CN114740388A - Lithium battery residual life state evaluation method based on improved TCN - Google Patents

Lithium battery residual life state evaluation method based on improved TCN Download PDF

Info

Publication number
CN114740388A
CN114740388A CN202210349477.5A CN202210349477A CN114740388A CN 114740388 A CN114740388 A CN 114740388A CN 202210349477 A CN202210349477 A CN 202210349477A CN 114740388 A CN114740388 A CN 114740388A
Authority
CN
China
Prior art keywords
capacity
cap
tcn
battery
model
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
CN202210349477.5A
Other languages
Chinese (zh)
Inventor
金心宇
陈民申
任泽华
金昀程
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.)
Zhejiang University ZJU
Original Assignee
Zhejiang University ZJU
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 Zhejiang University ZJU filed Critical Zhejiang University ZJU
Priority to CN202210349477.5A priority Critical patent/CN114740388A/en
Publication of CN114740388A publication Critical patent/CN114740388A/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
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/36—Arrangements for testing, measuring or monitoring the electrical condition of accumulators or electric batteries, e.g. capacity or state of charge [SoC]
    • G01R31/382—Arrangements for monitoring battery or accumulator variables, e.g. SoC
    • G01R31/3842—Arrangements for monitoring battery or accumulator variables, e.g. SoC combining voltage and current measurements
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E60/00—Enabling technologies; Technologies with a potential or indirect contribution to GHG emissions mitigation
    • Y02E60/10—Energy storage using batteries

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Secondary Cells (AREA)

Abstract

The invention discloses a method for evaluating the state of the remaining life of a lithium battery based on improved TCN (thyristor controlled reactor), which comprises the steps of obtaining actually-measured voltage and current data under the circulation working condition of the lithium battery, carrying out data slicing, ampere-hour integration and averaging treatment to obtain a circulation period and actual capacity sequence as the input of an A-TCN-DAE model, and predicting to obtain battery capacity data CapcurThen predicting the subsequent capacity of the battery in a recursion mode until Capcur≤CapEOLAnd ending the recursion prediction, and counting the residual cycle number reaching the EOL capacity to obtain the current residual life of the battery. The invention relates to a lithium battery residueThe residual life state evaluation method has strong local feature capture capability and adaptive capability for denoising and reconstructing input data, and compared with the existing deep learning network technology, the accuracy of estimating the residual service life RUL of the battery is obviously improved.

