CN107633331A - Time series models method for building up and device - Google Patents
Time series models method for building up and device Download PDFInfo
- Publication number
- CN107633331A CN107633331A CN201710878449.1A CN201710878449A CN107633331A CN 107633331 A CN107633331 A CN 107633331A CN 201710878449 A CN201710878449 A CN 201710878449A CN 107633331 A CN107633331 A CN 107633331A
- Authority
- CN
- China
- Prior art keywords
- data sequence
- model
- historical data
- final mask
- sequence
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Abstract
The present invention relates to statistical technique field, a kind of time series models method for building up and device are provided, by obtaining the historical data sequence of bank, and when historical data sequence meets preparatory condition, multiple models fitting is carried out to historical data sequence, obtains final mask group;Then final mask group is screened, select aicc values minimum is used as final mask, ensure that final mask reaches best fitting effect, the prediction data sequence in bank's a period of time in future finally obtained is more accurate, the adjustment trend of Asset-Liability Structure can be effectively guided, there is good application value.
Description
Technical field
The present invention relates to statistical technique field, in particular to a kind of time series models method for building up and device.
Background technology
Time series models as a kind of Short-term Forecasting Model, be can be widely applied to economy, medical science, meteorology, geography,
The every field such as the hydrology, geology.But because the algorithm of the model is complicated, have in modeling to data sequence extremely strict
Checking procedure and complex transformations process, and few stable data sequences in real life, so the model is in practical study
In seldom apply.At present, conventional time series models are the auto.arima algorithms that R language official announces, still
Auto.arima algorithms only include stationary test in time series models, determine rank and differential process, that is, only realize
Stationary test, difference and the automation for determining rank, cause the practicality of auto.arima algorithms very poor.
The content of the invention
It is an object of the invention to provide a kind of time series models method for building up and device, to improve above mentioned problem.
To achieve these goals, the technical scheme that the embodiment of the present invention uses is as follows:
In a first aspect, the invention provides a kind of time series models method for building up, applied to the electronic equipment of bank, use
It is predicted in the asset-liabilities to bank, methods described includes:Obtain the historical data sequence of the bank;Judge history number
Whether meet preparatory condition according to sequence;When historical data sequence meets preparatory condition, multiple mould is carried out to historical data sequence
Type is fitted, and obtains final mask group;To final mask, group screens, and obtains final mask;According to final mask, bank is obtained
Prediction data sequence.
Second aspect, the invention provides a kind of time series models to establish device, applied to the electronic equipment of bank, uses
It is predicted in the asset-liabilities to bank, described device includes historical data retrieval module, judge module, final mask
Group obtains module, final mask obtains module and prediction data sequence obtains module.Wherein, historical data retrieval module is used
In the historical data sequence for obtaining the bank;Judge module is used to judge whether historical data sequence meets preparatory condition;Most
Final cast group obtains module and is used for when historical data sequence meets preparatory condition, and multiple model plan is carried out to historical data sequence
Close, obtain final mask group;Final mask obtains module and is used to screen final mask group, obtains final mask;According to
Final mask, obtain the prediction data sequence of bank;Prediction data sequence obtains module and is used for according to final mask, obtains bank
Prediction data sequence.
Compared with the prior art, the invention has the advantages that:A kind of time series models provided by the invention are established
Method and device, by obtaining the historical data sequence of bank, and when historical data sequence meets preparatory condition, to history number
Multiple models fitting is carried out according to sequence, obtains final mask group;Then final mask group is screened, selects aicc values minimum
Model as final mask, it is ensured that final mask reaches best fitting effect, the bank that finally obtains following a period of time
Interior prediction data sequence is more accurate, can effectively guide the adjustment trend of Asset-Liability Structure, has good answer
With value.
To enable the above objects, features and advantages of the present invention to become apparent, preferred embodiment cited below particularly, and coordinate
Appended accompanying drawing, is described in detail below.
Brief description of the drawings
In order to illustrate the technical solution of the embodiments of the present invention more clearly, below by embodiment it is required use it is attached
Figure is briefly described, it will be appreciated that the following drawings illustrate only certain embodiments of the present invention, therefore be not construed as pair
The restriction of scope, for those of ordinary skill in the art, on the premise of not paying creative work, can also be according to this
A little accompanying drawings obtain other related accompanying drawings.
Fig. 1 shows the block diagram of electronic equipment provided in an embodiment of the present invention.
Fig. 2 shows time series models method for building up flow chart provided in an embodiment of the present invention.
Fig. 3 is the sub-step flow chart of the step S102 shown in Fig. 2.
Fig. 4 is the sub-step flow chart of the step S103 shown in Fig. 2.
Fig. 5 is the sub-step flow chart of the step S104 shown in Fig. 2.
Fig. 6 is the sub-step flow chart of the sub-step S1045 shown in Fig. 5.
Fig. 7 is the sub-step flow chart of the step S105 shown in Fig. 2.
Fig. 8 shows the parameter configuration interface of electronic equipment provided in an embodiment of the present invention.
Fig. 9 is the predicted value obtained using time series models method for building up provided in an embodiment of the present invention.
Figure 10 utilizes the prognostic chart that time series models method for building up provided in an embodiment of the present invention obtains.
Figure 11 shows that time series models provided in an embodiment of the present invention establish the block diagram of device.
Figure 12 establishes the block diagram of judge module in device for the time series models shown in Figure 11.
Figure 13 is that the time series models shown in Figure 11 establish the block diagram that final mask group in device obtains module.
Figure 14 is that the time series models shown in Figure 11 establish the block diagram that final mask in device obtains module.
Figure 15 obtains the block diagram of the second execution unit in module for the final mask shown in Figure 14.
Figure 16 is that the time series models shown in Figure 11 establish the square frame signal that prediction data sequence in device obtains module
Figure.
Icon:100- electronic equipments;101- memories;102- storage controls;103- processors;200- time series moulds
Type establishes device;201- historical data retrieval modules;202- judge modules;2021- white noise verification units;2022- is put down
Stability verification unit;2023- determines jump subdivision;2024- identifying units;203- final masks group obtains module;Among 2031-
Model Group obtaining unit;2032- residual error acquiring units;2033- final mask group's obtaining units;204- final masks obtain mould
Block;2041-aicc value computing units;2042- alternative model obtaining units;2043- conspicuousness judging units;2044- first is held
Row unit;The execution units of 2045- second;20451- sparse coefficient models establish unit;20452-aicc value acquiring units;20453-
First sub- execution unit;The second sub- execution units of 20454-;205- prediction data sequence obtains module;2051- model prediction lists
Member;2052- rejecting outliers units.
