CN117233635A - Echelon utilization battery performance evaluation method based on two-way parallel network - Google Patents
Echelon utilization battery performance evaluation method based on two-way parallel network Download PDFInfo
- Publication number
- CN117233635A CN117233635A CN202311162580.XA CN202311162580A CN117233635A CN 117233635 A CN117233635 A CN 117233635A CN 202311162580 A CN202311162580 A CN 202311162580A CN 117233635 A CN117233635 A CN 117233635A
- Authority
- CN
- China
- Prior art keywords
- network
- data
- battery
- module
- performance evaluation
- 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
Links
- 238000011156 evaluation Methods 0.000 title claims abstract description 32
- 238000000034 method Methods 0.000 claims abstract description 31
- 230000036541 health Effects 0.000 claims abstract description 25
- WHXSMMKQMYFTQS-UHFFFAOYSA-N Lithium Chemical compound [Li] WHXSMMKQMYFTQS-UHFFFAOYSA-N 0.000 claims abstract description 20
- 229910052744 lithium Inorganic materials 0.000 claims abstract description 20
- 230000004927 fusion Effects 0.000 claims abstract description 17
- 238000000605 extraction Methods 0.000 claims abstract description 8
- 230000006870 function Effects 0.000 claims description 29
- 230000004913 activation Effects 0.000 claims description 17
- 238000013528 artificial neural network Methods 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 8
- 238000010606 normalization Methods 0.000 claims description 7
- 238000007599 discharging Methods 0.000 claims description 6
- 230000007246 mechanism Effects 0.000 claims description 6
- 230000015654 memory Effects 0.000 claims description 5
- 230000003044 adaptive effect Effects 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 4
- 238000004458 analytical method Methods 0.000 claims description 3
- 210000004027 cell Anatomy 0.000 claims description 3
- 230000008034 disappearance Effects 0.000 claims description 3
- 230000007787 long-term memory Effects 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 210000002569 neuron Anatomy 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 230000006403 short-term memory Effects 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims description 3
- 230000003862 health status Effects 0.000 claims 1
- 239000000178 monomer Substances 0.000 abstract description 2
- 238000012360 testing method Methods 0.000 description 4
- 238000012549 training Methods 0.000 description 4
- 230000007547 defect Effects 0.000 description 3
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 description 2
- HBBGRARXTFLTSG-UHFFFAOYSA-N Lithium ion Chemical compound [Li+] HBBGRARXTFLTSG-UHFFFAOYSA-N 0.000 description 2
- 230000010339 dilation Effects 0.000 description 2
- 238000005265 energy consumption Methods 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 229910001416 lithium ion Inorganic materials 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 230000002123 temporal effect Effects 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000032683 aging Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000012854 evaluation process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E60/00—Enabling technologies; Technologies with a potential or indirect contribution to GHG emissions mitigation
- Y02E60/10—Energy storage using batteries
Landscapes
- Tests Of Electric Status Of Batteries (AREA)
Abstract
The invention discloses a cascade utilization lithium battery performance evaluation method based on a double-path parallel network, which comprises the steps of building the double-path parallel network and evaluating battery performance parameters of a cascade utilization battery system, namely the health state of a battery monomer or a battery pack. The two-way parallel network comprises a light-weight network and an enhanced intelligent network, wherein the light-weight network comprises a multi-layer residual convolution block and a full connection layer; the enhanced intelligent network comprises a data embedding module, a feature extraction module and a feature fusion module, the feature data is fully extracted and learned, when the health state of the battery is higher in echelon utilization, the enhanced intelligent network estimation result is taken as the main, and the accuracy of the battery performance evaluation is ensured; when the health state is low, the lightweight network estimation result is taken as the main basis, so that the safety of the battery used in a echelon manner is ensured; the enhanced intelligent network estimation result is used as an auxiliary, and the lightweight network is supplemented and corrected, so that the method has higher engineering practical application value.
