CN106528951A - Life prediction and safety warning methods and apparatuses for power battery - Google Patents
Life prediction and safety warning methods and apparatuses for power battery Download PDFInfo
- Publication number
- CN106528951A CN106528951A CN201610906400.8A CN201610906400A CN106528951A CN 106528951 A CN106528951 A CN 106528951A CN 201610906400 A CN201610906400 A CN 201610906400A CN 106528951 A CN106528951 A CN 106528951A
- Authority
- CN
- China
- Prior art keywords
- voltage
- battery
- item
- data
- predicted
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/06—Power analysis or power optimisation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Hardware Design (AREA)
- Evolutionary Computation (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Secondary Cells (AREA)
- Charge And Discharge Circuits For Batteries Or The Like (AREA)
Abstract
The invention discloses a method and an apparatus for predicting residual life of a power battery system by utilizing a computer. The method comprises the following steps of measuring and collecting battery voltage data and battery event data in a period of time by utilizing a battery management system; decomposing a voltage V of a given time region into an aging voltage Va and a fluctuation voltage Vf; predicting the aging voltage Va by utilizing a multi-instance multi-label learning method; predicting the fluctuation voltage Vf by utilizing an auto-regressive integrated moving average (ARIMA) model; and predicting residual life of a battery by utilizing the predicted aging voltage Va and fluctuation voltage Vf. According to the method, the performance and expected life of the battery system can be estimated more accurately through considering historical voltage data and event data of the battery by utilizing a big data analysis and processing technology, so that the prediction accuracy and reliability are improved.
Description
Technical field
The invention belongs to field of computer technology, more particularly to a kind of utilization computer system is using the mould of big data analysis
The method and apparatus that type and method carry out life prediction and safe early warning to electrokinetic cell systems such as lithium battery, lead-acid batteries, energy
It is enough in the concrete industrial circle such as new forms of energy, communication system, electronic information and computer technology.
Background technology
Currently, various aspects of the electrokinetic cell in people's daily life, are obtained for increasing application.Based on power
The new-energy automobile of battery, the back-up source of radio communication base station, energy storage technology etc. become countries in the world today and competitively develop
Most one of hot technology, the health and safety of thing followed electrokinetic cell also becomes the such as industry such as new-energy automobile health and sends out
The capacity of the key factor of exhibition, wherein electrokinetic cell is accurately estimated, accurately estimation, life-span accurate prediction are safe practices to health
Core, and current world-famous puzzle in the industry.
Currently in the industry, it is general all inside battery management system BMS to the volume calculation and health forecast of electrokinetic cell bag
Complete.It has been found that there is problems with existing battery management system BMS：Existing volume calculation and health forecast are calculated
Method is simple, less effective.For example, the battery backup system in radio communication base station, only records to the operational data of power supply,
Any prediction and warning is not done.The BMS of commercial lithium battery pack is based on cost reason, often all by the micro-control of the basic DSP functions of band
Device processed realizing the management of charging and discharging to battery bag, battery balanced etc., and generally without jumbo storage core inside BMS
Piece, almost or not makees to retain to the historical events or performance data of lithium battery pack, instant battery performance is estimated
Error is big, reliability is low, calculates instant internal resistance typically all based on the measurement of the voltage to moment, electric current and temperature, then and
It is compared when battery is initially used and obtains, does not consider historical information.
To solve drawbacks described above of the prior art, need to propose a kind of new to carry out life prediction and peace to electrokinetic cell
The method of full early warning, to overcome the above-mentioned problems in the prior art.
The content of the invention
It is an object of the invention to provide the model and method of a kind of analysis of utilization big data to carry out the life-span to electrokinetic cell pre-
Survey and safe early warning, to solve the problems, such as that evaluated error present in existing battery management system is big, effect on driving birds is not good.
According to an aspect of the invention, there is provided a kind of utilization computer is carried out to the residual life of electrokinetic cell system
The method of prediction, it is characterised in that comprise the following steps：Data acquisition step, measures and collects one section using battery management system
Battery voltage data and battery event data in time；Data processing step, specifically includes herein below：By preset time area
The voltage V in domain resolves into aging item voltage Va and fluctuation item voltage Vf；Using many example multi-tag (MIML) learning methods, to old
Change item voltage Va to be predicted；Moving average (ARIMA) model is integrated using autoregression, item voltage Vf is predicted to fluctuating；
The aging item voltage Va obtained using prediction and fluctuation item voltage Vf, is predicted to the residual life of battery.
Wherein, in the present invention as stated above, also including denoising step, making an uproar in the battery voltage data being collected by filtration
Sound, only retains floating voltage, to reduce interference of the voltage from other battery activities.
Wherein, in the present invention as stated above, described autoregression integration moving average model, is by moving average model and synthesis
The part autoregression of part composition integrates moving average model, and its concrete form is
Wherein,Fluctuation item magnitude of voltage during expression time t, θ1..., θqThe parameter of moving average model is represented,
ε1..., εt-1It is white noise error item, q is rolling average item number, εtFor comprehensive part.
Wherein, in the present invention as stated above, using equation below, calculate the residual life of i-th battery in time t
Wherein,WithAging item and the item magnitude of voltage that fluctuates when representing time t respectively, voltage slope is st, penalty values are
yt, i is battery sequence number, and θ is predefined battery failure threshold value.
Wherein, in the present invention as stated above, after the residual life to battery is predicted, also with warning step, point out
User is changed to the battery that will be failed.