Embodiment
Below in conjunction with accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Ground describes, it is clear that described embodiment is only part of the embodiment of the present invention, rather than whole embodiments.Generally exist
The component of the embodiment of the present invention described and illustrated in accompanying drawing can be configured to arrange and design with a variety of herein.Cause
This, the detailed description of the embodiments of the invention to providing in the accompanying drawings is not intended to limit claimed invention below
Scope, but it is merely representative of the selected embodiment of the present invention.Based on embodiments of the invention, those skilled in the art are not doing
The every other embodiment obtained on the premise of going out creative work, belongs to the scope of protection of the invention.
It should be noted that:Similar label and letter represents similar terms in following accompanying drawing, therefore, once a certain Xiang Yi
It is defined, then it further need not be defined and explained in subsequent accompanying drawing in individual accompanying drawing.Meanwhile the present invention's
In description, term " first ", " second " etc. are only used for distinguishing description, and it is not intended that instruction or hint relative importance.
Fig. 1 is refer to, Fig. 1 shows the block diagram of electronic equipment 100 provided in an embodiment of the present invention.Electronic equipment
100 desktop computers that may be, but not limited to, business bank staff, notebook computer, smart mobile phone, tablet personal computer etc..
The electronic equipment 100 establishes device 200, memory 101, storage control 102 and processor including time series models
103。
The memory 101, storage control 102 and 103 each element of processor are directly or indirectly electrical between each other
Connection, to realize the transmission of data or interaction.For example, these elements can pass through one or more communication bus or letter between each other
Number line, which is realized, to be electrically connected with.The time series models establish device 200 include it is at least one can be with software or firmware
(firmware) form is stored in the memory 101 or is solidificated in the operating system of the electronic equipment 100
Software function module in (operating system, OS).The processor 103 is used to perform what is stored in memory 101
Executable module, such as the time series models establish the software function module or computer program that device 200 includes.
Wherein, memory 101 may be, but not limited to, random access memory (Random Access Memory,
RAM), read-only storage (Read Only Memory, ROM), programmable read only memory (Programmable Read-Only
Memory, PROM), erasable read-only memory (Erasable Programmable Read-Only Memory, EPROM),
Electricallyerasable ROM (EEROM) (Electric Erasable Programmable Read-Only Memory, EEPROM) etc..
Wherein, memory 101 is used for storage program, and the processor 103 performs described program, this hair after execute instruction is received
Method performed by the server for the flow definition that bright any embodiment discloses can apply in processor 103, or by
Reason device 103 is realized.
Processor 103 can be a kind of IC chip, have signal handling capacity.Above-mentioned processor 103 can be with
It is general processor, including central processing unit (Central Processing Unit, CPU), network processing unit (Network
Processor, NP), speech processor and video processor etc.;Can also be digital signal processor, application specific integrated circuit,
Field programmable gate array either other PLDs, discrete gate or transistor logic, discrete hardware components.
It can realize or perform disclosed each method, step and the logic diagram in the embodiment of the present invention.General processor can be
Microprocessor or the processor 103 can also be any conventional processors etc..
First embodiment
Fig. 2 is refer to, Fig. 2 shows time series models method for building up flow chart provided in an embodiment of the present invention.Time
Series model method for building up comprises the following steps:
Step S101, obtain the historical data sequence of bank.
In embodiments of the present invention, historical data sequence can be the asset data or negative in bank's the past period
Debt data.The past period can be several years or some months, on the basis of the same day, for example, past 3 years from the same day, 3
Individual month etc., the length of the past period can be adjusted flexibly according to banking.To be predicted to the assets of bank, then
The asset data of the past period is obtained from the data system of bank;To be predicted to the debt of bank, then from silver
The debt data of the past period are obtained in capable data system.
Step S102, judges whether historical data sequence meets preparatory condition.
In embodiments of the present invention, preparatory condition is that historical data sequence is both stationary sequence non-white noise sequence again.
, it is necessary to first be tested to the historical data sequence of acquisition, the method for inspection can include before settling time series model:
First, white noise verification is carried out to historical data sequence, to judge whether historical data sequence is pure random sequence.
If it is determined that the historical data sequence is white noise, i.e., pure random sequence, then illustrate between the data item of the historical data sequence not
With correlation, it is impossible to time series models, now, the historical data sequence is predicted using moving average method;If
It is not white noise to judge the historical data sequence, i.e., is not pure random sequence, then illustrate the historical data sequence data item it
Between there is correlation, now to the historical data sequence carry out stationary test.
Secondly, stationary test is used to examine whether historical data sequence is stationary sequence, if historical data is not steady
Sequence, then the data sequence is carried out determining rank, and difference is carried out according to exponent number, be steady sequence the historical data series processing
Row;White noise verification is now carried out again to obtained stationary sequence, if the stationary sequence is white noise sequence, uses cunning
The dynamic method of average is predicted;If the stationary sequence non-white noise sequence, into the models fitting stage.
Fig. 3 is refer to, step S102 can include following sub-step:
Sub-step S1021, to historical data sequence carry out white noise verification, with judge historical data sequence whether be it is pure with
Machine sequence.
In embodiments of the present invention, white noise verification is carried out to historical data sequence, it can be determined that the historical data sequence
Data item between whether have correlation, at the beginning of settling time series model, do white noise to historical data sequence first
Examine, it can be ensured that the preciseness of whole computing.
Sub-step S1022, when historical data sequence is not pure random sequence, stationarity inspection is carried out to historical data sequence
Test, to judge whether historical data sequence is stationary sequence.
In embodiments of the present invention, stationary test can use ADF methods of inspection, and ADF examination requirementses have one to sample size
Provisioning request, but in actual applications, the sample size of the historical data sequence collected is attained by level needed for inspection,
There are 4 kinds of forms:
1. without drift term, intercept item:
2. contain drift term, without intercept item:
3. being free of drift term, contain intercept item:
4. contain drift term, intercept item:
Meanwhile ADF is examined and is used t statisticsWherein, T is sample total,Φ1、Φ2…ΦpEstimate for OLS
Evaluation,ForStandard deviation estimate amount, Δ ytRepresent the first-order difference of sample.
As a kind of embodiment, these four check systems are write as circulation on stream, according to data sequence
The method of inspection corresponding to the selection of the situation of middle drift term and intercept item, and calculates corresponding τ values, if all can not by inspection,
Then judge unit root be present, i.e. historical data sequence is unstable sequence.
Sub-step S1023, when historical data sequence is not stationary sequence, historical data sequence is carried out to determine order difference,
To obtain stationary sequence.
