CN103018673B - Method for predicating life of aerospace Ni-Cd storage battery based on improved dynamic wavelet neural network - Google Patents
Method for predicating life of aerospace Ni-Cd storage battery based on improved dynamic wavelet neural network Download PDFInfo
- Publication number
- CN103018673B CN103018673B CN201210468407.8A CN201210468407A CN103018673B CN 103018673 B CN103018673 B CN 103018673B CN 201210468407 A CN201210468407 A CN 201210468407A CN 103018673 B CN103018673 B CN 103018673B
- Authority
- CN
- China
- Prior art keywords
- data
- dwnn
- prediction
- life
- time series
- 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.)
- Expired - Fee Related
Links
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
A method for predicating life of an aerospace Ni-Cd storage battery based on an improved dynamic wavelet neural network is achieved by the steps: collecting life predication relevant data of all aerospace Ni-Cd storage batteries, pre-processing the life predication relevant data, analyzing data correlation, mapping data and obtaining equivalent data of discharge final voltage of the Ni-Cd storage batteries, improving the DWNN (dynamic wavelet neural network), building primary M-DWNN (1M-DWNN), training and predicating, building self-adaption iteration predicating model on the basis of a secondary M-DWNN (2M-DWNN), training and predicating, and adjusting a dynamic time window. According to the invention, the whole DWNN is adjusted dynamically in the life predication process, thereby ensuring that the predication precision in the whole life predication process is improved continually along with the prolonging of the time and the increment of the data volume.
Description
Technical field
The invention belongs to space flight Ni-Cd life of storage battery electric powder prediction, particularly a kind of space flight Ni-Cd life of storage battery Forecasting Methodology dynamic wavelet neural network based based on modified.
Background technology
The scope that forecasting technique in life span relates to and field are extremely extensive, from raw-material fatigue lifetime to the complicated shaped article life-span, from civil area to national defence, all need forecasting technique in life span.At present, the main forecasting technique in life span for space flight Ni-Cd accumulator can be summarized as follows:
A, the life prediction of physically based deformation model: the materializing procedure of the method to Ni-Cd internal storage battery is analyzed, thus set up the physical model of reflection object evolution process, by related data, model parameter is adjusted, finally obtain the Life Prediction Model needed;
B, the life prediction of Corpus--based Method model hypothesis: first these class methods suppose that the Ni-Cd life of storage battery obeys certain statistical distribution, and utilize a large amount of existing lifetime data to determine the parameter of this model, thus set up the Life Prediction Model of Ni-Cd accumulator;
C, life prediction based on aging effects factor analysis training: the method is mainly through studying and determining to affect each aging effects factor of Ni-Cd accumulator, and set up influence factor with the incidence relation between the life-span by a large amount of testing data of life-span, thus set up the Life Prediction Model of Ni-Cd accumulator.
For above-mentioned three kinds of methods, the life prediction of physically based deformation model needs the internal mechanism furtheing investigate Ni-Cd accumulator, and its workload is huge and portability is relatively poor, and the Ni-Cd accumulator for different model need set up Life Prediction Model respectively; Corpus--based Method model hypothesis and based on aging effects factor analysis training life prediction then need a large amount of Ni-Cd life of storage battery data to set up Life Prediction Model.Consider in practical engineering application, space flight Ni-Cd accumulator, be often subject to the restriction of various objective condition, a large amount of lifetime datas for life prediction can not be there is.Thus, study a kind of significant for the life-span prediction method in Ni-Cd accumulator few lifetime data situation.
Artificial neural network (Artificial Neural Network, ANN) is a simulation cerebral nervous system 26S Proteasome Structure and Function, the artificial network extensively connected to form by a large amount of simple process unit and neuron.It can from given data automatic sorting rule, obtain the inherent law of these data, there is very strong non-linear mapping capability.Artificial neural network has following outstanding advantages: 1. the concurrency of height; 2. the non-linear overall situation effect of height; 3. good fault-tolerance and function of associate memory; 4. self-adaptation, the self-learning function of ten points strong.
The patent that applicant had previously applied for, application number 20101022095.1, name is called: a kind of life-span prediction method of the Small Sample Database object based on dynamic bipolar MPNN, by collecting all data availables of life prediction object and carrying out the pre-service of life prediction related data and data dependence analysis to it; Obtain these aging effects factors with funtcional relationship between forecasting object life-span characterization parameter.Then data-mapping obtain the equivalent data value of forecasting object; The training of a MPNN network and prediction and the training of secondary MPNN network and prediction is carried out by the MPNN network improved; Finally according to the life value of the end-of-life criterion determination forecasting object of forecasting object life-span characterization parameter.This patent is predicted based on the life of storage battery of Small Sample Database by bipolar MPNN real-time performance, gives the solution being directed to a class problem with Small Sample Database feature, has stronger versatility.Due to this patent core be once with the precision of prediction of secondary MPNN, and wherein MPNN network is formed by the PNN network improvement with statistical property, make MPNN network while remaining PNN network advantage, have also been introduced the deficiency that can not embody single battery characteristic to be predicted very well; In addition, in this patent, the list that the iteration of secondary MPNN Web vector graphic is predicted as not repetition training props up single step iteration prediction, make life of storage battery precision of prediction on the low side, and be only in slump of disastrous proportions its life prediction precision in period at accumulator and could ensure to some extent, which greatly limits the engineer applied of life of storage battery prediction.
Dynamic wavelet neural network based (Dynamic Wavelet Neural Networks, DWNN) is the one of artificial neural network, this network by inputting, WNN, output and output feedack four parts form, as shown in Figure 2.
Wherein, U is outside input, and N is outside input dimension; Y is for exporting; M is output feedack node number; WNN is standard static wavelet neural network.The expression formula of DWNN is:
Y(t+1)＝WNN(Y(t),…,Y(t-M+1),U(t),…,U(t-N))
Because DWNN network builds network recurrence from multi-angle, strengthen the memory capacity to historical information, dynamic characteristic of the course is presented in computation process, than feedforward neural network and existing Recurrent Wavelet Neural Network, there is stronger dynamic behaviour and computing power, and there is and individual character overall to forecasting object simultaneously carry out the advantages such as the ability of resolved analysis, be widely used in practical engineering project.The present invention solves on the basis of Small Sample Database object lifetime prediction thought in succession patented claim 20101022095.1, first DWNN is improved, the DWNN model after improvement is enable to follow the tracks of the degenerative process of accumulator more subtly, simultaneously, adaptive iteration Forecasting Methodology is utilized to promote re prediction precision, and then, increase substantially life of storage battery precision of prediction, to meet the life prediction demand of Ni-Cd object and data characteristics thereof.
Summary of the invention
The technology of the present invention is dealt with problems: overcome the deficiencies in the prior art, a kind of space flight Ni-Cd life of storage battery Forecasting Methodology dynamic wavelet neural network based based on modified is provided, the method is on the basis remaining application number 20101022095.1 advantage, on the one hand, by having the improvement of the DWNN network of stronger resolved analysis ability to overall and individual information, make the present invention have resolution characteristic to monomer forecasting object, and enable the DWNN model after improvement follow the tracks of the degenerative process of accumulator more subtly; On the other hand, utilize adaptive iteration process to construct 2M-DWNN model, life of storage battery precision of prediction is ensured very well; And then, on comprehensive M-DWNN model and adaptive iteration model basis, significantly improve the precision of life prediction, further overcome the deficiency that the precision of application 20101022095.1 existence is on the low side.