Description

Lithium battery residual life state evaluation method based on improved TCN
Technical Field
The invention relates to the technical field of lithium batteries, in particular to a lithium battery residual life state evaluation method based on improved TCN.
Background
The lithium battery provides energy, actually, the chemical reaction process of the anode, the cathode and the internal components in the battery is realized, and inevitable loss can occur along with the increase of the time in the use process. The difference of the single batteries leaving factory and the inconsistency of the loss degree of the single batteries in the charging and discharging process can lead the service lives of the single batteries to be different finally. After the single battery reaches an End of Life (EOL) state, the performance of the normal single battery is affected according to the short plate effect after series-parallel connection, and then the reliability and safety of the whole vehicle operation are affected, and the consequences are very serious. Therefore, the Remaining service life (RUL) of the battery is predicted based on a data driving mode, the battery in which the single battery in the battery pack reaches the EOL state is sensed in time for maintenance, and the safe use of the whole vehicle system can be guaranteed.
Research shows that various RUL prediction methods exist at present. The method includes the steps that battery performance degradation data under different time intervals are built by means of particle filtering based on a battery model, the battery impedance data under different aging states are fitted, the relation between impedance and the aging states is obtained, further RUL prediction is conducted by means of impedance characteristics, model accuracy and prediction capability under stable conditions are very reliable, the model is easily affected by external factors, battery characteristics can be changed due to different working conditions and different environmental temperatures for a power battery system, and the model is difficult to obtain accurate battery mechanism representation capability under current actual operation. The capacity and cycle mapping relation is established for the aged battery close to the EOL state based on an SVM model, the RUL of the battery close to the EOL state can be predicted, but the SVM model has the defect that prediction under large data cannot be processed. From the battery impedance change. The existing battery RUL prediction method does not reasonably and effectively capture the capacity recovery characteristic in the battery attenuation process, only considers the general trend of the attenuation process and cannot ensure the accuracy of local prediction. On the other hand, most models are insufficient in robustness, and the predicted result is greatly influenced under the influence of environmental noise.
Accordingly, there is a need for improvements in the art.
Disclosure of Invention
The invention aims to provide a method for evaluating the state of the remaining service life of a lithium battery based on an improved TCN (thyristor controlled network), which is used for quickly and accurately estimating the remaining service life of the lithium battery.
In order to solve the technical problem, the invention provides a method for evaluating the residual life state of a lithium battery based on improved TCN, which comprises the following specific processes:
step S01, the lithium battery to be tested is charged to full charge by using a constant current and constant voltage charging mode, and then is discharged by using constant currents of 0.5C and 1C respectively until the voltage of the battery is reduced to 2.7V of a threshold value; repeating the operation for 5 times to obtain the actually measured voltage and current data of the lithium battery under 5 circulation working conditions to obtain voltage and current curves under the circulation working conditions;
step S02, taking the cut-off voltage 2.7V of the voltage data under the circulation working condition as a slice cut-off point, performing data slicing as the end of the current circulation period and the start of the next circulation period, and performing slice labeling on each data slice according to the sequence, wherein one data slice is a circulation period;
step S03, adopting an ampere-hour integration method for the discharging process in each cycle period and carrying out averaging treatment to obtain the actual capacity; summarizing according to the mapping relation between the fragment labels and the actual capacity to obtain a cycle period and an actual capacity sequence;
step S04, taking the cycle period and the actual capacity sequence as the input of an A-TCN-DAE model, wherein the A-TCN-DAE model comprises a noise reduction self-coding DAE network and an A-TCN model, and the cycle period and the actual capacity sequence output by the noise reduction self-coding DAE network after reconstruction optimization predict the battery capacity data Cap at the current moment through the A-TCN modelcurThen, the subsequent capacity of the battery is predicted in a recursion mode, and the capacity value Cap obtained once prediction is carried outcurPresetting a capacity value Cap corresponding to a battery EOLEOLThe following conditions are met: capcur≤CapEOLEnding the recursion prediction process, and then counting the residual cycle number reaching the EOL capacity to obtainThe current remaining life of the battery.
The invention relates to an improvement of a lithium battery residual life state evaluation method based on improved TCN, which comprises the following steps:
the A-TCN model is based on the TCN model, and each residual error module RB is improved as follows: 1) replacing causal convolution with mixed dilation convolution, 2) replacing an activation function ReLU with A-ReLU, 3) adding a 1 × 1 convolution structure between the input and output of a residual module RB; and each residual module RB calculates the respective residual and then transmits the residual layer by layer backwards.
The invention is further improved by the method for evaluating the residual life state of the lithium battery based on the improved TCN:
the de-noising auto-encoder DAE comprises mixing the input x with the noise v followed by the input samples
Figure BDA0003578899550000021
Inputting samples
Figure BDA0003578899550000022
Extracting data characteristics through an encoding network E (-) and reconstructing, optimizing and outputting a cycle period and an actual capacity sequence y after reconstruction and optimization through a decoding network D (-) after:
Figure BDA0003578899550000023
wherein,
Figure BDA0003578899550000024
in order to reconstruct the output that is optimized,
Figure BDA0003578899550000025
an offset term for the decoding network;
Figure BDA0003578899550000026
wherein, the output f (i) is sample data high-dimensional features after passing through the coding network, and W is the coding networkWeight matrix of the network, biTo encode the corresponding bias terms of the network.
The invention is further improved by the method for evaluating the residual life state of the lithium battery based on the improved TCN:
the recursive prediction process in step S04 is: the feature data set defining the inputs and outputs is:
CR＝{(X1,Cap2),(x2,Cap3),…,(XL,CapL+1) } (formula 4)
Wherein XLRepresenting model input at time L, CapL+1Indicating the capacity output at the L +1 th time;
capacity Cap predicted at time LL+1Sequence of input capacities X added to the next instantL+1Continuing to predict the subsequent capacity of the model; the new feature data set obtained by recursion is:
CRrec＝{(XL+1,CapL+1),…,(Xcur-1,Capcur) } (formula 5)
Once the capacity value Cap obtained is predictedcur≤CapEOLStopping the prediction process; counting the number of residual cycle periods reaching the EOL capacity in the whole prediction process to obtain the residual service life of the battery;
CapEOL＝Caprated70% (formula 3)
Wherein, CapratedIs the rated capacity of the battery.
The invention is further improved by the evaluation method of the residual life state of the lithium battery based on the improved TCN, which comprises the following steps:
the expansion coefficient of the hybrid expansion convolution is designed to be [1,2,4,8,16,32,64 ];
the activation function A-ReLU is:
f (x) max { ax, x }, (0< a <1) (formula 8)
Wherein a is an input correction coefficient, and x is an input layer characteristic value of the model.
The invention is further improved by the evaluation method of the residual life state of the lithium battery based on the improved TCN, which comprises the following steps:
the training and testing process of the A-TCN-DAE model comprises the following steps:
randomly initializing weight parameters and bias parameters of a network model, wherein training parameter settings comprise a learning rate, a dropout rate, a convolution kernel size and an expansion factor; inputting the training set into an A-TCN-DAE model; inputting a mini-batch of the sequence, and calculating parameters of a hidden layer and output of an output layer according to forward propagation of input data; calculating a loss function, performing back propagation by taking the minimized loss function as a target, calculating a hidden layer error, solving a partial derivative, updating a weight parameter and a bias parameter in the network in a gradient descending manner, reducing the error between a predicted value and a true value through continuous iteration until a preset iteration number is reached, and storing model parameters so as to obtain a trained A-TCN-DAE model;
then, recursion is carried out on the test set and input into the trained A-TCN-DAE model, the subsequent capacity is predicted according to the recursion, and once the obtained capacity value Cap is predictedcurCapacity value Cap corresponding to preset battery EOLEOLThe following conditions are met: capcur≤CapEOLEnding the recursion process; by mean absolute error MAE, root mean square error RMSE, R2And counting and verifying the accuracy of model prediction on a test set for measuring indexes so as to obtain the A-TCN-DAE model which can be used online.
The invention is further improved by the method for evaluating the residual life state of the lithium battery based on the improved TCN:
the acquisition process of the training set and the test set comprises the following steps:
1) the four groups of same lithium batteries are charged to a full charge state by adopting a constant current and constant voltage CCCV mode, then the No. 1 lithium battery and the No. 2 lithium battery are discharged at a constant current of 0.5C, and the No. 3 lithium battery and the No. 4 lithium battery are discharged at a constant current of 1C until the battery voltage is respectively reduced to the minimum threshold voltage of 2.7V, and then the battery voltage is taken as a cycle period;
2) repeating the operation to obtain the actually measured voltage and current data of the lithium battery under the circulation working condition until the current capacity of the lithium battery is attenuated to an EOL state, and obtaining a voltage and current curve of the circulation working condition;
3) and supplementing missing points or abnormal points by adopting a linear interpolation method on the obtained actually measured voltage and current data of the lithium battery under the circulating working condition:
Vt＝(Vt-1+Vt+1) /2 (formula 1)
Wherein VtRepresenting the current missing or outlier, Vt-1Data representing the last point in time, Vt+1Data representing a next point in time;
4) and establishing cycle periods by taking the cut-off voltage 2.7V as a slicing cut-off point for all voltage data, and sequentially carrying out slicing labeling on each cycle period:
5) performing ampere-hour integration on the discharge process of each time in the cycle period from full charge to discharge of the battery to obtain the capacity of the battery in the current cycle period;
Capt＝Capt-1+ [ integral ] I (t) dt (formula 2)
Wherein Capt-1The capacity of a certain battery at the time t-1, CaptThe capacity of a certain battery at the time t is shown;
6) averaging the capacity of the battery in the cycle period according to the sampling time interval to obtain the actual capacity in each cycle period;
7) summarizing according to the mapping relation between the fragment labels and the actual capacity of each cycle period to obtain a cycle period and an actual capacity sequence, then taking the data of the cycle periods and the actual capacity sequences of No. 2 and No. 4 lithium batteries as a training set, and taking the data of the cycle periods and the actual capacity sequences of No. 1 and No. 3 lithium batteries as a test set.
The invention has the following beneficial effects:
1. the method adopts mixed expansion convolution on the basis of a Time Convolution Network (TCN), thereby effectively expanding the experience visual field of the convolution process and enhancing the capture capability of the model on the local recovery characteristics of the capacity sequence;
2. the method adopts the A-RELU activation function on the basis of the Time Convolution Network (TCN), so that certain characteristic capture capacity is enhanced, and the prediction accuracy and the fitting goodness of a model are improved;
3. the A-TCN-DAE model after the A-TCN model is combined with the DAE has the best true value following capability, errors and fluctuation are effectively reduced, the DAE is equivalent to further extracting data characteristics and removing noise influence in the original capacity sequence reconstruction process, so that the reconstructed data has the essential characteristics, robustness and generalization capability of the data, and the accuracy and robustness of RUL prediction on the residual service life of the battery are improved;
4. compared with the prior art, the model provided by the invention has stronger local feature capturing capability and adaptive capability of noise reduction and reconstruction of input data, and the estimation accuracy of the residual service life RUL of the battery is obviously improved compared with the prior deep learning network technology.
Drawings
The following describes embodiments of the present invention in further detail with reference to the accompanying drawings.
FIG. 1 is a schematic flow chart of a method for evaluating the remaining life state of a lithium battery based on an improved TCN according to the present invention;
FIG. 2 is a schematic diagram of a data slicing process;
FIG. 3 is a schematic diagram of the overall architecture of the underlying TCN model;
FIG. 4 is a schematic diagram of a network structure of a noise reduction auto-encoder and a noise reduction process;
FIG. 5 is a schematic diagram of the structure of the residual module after introducing hybrid dilation convolution, A-ReLU and residual concatenation;
FIG. 6 is a schematic diagram of the A-TCN-DAE model of the present invention.
Detailed Description
The invention will be further described with reference to specific examples, but the scope of protection of the invention is not limited thereto:
embodiment 1, a method for evaluating a remaining life state of a lithium battery based on an improved TCN, as shown in fig. 1 to 6, includes the following specific steps:
step 1, obtaining a cycle period and actual capacity sequence through cyclic constant current discharge
Step 1.1, performing offline charging and discharging on the lithium battery by using a lithium battery charging and discharging level table, and collecting actually-measured voltage and current of the lithium battery under a circulating working condition by using battery management system equipment.
1) Four groups of same lithium batteries are selected as test batteries, and the four groups of test batteries of the lithium battery charging and discharging platform are all charged in a Constant Current and Constant Voltage (CCCV) mode to reach a full charge state. The No. 1 lithium battery and the No. 2 lithium battery are discharged at a constant current of 0.5C, and the No. 3 lithium battery and the No. 4 lithium battery are discharged at a constant current of 1C until the battery voltage is respectively reduced to the minimum threshold voltage of 2.7V, and then the battery voltage is taken as a cycle period;
2) repeating the experiment according to the step 1) to obtain actually measured voltage and current data of the lithium battery under the circulating working condition, and obtaining a voltage and current curve of the circulating working condition; according to the requirements of the IOS for the cycle life test standard of the lithium battery, the lithium battery with the current capacity attenuated to 70% of the rated capacity is considered to reach the EOL state, and in the charging and discharging experiment platform, the cycle working condition experiment termination condition of the single lithium battery is judged according to the set EOL standard.
Step 1.2, preprocessing of raw data
1) In the sampling and storing process, dirty data can be inevitably brought due to the fluctuation of data interaction between the acquisition hardware equipment and the communication module. Dirty data in test data mainly comprises a data loss problem and an abnormal jumping problem of some sampling points, an abnormal point is removed firstly in a processing mode, and then a linear interpolation method is adopted to supplement the data loss problem caused in the sampling process to the lost point or the abnormal point:
Vt＝(Vt-1+Vt+1) /2 (formula 1)
Wherein VtRepresenting the current missing or outlier, Vt-1Data representing the last point in time, Vt+1Data representing a next point in time;
2) data slicing according to cycle working condition times
According to the characteristics of the cycle condition cut-off voltage, the cut-off voltage point of each cycle condition can be used as the end of the current cycle period and the start of the next cycle period. As shown in fig. 2, data slicing is performed on all voltage data with the cutoff voltage of 2.7V as a slice cutoff point, and slice indexes are performed on each data slice in sequence: each data fragment is a cycle period, and each cycle period is provided with a fragment label;
3) performing ampere-hour integration on the discharge process of each time in the cycle period from full charge to discharge of the battery to obtain the capacity of the battery in the current cycle period;
Capt＝Capt-1+ [ integral ] I (t) dt (formula 2)
Wherein Capt-1The capacity of a certain battery at the time t-1, CaptThe capacity of a certain battery at the time t is shown;
4) the capacity of the battery in each cycle period obtained from equation 2 may have partial dirty data, so that the capacity of the battery in the cycle period is averaged according to the sampling time interval to obtain the actual capacity in each cycle period;
5) summarizing according to the mapping relation between the fragment labels and the actual capacity of each cycle period, and obtaining a cycle period and actual capacity sequence as the input of the A-TCN-DAE model.
Step 2, recursion prediction process and stopping condition
When the actual capacity of the lithium battery is reduced to 70% of the rated capacity, the battery reaches an EOL state. The threshold of the EOL is set to ensure that the battery capacity is attenuated to a certain extent and then the battery system is considered to enter an unreliable state, and at this time, maintenance is required to be performed on the battery system. Capacity Cap when Battery reached EOL StateEOLCan be expressed as:
CapEOL＝Caprated70% (formula 3)
Wherein CapratedIs the rated capacity of the battery.
In order to design a prediction model reasonably and make the model more suitable for predicting the remaining service life (RUL) of a battery, a characteristic data set of input and output is defined as follows:
CR＝{(X1,Cap2),(X2,Cap3),…,(XL,CapL+1) } (formula 4)
Wherein, XLRepresenting model input at time L, CapL+1Indicating the capacity output at time L + 1. According to the causal theorem of time series data, the output of the model at the current moment is only related to the characteristics of the current moment and the previous data, and is not related to the future data.
Capacity Cap predicted at time LL+1Sequence of input volumes X added to the next time instantL+1And continuing to predict the subsequent capacity of the model. The new feature data set obtained by recursion is:
CRrec＝{(XL+1,CapL+1),…,(Xcur-1,Capcur) } (formula 5)
Once the capacity value Cap obtained is predictedcur≤CapEOLThe prediction process may be stopped. And counting the residual cycle number reaching the EOL capacity to obtain the residual service life (RUL) of the battery.
Step 3, constructing an A-TCN-DAE model
The A-TCN-DAE model comprises a noise reduction self-coding (DAE) network and an A-TCN model (an improved TCN model), and the cycle period and the actual capacity sequence output by the noise reduction self-coding (DAE) network after reconstruction optimization are subjected to the A-TCN model to obtain the prediction result of the residual battery life RUL. Aiming at the problem that the capturing capability of a common sequence prediction model on the local capacity recovery characteristic is insufficient, on the basis of a TCN model, the following improvements are carried out inside each residual error module RB (residual Block) to construct an A-TCN model: 1) replacing causal convolution with mixed dilation convolution, 2) replacing an activation function ReLU with A-ReLU, 3) adding a 1 × 1 convolution structure between the input and output of a residual module RB; the A-TCN model is combined with the extended feature extraction capability of a noise reduction self-coding network (DAE) at the same time to improve the performance and robustness of the model.
Step 3.1 constructing the underlying TCN model
The overall architecture of the TCN model is shown in fig. 3: the method is characterized by comprising 7 residual modules RB (residual Block) which are externally connected with a full connection layer with a sigmoid activation function. Residual errorExpansion factor of the block obeys 2nIncreasing law to enlarge the perceived field of view of the model, example: 26If the convolution kernel is set to 2, the model can sense the field of view of 128, and for the capacity and cycle period data, it indicates that the influence factor of the current output contains the capacity information of the last 128 inputs. The residual module internally comprises two layers of causal convolutions, and a ReLU activation function and a dropout layer with the inactivation rate of 0.2 are used after each layer of causal convolution. Finally, through the full connection layer and through the gradient descent learning, the capacity prediction result at the next moment is finally output, that is, the output of the last full connection layer can be expressed as:
Capk＝σ(wkhk+bk) (formula 6)
Where σ denotes a sigmoid activation function, wkAnd bkWeight matrix and offset, hkIs the hidden layer tensor result of the last layer of residual error layer.
Step 3.2, convolution substitution
The causal convolution of the residual block RB in the TCN model established in step 3.1 is replaced with a mixed dilation convolution. In the process of extracting the features through convolution operation, the receptive field is used for describing the number of bottom layer network data units related to the output judgment of a single unit in a top layer network. The larger the perception field of view is, the smaller the number of network layers required for extracting all the features of the data features with the same input size is equivalent to the reduction of network parameters, so that the processing speed of the network can be improved, and the more historical data features can be contained. The expansion convolution is adopted to enlarge the receptive field, and the characteristics are as follows:
1) all the convolved expansion coefficients cannot have common divisor other than 1, for example, 2,4,6 is unreasonable expansion coefficient, and the grid effect is caused;
2) designing the expansion coefficient to be saw-toothed, such as: d ═ 1,2,4], used for extracting the characteristic information of different intervals;
3) and setting the maximum interval between non-zero eigenvalues in the convolution kernel to meet the following requirements:
Mi＝max{Mi+1-2ri,Mi+1-2(Mi+1-ri),ri} (formula 7)
Wherein r isiDenotes the coefficient of expansion, M, of the i-th layeriIndicates the maximum selectable expansion coefficient at the ith layer, and if n layers are provided, the default r isn＝MnCombining with capacity sequence characteristics, all historical capacity information needs to be correlated in order to capture local capacity recovery characteristics, and if the size of the selected convolution kernel is k × 1, the target is MkK, so that at least one layer of standard convolution with a coefficient of expansion of 1, designed as [1,2,4,8,16,32,64, can be used to ensure that no historical capacity characteristics are missing]；
Step 3.3, A-ReLU replaces original ReLU
The ReLU activation function feature allows negative output values to be filtered, taking into account only neuron features whose output is positive. The characteristic reduces a certain amount of calculation, so that the training and convergence of the model can be accelerated. However, the filtered neurons may cause partial data feature loss. For the capacity fading characteristic, although the overall characteristic is in a decreasing trend, due to the existence of the capacity recovery characteristic, a negative value may appear at a local characteristic point in the training process, the fitting capability of the model to the vicinity of the 0 value needs to be enhanced and the model cannot be directly set to 0, so that the activation function a-ReLU with parameters is introduced to replace the activation function ReLU of the residual module RB in the TCN model established in step 3.1, and the characteristic is defined as:
(x) max { ax, x }, (0< a <1) (equation 8)
Wherein a is an input correction coefficient, and x is an input layer characteristic value of the model.
The parameter a is added to the training process as a learnable variable in the network, although a little model calculation amount is added, the sensitivity of the model to the value near 0 is changed, and the problem that the characteristics are lost when the input amount is less than 0 is solved.
Step 3.4, residual error connection is introduced:
by introducing residual concatenation, the upper layer convolution can contain detailed local feature information and the feature information of the original input data is not lost.
In order to enable a network model of an L + n layer to have at least the capability of an L layer network and not to deepen the network but have worse effect due to network degradation and the like, a 1 × 1 convolution structure is added between the input and the output of a residual module RB in each residual module RB, and each residual module RB calculates respective residual and then transmits the residual layer by layer backwards, so that residual connection is formed between the residual modules RB, and all residual information is kept in the training process of the network, as shown in FIG. 5. Assuming that x is the input of the model, and f (x) is the process of characterizing the linear transformation and activation function processing, before the process of activating the second layer of linear processing, the result of f (x) is merged with the bottom layer input x, and the final output is expressed as:
o＝F2(x+F1(x) (formula 9)
After residual connection is introduced, if the feature extracted by hierarchical discrete correlation HDC can improve generalization capability, namely more detailed local features can be obtained on the basis of input data features, parameters can be updated normally, and if effective feature information is not learned, the network cannot be degraded.
Step 3.5, introducing a noise reduction automatic encoder DAE
The noise reduction process of the noise reduction automatic encoder DAE comprises mixing noise v into input x, referring to Dropout operation in deep learning by introducing the noise v, randomly setting data in a certain input sequence to be 0, and then utilizing input samples containing noise
Figure BDA0003578899550000091
Data features are extracted through an encoding network E (-) and finally restored into an original data structure through a decoding network D (-) and reconstructed and optimized, and output y is a cycle period and an actual capacity sequence after reconstruction and optimization, as shown in FIG. 4.
Model input X for time L defined in step 2LObtained by introducing noise v in the DAE process of a noise reduction automatic encoder
Figure BDA0003578899550000092
The process flow of the coding network can be expressed as:
Figure BDA0003578899550000093
wherein, the output f (i) is sample data high-dimensional features after passing through the coding network, W is a weight matrix of the coding network, biTo encode the corresponding bias terms for the network.
The decoding process can be expressed as:
Figure BDA0003578899550000094
wherein y and
Figure BDA0003578899550000095
namely the output of the reconstruction optimization is obtained,
Figure BDA0003578899550000096
to decode the bias term of the network. And reconstructing the optimized y as the input of the subsequent A-TCN.
Step 4, training the A-TCN-DAE model to predict the remaining service life of the battery
Step 4.1, constructing a training set and a testing set of the A-TCN-DAE model
And (2) testing four groups of lithium batteries respectively according to the method in the step 1 to obtain relation data of four groups of cycle periods and actual capacity sequences, then grouping the four groups of data according to a discharge working condition, taking the cycle periods and the actual capacity sequences of No. 2 and No. 4 lithium batteries as a training set, and taking the cycle periods and the actual capacity sequence data of No. 1 and No. 3 lithium batteries as a test set. The input of the A-TCN-DAE model is a cycle period and actual capacity sequence, and the output is predicted capacity.
Step 4.2, constructing a loss function
Defining a prediction value yiAnd true value
Figure BDA0003578899550000097
The error between is a loss function:
Figure BDA0003578899550000101
step 4.3, training and testing A-TCN-DAE model
Step 4.3.1, training Process of A-TCN-DAE model
(1) And training parameter setting: the expansion factor d of the expansion convolution is set to [1,2,4,8,16,32,64], the convolution kernel size is 3 x 1, the deactivation rate of the deep learning network is 0.2, and the Adam optimizer is used for optimization during training, the learning rate is dynamically adjusted, the initial learning rate is set to 0.005, and the first-order momentum parameter and the second-order momentum parameter are respectively set to 0.5 and 0.9. The number of training iterations was 80.
(2) The weight and bias parameters of the random initialization network model are as follows: the initial weight is a random number subject to a Gaussian distribution with a mean value of 0 and a standard deviation of 0.01, and the initial bias is all set to 0;
(3) inputting the cycle period and the actual capacity sequence in the training set into an A-TCN-DAE model; inputting a mini-batch of the sequence, and calculating parameters of a hidden layer and output of an output layer according to forward propagation of input data; calculating a loss function, performing back propagation by taking a minimized loss function (formula 12) as a target, calculating a hidden layer error, solving a partial derivative, updating a weight parameter and a bias parameter in the network in a gradient descending manner, reducing an error between a predicted value and a true value through continuous iteration until a preset iteration number is reached, and storing a model parameter so as to obtain a trained A-TCN-DAE model;
step 4.3.2, the test process of the A-TCN-DAE model:
recursion inputting the current cycle period and the actual capacity sequence in the test set into the trained A-TCN-DAE model, recursion predicting the subsequent capacity according to the formula 5 in the step 2, and once predicting the obtained capacity value CapcurCapacity value Cap corresponding to preset Battery EOLEOLMeeting the conditions: capcur≤CapEOLThen the recursion process is ended. Obtaining a capacity attenuation curve according to the recursion result, and counting the residual circulation reaching the EOL capacityThe number of cycles is the Remaining Useful Life (RUL) of the battery; and (4) counting and verifying the accuracy of model prediction by using MAE, RMSE and R2 as measurement indexes on a test set, thereby obtaining the A-TCN-DAE model which can be used online.
Step 5, on-line application process of A-TCN-DAE model
1) The lithium battery to be tested is charged in a constant-current constant-voltage charging mode to reach a full-charge state by using a lithium battery charging and discharging platform, and then is discharged by constant currents of 0.5C and 1C respectively until the voltage of the battery is reduced to a threshold value of 2.7V; repeating the operation for 5 times to obtain the actually measured voltage and current data of the lithium battery under 5 circulation working conditions to obtain voltage and current curves under the circulation working conditions;
2) segmenting the actually measured voltage and current data of the lithium battery under the circulating working condition, wherein the method comprises the steps of taking the cut-off voltage of 2.7V as a segment cut-off point of all voltage data as the end of the current circulating period and the beginning of the next circulating period, and sequentially carrying out segment marking on each circulating period;
3) adopting an ampere-hour integration method for the discharge process in each cycle period and carrying out averaging treatment to obtain the actual capacity; summarizing according to the mapping relation between the fragment labels and the actual capacity to obtain a cycle period and an actual capacity sequence;
4) predicting the battery capacity data Cap at the current moment by taking the cycle period and the actual capacity sequence obtained in the step 3) as the input of an A-TCN-DAE modelcurThen, the subsequent capacity of the battery is predicted according to the formula 5 in the step 2, and the capacity value Cap obtained once prediction is carried outcurCapacity value Cap corresponding to preset Battery EOLEOLThe following conditions are met: capcur≤CapEOLAnd ending the recursion process, and then counting the number of the residual cycle number reaching the EOL capacity to obtain the current residual life of the battery.
Experiment 1:
the experimental hardware was configured as: the processor is 12 × E5-2609v3@1.9 GHz; the memory is 16 GB; the display card is intel Tesla P48 GB (384.81); the operating system is CentOS7.3.1611; model construction and training used the python3.7+ tensoflow1.4 version.
The experimental data set uses the training set and the test set constructed in the step 4.1 in the example 1, and the experiment adopts four comparison models and the A-TCN-DAE model of the invention to carry out comparison experiments so as to verify the effectiveness of the A-TCN-DAE model of the invention on the prediction of the residual life RUL of the battery, wherein the four comparison networks are respectively as follows:
1) before the expansion convolution replacement: a basic TCN network was constructed as in step 3.1 of example 1.
2) Before the activation function is improved: the causal convolution of the TCN network is replaced with a mixed dilation convolution as in step 3.2 of example 1.
3) Residual concatenation is not used: the causal convolution of the TCN network is replaced with a mixed dilation convolution as in steps 3.2 and 3.3 of example 1 and the activation function ReLU is replaced with a-ReLU.
4) A-TCN: the causal convolution of the TCN network is replaced with the hybrid dilated convolution as in steps 3.2, 3.3 and 3.4 of example 1 and the activation function ReLU is replaced with a-ReLU, and then residual concatenation is introduced.
The experiment is carried out according to the average absolute error MAE, the mean square error MSE, the root mean square error RMSE and R2As a measure, the following is specifically defined:
mean absolute error MAE: the index is an expected value of absolute error loss, and is an average value of the whole sample number obtained after the absolute value of the difference between the predicted value and the true value of the sample is summed, so that the situation that the error is offset positively and negatively can be effectively avoided, and the average absolute error of N samples can be represented by the following formula:
Figure BDA0003578899550000111
Figure BDA0003578899550000112
is the true value of the sample, yiIs a predicted value of the model.
Mean square error MSE: the index is an expectation for square error, the value of the index is inconsistent with the dimension of a target variable, the average value of the whole number of samples is obtained after the square sum of the difference between the predicted value and the true value of the samples is based on, and the mean square error of N samples can be represented by the following formula:
Figure BDA0003578899550000121
Figure BDA0003578899550000122
is the true value of the sample, yiIs the predicted value of the model.
Root mean square error RMSE: is the square root of the mean square error of the model's predicted value versus the true value. The method is characterized by being very sensitive to showing of extra-large or extra-small errors in a group of error data and being capable of well describing the precision degree of a model evaluation result. The root mean square error of the N samples can be expressed by the following equation:
Figure BDA0003578899550000123
r square: also known as the coefficient of determinism or goodness of fit, reflects the degree of fit between the predicted value and the true value, the closer to 1, the better the model fits, and the R-squared of N samples can be represented by the following formula:
Figure BDA0003578899550000124
Figure BDA0003578899550000125
is the true value of the sample, yiIs the predicted value of the model, y0Is the mean of the real samples.
Respectively training 5 network models by using the training set constructed in the step 4.1 to obtain the trained network models, then respectively testing and counting average absolute errors MAE, RMSE and R by using the test set constructed in the step 4.12As shown in table 1 below
Table 1, comparative results of various measurement indexes of experiment 1
Model (model) MSE MAE RMSE R2
Before expansion convolution replacement 0.015443 0.092317 0.124270 0.674913
Before improvement of activation function 0.009657 0.064217 0.098272 0.796704
Without using residual concatenation 0.008588 0.057751 0.092670 0.819220
A-TCN 0.002470 0.033291 0.049695 0.9480133
A-TCN-DAE 0.000265 0.010763 0.016292 0.9944123
According to the result of the experiment 1, the introduction of the expansion convolution effectively expands the experience visual field of the convolution process and enhances the capture capability of the model on the local recovery characteristics of the volume sequence; after the activation function is improved, certain characteristic capture capacity is enhanced, and prediction accuracy and model goodness of fit are improved; the network prediction accuracy can be further improved by introducing residual connection, but the optimized A-TCN still has larger fluctuation between the optimized A-TCN and a true value and has defects in robustness, and an A-TCN-DAE model combined with a noise reduction automatic encoder DAE has the best true value following capability and effectively reduces errors and fluctuation, because the DAE is equivalent to further extracting data characteristics and removing noise influence in the original capacity sequence reconstruction process, the reconstructed data has essential characteristics of data, robustness and generalization capability.
Experiment 2:
selecting a representative network model for predicting the residual life of the battery, wherein the representative network model specifically comprises a Recurrent Neural Network (RNN)[1]Long and short term memory network LSTM[2]Gate controlled circulation unit network GRU[3]In order to laterally evaluate the accuracy and robustness of the model, the training sets constructed in the step 4.1 are used for training 4 network models respectively to obtain the trained network models, and then the test sets constructed in the step 4.1 are used for testing and counting the mean absolute errors MSE, MAE, RMSE and R respectively2As shown in table 2 below.
Table 2, comparative results of various measurement indexes of experiment 2
Model (model) MSE MAE RMSE R2
RNN 0.063097 0.161502 0.179966 0.326071
LSTM 0.032388 0.114287 0.118471 0.654070
GRU 0.005333 0.044645 0.073028 0.850089
A-TCN-DAE 0.000388 0.010185 0.019706 0.995852
Through quantitative analysis of the evaluation indexes, the A-TCN-DAE model has advantages on each evaluation index.
Reference documents:
[1]Yu W,Kim I Y,Mechefske C.An improved similarity-based prognostic algorithm for RUL estimation using an RNN autoencoder scheme[J].Reliability Engineering System Safety,2020,199:106926.
[2]Sayah M,Guebli D,Noureddine Z,et al.Deep LSTM Enhancement for RUL Prediction Using Gaussian Mixture Models[J].Automatic Control and Computer Sciences,2021,55(1):15-25.
[3]Lu Y W,Hsu C Y,Huang K C.An Autoencoder Gated Recurrent Unit for Remaining Useful Life Prediction.2020.
finally, it is also noted that the above-mentioned lists merely illustrate a few specific embodiments of the invention. It is obvious that the invention is not limited to the above embodiments, but that many variations are possible. All modifications which can be derived or suggested by a person skilled in the art from the disclosure of the present invention are to be considered within the scope of the invention.

Claims (7)

1. A lithium battery residual life state evaluation method based on improved TCN is characterized by comprising the following specific processes:
step S01, the lithium battery to be tested is charged to full charge by using a constant current and constant voltage charging mode, and then is discharged by using constant currents of 0.5C and 1C respectively until the voltage of the battery is reduced to 2.7V of a threshold value; repeating the operation for 5 times to obtain the actually measured voltage and current data of the lithium battery under 5 circulation working conditions to obtain voltage and current curves under the circulation working conditions;
step S02, taking the cut-off voltage 2.7V of the voltage data under the circulation working condition as a slice cut-off point, performing data slicing as the end of the current circulation period and the start of the next circulation period, and performing slice labeling on each data slice according to the sequence, wherein one data slice is a circulation period;
step S03, adopting an ampere-hour integration method for the discharging process in each cycle period and carrying out averaging treatment to obtain the actual capacity; summarizing according to the mapping relation between the fragment labels and the actual capacity to obtain a cycle period and an actual capacity sequence;
step S04, taking the cycle period and the actual capacity sequence as the input of an A-TCN-DAE model, wherein the A-TCN-DAE model comprises a noise reduction self-coding DAE network and an A-TCN model, and the cycle period and the actual capacity sequence output by the noise reduction self-coding DAE network after reconstruction optimization predict the battery capacity data Cap at the current moment through the A-TCN modelcurThen, the subsequent capacity of the battery is predicted in a recursion mode, and the capacity value Cap obtained once prediction is carried outcurPresetting a capacity value Cap corresponding to a battery EOLEOLThe following conditions are met: capcur≤CapEOLAnd ending the recursion prediction process, and then counting the number of the residual cycle periods reaching the EOL capacity to obtain the current residual life of the battery.
2. The improved TCN-based lithium battery remaining life state evaluation method as claimed in claim 1, wherein:
the A-TCN model is based on the TCN model, and each residual error module RB is improved as follows: 1) replacing causal convolution with hybrid dilation convolution, 2) replacing the activation function ReLU with a-ReLU, 3) adding a 1 × 1 convolution structure between the residual module RB input and output; and each residual error module RB calculates respective residual error and then transmits the residual errors layer by layer backwards.
3. The improved TCN-based lithium battery remaining life state evaluation method as claimed in claim 2, wherein:
the denoised auto-encoder DAE comprises mixing noise v into the input x and then taking the input samples
Figure FDA0003578899540000011
Inputting samples
Figure FDA0003578899540000012
Extracting data characteristics through an encoding network E (-) and reconstructing, optimizing and outputting a cycle period and an actual capacity sequence y after reconstruction and optimization through a decoding network D (-) after:
Figure FDA0003578899540000013
wherein,
Figure FDA0003578899540000014
in order to reconstruct the output that is optimized,
Figure FDA0003578899540000015
an offset term for the decoding network;
Figure FDA0003578899540000016
wherein, the output f (i) is sample data high-dimensional features after passing through the coding network, W is a weight matrix of the coding network, biTo encode the corresponding bias terms of the network.
4. The improved TCN-based lithium battery remaining life state evaluation method as claimed in claim 3, wherein:
the recursive prediction process in step S04 is: the feature data set defining the inputs and outputs is:
CR＝{(X1，Cap2)，(X2，Cap3)，…，(XL，CapL+1) } (formula 4)
Wherein, XLRepresenting model input at time L, CapL+1Indicating the capacity output at the L +1 th time;
capacity Cap predicted at time LL+1Sequence of input volumes X added to the next time instantL+1Continuing to predict the subsequent capacity of the model; new characteristic number obtained by recursionThe data set is as follows:
CRrec＝{(XL+1，CapL+1)，…，(Xcur-1，Capcur) } (formula 5)
Once the obtained capacity value Cap is predictedcur≤CapEOLStopping the prediction process; counting the number of residual cycle periods reaching the EOL capacity in the whole prediction process to obtain the residual service life of the battery;
CapEOL＝Caprated70% (formula 3)
Wherein, CapratedIs the rated capacity of the battery.
5. The improved TCN-based lithium battery remaining life state evaluation method as claimed in claim 4, wherein:
the expansion coefficient of the hybrid expansion convolution is designed to be [1,2,4,8,16,32,64 ];
the activation function A-ReLU is:
(x) max { ax, x }, (0< a <1) (equation 8)
Wherein a is an input correction coefficient, and x is an input layer characteristic value of the model.
6. The improved TCN-based lithium battery remaining life state evaluation method as claimed in claim 5, wherein:
the training and testing process of the A-TCN-DAE model comprises the following steps:
randomly initializing weight parameters and bias parameters of a network model, wherein training parameter settings comprise a learning rate, a dropout rate, a convolution kernel size and an expansion factor; inputting the training set into an A-TCN-DAE model; inputting a mini-batch of the sequence, and calculating parameters of a hidden layer and output of an output layer according to forward propagation of input data; calculating a loss function, performing back propagation by taking the minimized loss function as a target, calculating a hidden layer error, solving a partial derivative, updating a weight parameter and a bias parameter in the network in a gradient descending manner, reducing the error between a predicted value and a true value through continuous iteration until a preset iteration number is reached, and storing model parameters so as to obtain a trained A-TCN-DAE model;
then, recursion is carried out on the test set and input into the trained A-TCN-DAE model, the subsequent capacity is predicted according to the recursion, and once the obtained capacity value Cap is predictedcurCapacity value Cap corresponding to preset Battery EOLEOLThe following conditions are met: capcur≤CapEOLEnding the recursion process; by mean absolute error MAE, root mean square error RMSE, R2And counting and verifying the accuracy of model prediction on a test set for measuring indexes so as to obtain the A-TCN-DAE model which can be used online.
7. The improved TCN-based lithium battery remaining life state evaluation method as claimed in claim 6, wherein:
the acquisition process of the training set and the test set comprises the following steps:
1) the four groups of same lithium batteries are charged to a full charge state by adopting a constant current and constant voltage CCCV mode, then the No. 1 lithium battery and the No. 2 lithium battery are discharged at a constant current of 0.5C, and the No. 3 lithium battery and the No. 4 lithium battery are discharged at a constant current of 1C until the battery voltage is respectively reduced to the minimum threshold voltage of 2.7V, and then the battery voltage is taken as a cycle period;
2) repeating the operation to obtain the actually measured voltage and current data of the lithium battery under the circulation working condition until the current capacity of the lithium battery is attenuated to an EOL state, and obtaining a voltage and current curve of the circulation working condition;
3) and supplementing the missing points or abnormal points by adopting a linear interpolation method to the actually measured voltage and current data of the lithium battery under the acquired cycle working condition:
Vt＝(Vt-1+Vt+1) /2 (formula 1)
Wherein VtRepresenting the current missing or outlier, Vt-1Data representing the last point in time, Vt+1Data representing a next point in time;
4) establishing cycle periods by taking the cut-off voltage 2.7V as a fragment cut-off point for all the voltage data, and carrying out fragment labeling on each cycle period according to the sequence:
5) performing ampere-hour integration on the discharge process of each time in the cycle period from full charge to discharge of the battery to obtain the capacity of the battery in the current cycle period;
Capt＝Capt-1+ [ integral ] I (t) dt (formula 2)
Wherein Capt-1The capacity, Cap, of a certain cell at time t-1 is showntThe capacity of a certain battery at the time t is shown;
6) averaging the capacity of the battery in the cycle period according to the sampling time interval to obtain the actual capacity in each cycle period;
7) and summarizing according to the mapping relation between the fragment labels and the actual capacity of each cycle period to obtain a cycle period and an actual capacity sequence, then taking the data of the cycle periods and the actual capacity sequences of the No. 2 and No. 4 lithium batteries as a training set, and taking the data of the cycle periods and the actual capacity sequences of the No. 1 and No. 3 lithium batteries as a test set.
CN202210349477.5A 2022-04-01 2022-04-01 Lithium battery residual life state evaluation method based on improved TCN Pending CN114740388A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210349477.5A CN114740388A (en) 2022-04-01 2022-04-01 Lithium battery residual life state evaluation method based on improved TCN

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210349477.5A CN114740388A (en) 2022-04-01 2022-04-01 Lithium battery residual life state evaluation method based on improved TCN

Publications (1)

Publication Number Publication Date
CN114740388A true CN114740388A (en) 2022-07-12

Family

ID=82280165

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210349477.5A Pending CN114740388A (en) 2022-04-01 2022-04-01 Lithium battery residual life state evaluation method based on improved TCN

Country Status (1)

Country Link
CN (1) CN114740388A (en)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115129706A (en) * 2022-09-02 2022-09-30 中国气象局气象探测中心 Soil moisture observation data quality evaluation method considering periodic characteristics
CN115640706A (en) * 2022-12-22 2023-01-24 清陶(昆山)能源发展股份有限公司 Reliability test method and device for lithium battery and computer readable storage medium
CN116257972A (en) * 2022-11-29 2023-06-13 元始智能科技(南通)有限公司 Equipment state evaluation method and system based on field self-adaption and federal learning
CN116609668A (en) * 2023-04-26 2023-08-18 淮阴工学院 Lithium ion battery health state and residual life prediction method
CN117574762A (en) * 2023-11-17 2024-02-20 暨南大学 Construction method of digital twin model of power battery
CN118655473A (en) * 2024-08-21 2024-09-17 中创新航技术研究院(江苏)有限公司 Fault battery screening method, device, electronic equipment, medium and program product
CN118655473B (en) * 2024-08-21 2024-10-18 中创新航技术研究院(江苏)有限公司 Fault battery screening method, device, electronic equipment, medium and program product

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115129706A (en) * 2022-09-02 2022-09-30 中国气象局气象探测中心 Soil moisture observation data quality evaluation method considering periodic characteristics
CN116257972A (en) * 2022-11-29 2023-06-13 元始智能科技(南通)有限公司 Equipment state evaluation method and system based on field self-adaption and federal learning
CN116257972B (en) * 2022-11-29 2024-02-20 元始智能科技(南通)有限公司 Equipment state evaluation method and system based on field self-adaption and federal learning
CN115640706A (en) * 2022-12-22 2023-01-24 清陶(昆山)能源发展股份有限公司 Reliability test method and device for lithium battery and computer readable storage medium
CN116609668A (en) * 2023-04-26 2023-08-18 淮阴工学院 Lithium ion battery health state and residual life prediction method
CN116609668B (en) * 2023-04-26 2024-03-26 淮阴工学院 Lithium ion battery health state and residual life prediction method
CN117574762A (en) * 2023-11-17 2024-02-20 暨南大学 Construction method of digital twin model of power battery
CN118655473A (en) * 2024-08-21 2024-09-17 中创新航技术研究院(江苏)有限公司 Fault battery screening method, device, electronic equipment, medium and program product
CN118655473B (en) * 2024-08-21 2024-10-18 中创新航技术研究院(江苏)有限公司 Fault battery screening method, device, electronic equipment, medium and program product

Similar Documents

Publication Publication Date Title
CN114740388A (en) Lithium battery residual life state evaluation method based on improved TCN
CN112241608B (en) Lithium battery life prediction method based on LSTM network and transfer learning
CN111220921A (en) Lithium battery capacity estimation method based on improved convolution-long-and-short-term memory neural network
CN112904219B (en) Big data-based power battery health state prediction method
CN112990556A (en) User power consumption prediction method based on Prophet-LSTM model
CN113406521B (en) Lithium battery health state online estimation method based on feature analysis
CN112966432A (en) Method and device for predicting remaining effective life of lithium ion battery
CN113109715B (en) Battery health condition prediction method based on feature selection and support vector regression
CN110443377B (en) Support vector machine storage battery life prediction method based on immune algorithm optimization
CN112230154A (en) Lithium battery residual life prediction method
CN113376540B (en) LSTM battery health state estimation method based on evolutionary attention mechanism
CN112611976A (en) Power battery state of health estimation method based on double differential curves
CN116298935B (en) Lithium ion battery health state estimation method based on countermeasure encoder network
CN116359742B (en) Energy storage battery state of charge on-line estimation method and system based on deep learning combination extended Kalman filtering
CN112803398A (en) Load prediction method and system based on empirical mode decomposition and deep neural network
CN112379274A (en) Method for predicting residual life of power battery
CN116298936A (en) Intelligent lithium ion battery health state prediction method in incomplete voltage range
CN116736133A (en) Early prediction method for capacity degradation track of lithium ion battery in full life cycle
Zhang et al. Remaining useful life prediction of lithium-ion batteries based on TCN-DCN fusion model combined with IRRS filtering
CN114814707A (en) Intelligent ammeter stress error analysis method, equipment, terminal and readable medium
Li et al. State of health estimation and prediction of electric vehicle power battery based on operational vehicle data
CN117076866A (en) Lithium battery RUL prediction method, system, equipment and medium based on data driving and particle filtering fusion model
Yu et al. SOH estimation method for lithium-ion battery based on discharge characteristics
CN116736130A (en) Lithium battery residual service life prediction method and system
CN116186484A (en) Deep learning-based battery capacity attenuation trend prediction method and system

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