According to another aspect of the present invention, there is provided the dress that a kind of residual life to electrokinetic cell system is predicted
Put, the device includes：Battery management system, for measuring and collecting battery voltage data and battery event number in a period of time
According to；Data acquisition module, is used for subsequently for obtaining the battery voltage data and battery event data from battery management system
Process；Data processing module, specifically includes：The voltage V in preset time region is resolved into into aging item voltage Va and fluctuation item electricity
The unit of pressure Vf；Using many example multi-tag (MIML) learning methods, the unit being predicted to aging item voltage Va；Using certainly
Regression-Integral moving average (ARIMA) model, the unit is predicted by the item voltage Vf that fluctuates；The aging item obtained using prediction
Voltage Va and fluctuation item voltage Vf, the unit is predicted by the residual life of battery.
According to above-mentioned electrokinetic cell life prediction proposed by the present invention and the method and apparatus of safe early warning, using big data
The model and method of analysis, has taken into full account the shadow of battery history voltage data and history event data to the electrokinetic cell life-span
Ring.The present invention proposes a kind of event driven battery analysis method, and accurately extracting causes what battery pack working condition was degenerated
Characteristic, and set up forecast model.Based on the cell voltage in the present invention and the forecast model in life-span, can obtain it is more accurate,
Reliable prediction effect, can solve the problem that battery life evaluated error present in existing battery management system is big, effect on driving birds is not good
Problem.
Description of the drawings
Fig. 1 is the average voltage schematic diagram of new battery and used batteries；
Correlation schematic diagrames of the Fig. 2 for the average voltage and residual life of battery；
Fig. 3 is the big logotype of voltage variance of new used batteries；
Correlation schematic diagrames of the Fig. 4 for the variable quantity and residual life of voltage；
Relation schematic diagrams of the Fig. 5 for the quantity and battery life of generation " cell voltage is low " event；
Relation schematic diagrams of the Fig. 6 for the quantity and battery life of generation " battery discharge " event；
Relation schematic diagrams of the Fig. 7 for the quantity and battery life of generation " battery failures " event；
Fig. 8 is the schematic diagram of the EVENING frameworks of prediction cell operating status proposed by the present invention；
Fig. 9 is the distribution schematic diagram of the failure trend extracted from data set；
Figure 10 is that EVENING methods and the linear regression (LR) of the present invention, autoregression integrate moving average (ARIMA) and little
Four kinds of method predicated error schematic diagrames of wave analysis (Wavelet)；
Figure 11 be change quantity percentage and battery failure early warning after relation schematic diagram between the number of days survived；
Figure 12 is the comparison schematic diagram that the EVENING frameworks and simple ARIMA of the present invention predicts the outcome.
Specific embodiment
To make the object, technical solutions and advantages of the present invention of greater clarity, with reference to specific embodiment and join
According to accompanying drawing, the present invention is described in more detail.It should be understood that these descriptions are simply exemplary, and it is not intended to limit this
Bright scope.
1. background
With widely using for 3G/4G cellular networks, and 5G cybertimes in future are at hand, result in global mobile phone
The explosive growth of service market.A large amount of base stations have disposed environment, to meet to service quality and the demand of coverage, and
And base station number will the in a foreseeable future interior growth of category.In view of there are many base station deployments in remote countryside area, safeguard
High service availability becomes quite challenging.Particularly, they may meet with and frequently have a power failure.It is being subjected to hurricane or sudden and violent
After wind and snow, power recovery generally requires even several weeks a couple of days, and reserve battery becomes during this period uniquely supply electric resources.Although
It is rare that urban area has a power failure, and reserce cell is still required in a base station, because the interruption of any service leads to
The loss that can not be born.In view of in outage, the battery backup that base station is installed is typically unique electric power resource, therefore electricity
The working condition of pond group has vital impact to the availability service of base station.
For the availability of the service of improving, except being connected to utility network, each cell tower is further provided with for standby confession
The battery pack of electricity.When having a power failure on utility network, in order to avoid the interruption of any service, battery power discharge is supporting that communication sets
It is standby, until generating function provides enough supplies of electric power.This causes batteries and its condition of work to play an important role.
During this period, if the situation of battery deteriorates, power-off just can break service easily down.The arrival of emergency maintenance service may need
Even several weeks are taken days, this depends on the difficulty size for reaching and orienting maintenace point, especially in rural area or chance
To bad weather.Therefore, predict that the state of battery pack, for system maintenance, has significance technologically and commercially.
This is, in order to battery can be changed in advance before failure, to reduce the interference to availability service.One is successfully predicted, no
Only it should be understood that the knowledge of cell degradation process, will also have reason well to the stress event of possible induction and accelerated ageing process
Solution.
Although lithium ion battery and nickel-cadmium cell are shown with its less size, lighter weight and more preferable storage efficiency
The latest development of battery science and technology, but the major defect of this kind of battery is exactly that cost is too high.Traditional lead-acid battery has very big
Capacity, therefore the storage being widely used in the stand-by power supply of base station.The aging mechanism of lithium ion battery is caused a lot
Attempt, generate substantial amounts of daily record in the frequent activities of lithium battery, and there is provided being available for measuring the possibility of battery operating conditions
Property.However, lead-acid battery generally remains in floating charge state (floating charge state represents battery after fully charged from putting in a base station
Electricity maintains the ability of itself capacity), monitoring system collects float charge voltage twice daily from boosting battery, can so make data
Collection tails off.Therefore, data characteristics is extracted from such a rare data source, and predict that the working environment of lead-acid battery brings
Many challenges, in this application inventor attempt to solve the problem.