Space flight Ni-Cd life of storage battery Forecasting Methodology overall procedure provided by the invention as shown in Figure 1, realizes especially by following steps:
Step one, collect all Ni-Cd life of storage battery prediction related data;
By to Ni-Cd accumulator and like product analysis, collect utilizable all life prediction related datas.
Step 2, the pre-service of life prediction related data;
The life prediction data analysis that step one is obtained and screening, extract electric discharge final pressure supplemental characteristic (the Ni-Cd life of storage battery in the present invention stops criterion) required for the present invention and aging effects factor data, comprising: discharge current data, charging current data, charge and discharge cycles number of times and depth of discharge data etc.Meanwhile, the pretreatment work such as singular value rejecting, Noise reducing of data are carried out to the related data filtered out.
Step 3, data dependence analysis;
Consider the depth of discharge aging effects factor of like product and Ni-Cd accumulator, need to carry out correlation analysis by approximation of function or SPSS method to like product and pretreated parameter corresponding to Ni-Cd accumulator, thus obtain the correlationship between like product and Ni-Cd accumulator in electric discharge final pressure.
Described correlation analysis refers to by approximation of function or utilizes SPSS (statistical product and service solution---Statistical Product and Service Solutions), realize the correlation analysis to different parameters data, thus obtain the mapping relations between supplemental characteristic.
Step 4, data-mapping obtain the equivalent data value of Ni-Cd battery discharging final pressure;
Utilize step 3 to obtain under the effect of each aging effects factor, the like product obtained is with the correlationship between Ni-Cd battery discharging final pressure, based on the electric discharge final pressure data of the like product of reference, map and obtain the equivalent data value of Ni-Cd battery discharging final pressure, " equivalent-EoDV " curve as shown in Figure 4, wherein, dashed curve represents existing final pressure (EoDV) data of discharging in-orbit of Ni-Cd accumulator; Solid-line curve is final pressure (ED-EoDV) curve that discharges with the EoDV curve equivalent had under part same affect conditions in-orbit.
The improvement (M-DWNN) of step 5, DWNN network;
According to space flight Ni-Cd life of storage battery precision of prediction demand, improve DWNN, improvement project is as follows: utilize DWNN export data sequence Yi} build AR model (in the present invention: Yi} is EoDV_orbi}), comprise and determine AR model order p and coefficient a
i, i=0,1,2 ..., p-1, and then, using the exponent number of AR model as DWNN model feedback interstitial content, even the coefficient of M=p, AR model is as the weight coefficient of each feedback node, and as WNN mode input, and then complete the improvement to DWNN network model.Wherein, the corresponding relation of AR model coefficient and node is: a
iy (t-i), i=0,1,2 ..., p-1, as shown in Figure 5.
Step 6, a M-DWNN(1M-DWNN) foundation of network, training and prediction;
According to analysis of Influential Factors and association analysis result, determine 1M-DWNN input number of nodes, export as Ni-Cd battery discharging final pressure (single output) after treatment.Utilize Ni-Cd accumulator in-orbit EoDV and the ED-EoDV that obtained by step 4 do the process of difference ratio, as shown in Figure 4, the training sample of structure 1M-DWNN and test sample book.In order to reject the singular value in training sample, accelerate the speed of convergence of network, be normalized by the input vector of above-mentioned structure, object vector, then input 1M-DWNN network to train it, thus determine 1M-DWNN network parameter, and utilize the 1M-DWNN network trained to predict, divide Procedure Acquisition Ni-Cd battery discharging final pressure predicted value by renormalization and contrast, and then to realize with O-c section curve, for reference, obtaining electric discharge final pressure and predicting the outcome data segment a-e.
Step 7, based on secondary M-DWNN(2M-DWNN) the adaptive iteration forecast model of network is set up, training and prediction;
Determine 2M-DWNN input number of nodes and output node number (single output).Utilize the electric discharge final pressure of Ni-Cd battery discharging final pressure data and 1M-DWNN after pretreatment to predict the outcome, the training of structure 2M-DWNN and test sample book, i.e. adaptive iteration prediction, its flow process as shown in Figure 6.And then carry out prediction work, finally stop according to the Ni-Cd life of storage battery life value that criterion determines Ni-Cd accumulator.
Wherein, adaptive iteration forecast model is described below:
(1) data encasement (average-slope time series structure) of adaptive iteration prediction
Iteration forecast model in the present invention, segmentation average value processing is carried out to o-e section electric discharge final pressure value, and on this basis, the slope of computation of mean values, namely secondary M-DWNN prediction carries out adaptive iteration prediction to this average-slope time series, Figure 7 shows that average-slope sequence generation figure.
(j=1,2,…,n;n=fix(N/interval)) Equ.1
Wherein, { x
irepresent EoDV value sequence, length is N, interval is mean value interval, and { Avr (j) } is equal value sequence, and length is n, { s (k) } for average-slope sequence, length be n-1,
Subsequent content from step (2), is described the core content of adaptive iteration Forecasting Methodology for object with { s (k) } average-slope sequence.
(2) adaptive iteration process
After step (1) construction complete average-slope time series, start the core content performing adaptive iteration prediction, concrete grammar is as follows:
(2.1) auto-adaptive time sequence data double sampling
For the time series { A that given length is n
i j, wherein, i, j represent original time series iteration prediction number of times and seasonal effect in time series double sampling interval respectively, then A
0 1:={ s (k) }=s
1, s
2, s
3..., s
k..., s
n-1, s
n, represent raw data, iteration prediction number of times is 0, and sampling for time series interval is defined as 1, wherein ' :=' represent ' being defined as '; And A
1 2then represent that double sampling is spaced apart the time series data of 2 after an iteration prediction.Equally spaced from A according to the needs of 2M-DWNN training sample amount
0 1in carry out resampling, and then obtain new time series:
A
0 j:
S
n-(jn-3)，*j’…,S
n-(i+1)*j,S
n-i*j，…,S
n-2*j,S
n-j,S
n
[j＝1,2,3,…,jmax,n-(j
n-1)*j≥1，j
n≥Sample_size_min]
Wherein: n is original time series A
0 1length; A
0 jrepresent and be spaced apart j with resampling and obtain time series, and from last data (s of raw data
n) start sampling; j
nrepresent time series A
0 jlength, and meet: n-(j
n-1) * j>=1, and j
nalong with the increase of j constantly reduces.Suppose that the demand minimum value of 2M-DWNN training sample amount is Sample_size_min, then j
nj should be met
n>=Sample_size_min.Order, j
maxthe maximal value that=fix (n/ (Sample_size_min)) is j, wherein fix represents and rounds downwards.So complete self-adaptation sampling process again.
(2.2) Single-step Prediction of self-adaptation double sampling time series data
Utilize Single-step Prediction thought, build 2M-DWNN model (building thought identical with 1M-DWNN, is all adopt to determine rank and coefficient weights method based on AR model) respectively, Single-step Prediction is carried out to the time series data obtained through double sampling.A can be obtained
0 jone-step prediction value, and then obtain new time series A
0 j' can be expressed as:
…,S
n-(i+1)*j,S
n-i*j,…,S
n-2*j,S
n-j,S
n,S′
n+j
(j＝1,2,3,…jmax)
Same process completes all A
0 jtime series, wherein j=1,2,3 ..., j
max.Thus complete an iterative process, and obtain j
maxindividual predicted value, the new time series formed can be expressed as:
A
1 1:S
1,S
2,…,S
n-1,S
n,S′
n+1,S′
n+2,…,S′
n+jmax
Two steps achieve by random time sequence A above
0 1obtain j
maxnew time series A after individual predicted value
1 1, and then complete an iteration forecasting process.When with A
1 1substitute raw data A
0 1and repeating said process, iteration like this can obtain A
2 1, A
3 1..., A
k 1...
