CN117648557A - SOH prediction method and device based on SOH combined noise reduction - Google Patents
SOH prediction method and device based on SOH combined noise reduction Download PDFInfo
- Publication number
- CN117648557A CN117648557A CN202410121852.XA CN202410121852A CN117648557A CN 117648557 A CN117648557 A CN 117648557A CN 202410121852 A CN202410121852 A CN 202410121852A CN 117648557 A CN117648557 A CN 117648557A
- Authority
- CN
- China
- Prior art keywords
- data
- soh
- noise
- health
- noise reduction
- 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
- 238000000034 method Methods 0.000 title claims abstract description 55
- 230000009467 reduction Effects 0.000 title claims abstract description 51
- 230000036541 health Effects 0.000 claims abstract description 101
- 238000004364 calculation method Methods 0.000 claims abstract description 19
- 238000012216 screening Methods 0.000 claims abstract description 16
- 238000007781 pre-processing Methods 0.000 claims abstract description 8
- 238000000354 decomposition reaction Methods 0.000 claims description 39
- 230000006870 function Effects 0.000 claims description 28
- 238000005311 autocorrelation function Methods 0.000 claims description 19
- 238000001914 filtration Methods 0.000 claims description 19
- 238000011156 evaluation Methods 0.000 claims description 18
- 238000012545 processing Methods 0.000 claims description 12
- 238000004458 analytical method Methods 0.000 claims description 11
- 238000005516 engineering process Methods 0.000 claims description 11
- 238000003062 neural network model Methods 0.000 claims description 10
- 238000007667 floating Methods 0.000 claims description 9
- 238000007726 management method Methods 0.000 claims description 9
- 238000012549 training Methods 0.000 claims description 6
- 238000012217 deletion Methods 0.000 claims description 5
- 230000037430 deletion Effects 0.000 claims description 5
- 238000010219 correlation analysis Methods 0.000 claims description 4
- 238000011049 filling Methods 0.000 claims description 4
- 238000012937 correction Methods 0.000 claims description 3
- 230000003044 adaptive effect Effects 0.000 claims description 2
- 230000015556 catabolic process Effects 0.000 abstract description 16
- 238000006731 degradation reaction Methods 0.000 abstract description 16
- 230000000694 effects Effects 0.000 description 8
- 239000000470 constituent Substances 0.000 description 6
- 238000005562 fading Methods 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000002159 abnormal effect Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000008929 regeneration Effects 0.000 description 2
- 238000011069 regeneration method Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 241000845077 Iare Species 0.000 description 1
- HBBGRARXTFLTSG-UHFFFAOYSA-N Lithium ion Chemical compound [Li+] HBBGRARXTFLTSG-UHFFFAOYSA-N 0.000 description 1
- 230000003213 activating effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000005314 correlation function Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000003487 electrochemical reaction Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 229910001416 lithium ion Inorganic materials 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 238000012847 principal component analysis method Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
Landscapes
- Secondary Cells (AREA)
- Tests Of Electric Status Of Batteries (AREA)
- Charge And Discharge Circuits For Batteries Or The Like (AREA)
Abstract
The invention discloses an SOH prediction method and equipment based on SOH combined noise reduction, belongs to the technical field of power batteries of new energy automobiles, and is used for solving the technical problems that the state of health of the existing new energy automobile battery is difficult to accurately predict, the safe operation of the new energy automobile and the prediction of a battery degradation rule are not facilitated, and the fine management of the whole life cycle energy of the power battery is also not facilitated. The method comprises the following steps: carrying out data preprocessing on original data of real vehicle operation to obtain vehicle state data; SOH calculation is carried out on the battery state in the new energy automobile according to the parking charging data, and original SOH data are obtained; carrying out joint noise reduction on the original SOH data to obtain accurate SOH data; performing feature screening on the potential health features to obtain an optimal health feature subset; based on the accurate SOH data and the optimal health feature subset, a SOH prediction model is generated.
Description
Technical Field
The application relates to the field of new energy automobile power batteries, in particular to an SOH prediction method and SOH prediction equipment based on SOH combined noise reduction.
Background
The SOH prediction is used as a core function of a battery management system of the new energy automobile, can timely and accurately predict the SOH decline trend, and has important significance in the aspects of energy fine management, maintenance and maintenance, safety early warning, gradient utilization battery residual value evaluation and the like of the whole life cycle of the power battery of the new energy automobile.
Currently, most students conduct four parts of data processing, feature extraction and screening, SOH calculation and noise reduction, and prediction model building aiming at laboratory data when researching the State of Health (SOH). However, considering that feature selection selects a feature set which is the most rich in information for SOH, rather than a single feature, some single features have high correlation with SOH, but the constituent feature subset has poor effect, some single features have low correlation with SOH, but the constituent feature subset has good effect, so the correlation magnitude of the single features cannot be used as the only standard for feature selection. In SOH calculation and noise reduction, only the main SOH fading trend can be saved, and the information such as the local capacity regeneration phenomenon in the original data cannot be saved, so that key information of SOH fading can be lost, and the SOH prediction accuracy is affected.
For a complex battery management system of a new energy automobile, the state of health of the battery of the new energy automobile is difficult to accurately predict by the prior technical means, so that the safe operation of the new energy automobile and the prediction of the degradation rule of the battery are not facilitated, and the fine management of the energy of the whole life cycle of the power battery is also not facilitated.
Disclosure of Invention
The embodiment of the application provides an SOH prediction method and equipment based on SOH combined noise reduction, which are used for solving the following technical problems: the state of health of the existing new energy automobile battery is difficult to accurately predict, which is not beneficial to the safe operation of the new energy automobile and the prediction of the battery degradation rule, and is also not beneficial to the fine management of the full life cycle energy of the power battery.