In embodiments of the present invention, can useAs acf functions:rk=cov
(xt, xt-k)=E (xt-μ)(xt-k- μ) Consistent Estimation, wherein, i is lag order, and n is historical data sequence observation,For
Average value is observed, auto-correlation coefficient is used for the exponent number for judging ma functions.In general, as k growth, auto-correlation function are gradual
Decay, finally not significantly different from zero, drafting function image, if function is to drop to zero suddenly, function truncation, if function
It is slowly to drop to zero, then function trails.
Pass through and compareValue withSize, it can be determined that whether be number of samples significantly different from zero, n.Make
With R language draw autocorrelogram as when can automatically give tacit consent to and mark two height and beDotted line, in general, model order
To lack as far as possible, the exponent number sometimes havingAlthough having exceeded dotted line, it can consider to be treated as accidental value processing, take smaller
Exponent number.
Pacf functions:Use rt-k(k>0) it is multiplied by p rank autoregressive process simultaneously in both sides
rt=Φ1rt-1+Φ2rt-2+…+Φprt-p+at
Obtain rtrt-k=Φ1rt-1rt-k+Φ2rt-2rt-k+…+Φprt-prt-k+atrt-k;
Expectation is asked to obtain γ above formula both sidesk=Φ1γk-1+Φ2γk-2+…+Φpγk-p, wherein, wherein γkFor sequence
Lag the coefficient correlation of k phases, equation both sides divided by γ0Obtain Yule-Walker equations:
ρk=Φ1ρk-1+Φ2ρk-2+…+Φpρk-p
For p rank autoregressive process rt=Φk1rt-1+Φk2rt-2+…+Φkkrt-k+at, ΦkkIt is considered as historical data sequence
The partial autocorrelation function of row hysteresis k ranks, because this is precisely to lag k ranks to be worth before exclusion after lag order to when time value
Influence size measurement.Now Yule-Walker equations are:
ρj=Φk1ρk-1+Φk2ρk-2+…+Φkkρj-k
K=1,2,3 ... is brought into above formula successively can solve PARCOR coefficients.
Sub-step S1024, when historical data sequence is stationary sequence, judge that historical data sequence meets preparatory condition.
Step S103, when historical data sequence meets preparatory condition, multiple models fitting is carried out to historical data sequence,
Obtain final mask group.
In embodiments of the present invention, obtaining the process of final mask group can include:First, to meeting that preparatory condition is gone through
History data sequence carries out multiple models fitting, establishes multiple models, forms a mid-module group.It is poor in the models fitting stage
All combinations one of gradation number and exponent number share 192 kinds, and 192 models, shape are fitted according to this 192 kinds of parameter combination modes
Into mid-module group;Then, the residual error of each model in mid-module group is obtained, and white noise verification is carried out to residual error, if residual
Difference is not white noise, then model corresponding to the residual error is removed into Model Group, if residual error is white noise, retain, obtain final mould
Type group.
Fig. 4 is refer to, step S103 can include following sub-step:
Sub-step S1031, multiple models fitting is carried out to historical data sequence, establishes multiple models, forms a centre
Model Group.
Sub-step S1032, obtain the residual error of each model in mid-module group.
Sub-step S1033, residual test is carried out to the residual error of each model in middle Model Group, if residual error is white noise
Retain the model, the model is rejected if residual error nonwhite noise, to obtain final mask group.
In embodiments of the present invention, residual test is carried out namely to each to the residual error of each model in middle Model Group
The residual error of model carries out white noise verification, and also known as Ljung-Box is examined, and the method for inspection lags between examining each issue of residual error with it
It whether there is correlation between the residual error of some exponent numbers, if it is present model is not fitted completely, if it does not, and just
State property is upchecked, then it is white noise sequence to prove residual error, and models fitting is successful.
Ljung-Box normalized set modes are as follows:
Wherein, n is sequence observation number, is also called sequence length, lag is
Lag order, acfiFor sequence i rank auto-correlation function values:
rtFor the random variable values of sequence t phases, acf functions can be realized with stats bags in R language.
The approximate free degree of obeying of Ljung-Box statistics (due to being residual error, reduces one for lag-1 chi square distribution
The free degree, in pure randomness test, the free degree of statistic is lag), p value thus can be calculated.
If recent residual error is without auto-correlation it is at a specified future date in general also will not auto-correlation, the inspection can be by correlative code meter
The p value of 12 rank statistics before calculation, when p value is more than 0.05, no auto-correlation is assumed to set up.
Step S104, to final mask, group screens, and obtains final mask.
In embodiments of the present invention, the process screened to final mask group can include:First, final mask is detected
Whether group is empty, if so, being then predicted using moving average method;If it is not, then select the minimum model of aicc values alternately
Model;Then, t inspections are carried out to the alternative model, to judge whether the autoregressive coefficient of alternative model has conspicuousness, if from
Regression coefficient has conspicuousness, then using the alternative model as final mask, if it is not, then showing historical data sequence in time
It is interval, now needs to carry out sparse coefficient analysis, to obtain final mask.
As a kind of embodiment, sparse coefficient analysis can include:First, according to alternative model, sparse coefficient model is established;
Then, white noise verification is done to the residual error of the sparse coefficient model, if residual error nonwhite noise, abandons the sparse coefficient model, returned
Alternative model;If residual error is white noise, compare the aicc values of sparse coefficient model and alternative model, and select aicc value smallers
As final mask.
Fig. 5 is refer to, step S104 can include following sub-step:
Sub-step S1041, calculate the aicc values of each model in final mask group.
Sub-step S1042, obtains the model of aicc values minimum, and by the model alternately model.
In embodiments of the present invention, aicc values (also referred to as AICc) minimum model is obtained using AIC/AICc/BIC to believe
Criterion is ceased, AIC=-2log Maximum-likelihood estimation+2k, for common arima models, p, q are respectively ar models, ma models
Exponent number, if model includes intercept or constant term, k=p+q+1, otherwise k=p+q.For the arima moulds containing season item
Type k will also add season exponent number.AIC estimators are by increasing penalty, it is ensured that select succinct model, but it is one and had
Estimation partially, the ratio for larger number of parameters relative to data capacity, deviation can be quite big.
AICc adds a nonrandom penalty term in AIC, can eliminate approximate deviation, and n is effective sample capacity, real
Verify it is bright, k/n be more than 10% in the case of, AICc performance is better than many other model selection criterias, including AIC, BIC.
In time series models modeling, Aicc values are selected as final choice foundation.
Sub-step S1043, t inspections are carried out to alternative model, to judge it is aobvious whether the autoregressive coefficient of alternative model has
Work property.
In embodiments of the present invention, t is verified asWherein,It is estimating for i-th autoregressive coefficient
Evaluation,It is the standard error of i-th of autoregressive coefficient, i.e.,The estimation of standard deviation,For i-th autoregressive coefficient
Actual value, null hypothesisAlternative hypothesisFirst assume that null hypothesis is set up, thenSet up, obey the free degree
It is distributed for n-2 t, n is independent variable observation number.If α is confidence level, general α=0.05, if t*>t1-α(n-2) it is then former
Assuming that invalid, autoregressive coefficient is notable.
Sub-step S1044, when autoregressive coefficient has conspicuousness, setting alternative model is final mask.
Sub-step S1045, when autoregressive coefficient does not have conspicuousness, sparse coefficient analysis is carried out to alternative model, with
To final mask.
Fig. 6 is refer to, sub-step S1045 can include following sub-step:
Sub-step S10451, according to alternative model, establish sparse coefficient model.
Sub-step S10452, obtain the aicc values of sparse coefficient model.
Sub-step S10453, when the aicc values of alternative model are less than the aicc values of sparse coefficient model, alternative model is set
For final mask.
Sub-step S10454, when the aicc values of sparse coefficient model are less than the aicc values of alternative model, sparse coefficient mould is set
Type is final mask.
Step S105, according to final mask, obtain the prediction data sequence of bank.
In embodiments of the present invention, prediction data sequence can be the assets or debt of bank in following a period of time
Data sequence, it can be several days following or some months, on the basis of the same day, for example, being following 30 days, 2 from the same day
Individual month etc., specific time length can be adjusted flexibly according to banking.According to final mask, the prediction data sequence of bank is obtained
The method of row can include:First, using final mask, the prediction data sequence of bank is exported, prediction data sequence can wrap
Include the coefficient and aicc values of predicted value, prognostic chart and final mask;Then, Bonferroni algorithms, detection history data are utilized
Exceptional value in sequence and prediction data sequence, and outlier is labeled.
As a kind of embodiment, final output includes prediction data sequence and its 80% and 95% confidential interval, pre-
Survey the line chart area-graph corresponding with its 80% and 95% confidential interval of data sequence, final mask autoregressive coefficient and
The coefficient of each model and corresponding aicc values in aicc values, final mask group.Each model and most is exported in final mask group simultaneously
The autoregressive coefficient of final cast and aicc values be for the ease of final mask and other models in final mask group are contrasted,
To confirm that model selects excellent process in contrast.
Fig. 7 is refer to, step S105 can include following sub-step:
Sub-step S1051, using final mask, export the prediction data sequence of bank.
Sub-step S1052, using Bonferroni algorithms, the exception in detection history data sequence and prediction data sequence
Value.
In embodiments of the present invention, first, new breath exceptional value is judged with Bonferroni algorithms.If it is in time t
When error (also referred to as new breath) it is perturbed that (error is e't=et+ω1Pt T, wherein etIt is a zero-mean white noise
Process, Pt TRefer to the T moment be 1, remaining moment be 0 impulse response function), then just occur in t one newly cease it is different
Constant value, therefore, e'T=eT+ωI, e' in the rest of the casesT=eT, it is assumed that in the case where not being disturbed:
Then in the case where being disturbed:
EitherWherein,And when j is negativeTherefore, even as observation
It is gradually distance from the generation point disturbance effect diminuendo of exceptional value, moment T new breath exceptional value still can be to moment T and its later all
Observation produces disturbance.
Residual error is represented with the process AR (∞) not being disturbed:
at=Y't-π1Y't-1-π2Y't-2-…
For simplicity, it is assumed that the average of process is zero, and known to all parameters.The parameter that typically we are obtained with estimation
Instead of actual parameter, the possible perturbed influence of these results.In null hypothesis and in the case of being no different constant value and large sample,
This approximate substitution can be ignored to the validity of following inspection processes.If sequence is only carved with new breath exceptional value in T,
So residual error is aT=ωI+et, a in the case of remainingT=et, therefore ωICan beVariance is equal to σ2.Therefore, for examining
The new breath exceptional value tested on moment T has following statistic:
Null hypothesis is not have exceptional value in time series, and test statistics is approximately obeyed standard and is just distributed very much.Realized in T
When known, determine that observation exceedes for the size that the condition of exceptional value is the respective standard residual error in 5% significance
1.96.Due to not knowing which T be, therefore need to test to whole observations at moment, also need to estimate σ in addition.Simply
And conservative way is to control the global error rate of multiple check using Bonferroni algorithms, make
λ1=max | λ1,t|
Maximum obtains at the t=T moment, if λ1>0.025/n × 100, that is, the upper hundredths of standardized normal distribution is exceeded
Number, then the T observation necessarily newly ceases exceptional value, and the flow ensure that the probability of misattribution is no more than 5%.
Then, judge that exceptional value can be added with Bonferroni algorithms.Assuming that the presence of T moment can add exceptional value, at other
Time point does not have exceptional value, then provable
aT=-ωAπt-T+et, wherein, π0π when=- 1, and if only if j are negativej=0, therefore work as t<A during Tt=etWhen, aT=
ωA+eT, aT+1=-ωAπ1+eT+1, aT+2=-ωAπ2+eT+2, by that analogy, ωALeast squares estimator be:
Wherein,And Estimator Variance is ρ2σ2,
Test statistics can then be defined
Null hypothesis is that prediction data sequence is no different constant value, and alternative hypothesis is that prediction data sequence has adding in T
The situation of exceptional value, the progressive obedience standardized normal distribution of test statistics, Bonferroni algorithms can be applied to control overall mistake
Difference, in addition the property of exceptional value can not learn in advance.When on time T examine arrive exceptional value when, if | λ1, T|>|λ2, T|, see
Measured value is new breath exceptional value, otherwise can to add exceptional value.
In embodiments of the present invention, time series models method for building up has been encapsulated as time sequence function, and can be in electricity
Sub- equipment 100 sets the parameter configuration interface of time series models method for building up, refer to Fig. 8, following 30 days to calculate bank
Interior debt change conditions, it need to only input this previous year bank's items debt data and option date, the processor of electronic equipment 100
103 can allocating time ordinal functions, it is possible to it is quick to calculate predicted value of being in debt in 30 days futures of bank, and with line chart
Form exports, and if desired changes the data of a certain debt, only needs click " editor " to re-enter the data of this assets.
As a kind of embodiment, it is assumed that certain bank can be obtained and go over the current deposit data of 2 years (24 months) such as
Under;
After allocating time ordinal function, incorporated by reference to reference picture 9 and Figure 10, the assets that can export bank's following 10 phases are pre-
Measured value, prognostic chart and its confidential interval.
In embodiments of the present invention, compared with existing time series models, time series models provided by the invention are built
Cube method has advantages below:First, white noise verification is carried out to historical data sequence before time series models foundation, can
To ensure the preciseness of the accuracy of prediction result and whole computing;Second, many historical data sequences are not in practical application
Stable or white noise sequence, therefore be predicted for pure random sequence using moving average method, it can be ensured that time sequence
The robustness of row model;3rd, use multi-difference and repeatedly determined the framework of the loop computation being mutually combined of rank, calculated every
A kind of error of parameter combination, the minimum model of error is finally selected as final mask, export the operation result of the model.
During this, 192 models are established, and select the minimum conduct final mask of aicc values, are significantly better than auto.arima
Only calculate the computing mode for once determining rank and difference;4th, time series models method for building up provided by the invention, final output
Including prediction data sequence and its 80% and 95% confidential interval, the line chart of prediction data sequence and its 80% and 95% confidence
In area-graph, the autoregressive coefficient of final mask and aicc values, final mask group corresponding to section the coefficient of each model with it is corresponding
Aicc values, be significantly better than only output predicted value and prognostic chart an auto.arima.
It should be noted that time series models method for building up provided in an embodiment of the present invention, except comprising
Stationary test in auto.arima algorithms, determine beyond order difference process, other processes and whole time series models are established
The framework of method, is creative content, all any modification, equivalent substitution and improvements made in the creative content
Deng should be included in the scope of the protection.
Second embodiment
Figure 11 is refer to, Figure 11 shows that time series models provided in an embodiment of the present invention establish the square frame of device 200
Schematic diagram.Time series models, which establish device 200, includes historical data retrieval module 201, judge module 202, final mould
Type group obtains module 203, final mask obtains module 204 and prediction data sequence obtains module 205.
Historical data retrieval module 201, for obtaining the historical data sequence of bank.
In embodiments of the present invention, historical data retrieval module 201 can be used for performing step S101.
Judge module 202, for judging whether historical data sequence meets preparatory condition.
In embodiments of the present invention, judge module 202 can be used for performing step S102.
Figure 12 is refer to, Figure 12 establishes the square frame of judge module 202 in device 200 for the time series models shown in Figure 11
Schematic diagram.Judge module 202 includes white noise verification unit 2021, stationary test unit 2022, determines jump subdivision 2023
And identifying unit 2024.
White noise verification unit 2021, for carrying out white noise verification to historical data sequence, to judge historical data sequence
Whether row are pure random sequence.
In embodiments of the present invention, white noise verification unit 2021 can be used for performing sub-step S1021.
Stationary test unit 2022, for when historical data sequence is not pure random sequence, to historical data sequence
Stationary test is carried out, to judge whether historical data sequence is stationary sequence.
In embodiments of the present invention, stationary test unit 2022 can be used for performing sub-step S1022.
Jump subdivision 2023 is determined, for when historical data sequence is not stationary sequence, being carried out to historical data sequence
Order difference is determined, to obtain stationary sequence.
In embodiments of the present invention, determine jump subdivision 2023 to can be used for performing sub-step S1023.
Identifying unit 2024, it is default for when historical data sequence is stationary sequence, judging that historical data sequence meets
Condition.
In embodiments of the present invention, identifying unit 2024 can be used for performing sub-step S1024.
Final mask group obtains module 203, for when historical data sequence meets preparatory condition, to historical data sequence
Multiple models fitting is carried out, obtains final mask group.
In embodiments of the present invention, final mask group obtains module 203 and can be used for performing step S103.
Figure 13 is refer to, Figure 13 is that the time series models shown in Figure 11 establish final mask group acquisition mould in device 200
The block diagram of block 203.Final mask group, which obtains module 203, includes mid-module group obtaining unit 2031, residual error acquisition list
Member 2032 and final mask group obtaining unit 2033.
Mid-module group obtaining unit 2031, for carrying out multiple models fitting to historical data sequence, establish multiple moulds
Type, form a mid-module group.
In embodiments of the present invention, mid-module group obtaining unit 2031 can be used for performing sub-step S1031.
Residual error acquiring unit 2032, for obtaining the residual error of each model in mid-module group.
In embodiments of the present invention, residual error acquiring unit 2032 can be used for performing sub-step S1032.
Final mask group obtaining unit 2033, for carrying out residual test to the residual error of each model in middle Model Group,
Retain the model if residual error is white noise, the model is rejected if residual error nonwhite noise, to obtain final mask group.
In embodiments of the present invention, final mask group obtaining unit 2033 can be used for performing sub-step S1033.
Final mask obtains module 204, for being screened to final mask group, obtains final mask.
In embodiments of the present invention, final mask obtains module 204 and can be used for performing step S104.
Figure 14 is refer to, Figure 14 is that the time series models shown in Figure 11 establish final mask acquisition module in device 200
204 block diagram.Final mask, which obtains module 204, includes aicc values computing unit 2041, alternative model obtaining unit
2042nd, conspicuousness judging unit 2043, the first execution unit 2044 and the second execution unit 2045.
Aicc values computing unit 2041, for calculating the aicc values of each model in final mask group.
In embodiments of the present invention, aicc values computing unit 2041 can be used for performing sub-step S1041.
Alternative model obtaining unit 2042, the model minimum for obtaining aicc values, and by the model alternately model.
In embodiments of the present invention, alternative model obtaining unit 2042 can be used for performing sub-step S1042.
Conspicuousness judging unit 2043, for carrying out t inspections to alternative model, to judge the autoregressive coefficient of alternative model
Whether there is conspicuousness.
In embodiments of the present invention, conspicuousness judging unit 2043 can be used for performing sub-step S1043.
First execution unit 2044, for when autoregressive coefficient has conspicuousness, setting alternative model to be final mask.
In embodiments of the present invention, the first execution unit 2044 can be used for performing sub-step S1044.
Second execution unit 2045, for when autoregressive coefficient does not have conspicuousness, sparse coefficient to be carried out to alternative model
Analysis, to obtain final mask.
In embodiments of the present invention, the second execution unit 2045 can be used for performing sub-step S1045.
Figure 15 obtains the block diagram of the second execution unit 2045 in module 204 for the final mask shown in Figure 14.The
Two execution units 2045 establish unit 20451, aicc values acquiring unit 20452, the first sub- execution unit including sparse coefficient model
20453 and the second sub- execution unit 20454.
Sparse coefficient model establishes unit 20451, for according to alternative model, establishing sparse coefficient model.
In embodiments of the present invention, sparse coefficient model establishes unit 20451 and can be used for performing sub-step S10451.
Aicc values acquiring unit 20452, for obtaining the aicc values of sparse coefficient model.
In embodiments of the present invention, aicc values acquiring unit 20452 can be used for performing sub-step S10452.
First sub- execution unit 20453, for when alternative model aicc values be less than sparse coefficient model aicc values when, if
It is final mask to put alternative model.
In embodiments of the present invention, the first sub- execution unit 20453 can be used for performing sub-step S10453.
Second sub- execution unit 20454, for when sparse coefficient model aicc values be less than alternative model aicc values when, if
It is final mask to put sparse coefficient model.
In embodiments of the present invention, the second sub- execution unit 20454 can be used for performing sub-step S10454.
Prediction data sequence obtains module 205, for according to final mask, obtaining the prediction data sequence of bank.
In embodiments of the present invention, prediction data sequence obtains module 205 and can be used for performing step S105.
Figure 16 is refer to, Figure 16 is that the time series models shown in Figure 11 establish prediction data sequence acquisition in device 200
The block diagram of module 205.Prediction data sequence, which obtains module 205, includes model prediction unit 2051 and rejecting outliers list
Member 2052.
Model prediction unit 2051, for using final mask, exporting the prediction data sequence of bank.
In embodiments of the present invention, model prediction unit 2051 can be used for performing sub-step S1051.
Rejecting outliers unit 2052, for utilizing Bonferroni algorithms, detection history data sequence and prediction data
Exceptional value in sequence.
In embodiments of the present invention, rejecting outliers unit 2052 can be used for performing sub-step S1052.
In summary, a kind of time series models method for building up and device provided by the invention, the electronics applied to bank
Equipment, for being predicted to the asset-liabilities of bank, methods described includes:Obtain the historical data sequence of the bank;Sentence
Whether disconnected historical data sequence meets preparatory condition;When historical data sequence meets preparatory condition, historical data sequence is entered
The multiple models fitting of row, obtains final mask group;To final mask, group screens, and obtains final mask;According to final mask,
Obtain the prediction data sequence of bank.Time series models method for building up provided by the invention, first, builds in time series models
White noise verification is carried out to historical data sequence before vertical, it can be ensured that the accuracy of prediction result and whole computing it is rigorous
Property;Secondly, many historical data sequences are jiggly or white noise sequence in practical application, therefore are directed to pure random sequence
It is predicted using moving average method, it can be ensured that the robustness of time series models;Again, used multi-difference and repeatedly
Determine the framework of the loop computation being mutually combined of rank, calculate the error of each parameter combination, finally select the minimum mould of error
Type exports the operation result of the model as final mask.In this process, 192 models are established, and select aicc values
Minimum is used as final mask, is significantly better than auto.arima and only calculates the computing mode for once determining rank and difference;Finally, this hair
The time series models method for building up of bright offer, final output include prediction data sequence and its 80% and 95% confidential interval,
The line chart of prediction data sequence area-graph corresponding with its 80% and 95% confidential interval, final mask autoregressive coefficient and
The coefficient of each model and corresponding aicc values in aicc values, final mask group, it is significantly better than only output predicted value and prognostic chart
auto.arima.Therefore, the present invention may insure the silver that final mask reaches best fitting effect, obtains in calculating process
Prediction data sequence in row following a period of time is more accurate, can effectively guide the adjustment of Asset-Liability Structure to become
Gesture, there is good application value.
In several embodiments provided herein, it should be understood that disclosed apparatus and method, can also pass through
Other modes are realized.Device embodiment described above is only schematical, for example, flow chart and block diagram in accompanying drawing
Show the device of multiple embodiments according to the present invention, method and computer program product architectural framework in the cards,
Function and operation.At this point, each square frame in flow chart or block diagram can represent the one of a module, program segment or code
Part, a part for the module, program segment or code include one or more and are used to realize holding for defined logic function
Row instruction.It should also be noted that at some as in the implementation replaced, the function that is marked in square frame can also with different from
The order marked in accompanying drawing occurs.For example, two continuous square frames can essentially perform substantially in parallel, they are sometimes
It can perform in the opposite order, this is depending on involved function.It is it is also noted that every in block diagram and/or flow chart
The combination of individual square frame and block diagram and/or the square frame in flow chart, function or the special base of action as defined in performing can be used
Realize, or can be realized with the combination of specialized hardware and computer instruction in the system of hardware.
In addition, each functional module in each embodiment of the present invention can integrate to form an independent portion
Point or modules individualism, can also two or more modules be integrated to form an independent part.
If the function is realized in the form of software function module and is used as independent production marketing or in use, can be with
It is stored in a computer read/write memory medium.Based on such understanding, technical scheme is substantially in other words
The part to be contributed to prior art or the part of the technical scheme can be embodied in the form of software product, the meter
Calculation machine software product is stored in a storage medium, including some instructions are causing a computer equipment (can be
People's computer, server, or network equipment etc.) perform all or part of step of each embodiment methods described of the present invention.
And foregoing storage medium includes:USB flash disk, mobile hard disk, read-only storage (ROM, Read-Only Memory), arbitrary access are deposited
Reservoir (RAM, Random Access Memory), magnetic disc or CD etc. are various can be with the medium of store program codes.Need
Illustrate, herein, such as first and second or the like relational terms be used merely to by an entity or operation with
Another entity or operation make a distinction, and not necessarily require or imply between these entities or operation any this reality be present
The relation or order on border.Moreover, term " comprising ", "comprising" or its any other variant are intended to the bag of nonexcludability
Contain, so that process, method, article or equipment including a series of elements not only include those key elements, but also including
The other element being not expressly set out, or also include for this process, method, article or the intrinsic key element of equipment.
In the absence of more restrictions, the key element limited by sentence "including a ...", it is not excluded that including the key element
Process, method, other identical element also be present in article or equipment.
The preferred embodiments of the present invention are the foregoing is only, are not intended to limit the invention, for the skill of this area
For art personnel, the present invention can have various modifications and variations.Within the spirit and principles of the invention, that is made any repaiies
Change, equivalent substitution, improvement etc., should be included in the scope of the protection.It should be noted that:Similar label and letter exists
Similar terms is represented in following accompanying drawing, therefore, once being defined in a certain Xiang Yi accompanying drawing, is then not required in subsequent accompanying drawing
It is further defined and explained.
Claims (10)
1. a kind of time series models method for building up, it is characterised in that applied to the electronic equipment of bank, for the bank
Asset-liabilities be predicted, methods described includes:
Obtain the historical data sequence of the bank;
Judge whether historical data sequence meets preparatory condition;
When the historical data sequence meets preparatory condition, multiple models fitting is carried out to the historical data sequence, obtained
Final mask group;
The final mask group is screened, obtains final mask;
According to the final mask, the prediction data sequence of the bank is obtained.
2. the method as described in claim 1, it is characterised in that described to judge whether historical data sequence meets preparatory condition
Step, in addition to:
White noise verification is carried out to the historical data sequence, to judge whether the historical data sequence is pure random sequence;
When the historical data sequence is not pure random sequence, stationary test is carried out to the historical data sequence, to sentence
Whether the historical data sequence of breaking is stationary sequence;
When the historical data sequence is not stationary sequence, the historical data sequence is carried out determining order difference, to be put down
Steady sequence;
When the historical data sequence is stationary sequence, judge that the historical data sequence meets preparatory condition.
3. the method as described in claim 1, it is characterised in that it is described that multiple models fitting is carried out to historical data sequence, obtain
The step of to final mask group, including:
Multiple models fitting is carried out to the historical data sequence, establishes multiple models, forms a mid-module group;
Obtain the residual error of each model in the mid-module group;
Residual test is carried out to the residual error of each model in the mid-module group, retains the model if residual error is white noise,
The model is rejected if residual error nonwhite noise, to obtain final mask group.
4. the method as described in claim 1, it is characterised in that it is described that final mask group is screened, obtain final mask
The step of, including:
Calculate the aicc values of each model in final mask group;
Obtain the model of aicc values minimum, and by the model alternately model;
T inspections are carried out to the alternative model, to judge whether the autoregressive coefficient of the alternative model has conspicuousness;
When the autoregressive coefficient has conspicuousness, it is the final mask to set the alternative model;
When the autoregressive coefficient does not have conspicuousness, sparse coefficient analysis is carried out to the alternative model, with obtain it is described most
Final cast.
5. method as claimed in claim 4, it is characterised in that it is described that sparse coefficient analysis is carried out to alternative model, to obtain
The step of stating final mask, including:
According to the alternative model, sparse coefficient model is established;
Obtain the aicc values of the sparse coefficient model;
When the aicc values of the alternative model are less than the aicc values of the sparse coefficient model, it is final to set the alternative model
Model;
When the aicc values of the sparse coefficient model are less than the aicc values of the alternative model, the sparse coefficient model is set for most
Final cast.
6. the method as described in claim 1, it is characterised in that it is described according to final mask, obtain the prediction number of the bank
The step of according to sequence, including:
Using the final mask, the prediction data sequence of the bank is exported;
Using Bonferroni algorithms, the exceptional value in the historical data sequence and the prediction data sequence is detected.
7. a kind of time series models establish device, it is characterised in that applied to the electronic equipment of bank, for the bank
Asset-liabilities be predicted, described device includes:
Historical data retrieval module, for obtaining the historical data sequence of the bank;
Judge module, for judging whether historical data sequence meets preparatory condition;
Final mask group obtains module, for when the historical data sequence meets preparatory condition, to the historical data sequence
Row carry out multiple models fitting, obtain final mask group;
Final mask obtains module, for being screened to the final mask group, obtains final mask;
Prediction data sequence obtains module, for according to the final mask, obtaining the prediction data sequence of the bank.
8. device as claimed in claim 7, it is characterised in that the judge module includes:
White noise verification unit, for carrying out white noise verification to the historical data sequence, to judge the historical data sequence
Whether row are pure random sequence;
Stationary test unit, for when the historical data sequence is not pure random sequence, to the historical data sequence
Stationary test is carried out, to judge whether the historical data sequence is stationary sequence;
Jump subdivision is determined, for when the historical data sequence is not stationary sequence, being carried out to the historical data sequence
Order difference is determined, to obtain stationary sequence;
Identifying unit, when the historical data sequence is stationary sequence, judge that the historical data sequence meets preparatory condition.
9. device as claimed in claim 7, it is characterised in that the final mask, which obtains module, to be included:
Aicc value computing units, for calculating the aicc values of each model in final mask group;
Alternative model obtaining unit, the model minimum for obtaining aicc values, and by the model alternately model;
Conspicuousness judging unit, for carrying out t inspections to the alternative model, to judge the autoregressive coefficient of the alternative model
Whether there is conspicuousness;
First execution unit, it is described final for when the autoregressive coefficient has conspicuousness, setting the alternative model
Model;
Second execution unit, for when the autoregressive coefficient does not have conspicuousness, sparse coefficient to be carried out to the alternative model
Analysis, to obtain the final mask.
10. device as claimed in claim 9, it is characterised in that second execution unit includes:
Sparse coefficient model establishes unit, for according to the alternative model, establishing sparse coefficient model;
Aicc value acquiring units, for obtaining the aicc values of the sparse coefficient model;
First sub- execution unit, for when the alternative model aicc values be less than the sparse coefficient model aicc values when, if
It is final mask to put the alternative model;
Second sub- execution unit, for when the sparse coefficient model aicc values be less than the alternative model aicc values when, if
It is final mask to put the sparse coefficient model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710878449.1A CN107633331A (en) | 2017-09-26 | 2017-09-26 | Time series models method for building up and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710878449.1A CN107633331A (en) | 2017-09-26 | 2017-09-26 | Time series models method for building up and device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107633331A true CN107633331A (en) | 2018-01-26 |
Family
ID=61103798
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710878449.1A Pending CN107633331A (en) | 2017-09-26 | 2017-09-26 | Time series models method for building up and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107633331A (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108389631A (en) * | 2018-02-07 | 2018-08-10 | 平安科技(深圳)有限公司 | Varicella morbidity method for early warning, server and computer readable storage medium |
CN109783877A (en) * | 2018-12-19 | 2019-05-21 | 平安科技(深圳)有限公司 | Time series models method for building up, device, computer equipment and storage medium |
CN110163122A (en) * | 2019-04-30 | 2019-08-23 | 中国科学院西安光学精密机械研究所 | A kind of crowded crowd's method for detecting abnormality and system based on semi-supervised dictionary learning |
CN110365518A (en) * | 2019-06-05 | 2019-10-22 | 华南理工大学 | Virtual machine bandwidth allocation methods of the OpenStack based on application service |
CN110378541A (en) * | 2019-08-08 | 2019-10-25 | 上海交通大学 | A kind of short-term multistep forecasting method of wind power and device |
CN110570322A (en) * | 2019-09-17 | 2019-12-13 | 西北农林科技大学 | Agricultural meteorological index insurance rate determination method based on time series simulation |
CN110930258A (en) * | 2019-11-15 | 2020-03-27 | 安徽海汇金融投资集团有限公司 | Receivable financial financing cash scale prediction method and system |
CN111414999A (en) * | 2020-04-27 | 2020-07-14 | 新智数字科技有限公司 | Method and device for monitoring running state of equipment |
CN112508283A (en) * | 2020-12-12 | 2021-03-16 | 广东电力信息科技有限公司 | Method and device for constructing time series model |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104657788A (en) * | 2015-02-04 | 2015-05-27 | 国家电网公司 | Method for predicting key industrial electricity consumption based on industrial condition index |
CN104794112A (en) * | 2014-01-16 | 2015-07-22 | 中国移动通信集团山西有限公司 | Time series processing method and device |
CN104899663A (en) * | 2015-06-17 | 2015-09-09 | 北京奇虎科技有限公司 | Data prediction method and apparatus |
CN105046045A (en) * | 2015-06-05 | 2015-11-11 | 河海大学 | Bayesian combination-based Web service QoS prediction method |
US20160055494A1 (en) * | 2014-08-19 | 2016-02-25 | Boyi Ni | Booking based demand forecast |
CN105974495A (en) * | 2016-04-29 | 2016-09-28 | 中国科学院遥感与数字地球研究所 | Method for pre-judging future average cloud amount of target area by using classification fitting method |
CN106295888A (en) * | 2016-08-12 | 2017-01-04 | 东南大学 | A kind of public building based on measured data is joined and is built parking position and share time window and determine method |
-
2017
- 2017-09-26 CN CN201710878449.1A patent/CN107633331A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104794112A (en) * | 2014-01-16 | 2015-07-22 | 中国移动通信集团山西有限公司 | Time series processing method and device |
US20160055494A1 (en) * | 2014-08-19 | 2016-02-25 | Boyi Ni | Booking based demand forecast |
CN104657788A (en) * | 2015-02-04 | 2015-05-27 | 国家电网公司 | Method for predicting key industrial electricity consumption based on industrial condition index |
CN105046045A (en) * | 2015-06-05 | 2015-11-11 | 河海大学 | Bayesian combination-based Web service QoS prediction method |
CN104899663A (en) * | 2015-06-17 | 2015-09-09 | 北京奇虎科技有限公司 | Data prediction method and apparatus |
CN105974495A (en) * | 2016-04-29 | 2016-09-28 | 中国科学院遥感与数字地球研究所 | Method for pre-judging future average cloud amount of target area by using classification fitting method |
CN106295888A (en) * | 2016-08-12 | 2017-01-04 | 东南大学 | A kind of public building based on measured data is joined and is built parking position and share time window and determine method |
Non-Patent Citations (3)
Title |
---|
刘明: "《线性回归模型基础应用问题研究》", 31 December 2016, 北京:中国统计出版社 * |
孙继国等: "我国银行间同业拆借利率的时间序列预测模型", 《统计与决策》 * |
王德发: "《计量经济学》", 30 September 2016, 上海:上海财经大学出版社 * |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108389631A (en) * | 2018-02-07 | 2018-08-10 | 平安科技(深圳)有限公司 | Varicella morbidity method for early warning, server and computer readable storage medium |
CN109783877A (en) * | 2018-12-19 | 2019-05-21 | 平安科技(深圳)有限公司 | Time series models method for building up, device, computer equipment and storage medium |
CN109783877B (en) * | 2018-12-19 | 2024-03-01 | 平安科技(深圳)有限公司 | Time sequence model establishment method, device, computer equipment and storage medium |
CN110163122A (en) * | 2019-04-30 | 2019-08-23 | 中国科学院西安光学精密机械研究所 | A kind of crowded crowd's method for detecting abnormality and system based on semi-supervised dictionary learning |
CN110365518A (en) * | 2019-06-05 | 2019-10-22 | 华南理工大学 | Virtual machine bandwidth allocation methods of the OpenStack based on application service |
CN110378541A (en) * | 2019-08-08 | 2019-10-25 | 上海交通大学 | A kind of short-term multistep forecasting method of wind power and device |
CN110570322A (en) * | 2019-09-17 | 2019-12-13 | 西北农林科技大学 | Agricultural meteorological index insurance rate determination method based on time series simulation |
CN110570322B (en) * | 2019-09-17 | 2023-09-12 | 西北农林科技大学 | Agricultural meteorological index insurance rate calibrating method based on time sequence simulation |
CN110930258A (en) * | 2019-11-15 | 2020-03-27 | 安徽海汇金融投资集团有限公司 | Receivable financial financing cash scale prediction method and system |
CN111414999A (en) * | 2020-04-27 | 2020-07-14 | 新智数字科技有限公司 | Method and device for monitoring running state of equipment |
CN111414999B (en) * | 2020-04-27 | 2023-08-22 | 新奥新智科技有限公司 | Method and device for monitoring running state of equipment |
CN112508283A (en) * | 2020-12-12 | 2021-03-16 | 广东电力信息科技有限公司 | Method and device for constructing time series model |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107633331A (en) | Time series models method for building up and device | |
Haslbeck et al. | A tutorial on estimating time-varying vector autoregressive models | |
CN107818344B (en) | Method and system for classifying and predicting user behaviors | |
CN106651057B (en) | Mobile terminal user age prediction method based on installation package sequence list | |
US10528533B2 (en) | Anomaly detection at coarser granularity of data | |
TWI723476B (en) | Interpretation feature determination method, device and equipment for abnormal detection | |
CN111309614B (en) | A/B test method and device and electronic equipment | |
JP2020501232A (en) | Risk control event automatic processing method and apparatus | |
US11425012B2 (en) | Dynamically generating visualizations of data based on correlation measures and search history | |
CN110472802B (en) | Data characteristic evaluation method, device and equipment | |
CN111797861A (en) | Information processing method, information processing apparatus, storage medium, and electronic device | |
CN112131322A (en) | Time series classification method and device | |
CN109213554A (en) | A kind of icon layout method, computer readable storage medium and terminal device | |
CN116684330A (en) | Traffic prediction method, device, equipment and storage medium based on artificial intelligence | |
CN107067276A (en) | Determine the method and device of object influences power | |
CN114418187A (en) | River channel hydrological information prediction method and system, terminal equipment and storage medium | |
CN109308660B (en) | Credit assessment scoring model evaluation method, apparatus, device and storage medium | |
CN113762421B (en) | Classification model training method, flow analysis method, device and equipment | |
CN113051911B (en) | Method, apparatus, device, medium and program product for extracting sensitive words | |
CN108429632B (en) | Service monitoring method and device | |
CN113486933B (en) | Model training method, user identity information prediction method and device | |
CN114417982A (en) | Model training method, terminal device and computer readable storage medium | |
CN113626734A (en) | User behavior guiding method and device, electronic equipment and storage medium | |
CN115221663A (en) | Data processing method, device, equipment and computer readable storage medium | |
CN113393303A (en) | Article recommendation method, device, equipment and storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180126 |