(3) adaptive prediction iterative data result inverse transformation and Ni-Cd life value judge
By iterative process, acquisition average-slope sequence A that can be infinitely many
k 1, in order to finally judge the life-span of accumulator, need predicting the average-slope sequence A obtained
k 1carry out inverse transformation, and obtain Ni-Cd accumulator electric discharge final pressure sequence x '
i, data that this sequence comprises existing final pressure data of discharging in-orbit, 1M-DWNN predicts the outcome data and 2M-DWNN iteration predicts the outcome.Based on x '
isequence, determine Ni-Cd accumulator predicting residual useful life value by life of storage battery criterion.
Step 8, dynamic time windows adjust;
In the process utilizing M-DWNN to predict Ni-Cd accumulator, due to the passing of time, constantly new data can be obtained.Now, needing dynamically (every N days is an adjustment cycle, N determines according to actual needs) repeat above-mentioned steps two to step 7, rebuild and train M-DWNN network (comprising 1M-DWNN and 2M-DWNN model) to upgrade network parameter and again to predict, guarantee to ensure the prolongation along with time in orbit, precision of prediction can constantly promote.
Illustrate: in life prediction in earlier stage, stable system performance, thus, final pressure of discharging in long period section change slowly.During this period of time, ' the dynamic conditioning time window ' of network training can suitably relax, as: two weeks or 1 month etc.Along with the carrying out of prediction, the performance of system starts progressively to fail, and suitably should reduce the time window width (as: a week) of life prediction, to obtain accumulator remaining lifetime value more in time and accurately.
The present invention's advantage is compared with prior art:
(1) DWNN has the ability that and individual character overall to forecasting object carries out resolved analysis, makes the present invention effectively in conjunction with single battery and overall information, and can provide the result of cell batteries life prediction accurately;
(2) DWNN is improved, enable the DWNN model after improvement follow the tracks of the degenerative process of accumulator more subtly, thus in model is set up, first ensure that the precision of prediction;
(3) the adaptive iteration technology adopted has carries out iteration predictive ability according to real data self-adaptation, improves re prediction precision;
(4) Comprehensive Model comprising 1M-DWNN and 2M-DWNN of the present invention's proposition, significantly improve life prediction precision on the whole, especially run on non-slump of disastrous proportions period at accumulator, its life of storage battery precision of prediction can promote about order of magnitude.
Accompanying drawing explanation
Fig. 1 is space flight Ni-Cd life of storage battery prediction process flow diagram;
Fig. 2 is the dynamic wavelet neural network based illustratons of model of existing classics;
Fig. 3 is the space flight Ni-Cd life of storage battery PREDICTIVE CONTROL logical diagram based on modified DWNN;
Fig. 4 is space flight Ni-Cd life of storage battery forecast model process of establishing schematic diagram in the present invention;
Fig. 5 is the dynamic wavelet neural network based illustraton of model of modified;
Fig. 6 is the adaptive iteration prediction process flow diagram in the present invention;
Fig. 7 is average-slope sequence generation figure;
Fig. 8 is electric discharge final pressure raw-data map;
Fig. 9 for discharging final pressure data plot after outlier processing and noise reduction;
Figure 10 is 1M-DWNN model structure figure;
Figure 11 is that 1M-DWNN predicts measuring accuracy curve map;
Figure 12 is the Ni-Cd battery discharging final pressure decline prediction curve figure based on 1M-DWNN model;
Figure 13 is that the comprehensive M-DWNN comprising 2M-DWNN model prediction result predicts the outcome figure.
Embodiment
Below in conjunction with drawings and Examples, the present invention is described in detail.
The present invention is a kind of space flight Ni-Cd life of storage battery Forecasting Methodology dynamic wavelet neural network based based on modified, described life-span prediction method is a kind of nonparametric technique, the method obtains M-DWNN network after improving existing DWNN network, after to the analysis of Ni-Cd battery data and pretreatment work, the data processed are utilized to form training set and the test set of 1M-DWNN network, after training study, 1M-DWNN neural network forecast is utilized to supplement historical data sample; After completing these work, 2M-DWNN network namely can be utilized to carry out iteration life prediction work, finally determine Ni-Cd life of storage battery value according to end-of-life criterion, Figure 1 shows that the overview flow chart of life-span prediction method of the present invention, Fig. 3 is life prediction steering logic figure, and concrete implementation step is as follows:
Step one, all available life prediction related datas of collection Ni-Cd life of storage battery prediction;
By to space flight Ni-Cd accumulator and like product analysis thereof, collect utilizable all life prediction related datas.Comprise: the supplemental characteristics such as temperature, charging current, discharge current, discharge electricity amount, load current data, charge capacity, dump energy, voltage, discharge and recharge ratio; For solar battery array, its data available is: the supplemental characteristics such as temperature, square formation current data, load current data, power, busbar voltage, stepup transformer output voltage.
Described like product refers in physical arrangement, logical organization and functional structure, with the product that Ni-Cd accumulator is similar or identical.
Step 2, the pre-service of life prediction related data;
The life prediction related data obtained in step one is analyzed and screened, extracts electric discharge final pressure data and the aging effects factor of Ni-Cd accumulator, and aging effects factor is classified.
The electric discharge final pressure (EoDV) of described Ni-Cd accumulator and supplemental characteristic as follows:
(1) the life-cycle data of the electric discharge final pressure (EoDV_sim) of like product, namely bring into use the historical time sequence data to this like product life termination from like product, its time series data can be expressed as: { EoDV_simi};
(2) Ni-Cd accumulator discharges the part probabilistic information of final pressure (EoDV_orb) in orbit, and namely bring into use up to the present all time series datas from Ni-Cd accumulator, its time series data can be expressed as: { EoDV_orbi}.
Described aging effects factor and aging effects factor data can be divided into following two classes:
(1) time series aging effects factor data: the aging effects factor data with Ni-Cd battery discharging final pressure in the present invention with same time yardstick, these influence factor data comprise: charging current data (Charge Current:CC), discharge current (Discharge Current:DC), charge and discharge cycles number of times, described aging effects factor data comprises life-cycle data and the existing time series data of Ni-Cd life of storage battery influence factor in-orbit of like product aging effects factor, and the form of the composition of like product data and expression way and in-orbit Ni-Cd battery discharging final pressure and aging effects factor similar, specifically can be divided into: like product charging current, like product discharge current, like product charge and discharge cycles data and in-orbit Ni-Cd battery charging current, Ni-Cd battery discharging electric current in-orbit, Ni-Cd accumulator cell charging and discharging cycle index in-orbit,
(2) data point reuse aging effects factor data: be different from time series aging effects factor data, these type of influence factor data are that finite number is according to right, these data to the historical data being like product, reaction be that corresponding aging effects factor parameter is with the corresponding relation between the life-span.For local directed complete set Ni-Cd battery discharging final pressure (EoDV_sim, EoDV_orb) and the time series data such as charging current data (Charge Current:CC), discharge current (Discharge Current:DC).For Ni-Cd accumulator, depth of discharge is this type of aging effects factor, and when depth of discharge (Depth of Discharge:DOD) is 17%, the life value of its correspondence is 20,000 charge and discharge cycles, forms data thus to (17%-20000).
After completing aging effects factor classification, to the electric discharge final pressure data of above-mentioned Ni-Cd accumulator (EoDVi: the electric discharge final pressure data (EoDV_orbi) comprising like product (EoDV_simi) and Ni-Cd accumulator in-orbit), charging current data discharge current, depth of discharge (comprise like product and in-orbit data) carries out singular value rejecting, Noise reducing of data pre-service, and then the data { EoDV_p_simi} obtained through data prediction, { EoDV_p_orbi}, { CC_p_simi}, { CC_p_orbi}, { DC_p_simi}, { DC_p_orbi}, { C_simi}, { C_orbi}, { DOD_simi}, { DOD_orbi}.In the process of carrying out life prediction, the aging effects factor data of Ni-Cd accumulator and in forecasting process to the operation of these data, must with the aging effects factor data of like product and operation one_to_one corresponding thereof.
Step 3, data dependence analysis;
Utilize SPSS to carry out correlation analysis to the depth of discharge aging effects factor data screened in step 2, thus obtain these aging effects factors with funtcional relationship f between Ni-Cd battery discharging final pressure data.
Step 4, data-mapping obtain the equivalent time sequence data value of forecasting object;
Correlationship f between the like product utilizing step 3 to obtain and Ni-Cd accumulator in electric discharge final pressure, based on like product electric discharge final pressure data, map and obtain the equivalent electric discharge final pressure data of Ni-Cd battery discharging final pressure { EDEoDVi}, if Fig. 4 solid line is ' shown in equivalent-EoDV '.
The improvement (M-DWNN) of step 5, DWNN network;
With the time series data of battery discharging final pressure in-orbit, { EoDV_p_orbi}, for data object, builds AR model, determines AR model order p and coefficient a
i, i=0,1,2 ..., p-1, even the coefficient of M=p, AR model is as the weight coefficient of each feedback node, and as WNN mode input, and then completes the improvement to DWNN network model, and wherein, the corresponding relation of AR model coefficient and node is: a
iy (t-i), i=0,1,2 ..., p-1, as shown in Figure 5.All need to build AR model respectively in 1M-DWNN and 2M-DWNN, to realize the improvement to DWNN.Wherein, 1M-DWNN needs structure AR model, and 2M-DWNN needs to build j in each iterative process
maxindividual AR model, and process of establishing is all based on available electric discharge final pressure in-orbit data (for iteration prediction, its time series data is average-slope sequence that is existing and prediction).Described j
maximplication is defined by summary of the invention step 7 (2.1).
Step 6, a M-DWNN(1M-DWNN) foundation of network, training and prediction;
{ { ED-EoDVi} carries out the process of difference ratio, and exports as 1M-DWNN network for EoDV_p_orbi} and the equivalent data value of Ni-Cd accumulator that obtained by step 4 to put a final pressure to the Ni-Cd accumulator obtained by step 2.To the time series aging effects factor data obtained by step 2 (comprise like product and in-orbit Ni-Cd accumulator charging current, put an electric current, charge and discharge cycles number of times) carry out the process of difference ratio, as 1M-DWNN network input parameter, so far, input vector and the object vector of 1M-DWNN network is constructed.
In order to realize rejecting the singular value in training sample, accelerating the speed of convergence of network, being normalized by the input vector of above-mentioned structure and object vector; Then input 1M-DWNN network to train it; Finally, utilize the 1M-DWNN neural network trained, index of aging parameter is predicted.Predicted value sequence { the EoDV_pre_i} of 1M-DWNN is obtained after renormalization and contrast divide.Detailed process is described below:
(1) AR model is set up
To the Ni-Cd accumulator obtained by step 2 put a final pressure equivalent data value of EoDV_p_orbi} and the Ni-Cd accumulator that obtained by step 4 ED-EoDVi} carries out the process of difference ratio, and for setting up AR model, and then, obtain AR model order p and coefficient { a
i, the basic representation of AR model is as follows:
AR:Y(t+1)＝a
0Y(t)+a
1Y(t-1)+…+a
pY(t-p)+e(t)
Wherein, e (t) for average be the white noise signal of 0.
(2) 1M-DWNN difference ratio I/O
As shown in Figure 3, after above-mentioned steps is analyzed, 1M-DWNN of the present invention is that (outside is three inputs to triple input single output network, inner input number is determined, under not adding specified otherwise situation in this patent by AR model order p, all inputs all refer to outside input), its input can be described as:
1) the difference CC(charging current that EoDV is corresponding in-orbit { CC_p_orbi}, the charging current corresponding with EDEoDV curve { between CC_p_simi} difference) (linear normalization);
2) the difference DC(discharge current that EoDV is corresponding in-orbit { discharge current { between DC_p_simi} difference) (linear normalization) that DC_p_orbi} and ED-EoDV curve is corresponding;
3) cycle index (charge and discharge cycles number of times that ED-EoDV curve is corresponding { C_orbi}) (arc tangent normalization);
Output is: difference EoDV(in-orbit EoDV_p_orbi} and { between EoDV_p_simi} difference) (linear normalization), the expression after normalized is:
Wherein: CC_p_orbi and CC_p_simi represents in i-th cycle period respectively, the discharge current value that EoDV curve is corresponding with ED-EoDV curve in-orbit; DC_p_orbi and DC_p_simi represents in i-th cycle period respectively, the discharge current value that EoDV curve is corresponding with ED-EoDV curve in-orbit; C_orbi is charge and discharge cycles number of times, EoDV_p_orbi and ED_EoDVi is respectively the electric discharge final pressure value on EoDV curve and ED-EoDV curve in-orbit.As shown in Figure 3, corresponding Δ EoDV is also existed for arbitrfary point C_orbi
i, and namely the existence of this species diversity is by charging current, discharge current and the coefficient result of charge and discharge cycles number of times.Thus, Δ EoDV can be thought
ibe the function of input vector, and this funtcional relationship can be expressed by Equ.2.
(3) 1M-DWNN training, test and prediction
Utilize by EoDV data (from O-a section) in-orbit existing in Fig. 4 and corresponding ED-EoDV data (from O point to b point section) thereof, the input vector of discharge current, charging current and charge and discharge cycles number of times structure 1M-DWNN network and object vector, train/test and set up 1M-DWNN network.Prediction based on 1M-DWNN inputs, and prediction obtains a-e segment data curve based on b-c segment data.
Step 7, based on secondary M-DWNN(2M-DWNN) the adaptive iteration forecast model of network is set up, training and prediction;
By through the pretreated Ni-Cd accumulator of step 2 electric discharge final pressure data o-a segment data corresponding in EoDV_p_i}(Fig. 4) and obtained by step 6 1M-DWNN neural network forecast electric discharge final pressure data a-e segment data corresponding in EoDV_pre_i}(Fig. 4) based on, after determining Sample_size_min and interval, generate average-slope time series A according to average-slope seasonal effect in time series building method
j 0.
Described from 2M-DWNN forecasting process, predicted by adaptive iteration, can be implemented in a large amount of predicted values under Single-step Prediction.In whole 2M-DWNN model process of establishing, because the data prediction after double sampling and iteration prediction are all identical in itself, thus, here only with A
j isequence is that the implementation process of example to 2M-DWNN is described.
(1) AR model is set up
With the A generated
j i(under given iterations i and double sampling are spaced apart j condition, the sequence obtained, average as shown in Figure 6)-slope time series is that to carry out single step time series be data bases on data basis, sets up classical AR model, and the exponent number p obtained in AR model and coefficient { a
i, the basic representation of AR model is as follows:
AR:X(t+1)＝a
0X(t)+a
1X(t-1)+…+a
pX(t-p)+e(t)
Wherein, e (t) for average be the white noise signal of 0.Thus, determine feedback element nodes and the weight coefficient of 2M-DWNN model, thus build A
j i2M-DWNN model under data qualification.
(2) the data I/O of 2M-DWNN
With the A generated
j iaverage-slope time series is that single step time series forecasting is carried out on data basis, and as shown in Figure 3, the input number of nodes of 2M-DWNN network is determined in the experience predicted according to the Ni-Cd life of storage battery or by experiment test, and its output node number is 1.
(3) 2M-DWNN training, test and prediction
Export so that 4 inputs are single, now, the single step function logic relation of 2M-DWNN network can be expressed as follows:
k＝EoDV_end,(EoDV_end-j),(EoDV_end-2·j),…,(EoDV_end-(Len_j-5)·j)
Wherein, EoDV_end, Len_j are respectively last value of the EoDV sequence formed after i iteration prediction, and on EoDV basis, the time sequential length of the new sequence obtained of sampling with interval j.So, progressively based on the average-slope time series A set up
1 icomplete adaptive iteration forecasting process, finally average-slope inverse transformation is carried out to the predicted value obtained, obtain electric discharge final pressure sequence { x (i) }, coordinate figure d (the C_end1 of the e-d section shown in Fig. 4 EoDV data and curves and mark end-of-life (life-span criterion) position d point in-orbit can be obtained, EoDV_threshold), wherein C_end1 is the bimetry of Ni-Cd accumulator.
Step 8, dynamic time windows adjust;
In Practical Project, As time goes on, the data volume of Ni-Cd accumulator on above-mentioned parameter increases gradually, and new data effectively can promote life prediction precision.According to Ni-Cd accumulator feature and request for utilization, corresponding dynamic time window values is set, as: one month, two weeks, one week etc.Afterwards according to the time window values arranged, repeat above-mentioned steps two to step 7, re-training and prediction are carried out to 1M-DWNN and 2M-DWNN network, redefines the life value of Ni-Cd accumulator.
The present invention on the basis of modified dynamic wavelet neural network based (DWNN network) by reasonably analyze and organization network input and output data, modified of structure respectively stage by stage dynamic wavelet neural network based (1M-DWNN) and secondary modified dynamic wavelet neural network based (2M-DWNN), utilize these two the modified neural network predictions built, and obtain the life value of Ni-Cd accumulator.In addition, the present invention can upgrade life prediction data under given time window prerequisite, and neural network training again, and then obtains the life value of Ni-Cd accumulator under new data.User can understand the remaining life of forecasting object by life prediction, thus by the configuration environment for use control Ni-Cd life of storage battery and foundation can be provided for logistics management carries out decision-making, make ensureing under the prerequisite that the topmost task of Ni-Cd accumulator is implemented, utilize to greatest extent, reliably and use Ni-Cd accumulator, thus giving full play to the usefulness of Ni-Cd accumulator.
Below with China space flight HY-1B moonlet 30AhNi-Cd accumulator for object, like product is ground experiment 45Ah Ni-Cd accumulator is that example further illustrates.Because the life-span related data of small satellite origin system is few, meet the Small Sample Database life prediction problem that will solve required for the present invention.Elaborating by the present embodiment, further illustrates implementation process of the present invention and engineer applied process.
For HY-1B moonlet Ni-Cd accumulator, the data that may be used for life prediction are: battery discharging voltage data (fragmentary data) in-orbit, battery discharging current data in-orbit (fragmentary data), battery charging current data (fragmentary data) in-orbit, battery discharging depth data (fragmentary data) in-orbit, ground accumulator test electric discharge final pressure data (complete data), ground accumulator test discharge current data (constant), ground accumulator test charge current data (constant), depth of discharge and the corresponding life value 5 of ground accumulator test depth of discharge data (constant) and history are right.
Described " fragmentary data " represents because accumulator still can normally work in-orbit, is only the front portion time series data of complete lifetime data.
Utilizing the basis of the above-mentioned data for life prediction, the life-span of life-span prediction method to moonlet accumulator that application the present invention proposes is predicted, the step and method of its application is as follows:
Step one, all available life prediction related datas of collection Ni-Cd life of storage battery prediction;
By analyzing moonlet accumulator, it is as follows to collect utilizable all related datas:
Forecasting object data---moonlet Ni-Cd accumulator in-orbit:
(1) battery discharging voltage data (fragmentary data) in-orbit;
(2) charge in batteries voltage data (fragmentary data) in-orbit;
(3) battery discharging current data in-orbit (fragmentary data);
(4) battery charging current data (fragmentary data) in-orbit;
(5) battery discharging electric quantity data (fragmentary data) in-orbit;
(6) charge in batteries electric quantity data (fragmentary data) in-orbit;
(7) battery temp data (fragmentary data) in-orbit;
(8) accumulator cell charging and discharging ratio (fragmentary data) in-orbit;
Like product data---ground experiment Ni-Cd accumulator:
(1) ground experiment Ni-Cd battery discharging final pressure data (complete data);
(2) ground experiment Ni-Cd battery discharging current data (constant);
(3) ground experiment Ni-Cd battery charging current data (constant);
(4) ground experiment Ni-Cd battery discharging depth data (constant);
(5) depth of discharge of history and corresponding life value 5 right;
Step 2, the pre-service of life prediction related data;
The all related datas obtained in step one are reasonably analyzed, in conjunction with the above-mentioned description of this invention, extract can be used for electric discharge final pressure data and aging effects factor data as follows:
(1) life-span characterization parameter:
The data of battery discharging final pressure in-orbit EoDV_orbi}---can be extracted by ' in-orbit moonlet battery discharging voltage ' and obtain;
Ground experiment Ni-Cd battery discharging final pressure data { EoDV_simi};
(2) aging effects factor:
Ground experiment Ni-Cd battery charging current (CC_sim);
Ground experiment Ni-Cd battery discharging electric current (DC_sim);
Ground experiment Ni-Cd accumulator cell charging and discharging loop-around data (C_sim);
Ni-Cd battery charging current (CC_orb) in-orbit;
Ni-Cd battery discharging electric current (DC_orb) in-orbit;
Ni-Cd accumulator cell charging and discharging cycle index (C_orb) in-orbit;
Battery discharging depth data (DOD_orb) in-orbit---can be extracted by ' in-orbit battery discharging electric quantity data ' and obtain;
Ground experiment Ni-Cd battery discharging depth data (DOD_sim);
Depth of discharge and the corresponding life value 5 of history are right;
Singular value rejecting is carried out to above-mentioned data, Noise reducing of data pretreatment work.Fig. 8, Fig. 9 are for picking ground accumulator test electric discharge final pressure data plot before and after process and noise reduction through singular value.Contrast as can be seen from figure, the data after picking value and noise reduction are more regular, undulatory property is less.In addition, need to obtain the data of battery discharging final pressure in-orbit, to characterize performance and the life status of accumulator in-orbit from battery discharging voltage in-orbit.
Step 3, data dependence analysis;
By approximation of function or SPSS, correlation analysis is carried out to ground experiment Ni-Cd accumulator and Ni-Cd accumulator is corresponding in-orbit pretreated parameter.According to actual in-orbit 17% depth of discharge of Ni-Cd accumulator and ground experiment Ni-Cd accumulator 30% depth of discharge set up the Ni-Cd life of storage battery with reference the ground experiment Ni-Cd life of storage battery between correlationship f:1:1.7.
Step 4, data-mapping obtain the equivalent time sequence data value of forecasting object;
Utilize the correlationship f:1:1.7 that step 3 obtains, based on ground experiment Ni-Cd battery discharging final pressure data, map and obtain the equivalent electric discharge final pressure data of Ni-Cd battery discharging final pressure { EDEoDVi}, if Fig. 4 solid line is ' shown in equivalent EoDV '.
The improvement (M-DWNN) of step 5, DWNN network;
The concrete improvement of this part and optimum configurations are set up see the AR model in step 6, seven.
Step 6, a M-DWNN(1M-DWNN) foundation of network, training and prediction;
With { EoDV_p_orbi} and { the differential data result of ED-EoDVi} for object, constructed AR model parameter is as being p=4, a
i=-0.632,0.545 ,-0.021,0.162}, and then, 1M-DWNN feedback node number M=4, and weighting coefficient is known.Because the input number of nodes of 1M-DWNN network is three, based on Equ.2, establish institute and set up 1M-DWNN model as shown in Figure 10.
According to the constrained input data demand of 1M-DWNN network, utilize the battery charging current of Ni-Cd in-orbit (CC_p_orb) after normalized, Ni-Cd battery discharging electric current (DC_p_orb) in-orbit, final pressure data of discharging in-orbit { EoDV_p_orbi}, ground experiment Ni-Cd battery charging current (CC_p_sim), ground experiment Ni-Cd battery discharging electric current (DC_p_sim) charge and discharge cycles number of times and the equivalent electric discharge final pressure data { input vector of ED_EoDVi} structure 1M-DWNN network and the object vector that are obtained by step 4, train/test and set up 1M-DWNN network, train and predict that relevant design parameter is as follows:
Training time: 2.3153s
Precision of prediction: square error MSE=0.0143
Figure 11 shows that 1M-DWNN predicts measuring accuracy curve, as can be seen from the figure 1M-DWNN can be good at following the tracks of accumulator property degenerative process.
1M-DWNN network exports as difference normalized value corresponding under corresponding charge and discharge cycles number of times, final pressure value of discharging in-orbit can be obtained after renormalization and contrast divide, Figure 12 is the Ni-Cd battery discharging final pressure decline prediction curve based on 1M-DWNN model, has followed the tracks of existing final pressure data trend of discharging in-orbit generally well.
Step 7, based on secondary M-DWNN(2M-DWNN) prediction of the adaptive iteration of network is set up, training and prediction;
Make Sample_size_min=100, interval=8, by through the pretreated Ni-Cd accumulator of step 2 electric discharge final pressure data o-a segment data corresponding in EoDV_p_i}(Fig. 4) and obtained by step 6 1M-DWNN neural network forecast electric discharge final pressure data a-e segment data corresponding in EoDV_pre_i}(Fig. 4) based on, generate average-slope time series A according to average-slope seasonal effect in time series building method
j 0.
For 2M-DWNN, all need to build j in each iterative process
maxindividual AR model, and process of establishing is all with A
j 0(for the data that iterations is greater than 0, the data basis of AR model construction is A
j i, wherein i>0) and be data basis, and limit for any A
j idouble sampling sequence j
maxfor definite value, so ensure that in each iterative process, the amount of constructed AR model is identical.Based on the AR model built, the 2M-DWNN model of 4 inputs/mono-output is set up according to Equ.3, complete adaptive iteration forecasting process, finally average slope inverse transformation is carried out to the predicted value obtained, obtain electric discharge final pressure sequence { x (i) }, can obtain the coordinate figure d (C_end1, EoDV_threshold) of the e-d section shown in Fig. 4 EoDV data and curves and mark end-of-life (life-span criterion) position d point in-orbit, wherein C_end1 is the bimetry of Ni-Cd accumulator.
Figure 13 is that the comprehensive M-DWNN comprising 2M-DWNN model prediction result predicts the outcome, and wherein right side thick line portion is depicted as second iteration and predicts the outcome.According to the life-span criterion of HY-1B moonlet accumulator, the bimetry value obtaining the embodiment of the present invention is: 5.21, that is: 5 years 2 months 15 days 14 hours 24 points.
Step 8, dynamic time windows adjust;
The time window of the embodiment of the present invention is 1week, when the week age arranged arrives, utilizes the new battery data of moonlet in-orbit collected, and repeats above-mentioned steps two and can obtain new life prediction value to step 7 process.
Instructions of the present invention does not elaborate part and belongs to techniques well known.
The above is only the preferred embodiment of the present invention; it should be pointed out that for those skilled in the art, under the premise without departing from the principles of the invention; can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.
Claims (1)
1., based on the space flight Ni-Cd life of storage battery Forecasting Methodology that modified is dynamic wavelet neural network based, it is characterized in that performing step is as follows:
Step one, collect all space flight Ni-Cd life of storage battery prediction related data;
By to space flight Ni-Cd accumulator and like product analysis thereof, collect utilizable all life prediction related datas, comprising: the supplemental characteristics such as temperature, charging current, discharge current, discharge electricity amount, load current data, charge capacity, dump energy, voltage, discharge and recharge ratio; For solar battery array, its data available is: the supplemental characteristics such as temperature, square formation current data, load current data, power, busbar voltage, stepup transformer output voltage;
Described like product refers in physical arrangement, logical organization and functional structure, with the product that Ni-Cd accumulator is similar or identical;
Step 2, the pre-service of life prediction related data;
All space flight Ni-Cd life of storage battery predicted data that step one obtains are analyzed and screened, extract required space flight Ni-Cd battery discharging final pressure supplemental characteristic and aging effects factor data, described factor data comprises: discharge current data, charging current data, charge and discharge cycles number of times and depth of discharge data; Meanwhile, singular value rejecting, Noise reducing of data pretreatment work are carried out to the correlation factor data filtered out;
Step 3, data dependence analysis;
By approximation of function or SPSS (statistical product and service solution, Statistical Product and Service Solutions) method carries out correlation analysis to like product and pretreated data corresponding to space flight Ni-Cd accumulator, obtains the correlationship in electric discharge final pressure between like product and Ni-Cd accumulator;
Described correlation analysis refers to by approximation of function or utilizes SPSS method, realizes the correlation analysis to different parameters data, obtains the mapping relations between supplemental characteristic;
Step 4, data-mapping obtain the equivalent data value of Ni-Cd battery discharging final pressure;
Correlationship between the like product obtained with step 3 and Ni-Cd battery discharging final pressure, based on the electric discharge final pressure data of the like product of reference, maps and obtains the equivalent data value of Ni-Cd battery discharging final pressure;
The improvement (M-DWNN) of step 5, DWNN network;
According to space flight Ni-Cd life of storage battery precision of prediction demand, improve DWNN, improvement project is as follows: { Yi} builds AR model, comprises and determines AR model order p and coefficient sequence a to utilize DWNN to export data sequence
i, i=0,1,2 ..., p-1, and then using the exponent number of AR model as DWNN model feedback interstitial content, even M=p, wherein, M represents DWNN network-feedback nodes, the coefficient sequence a of AR model
ias the corresponding weight coefficient of each feedback node, and as WNN mode input, and then complete the improvement to DWNN network model; Wherein the corresponding relation of AR model coefficient and node is: a
iy (t-i), i=0,1,2 ..., p-1;
The foundation of step 6, M-DWNN (1M-DWNN) network, training and prediction;
According to analysis of Influential Factors and association analysis result, determine 1M-DWNN input number of nodes, export as Ni-Cd battery discharging final pressure after treatment, be i.e. single output; Utilize Ni-Cd accumulator in-orbit EoDV and the ED-EoDV that obtained by step 4 do the process of difference ratio, the training sample of structure 1M-DWNN and test sample book; In order to reject the singular value in training sample, accelerate the speed of convergence of network, be normalized by the input vector of above-mentioned structure, object vector, then input 1M-DWNN network to train it, determine 1M-DWNN network parameter, and utilize the 1M-DWNN network trained to predict, divide Procedure Acquisition Ni-Cd battery discharging final pressure predicted value by renormalization and contrast, and then realizing with ground test data section for reference, final pressure of being discharged in-orbit predicts the outcome data segment;
Step 7, set up based on the adaptive iteration forecast model of secondary M-DWNN (2M-DWNN) network, train and predict, performing step is as follows:
(1) data encasement of adaptive iteration prediction, i.e. average-slope time series structure
To by discharging final pressure and predict that the data sequence that the electric discharge final pressure value obtained is formed carries out segmentation average value processing by 1M-DWNN in-orbit, and on this basis, the slope of computation of mean values, namely secondary M-DWNN prediction carries out adaptive iteration prediction to this average-slope time series, and average-slope time series structural formula is as follows:
(j＝1,2,…,n；n＝fix(N/interval))
Wherein, { x
irepresent EoDV value sequence, length is N, interval is mean value interval, and { Avr (j) } is equal value sequence, and length is n, { s (k) } for average-slope sequence, length be n-1;
Subsequent content from step (2), is described the core content of adaptive iteration Forecasting Methodology for object with { s (k) } average-slope sequence;
(2) adaptive iteration process
After step (1) construction complete average-slope time series, start the core content performing adaptive iteration prediction, concrete grammar is as follows:
(2.1) auto-adaptive time sequence data double sampling
For the time series { A that given length is n
i j, wherein, i, j represent original time series iteration prediction number of times and seasonal effect in time series double sampling interval respectively, then A
0 1:={ s (k) }=s
1, s
2, s
3..., s
k..., s
n-1, s
n, wherein, s (k) is the s (k) in Equ.1, represents the average-slope sequence of construction complete, s
kfor a kth data of this sequence, represent raw data, iteration prediction number of times is 0, and sampling for time series interval is defined as 1, wherein ' :=' represent ' being defined as '; And A
1 2then represent that double sampling is spaced apart the time series data of 2, equally spaced from A according to the needs of 2M-DWNN training sample amount after an iteration prediction
0 1in carry out resampling, and then obtain new time series:
Wherein: n is original time series A
0 1length; A
0 jrepresent and be spaced apart j with resampling and obtain time series, and from last data (s of raw data
n) start sampling; j
nrepresent time series A
0 jlength, and meet: n-(j
n-1) * j>=1, and j
nalong with the increase of j constantly reduces, suppose that the demand minimum value of 2M-DWNN training sample amount is Sample_size_min, then j
nj should be met
n>=Sample_size_min, order, j
maxthe maximal value that=fix (n/ (Sample_size_min)) is j, wherein fix represents and rounds downwards, so completes self-adaptation sampling process again;
(2.2) Single-step Prediction of self-adaptation double sampling time series data
Utilize Single-step Prediction thought, build 2M-DWNN model respectively, Single-step Prediction is carried out to the time series data obtained through double sampling, obtains A
0 jone-step prediction value, and then obtain new time series A
0'
jbe expressed as:
Same process completes all A
0 jtime series, wherein j=1,2,3 ..., j
max, thus complete an iterative process, and obtain j
maxindividual predicted value, the new time series formed can be expressed as:
A
1 1:s
1,s
2,…,s
n-1,s
n,s'
n+1,s'
n+2,…,s'
n+jmax
Two steps achieve by random time sequence A above
0 1obtain j
maxnew time series A after individual predicted value
1 1, and then complete an iteration forecasting process; When with A
1 1substitute raw data A
0 1and repeating said process, namely iteration like this obtains A
2 1, A
3 1..., A
k 1,
(3) adaptive prediction iterative data result inverse transformation and Ni-Cd life value judge
By iterative process, infinitely many obtains average-slope sequence A
k 1, in order to finally judge the life-span of accumulator, need predicting the average-slope sequence A obtained
k 1carry out inverse transformation, and obtain Ni-Cd accumulator electric discharge final pressure sequence x '
i, data that this sequence comprises existing final pressure data of discharging in-orbit, 1M-DWNN predicts the outcome data and 2M-DWNN iteration predicts the outcome; Finally based on x '
isequence, determine Ni-Cd accumulator predicting residual useful life value by life of storage battery criterion;
Step 8, dynamic time windows adjust
In the process utilizing M-DWNN to predict Ni-Cd accumulator, due to the passing of time, constantly can obtain new data, now need to repeat above-mentioned steps two dynamically to step 7, rebuild and train M-DWNN network, M-DWNN network comprises 1M-DWNN and 2M-DWNN model to upgrade network parameter again predicting, ensure the prolongation along with time in orbit, precision of prediction can constantly promote.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210468407.8A CN103018673B (en) | 2012-11-19 | 2012-11-19 | Method for predicating life of aerospace Ni-Cd storage battery based on improved dynamic wavelet neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210468407.8A CN103018673B (en) | 2012-11-19 | 2012-11-19 | Method for predicating life of aerospace Ni-Cd storage battery based on improved dynamic wavelet neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103018673A CN103018673A (en) | 2013-04-03 |
| CN103018673B true CN103018673B (en) | 2015-01-21 |
Family
ID=47967499
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210468407.8A Expired - Fee Related CN103018673B (en) | 2012-11-19 | 2012-11-19 | Method for predicating life of aerospace Ni-Cd storage battery based on improved dynamic wavelet neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103018673B (en) |
Families Citing this family (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104102804A (en) * | 2013-04-11 | 2014-10-15 | 华为技术有限公司 | Method and device for predicting service life of component of equipment |
| CN103439666B (en) * | 2013-09-02 | 2016-01-20 | 北京航空航天大学 | A kind of method of geometry of capacity of lithium ion battery slump evaluations |
| CN103633351A (en) * | 2013-11-15 | 2014-03-12 | 中国科学院电工研究所 | Method for establishing temperature control strategy for fuel battery |
| CN105759215B (en) * | 2016-02-26 | 2019-09-27 | 江苏快乐电源(涟水)有限公司 | A kind of charged capacity prediction methods of the lead-acid accumulator of data-driven |
| CN107797067B (en) * | 2016-09-05 | 2019-09-27 | 北京航空航天大学 | Lithium ion battery life migration prediction method based on deep learning |
| CN106569138A (en) * | 2016-10-21 | 2017-04-19 | 北京空间飞行器总体设计部 | Method for predicting service life of satellite subsample cadmium-nickel storage battery based on performance degradation |
| CN107480395B (en) * | 2017-08-29 | 2019-12-13 | 燕山大学 | Method and system for constructing vehicle steering knuckle load spectrum prediction model |
| CN107748332B (en) * | 2017-09-29 | 2019-05-17 | 北京航空航天大学 | A kind of battery charge state (SOC) measuring system based on mechanical wave |
| CN108279383B (en) * | 2017-11-30 | 2020-07-03 | 深圳市科列技术股份有限公司 | Battery life prediction method, battery data server and battery data processing system |
| CN109034469A (en) * | 2018-07-20 | 2018-12-18 | 成都中科大旗软件有限公司 | A kind of tourist flow prediction technique based on machine learning |
| CN110018423A (en) * | 2019-05-07 | 2019-07-16 | 江苏吉意信息技术有限公司 | Battery life Prediction System and battery life predictor method |
| CN116956174B (en) * | 2019-05-13 | 2024-06-18 | 北京绪水互联科技有限公司 | Classification model for cold head state classification detection and life prediction and generation method of prediction model |
| CN110287640B (en) * | 2019-07-03 | 2023-10-13 | 辽宁艾特斯智能交通技术有限公司 | Lighting equipment service life prediction method and device, storage medium and electronic equipment |
| CN110659776B (en) * | 2019-09-25 | 2022-04-19 | 南京国电南自维美德自动化有限公司 | New energy power generation power prediction method and system with self-adaptive time scale |
| CN111090051B (en) * | 2020-01-21 | 2020-11-10 | 北京空间飞行器总体设计部 | Method for automatically diagnosing discharge final voltage of cadmium-nickel storage battery for satellite |
| CN111190112B (en) * | 2020-02-10 | 2020-10-09 | 宜宾职业技术学院 | Battery charging and discharging prediction method and system based on big data analysis |
| CN111190113B (en) * | 2020-04-15 | 2020-07-07 | 中国人民解放军国防科技大学 | Spacecraft storage battery performance degradation abnormity detection method |
| CN111474485A (en) * | 2020-04-28 | 2020-07-31 | 上海空间电源研究所 | Method and system for evaluating on-orbit real-time capacity of spacecraft storage battery pack |
| CN111537889B (en) * | 2020-05-09 | 2022-11-11 | 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 |
| CN112540309B (en) * | 2020-12-10 | 2024-02-06 | 广州能源检测研究院 | Battery monitoring system and method based on battery cycle data similarity analysis |
| CN116070140B (en) * | 2023-04-03 | 2023-07-14 | 国网冀北电力有限公司 | Power distribution substation safe operation state monitoring system and method |
| CN116736174B (en) * | 2023-08-15 | 2023-12-26 | 中国华能集团清洁能源技术研究院有限公司 | Method, apparatus, computer device and storage medium for predicting remaining life of battery |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH04274776A (en) * | 1991-03-01 | 1992-09-30 | Oki Electric Ind Co Ltd | Detecting device of lifetime of ni-cd storage battery |
| JPH07280898A (en) * | 1994-04-11 | 1995-10-27 | Matsushita Electric Ind Co Ltd | Battery life notifying unit |
| CN1890574A (en) * | 2003-12-18 | 2007-01-03 | 株式会社Lg化学 | Apparatus and method for estimating state of charge of battery using neural network |
| CN101894185A (en) * | 2010-06-29 | 2010-11-24 | 北京航空航天大学 | Method for predicting life of small sample data object based on dynamic bipolar modified probabilistic neural network (MPNN) |
-
2012
- 2012-11-19 CN CN201210468407.8A patent/CN103018673B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH04274776A (en) * | 1991-03-01 | 1992-09-30 | Oki Electric Ind Co Ltd | Detecting device of lifetime of ni-cd storage battery |
| JPH07280898A (en) * | 1994-04-11 | 1995-10-27 | Matsushita Electric Ind Co Ltd | Battery life notifying unit |
| CN1890574A (en) * | 2003-12-18 | 2007-01-03 | 株式会社Lg化学 | Apparatus and method for estimating state of charge of battery using neural network |
| CN101894185A (en) * | 2010-06-29 | 2010-11-24 | 北京航空航天大学 | Method for predicting life of small sample data object based on dynamic bipolar modified probabilistic neural network (MPNN) |
Non-Patent Citations (3)
| Title |
|---|
| Life Prediction of Small Satellite Storage Battery Based on Dynamic Bi-Step MPNN Method;Tao Lai-fa等;《2010 International Conference on Electrical and Control Engineering》;20100627;全文 * |
| Prediction on Moonlet Power System Data Based on Modified Probability Neural Network;Tao Lai-fa等;《2009 8th International Conference on reliability、Maintainability and safety》;20091231;全文 * |
| 复合材料疲劳剩余寿命预测的动态小波神经网络方法;谢建宏等;《仪器仪表学报》;20040831;第25卷(第4期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103018673A (en) | 2013-04-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103018673B (en) | Method for predicating life of aerospace Ni-Cd storage battery based on improved dynamic wavelet neural network | |
| CN112034356B (en) | GP-UKF-based online SOH estimation method for electric vehicle power battery | |
| Ji et al. | An RUL prediction approach for lithium-ion battery based on SADE-MESN | |
| CN110488204B (en) | Energy storage element SOH-SOC combined online estimation method | |
| CN108181591B (en) | Battery SOC value prediction method based on improved BP neural network | |
| CN112782591B (en) | Lithium battery SOH long-term prediction method based on multi-battery data fusion | |
| CN103033761B (en) | Lithium ion battery residual life forecasting method of dynamic gray related vector machine | |
| CN111736084B (en) | Valve-regulated lead-acid storage battery health state prediction method based on improved LSTM neural network | |
| CN112990556A (en) | User power consumption prediction method based on Prophet-LSTM model | |
| CN112686464A (en) | Short-term wind power prediction method and device | |
| CN101894185A (en) | Method for predicting life of small sample data object based on dynamic bipolar modified probabilistic neural network (MPNN) | |
| CN105911476B (en) | A kind of battery energy storage system SOC prediction techniques based on data mining | |
| CN102968573A (en) | Online lithium ion battery residual life predicting method based on relevance vector regression | |
| CN110687450B (en) | Lithium battery residual life prediction method based on phase space reconstruction and particle filtering | |
| CN107609671A (en) | A kind of Short-Term Load Forecasting Method based on composite factor evaluation model | |
| CN105160423A (en) | Photovoltaic power generation prediction method based on Markov residual error correction | |
| CN112834927A (en) | Lithium battery residual life prediction method, system, device and medium | |
| CN115201686B (en) | Lithium ion battery health state assessment method under incomplete charge and discharge data | |
| CN114384435A (en) | WSA-LSTM algorithm-based self-adaptive prediction method for residual service life of new energy automobile power battery | |
| CN113361692A (en) | Lithium battery residual life combined prediction method | |
| CN114280490A (en) | Lithium ion battery state of charge estimation method and system | |
| CN106056305A (en) | Power generation system reliability rapid assessment method based on state clustering | |
| Gou et al. | Remaining useful life prediction for lithium-ion battery using ensemble learning method | |
| CN105447596A (en) | Optimized power data short-term prediction method | |
| Zhang et al. | Remaining useful life prediction of lithium-ion batteries based on TCN-DCN fusion model combined with IRRS filtering |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20150121 Termination date: 20201119 |