The present invention is by advanced cloud computing model and the big data information processing technology, it is proposed that a kind of to be applied to power electric
The complex mathematical model framework in pond solving the technical barrier of the performance of the accurate estimation battery system of how dynamic, and to power electric
The unsafe factor in pond carries out real-time early warning, so as to prevent electrokinetic cell from producing various dangerous or burning factor.Battery operation
Real time data completed by coupled data collection station.
The big data processing method that we invent comes accurately pre- EVENING, the i.e. analytical framework of event driven battery
Electrokinetic cell bag real-time operating conditions are surveyed, and arranges corresponding maintenance.We have collected 46913 equipment altogether from July, 2014
28 days on 2 17th, 2,016 1550032984 row data, including 112 kinds of different events, by our cloud computing platform
Hadoop 1.2.1 and Hive 1.2.1 are further processed and analyze.Particularly, we are by seasonal effect in time series voltage decomposition
To aging item and fluctuation item in terms of.Based on aging item and log event, we analyze the aging tendency of battery to state battery
Differentiation, including represent battery condition of work slope and variance.We are using event driven problem of aging as more than one
The target of many example problems of label, minimizes root mean square error between predicted value and actual value, so as to ensure that cell voltage is pre-
The accuracy of survey.Based on the framework, we are to define forecast model in cell voltage and life-span, and propose a series of solution
Scheme is accurately exported to increase.
2 data analyses
The voltage readings of 2.1 batteries
The voltage of each battery is most important characteristic, because it has reacted the power mode output of battery.Our roots
The representational battery of two classes is observed according to data record：New battery and frequency face the battery of failure, then select 1578 batteries
As new battery pack, and 1459 batteries are put into the group for being on the brink of to fail.The rated voltage of battery is 2.23V, 24 battery quilts
Connect as a battery bag, the rated voltage of whole battery backup system is 53.5V.On this basis, we further analyze
The typicalness of the voltage mode in two class representative cell voltage class.Fig. 1 illustrates new battery (good battery) and is on the verge of
The marked difference of the battery (bad battery) of failure.Solid line depicts the average voltage of new battery pack, substantially remain in 2.21V and
2.25V between.Dotted line represents the attenuation trend for being on the verge of dead battery average voltage.Under having one when being close to the Expiration Date significantly
Drop trend, in this stage, the power of battery frequently declines, and rapid exhaustion, causes numerous events and alarm.Fig. 2 is further
Show average voltage is how to be mutually related with the length of residual life, this shows that voltage and battery life have very strong association
Property.As shown in figure 3, solid line represents that new battery of installing can export stable power, voltage variance is close to 0.Dotted line shows to be on the point of
The new battery of variance ratio of dead battery it is much bigger.Fig. 4 shows that voltage difference is also relevant with the length of remaining battery life.Battery
Output voltage change with the time, reflect the aging trend of battery quality.
The activity of 2.2 batteries
We are analyzed to the battery activity in record, inquire into battery activity and associating between remaining battery life
Property, the event of selection is low cell voltage, battery discharge and trouble unit respectively, as shown in Fig. 5,6 and 7.When battery failure
When, we calculate the number of particular event for each battery, and dotted line represents the battery that residual life was more than 576 days.Fig. 5 and Tu
6 depict the low relevance between electric discharge and residual life of cell voltage.This clearly illustrates, in remaining battery life and low
There is a kind of very strong relevance between floating voltage, between residual life and battery discharge, there is also certain correlation.
In Fig. 7, we further draw the crash rate of fail battery activity number in the system, therefrom do not see there is significantly association
Property.It means that failure is affected by some particular events.Observation indicate that, different events are to battery operated shape
State has opposite impacts on, therefore, it is in order to accurately be predicted, necessary that these events are treated with a certain discrimination.
3. forecast model and solution
For ease of understanding, the parameter declaration and definition occurred in the application refers to table 1 below：
1 parameter declaration of table and definition
Observation shows that cell operating status can be gone out by related history voltage and event prediction, and this observation can be made
To predict the basis of remaining battery life.Monitoring system collects initial data from the voltage and event of all batteries, such to obtain
To voltage collection V={ v1,v2,…vNAnd event set X={ x1,x2,…xN, wherein N represents battery sum.Each voltage is recordedIt is magnitude of voltage vs of the battery i in time TT.Our target is to extract to summarize in cell operating status
Aging tendency, and predict remaining battery life.As remaining battery life is on the basis of foregoing magnitude of voltage
Prediction is obtained, so voltage timeliness problem can be defined as follows：
In view of battery voltage data V and historical events X on time shaft from 1 to T, we are calculated based on predicted voltage
Go out remaining battery life.The predicted voltage of bay when defining following K, with virtual voltageContrasted.Overall goals function can be written as mode：
By professional knowledge and our observation, it can be found that voltage is the standard of battery condition, residual life can be very
Obtain easily by the prediction of predefined battery voltage threshold θ.
For solve problem, we have proposed EVENING frameworks shown in Fig. 8 to predict cell operating status, Tu Zhongbao
Following several stages are contained：Data preparation, voltage decomposition, voltage prediction and predicting residual useful life.EVENING frameworks are filtered first
Fall the much noise in history voltage, only with floating voltage reducing interference of the voltage from other batteries activities, such as battery
Electric discharge.Although historical events is related to cell operating status, individual event record is not the responsible factor of prediction.