The embodiment of the application adopts the following technical scheme:
in one aspect, an embodiment of the present application provides an SOH prediction method based on SOH joint noise reduction, including: carrying out data preprocessing on original data related to actual vehicle operation in the new energy vehicle to obtain vehicle state data; wherein the vehicle state data includes: parking charging data and driving data; according to the parking charging data, SOH calculation is carried out on the battery state in the new energy automobile, and original SOH data are obtained; the original SOH data is subjected to joint noise reduction through a preset complete set empirical mode decomposition algorithm and an SG filtering technology, so that accurate SOH data after noise reduction is obtained; extracting potential health features in the vehicle state data; according to a preset forward floating sequence algorithm, carrying out feature screening on the potential health features to obtain an optimal health feature subset; and generating an SOH prediction model for predicting the state of health of the battery in the new energy automobile based on the accurate SOH data and the optimal health feature subset.
According to the embodiment of the application, through accurately processing original data in the new energy automobile, the actual automobile running environment can be considered, battery degradation health characteristics can be extracted from the angles of the charging working condition and the running working condition, the CEEDMAN-SG combined noise reduction scheme is utilized to reduce noise of the original SOH data, the SOH degradation trend can be saved at the same time when noise is reduced to the greatest extent, the SFFS (forward floating sequence) algorithm is adopted to conduct characteristic screening, the optimal health characteristic subset is constructed, the influence of SOH degradation information in the original SOH data on SOH prediction precision is saved to the greatest extent, the correlation among various characteristics is utilized, namely the correlation of single characteristics cannot be used as the unique standard of characteristic selection, the SOH prediction model which is accurately predicted is constructed, and therefore the SOH prediction model is used to accurately predict the SOH of the target battery of the new energy automobile.
In a possible implementation manner, the data preprocessing is performed on the original data related to the actual vehicle operation in the new energy vehicle to obtain the vehicle state data, and specifically includes: collecting new energy automobile operation data through the battery management system of the new energy automobile; uploading the new energy automobile operation data to a cloud server, and analyzing to obtain the original data of the actual automobile operation in the new energy automobile; wherein the raw data comprises at least: vehicle speed, accumulated driving range, voltage, current, state of charge, and temperature value; performing outlier deletion and missing value filling processing on the original data to obtain high-quality data; dividing the high-quality data into data related to the running state of the vehicle to obtain parking data and running data; wherein the vehicle state includes: flameout and non-flameout; according to the vehicle current information in the flameout state, dividing the parking data into data related to the parking state to obtain the parking charging data and the common parking data; the travel data and the parking charge data are determined as the vehicle state data.
In a possible implementation manner, according to the parking charging data, SOH calculation is performed on the battery state in the new energy automobile to obtain original SOH data, which specifically includes: according toObtaining the maximum available capacity of the battery at the current time>The method comprises the steps of carrying out a first treatment on the surface of the Wherein (1)>And->Respectively representA charge start SOC and a charge end SOC in the parking charge data; />Representing the current value; />The coulomb efficiency is represented and is related to the current multiplying factor and the temperature correction factor;t 1 and (3) witht 2 All representing different moments of the battery state; according to->Obtaining the original SOH data; wherein (1)>Is the rated capacity of the power battery pack when leaving the factory.
In a possible implementation manner, the original SOH data is subjected to joint noise reduction through a preset complete set empirical mode decomposition algorithm and an SG filtering technology to obtain accurate SOH data after noise reduction, which specifically includes: performing empirical mode decomposition of the self-adaptive noise set on the original SOH data through the complete set empirical mode decomposition algorithm to obtain a plurality of eigenmode functions and residual signals; according to a preset autocorrelation function, carrying out autocorrelation analysis on the eigenmode function and the residual signal, and determining an autocorrelation function curve; based on the distribution characteristics of the autocorrelation function curve, carrying out first noise division on the data component corresponding to the eigenmode function and the data component corresponding to the residual signal to obtain a first noise division result; wherein the first noise division result includes: a noise-containing component and a useful signal component; performing correlation analysis and calculation between the data components in the first noise division result and the battery health state, and determining the correlation degree of the data components in the first noise division result; performing secondary noise division on the data component in the first noise division result based on the association degree through the autocorrelation function to obtain a second noise division result; wherein the second noise division result includes: a full noise component, a noise dominant component, and a useful signal dominant component; removing the total noise component; performing polynomial fitting processing on the noise dominant component through the SG filtering technology to obtain a filtered noise dominant component; and carrying out data reconstruction between the filtering noise dominant component and the useful signal dominant component to obtain the accurate SOH data.
In a possible implementation manner, through the complete set empirical mode decomposition algorithm, performing empirical mode decomposition of a self-adaptive noise set on the original SOH data to obtain a plurality of eigenmode functions and residual signals, which specifically includes: according toObtaining the third under the empirical mode decomposition algorithm based on the complete setkIntrinsic mode component after +1 decomposition +.>The method comprises the steps of carrying out a first treatment on the surface of the Wherein,Ia mathematical constant;EMDis an EMD algorithm; />Is thatkSignal-to-noise ratio at the next time; />Is thatkWhite noise sequence next; according to->ObtainingkResidual component after +1 decomposition +.>The method comprises the steps of carrying out a first treatment on the surface of the Wherein (1)>Is the firstkResidual margin after secondary decomposition; according to->Obtaining the original signal->The method comprises the steps of carrying out a first treatment on the surface of the Wherein (1)>Terminating the calculated residual signal for the full set empirical mode decomposition algorithm; />Terminating the calculated plurality of eigenmode functions for the full set empirical mode decomposition algorithm; wherein the original signal is composed of the plurality of eigenmode functions and the residual signal.
In a possible implementation manner, the filtering noise dominant component is obtained by performing polynomial fitting processing on the noise dominant component through the SG filtering technology, and specifically includes: according toObtaining the filtered noise dominant component after denoising +.>The method comprises the steps of carrying out a first treatment on the surface of the Wherein N is polynomial fitting degree, < + >>Is a polynomial coefficient, i is a mathematical constant,/->Capacity data for each convolution window in the noise-dominant component.
In a possible embodiment, extracting potential health features in the vehicle state data specifically includes: performing feature analysis on the parking charging data, and extracting first potential health features; wherein the first potential health feature comprises one or more of: SOC, charging temperature, charging voltage, charging power, and charging magnification; performing feature analysis on the driving data, and extracting a second potential health feature; wherein the second potential health feature comprises one or more of: accumulating driving mileage, discharge temperature, discharge power and discharge multiplying power; wherein the potential health feature consists of the first potential health feature and the second potential health feature.
In a possible implementation manner, feature screening is performed on the potential health features according to a preset forward floating sequence algorithm to obtain an optimal health feature subset, which specifically includes: traversing the subset of health features of the potential health features; performing addition and deletion evaluation on the health features in the health feature subset through a preset evaluation function, and determining an evaluation index of the health feature subset; and screening the health feature subsets of the potential health features according to the evaluation index to obtain the optimal health feature subsets.
In a possible embodiment, generating an SOH prediction model for predicting a state of health of a battery in the new energy automobile based on the accurate SOH data and the optimal subset of health features specifically includes: determining the optimal health feature subset as an input of an LSTM neural network model and determining the accurate SOH data as an output of the LSTM neural network model; model training related to rated recursive link weights is carried out on each unit structure in the LSTM neural network model, and the SOH prediction model after training is obtained; wherein each unit structure comprises: forget gate, input gate, output gate, and cell state.
On the other hand, the embodiment of the application also provides SOH prediction equipment based on SOH joint noise reduction, which comprises: at least one processor; and a memory communicatively coupled to the at least one processor; the memory stores instructions executable by the at least one processor to enable the at least one processor to perform an SOH prediction method based on SOH joint noise reduction as described in any one of the above embodiments.
Compared with the prior art, the SOH prediction method and the SOH prediction device based on SOH combined noise reduction have the following beneficial technical effects:
according to the embodiment of the application, through accurately processing original data in the new energy automobile, the actual automobile running environment can be considered, battery degradation health characteristics can be extracted from the angles of the charging working condition and the running working condition, the CEEDMAN-SG combined noise reduction scheme is utilized to reduce noise of the original SOH data, the SOH degradation trend can be saved at the same time when noise is reduced to the greatest extent, the SFFS (forward floating sequence) algorithm is adopted to conduct characteristic screening, the optimal health characteristic subset is constructed, the influence of SOH degradation information in the original SOH data on SOH prediction precision is saved to the greatest extent, the correlation among various characteristics is utilized, namely the correlation of single characteristics cannot be used as the unique standard of characteristic selection, the SOH prediction model which is accurately predicted is constructed, and therefore the SOH prediction model is used to accurately predict the SOH of the target battery of the new energy automobile.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments described in the present application, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art. In the drawings:
fig. 1 is a flowchart of an SOH prediction method based on SOH joint noise reduction according to an embodiment of the present application;
FIG. 2 is a correlation strength comparison schematic diagram according to an embodiment of the present application;
fig. 3 is a schematic structural diagram of an SOH prediction apparatus based on SOH joint noise reduction according to an embodiment of the present application.
Detailed Description
In order to better understand the technical solutions in the present application, the following description will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only some embodiments of the present application, not all embodiments. All other embodiments, which can be made by one of ordinary skill in the art based on the embodiments herein without making any inventive effort, shall fall within the scope of the present application.
The embodiment of the application provides an SOH prediction method based on SOH joint noise reduction, as shown in fig. 1, specifically comprising steps S101-S106:
it should be noted that, in the real vehicle environment, the real vehicle running environment of the new energy automobile is very complex, and the internal state of the power battery is difficult to directly obtain after the power battery is loaded. The running condition of the power battery is complex and changeable, and the performance influence factors are numerous. In this context, battery SOH prediction for real vehicle data presents a significant challenge. The real vehicle SOH prediction faces the following problems:
(1) Influence of original data on SOH prediction in real vehicle operation environment
The real vehicle running data of the new energy automobile are collected through the vehicle-mounted sensor and transmitted by the vehicle-mounted communication equipment. The equipment has working state fluctuation in the actual use process, a large amount of missing data, error data and abnormal data exist in the original data, the data quality is seriously influenced, the data mining difficulty is increased, the reliability of model evaluation and prediction results is reduced, and when SOH prediction is carried out, data preprocessing is carried out firstly to obtain high-quality data meeting SOH analysis and modeling requirements.
(2) Effects of SOH denoising on SOH prediction
The primarily calculated SOH contains important SOH degradation information and noise information, and when SOH noise reduction is performed, the SOH degradation information in original SOH data is saved as much as possible while noise is removed to the greatest extent, so that SOH prediction accuracy is crucial.
(3) Influence of feature selection on SOH prediction
In the current feature selection research, most of the features with higher correlation are adopted to construct a feature subset to be input. However, considering that feature selection selects a feature set which is the most rich in information for SOH, rather than a single feature, some single features have high correlation with SOH, but the constituent feature subset has poor effect, some single features have low correlation with SOH, but the constituent feature subset has good effect, so the correlation magnitude of the single features cannot be used as the only standard for feature selection. How to determine the best feature subset as the input of the subsequent prediction model is important to ensure the SOH prediction accuracy.
