CN116106773A - Mixed model lithium battery health state monitoring method based on data enhancement - Google Patents

Mixed model lithium battery health state monitoring method based on data enhancement Download PDF

Info

Publication number
CN116106773A
CN116106773A CN202211612664.4A CN202211612664A CN116106773A CN 116106773 A CN116106773 A CN 116106773A CN 202211612664 A CN202211612664 A CN 202211612664A CN 116106773 A CN116106773 A CN 116106773A
Authority
CN
China
Prior art keywords
data
model
lithium battery
representing
monitoring
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
CN202211612664.4A
Other languages
Chinese (zh)
Inventor
李鹏华
李皃
单康恒
余江
高代林
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.)
Chongqing University of Post and Telecommunications
Original Assignee
Chongqing University of Post and Telecommunications
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 Chongqing University of Post and Telecommunications filed Critical Chongqing University of Post and Telecommunications
Priority to CN202211612664.4A priority Critical patent/CN116106773A/en
Publication of CN116106773A publication Critical patent/CN116106773A/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
    • 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
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • 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
    • 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

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

Abstract

The invention relates to a hybrid model lithium battery health state monitoring method based on data enhancement, and belongs to the technical field of battery detection. The method takes a multi-head attention mechanism and a time sequence convolution mixed model as a basic model, integrates a generated countermeasure method into a mixed neural network model, designs a countermeasure model generated aiming at time sequence data, adds supervised learning of an embedding and generating module on the basis of a traditional unsupervised generated countermeasure module, and the serial-parallel connection of the mixed neural network breaks through unsupervised characteristic of the generated countermeasure learning and the limitation of single model on data characteristic extraction in the traditional neural network method, compensates for information loss caused by insufficient training sets, and realizes high-precision lithium battery health condition monitoring based on a limited data set. According to the invention, under the condition of a small amount of battery data under abnormal working conditions, enough data can be provided in a data enhancement mode, so that high-precision battery health state prediction is realized.

Description

Mixed model lithium battery health state monitoring method based on data enhancement
Technical Field
The invention belongs to the technical field of lithium ion battery detection, and relates to a hybrid model lithium battery health state monitoring method based on data enhancement.
Background
Along with the gradual increase of the demand of new energy automobile markets, the key technology of the electric automobile is urgently required to break through. As a power source of the electric vehicle, the development of the power lithium ion battery determines the future of the electric vehicle, and in order to ensure the safety and reliability of the battery, a battery management system (Battery Management System, BMS) is used as an important component of the electric vehicle and performs functions of data acquisition, balanced management, state estimation and the like. Among them, state of Health (SOH) is one of the most important indexes of the BMS, and it is important to accurately diagnose the SOH of the battery.
In the SOH diagnosis technology, a model-based method, such as an equivalent circuit model, an electrochemical model, a mathematical model, etc., is mostly adopted, and besides, the neural network model is also widely applied to monitoring the health state of the lithium ion battery. The neural network model realizes direct mapping from input to output by learning a nonlinear relation between data. Some conventional neural networks, such as Artificial Neural Networks (ANNs), probabilistic Neural Networks (PNNs), convolutional Neural Networks (CNNs), a priori-based neural networks, gated neural networks (GRUs), long-short term neural networks (LSTM), etc., have been widely used for SOH monitoring of batteries, and, among many hybrid models, neural networks are the most widely used single model in combination with other models. The method comprises the steps that a convolutional neural network (GRU-CNN) based hybrid network is utilized to expand CNN by utilizing GRU-RNN sub-modules, a convolutional block of CNN utilizes a shared weight structure to reduce the number of weights, and an attempt is made to find shared information from actually measured charging voltage, current and temperature data; on the other hand, GRU-RNN blocks use their internal states to learn characteristics and time dependencies from sequential data. In fact, the combined structure can simultaneously utilize the advantages of CNN and GRU-RNN networks to capture the shared space-time characteristics of charging data. Similarly, an integration algorithm composed of ELM and LSTM may capture the underlying correspondence between SOHs. There are also many studies demonstrating that it is equally feasible to combine the advantages of CNN and LSTM for lithium battery state of health prediction.
However, most of the data measured at present are battery data from normal conditions, while a small portion of battery data from abnormal conditions, such as battery swelling, runaway, burning, explosion, etc., in extreme cases are quite rare. Repeated battery-related experiments to collect these data require expensive equipment and a significant amount of time, and furthermore, the lack of diversity in the data in the public dataset also limits research in this area. Therefore, how to process a small amount of lithium battery data under abnormal working conditions, and to improve the model prediction accuracy are urgent problems in monitoring the health state of the lithium battery.
Disclosure of Invention
Therefore, the invention aims to provide a lithium battery health state monitoring method based on a data enhancement hybrid model, which takes a multi-head attention mechanism and a time sequence convolution hybrid model as basic models, combines an anti-countermeasures method with an unsupervised learning and a supervised learning method, compensates for information loss caused by the shortage of a training set, and realizes high-precision lithium battery health state monitoring based on a limited data set.
In order to achieve the above purpose, the present invention provides the following technical solutions:
a method for monitoring the health state of a lithium battery based on a data enhancement hybrid model takes a hybrid neural network model based on a multi-head attention mechanism and a time sequence convolutional neural network as a basic model, simultaneously fuses a time sequence generation countermeasure method, embeds a supervised learning method, and combines a regression model to realize the prediction of the health state of the lithium battery, and the method comprises the following implementation steps:
s1, acquiring a training data set and randomly generating a noise data set;
s2, constructing a data enhancement model by using a time sequence generation countermeasure network as a data enhancement model basis and adopting a training data set and a noise data set to generate an enhancement data set;
s3, respectively extracting time effective features and space effective features of battery data through a hybrid neural network model;
s4, integrating the time effective features and the space effective features, and inputting the integrated feature vectors into a regression model to predict the health state of the lithium battery.
Further, in step S1, the training data set is represented as
Figure BDA0004000695500000021
The noise dataset is represented as
Figure BDA0004000695500000022
Where T represents the sequence length, N e {1,..n } represents the single sample index, and N represents the number of samples.
Further, in step S2, the data enhancement model includes a generator G for generating effective data approximated to the real data and a discriminator D for classifying the real data and the effective data.
Further, the data enhancement model construction process is as follows: training a generator G by using the training data set and the noise data set, and obtaining a generator G of a minimum maximum value by means of countermeasure evaluation; meanwhile, the discriminator D is optimized through JS divergence, and the classification accuracy of the discriminator D is maximized; in the construction process of the data enhancement model, the discriminator D maximizes the classification accuracy of the training data and the effective data, the generator G minimizes the maximum value, and when the maximization of the discriminator D and the minimization of the generator G reach Nash equilibrium, the construction of the data enhancement model is completed.
Further, in step S3, extracting time-efficient features of lithium battery data using an encoder with a multi-head attention mechanism;
input a of the encoder (l) The method comprises the following steps:
A (l) ＝PE (l) +X (l)
wherein X is (l) Input samples representing a transducer, PE (l) Representing a position encoding of the input samples, wherein the position encoding is of the formula:
Figure BDA0004000695500000031
where pos represents the position of the current element in the whole vector, i represents the index of each element in the vector, and d represents the dimension of the vector;
output B of encoder (l) The method comprises the following steps:
Figure BDA0004000695500000032
wherein h represents the number of multi-head attention mechanisms, C is a splicing function, W q 、W k And W is v All represent a weight matrix, W i Q 、W i K And W is i V All represent a multi-headed mapping parameter matrix, W O Representing the fusion matrix.
Further, in step S3, 1d-TCN is adopted to extract space effective characteristics of lithium battery data;
for the nth convolution layer, its input X n And output Y n Expressed as:
Figure BDA0004000695500000033
/>
Figure BDA0004000695500000034
wherein L and W represent the length and width of the input data, respectively; the correspondence between inputs and outputs is shown as follows:
Figure BDA0004000695500000035
where, σ represents the activation function,
Figure BDA0004000695500000036
represents the offset of the convolution kernel, d represents the step size of the convolution kernel, s v 、c v Position coordinates, k, representing the extracted features of the convolution kernel v Representing a convolution kernel number; y is set to n Inputting the maximum pooling layer to obtain a pooling vector:
Figure BDA0004000695500000037
Figure BDA0004000695500000038
wherein P is n Representing the pooling vector, x l 、y w Representing the position coordinates of the pooling output, d representing the pooling step size.
Further, the length of the input data is the length of the sample, and the width of the input data is the lithium battery data sample type.
Further, in step S4, a regression model based on a weight distribution mechanism is used to predict the health status of the lithium battery;
for the feature vector obtained by integration
Figure BDA0004000695500000039
The weight distribution mechanism is as follows:
Figure BDA00040006955000000310
wherein d model′ Xn' represents the feature size of the output of the hybrid model, W 1 ′、W 2 'the' representation of the parameter matrix,
Figure BDA00040006955000000311
represents Hadamard product point multiplication, and h' represents a weight correction coefficient.
The invention has the beneficial effects that: according to the invention, the countermeasure generation method is integrated into the hybrid neural network model, the countermeasure model generated aiming at the time sequence data is designed, the supervised learning of the embedding and generating module is added on the basis of the traditional unsupervised countermeasure generation module, the limitations of unsupervised property of the countermeasure generation in the traditional neural network method and the limitation of single model on the extraction of the data characteristics are broken, the information deficiency caused by the deficiency of the training set is overcome, the defects of unbalanced data and difficult acquisition of experimental data in the lithium battery research are overcome, and the high-precision lithium battery health condition monitoring based on the limited data set is realized.
Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention. The objects and other advantages of the invention may be realized and obtained by means of the instrumentalities and combinations particularly pointed out in the specification.
Drawings
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in the following preferred detail with reference to the accompanying drawings, in which:
FIG. 1 is a diagram of a frame model of the present invention.
Detailed Description
Other advantages and effects of the present invention will become apparent to those skilled in the art from the following disclosure, which describes the embodiments of the present invention with reference to specific examples. The invention may be practiced or carried out in other embodiments that depart from the specific details, and the details of the present description may be modified or varied from the spirit and scope of the present invention. It should be noted that the illustrations provided in the following embodiments merely illustrate the basic idea of the present invention by way of illustration, and the following embodiments and features in the embodiments may be combined with each other without conflict.
Wherein the drawings are for illustrative purposes only and are shown in schematic, non-physical, and not intended to limit the invention; for the purpose of better illustrating embodiments of the invention, certain elements of the drawings may be omitted, enlarged or reduced and do not represent the size of the actual product; it will be appreciated by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.
The same or similar reference numbers in the drawings of embodiments of the invention correspond to the same or similar components; in the description of the present invention, it should be understood that, if there are terms such as "upper", "lower", "left", "right", "front", "rear", etc., that indicate an azimuth or a positional relationship based on the azimuth or the positional relationship shown in the drawings, it is only for convenience of describing the present invention and simplifying the description, but not for indicating or suggesting that the referred device or element must have a specific azimuth, be constructed and operated in a specific azimuth, so that the terms describing the positional relationship in the drawings are merely for exemplary illustration and should not be construed as limiting the present invention, and that the specific meaning of the above terms may be understood by those of ordinary skill in the art according to the specific circumstances.
Aiming at the problem of lack of abnormal working condition samples in a lithium battery health state prediction task, the invention learns the sharing information and time dependency relationship of a discharge curve by a deep learning method, thereby realizing accurate prediction of the battery health state. As shown in fig. 1, at a data end, an countermeasure network model is generated by constructing a time sequence, battery usage data under abnormal working conditions is processed, and a training set after data enhancement is obtained; in the modeling process, an end-to-end high-performance learning model is built by researching a multi-head attention mechanism and a time sequence convolutional neural network parallel hybrid model; at the prediction end, the output of the hybrid model is connected, corrected and gives the prediction result.
The invention comprises the following specific contents:
1. battery anomaly data enhancement based on time series generation countermeasure model
Acquiring an original training dataset
Figure BDA0004000695500000051
Randomly generated noise dataset +>
Figure BDA0004000695500000052
Where T represents the sequence length, N e {1,..n } represents the single sample index, and N represents the number of samples. Definition of feature space->
Figure BDA0004000695500000053
Distribution p (X) 1:T ) Generator G distribution->
Figure BDA0004000695500000054
Utilize training dataset->
Figure BDA0004000695500000055
Training as close as possible to the feature space distribution p (X 1:T ) Is a generated distribution of (1)
Figure BDA0004000695500000056
Thereby generating an enhanced data set->
Figure BDA0004000695500000057
The process is as follows:
data for lithium battery
Figure BDA0004000695500000058
The time sequence generating countermeasure network is used as a data enhancement model for generating effective data which is approximate to training data as far as possible. In this model, feature space +_ is realized by embedding function e>
Figure BDA0004000695500000059
And potential space->
Figure BDA00040006955000000510
Completely reversible mapping between:
Figure BDA00040006955000000511
and potentially represents h by recovering the r function 1:T Reconstruction as an accurate representation of feature space
Figure BDA00040006955000000512
Figure BDA00040006955000000513
Embedding function e, reconstruction function r and mapping-reconstruction loss function of the process
Figure BDA00040006955000000514
Can be expressed as:
Figure BDA00040006955000000515
/>
Figure BDA00040006955000000516
in the formula, h t Representing potential space
Figure BDA00040006955000000517
Output vector of x t Representing the original input data, < >>
Figure BDA00040006955000000518
Representing the predicted network input vector, p representing the feature spatial distribution.
During the generation of enhanced data, the original training data set
Figure BDA00040006955000000519
And noise data set->
Figure BDA00040006955000000520
Training for generator G. During training, generator G operates in two states, an open loop mode and a closed loop mode.
In open loop mode (unsupervised learning), the discriminator D is from potential space
Figure BDA00040006955000000521
Extracts the input and outputs the true and false classification +.>
Figure BDA00040006955000000522
Figure BDA0004000695500000061
Wherein the generator function and the discriminator function are:
Figure BDA0004000695500000062
Figure BDA0004000695500000063
in the method, in the process of the invention,
Figure BDA0004000695500000064
generator function, ++>
Figure BDA0004000695500000065
G () is the discriminator function, g () is the discriminator network, d () is z t Is noise data.
The final goal of the generator can be expressed as:
G * ＝argminmaxV(G,D)
it is desirable to obtain a generator G that minimizes the maximum by countering the differences between the evaluation data sets, so vectors obtained in two new feature spaces using JS divergence measures, the JS divergence theory is expressed as follows:
Figure BDA0004000695500000066
for a given set
Figure BDA0004000695500000067
Generating sample by G->
Figure BDA0004000695500000068
Similarly, the output feature vector X is generated by embedding part of the network E i ＝E(x i ) Obtaining x i ＝{x 1 ,x 2 ,...,x n -it can be seen that:
Figure BDA0004000695500000069
wherein p is date Is the real data, p g Is dummy data. Optimization discriminator
Figure BDA00040006955000000610
Update by gradient descent->
Figure BDA00040006955000000611
/>
Figure BDA00040006955000000612
Where m represents the total number of data samples. Finally, the antagonistic unsupervised loss function between generator G and discriminator D can be expressed as:
Figure BDA00040006955000000613
wherein y is t Representing the output characteristics of the discriminator and t representing the tag.
In closed loop mode (supervised learning), noise data set
Figure BDA0004000695500000071
Conversion to potential space by generator>
Figure BDA0004000695500000072
Figure BDA0004000695500000073
To further capture the distribution of the raw data, a supervised learning method is introduced, in which a generator receives an embedded sequence h of actual data 1:t-1 Thus embedding the potential distribution p (H t |H 1:t-1 ) And generating potential distributions
Figure BDA0004000695500000074
The difference distribution between them can be expressed by a maximum likelihood function as:
Figure BDA0004000695500000075
during the entire data enhancement model construction process, the discriminator attempts to maximize its classification accuracy for the training data and the generated data; while the generator tries to minimize, when both reach nash equilibrium, the data enhancement model construction is complete.
2. Lithium battery health state monitoring model based on multi-head attention mechanism and time sequence convolution hybrid model
1) Battery data feature extraction
In the hybrid neural network model, a 1d-TCN and an encoder with a multi-head attention mechanism are used as feature extractors for simultaneously extracting spatial information and time-dependent information in lithium battery data.
(1) Aiming at the characteristics of one-dimensional lithium battery sequence, the device is provided with a multi-head attention mechanismThe encoder extracts the time-efficient features of lithium battery data (including raw data and enhanced data) in parallel using global information retrieval, for the layer L (L e 1,2, l.) encoder model, let
Figure BDA0004000695500000076
And->
Figure BDA0004000695500000077
Is the input and output of the layer i encoder, where D represents the dimension of the vector. In particular, the +>
Figure BDA0004000695500000078
And->
Figure BDA0004000695500000079
Position coding of input samples and input samples, respectively, of a transducer +.>
Figure BDA00040006955000000710
Is the input of the layer 1 encoder after position embedding. The position code stores the relative and absolute positions of each element of the input sequence in the sequence, and the position code is calculated as follows:
A (1) ＝PE (1) +X (1)
Figure BDA00040006955000000711
Figure BDA00040006955000000712
where pos represents the position of the current element in the entire vector; i represents the index of each element in the vector, i.e. 2i represents even positions and 2i+1 represents odd positions; d represents the dimension of the vector, which coincides with the encoded vector. By trigonometric function properties:
Figure BDA00040006955000000713
the method can obtain the following steps:
Figure BDA0004000695500000081
where k represents a relative position vector, it can be seen that pos+k position vectors can be represented as a linear combination of pos and k position vectors, which means that the absolute position vector contains a relative position vector, and thus the time sequence of the lithium battery sequence is represented. After position coding, A (l) Input to the encoder results in an encoder output:
Figure BDA0004000695500000082
wherein h represents the number of multi-head attention mechanisms, C is a splicing function,
Figure BDA0004000695500000083
and
Figure BDA0004000695500000084
all represent weight matrix, ">
Figure BDA0004000695500000085
And->
Figure BDA0004000695500000086
All represent a multi-headed mapping parameter matrix,>
Figure BDA0004000695500000087
representing the fusion matrix.
(2) In addition to the multi-headed attention mechanism, 1d-TCN was introduced as a feature extractor for extracting the space-efficient features of the battery data. For the layer N (N e 1,2,., N) TCN model, let
Figure BDA0004000695500000088
And->
Figure BDA0004000695500000089
Input and output of one-dimensional convolution layer, convolution kernel +.>
Figure BDA00040006955000000810
The number of (2) is k m Wherein L and W represent the length and width of the input data, respectively, and since the lithium battery data is a one-dimensional time series, the input data X 1 I.e., the length of the sample, and the width, i.e., the type of lithium battery data sample, such as current, voltage, temperature, sample time, capacity, etc. Thus, for a given input X n-1 Output Y n Can be expressed as:
Figure BDA00040006955000000811
Figure BDA00040006955000000812
where, σ represents the activation function,
Figure BDA00040006955000000813
represents the offset of the convolution kernel, d represents the step size of the convolution kernel, s v 、c v Position coordinates, k, representing the extracted features of the convolution kernel v Represents the kth v And a convolution kernel.
After l convolution, Y n Input to the maximum pooling layer to obtain:
Figure BDA00040006955000000814
Figure BDA00040006955000000815
wherein P is n For the purpose of pooling the vectors,o represents a pooling core; x is x l 、y w Representing position coordinates of the pooled output; d represents the pooling step size.
Briefly, training a one-dimensional convolutional network from input to output can be expressed as learning a parameter Θ of a complex nonlinear function F (x|Θ):
Y＝F(x|Θ)＝f L (L f 2 (f 1 (x|θ 1 )|θ 2 |θ L )
2) Regression model-based battery state of health prediction
A regression model based on a weight distribution mechanism is proposed to predict the state of health of the lithium battery. The input of the regression model is the characteristic vector V output by the hybrid neural network model, and the characteristic vector is the output B of the transducer (l) And an output P of 1d-TCN n And (5) splicing to obtain the product. Feature vectors for mixed model output
Figure BDA0004000695500000091
The weight allocation mechanism can be expressed as:
Figure BDA0004000695500000092
wherein d model X n' represents the feature size of the mixed model output,
Figure BDA0004000695500000093
parameter matrix determined for training +_>
Figure BDA0004000695500000094
Represents Hadamard product point multiplication, and h' represents a weight correction coefficient.
Finally, it is noted that the above embodiments are only for illustrating the technical solution of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications and equivalents may be made thereto without departing from the spirit and scope of the present invention, which is intended to be covered by the claims of the present invention.

Claims (8)

1. A method for monitoring the health state of a lithium battery based on a data enhancement hybrid model is characterized by comprising the following steps: the method takes a mixed neural network model based on a multi-head attention mechanism and a time sequence convolutional neural network as a basic model, simultaneously fuses a time sequence generation countermeasure method, embeds a supervised learning method, combines a regression model to realize the prediction of the health state of the lithium battery, and comprises the following implementation steps:
s1, acquiring a training data set and randomly generating a noise data set;
s2, constructing a data enhancement model by using a time sequence generation countermeasure network as a data enhancement model basis and adopting a training data set and a noise data set to generate an enhancement data set;
s3, respectively extracting time effective features and space effective features of battery data through a hybrid neural network model;
s4, integrating the time effective features and the space effective features, and inputting the integrated feature vectors into a regression model to predict the health state of the lithium battery.
2. The method for monitoring the health status of a lithium battery according to claim 1, wherein: in step S1, the training data set is represented as
Figure FDA0004000695490000011
The noise dataset is denoted +.>
Figure FDA0004000695490000012
Where T represents the sequence length, N e {1,..n } represents the single sample index, and N represents the number of samples.
3. The method for monitoring the health status of a lithium battery according to claim 1, wherein: in step S2, the data enhancement model includes a generator G for generating valid data approximated to the real data and a discriminator D for classifying the real data and the valid data.
4. The lithium battery health status monitoring method of claim 3, wherein: the data enhancement model construction process comprises the following steps: training a generator G by using the training data set and the noise data set, and obtaining a generator G of a minimum maximum value by means of countermeasure evaluation; meanwhile, the discriminator D is optimized through JS divergence, and the classification accuracy of the discriminator D is maximized;
in the construction process of the data enhancement model, the discriminator D maximizes the classification accuracy of the training data and the effective data, the generator G minimizes the maximum value, and when the maximization of the discriminator D and the minimization of the generator G reach Nash equilibrium, the construction of the data enhancement model is completed.
5. The method for monitoring the health status of a lithium battery according to claim 1, wherein: in step S3, extracting time effective characteristics of lithium battery data by adopting an encoder with a multi-head attention mechanism;
input a of the encoder (l) The method comprises the following steps:
A (l) ＝PE (l) +X (l)
wherein X is (l) Input samples representing a transducer, PE (l) Representing a position encoding of the input samples, wherein the position encoding is of the formula:
Figure FDA0004000695490000021
where pos represents the position of the current element in the whole vector, i represents the index of each element in the vector, and d represents the dimension of the vector;
output B of encoder (l) The method comprises the following steps:
Figure FDA0004000695490000022
wherein h represents the number of multi-head attention mechanisms, C is a splicing function,W q 、W k and W is v All represent a weight matrix, W i Q 、W i K And W is i V All represent a multi-headed mapping parameter matrix, W O Representing the fusion matrix.
6. The method for monitoring the health status of a lithium battery according to claim 1, wherein: in step S3, extracting space effective features of lithium battery data by adopting 1 d-TCN;
for the nth convolution layer, its input X n And output Y n Expressed as:
Figure FDA0004000695490000023
Figure FDA0004000695490000024
wherein L and W represent the length and width of the input data, respectively; the correspondence between inputs and outputs is shown as follows:
Figure FDA0004000695490000025
where, σ represents the activation function,
Figure FDA0004000695490000026
represents the offset of the convolution kernel, d represents the step size of the convolution kernel, s v 、c v Position coordinates, k, representing the extracted features of the convolution kernel v Representing a convolution kernel number; y is set to n Inputting the maximum pooling layer to obtain a pooling vector:
Figure FDA0004000695490000027
Figure FDA0004000695490000028
wherein P is n Representing the pooling vector, x l 、y w Representing the position coordinates of the pooling output, d representing the pooling step size.
7. The method for monitoring the health status of a lithium battery according to claim 6, wherein: the length of the input data is the length of the sample, and the width of the input data is the type of the lithium battery data sample.
8. The method for monitoring the health status of a lithium battery according to claim 1, wherein: in step S4, predicting the health state of the lithium battery by adopting a regression model based on a weight distribution mechanism;
for the feature vector obtained by integration
Figure FDA0004000695490000029
The weight distribution mechanism is as follows:
Figure FDA0004000695490000031
wherein d model Xn' represents the feature size of the output of the hybrid model, W 1 ′、W 2 'the' representation of the parameter matrix,
Figure FDA0004000695490000032
represents Hadamard product point multiplication, and h' represents a weight correction coefficient. />
CN202211612664.4A 2022-12-15 2022-12-15 Mixed model lithium battery health state monitoring method based on data enhancement Pending CN116106773A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211612664.4A CN116106773A (en) 2022-12-15 2022-12-15 Mixed model lithium battery health state monitoring method based on data enhancement

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211612664.4A CN116106773A (en) 2022-12-15 2022-12-15 Mixed model lithium battery health state monitoring method based on data enhancement

Publications (1)

Publication Number Publication Date
CN116106773A true CN116106773A (en) 2023-05-12

Family

ID=86264818

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211612664.4A Pending CN116106773A (en) 2022-12-15 2022-12-15 Mixed model lithium battery health state monitoring method based on data enhancement

Country Status (1)

Country Link
CN (1) CN116106773A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116298934A (en) * 2023-05-19 2023-06-23 河南科技学院 Modeling method of prediction network for lithium battery health state estimation
CN116466243A (en) * 2023-06-16 2023-07-21 国网安徽省电力有限公司电力科学研究院 Method and device for evaluating health state of lithium battery based on generation countermeasure network
CN117607756A (en) * 2024-01-18 2024-02-27 杭州布雷科电气有限公司 Fuse performance test platform based on antagonistic neural network

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116298934A (en) * 2023-05-19 2023-06-23 河南科技学院 Modeling method of prediction network for lithium battery health state estimation
CN116298934B (en) * 2023-05-19 2023-08-04 河南科技学院 Modeling method of prediction network for lithium battery health state estimation
CN116466243A (en) * 2023-06-16 2023-07-21 国网安徽省电力有限公司电力科学研究院 Method and device for evaluating health state of lithium battery based on generation countermeasure network
CN117607756A (en) * 2024-01-18 2024-02-27 杭州布雷科电气有限公司 Fuse performance test platform based on antagonistic neural network
CN117607756B (en) * 2024-01-18 2024-04-05 杭州布雷科电气有限公司 Fuse performance test platform based on antagonistic neural network

Similar Documents

Publication Publication Date Title
Ren et al. A data-driven auto-CNN-LSTM prediction model for lithium-ion battery remaining useful life
CN112904220B (en) UPS (uninterrupted Power supply) health prediction method and system based on digital twinning and machine learning, electronic equipment and storable medium
CN116106773A (en) Mixed model lithium battery health state monitoring method based on data enhancement
Gao et al. Machine learning toward advanced energy storage devices and systems
Ren et al. A review of machine learning state-of-charge and state-of-health estimation algorithms for lithium-ion batteries
Khan et al. Batteries state of health estimation via efficient neural networks with multiple channel charging profiles
CN106908736B (en) Lithium battery method for predicting residual useful life
Wang et al. A transferable lithium-ion battery remaining useful life prediction method from cycle-consistency of degradation trend
Zhou et al. Battery health prognosis using improved temporal convolutional network modeling
CN116679211A (en) Lithium battery health state prediction method
Li et al. A hybrid framework for predicting the remaining useful life of battery using Gaussian process regression
Zhao et al. Battery fault diagnosis and failure prognosis for electric vehicles using spatio-temporal transformer networks
CN114580262B (en) Lithium ion battery health state estimation method
CN114578234A (en) Lithium ion battery degradation and capacity prediction model considering causality characteristics
Cui et al. Ultra-early prediction of lithium-ion battery performance using mechanism and data-driven fusion model
CN117825965A (en) State prediction method and system for lithium ion battery
Song et al. Capacity estimation method of lithium-ion batteries based on deep convolution neural network
Cui et al. Long-sequence voltage series forecasting for internal short circuit early detection of lithium-ion batteries
CN117233635A (en) Echelon utilization battery performance evaluation method based on two-way parallel network
CN112557908A (en) SOC and SOH joint estimation method for lithium ion power battery
Li et al. A framework for predicting network security situation based on the improved LSTM
CN116643190A (en) Real-time monitoring method and system for lithium battery health state
Li et al. Remaining useful life prediction of lithium battery based on ACNN-Mogrifier LSTM-MMD
Fu et al. Lithium-ion battery SOH prediction based on VMD-PE and improved DBO optimized temporal convolutional network model
CN114924192A (en) Parallel battery pack safety early warning method based on neural network

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination