CN109934408A - A kind of application analysis method carrying out automobile batteries RUL prediction based on big data machine learning - Google Patents
A kind of application analysis method carrying out automobile batteries RUL prediction based on big data machine learning Download PDFInfo
- Publication number
- CN109934408A CN109934408A CN201910201346.0A CN201910201346A CN109934408A CN 109934408 A CN109934408 A CN 109934408A CN 201910201346 A CN201910201346 A CN 201910201346A CN 109934408 A CN109934408 A CN 109934408A
- Authority
- CN
- China
- Prior art keywords
- data
- battery
- model
- rul
- soh
- 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.)
- Withdrawn
Links
Abstract
Patent is related to a kind of method for carrying out batteries of electric automobile RUL (Remaining Useful Life) prediction based on big data machine learning, and this method is made of corresponding application architecture, process, computation model.This method is first to the battery real time data acquired in batteries of electric automobile operational process, and electric car vehicle others operation data, carry out data preparation and cleaning, and characterization is carried out to data, model and training verification algorithm are established by big data machine learning, wherein modeling has mainly used non-linear hybrid algorithm model and survival model, and different angle is carried out to result and is evaluated and optimized, to establish the model of batteries of electric automobile RUL prediction, optimize the maintenance and replacement of battery, improve the safety indexes of car owner, reach the balance of system performance and economic benefit.
Description
Technical field
The present invention relates to a kind of application analysis methods that automobile batteries RUL prediction is carried out based on big data machine learning, answer
Field is the prediction for carrying out the electric car vehicle battery retired time and the value assessment of battery pack.
Background technique
As electric car is in the popularization of China and the application of car networking technology, more and more electric cars, which enter, to disappear
Expense person market and travelling data is acquired according to national standard (GBT32960) in real time.Power battery is as the dynamic of electric car
Power source, with the increase of charge and discharge number and mileage travelled, the capacity of battery is constantly decayed, this reaction is typical dynamic
Nonlinear electro-chemical systems, inner parameter is difficult to measure when application on site, and degenerate state identification and state estimation are still
There are huge challenges.
The remaining life (RUL, Remaining Useful Life) of battery, refers to that under certain condition, battery makes
With the service life of remaining battery after a period of time, for judging battery health.RUL not only with the electrification of battery itself
System is related to battery manufacturing process, also related to the working environment of vehicle driving-cycle and internal battery pack.
The prediction for carrying out battery RUL using a kind of method using machine learning herein is depending on electric car number
In the case of the long period of acquisition, excavated from the rating blocks and Condition Monitoring Data (voltage, electric current, temperature, SOC etc.) of battery
Wherein implicit cell health state information and its development law, extrapolate RUL on the basis of battery SOH.
Summary of the invention
In order to solve this problem, the present invention provides a kind of batteries of electric automobile RUL prediction techniques, which is characterized in that
The described method includes: step 001 data preparation step, obtains and uses relevant data, the electronic vapour to batteries of electric automobile
Vehicle battery includes the use data of breakdown maintenance data and battery using related data;Wherein, the breakdown maintenance data include
The mantenance data of data record and/or battery before cell malfunctions;The use data of the battery are included in normal use
When relevant to battery battery itself data and vehicle condition data;The breakdown maintenance data, the use data of battery are equal
It is the stream data based on time series;And the t moment RUL that empirically formula is calculatedt;Step 002 data preparation
Step carries out cleaning using relevant data to the batteries of electric automobile and the related data after cleaning is based on time quantum
Carry out data building;The data cleansing includes, using the average value or median or neighbor interpolation for taking a trip variable
Carry out the assignment of vacant variable;Check that data are using the threshold value of each variable of related data by setting batteries of electric automobile
No meet the requirements will exceed the data of normal range (NR) and be deleted or be corrected;Related data is used by setting batteries of electric automobile
It is mutual constraint and dependence, data unreasonable or conflicting in logic are deleted or are corrected;The data
Building includes sequentially in time integrating the data collected;Step 003 data characterization step, to passing through data
It arranges the data that step obtains to summarize and extract, obtains the data after characterizing;Summary and extraction for data include
Polymerization is rolled, the rolling polymerization refers to one time window of setting, calculates poly- in the time window in scheduled variable
Conjunction value, the polymerizing value can be the summation of data, average value either standard deviation;The summary and extraction further include by feature
Variable is extended, and the extension includes to initial characteristics variable according to the corresponding number of mean value increase for rolling polymerization, and
Corresponding number is increased according to the standard deviation for rolling polymerization to initial characteristics variable;Step 004 target determines step, calculates for learning
The RUL value of habit;For the data that the step 003 obtains, the calculating of SOH is carried out, the SOH is target value, the step
004 includes: step 1: battery master data is obtained, for calculating the SOH in second step and third step, the fundamental packets
It includes: cycle-index and capacity attenuation under battery capacity, the mapping table of battery capacity and temperature and battery ideal operating condition
Mapping table;Step 2: the SOHt of statistics t moment, counts handling capacity since when battery factory brings into operationWherein △ t is sampling time interval, ItElectric current when for charge and discharge, I when chargingtIt is negative, I when electric dischargetFor
Just;According to currently practical remaining capacity, temperature, battery discharge rates, by the capacity and vs. temperature of looking into the first step
Table obtains attenuation coefficient P;Goodput isThe ideally charge and discharge cycles of battery at this time
Number isThen N is found according to cycle-index and capacity attenuation Cap Fade CurvetCorresponding Capt, t
The SOH at moment is represented byWherein Cap_BOL is battery capacity;Step 3: calculating RUL, cycles left
Times N 1 is calculated as follows: N1=N2-Nt, wherein N2 is that initial cycle-index is infused;Residue is using number of days RUL according to such as
Lower formula calculates:Wherein RUL is the remaining number of days used, NtTo add up access times, as initial cycle time
Several and cycles left number interpolation, D are to add up to have used number of days;RUL obtained above is the learning objective of subsequent step;
S005 data calculate step, battery RUL prediction model are established based on the data after characterization, with the SOH of t momenttIt is right as Y
Each data are from time enterprising row label;The data obtained after step S001, S002 and S003 are set as x, establish model Y
=f (x), the function between x and Y that wherein f () is learnt for machine based on big data;The input of the model be time t with
And the data of t moment acquisition, the output of model are t moment battery SOHt;The f () uses nonlinear mixed-effect model and life
Model is deposited to establish;Wherein, nonlinear mixed-effect model algorithm is Y=f (x+ Φ)+e, and wherein f () is nonlinear function,
A in Φ=A β+Bb, B are the matrix of design, and β is fixed effect parameter vector and b is stochastic effects parameter vector, e be error to
Amount, wherein β is to predict relevant fixed effect data for battery SOH in input data x, and b is then to predict not phase for SOH
The stochastic effects data of pass;The estimation of parameter A and B are complete by the iteration that pseudo- data step and linear hybrid effect walk between two steps
At can be solved respectively using Gauss-Newton iterative method and EM algorithm;Wherein, the survival model algorithm isWherein t is the time that uses of battery, and x is the data acquired based on time series, and f (x) is research pair
As the probability density function that life span is distributed, S (t) is the probability that research object life span is longer than t;The algorithm model of RUL
For Y=f (S (t), x), wherein f () algorithm model for survival;In this step S005, nonlinear mixed-effect model and existence
Model carries out parallel;Step 006 trains verification step, the model of foundation is trained and is verified to optimize the model;It is described
Training verification step preferably includes cross validation, determines optimal data classification based on the experimental result;Step 007 algorithm is commented
Estimate step, the result that prediction models various in the step 006 obtain is compared with RUL obtained in step 004, assesses
Prediction result of the various prediction model data under algorithms of different, selects optimal prediction model.
Preferably, it after completing data building, is assessed and is corrected to based on the data that time quantum is constructed;Institute
Commentary estimates that there are those of mistake data including garbled data itself;After evaluation, the wrong data is corrected;
The correction includes: to set 0 for missing values for missing values;For exceptional value, 0 is set by negative value;For the time cycle
The numerical value of mistake should clearly obtain time cycle, adjustment and again operation data;It is bright for calculating the numerical value of specification mistake
True bore adjustment and again operation data.
Preferably, in the step 006, when data a kind of in sample only have a small amount of training sample, by that will lack
Several sample datas synthesizes new minority class sample data to carry out the training of model;To each minority class sample A, from it away from
From a sample B is selected in arest neighbors at random, the distance is calculated according to the distance in time and variogram, then in A
Random selection is a little used as newly synthesized minority class sample on line between B;By constantly synthesizing, by a small amount of sample
A becomes have multidata sample A+.
Preferably, the second step of the step 004 further include: capture SOC from 20% or less be charged to 100% characteristic
According to examining the learning objective SOHt;From 20% or less be charged to 100% during, the information for beginning of charging: time t0,
SOC0,;Charge the information terminated: time t1, SOC1=100, temperature T1, voltage V1;Battery capacity: Cap0=∑ is calculated firsttIt* △ t, wherein △ t is acquisition time interval, to electric current I in charging processtTemporally t is integrated;And it does primary conversion and obtains electricity
Tankage Cap1, formula are as follows:
Influence of the temperature T1 for battery capacity is calculated again, obtains final modified battery capacity Cap2 are as follows:
Wherein, coefficient q is searched according to the mapping table of battery capacity and temperature;
Then, voltage consistency when being full of according to monomer voltage is very poor with the assessment of voltage standard difference;If consistency
It is good, then obtain SOH when this is full of are as follows:
By above-mentioned SOH numerical value to the SOH of acquisitiontIt is verified, the SOH that then will be upcheckedtAs calculating study
The numerical value of target RUL.
This method has determined the problem of battery RUL prediction in batteries of electric automobile management, carries out for the key problem
The acquisition and calibration of data and progress Data Integration and Feature Engineering, explicit data defines and carries out preliminary treatment, by pre-
The rule of definition carries out the definition of feature and label.It is finally to carry out model training and assessment, is imported by data, utilize machine
The different models of study select algorithms of different to carry out matching verifying, and are issued, and become the product of structuring, and with when
Between accumulate and data rich, the forecasting accuracy of model can constantly be promoted.
Detailed description of the invention
Fig. 1 is batteries of electric automobile RUL prediction embodiment;
Fig. 2 is system structure diagram of the invention;
Fig. 3 is big data machine learning block diagram of the invention;
Fig. 4 is that polymerization schematic diagram is rolled in the present invention.
Specific embodiment
Specific implementation of the patent mode is described in detail in conjunction with the following figure, it should be pointed out that the specific embodiment party
Formula is only the citing to optimal technical scheme of the present invention, can not be interpreted as limiting the scope of the invention.
Fig. 1-4 shows the step of one of this patent specific embodiment batteries of electric automobile RUL prediction.Wherein:
Step S001 data preparation step obtains and uses relevant data to batteries of electric automobile.
In this step, the data of the batteries of electric automobile include the monitoring data of electric car, monitoring data every ten
Second acquisition is primary, in the different whole vehicle states of electric car, such as in traveling, charging process, can all generate.The battery
Monitoring data includes battery itself data relevant to battery and vehicle condition data in normal use, altogether more than 200
A data variable.
The stream data that time series is all based on using data of the battery, including carry out the relevant electricity of machine learning
Stream, voltage, temperature, remaining capacity (SOC) etc..Relevant data content is as shown in the table.
S002 data preparation step, to the batteries of electric automobile using relevant data carry out cleaning and will be after cleaning
The batteries of electric automobile is based on time quantum using relevant data and carries out data building.
In the present embodiment, due to being mainly based upon data processing realization, guarantee that the data of high quality are conducive to mention
The accuracy of high result, it is therefore desirable to which data preparation is carried out to the data of acquisition.The data preparation first has to carry out data
Cleaning, the present invention have formulated corresponding cleaning rule and have converted data of low quality to the data for meeting quality of data requirement.
Cleaning rule includes:
Vacant assignment: battery data is in transmission process, it is easy to and occurring to exchange causes variable to lack, in the present invention,
The main assignment that vacant variable is carried out using the average value or median or neighbor interpolation that take a trip variable.
Mistake value removal: by setting batteries of electric automobile using the reasonable value range of each variable of related data, i.e.,
Threshold value checks data whether meet the requirement, and the data that will exceed normal range (NR) are deleted or corrected.
Crosscheck: by setting batteries of electric automobile using the mutual constraint of related data and dependence, by logic
Upper unreasonable or conflicting data are deleted or are corrected.
It cleans after data, data building, i.e., the data that will be collected according to the sequence of time is carried out based on time quantum
It is integrated.Time quantum can be based on millisecond, second, minute etc., and time quantum can be inconsistent with the frequency of collection.
After completing data building, need to be assessed and corrected to based on the data that time quantum is constructed.Institute
Commentary is estimated including filtering out wrong data, i.e., there are those of mistake data for data itself.E.g., including but be not limited to, it lacks
Value, exceptional value, time cycle mistake and calculating specification mistake etc..After evaluation, the wrong data is corrected.Example
Such as missing values, the value that null will be present is set as 0, supplements the data of missing;For exceptional value, 0 is set by negative value, is kept away
Exempt from occur mistake in training process;For the numerical value of time cycle mistake, the time cycle should be clearly obtained, adjusts and transports again
Row data;For calculating the numerical value of specification mistake, bore adjustment and again operation data are specified.
The data obtained by data preparation step are summarized and are extracted by S003 data characterization step, obtain special
Data after signization.
Due to needing to be handled data and calculated in subsequent processing step, for ease of calculation with identification data
Feature, it is necessary first to reduced data is characterized in order to show the various features of the data consequently facilitating meter
It calculates and identifies.
It in this step, include rolling polymerization for the summary of data and extraction.The rolling polymerization refers to setting one
Time window, calculates the polymerizing value in scheduled variable in the time window, and the polymerizing value can be the summation of data, put down
Mean value either standard deviation.As shown in figure 4, such as t1 node, setting time window are 3, its rolling polymerization is exactly to calculate t1 section
Summation, mean value or the standard deviation of point and 3 nodes between the t1 node.
In this step, more preferable in order to provide learning algorithm, even additional study and predictive ability need
More multivariate data, invention are summarized and are extracted from the battery data based on time series, thus by initial S001
Characteristic variable is extended.For example, when there is 126 characteristic variables in step S001, in this example, the number being extended
According to mainly two classes: first major class is to increase 126-2=124 according to the mean value for rolling polymerization to initial 126 characteristic variables
It is a;Second class is to increase 126-2=124 according to the standard deviation for rolling polymerization to 126 initial characteristic variables;So most
The variable obtained afterwards is 126+124+124=374.This makes it possible to provide more multivariate data, calculated to be conducive to study
Method provides more preferable and predictive ability.
S004 target determines step, calculates the RUL value for study
Need to carry out target value, the i.e. calculating of SOH after characterizing for the acquisition and recording of each battery data.
Step 1: battery master data is obtained, for calculating the SOH in second step and third step
The master data, is referred to as factory data, comprising: battery capacity (Cap_BOL), battery capacity and temperature
Mapping table and battery ideal operating condition under cycle-index and capacity attenuation mapping table.
The battery capacity can be provided by Battery Plant, because general battery can all mark capacity;Battery capacity and temperature
The mapping table of degree can also be provided by Battery Plant, if can not provide, remove study temperature T and battery capacity Cap by data
Relation table (when charging SOC from 20% or less to 100%);The corresponding relationship of cycle-index and capacity attenuation under battery ideal operating condition
Ideal situation in table refers to that battery 1C discharges, and 0.5C charges (wherein the C refers to battery discharge rates), in 25 DEG C of ring
Under border, 0%SOC, a charge and discharge zooming circulation primary are discharged into.
Step 2: the SOH of statistics t moment
Handling capacity is counted since when battery factory brings into operationWherein △ t is between the sampling time
Every containing all charging and discharging processes, ItElectric current when for charge and discharge, I when chargingtIt is negative, I when electric dischargetIt is positive.Due to
25 DEG C are not in during the actual operation of battery, 1C electric discharge, 0.5C charging is completely full of and puts ideally, so
Need according to currently practical SOC (remaining capacity), T (temperature), C (battery discharge rates), by look into the first step capacity and
Vs. temperature table obtains attenuation coefficient P, therefore goodput isThe ideal shape of battery at this time
Charge and discharge cycles number is under stateThen it is found according to cycle-index with capacity attenuation Cap Fade Curve
NtCorresponding Capt, the SOH of t moment is represented by
100% is charged to from 20% or less step 3: capturing SOC
This process is mainly used for verification and uses, and verifies to the SOH of second step.
For once effectively capturing: the information that note charging starts: time t0, SOC0, the information of note charging end: time
t1, SOC1=100, temperature T1, voltage V1Steps are as follows for calculating:
Battery capacity: cap0=∑ is calculated firsttIt* △ t, wherein △ t is acquisition time interval, to electricity in charging process
Flow ItTemporally t is integrated, because battery SOC is from SOC0(non-emptying state) is charged to 100%, therefore needs with ideally
SOC is charged to 100% from 0 and compares, therefore needs to do primary conversion and obtain battery capacity Cap1, and formula is as follows:
Because temperature is T1 when SOC is charged to 100%, and need to compare with ideally 25 DEG C, therefore needs to do the
Secondary operation, coefficient q obtain final modified battery capacity Cap2 according to the mapping table of battery capacity and temperature are as follows:
Voltage consistency of assessment when being full of, monomer voltage is very poor poor with voltage standard, the reason of to capacity attenuation
The analysis in terms of consistency is carried out, if it may be that consistency is poor, therefore need to first assess consistency that Cap2 is low, if consistency is good, Cap2
Low is because of deterioration of cell properties itself;Obtain SOH when this is full of are as follows:
By above three step, the calculating of SOH is carried out to each battery data, has wherein been obtained in third step
SOH is mainly used for the SOH obtained in second steptIt is verified, the SOH that then will be obtained in second steptMesh as study
Mark.
Step 4: calculating RUL
It converts to obtain RUL by SOH using following formula
1. obtaining cycle-index N until t moment according to the SOH prediction model of foundationtAnd SOHt;
2. cycles left times N 1 is calculated as follows:
N1=N2-Nt;
Wherein N2 is initial cycle-index note:
3. wherein remaining calculated using number of days RUL according to following formula:
Wherein RUL is the remaining number of days used, NtTo add up access times, as initial cycle number and cycles left time
Several interpolation, D are to add up to have used number of days.
By above three step, the calculating of RUL is carried out to each battery data, it is determined that the target of study.
S005 data calculate step, and the model of battery RUL prediction is established based on the data after characterization.
The problem of for battery RUL prediction, using nonlinear mixed-effect model and existence mould in present embodiment
Type establishes the battery predictive RUL model.
The model determines the relationship between variable to the credible journey of these relational expressions from one group of sample data
Degree carries out various statistical checks, and the influence for finding out from all multivariables for influencing a certain particular variables which variable is significant, which
It is not significant a bit.
With the RUL of t momenttAs Y, to each data from time enterprising row label;By step S001, S002 and
The data obtained after S003 are set as x, establish model Y=f (x), and wherein f () is the model that machine is learnt based on big data;
In the actual operation of electric car and use process, battery RUL is difficult to real-time monitoring, and method traditional at present is mainly
RUL is substantially predicted based on existing empirical equation, the main disadvantage of such method is that cannot calculate RUL and essence in real time first
Degree is not high, and due to monomer otherness, cannot predict well each monomer RUL.The model established based on big data
Problem more than can well solving.The data that the input of model is time t and t moment acquires, when the output of model is t
Carve battery RULt, in electric car real time execution and use process, the data x according to acquisition can accurately utilize model
Release RULt。
Wherein nonlinear mixed-effect model, is a kind of extension of linear assembly language, fixed effect and random
Effect part can be included in model in a non-linear fashion, relative to linear model normal state it is assumed that nonlinear model to money
The distribution of material can be normal distribution without particular/special requirement, data, be also possible to bi-distribution, Poisson distribution, same to non-linear hour
Mixed effect model has better robustness in the processing to missing data.The model of its algorithm is Y=f (x+ Φ)+e,
Wherein f () is nonlinear function, and A in Φ=A β+Bb, B are the matrix of design, and β is fixed effect parameter vector and b is random
Effect parameter vector, e are error vector, and wherein β is to predict relevant fixed effect data for battery RUL in input data x,
And b is then to predict incoherent stochastic effects data for RUL.The estimation of its parameter A and B can be walked and linear by pseudo- data
Melange effect walks the iteration between two steps and completes, and can be solved respectively using Gauss-Newton iterative method and EM algorithm.Due to electricity
Electrical automobile is in daily use process, and it and is the variation of kinematic nonlinearity that battery capacity, which is constantly to decay, so non-
Nonlinear function in linear assembly language model can preferably be fitted the battery capacity variation of kinematic nonlinearity, and
Since with the presence of partial parameters and battery capacity correlation, and part does not belong to and is distributed immediately, institute in the battery parameter of acquisition
Two class parameters of description can be then gone well with the fixed effect item in nonlinear mixed-effect model and immediately effect item.
Wherein survival analysis is exactly between the regularity of distribution and life span and correlative factor for studying life span
Relationship is analyzed and is inferred to the life span of biology or people etc. according to the data that test or investigation obtain.Survival analysis
Research concentrates on the probability of prediction reaction, survival probability, average life span.Main method has: description, nonparametric method, parametric method,
Semi-parametric approach.The wherein information that description is provided according to sample observations directly calculates each time point or every with formula
Survival function on one time interval, dead function, risk function etc., and shown using list or drawing in the form of when surviving
Between the regularity of distribution;Nonparametric method is estimated not require the distribution of life span when survival function, and examines risk factor
Using non-parametric test method when influence to life span;Parametric method estimates the distribution assumed according to sample observations
Parameter in model obtains the probability Distribution Model of life span;Semi-parametric approach does not need to make vacation to the distribution of life span
It is fixed, but the regularity of distribution and risk factor that life span can be analyzed by a model are to the shadow of life span
It rings.In survival model algorithmWherein t is battery using the time, and x is to be acquired based on time series
Data, f (x) are the probability density function of research object life span distribution, and S (t) is that research object life span is longer than the general of t
Rate.The algorithm model of RUL is Y=f (S (t), x), wherein f () algorithm model for survival.Since the RUL of battery is from initial
100% to it is retired when 80%, be equivalent to the process survived from birth to death, therefore in big data modeling process,
According to the probability density function that modeling obtains, then the battery life based on RUL parameter can be predicted well.
In the present embodiment, nonlinear mixed-effect model and survival model carry out parallel, according to the effect of last S007
Fruit selects most suitable model, and this selection is also dynamic adjustment.
S006 trains verification step, is trained and verifies to model to optimize the adaptive model.
On the basis of establishing above-mentioned model, the work for needing to be trained and verify carrys out Optimized model.To improve mould
The accuracy of type.
In this embodiment, the trained verification step preferably includes cross validation and minority class sampling.
The cross validation method of parameter frame in to(for) each model optimizes.The reliability of algorithm relies on parameter
Frame, that is which battery data is for generation the result is that most effective.
In this embodiment, in order to improve the quality of parameter frame, original data are randomly divided into K first
Part.In this K part, select one of part as test data, remaining K-1 part is obtained as training data
To corresponding experimental result.Then, another part is selected as test data, and remaining K-1 part is as training number
According to, and so on, repeat K crosscheck.Every time experiment all selected from K part a different part as
Test data guarantees that the data of K part all did test data respectively, and remaining K-1 are tested as training data.
Finally K obtained experimental result is averaged, the experimental result can be the difference of predicted value and check value, so that difference is got over
It is small better, so that it is determined that optimal classification, the training of implementation model.In this application, the data of electric car will can be obtained
It is divided into K part immediately, the data of wherein K-1 part is used to establish RUL prediction model first, then utilizes new established model
Go whether the data for verifying remaining last part meet the model.And so on.
The minority class sampling is when only having small number of training sample for a kind of data, and data set is unbalanced
It is used when situation.It, can be by will be a small number of in present embodiment when a kind of data only have a small amount of training sample
The new minority class sample data of fault sample Data Synthesis carry out the training of model.Such as in the data collection of battery,
When only collecting a small amount of sample, in order to generate more data for carrying out machine learning from low volume data, need to carry out data
Synthesis.Specifically, selecting a sample B at random from its arest neighbors to each minority class sample A, distance here is root
It is calculated, is then randomly choosed on the line between A and B a little as newly synthesized according to the distance in time and variogram
Minority class sample.Continuous synthesis in this way, can become have multidata sample A+, to reach for a small amount of sample A
To the data demand of prediction battery RUL, i.e., will not generate in calculating because of over-fitting or distortion caused by data nonbalance.
S007 algorithm evaluation step is assessed prediction result of the data under algorithms of different, optimal calculation is selected based on assessment
Method.
In battery RUL prediction, based on different prediction targets or it is different data source, using different algorithm institutes
Obtained result is also different, and thus needs to select preferable algorithm for different situations.
Usually RUL prediction in, the difference of the check value in predicted value and S004 can be used, assessment prediction as a result, than
It is whether optimal using the different obtained results of algorithm more in varied situations, to select optimal algorithm.
Wherein, difference is that the gap of the model prediction battery RUL and check value is how many for prediction result, one
As be more lower better.
Claims (4)
1. a kind of batteries of electric automobile RUL prediction technique, which is characterized in that the described method includes:
Step 001 data preparation step obtains and uses relevant data to batteries of electric automobile,
The batteries of electric automobile includes the use data of breakdown maintenance data and battery using related data;Wherein, the event
Barrier mantenance data includes the mantenance data of the data record and/or battery before cell malfunctions;The use data of the battery
It is included in battery itself data relevant to battery and vehicle condition data when normal use;The breakdown maintenance data, electricity
The stream data that time series is all based on using data in pond;And the t moment RUL that empirically formula is calculatedt;
Step 002 data preparation step, to the batteries of electric automobile using relevant data carry out cleaning and will be after cleaning
Related data is based on time quantum and carries out data building;The data cleansing includes, using taking being averaged for a trip variable
Value or median or neighbor interpolation carry out the assignment of vacant variable;Each of related data is used by setting batteries of electric automobile
The data that the threshold value inspection data of variable will exceed normal range (NR) whether meet the requirement are deleted or are corrected;It is electronic by setting
Automobile batteries is given data unreasonable or conflicting in logic using the mutual constraint of related data and dependence
It deletes or corrects;The data building includes sequentially in time integrating the data collected;
Step 003 data characterization step is summarized and is extracted to the data obtained by data preparation step, and feature is obtained
Data after change;Summary and extraction for data include rolling polymerization, and the rolling polymerization, which refers to, sets a time window,
Calculate polymerizing value in scheduled variable in the time window, the polymerizing value can be the summations of data, average value or
It is standard deviation;The summary and extraction further include being extended characteristic variable, and the extension includes to initial characteristics variable root
Increase corresponding number according to the mean value polymerizeing is rolled, and initial characteristics variable is increased accordingly according to the standard deviation for rolling polymerization
Number;
Step 004 target determines step, calculates the RUL value for study;
For the data that the step 003 obtains, the calculating of SOH is carried out, the SOH is target value, step 004 packet
It includes:
Step 1: obtaining battery master data, for calculating the SOH in second step and third step, the master data includes: electricity
The corresponding pass of tankage, battery capacity and cycle-index under the mapping table of temperature and battery ideal operating condition and capacity attenuation
It is table;
Step 2: the SOHt of statistics t moment, counts handling capacity since when battery factory brings into operation Its
Middle △ t is sampling time interval, ItElectric current when for charge and discharge, I when chargingtIt is negative, I when electric dischargetIt is positive;According to currently practical
Remaining capacity, temperature, battery discharge rates obtain attenuation coefficient P by the capacity and vs. temperature table of looking into the first step;It is real
Border handling capacity isThe ideally charge and discharge cycles number of battery is at this time
Then N is found according to cycle-index and capacity attenuation Cap Fade CurvetCorresponding Capt, the SOH of t moment is represented byWherein Cap_BOL is battery capacity;
Step 3: calculating RUL, cycles left times N 1 is calculated as follows: N1=N2-Nt, wherein N2 is initial circulation time
Number note;Residue is calculated using number of days RUL according to following formula:Wherein RUL is the remaining number of days used, NtIt is tired
Access times, the as interpolation of initial cycle number and cycles left number are counted, D is to add up to have used number of days;It is above-mentioned to obtain
RUL be subsequent step learning objective;
S005 data calculate step, establish battery RUL prediction model based on the data after characterization
With the SOH of t momenttAs Y, to each data from time enterprising row label;After step S001, S002 and S003
Obtained data are set as x, establish model Y=f (x), between the x and Y that wherein f () is learnt for machine based on big data
Function;The data that the input of the model is time t and t moment acquires, the output of model are t moment battery SOHt;
The f () is established using nonlinear mixed-effect model and survival model;
Wherein, nonlinear mixed-effect model algorithm is Y=f (x+ Φ)+e, and wherein f () is nonlinear function, in Φ=A β+Bb
A, B are the matrix of design, and β is fixed effect parameter vector and b is stochastic effects parameter vector, and e is error vector, and wherein β is
Relevant fixed effect data are predicted for battery SOH in input data x, and b is then to predict incoherent random effect for SOH
Answer data;The estimation of parameter A and B walk the iteration between two steps by pseudo- data step and linear hybrid effect and complete, and can make respectively
It is solved with Gauss-Newton iterative method and EM algorithm;
Wherein, the survival model algorithm isWherein t is the time that uses of battery, and x is based on time sequence
The data of acquisition are arranged, f (x) is the probability density function of research object life span distribution, and S (t) is research object life span
It is longer than the probability of t;The algorithm model of RUL is Y=f (S (t), x), wherein f () algorithm model for survival;
In this step S005, nonlinear mixed-effect model and survival model carry out parallel;
Step 006 trains verification step
The model of foundation is trained and is verified to optimize the model;The trained verification step preferably includes cross validation,
Optimal data classification is determined based on the experimental result;
Step 007 algorithm evaluation step
The result that prediction models various in the step 006 obtain is compared with RUL obtained in step 004, assessment is each
Prediction result of the kind prediction model data under algorithms of different, selects optimal prediction model.
2. a kind of batteries of electric automobile RUL prediction technique according to claim 1, which is characterized in that complete data building
Later, it is assessed and is corrected to based on the data that time quantum is constructed;The assessment includes that garbled data itself exists
Those of mistake data;After evaluation, the wrong data is corrected;The correction includes: that missing values are incited somebody to action
Missing values are set as 0;For exceptional value, 0 is set by negative value;For the numerical value of time cycle mistake, when should clearly obtain
Between the period, adjust and operation data again;For calculating the numerical value of specification mistake, bore adjustment and again operation data are specified.
3. a kind of batteries of electric automobile RUL prediction technique described in -2 according to claim 1, which is characterized in that the step
In 006, when data a kind of in sample only have a small amount of training sample, by the way that a small number of sample datas is synthesized new minority
Class sample data carries out the training of model;To each minority class sample A, a sample is selected at random in arest neighbors from it
B, the distance are calculated according to the distance in time and variogram, then randomly choose one on the line between A and B
Point is used as newly synthesized minority class sample;By constantly synthesizing, by a small amount of sample A, become have multidata sample A+.
4. a kind of batteries of electric automobile RUL prediction technique according to claim 1, which is characterized in that the step 004
Second step further include: capture SOC from 20% or less be charged to 100% characteristic, examine the learning objective SOHt;
From 20% or less be charged to 100% during, the information for beginning of charging: time t0, SOC0,;Charge the information terminated:
Time t1, SOC1=100, temperature T1, voltage V1;
Battery capacity: Cap0=∑ is calculated firsttIt* △ t, wherein △ t is acquisition time interval, to electric current I in charging processt
Temporally t is integrated;And do primary conversion and obtain battery capacity Cap1, formula is as follows:
Influence of the temperature T1 for battery capacity is calculated again, obtains final modified battery capacity Cap2 are as follows:
Wherein, coefficient q is searched according to the mapping table of battery capacity and temperature;
Then, voltage consistency when being full of according to monomer voltage is very poor with the assessment of voltage standard difference;If consistency is good,
Obtain SOH when this is full of are as follows:
By above-mentioned SOH numerical value to the SOH of acquisitiontIt is verified, the SOH that then will be upcheckedtAs the target for calculating study
The numerical value of RUL.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910201346.0A CN109934408A (en) | 2019-03-18 | 2019-03-18 | A kind of application analysis method carrying out automobile batteries RUL prediction based on big data machine learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910201346.0A CN109934408A (en) | 2019-03-18 | 2019-03-18 | A kind of application analysis method carrying out automobile batteries RUL prediction based on big data machine learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109934408A true CN109934408A (en) | 2019-06-25 |
Family
ID=66987411
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910201346.0A Withdrawn CN109934408A (en) | 2019-03-18 | 2019-03-18 | A kind of application analysis method carrying out automobile batteries RUL prediction based on big data machine learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109934408A (en) |
Cited By (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110286668A (en) * | 2019-07-15 | 2019-09-27 | 广东毓秀科技有限公司 | A kind of rail friendship signal system VIM board faults prediction technique based on big data |
CN110457833A (en) * | 2019-08-15 | 2019-11-15 | 深圳普瑞赛思检测技术有限公司 | A kind of data construction method for cell safety and performance evaluation |
CN110596595A (en) * | 2019-09-25 | 2019-12-20 | 广东毓秀科技有限公司 | Method for predicting RUL of rail-traffic lithium battery through big data |
CN110610016A (en) * | 2019-07-15 | 2019-12-24 | 广东毓秀科技有限公司 | Method for predicting rail transit stopping problem based on big data machine learning |
CN110850297A (en) * | 2019-09-23 | 2020-02-28 | 广东毓秀科技有限公司 | Method for predicting SOH of rail-traffic lithium battery through big data |
CN110861496A (en) * | 2019-11-27 | 2020-03-06 | 安徽江淮汽车集团股份有限公司 | Battery SOH value adjusting method and system |
CN111537889A (en) * | 2020-05-09 | 2020-08-14 | 国网福建省电力有限公司莆田供电公司 | Data-driven echelon battery RUL prediction and classification method |
CN111579121A (en) * | 2020-05-08 | 2020-08-25 | 上海电享信息科技有限公司 | Method for diagnosing temperature fault in new energy automobile battery pack on line based on big data |
CN111707957A (en) * | 2020-04-23 | 2020-09-25 | 北京邮电大学 | Method and device for estimating residual value of battery of electric vehicle |
CN111948561A (en) * | 2020-08-04 | 2020-11-17 | 上海安趋信息科技有限公司 | Battery life prediction method based on actual measurement big data and artificial intelligence learning algorithm |
CN112327170A (en) * | 2020-11-13 | 2021-02-05 | 中汽研(天津)汽车工程研究院有限公司 | Power battery full-period residual life estimation method based on neural network |
CN112542848A (en) * | 2020-11-02 | 2021-03-23 | 中国南方电网有限责任公司超高压输电公司广州局 | State estimation method and device for extra-high voltage flexible direct current transmission system |
CN112550066A (en) * | 2020-11-30 | 2021-03-26 | 蜂巢能源科技有限公司 | Method and device for determining whether vehicle battery temperature is balanced or not, storage medium and vehicle |
CN112582058A (en) * | 2019-11-07 | 2021-03-30 | 广州医科大学 | Slow obstructive pulmonary disease prediction method and system based on air quality |
CN112666464A (en) * | 2021-01-27 | 2021-04-16 | 北京嘀嘀无限科技发展有限公司 | Battery health state prediction method and device, electronic equipment and readable storage medium |
CN113240359A (en) * | 2021-03-30 | 2021-08-10 | 中国科学技术大学 | Demand prediction method for coping with external serious fluctuation |
CN113459897A (en) * | 2021-07-27 | 2021-10-01 | 四川嘉垭汽车科技有限公司 | Charge state correction and prediction method based on electric vehicle charging big data |
CN113487762A (en) * | 2021-07-22 | 2021-10-08 | 东软睿驰汽车技术(沈阳)有限公司 | Coding model generation method and charging data acquisition method and device |
CN113805087A (en) * | 2020-07-25 | 2021-12-17 | 广东毓秀科技有限公司 | Method for evaluating and predicting overall health degree of UPS (uninterrupted Power supply) by data driving |
CN114216558A (en) * | 2022-02-24 | 2022-03-22 | 西安因联信息科技有限公司 | Method and system for predicting remaining life of battery of wireless vibration sensor |
CN114236415A (en) * | 2021-12-16 | 2022-03-25 | 湖北亿纬动力有限公司 | Power battery health state detection method, device, equipment and storage medium |
-
2019
- 2019-03-18 CN CN201910201346.0A patent/CN109934408A/en not_active Withdrawn
Cited By (24)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110610016A (en) * | 2019-07-15 | 2019-12-24 | 广东毓秀科技有限公司 | Method for predicting rail transit stopping problem based on big data machine learning |
CN110286668A (en) * | 2019-07-15 | 2019-09-27 | 广东毓秀科技有限公司 | A kind of rail friendship signal system VIM board faults prediction technique based on big data |
CN110457833A (en) * | 2019-08-15 | 2019-11-15 | 深圳普瑞赛思检测技术有限公司 | A kind of data construction method for cell safety and performance evaluation |
CN110850297A (en) * | 2019-09-23 | 2020-02-28 | 广东毓秀科技有限公司 | Method for predicting SOH of rail-traffic lithium battery through big data |
CN110596595A (en) * | 2019-09-25 | 2019-12-20 | 广东毓秀科技有限公司 | Method for predicting RUL of rail-traffic lithium battery through big data |
CN112582058A (en) * | 2019-11-07 | 2021-03-30 | 广州医科大学 | Slow obstructive pulmonary disease prediction method and system based on air quality |
CN110861496A (en) * | 2019-11-27 | 2020-03-06 | 安徽江淮汽车集团股份有限公司 | Battery SOH value adjusting method and system |
CN111707957A (en) * | 2020-04-23 | 2020-09-25 | 北京邮电大学 | Method and device for estimating residual value of battery of electric vehicle |
CN111579121A (en) * | 2020-05-08 | 2020-08-25 | 上海电享信息科技有限公司 | Method for diagnosing temperature fault in new energy automobile battery pack on line based on big data |
CN111537889A (en) * | 2020-05-09 | 2020-08-14 | 国网福建省电力有限公司莆田供电公司 | Data-driven echelon battery RUL prediction and classification method |
CN113805087A (en) * | 2020-07-25 | 2021-12-17 | 广东毓秀科技有限公司 | Method for evaluating and predicting overall health degree of UPS (uninterrupted Power supply) by data driving |
CN111948561B (en) * | 2020-08-04 | 2022-12-27 | 上海安趋信息科技有限公司 | Battery life prediction method based on actual measurement big data and artificial intelligence learning algorithm |
CN111948561A (en) * | 2020-08-04 | 2020-11-17 | 上海安趋信息科技有限公司 | Battery life prediction method based on actual measurement big data and artificial intelligence learning algorithm |
CN112542848A (en) * | 2020-11-02 | 2021-03-23 | 中国南方电网有限责任公司超高压输电公司广州局 | State estimation method and device for extra-high voltage flexible direct current transmission system |
CN112327170A (en) * | 2020-11-13 | 2021-02-05 | 中汽研(天津)汽车工程研究院有限公司 | Power battery full-period residual life estimation method based on neural network |
CN112550066A (en) * | 2020-11-30 | 2021-03-26 | 蜂巢能源科技有限公司 | Method and device for determining whether vehicle battery temperature is balanced or not, storage medium and vehicle |
CN112666464A (en) * | 2021-01-27 | 2021-04-16 | 北京嘀嘀无限科技发展有限公司 | Battery health state prediction method and device, electronic equipment and readable storage medium |
CN113240359A (en) * | 2021-03-30 | 2021-08-10 | 中国科学技术大学 | Demand prediction method for coping with external serious fluctuation |
CN113240359B (en) * | 2021-03-30 | 2024-02-23 | 中国科学技术大学 | Demand prediction method for coping with external major changes |
CN113487762A (en) * | 2021-07-22 | 2021-10-08 | 东软睿驰汽车技术(沈阳)有限公司 | Coding model generation method and charging data acquisition method and device |
CN113459897A (en) * | 2021-07-27 | 2021-10-01 | 四川嘉垭汽车科技有限公司 | Charge state correction and prediction method based on electric vehicle charging big data |
CN114236415A (en) * | 2021-12-16 | 2022-03-25 | 湖北亿纬动力有限公司 | Power battery health state detection method, device, equipment and storage medium |
CN114216558B (en) * | 2022-02-24 | 2022-06-14 | 西安因联信息科技有限公司 | Method and system for predicting remaining life of battery of wireless vibration sensor |
CN114216558A (en) * | 2022-02-24 | 2022-03-22 | 西安因联信息科技有限公司 | Method and system for predicting remaining life of battery of wireless vibration sensor |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109934408A (en) | A kind of application analysis method carrying out automobile batteries RUL prediction based on big data machine learning | |
CN109934294A (en) | A method of batteries of electric automobile SOH prediction is carried out based on big data machine learning | |
CN109978229A (en) | The method that the full battery core multi-point temperature of a kind of pair of power battery pack and tie point temperature carry out thermal runaway prediction | |
CN112241608B (en) | Lithium battery life prediction method based on LSTM network and transfer learning | |
CN109993270B (en) | Lithium ion battery residual life prediction method based on gray wolf group optimization LSTM network | |
CN110516813A (en) | A method of batteries of electric automobile RDR prediction is carried out based on big data machine learning | |
CN111680848A (en) | Battery life prediction method based on prediction model fusion and storage medium | |
CN106453293A (en) | Network security situation prediction method based on improved BPNN (back propagation neural network) | |
CN110687450B (en) | Lithium battery residual life prediction method based on phase space reconstruction and particle filtering | |
CN111999648A (en) | Lithium battery residual life prediction method based on long-term and short-term memory network | |
CN112434848B (en) | Nonlinear weighted combination wind power prediction method based on deep belief network | |
CN114580706A (en) | Power financial service wind control method and system based on GRU-LSTM neural network | |
CN109523021A (en) | A kind of dynamic network Structure Prediction Methods based on long memory network in short-term | |
CN110850297A (en) | Method for predicting SOH of rail-traffic lithium battery through big data | |
CN113536670B (en) | Federal migration learning method for predicting residual life of group product | |
CN115598557B (en) | Lithium battery SOH estimation method based on constant-voltage charging current | |
CN112611976A (en) | Power battery state of health estimation method based on double differential curves | |
CN115586444A (en) | Lithium battery residual life prediction method based on VMD and BP neural network | |
CN113866659A (en) | Lithium battery SOH prediction method based on machine learning and ampere-hour integration method | |
CN114384435A (en) | WSA-LSTM algorithm-based self-adaptive prediction method for residual service life of new energy automobile power battery | |
CN114189047B (en) | False data detection and correction method for active power distribution network state estimation | |
CN115577637A (en) | Knowledge and data fused power supply system residual life prediction method | |
Liu et al. | Research on the strategy of locating abnormal data in IOT management platform based on improved modified particle swarm optimization convolutional neural network algorithm | |
CN117034762A (en) | Composite model lithium battery life prediction method based on multi-algorithm weighted sum | |
CN110596595A (en) | Method for predicting RUL of rail-traffic lithium battery through big data |
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 | ||
WW01 | Invention patent application withdrawn after publication | ||
WW01 | Invention patent application withdrawn after publication |
Application publication date: 20190625 |