Meanwhile, in the feature extraction and screening work, the feature screening method mainly adopted in the current research is as follows: pearson correlation coefficient method, gray correlation degree analysis method, and PCA principal component analysis method. And selecting the first few features with higher correlation by using the method to construct the feature subset to be input. However, considering that feature selection is to select a feature set which is the most rich in information for SOH, rather than a single feature, some single features have high correlation with SOH, but the constituent feature subset has poor effect, some single features have low correlation with SOH, but the constituent feature subset has good effect, so the correlation magnitude of the single features cannot be used as the only standard for feature selection, and the method cannot determine the number of input features.
In the calculation and noise reduction of the SOH, methods such as SG filtering noise reduction, EMD noise reduction, wavelet noise reduction, kalman filtering noise reduction and the like are mostly adopted, but after the noise reduction is carried out by using the method, only the main fading trend of the SOH can be saved, and the information such as the local capacity regeneration phenomenon in the original data cannot be saved, so that key information of SOH fading can be lost, and the SOH prediction precision is affected.
In the construction of a prediction model, a model-based method and a data driving method are mainly divided. The model-based method can model the battery performance parameters through an electrochemical model, an equivalent circuit model, an empirical model and the like, describe the internal degradation mechanism of the lithium ion battery, and realize SOH prediction. The data driving method does not need to master complex electrochemical reaction and physical degradation mechanisms in the battery, but utilizes algorithms such as an Artificial Neural Network (ANN), a Support Vector Machine (SVM), a correlation vector machine (RVM), particle Filtering (PF), gaussian regression (GPR) and the like to learn the mapping relation between health features and SOH to realize SOH prediction.
S101, preprocessing data of original data related to actual vehicle operation in the new energy vehicle to obtain vehicle state data. Wherein the vehicle state data includes: parking charging data and travel data.
Specifically, the new energy automobile operation data is collected through a battery management system of the new energy automobile. And uploading the operation data of the new energy automobile to a cloud server, and analyzing to obtain the original data of the actual automobile operation in the new energy automobile. Wherein, the original data at least comprises: vehicle speed, accumulated driving range, voltage, current, state of charge, and temperature values.
Further, the original data is subjected to outlier deletion and missing value filling processing, and high-quality data are obtained.
Further, the high-quality data is subjected to data division related to the running state of the vehicle, and parking data and running data are obtained. Wherein the vehicle state includes: flameout and non-flameout conditions.
Further, according to the vehicle current information in the flameout state, the parking data is divided into data related to the parking state, and the parking charging data and the common parking data are obtained.
Further, the running data and the parking charge data are determined as vehicle state data.
In one embodiment, the battery management system collects the operation data of the new energy automobile, packages the data and sends the data to the cloud server, and downloads the analyzed data from the cloud server to obtain the actual operation original data of the new energy automobile. The original data of the actual vehicle operation of the new energy automobile comprises information such as vehicle speed, accumulated driving mileage, voltage, current, state of charge, temperature value and the like. And then deleting the abnormal value in the actual vehicle running original data of the new energy vehicle, filling the missing value, and obtaining high-quality data. And dividing the data types according to the high-quality data and the vehicle state to obtain parking data and driving data. Wherein the vehicle state includes a flameout state and a non-flameout state. And then dividing the parking data into parking charging data and ordinary parking data according to whether the vehicle is in a flameout state or not, and finally obtaining the parking charging data and the driving data according to the current information during parking.
S102, SOH calculation is carried out on the battery state in the new energy automobile according to the parking charging data, and original SOH data are obtained.
In particular according toObtaining the maximum available capacity of the battery at the current time>. Wherein (1)>And->The charge start SOC and the charge end SOC in the parking charge data are indicated, respectively. />Indicating the current value. />The coulomb efficiency is expressed and is related to the current-multiplying factor and the temperature correction factor.t 1 And (3) witht 2 Each representing a different time of the battery state.
Further according toAnd obtaining original SOH data. Wherein (1)>Is the rated capacity of the power battery pack when leaving the factory.
S103, carrying out joint noise reduction on the original SOH data through a preset complete set empirical mode decomposition algorithm and an SG filtering technology to obtain accurate SOH data after noise reduction.
Specifically, through a complete set empirical mode decomposition algorithm, the original SOH data is subjected to adaptive noise set empirical mode decomposition to obtain a plurality of eigenmode functions and residual signals.
In one embodiment, the CEEMDAN is calculated as follows: first to the signal to be processedAdding different amplitude white noise sequences +.>Expressed as->，/>Is white noise figure>. Then using EMD algorithm to add white noise signal>Performing first decomposition to obtain IMF component after the first decompositionAnd residual margin->：/>And->The method comprises the steps of carrying out a first treatment on the surface of the In the foregoing, ->For signal to noise ratio +.>For the added white noise sequence, +.>Is the first decomposed IMF component. />Is the residual margin after the first decomposition.
In one embodiment, the splitting is continuedAdding white noise to the residual margin after the solution, and continuously applying EMD (empirical mode decomposition): according toObtaining the third under the empirical mode decomposition algorithm based on complete setkIntrinsic mode component after +1 decomposition +.>. Wherein,Iare mathematical constants.EMDIs an EMD algorithm. />Is thatkSignal to noise ratio at the next time. />Is thatkWhite noise sequence next. Then will be according to +.>ObtainingkResidual component after +1 decomposition +.>. Wherein (1)>Is the firstkResidual margin after sub-decomposition. And stopping the CEEMDAN algorithm until the residual signal does not exceed two extreme points and the decomposition cannot be continued. At this time, the original signal is decomposed into I eigenmode components IMF and residual signals +.>: according to->Obtaining the original signal->. Wherein (1)>The calculated residual signal is terminated for the complete set of empirical mode decomposition algorithms. />Terminating the computed plurality of eigenmode functions for the complete set of empirical mode decomposition algorithms. The original signal is composed of a plurality of eigenmode functions and a residual signal.
Further, according to a preset autocorrelation function, the intrinsic mode function and the residual signal are subjected to autocorrelation analysis, and an autocorrelation function curve is determined.
In one embodiment, the autocorrelation function is used to describe a random signalx(t) The linear correlation function of the correlation degree between values at any two different moments, the autocorrelation function curve of the noise signal has an irregular state, the maximum value is reached at the middle point, the maximum value is smaller, and the autocorrelation function values at the two sides rapidly attenuate to 0; the autocorrelation function curve of the dominant component of the useful signal can be regarded as a stable signal in a short period, and has the characteristics of certain periodicity and slow decay.
As a possible implementation, the components are divided into noise-containing components and useful signal dominant components by analyzing the autocorrelation function curves of the components to determine how much noise is contained in the components themselves. Specifically, the autocorrelation function calculation method is as follows:wherein->As a function of a random signal template.kIs the delay variable of the autocorrelation function.NIs a sampling point.
Further, based on the distribution characteristics of the autocorrelation function curve, the data component corresponding to the eigenmode function and the data component corresponding to the residual signal are subjected to first noise division, and a first noise division result is obtained. Wherein, the first noise division result includes: containing noise components and useful signal components.
Further, correlation analysis calculation is carried out between the data components in the first noise division result and the battery health state, and the correlation degree of the data components in the first noise division result is determined.
In one implementation, fig. 2 is a schematic diagram of correlation strength comparison provided in an embodiment of the present application, as shown in fig. 2, in the correlation analysis calculation, the Pearson correlation coefficient method is a linear correlation coefficient, which is a ratio of covariance to standard deviation, and the correlation between each component and SOH can be shown by calculation, and according to the correlation size, whether each component contains noise or not can be determined, whether the IMF and SOH correlation coefficient is greater than 0.5 is determined as a dominant component of the useful signal, noise components greater than 0.2 and less than 0.5 are divided into dominant noise components, and noise components less than 0.2 are divided into full noise components. The correlation calculation formula is as follows:wherein X is i And Y i Respectively representing the ith data component and SOH,>and->Mean values of data component and SOH, respectively, +.>Indicating the degree of association between X and Y. According to the Pearson correlation coefficient absolute value range, the correlation intensity can be divided into the following 5 grades, and then the correlation degree of the data components in the first noise division result is determined.
Further, the data component in the first noise division result is subjected to secondary noise division through an autocorrelation function and based on the association degree, and a second noise division result is obtained. Wherein the second noise division result includes: a full noise component, a noise dominant component, and a useful signal dominant component.
Further, the full noise component is removed. And performing polynomial fitting processing on the noise dominant component through SG filtering technology to obtain a filtered noise dominant component.
As a possible embodiment, according toObtaining the filtered noise dominant component after denoising +.>. Wherein N is polynomial fitting degree, < + >>Is a polynomial coefficient, i is a mathematical constant,/->Capacity data for each convolution window in the noise-dominant component. The SG filtering technology is a low-pass filter, and is a method for realizing best fitting by using a polynomial through a moving window in a time domain. The method carries out convolution calculation on the time sequence, and carries out polynomial fitting on the capacity data in each convolution window, thereby eliminating high-frequency noise in the capacity data and realizing signal smoothing.
Further, data reconstruction is carried out between the filtered noise dominant component and the useful signal dominant component, and accurate SOH data is obtained.
S104, extracting potential health features in the vehicle state data.
Specifically, it is desirable to perform a feature analysis on the parking charge data and extract a first potential health feature. Wherein the first potential health feature comprises one or more of: SOC, charging temperature, charging voltage, charging power, and charging magnification.
Further, a feature analysis is performed on the travel data and a second potential health feature is extracted. Wherein the second potential health feature comprises one or more of: the driving range, the discharge temperature, the discharge power and the discharge multiplying power are accumulated. The potential health feature is comprised of a first potential health feature and a second potential health feature.
And S105, carrying out feature screening on the potential health features according to a preset forward floating sequence algorithm to obtain an optimal health feature subset.
Specifically, a subset of the health features of the potential health features is traversed first.
Further, adding and deleting the health features in the health feature subset through a preset evaluation function, and determining the evaluation index of the health feature subset.
Further, screening the health feature subsets of the potential health features according to the evaluation index to obtain the optimal health feature subsets.
In one embodiment, the set of potential health features is traversed using an SFFS (Forward float sequence) algorithm, and then the best health feature subset is determined from an evaluation function. The evaluation index in the evaluation function is the number of elements/total number of elements of the accury=classification pair, and the SFFS algorithm is specifically implemented as follows:
(1) the potential health feature total set of the battery is Y, and the SFFS initial feature search subset is
(2) Sequentially adding health features to the initial search feature subset to form a new feature subsetAnd calculate its evaluation function if +.>The feature is added until no pluggable health features exist, the search completion feature subset is defined as +.>。
(3) From the slaveSequentially removing a feature and inputting an evaluation function if +.>The feature is subtracted until no deletable feature exists.
(4) Checking whether all the features are traversed, and if so, taking the feature subset (the best health feature subset) at the moment as a final result; otherwise, repeating the step (2) and the step (3).
S106, generating an SOH prediction model for predicting the state of health of the battery in the new energy automobile based on the accurate SOH data and the optimal health feature subset.
Specifically, the optimal health feature subset is determined as the input of the LSTM neural network model and the precise SOH data is determined as the output of the LSTM neural network model.
Further, model training related to rated recursive link weights is carried out on each unit structure in the LSTM neural network model, and a trained SOH prediction model is obtained. Wherein each unit structure includes: forget gate, input gate, output gate, and cell state.
In one embodiment, the structural calculation formula of each part of the LSTM unit in the LSTM neural network model is as follows: comprising、/>、/>、、/>And->。
In the above-mentioned formulae, the first and second light-emitting elements,、/>、/>、/>and->Representing forget gate, input gate, output gate, intermediate output and cell state respectively; />Rated recursive link weight representing its relative gate,/->Is the corresponding bias parameter. />And->The sigmoid activating function and the hyperbolic tangent function are respectively represented, the weight and the bias parameters in the LSTM network are updated by adopting an Adam optimizer, a trained SOH prediction model is obtained after model training, and then the SOH prediction model is used for accurately predicting the SOH of the target battery of the new energy automobile.
In addition, the embodiment of the application further provides an SOH prediction device based on SOH joint noise reduction, as shown in fig. 3, an SOH prediction device 300 based on SOH joint noise reduction specifically includes:
at least one processor 301. And a memory 302 communicatively coupled to the at least one processor 301. Wherein the memory 302 stores instructions executable by the at least one processor 301 to enable the at least one processor 301 to perform:
carrying out data preprocessing on original data related to actual vehicle operation in the new energy vehicle to obtain vehicle state data; wherein the vehicle state data includes: parking charging data and driving data;
SOH calculation is carried out on the battery state in the new energy automobile according to the parking charging data, and original SOH data are obtained;
the original SOH data is subjected to joint noise reduction through a preset complete set empirical mode decomposition algorithm and an SG filtering technology, so that accurate SOH data after noise reduction is obtained;
extracting potential health features in the vehicle state data;
according to a preset forward floating sequence algorithm, carrying out feature screening on potential health features to obtain an optimal health feature subset;
based on the accurate SOH data and the optimal health feature subset, an SOH prediction model for predicting the state of health of the battery in the new energy automobile is generated.
According to the SOH prediction method and the SOH prediction device based on SOH combined noise reduction, through accurately processing original data in a new energy automobile, battery degradation health features can be extracted from the angles of a charging working condition and a driving working condition by considering the running environment of the new energy automobile, and a CEEDMAN-SG combined noise reduction scheme is utilized to reduce noise of the original SOH data, so that SOH degradation trend can be saved while noise is reduced to the greatest extent, an SFFS (forward floating sequence) algorithm is adopted to conduct feature screening, an optimal health feature subset is constructed, the influence of SOH degradation information in the original SOH data on SOH prediction precision is furthest saved, the correlation among various features is utilized, namely the correlation of a single feature cannot be used as a unique standard of feature selection, and therefore the SOH of a target battery of the new energy automobile is accurately predicted by using the SOH prediction model.
All embodiments in the application are described in a progressive manner, and identical and similar parts of all embodiments are mutually referred, so that each embodiment mainly describes differences from other embodiments. In particular, for the apparatus embodiments, since they are substantially similar to the method embodiments, the description is relatively simple, and reference is made to the description of the method embodiments in part.
The foregoing describes specific embodiments of the present application. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
The foregoing is merely exemplary of the present application and is not intended to limit the present application. Various modifications and variations of the embodiments of the present application will be apparent to those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the embodiments of the present application should be included in the scope of the claims of the present application.
Claims (10)
1. An SOH prediction method based on SOH joint noise reduction, the method comprising:
carrying out data preprocessing on original data related to actual vehicle operation in the new energy vehicle to obtain vehicle state data; wherein the vehicle state data includes: parking charging data and driving data;
according to the parking charging data, SOH calculation is carried out on the battery state in the new energy automobile, and original SOH data are obtained;
the original SOH data is subjected to joint noise reduction through a preset complete set empirical mode decomposition algorithm and an SG filtering technology, so that accurate SOH data after noise reduction is obtained;
extracting potential health features in the vehicle state data;
according to a preset forward floating sequence algorithm, carrying out feature screening on the potential health features to obtain an optimal health feature subset;
and generating an SOH prediction model for predicting the state of health of the battery in the new energy automobile based on the accurate SOH data and the optimal health feature subset.
2. The SOH prediction method based on SOH joint noise reduction according to claim 1, wherein the method is characterized by performing data preprocessing on raw data related to actual vehicle operation in a new energy vehicle to obtain vehicle state data, and specifically comprises:
collecting new energy automobile operation data through the battery management system of the new energy automobile; uploading the new energy automobile operation data to a cloud server, and analyzing to obtain the original data of the actual automobile operation in the new energy automobile; wherein the raw data comprises at least: vehicle speed, accumulated driving range, voltage, current, state of charge, and temperature value;
performing outlier deletion and missing value filling processing on the original data to obtain high-quality data;
dividing the high-quality data into data related to the running state of the vehicle to obtain parking data and running data; wherein the vehicle state includes: flameout and non-flameout;
according to the vehicle current information in the flameout state, dividing the parking data into data related to the parking state to obtain the parking charging data and the common parking data;
the travel data and the parking charge data are determined as the vehicle state data.
3. The SOH prediction method based on SOH joint noise reduction according to claim 1, wherein SOH calculation is performed on the battery state in the new energy automobile according to the parking charging data, so as to obtain original SOH data, and specifically comprising:
according toObtaining the maximum available capacity of the battery at the current time>The method comprises the steps of carrying out a first treatment on the surface of the Wherein,and->Respectively represent the stopA charge start SOC and a charge end SOC in the vehicle charge data; />Representing the current value; />The coulomb efficiency is represented and is related to the current multiplying factor and the temperature correction factor;t 1 and (3) witht 2 All representing different moments of the battery state;
according toObtaining the original SOH data; wherein (1)>Is the rated capacity of the power battery pack when leaving the factory.
4. The SOH prediction method based on SOH joint noise reduction according to claim 1, wherein the method comprises the steps of performing joint noise reduction on the original SOH data through a preset complete set empirical mode decomposition algorithm and an SG filtering technology to obtain accurate SOH data after noise reduction, and specifically comprises the following steps:
performing empirical mode decomposition of the self-adaptive noise set on the original SOH data through the complete set empirical mode decomposition algorithm to obtain a plurality of eigenmode functions and residual signals;
according to a preset autocorrelation function, carrying out autocorrelation analysis on the eigenmode function and the residual signal, and determining an autocorrelation function curve;
based on the distribution characteristics of the autocorrelation function curve, carrying out first noise division on the data component corresponding to the eigenmode function and the data component corresponding to the residual signal to obtain a first noise division result; wherein the first noise division result includes: a noise-containing component and a useful signal component;
performing correlation analysis and calculation between the data components in the first noise division result and the battery health state, and determining the correlation degree of the data components in the first noise division result;
performing secondary noise division on the data component in the first noise division result based on the association degree through the autocorrelation function to obtain a second noise division result; wherein the second noise division result includes: a full noise component, a noise dominant component, and a useful signal dominant component;
removing the total noise component; performing polynomial fitting processing on the noise dominant component through the SG filtering technology to obtain a filtered noise dominant component;
and carrying out data reconstruction between the filtering noise dominant component and the useful signal dominant component to obtain the accurate SOH data.
5. The SOH prediction method based on SOH joint noise reduction according to claim 4, wherein the empirical mode decomposition of the adaptive noise set is performed on the original SOH data by the complete set empirical mode decomposition algorithm to obtain a plurality of eigenmode functions and residual signals, and specifically comprising:
according toObtaining the third under the empirical mode decomposition algorithm based on the complete setkIntrinsic mode component after +1 decomposition +.>The method comprises the steps of carrying out a first treatment on the surface of the Wherein,Ia mathematical constant;EMDis an EMD algorithm;is thatkSignal-to-noise ratio at the next time; />Is thatkWhite noise sequence next;
according toObtainingkResidual component after +1 decomposition +.>The method comprises the steps of carrying out a first treatment on the surface of the Wherein (1)>Is the firstkResidual margin after secondary decomposition;
according toObtaining the original signal->The method comprises the steps of carrying out a first treatment on the surface of the Wherein (1)>Terminating the calculated residual signal for the full set empirical mode decomposition algorithm; />Terminating the calculated plurality of eigenmode functions for the full set empirical mode decomposition algorithm;
wherein the original signal is composed of the plurality of eigenmode functions and the residual signal.
6. The SOH prediction method based on SOH joint noise reduction according to claim 4, wherein the filtering noise dominant component is obtained by performing polynomial fitting on the noise dominant component by using the SG filtering technique, and specifically comprises:
according toObtaining the filtered noise dominant component after denoising +.>The method comprises the steps of carrying out a first treatment on the surface of the Wherein N is polynomial fitting degree, < + >>Is a polynomial coefficient, i is a mathematical constant,/->Capacity data for each convolution window in the noise-dominant component.
7. The SOH prediction method based on SOH joint noise reduction according to claim 1, wherein extracting potential health features in the vehicle state data specifically comprises:
performing feature analysis on the parking charging data, and extracting first potential health features; wherein the first potential health feature comprises one or more of: SOC, charging temperature, charging voltage, charging power, and charging magnification;
performing feature analysis on the driving data, and extracting a second potential health feature; wherein the second potential health feature comprises one or more of: accumulating driving mileage, discharge temperature, discharge power and discharge multiplying power;
wherein the potential health feature consists of the first potential health feature and the second potential health feature.
8. The SOH prediction method based on SOH joint noise reduction according to claim 1, wherein the feature screening is performed on the potential health features according to a preset forward floating sequence algorithm to obtain an optimal health feature subset, and specifically comprises:
traversing the subset of health features of the potential health features;
performing addition and deletion evaluation on the health features in the health feature subset through a preset evaluation function, and determining an evaluation index of the health feature subset;
and screening the health feature subsets of the potential health features according to the evaluation index to obtain the optimal health feature subsets.
9. The SOH prediction method based on SOH joint noise reduction according to claim 1, wherein generating the SOH prediction model for predicting the state of health of the battery in the new energy automobile based on the accurate SOH data and the optimal health feature subset specifically comprises:
determining the optimal health feature subset as an input of an LSTM neural network model and determining the accurate SOH data as an output of the LSTM neural network model;
model training related to rated recursive link weights is carried out on each unit structure in the LSTM neural network model, and the SOH prediction model after training is obtained; wherein each unit structure comprises: forget gate, input gate, output gate, and cell state.
10. An SOH prediction apparatus based on SOH joint noise reduction, the apparatus comprising:
at least one processor; the method comprises the steps of,
a memory communicatively coupled to the at least one processor; wherein,
the memory stores instructions executable by the at least one processor to enable the at least one processor to perform an SOH prediction method based on SOH joint noise reduction according to any one of claims 1-9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410121852.XA CN117648557A (en) | 2024-01-30 | 2024-01-30 | SOH prediction method and device based on SOH combined noise reduction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410121852.XA CN117648557A (en) | 2024-01-30 | 2024-01-30 | SOH prediction method and device based on SOH combined noise reduction |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117648557A true CN117648557A (en) | 2024-03-05 |
Family
ID=90043729
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410121852.XA Pending CN117648557A (en) | 2024-01-30 | 2024-01-30 | SOH prediction method and device based on SOH combined noise reduction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117648557A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109871733A (en) * | 2018-09-27 | 2019-06-11 | 南京信息工程大学 | A kind of adaptive sea clutter signal antinoise method |
| CN112766127A (en) * | 2021-01-12 | 2021-05-07 | 南京信息工程大学 | Thundercloud point charge positioning method based on complementary set modal decomposition and SG filtering |
| CN116744843A (en) * | 2020-11-24 | 2023-09-12 | 佐治亚技术研究公司 | Systems and methods for measuring hemodynamic parameters with wearable cardiovascular sensing |
| CN116794543A (en) * | 2023-06-07 | 2023-09-22 | 广东工业大学 | Lithium battery performance prediction method based on GBLS boost multitask learning model |
| US20230333172A1 (en) * | 2022-04-13 | 2023-10-19 | Beijing Institute Of Technology | Method and device for capacity degradation prediction of lithium-ion battery |
-
2024
- 2024-01-30 CN CN202410121852.XA patent/CN117648557A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109871733A (en) * | 2018-09-27 | 2019-06-11 | 南京信息工程大学 | A kind of adaptive sea clutter signal antinoise method |
| CN116744843A (en) * | 2020-11-24 | 2023-09-12 | 佐治亚技术研究公司 | Systems and methods for measuring hemodynamic parameters with wearable cardiovascular sensing |
| CN112766127A (en) * | 2021-01-12 | 2021-05-07 | 南京信息工程大学 | Thundercloud point charge positioning method based on complementary set modal decomposition and SG filtering |
| US20230333172A1 (en) * | 2022-04-13 | 2023-10-19 | Beijing Institute Of Technology | Method and device for capacity degradation prediction of lithium-ion battery |
| CN116794543A (en) * | 2023-06-07 | 2023-09-22 | 广东工业大学 | Lithium battery performance prediction method based on GBLS boost multitask learning model |
Non-Patent Citations (4)
| Title |
|---|
| FEI XIA等: "State-of-Health Prediction for Lithium-Ion Batteries Based on Complete Ensemble Empirical Mode Decomposition with Adaptive Noise-Gate Recurrent Unit Fusion Model", ENERGY TECHNOLOGY, 31 December 2022 (2022-12-31), pages 1 - 16 * |
| 王茂男: "基于深度学习的锂离子电池剩余寿命预测方法研究", 中国优秀硕士学位论文全文数据库工程科技Ⅱ辑, no. 03, 15 March 2022 (2022-03-15), pages 042 - 3565 * |
| 王辉等: "基于改进ICEEMDAN的航班延误短期预测研究", 中国民航大学学报, vol. 38, no. 4, 31 August 2020 (2020-08-31), pages 37 - 42 * |
| 王鹏: "基于数据驱动的动力电池健康状态评估与预测研究", 万方学位论文数据库, 1 December 2023 (2023-12-01), pages 1 - 116 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110457789B (en) | Lithium ion battery residual life prediction method | |
| CN112034356B (en) | GP-UKF-based online SOH estimation method for electric vehicle power battery | |
| CN112098848B (en) | Battery discharge capacity prediction method and system and readable storage medium | |
| CN112904219B (en) | Big data-based power battery health state prediction method | |
| CN112540317A (en) | Battery health state estimation and residual life prediction method based on real vehicle data | |
| CN113866642A (en) | Lithium ion battery fault diagnosis method based on gradient lifting tree | |
| CN112098874A (en) | Lithium ion battery electric quantity prediction method considering aging condition | |
| CN112686380A (en) | Neural network-based echelon power cell consistency evaluation method and system | |
| CN114384435A (en) | WSA-LSTM algorithm-based self-adaptive prediction method for residual service life of new energy automobile power battery | |
| CN115097312A (en) | Lithium ion battery fusion life prediction model combining data driving model and empirical model | |
| CN115586444A (en) | Lithium battery residual life prediction method based on VMD and BP neural network | |
| CN109299552A (en) | A kind of appraisal procedure and its assessment system of battery power status | |
| CN116298936A (en) | Intelligent lithium ion battery health state prediction method in incomplete voltage range | |
| CN116106761A (en) | Real-time lithium ion battery electric quantity estimation method based on typical correlation analysis | |
| CN113406500B (en) | Method for estimating residual electric quantity of power lithium battery | |
| CN116653608B (en) | Electric automobile charging protection and control method, device and storage medium | |
| Zhang et al. | Remaining useful life prediction of lithium-ion batteries based on TCN-DCN fusion model combined with IRRS filtering | |
| CN116973761A (en) | Method for predicting remaining service life of power battery based on particle swarm optimization | |
| CN117471320A (en) | Battery state of health estimation method and system based on charging fragments | |
| CN116754960A (en) | Cloud SOH estimation and residual life prediction method, device and program for sodium ion battery | |
| CN117648557A (en) | SOH prediction method and device based on SOH combined noise reduction | |
| CN116859259A (en) | Battery health state estimation method based on integrated learning and real vehicle big data | |
| CN115808627A (en) | Lithium battery SOH prediction method and device | |
| Zhao et al. | Lithium-ion battery state-of-health estimation method using isobaric energy analysis and pso-lstm | |
| CN110750848A (en) | Method for estimating remaining life of software-hardware degradation system by considering software operation |
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 |