Description
Technical Field
The invention relates to the technical field of power battery management, in particular to a gradient utilization battery performance evaluation method based on a two-way parallel network.
Background
The power lithium battery of the new energy automobile decays along with the increase of the charge and discharge times, and when the rated capacity of the power battery is reduced to below 80%, the power battery does not accord with the daily operation standard of the vehicle-mounted power, and the power lithium battery is required to be retired from the electric car. But the retired battery can still be used in a ladder way in the field with low capacity requirement. However, the retired electric power battery has the problems of rapid degradation of performance, increased risk of safety failure and the like. Therefore, it is necessary to perform accurate performance evaluation of the cascade utilization battery.
The state of health of the battery is an extremely important performance index, and the current estimation method for the state of health is more and can be roughly divided into a method of formula calculation, a method based on a model and a method driven by data. The formula calculation is only rough calculation, has no complete theory and method, and cannot accurately estimate the actual performance of the battery; model-based methods suffer from modeling difficulties. In addition, the safety state, the electrical property state and the like of the power lithium battery for the vehicle are complex after the power lithium battery is retired, the gradient utilization scene is not fixed, the application form is various, the requirement of gradient utilization cannot be met by evaluating and testing through a formula or modeling, and the safety is difficult to reliably guarantee.
Based on the defects of the method, the invention provides a method for evaluating the performance of a battery by echelon utilization based on a two-way parallel network.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, and provides a cascade utilization battery performance evaluation method based on a two-way parallel network, which mainly adopts a data driving method, utilizes historical big data and test data as neural network input, comprehensively and adaptively evaluates the cascade utilization battery, is a more effective method, and has great significance on the safe use and the service life extension of the cascade utilization battery.
In order to solve the technical problems, the technical scheme of the invention is as follows:
a cascade utilization battery performance evaluation method based on a two-way parallel network comprises the following steps:
step (1): and acquiring and preprocessing lithium battery data in a gradient manner.
The lithium battery data used in the cascade comprises the discharge characteristics of the lithium ion battery system, namely the terminal voltage X in the discharge process V Current X I Temperature X T Data and corresponding battery capacity X C Data. Correlation between health factor (feature) and capacity of three data was calculated by pearson correlation coefficient analysis:
wherein X is V,I,T Representative voltage, current, temperature input, X C Representing corresponding capacity data; n represents the number of sample data and selects the correlation coefficientThe data closest to 1 is all over the discharge periodThe data value is used as the original input of the network, and normalization operation is carried out to obtain network input data X i 。
Step (2): a lightweight network is constructed.
The lightweight network comprises a multi-layer residual convolution block and a full connection layer; the characteristics are extracted and learned by the multi-layer residual convolution block and then input into a subsequent full-connection layer for output. Each residual convolution block comprises three layers of expansion convolution with different expansion rates and one layer of one-dimensional convolution, and the expansion convolution increases the receptive field and reduces the calculated amount on the premise of not carrying out pooling and not losing information. The step length of the first expansion convolution is k, the step length of the other two expansion convolutions is 1, the expansion rates are 1, n and n2 respectively, the activation function adopts a ReLU function, and three layers of expansion convolutions are connected in series:
wherein F is DCNN1 、F DCNN2 、F DCNN3 Representing the first, second, and third layer dilation convolutions, respectively. The step length of one-dimensional convolution is k, and the activation function adopts a ReLU6 function:
the one-dimensional convolution and the expansion convolution are connected in parallel to form a residual block:
the residual connection accelerates the network convergence, reduces the model training time and improves the estimation efficiency. After passing through the multi-layer residual convolution block, the data characteristics are input to a full connection layer, and the full connection layer adopts a ReLU6 activation function:
and the full connection layer maps the input characteristics to an output result to obtain a final cascade utilization battery performance evaluation initial value.
Step (3): high accuracy enhanced intelligent network is built.
The enhanced intelligent network comprises a data embedding module, a feature extraction module and a feature fusion module.
In particular, the method comprises the steps of,
the data embedding module comprises a one-dimensional convolution network and a two-way long-short-term memory network, wherein the step length of the one-dimensional convolution network is 1, and the activation function is a ReLU function. Obtaining initial characteristics through a one-dimensional convolution network:
initial feature pairs through a two-way long and short term memory networkThe feature embedding is carried out, and the two-way long-short-term memory network processes the past and future feature information at the same time, so that the model performance of time sequence prediction can be improved. The calculation steps are as follows:
the feature extraction module includes a temporal attention module and a spatial attention module. The time attention module firstly exchanges the input dimension:
and then inputting the features into a multi-head sparse self-attention layer, optimizing the point-by-point dot product operation of an original multi-head self-attention mechanism by multi-head sparse self-attention, and reducing the energy consumption of the original self-attention secondary dot product calculation. Three scalar quantities in the attention mechanism are defined, query (Q), key (K) and Value (V), respectively. The formula is as follows:
wherein the method comprises the steps ofIs obtained by probability sparsification of Q, K T The K is transposed to form a Q matrix with only a few large contributions to the attention value, ++>Is a scale factor that prevents the gradient from increasing due to disappearance, softmax () is an activation function;
summing and normalizing the original features input into the module and the features passing through the multi-head sparse self-attention layer:
and then output to a feedforward neural network:
finally input intoAnd carrying out summation and normalization operation on the processed characteristics of the feedforward neural network to obtain the final output of the time attention module:
similar to the time attention module, the space attention module inputs original features on the premise of not carrying out feature dimension exchangeProcessing the multi-head attention and the feedforward neural network to finally obtain the output +.>
The feature fusion module comprises a feature addition operation, three expansion convolution networks with different expansion rates and a full connection layer, wherein an activation function of each expansion convolution adopts a ReLU function, and the expansion rates are respectively 1, n and n2. Before feature stitching, the output of the time attention module is subjected to feature dimension transformation so as to facilitate the subsequent adding operation:
output of time attention moduleAnd the output of the spatial attention module->Adding to obtain the initial fusion feature->And then carrying out further adaptive fusion on the characteristics through three series expansion convolutions to realize deep information interaction. And finally, outputting the characteristics through a full-connection layer with the number of neurons set to be 1, and obtaining a final estimation result of the enhanced intelligent network.
Step (4): and constructing a two-way parallel network.
And constructing a two-way parallel network for gradient battery performance evaluation, wherein one path is a lightweight network, and the other path is an enhanced intelligent network. The two paths of networks are connected in parallel, and the performance of the battery is evaluated. And determining super parameters of the two paths of networks by adopting a cross verification method, wherein the super parameters comprise the number of hidden layers, the learning rate, the iteration times and the like.
And (5) performing performance evaluation of the cascade utilization batteries based on the two-way parallel network, namely estimating the health state of the battery cells or the battery packs.
When the battery state of health is greater than 60%, the enhanced intelligent network estimation result is taken as a main part, and the lightweight network is taken as an auxiliary part. When the battery health state is lower than 60%, the lightweight network estimation result is taken as the main basis, the intelligent network estimation result is enhanced, and the lightweight network estimation result is corrected; that is, when the estimated difference between the two is greater than 2%, the output of the two is combined and output by giving different weight levels at the time of the update of the enhanced intelligent network, and the output of the enhanced intelligent network at the moment is used for the update of the subsequent time (the time of the update of the real-time lightweight network). Specifically, when the prediction difference between two networks is more than 2% and less than 3%, the output of the enhanced intelligent network is multiplied by 25%, and the output of the lightweight network is multiplied by 75%; when the output of the enhanced intelligent network is more than 3 percent and less than 4 percent, the output of the enhanced intelligent network is multiplied by 30 percent, and the output of the lightweight network is multiplied by 70 percent; similarly, when the difference is more than 8%, alarm information is given. For the battery health state, selecting all data values of data with correlation coefficients closest to 1 in voltage factor-capacity, current factor-capacity and temperature factor-capacity in the whole discharging period as network original input, taking capacity data at corresponding moments as input labels, and calculating the performance of the battery used in a cascade, namely the health state:
wherein C is i And C 0 Representing the battery capacity and the battery initial capacity of the i-th cycle, respectively.
The invention has the following characteristics and beneficial effects:
by adopting the technical scheme, the invention fully preprocesses four time sequences of voltage, current, temperature and capacity which are directly measured by the battery in echelon by utilizing the two-way parallel network, and inputs the four time sequences into the two-way parallel network for self-adaptive feature extraction and fusion. By a brand new method based on the two-way parallel network, the defects of low accuracy and poor real-time performance of a single network in performance evaluation of the cascade utilization battery are avoided, and the use safety of the cascade utilization battery in various application scenes is improved. The evaluation process is simple, and has important significance for prolonging the service life of the battery used in the echelon.
In addition, the health state of the battery monomer or the battery pack is estimated in a gradient manner through two parallel networks, namely a light-weight network and an enhanced intelligent network. Considering the importance of using the battery in a gradient manner, when the health state is low, the lightweight network estimation result is taken as the standard, the high-accuracy enhanced intelligent network estimation result is taken as the auxiliary, and the lightweight network is corrected in real time. And finally, obtaining the performance evaluation result of the gradient utilization battery with high accuracy and high safety.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions of the prior art, the drawings which are used in the description of the embodiments or the prior art will be briefly described, it being obvious that the drawings in the description below are only some embodiments of the invention, and that other drawings can be obtained according to these drawings without inventive faculty for a person skilled in the art.
Fig. 1 is a frame diagram of a method for evaluating performance of a battery with gradient utilization based on a two-way parallel network according to an embodiment of the present invention.
Detailed Description
It should be noted that, without conflict, the embodiments of the present invention and features of the embodiments may be combined with each other.
The invention provides a cascade utilization battery performance evaluation method based on a two-way parallel network, which is shown in figure 1 and comprises the following steps:
step (1): and acquiring and preprocessing lithium battery data in a gradient manner.
In this embodiment, experiments were conducted on the public dataset NASA aging dataset for the proposed battery performance assessment method.
The lithium battery data used in the cascade comprises the discharge characteristics of the lithium ion battery system, namely the terminal voltage in the discharge processX V Current X I Temperature X T Data and corresponding battery capacity X C Data. Correlation between health factor (feature) and capacity of three data was calculated by pearson correlation coefficient analysis:
wherein X is V,I,T Representative voltage, current, temperature input, X C Representing corresponding capacity data; n represents the number of sample data and selects the correlation coefficientAll data values of the data closest to 1 in the whole discharging period are taken as network original inputs, and normalization operation is carried out to obtain network input data X i 。
Step (2): a lightweight network is constructed.
The lightweight network comprises a multi-layer residual convolution block and a full connection layer; the characteristics are extracted and learned by the multi-layer residual convolution block and then input into a subsequent full-connection layer for output. Each residual convolution block comprises three layers of expansion convolution with different expansion rates and one layer of one-dimensional convolution, and the expansion convolution increases the receptive field and reduces the calculated amount on the premise of not carrying out pooling and not losing information. The step length of the first expansion convolution is k, the step length of the other two expansion convolutions is 1, the expansion rates are 1, n and n2 respectively, the activation function adopts a ReLU function, and three layers of expansion convolutions are connected in series:
wherein F is DCNN1 、F DCNN2 、F DCNN3 Representing the first, second, and third layer dilation convolutions, respectively. The step length of one-dimensional convolution is k, and the activation function adopts a ReLU6 function:
the one-dimensional convolution and the expansion convolution are connected in parallel to form a residual block:
the residual connection accelerates the network convergence, reduces the model training time and improves the estimation efficiency. After passing through the multi-layer residual convolution block, the data characteristics are input to a full connection layer, and the full connection layer adopts a ReLU6 activation function:
and the full connection layer maps the input characteristics to an output result to obtain a final cascade utilization battery performance evaluation initial value.
Step (3): high accuracy enhanced intelligent network is built.
The enhanced intelligent network comprises a data embedding module, a feature extraction module and a feature fusion module.
The data embedding module comprises a one-dimensional convolution network and a two-way long-short-term memory network, wherein the step length of the one-dimensional convolution network is 1, and the activation function is a ReLU function. Obtaining initial characteristics through a one-dimensional convolution network:
initial feature pairs through a two-way long and short term memory networkThe feature embedding is carried out, and the two-way long-short-term memory network processes the past and future feature information at the same time, so that the model performance of time sequence prediction can be improved. The calculation steps are as follows:
the feature extraction module includes a temporal attention module and a spatial attention module. The time attention module firstly exchanges the input dimension:
and then inputting the features into a multi-head sparse self-attention layer, optimizing the point-by-point dot product operation of an original multi-head self-attention mechanism by multi-head sparse self-attention, and reducing the energy consumption of the original self-attention secondary dot product calculation. Three scalar quantities in the attention mechanism are defined, query (Q), key (K) and Value (V), respectively. The formula is as follows:
wherein the method comprises the steps ofIs obtained by probability sparsification of Q, K T The K is transposed to form a Q matrix with only a few large contributions to the attention value, ++>Is a scale factor that prevents the gradient from increasing due to disappearance, softmax () is an activation function;
summing and normalizing the original features input into the module and the features passing through the multi-head sparse self-attention layer:
and then output to a feedforward neural network:
finally input intoAnd carrying out summation and normalization operation on the processed characteristics of the feedforward neural network to obtain the final output of the time attention module:
similar to the time attention module, the space attention module inputs original features on the premise of not carrying out feature dimension exchangeProcessing the multi-head attention and the feedforward neural network to finally obtain the output +.>
The feature fusion module comprises a feature addition operation, three expansion convolution networks with different expansion rates and a full connection layer, wherein an activation function of each expansion convolution adopts a ReLU function, and the expansion rates are respectively 1, n and n2. Before feature stitching, the output of the time attention module is subjected to feature dimension transformation so as to facilitate the subsequent adding operation:
output of time attention moduleAnd the output of the spatial attention module->Adding to obtain the initial fusion feature->And then carrying out further adaptive fusion on the characteristics through three series expansion convolutions to realize deep information interaction. And finally, outputting the characteristics through a full-connection layer with the number of neurons set to be 1, and obtaining a final estimation result of the enhanced intelligent network.
Step (4): and constructing a two-way parallel network.
And constructing a two-way parallel network for gradient battery performance evaluation, wherein one path is a lightweight network, and the other path is an enhanced intelligent network. The two paths of networks are connected in parallel, and the performance of the battery is evaluated. And determining super parameters of the two paths of networks by adopting a cross verification method, wherein the super parameters comprise the number of hidden layers, the learning rate, the iteration times and the like.
And (5) performing performance evaluation of the cascade utilization batteries based on the two-way parallel network, namely estimating the health state of the battery cells or the battery packs.
When the battery state of health is greater than 60%, the enhanced intelligent network estimation result is taken as a main part, and the lightweight network is taken as an auxiliary part. The estimation result of the enhanced intelligent network is updated and displayed in real time, and the estimation result of the lightweight network is updated in the background but is not displayed for the user; when the enhanced intelligent network estimation result is abnormal (greatly reduced/spanned, etc.), the lightweight network estimation result is adopted as output, and alarm information is given at the same time. When the battery health state is lower than 60%, the lightweight network estimation result is taken as the main basis, the intelligent network estimation result is enhanced, and the lightweight network estimation result is corrected; that is, when the estimated difference between the two is greater than 2%, the output of the two is combined and output by giving different weight levels at the time of the update of the enhanced intelligent network, and the output of the enhanced intelligent network at the moment is used for the update of the subsequent time (the time of the update of the real-time lightweight network).
Specifically, in this embodiment, when the difference between two network predictions is greater than 2% and less than 3%, the intelligent network output is enhanced by 25% and the lightweight network output by 75%; when the output of the enhanced intelligent network is more than 3 percent and less than 4 percent, the output of the enhanced intelligent network is multiplied by 30 percent, and the output of the lightweight network is multiplied by 70 percent; similarly, when the difference is more than 8%, alarm information is given. For the battery health state, selecting all data values of data with correlation coefficients closest to 1 in voltage factor-capacity, current factor-capacity and temperature factor-capacity in the whole discharging period as network original input, taking capacity data at corresponding moments as input labels, and calculating the performance of the battery used in a cascade, namely the health state:
wherein C is i And C 0 Representing the battery capacity and the battery initial capacity of the i-th cycle, respectively.
In this embodiment, on the basis of verifying the battery health state estimation precision of the lightweight network and the enhanced intelligent network, the training time and the testing time of the two networks are tested, and the results are shown in the following table:
TABLE 1
Wherein MAE and RMSE are average absolute error and root mean square error loss functions respectively. As can be seen from Table 1, the lightweight network and the enhanced intelligent network provided by the invention can well realize the task of the gradient battery performance evaluation method, the training time and the testing time of the lightweight network are obviously lower than those of the enhanced intelligent network, the evaluation accuracy of the enhanced intelligent network is obviously higher than that of the lightweight network, but both networks can ensure higher accuracy and faster evaluation accuracy, and can realize accurate performance evaluation on the premise of ensuring the safety of gradient battery.
The embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. It will be apparent to those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments, including the components, without departing from the principles and spirit of the invention, yet fall within the scope of the invention.
Claims (9)
1. The echelon utilization lithium battery performance evaluation method based on the two-way parallel network is characterized by comprising the following steps of:
step (1): gradient utilization lithium battery data acquisition and preprocessing
The data of the lithium battery used in the echelon comprises terminal voltage X in the discharging process V Current X I Temperature X T Data and corresponding battery capacity X C Data;
step (2): constructing a lightweight network
The light-weight network comprises a plurality of layers of residual convolution blocks and a full-connection layer, each residual convolution block comprises three layers of expansion convolution blocks and one layer of one-dimensional convolution, the three layers of expansion convolution blocks are connected in series, the one-dimensional convolution blocks and the expansion convolution blocks are connected in parallel to form a residual convolution block, and an activation function is a ReLU6 function;
step (3): building an enhanced intelligent network
The enhanced intelligent network comprises a data embedding module, a feature extraction module and a feature fusion module;
the data embedding module comprises a one-dimensional convolution network and a two-way long-short-term memory network, wherein the step length of the one-dimensional convolution network is 1, and the activation function is a ReLU function;
the feature extraction module comprises a time attention module and a space attention module;
the feature fusion module comprises a feature addition operation, three expansion convolutions and a full connection layer, wherein an activation function of each expansion convolution adopts a ReLU function, and the expansion rates are respectively 1, n and n2;
step (4): building two-way parallel network
Connecting a lightweight network and an enhanced intelligent network in a parallel manner to form a two-way parallel network; determining super parameters of the two paths of networks by adopting a cross verification method, wherein the super parameters comprise the number of hidden layers, the learning rate and the iteration times;
and (5) performing performance evaluation of the cascade utilization batteries based on the two-way parallel network, namely estimating the health state of the battery cells or the battery packs.
2. The two-way parallel network-based echelon utilization lithium battery performance evaluation method according to claim 1, wherein the preprocessing method in the step (1) is as follows:
the correlation between health factor and capacity of three data was calculated by pearson correlation coefficient analysis:
wherein X is V,I,T Representative voltage, current, temperature input, X C Representing corresponding capacity data; n represents the number of sample data and selects the correlation coefficientAll data values of the data closest to 1 in the whole discharging period are taken as network original inputs, and normalization operation is carried out to obtain network input data X i 。
3. The two-way parallel network-based echelon utilization lithium battery performance evaluation method according to claim 2, wherein the step size of the first expansion convolution is k, the step sizes of the other two expansion convolutions are 1, the expansion rates of the three expansion convolutions are 1, n and n2 respectively, and three layers of expansion convolutions are connected in series:
the step size of the one-dimensional convolution is k.
4. The two-way parallel network-based echelon utilization lithium battery performance evaluation method according to claim 3, wherein in the step (2), the preprocessed data is subjected to a multi-layer residual convolution block to extract data features, the data features are input to a full connection layer, and the full connection layer adopts a ReLU6 activation function:
and the full connection layer maps the input characteristics to an output result to obtain a final cascade utilization battery performance evaluation initial value.
5. The two-way parallel network-based echelon utilization lithium battery performance evaluation method according to claim 2, wherein the implementation method of the data embedding module is as follows:
obtaining initial characteristics through a one-dimensional convolution network:
initial feature pairs through a two-way long and short term memory networkThe characteristic embedding is carried out, and the calculation steps are as follows:
6. the two-way parallel network-based echelon utilization lithium battery performance evaluation method according to claim 2, wherein the implementation method of the time attention module is as follows:
firstly, exchanging input dimensions:
then inputting the features into a multi-head sparse self-attention layer, optimizing the dot product operation of an original multi-head self-attention mechanism, and defining three scalar quantities in the attention mechanism, namely Query (Q), key (K) and Value (V), wherein the formulas are as follows:
wherein the method comprises the steps ofIs obtained by probability sparsification of Q, K T The K is transposed to form a Q matrix with only a few large contributions to the attention value, ++>Is a scale factor that prevents the gradient from increasing due to disappearance, softmax () is an activation function;
features of attention to input time moduleAnd carrying out summation and normalization operation on the characteristics of the multi-head sparse self-attention layer:
and then output to a feedforward neural network:
finally input intoAnd carrying out summation and normalization operation on the processed characteristics of the feedforward neural network to obtain the final output of the time attention module:
time attention moduleSimilarly, the spatial attention module inputs features without feature dimension exchangeProcessing the multi-head attention and the feedforward neural network to finally obtain the output +.>
7. The two-way parallel network-based echelon utilization lithium battery performance evaluation method according to claim 6, wherein the implementation method of the feature fusion module is as follows:
before feature stitching, firstly, performing feature dimension transformation on the output of the time attention module:
output of time attention moduleAnd the output of the spatial attention module->Adding to obtain initial fusion characteristicsThe initial fusion feature is then->And carrying out adaptive fusion through three series expansion convolutions, and outputting the characteristics after the adaptive fusion through a full-connection layer with the neuron number set to be 1 to obtain the final estimation result of the enhanced intelligent network.
8. The two-way parallel network-based echelon utilization lithium battery performance evaluation method according to claim 7, wherein the specific method of the step (5) is as follows:
when the battery health state is more than 60%, the enhanced intelligent network estimation result is taken as a main part, and the lightweight network is taken as an auxiliary part; when the battery state of health is lower than 60%, the lightweight network estimation result is taken as the main basis, and the enhanced intelligent network estimation result is used for correcting the lightweight network estimation result in a weight mode.
9. The two-way parallel network-based echelon utilization lithium battery performance evaluation method according to claim 8, wherein the battery health status quantifying method comprises the steps of: selecting all data values of data with correlation coefficients closest to 1 in voltage factor-capacity, current factor-capacity and temperature factor-capacity in the whole discharging period as network original input, taking capacity data at corresponding time as input labels, and calculating the performance of the cascade utilization battery, namely the health state:
wherein C is i And C 0 Representing the battery capacity and the battery initial capacity of the i-th cycle, respectively.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311162580.XA CN117233635A (en) | 2023-09-08 | 2023-09-08 | Echelon utilization battery performance evaluation method based on two-way parallel network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311162580.XA CN117233635A (en) | 2023-09-08 | 2023-09-08 | Echelon utilization battery performance evaluation method based on two-way parallel network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117233635A true CN117233635A (en) | 2023-12-15 |
Family
ID=89095969
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311162580.XA Pending CN117233635A (en) | 2023-09-08 | 2023-09-08 | Echelon utilization battery performance evaluation method based on two-way parallel network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117233635A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117849628A (en) * | 2024-03-08 | 2024-04-09 | 河南科技学院 | Lithium ion battery health state estimation method based on time sequence transformation memory network |
-
2023
- 2023-09-08 CN CN202311162580.XA patent/CN117233635A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117849628A (en) * | 2024-03-08 | 2024-04-09 | 河南科技学院 | Lithium ion battery health state estimation method based on time sequence transformation memory network |
| CN117849628B (en) * | 2024-03-08 | 2024-05-10 | 河南科技学院 | Lithium ion battery health state estimation method based on time sequence transformation memory network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113009349B (en) | Lithium ion battery health state diagnosis method based on deep learning model | |
| CN114372417A (en) | Electric vehicle battery health state and remaining life evaluation method based on charging network | |
| CN112630662B (en) | Power battery SOH estimation method based on data driving and multi-parameter fusion | |
| CN109375116B (en) | Battery system abnormal battery identification method based on self-encoder | |
| CN117233635A (en) | Echelon utilization battery performance evaluation method based on two-way parallel network | |
| CN112881914B (en) | Lithium battery health state prediction method | |
| CN113687242A (en) | Lithium ion battery SOH estimation method for optimizing and improving GRU neural network based on GA algorithm | |
| CN114791993B (en) | Power battery pack SOH prediction method and system | |
| CN114726045B (en) | Lithium battery SOH estimation method based on IPEA-LSTM model | |
| CN115453399B (en) | Battery pack SOH estimation method considering inconsistency | |
| CN115366683A (en) | Fault diagnosis strategy for new energy automobile power battery multi-dimensional model fusion | |
| CN115598557B (en) | Lithium battery SOH estimation method based on constant-voltage charging current | |
| CN115980584A (en) | Lithium battery RUL estimation method based on multi-feature fusion LSTM network | |
| CN116679211A (en) | Lithium battery health state prediction method | |
| CN116298936A (en) | Intelligent lithium ion battery health state prediction method in incomplete voltage range | |
| CN117148197A (en) | Lithium ion battery life prediction method based on integrated transducer model | |
| CN113791351B (en) | Lithium battery life prediction method based on transfer learning and difference probability distribution | |
| CN114460481A (en) | Energy storage battery thermal runaway early warning method based on Bi-LSTM and attention mechanism | |
| CN112014757A (en) | Battery SOH estimation method integrating capacity increment analysis and genetic wavelet neural network | |
| CN116679213A (en) | SOH estimation method for electric vehicle power battery based on integrated deep learning | |
| CN117435909A (en) | Non-invasive load decomposition method based on transfer learning and multidimensional feature extraction model | |
| CN116596129A (en) | Electric vehicle charging station short-term load prediction model construction method | |
| CN115481796A (en) | Method for predicting remaining service life of battery based on Bayesian hybrid neural network | |
| CN115308623A (en) | Battery state of charge estimation method based on particle resampling and searcher optimization algorithm | |
| CN114545279A (en) | Lithium battery health state estimation method based on neural network ordinary differential equation |
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 |