In terms of EVENING frameworks resolve into aging item and fluctuation item two history floating voltage in temporal sequence.For aging item, we
A kind of punishment idea is proposed, for describing the impact of event in one group of voltage aging trend.Framework utilizes time algorithm and event
Data come predict future aging tendency, and further combine ARIMA predicted voltage fluctuation item.By the following months
Predicted voltage, EVENING frameworks are estimated that the residual life of battery.
As being previously mentioned in Fig. 8, we resolve into aging item v the voltage v in preset time regionaWith fluctuation item vf：
V=va+vf (2)
In order to ensure vaTrend be dull, we can simply write out the constraint to object function：
WithDescription voltage pulsation, with the growth of t, fluctuation also increases.Optimization during we further expand formula (1) is asked
The object function of topic, reduces reconstructed error as far as possible | | v-va’-vf’||2With the flatness for guaranteeing to fluctuate, this is by formula 3
Monotonicity constraint control.
(1) aging item
In order to extract voltage aging item and make fluctuation item steady, we in time series are segmented voltage, and guarantee
Per section of mean difference as far as possible it is little.Specifically, it is contemplated that the voltage v in a time seriesi, we draw it
It is divided into section, is expressed as follows：
L is the length of section, and each time period k has mean valueWe match each section with polynomial regression, s be
The voltage slope of time period k battery i.Aging item can be defined as：
Then, we have proposed a loss voltage conceptTo describe the growth of aging item：
For the battery i in time period k, we have aging itemWithHistory electricity in our usage records
Pond event is predicting aging item viaPenalty valuesIn view of history battery event and the training set of voltage loss value, we
One predictable problem can be illustrated becomes predicted voltage penalty values and builds a time sorter.One of them is main
Challenge be exactly monitoring system for each battery pack Collection Events, the cell in battery pack has identical event note
Record, although cell has the voltage data value of oneself uniqueness.Therefore, we for each battery pack predicted voltage penalty values not
It is predicted for each cell, although the residual life of cell can be estimated by the battery penalty values based on battery bag
Arrive.
Our problem can change into many example multi-tag (MIML) problems concerning study, each training sample not only with it is many
Individual example association, is also associated with multiple labels.With receive one group of independence mark example as criteria classification conversely, in Fig. 8
MIML models receive one group of bag, this group bag is simultaneously associated with multiple labels.In our problem, each event is defined
For an example, each loss of voltage is defined as a label.Concept bag represents one group of event xi, each bag xiThere is miIndividual reality
Example, xi={ xI, 1, xI, 2..., xI, mi}.Obtain from different batteries in view of bag, our target is to set up a grader,
Energy correct labeling goes out invisible bag.
For formal, X represents instance space, and Y defines the set of loss label.We use the set containing N number of sample
{(x1,y1),(x2,y2),…,(xN,yN) representing training set, yiIt is and xiRelated loss label.Carried above, each
Battery is surrounded by niThe battery of individual different voltage aging trend, yi={ yI, 1, yI, 2..., yI, niIt is the one of all possible label
Individual subset, yi∈Y.Target is from one group of given data set { (x1,y1),(x2,y2),…,(xN,yN) in know a function
fMIML:2X→2Y, for accurately predicting sequence of events xiLoss label yi。
Based on MIML fast methods (S.-J.Huang and Z.-H.Zhou, " Fast multi-instance
multi-label learning,”arXiv preprint arXiv:1310.2049,2013.), it is proposed that a kind of and former
The effectively approximate improved method of beginning MIML problem.Specifically, using the relation between multiple labels, MIMLfast is first
From the communal space of all labels of primitive character learning, then train from shared label and mark specific linear model.For
Showing the bag of specific label, disaggregated model is trained identification critical instance in instance layer, then selects the pre- of maximum
Survey example value.In order to strengthen learning efficiency, optimize the loss of approximate sorting with stochastic gradient descent.In test phase,
MIMLfast returns a subset with the be possible to label of test value, and by selecting maximum prediction data, ENENING can
To obtain the penalty values of each bag.
(2) fluctuate item
Voltage change is also prediction one important feature of battery life.Obtain out floating from initial data using the eyes of people
The fluctuation of dynamic voltage is not an easy thing.It is understood that most of battery keeps floating charge state for a long time, this causes battery
Weak current discharge and recharge is accompanied by periodically in fixed intervals.Sequence of events method is applied to random process as description,
Also easily establish the forecast model of the item that can fluctuate for predicted voltage.Autoregression integration moving average model (ARIMA) is most to flow
It is capable for predicting one of time series models of future value, this may be widely used for finance, economy and domain of the social sciences.
ARIMA can be used for representing past observation and a linear combination of mistake, comprising three parts, an autoregression model, and one
Individual moving average model and a comprehensive part.ARIMA (p, q) model is expressed as follows：
Y (t) represents the amount of views in the t times.β1..., βpRepresent the parameter of autoregression model, θ1..., θqRepresent
The parameter of moving average model.ε1..., εt-1It is white noise error item, this is typically considered zero-mean with lasting variance
Gaussian random variable.P is autoregression item；Q is rolling average item number.In our EVENING frameworks, we are using a part
ARIMA models are representing fluctuation item.
In view of the floating voltage v in past several daysiTime series, it can utilize historical information in becoming
The performance of gesture, periodicity and autocorrelation, is that fine granularity prediction is made in the evolution of voltage.EVENING frameworks have been used such as formula (8)
Described part ARIMA models are based on voltage v to extract and predictiFluctuation item vif。
(3) residual life is estimated
By voltage aging trend term vaWith fluctuation tendency item vf, we can pass through the magnitude of voltage that formula 2 predicts future
v’.When floating voltage value is higher than predefined battery failure threshold θ, one can consider that the working condition of battery i is good
, form can be written as follows：
via+vif-θ>0 (9)
In time t, magnitude of voltage isVoltage slope is st, penalty values are yt, residual life r can be expressed as follows：
4. test
Before quantization performance evaluation is carried out, we provide an example as case study, for proving EVENING frames
The validity of frame.We apply to EVENING and ARIMA the battery data of 365 days, old in following 120 days for predicting
Change trend and battery operating regime.
(1) experimental facilities
The analysis of mass data needs time enough and capacity, and we have processed the data of 323GB in cloud platform, make
The inquiry for achieving data is more efficient, and the platform in a specific embodiment of the invention contains 8 Dell Dell PowerEdge
R430 servers, each server are equipped with the DDR3 internal memories and of Xeon E5-2630v3,256GB of two 2.4GHz8 cores
The NIC (NIC) of individual 10Gbits/ seconds.CPU is enabled with hyperthread, each CPU core can support that two threads are (empty like that
Intend core cpu).All of physical server is connected with each other by 16 port of netware, 10 gigabit switch.Using special thing
Reason equipment to main frame host node rather than use a virtual machine, it is ensured that minimize contention for resources fast response time.Its
The physical equipment of his main frame dummy node runs the Xen Hypervisor (version 4.1.3) of latest edition.With regard to operating in
Operating system on virtual machine (Domain0 and DomainU), we use the Ubuntu 12.04LTS (kernels of popular 64
Version 3 .11.0-12).Distribute the internal memory of 8 virtual cpu cores and 4GB for all of virtual machine.We run Hive 1.2.1,
The data warehouse being built upon on Apache Hadoop 1.2.1, it is adaptable to data management and looked into based on the analysis of HiveQL
Ask, wherein, HiveQL is a kind of query language based on SQL.
(2) assess
In this section, we carry out the accuracy of appraisal framework using the real data that obtain actually are collected, and deduce
Residual life.For assessment prediction quality, we are tested to 1691 battery datas, wherein there is 1282 sample training collections
With 409 test sample collection.
We calculate previous 365 days and the difference between the slope of 120 days afterwards, are that cell degradation item generates a failure
Label.Fig. 9 summarises the distribution of the failure trend extracted from data set, coverage from 0 to -5 × 10-4, and mostly
Number battery is in slow degradation mode.In table 2, we compared for EVENING methods and linear regression (LR), ARIMA and small echo
Analysis (Wavelet), and calculate the root-mean-square error of predicted voltage value and real data.As shown in Figure 10, we can be with
| 20 | 40 | 60 | 80 | 100 | 120 | |
| MIML | 0.0022 | 0.0020 | 0.0013 | 0.0003 | 0.0008 | 0.0020 |
| ARIMA | 0.0041 | 0.0054 | 0.0122 | 0.0190 | 0.0290 | 0.0401 |
| LR | 0.0307 | 0.0272 | 0.0286 | 0.0287 | 0.0298 | 0.0300 |
| Wavelet | 0.0601 | 0.0670 | 0.0668 | 0.0737 | 0.0694 | 0.0698 |
The root-mean-square error of 2 real data of table and prediction data
Find out, in most contrast scheme, effect that our method presents preferably, the trend of extraction with it is true
Error between development trend is minimum.Especially when slope very little, when aging phenomenon is very small, our method than linear analysis,
Autoregression integrates moving average model and wavelet analysis effect will be got well.Additionally, the aging tendency extracted from scheme is dull
, meet the essence of aging phenomenon, however, linear analysis, autoregression integration moving average model and wavelet analysis just do not have this
The advantage of aspect, because they do not carry out monotonicity constraint in terms of extraction trend.
Have that one is rational it is assumed that the consequence of battery failure is to change battery, we select to have 112 that change record it is electric
Pond, for verifying prediction residual life.Relation such as Figure 11 institutes between the number of days survived after percentage and battery failure early warning
Show.87% battery was replaced within three months, and this shows that our method can forcefully predict battery failure.With this
Meanwhile, only a few battery is still mounted after early warning failure, and we are examined to them, it is found that these batteries do not have
Have to be replaced and be because there is unnecessary battery pack, this shows that our EVENING frameworks can effectively assist Maintenance Engineer
Potential problem in investigation battery pack.
(3) case study
As shown in figure 12, comparative result is met with our guess, that is, thing for the comparison of EVENINGE frameworks and ARIMA
Part can bring permanent change to cell operating conditions.ARIMA is only using the floating voltage of single battery, and it can not be by
Model is made in the impact of historical events.In view of the impact of event, EVENING frameworks can quick response these events, and carry
For an accurate anticipation trend.In fig. 12, we are it is clear that EVENING frameworks can go out battery with success prediction
The downward trend of voltage, can also predict working condition of the battery in following three months.FOR ALL WE KNOW,
The major defect of ARIMA models is that it needs the historical information of long periods to be predicted.In an experiment, ARIMA is applied to
The fluctuation item of the data based on first 365 days is predicted, although it can not accurately estimate aging condition.We can be pre- according to battery
The threshold value of definition, predicts remaining battery life.
In sum, a kind of the present processes, it is proposed that event driven battery analysis method, can accurately to extract
With the feature for causing battery pack working condition to be degenerated.We are that cell voltage and service life have formulated forecast model, and are carried
A series of solutions that can obtain accurate output valve are gone out, this let us is it is further proposed that a scheme, arranges battery in time
Safeguard and substitute, so that the service disruption brought that has a power failure is minimized.Assessed according to the tracking with True Data, it was demonstrated that this Shen
Method please can obtain higher precision in terms of voltage and life prediction, it is possible to which being obviously improved taking for cellular network can
The property used.
The above-mentioned specific embodiment of the present invention is used only for exemplary illustration or explains the principle of the present invention, and does not constitute
Limitation of the present invention.Therefore, any modification for being made in the case of without departing from the spirit and scope of the present invention, equivalent are replaced
Change, improve, should be included within the scope of the present invention.Additionally, claims of the present invention are intended to fall into
Whole in the equivalents on scope and border or this scope and border changes and modifications example.
Claims (10)
1. a kind of method that utilization computer is predicted to the residual life of electrokinetic cell system, it is characterised in that including following
Step：
Data acquisition step, measures and collects the battery voltage data and battery event in a period of time using battery management system
Data；
Data processing step, specifically includes herein below：
The voltage V in preset time region is resolved into into aging item voltage Va and fluctuation item voltage Vf；
Using many example multi-tag (MIML) learning methods, aging item voltage Va is predicted；
Moving average (ARIMA) model is integrated using autoregression, item voltage Vf is predicted to fluctuating；
The aging item voltage Va obtained using prediction and fluctuation item voltage Vf, is predicted to the residual life of battery.
2. method according to claim 1, it is characterised in that the data acquisition step, also including denoising step, filters
Noise in the battery voltage data collected, only retains floating voltage, to reduce voltage from other battery activities
Interference.
3. method according to claim 1, it is characterised in that described autoregression integration moving average model, is by sliding
The part autoregression that dynamic averaging model and Synthesis Department are grouped into integrates moving average model, and its concrete form is
Wherein,Fluctuation item magnitude of voltage during expression time t, θ1..., θqRepresent the parameter of moving average model, ε1...,
εt-1It is white noise error item, q is rolling average item number, εtFor comprehensive part.
4. method according to claim 1, it is characterised in that utilize equation below, calculates i-th battery in time t
Residual life
Wherein,WithAging item and the item magnitude of voltage that fluctuates when representing time t respectively, voltage slope is st, penalty values are yt, i
For battery sequence number, θ is predefined battery failure threshold value.
5. the method according to claim 1-4, it is characterised in that after the residual life to battery is predicted, also
With warning step, user is pointed out to change the battery that will be failed.
6. the device that a kind of residual life to electrokinetic cell system is predicted, it is characterised in that include with lower module：
Battery management system, for measuring and collecting battery voltage data and battery event data in a period of time；
Data acquisition module, is used for subsequently for obtaining the battery voltage data and battery event data from battery management system
Process；
Data processing module, specifically includes：
The voltage V in preset time region is resolved into into the unit of aging item voltage Va and fluctuation item voltage Vf；
Using many example multi-tag (MIML) learning methods, the unit being predicted to aging item voltage Va；
Moving average (ARIMA) model is integrated using autoregression, the unit being predicted to the item voltage Vf that fluctuates；
The aging item voltage Va obtained using prediction and fluctuation item voltage Vf, the unit is predicted by the residual life of battery.
7. device according to claim 6, it is characterised in that the data acquisition module, also including denoising unit, is used for
Noise in the battery voltage data being collected by filtration, only retains floating voltage, is lived from other batteries with reducing voltage
Dynamic interference.
8. device according to claim 6, it is characterised in that described autoregression integration moving average model, is by sliding
The part autoregression that dynamic averaging model and Synthesis Department are grouped into integrates moving average model, and its concrete form is
Wherein,Fluctuation item magnitude of voltage during expression time t, θ1..., θqRepresent the parameter of moving average model, ε1...,
εt-1It is white noise error item, q is rolling average item number, εtFor comprehensive part.
9. device according to claim 6, it is characterised in that utilize equation below, calculates i-th battery in time t
Residual life
Wherein,WithAging item and the item magnitude of voltage that fluctuates when representing time t respectively, voltage slope is st, penalty values are yt, i
For battery sequence number, θ is predefined battery failure threshold value.
10. the device according to claim 6-9, it is characterised in that also with warning module, points out user to failing
Battery changed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610906400.8A CN106528951B (en) | 2016-10-18 | 2016-10-18 | A kind of method and apparatus of power battery life prediction and safe early warning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610906400.8A CN106528951B (en) | 2016-10-18 | 2016-10-18 | A kind of method and apparatus of power battery life prediction and safe early warning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106528951A true CN106528951A (en) | 2017-03-22 |
| CN106528951B CN106528951B (en) | 2019-10-25 |
Family
ID=58332266
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610906400.8A Active CN106528951B (en) | 2016-10-18 | 2016-10-18 | A kind of method and apparatus of power battery life prediction and safe early warning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106528951B (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107273632A (en) * | 2017-06-28 | 2017-10-20 | 北京航空航天大学 | A kind of non-accumulated impact pectus part lifetime of system Forecasting Methodology and device |
| WO2018227986A1 (en) * | 2017-06-14 | 2018-12-20 | 上海蔚来汽车有限公司 | Power-up optimization method and apparatus, terminal, facility, device, and storage medium |
| CN109116244A (en) * | 2018-06-29 | 2019-01-01 | 深圳市科列技术股份有限公司 | A kind of judgment method and device of power battery aging tendency |
| CN110208705A (en) * | 2019-05-09 | 2019-09-06 | 赛尔网络有限公司 | A kind of lithium battery method for predicting residual useful life and device |
| CN110888077A (en) * | 2019-10-30 | 2020-03-17 | 无锡市产品质量监督检验院 | Accelerated lithium ion battery life evaluation method based on ARIMA time sequence |
| CN110988717A (en) * | 2019-12-10 | 2020-04-10 | 惠州Tcl移动通信有限公司 | Battery detection method, storage medium and terminal device |
| CN111190112A (en) * | 2020-02-10 | 2020-05-22 | 宜宾职业技术学院 | Battery charging and discharging prediction method and system based on big data analysis |
| CN112563543A (en) * | 2021-02-23 | 2021-03-26 | 北京亿华通科技股份有限公司 | Method for predicting and maintaining service life of fuel cell system |
| CN112991574A (en) * | 2019-12-13 | 2021-06-18 | 北京亿华通科技股份有限公司 | Method for analyzing attenuation of electric pile |
| US11065978B2 (en) | 2019-02-25 | 2021-07-20 | Toyota Research Institute, Inc. | Systems, methods, and storage media for adapting machine learning models for optimizing performance of a battery pack |
| US11084387B2 (en) | 2019-02-25 | 2021-08-10 | Toyota Research Institute, Inc. | Systems, methods, and storage media for arranging a plurality of cells in a vehicle battery pack |
| CN113589183A (en) * | 2021-08-09 | 2021-11-02 | 湖北亿纬动力有限公司 | Self-discharge detection method, device and system based on big data |
| CN116401604A (en) * | 2019-05-13 | 2023-07-07 | 北京绪水互联科技有限公司 | Method for classifying and detecting cold head state and predicting service life |
| US11899441B2 (en) | 2018-10-10 | 2024-02-13 | Siemens Aktiengesellschaft | Remaining service life prediction for switches |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103399279A (en) * | 2013-08-01 | 2013-11-20 | 哈尔滨工业大学 | Method for predicting cycle life of fused lithium ion battery based on EKF (Extended Kalman Filter) method and AR (AutoRegressive) model |
| CN103927597A (en) * | 2014-04-22 | 2014-07-16 | 国家电网公司 | Ultra-short-term wind power prediction method based on autoregression moving average model |
| CN104680024A (en) * | 2015-03-11 | 2015-06-03 | 南京航空航天大学 | Method for predicting remaining useful life of lithium ion battery based on GA (Genetic Algorithms) and ARMA (Auto Regressive and Moving Average) models |
| CN105116337A (en) * | 2015-06-30 | 2015-12-02 | 中国人民解放军国防科学技术大学 | Lithium ion battery full charge storage service life evaluation method |
| CN105137358A (en) * | 2015-08-27 | 2015-12-09 | 张家港莫特普数据科技有限公司 | Power battery SOC/SOH prediction method based on large data self-learning mechanism |
| US20160259014A1 (en) * | 2015-03-03 | 2016-09-08 | Samsung Electronics Co., Ltd. | Method and apparatus for automatically estimating remaining useful life (rul) of battery in real time |
| KR20160107093A (en) * | 2015-03-03 | 2016-09-13 | 삼성전자주식회사 | Method and system for predicting remaining useful life(rul) of battery |
-
2016
- 2016-10-18 CN CN201610906400.8A patent/CN106528951B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103399279A (en) * | 2013-08-01 | 2013-11-20 | 哈尔滨工业大学 | Method for predicting cycle life of fused lithium ion battery based on EKF (Extended Kalman Filter) method and AR (AutoRegressive) model |
| CN103927597A (en) * | 2014-04-22 | 2014-07-16 | 国家电网公司 | Ultra-short-term wind power prediction method based on autoregression moving average model |
| US20160259014A1 (en) * | 2015-03-03 | 2016-09-08 | Samsung Electronics Co., Ltd. | Method and apparatus for automatically estimating remaining useful life (rul) of battery in real time |
| KR20160107093A (en) * | 2015-03-03 | 2016-09-13 | 삼성전자주식회사 | Method and system for predicting remaining useful life(rul) of battery |
| CN104680024A (en) * | 2015-03-11 | 2015-06-03 | 南京航空航天大学 | Method for predicting remaining useful life of lithium ion battery based on GA (Genetic Algorithms) and ARMA (Auto Regressive and Moving Average) models |
| CN105116337A (en) * | 2015-06-30 | 2015-12-02 | 中国人民解放军国防科学技术大学 | Lithium ion battery full charge storage service life evaluation method |
| CN105137358A (en) * | 2015-08-27 | 2015-12-09 | 张家港莫特普数据科技有限公司 | Power battery SOC/SOH prediction method based on large data self-learning mechanism |
Non-Patent Citations (1)
| Title |
|---|
| 刘月峰等: ""锂离子电池循环寿命的融合预测方法"", 《仪器仪表学报》 * |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018227986A1 (en) * | 2017-06-14 | 2018-12-20 | 上海蔚来汽车有限公司 | Power-up optimization method and apparatus, terminal, facility, device, and storage medium |
| CN107273632B (en) * | 2017-06-28 | 2019-09-27 | 北京航空航天大学 | The non-accumulated impact pectus part lifetime of system prediction technique of one kind and device |
| CN107273632A (en) * | 2017-06-28 | 2017-10-20 | 北京航空航天大学 | A kind of non-accumulated impact pectus part lifetime of system Forecasting Methodology and device |
| CN109116244B (en) * | 2018-06-29 | 2021-03-02 | 深圳市科列技术股份有限公司 | Method and device for judging aging trend of power battery |
| CN109116244A (en) * | 2018-06-29 | 2019-01-01 | 深圳市科列技术股份有限公司 | A kind of judgment method and device of power battery aging tendency |
| US11899441B2 (en) | 2018-10-10 | 2024-02-13 | Siemens Aktiengesellschaft | Remaining service life prediction for switches |
| US11065978B2 (en) | 2019-02-25 | 2021-07-20 | Toyota Research Institute, Inc. | Systems, methods, and storage media for adapting machine learning models for optimizing performance of a battery pack |
| US11084387B2 (en) | 2019-02-25 | 2021-08-10 | Toyota Research Institute, Inc. | Systems, methods, and storage media for arranging a plurality of cells in a vehicle battery pack |
| CN110208705A (en) * | 2019-05-09 | 2019-09-06 | 赛尔网络有限公司 | A kind of lithium battery method for predicting residual useful life and device |
| CN116401604B (en) * | 2019-05-13 | 2024-05-28 | 北京绪水互联科技有限公司 | Method for classifying and detecting cold head state and predicting service life |
| CN116401604A (en) * | 2019-05-13 | 2023-07-07 | 北京绪水互联科技有限公司 | Method for classifying and detecting cold head state and predicting service life |
| CN110888077A (en) * | 2019-10-30 | 2020-03-17 | 无锡市产品质量监督检验院 | Accelerated lithium ion battery life evaluation method based on ARIMA time sequence |
| CN110988717B (en) * | 2019-12-10 | 2022-03-25 | 惠州Tcl移动通信有限公司 | Battery detection method, storage medium and terminal device |
| CN110988717A (en) * | 2019-12-10 | 2020-04-10 | 惠州Tcl移动通信有限公司 | Battery detection method, storage medium and terminal device |
| CN112991574A (en) * | 2019-12-13 | 2021-06-18 | 北京亿华通科技股份有限公司 | Method for analyzing attenuation of electric pile |
| CN111190112A (en) * | 2020-02-10 | 2020-05-22 | 宜宾职业技术学院 | Battery charging and discharging prediction method and system based on big data analysis |
| CN112563543B (en) * | 2021-02-23 | 2021-06-25 | 北京亿华通科技股份有限公司 | Method for predicting and maintaining service life of fuel cell system |
| CN112563543A (en) * | 2021-02-23 | 2021-03-26 | 北京亿华通科技股份有限公司 | Method for predicting and maintaining service life of fuel cell system |
| CN113589183A (en) * | 2021-08-09 | 2021-11-02 | 湖北亿纬动力有限公司 | Self-discharge detection method, device and system based on big data |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106528951B (en) | 2019-10-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106528951B (en) | A kind of method and apparatus of power battery life prediction and safe early warning | |
| CN111584952B (en) | Method and system for online evaluation of electrochemical cells of energy storage power station | |
| CN105938183B (en) | A kind of intelligent monitor system in prediction UPS battery service life | |
| CN108537394B (en) | Real-time safety early warning method and device for smart power grid | |
| CN102880989B (en) | A kind of wind power output time series modeling method | |
| CN107301118B (en) | A kind of fault indices automatic marking method and system based on log | |
| CN105445671A (en) | Lithium ion battery service life prediction method based on traceless particle filtering | |
| WO2019196869A1 (en) | Method for determining list of patrolling base stations, and patrolling apparatus | |
| CN105300692A (en) | Bearing fault diagnosis and prediction method based on extended Kalman filtering algorithm | |
| CN103136539A (en) | Grounding grid corrosion rate level prediction method | |
| US11619674B2 (en) | State estimation method and state estimation apparatus | |
| CN105868073A (en) | Data center energy saving strategy implementation method based on time series analysis | |
| CN115600824B (en) | Carbon emission early warning method and device, storage medium and electronic equipment | |
| CN104793607A (en) | Online fault diagnosis, health analysis and failure prediction system and online fault diagnosis, health analysis and failure prediction method for servers | |
| CN108205114A (en) | The Forecasting Methodology and system of battery life | |
| CN110445939A (en) | The prediction technique and device of capacity resource | |
| CN111856309A (en) | Quantitative judgment method for health state of battery | |
| CN110187281A (en) | The method of lithium battery health status estimation based on charging stage health characteristics | |
| US20220283242A1 (en) | Method and device for managing battery data | |
| CN115792505A (en) | Rail transit power supply management method and system | |
| CN103324777A (en) | Android application program energy consumption modeling method based on time | |
| CN112328851A (en) | Distributed power supply monitoring method and device and electronic equipment | |
| CN104618133A (en) | Power grid voltage quality monitoring data collecting method and system | |
| CN116029116A (en) | Intelligent substation wisdom management and control digital twin system | |
| CN103837772B (en) | Low-power-consumption acquisition system life evaluation method based on accelerated life test |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20190429 Address after: 201802 Shanghai Boqiang Microelectronics Co., Ltd. 1255 C-22 Shanghai Jiading District Huyi Highway Applicant after: Shanghai Richpower Microelectronic Co., Ltd. Address before: 215625 Science and Technology Pioneer Park A03, Jinfeng Town, Zhangjiagang City, Suzhou City, Jiangsu Province Applicant before: ZHANGJIAGANG MOTOPU DATA TECHNOLOGY CO., LTD. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |