As a noise term of the fourth component of the state variable at the k-th duty cycle, v k To measure the noise term, v k ∈R 1×1 1 × 1 is a matrix with a length and width of 1,
Figure BDA0003712408280000086
q k is the k-thBattery capacity of the working cycle, g (x) k ) Is a measurement equation;
wherein the content of the first and second substances,
Figure BDA0003712408280000087
to measure the variance of the noise, N represents a normal distribution, representing obeying a certain distribution;
accordingly, the measurement equation can be written in the form of equation (14)
Figure BDA0003712408280000088
According to an initial state variable x 0 By a predetermined mean value of u 0 Variance is
Figure BDA0003712408280000089
Normal distribution of (x) 0 ) Can produce a collection of particles
Figure BDA00037124082800000810
N p Is the number of particles and the initial weight value of each particle is
Figure BDA0003712408280000091
The essence of the particle filtering algorithm is that Bayes filtering and Monte Carlo algorithms are fused, and the posterior probability density function of the particle set is solved through state variable updating and measurement updating. Specifically, for the kth battery operation process, the prior probability density function p (x) k |q 1:k-1 ) Can be expressed in the form shown in equation (15):
p(x k |q 1:k-1 )＝∫p(x k |x k-1 )p(x k-1 |q 1:k-1 )dx k-1 (15)
wherein q is 1:k-1 ＝[q 1 ,q 2 ,…,q k-1 ]Representing the battery capacity data from the initial state to the k-1 th working cycle, p (x) k |q 1:k-1 ) Is x k At q 1:k-1 Probability density function under the condition, p (x) k |x k-1 ) Is x k At x k-1 A probability density function under the condition;
after the k-th time battery capacity q is obtained k Then, according to bayesian filtering, the probability density function of the state variables under the posterior condition can be obtained, as shown in equation (16):
Figure BDA0003712408280000092
wherein, p (x) k |q 1:k ) Is x k At q 1:k Probability density function under the condition, p (q) k |x k ) Is q k At x k Probability density function under the condition, p (x) k |q 1:k-1 ) Is x k At q 1:k-1 Probability density function under the condition, p (q) k |q 1:k-1 ) Is q k At q 1:k-1 Probability density function under the condition, p (x) k |q 1:k-1 ) Is x k At q 1:k-1 A probability density function under the condition;
considering the difficulty of the integral operation in equation (16), it is necessary to convert the calculation of the posterior probability density function into the summation of particles by generating a large number of random particles in a manner as shown in equation (17):
Figure BDA0003712408280000093
where delta (.) represents the dirichlet function,
Figure BDA0003712408280000094
the weight represented by the ith particle in the k working process after normalization;
Figure BDA0003712408280000095
a state variable representing the ith particle;
to solve
Figure BDA0003712408280000096
The weight of the particle may be updated first, as shown in equation (18):
Figure BDA0003712408280000097
wherein the content of the first and second substances,
Figure BDA0003712408280000101
is q k In that
Figure BDA0003712408280000102
The probability density function under the conditions of the condition,
Figure BDA0003712408280000103
is composed of
Figure BDA0003712408280000104
In that
Figure BDA0003712408280000105
The probability density function under the conditions of the condition,
Figure BDA0003712408280000106
is composed of
Figure BDA0003712408280000107
In that
Figure BDA0003712408280000108
The probability density function under the conditions of the condition,
Figure BDA0003712408280000109
all state variables for the ith particle from the initial state to the k-1 th duty cycle,
Figure BDA00037124082800001010
is the weight of the ith particle at k-1 duty cycles,
Figure BDA00037124082800001011
the weight of the ith particle under k work cycles;
Figure BDA00037124082800001012
through weight normalization of the particles, a normalized particle weight expression can be obtained as shown in equation (19):
Figure BDA00037124082800001013
in order to remove the particles with low weight and keep the particles with high weight to continue the iteration process, the invention adopts a random resampling mode to carry out particle filtering to obtain
Figure BDA00037124082800001014
On the basis of the state variable, the state variable under the posterior probability density can be obtained
Figure BDA00037124082800001015
And measuring the variable
Figure BDA00037124082800001016
As shown in formulas (20) to (21):
Figure BDA00037124082800001017
Figure BDA00037124082800001018
wherein the content of the first and second substances,
Figure BDA00037124082800001019
is a state variable under the a-posteriori conditions,
Figure BDA00037124082800001020
for the capacity of the battery under the a posteriori conditions,
Figure BDA00037124082800001021
is the operation result of the posterior state variable through the measurement equation.
Other steps and parameters are the same as those in one of the first to third embodiments.
The fifth concrete implementation mode: the difference between the present embodiment and one of the first to the fourth embodiments is that the remaining service life is predicted online in the fourth step; the specific process is as follows:
assuming that the kth working process of the lithium ion battery is a predicted initial prediction point, a dual-exponential battery degradation model based on a particle filter shape as shown in formula (13) can be constructed, specifically, model parameters of the dual-exponential model are updated according to a particle filter algorithm by using online battery capacity data of the lithium ion battery from the 1 st working process to the kth working process, and the specific process is shown in formula (14) - (21). Next, after the k-th working process, the bidirectional gated cyclic unit network of the time attention mechanism will measure the battery capacity of the k + 1-th working process through the form of a time window
Figure BDA00037124082800001022
The prediction can be expressed in the form shown in equation (24):
Figure BDA0003712408280000111
wherein, g TAM-BiGRU Mapping of bi-directional gated cyclic units, q, representing a temporal attention mechanism k Represents the battery capacity of the kth work cycle;
on the basis of the formula (24), the predicted value of the battery capacity of the bidirectional gating cycle unit of the time attention mechanism is used as the posterior value of the particle filter double-exponential degradation model at the current moment, so as to guide the updating of the parameters of the particle filter model, and the updated value under the posterior condition of the battery capacity can be obtained through the formulas (17) - (21)
Figure BDA0003712408280000112
This updated mapping relationship can be expressed as shown in equation (25):
Figure BDA0003712408280000113
wherein, g PF Representing the mapping relation of the particle filter algorithm, and then, updating the particle filter
Figure BDA0003712408280000114
Adding the data into the time window data, correspondingly deleting the first data in the time window, and continuously performing the process until the predicted value of the battery capacity reaches the failure threshold value; the number of battery cycles experienced in the middle is the remaining useful life of the battery.
Suppose the v th (v)>L) Battery Capacity prediction value obtained by iterative prediction
Figure BDA0003712408280000115
For the first time below the failure threshold of the battery, the entire online prediction process can be combined into the form shown in expression (26):
Figure BDA0003712408280000116
wherein, g TAM-BiGRU Mapping of bidirectional gated cyclic units, g, representing a temporal attention mechanism PF Representing the mapping of the particle filter algorithm, q k Represents the battery capacity of the kth work cycle;
the number of cycles v is a predicted value of the remaining service life of the battery during the kth operation.
Other steps and parameters are the same as in one of the first to fourth embodiments.
The sixth specific implementation mode: the difference between the present embodiment and one of the first to fifth embodiments is that the bidirectional gated loop unit network model based on the time attention mechanism sequentially includes three parts, namely an attention mechanism network for time, a 3-layer bidirectional gated loop unit network and a 2-layer fully-connected layer;
the number of neurons of the 3-layer bidirectional gating circulation unit network is 128;
wherein the number of layer 1 fully-linked layer neurons is 64;
the number of layer 2 full connectivity layer neurons was 128.
The structure of the deep neural network of the time attention mechanism bidirectional gating cycle unit of the invention is shown in FIG. 4.
Other steps and parameters are the same as those in one of the first to fifth embodiments.
The following examples were used to demonstrate the beneficial effects of the present invention:
the first embodiment is as follows:
the invention provides a residual service life prediction method based on digital-analog linkage, which is provided by lithium ion battery data verification provided by NASA Ames prediction center. The data set contains data generated from 4 lithium ion cells (nos. B0005, B0006, B0007, and B0018) rated for 2Ah operating at room temperature. The 4 batteries include two modes of operation, a discharge mode and a charge mode. For the discharge mode, 4 cells will be discharged under a constant current condition of 2A, and the discharge will be stopped until the voltage of each cell drops to a certain threshold. For the charge mode, 4 cells will be charged under a constant current condition of 1.5A until the voltage of the cells rises to 4.2V. Then, the charging was continued again in a constant voltage manner until the charging current dropped to 20 mA. After each charge and discharge cycle is finished, the lithium ion battery obtains internal parameters of the battery in a mode of impedance test, and therefore the battery capacity of the battery at the end of the corresponding cycle is obtained. The cell capacity fade curves for these 4 cells are shown in fig. 3. The related literature indicates that when the capacity of the lithium ion battery is reduced to 70% -80% of the initial capacity, the performance of the battery is difficult to achieve the performance of normal operation. Considering that the B0007 cell did not reach 70% of the initial capacity, i.e., 1.4Ah, the present invention sets the degradation threshold of 4 cells to 1.45Ah for the convenience of study.
Step 1, constructing a bidirectional gating cycle unit network based on a time attention mechanism: the bidirectional gated cycle unit network of the time attention mechanism comprises three parts of an attention mechanism sub-network, a bidirectional gated cycle unit sub-network and a full connection layer sub-network. Thus, a bidirectional gated cyclic cell network based on a temporal attention mechanism can be constructed according to the structure shown in FIG. 4.
Step 2, training a bidirectional gating circulation unit network based on a time attention mechanism: and (3) inputting the historical data of the battery into the neural network built in the step (1) as a training data set of the neural network, and constructing a mapping relation between the input past battery capacity data and the future capacity data. And randomly dividing 80% of the whole data set to serve as a training data set training model, and using the rest 20% of the whole data set as a verification data set to check the prediction effect of the model. The model with the best prediction effect on the verification data set is selected and stored for final prediction. In this embodiment, when one battery is used as test data, the capacity degradation data of the other three batteries is used as a training data training model. For example, when B0005 batteries are used as test data, B0006, B0007 and B0018 batteries are used as training data.
And 3, constructing a battery degradation model based on particle filtering: a dynamic degradation model based on particle filtering is constructed to describe the degradation process of the battery through online data of the battery capacity, and the battery degradation model is constructed through the steps of particle initialization, importance sampling, weight updating and normalization and state variable updating. Assuming that the k-th work cycle is taken as a prediction starting point, a battery degradation model based on particle filtering is constructed from online data from an initial state to the k-th work cycle for the remaining service life prediction of the k-th process as the prediction starting point.
And 4, predicting the remaining service life on line: and combining the particle filter algorithm with the TAM-BiGRU, and mutually correcting the two methods in the prediction process. Specifically, from the set remaining service life prediction start point, the particle filter model parameters are updated with the predicted value of the remaining service life of the TAM-BiGRU as the posterior value of the particle filter, and the predicted value of the battery capacity of the particle filter model is input into the time window for predicting the future battery capacity. The above process is repeated until the predicted battery capacity is less than the battery threshold, and the number of battery cycles that are passed in the middle is the remaining service life of the battery. Finally, each particle is respectively substituted into the model to obtain a confidence interval of the remaining service life, so that the uncertainty of the remaining service life of the battery in the degradation process is well described. Taking the 60 th cycle of each battery as an example, fig. 5a, 5b, 5c, 5d respectively show the true and predicted values of the remaining service life of 4 batteries on the NASA battery data set by the proposed method. It can be seen that the predicted value and the true value of the residual service life in the method provided by the invention are very close, so that the method provided by the invention has a good prediction effect. Table 1 shows the confidence interval of the residual service life prediction of 95%, and it can be seen that the true values of the residual service life substantially fall within the confidence interval, so that it can be known that the method provided by the present invention can well describe the uncertainty of the battery in the degradation process.
TABLE 1 confidence intervals for prediction of 95% of remaining useful life
Figure BDA0003712408280000131
And 5, evaluating the prediction effect of the residual service life: the prediction effect of the residual service life of the lithium ion battery based on digital-analog linkage is measured by adopting Absolute Error (AE), and the specific result is summarized as shown in Table 2.
TABLE 2 Absolute error of remaining life prediction
Figure BDA0003712408280000132
The present invention is capable of other embodiments and its several details are capable of modifications in various obvious respects, all without departing from the spirit and scope of the present invention.

Claims (6)

1. A method for predicting the remaining service life of a lithium ion battery based on digital-analog linkage is characterized by comprising the following steps: the method comprises the following specific processes:
step one, building a bidirectional gating cycle unit network model based on a time attention mechanism; the specific process is as follows:
the bidirectional gating circulation unit network model based on the time attention mechanism sequentially comprises an attention mechanism network, a bidirectional gating circulation unit network and a full connection layer;
step two, training a bidirectional gating cycle unit network model based on a time attention mechanism; the specific process is as follows:
training the model by taking 80% of historical data of the battery as a training data set, and taking the rest 20% of the historical data as a verification data set to check the prediction effect of the model;
inputting a training data set into the bidirectional gating cycle unit network model which is built in the step one and is based on the time attention mechanism, and constructing a mapping relation between input past battery capacity data and input future capacity data;
selecting and storing a bidirectional gating cycle unit network model with the best prediction effect on the verification data set based on a time attention mechanism for online battery capacity prediction;
step three, constructing a battery degradation model based on particle filtering;
and fourthly, predicting the residual service life on line.
2. The method for predicting the remaining service life of the lithium ion battery based on digital-analog linkage according to claim 1, wherein the method comprises the following steps: building a bidirectional gating cycle unit network model based on a time attention mechanism in the first step;
the bidirectional gating circulation unit network model based on the time attention mechanism sequentially comprises an attention mechanism network, a bidirectional gating circulation unit network and a full connection layer;
the specific process is as follows:
the time window data input to the attention mechanism network may be expressed as:
Q＝[q 1 ,q 2 ,…,q j ,…,q L ]
wherein L is the length of the time window; q. q.s j Is the battery capacity at the jth moment;
constructing battery capacity q at jth moment in time window j And its importance s j The mapping relationship between the two can be expressed as shown in formula (1):
Figure FDA0003712408270000011
wherein s is j Is the importance degree of the jth moment, e is the natural logarithm, T is the transposition operation of the matrix, sigma is the sigmoid activation function, W j And b j Weights and offsets representing respective time instants;
the weights of the attention mechanism network can be written as W s ＝[W 1 ,…,W j ,…,W L ]，b s ＝[b 1 ,…,b j ,…,b L ]；
The importance degree is normalized by a softmax function, and the expression is shown as (2):
Figure FDA0003712408270000021
wherein alpha is j Representing the degree of importance of the respective moment;
on the basis, the output of the attention mechanism network can be obtained
Figure FDA0003712408270000022
The expression is shown as formula (3)：
Figure FDA0003712408270000023
Wherein the content of the first and second substances,
Figure FDA0003712408270000024
the output of the network layer is controlled for attention at the jth moment;
will be provided with
Figure FDA0003712408270000025
Inputting the data into a bidirectional gating circulation unit network;
the specific process of the bidirectional gating circulation unit network is as follows:
the calculation formula of the gating cycle unit is shown as the formula (4) to the formula (7):
Figure FDA0003712408270000026
Figure FDA0003712408270000027
Figure FDA0003712408270000028
Figure FDA0003712408270000029
wherein the content of the first and second substances,
Figure FDA00037124082700000210
representing the input vector of the cell at time t, h t And h t-1 Representing the outputs of the network elements of the unit at time t and time t-1, respectively, z t 、r t 、c t Are respectively provided withRepresenting the outputs of the refresh gate, reset gate and memory cell, W z 、W r 、W c Connection matrix, U, representing the input information of the refresh gate, reset gate and memory cell, respectively z And b z Respectively representing the weight and offset vector, U, of the update gate r And b r Respectively representing the weight and offset vector, U, of the reset gate c And b c Respectively representing the weight and the offset vector of the memory unit, sigma representing sigmoid activation function, tanh representing hyperbolic tangent function,
Figure FDA00037124082700000211
representing a dot product operation;
with a bidirectional GRU network, the input to the forward GRU is
Figure FDA00037124082700000212
The forward output sequence for obtaining the hidden layer is shown as formula (8):
Figure FDA00037124082700000213
wherein the content of the first and second substances,
Figure FDA00037124082700000214
representing the mapping relation of forward GRU units;
accordingly, the inputs to the GRU are
Figure FDA00037124082700000215
Obtaining the reverse output sequence of the hidden layer is shown as formula (9):
Figure FDA00037124082700000216
wherein the content of the first and second substances,
Figure FDA0003712408270000031
representing backward GRU unit mappingA relationship;
the hidden layer output at the current moment is obtained as shown in formula (10):
Figure FDA0003712408270000032
inputting the output of the network layer of the bidirectional gating circulation unit into a full connection layer;
the specific process of the full connection layer is as follows:
equation (11) accomplishes the transformation from hidden layer to future battery capacity
Figure FDA0003712408270000033
The mapping relationship of (1):
Figure FDA0003712408270000034
wherein
Figure FDA0003712408270000035
The battery capacity value at the L +1 th moment predicted by the bidirectional gating circulation unit network model based on the time attention mechanism is represented by eta (eta)) in a mapping function of the whole full-connection layer, xi (eta)) in an activation function of the full-connection layer, and W 1 And b 1 Weight matrix and offset vector, W, representing the 1 st fully-connected layer, respectively u And b u Respectively representing the weight matrix and the offset vector of the u-th fully-connected layer.
3. The method for predicting the remaining service life of the lithium ion battery based on digital-analog linkage according to claim 2, wherein the method comprises the following steps: training a bidirectional gating cycle unit network model based on a time attention mechanism in the second step; the specific process is as follows:
training the model by taking 80% of historical data of the battery as a training data set, and taking the rest 20% of the historical data as a verification data set to check the prediction effect of the model;
inputting a training data set into the bidirectional gating cycle unit network model which is built in the step one and is based on the time attention mechanism, and constructing a mapping relation between input past battery capacity data and input future capacity data;
selecting and storing a bidirectional gating cycle unit network model with the best prediction effect on the verification data set based on a time attention mechanism for online battery capacity prediction;
the specific process is as follows:
the network parameters of the bidirectional gated cyclic unit network model based on the time attention mechanism are subjected to parameter updating by a mean square error loss function shown in formula (12):
Figure FDA0003712408270000036
where T' is the number of training data samples, a is the sample number, and W and b are the set of weight matrix and offset vector W ═ W, respectively s ,W z ,W r ,W c ,U z ,U r ,U c ,W 1 ,…,W u }，b＝{b s ,b z ,b r ,b c ,b 1 ,…,b u }；q a Is the actual battery capacity of the a-th sample,
Figure FDA0003712408270000037
for the predicted battery capacity of the a-th sample,
Figure FDA0003712408270000038
the square operation of two norms is carried out;
the loss function of the network model training is a mean square error loss function, the optimization algorithm is an Adam optimization algorithm, the learning rate is 0.001, and the network model training process is carried out in the hardware environment of 1 GPU.
4. The method for predicting the remaining service life of the lithium ion battery based on digital-analog linkage according to claim 3, wherein the method comprises the following steps: constructing a battery degradation model based on particle filtering in the third step;
the specific process is as follows:
the particle filter based battery degradation model may be expressed in the form of a state space in the form shown in equation (13):
Figure FDA0003712408270000041
wherein x is k Is a state variable at the kth duty cycle, x k ＝[a k ,b k ,c k ,d k ] T ，a k As the first component of the state variable at the k-th duty cycle, b k As a second component of the state variable at the k-th duty cycle, c k Is the third component of the state variable at the k-th duty cycle, d k Is the fourth component of the state variable at the kth duty cycle; f (x) k ) Is the state variable at the kth duty cycle, f (x) k )＝x k ；u k ＝[u a ,u b ,u c ,u d ] T For the noise term of the state transition equation, u a Is the noise term, u, of the first component of the state variable at the k-th duty cycle b As a noise term, u, of the second component of the state variable at the k-th duty cycle c Noise term, u, being the third component of the state variable at the k-th duty cycle d As a noise term of the fourth component of the state variable at the k-th duty cycle, v k In order to measure the noise term(s),
Figure FDA0003712408270000042
q k battery capacity, g (x) for the k-th duty cycle k ) Is a measurement equation;
wherein the content of the first and second substances,
Figure FDA0003712408270000043
to measure the variance of the noise, N represents a normal distribution, representing obeying a certain distribution;
accordingly, the measurement equation can be written in the form of equation (14)
Figure FDA0003712408270000044
According to an initial state variable x 0 By a predetermined mean value of u 0 Variance is
Figure FDA0003712408270000045
Normal distribution of (x) 0 ) A collection of particles can be generated
Figure FDA0003712408270000046
N p Is the number of particles and the initial weight value of each particle is
Figure FDA0003712408270000047
For the kth battery operation, the prior probability density function p (x) k |q 1:k-1 ) Can be expressed in the form shown in equation (15):
p(x k |q 1:k-1 )＝∫p(x k |x k-1 )p(x k-1 |q 1:k-1 )dx k-1 (15)
wherein q is 1:k-1 ＝[q 1 ,q 2 ,…,q k-1 ]Representing the battery capacity data from the initial state to the k-1 th working cycle, p (x) k |q 1:k-1 ) Is x k At q 1:k-1 Probability density function under the condition, p (x) k |x k-1 ) Is x k At x k-1 A probability density function under the condition;
after the k-th battery capacity q is obtained k Then, according to bayesian filtering, the probability density function of the state variables under the posterior condition can be obtained, as shown in equation (16):
Figure FDA0003712408270000051
wherein, p (x) k |q 1:k ) Is x k At q 1:k Probability density function under the condition, p (q) k |x k ) Is q k At x k Probability density function under the condition, p (x) k |q 1:k-1 ) Is x k At q 1:k-1 Probability density function under the condition, p (q) k |q 1:k-1 ) Is q k At q 1:k-1 Probability density function under the condition, p (x) k |q 1:k-1 ) Is x k At q 1:k-1 A probability density function under the condition;
the calculation of the posterior probability density function is converted to a summation of particles, as shown in equation (17):
Figure FDA0003712408270000052
where delta (.) represents the dirichlet function,
Figure FDA0003712408270000053
the weight represented by the ith particle in the k working process after normalization;
Figure FDA0003712408270000054
a state variable representing the ith particle;
to solve
Figure FDA0003712408270000055
The weight of the particle may be updated first, as shown in equation (18):
Figure FDA0003712408270000056
wherein the content of the first and second substances,
Figure FDA0003712408270000057
is q k In that
Figure FDA0003712408270000058
The probability density function under the conditions of the condition,
Figure FDA0003712408270000059
is composed of
Figure FDA00037124082700000510
In that
Figure FDA00037124082700000511
The probability density function under the conditions of the condition,
Figure FDA00037124082700000512
is composed of
Figure FDA00037124082700000513
In that
Figure FDA00037124082700000514
The probability density function under the conditions of the condition,
Figure FDA00037124082700000515
all state variables for the ith particle from the initial state to the k-1 th duty cycle,
Figure FDA00037124082700000516
is the weight of the ith particle at k-1 duty cycles,
Figure FDA00037124082700000517
the weight of the ith particle under k work cycles;
Figure FDA00037124082700000518
through weight normalization of the particles, a normalized particle weight expression can be obtainedAs shown in equation (19):
Figure FDA0003712408270000061
the particle filtering is carried out by adopting a random resampling mode to obtain
Figure FDA0003712408270000062
The state variable under the posterior probability density can be obtained
Figure FDA0003712408270000063
And measuring the variable
Figure FDA0003712408270000064
As shown in formulas (20) to (21):
Figure FDA0003712408270000065
Figure FDA0003712408270000066
wherein the content of the first and second substances,
Figure FDA0003712408270000067
is a state variable under the a-posteriori conditions,
Figure FDA0003712408270000068
for the capacity of the battery under the a posteriori conditions,
Figure FDA0003712408270000069
is the operation result of the posterior state variable through the measurement equation.
5. The method for predicting the remaining service life of the lithium ion battery based on digital-analog linkage according to claim 4, wherein the method comprises the following steps: predicting the remaining service life on line in the fourth step; the specific process is as follows:
suppose the v th (v)>L) Battery Capacity prediction value obtained by iterative prediction
Figure FDA00037124082700000610
For the first time below the failure threshold of the battery, the entire online prediction process can be combined into the form shown in expression (26):
Figure FDA00037124082700000611
wherein, g TAM-BiGRU Mapping of bidirectional gated cyclic units, g, representing a temporal attention mechanism PF Representing the mapping of the particle filter algorithm, q k Represents the battery capacity of the kth work cycle;
the number of cycles v is a predicted value of the remaining service life of the battery during the kth operation.
6. The method for predicting the remaining service life of the lithium ion battery based on digital-analog linkage according to claim 5, wherein the method comprises the following steps: the bidirectional gating circulation unit network model based on the time attention mechanism sequentially comprises three parts of an attention mechanism network, a 3-layer bidirectional gating circulation unit network and a 2-layer full-connection layer;
the number of neurons of the 3-layer bidirectional gating circulation unit network is 128;
wherein the number of layer 1 fully-linked layer neurons is 64;
the number of layer 2 full connectivity layer neurons was 128.
CN202210729394.9A 2022-06-24 2022-06-24 Digital-analog linkage based lithium ion battery remaining service life prediction method Active CN115047350B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210729394.9A CN115047350B (en) 2022-06-24 2022-06-24 Digital-analog linkage based lithium ion battery remaining service life prediction method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210729394.9A CN115047350B (en) 2022-06-24 2022-06-24 Digital-analog linkage based lithium ion battery remaining service life prediction method

Publications (2)

Publication Number Publication Date
CN115047350A true CN115047350A (en) 2022-09-13
CN115047350B CN115047350B (en) 2023-04-18

Family

ID=83164276

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210729394.9A Active CN115047350B (en) 2022-06-24 2022-06-24 Digital-analog linkage based lithium ion battery remaining service life prediction method

Country Status (1)

Country Link
CN (1) CN115047350B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115718263A (en) * 2023-01-09 2023-02-28 北京科技大学 Attention-based lithium ion battery calendar aging prediction model and method

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170205466A1 (en) * 2014-12-29 2017-07-20 Hefei University Of Technology Method for predicting remaining useful life of lithium battery based on wavelet denoising and relevance vector machine
US20180143257A1 (en) * 2016-11-21 2018-05-24 Battelle Energy Alliance, Llc Systems and methods for estimation and prediction of battery health and performance
CN111103544A (en) * 2019-12-26 2020-05-05 江苏大学 Lithium ion battery remaining service life prediction method based on long-time and short-time memory LSTM and particle filter PF
CN112784798A (en) * 2021-02-01 2021-05-11 东南大学 Multi-modal emotion recognition method based on feature-time attention mechanism
CN113094822A (en) * 2021-03-12 2021-07-09 华中科技大学 Method and system for predicting residual life of mechanical equipment
CN113204921A (en) * 2021-05-13 2021-08-03 哈尔滨工业大学 Method and system for predicting remaining service life of airplane turbofan engine
CN113971489A (en) * 2021-10-25 2022-01-25 哈尔滨工业大学 Method and system for predicting remaining service life based on hybrid neural network
CN114186500A (en) * 2022-02-16 2022-03-15 华中科技大学 Marine bearing residual life prediction method based on transfer learning and multiple time windows
CN114266278A (en) * 2021-12-29 2022-04-01 合肥工业大学 Dual-attention-network-based method for predicting residual service life of equipment

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170205466A1 (en) * 2014-12-29 2017-07-20 Hefei University Of Technology Method for predicting remaining useful life of lithium battery based on wavelet denoising and relevance vector machine
US20180143257A1 (en) * 2016-11-21 2018-05-24 Battelle Energy Alliance, Llc Systems and methods for estimation and prediction of battery health and performance
CN111103544A (en) * 2019-12-26 2020-05-05 江苏大学 Lithium ion battery remaining service life prediction method based on long-time and short-time memory LSTM and particle filter PF
CN112784798A (en) * 2021-02-01 2021-05-11 东南大学 Multi-modal emotion recognition method based on feature-time attention mechanism
CN113094822A (en) * 2021-03-12 2021-07-09 华中科技大学 Method and system for predicting residual life of mechanical equipment
CN113204921A (en) * 2021-05-13 2021-08-03 哈尔滨工业大学 Method and system for predicting remaining service life of airplane turbofan engine
CN113971489A (en) * 2021-10-25 2022-01-25 哈尔滨工业大学 Method and system for predicting remaining service life based on hybrid neural network
CN114266278A (en) * 2021-12-29 2022-04-01 合肥工业大学 Dual-attention-network-based method for predicting residual service life of equipment
CN114186500A (en) * 2022-02-16 2022-03-15 华中科技大学 Marine bearing residual life prediction method based on transfer learning and multiple time windows

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115718263A (en) * 2023-01-09 2023-02-28 北京科技大学 Attention-based lithium ion battery calendar aging prediction model and method
CN115718263B (en) * 2023-01-09 2023-04-07 北京科技大学 Attention-based lithium ion battery calendar aging prediction model and method

Also Published As

Publication number Publication date
CN115047350B (en) 2023-04-18

Similar Documents

Publication Publication Date Title
CN110824364B (en) Lithium battery SOH estimation and RUL prediction method based on AST-LSTM neural network
Liu et al. Remaining useful life estimation for proton exchange membrane fuel cells using a hybrid method
CN112241608B (en) Lithium battery life prediction method based on LSTM network and transfer learning
Wang et al. Improved anti-noise adaptive long short-term memory neural network modeling for the robust remaining useful life prediction of lithium-ion batteries
Khalid et al. Prediction of li-ion battery state of charge using multilayer perceptron and long short-term memory models
Ji et al. An RUL prediction approach for lithium-ion battery based on SADE-MESN
CN110738344B (en) Distributed reactive power optimization method and device for load prediction of power system
Liu et al. A novel remaining useful life prediction method for lithium-ion battery based on long short-term memory network optimized by improved sparrow search algorithm
CN111680848A (en) Battery life prediction method based on prediction model fusion and storage medium
CN111948563B (en) Electric forklift lithium battery residual life prediction method based on multi-neural network coupling
CN111736084B (en) Valve-regulated lead-acid storage battery health state prediction method based on improved LSTM neural network
CN113406521B (en) Lithium battery health state online estimation method based on feature analysis
Zhang et al. Weight optimized unscented Kalman filter for degradation trend prediction of lithium-ion battery with error compensation strategy
CN115047350B (en) Digital-analog linkage based lithium ion battery remaining service life prediction method
CN115201686B (en) Lithium ion battery health state assessment method under incomplete charge and discharge data
CN115453399A (en) Battery pack SOH estimation method considering inconsistency
CN114839542A (en) Sample data set generation method and SOC estimation method for power lithium battery
Zhou et al. Remaining useful life prediction and state of health diagnosis for lithium-ion batteries based on improved grey wolf optimization algorithm-deep extreme learning machine algorithm
CN116047314B (en) Rechargeable battery health state prediction method
CN113361692A (en) Lithium battery residual life combined prediction method
Ansari et al. Jellyfish optimized recurrent neural network for state of health estimation of lithium-ion batteries
De Lima et al. State-of-charge estimation of a li-ion battery using deep forward neural networks
Song et al. Capacity estimation method of lithium-ion batteries based on deep convolution neural network
CN113093014B (en) Online collaborative estimation method and system for SOH and SOC based on impedance parameters
CN109977622B (en) Method for predicting residual life of power battery

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