CN105654381A - Predicting system for business transaction volume - Google Patents

Abstract

The invention relates to the field of data mining sequence, and particularly to a predicting system for business transaction volume. The predicting system comprises the components of a business abnormity detecting module which determines whether historical business transaction volume data in a certain time period are normal; a trend index measuring module which is connected with the business abnormity detecting module and obtains a trend index according to the historical business transaction volume data when the historical business transaction volume data are normal; a period index measuring module which is connected with the business abnormity detecting module and obtains a period index according to the historical business transaction volume data when the historical business transaction volume data are normal; a season index measuring module which is connected with the business abnormity detecting module and obtains a season index according to the historical business transaction volume data when the historical business transaction volume data are normal; and a transaction volume predicting module which is connected with the trend index measuring module, the period index measuring module and the season index measuring module and performs business transaction volume prediction according to the trend index, the period index and the season index, thereby obtaining a predicted value of the business transaction volume.

Description

The pre-examining system of business turnover

Technical field

The present invention relates to data mining sequential field, particularly relate to the pre-examining system of a kind of business turnover.

Background technology

T+0 settles accounts transaction business and requires that enterprise provides fund pool for oneself, for paying for first of being deducted a percentage by each commission merchant's transaction business. Providing fund pool for oneself and arranged the little development that can affect business, arrange the excessive waste that can cause again fund cost, therefore look-ahead for turnover becomes to be necessary very much. Existing statistical method is as all too coarse in moving average, result that index is level and smooth, other require that the time of data has continuity on the one hand such as Seasonal decomposition method, ARIMA method, become again not to be very flexible when considering related factor on the other hand. Have the handiness in operation due to T+0 business, when business at weekend business at the early-stage, Holidays is closed, above-mentioned algorithm just can not meet actual needs completely.

Summary of the invention

For prior art Problems existing, the present invention provides the pre-examining system of a kind of business turnover, it is contemplated that the influence factor of multiple business volume, it is to increase the precision of turnover prediction.

The present invention adopts following technical scheme:

A pre-examining system for business turnover, described pre-examining system comprises:

Service exception detection module, judges that whether the history business turnover data in the period are normal;

Trend index measuring and calculating module, is connected with described service exception detection module, when described history business turnover data are normal, obtains Trend index according to described history business turnover data;

Periodic index measuring and calculating module, is connected with described service exception detection module, when described history business turnover data are normal, obtains periodic index according to described history business turnover data;

Season, index measurement method module, was connected with described service exception detection module, when described historical trading volume is normal, obtained index in season according to described history business turnover data;

Turnover prediction module, calculate with described Trend index respectively module, described periodic index measuring and calculating module, described season index measurement method model calling, according to described Trend index, described periodic index, described season exponent pair business turnover predict, obtain business turnover predictor.

Preferably, described pre-examining system also comprises:

Abnormal data correcting module, is connected with described service exception detection module, described turnover prediction module respectively; And,

When described history business turnover is abnormal, described history business turnover data are revised by described abnormal data correcting module, obtain revising history business turnover data, described turnover prediction module according to described correction history business historical trading volume, described Trend index, described periodic index, described season exponent pair business turnover predict, obtain business turnover predictor.

Preferably, described history business turnover data comprise date, period and the history business turnover corresponding with described date, period.

Preferably, described service exception detection module comprises,

Initialization unit, history business turnover data described in initialize;

Training unit, is connected with described initialization unit, learns the statistics rule of described history business turnover data, obtains having the rule collection of upper limit index and lower limit;

Abnormality detecting unit, is connected with described training unit, judges described history business datum volume whether in the scope of reach the standard grade index and the described index that rolls off the production line of described rule collection; And

If then described history business turnover data are normal, otherwise described history business turnover data are abnormal.

Preferably, described training unit comprises:

Reading unit, reads described history business turnover data;

Ratio index unit, is connected with described reading unit, calculates the ratio index of the history business turnover data of each period according to the date;

Calculate unit, it is connected with described ratio index unit, described reading unit respectively, calculates average and the standard deviation of the described ratio index of each period, and calculate average and the standard deviation of the described history business turnover data of each period;

Rule collection unit, according to the average of the average of described ratio index and standard deviation, described history business turnover data and standard deviation, obtains having the rule collection of upper limit index and lower limit.

Preferably, described calculating unit comprises:

Ratio exponent calculation unit, calculates average and the standard deviation of the described ratio index of each period; Wherein,

The calculation formula of the average PrBAR of described ratio index is:

The calculation formula of the standard deviation PrSD of described ratio index is:

$P_{r}SD=\sqrt{\frac{{\Σ}_{d\∈\[1,n\]}(P_{dr}-\stackrel{\‾}{P_r})}{n}};$

Turnover Data Computation Unit, calculates average and the standard deviation of the described history business turnover data of each period; Wherein,

The calculation formula of the average TrBAR of described history business turnover data is:

$\stackrel{\‾}{T_r}BAR=\frac{{\Σ}_{d\∈\[1,n\]}{T}_{d,r}}{n};$

The calculation formula of the standard deviation TrSD of described history business turnover data is:

$T_{r}SD=\sqrt{\frac{{\Σ}_{d\∈\[1,n\]}(T_{d,r}-\stackrel{\‾}{T_r})}{n}};$

Wherein, d is date codes; R is period coding; N is quantity on working days corresponding to described history business turnover data.

Preferably, described abnormality detecting unit comprises:

Full-time abnormality detecting unit, calculates the ratio index of each period of a day of trade, obtains the first ratio index, judges described first ratio index whether in the scope of reach the standard grade index and the described index that rolls off the production line of described rule collection;

Abnormality detecting unit on the half, calculates the ratio index of each period of half day of trade, obtains the 2nd ratio index, judges described 2nd ratio index whether in the scope of reach the standard grade index and the described index that rolls off the production line of described rule collection.

Preferably, described Trend index measuring and calculating module comprises:

Transfer unit, transfer described history business turnover data;

Period index unit, is connected with described unit of transferring, and according to the period, described history business turnover data is divided into multiple part, calculates the period index of each day of trade of each part;

Trend index unit, is connected with described period index unit, calculates the Trend index of each part according to the tantile of the period index of each part described;

Trend amending unit, is connected with described Trend index unit, is revised by the described Trend index of each part so that the described Trend index of each part and be 1.

Preferably, the tantile of each part described and be 1, and the tantile of each part is all at the two ends of median of tantile.

Preferably, described periodic index measuring and calculating module comprises:

Mean value computation unit, the average of each period ratio index in computation period;

Cycle measuring and calculating unit, is connected with described mean value computation unit, periodic index according to the mean value computation of each period ratio index described.

Preferably, index measurement method module in described season comprises:

Rejected unit, calculates the ratio index in multiple stages of seasonal effect;

Season, index unit, was connected with described rejected unit, index in season according to the ratio Index for Calculation in multiple stages of described seasonal effect.

The invention has the beneficial effects as follows:

The system of the present invention is on the basis that with reference to the multiple algorithm of time series, incorporate the trend that comprises T+0 account settlement business, the moon seasonal, periodically, the data of the many aspects such as period acceleration and abnormal correction, define the brand-new pre-examining system of most of scene that often can occur in reply business. This system can significant response business trend, the season moon sexual factor and the transaction change that brings of the impact of periodically factor, the impact of the abnormal index that the response of abnormal correction algorithm brings can also be utilized due to systematicness technology fault, period acceleration factor detection can improve the business trading volume sudden change that singularity factor is brought, and reaches generally the good prediction effect of turnover.

Accompanying drawing explanation

Fig. 1 is the structural representation of the pre-examining system of a kind of turnover of the present invention;

Fig. 2 is the principle of work figure of training unit of the present invention;

Fig. 3 is the principle of work figure of Trend index of the present invention measuring and calculating module;

Fig. 4 is the data structure figure in the present invention's rule training process;

Fig. 5 is the changing trend diagram of day part turnover ratio of the present invention;

Fig. 6 is the data structure figure of turnover trend of the present invention;

Fig. 7 is the design sketch that the present invention totally predicts.

Embodiment

It should be noted that, when not conflicting, following technical proposals, can combine between technology feature mutually.

Below in conjunction with accompanying drawing, the specific embodiment of the present invention is further described:

The prediction of turnover is mainly provided a kind of predictive model by the present embodiment, the model of the present embodiment has achieved good prediction effect, its overall prediction effect is as shown in Figure 7, this predictive model is that the pre-examining system based on the present invention realizes, the target of the present embodiment predictive model: realize the prediction to T0 account settlement business whole day turnover.

The input data of this model are: date, period, turnover;

The output data of this model are: the turnover of date, prediction;

The data area of constrained input can be: normal working day, gets rid of weekend, the legal festivals and holidays;

The prediction time point of this model: between same day 14:00 to 14:30, after when 13, data produce; Above-mentioned model parameter is only a kind of embodiment of the present invention, and the concrete numerical value of its parameter can set according to practical situation, so not forming the restriction of the present invention.

Embodiment one

As shown in Figure 1, the pre-examining system of a kind of business turnover, described pre-examining system comprises:

Service exception detection module, judges that whether the history business turnover data in the period are normal;

Trend index measuring and calculating module, is connected with described service exception detection module, when described history business turnover data are normal, obtains Trend index according to described history business turnover data;

Periodic index measuring and calculating module, is connected with described service exception detection module, when described history business turnover data are normal, obtains periodic index according to described history business turnover data;

Season, index measurement method module, was connected with described service exception detection module, when described historical trading volume is normal, obtained index in season according to described history business turnover data;

Turnover prediction module, calculate with described Trend index respectively module, described periodic index measuring and calculating module, described season index measurement method model calling, according to described Trend index, described periodic index, described season exponent pair business turnover predict, obtain business turnover predictor.

Abnormal data correcting module, is connected with described service exception detection module, described turnover prediction module respectively; And,

When described history business turnover is abnormal, described history business turnover data are revised by described abnormal data correcting module, obtain revising history business turnover data, described turnover prediction module according to described correction history business historical trading volume, described Trend index, described periodic index, described season exponent pair business turnover predict, obtain business turnover predictor.

As from the foregoing, the pre-examining system of the present embodiment comprises: service exception detection module, Trend index measuring and calculating module, periodic index measuring and calculating module, season index measurement method module, turnover prediction module, abnormal data correcting module, service exception detection module is to whether history business turnover data normally detect, the possibility of result of detection has two kinds of situations, one is normal, one is abnormal, when the result detected is normal time, calculate module by Trend index respectively according to history business turnover data and calculate Trend index, periodic index measuring and calculating module computation period index, season, index measurement method module calculated index in season, by turnover prediction module according to Trend index, periodic index, season, index carried out the prediction of turnover, when the result of history business turnover Data Detection is abnormal by service exception detection module time, history business turnover data can be revised by abnormal data correcting module, obtain revise history business turnover data, afterwards again by turnover prediction module according to correction history business historical trading volume, Trend index, periodic index, season exponent pair business turnover predict.

Abnormal data correcting module in the present embodiment, the correction of its outlier only for the time period value on current date, if service exception judge in be judged to abnormal data, it is necessary to these data are revised.

If the part period is abnormal, its correction formula is:

$\begin{array}{cc}{T}_{d+1,r}=\frac{\stackrel{\‾}{P_r}}{{\Σ}_{i=1}^{13}\stackrel{\‾}{P_{-i}}}\×{\Σ}_{i=1}^{13}{T}_{d+1,-i}& r\∈\[0,13\]\end{array};$

R code:

T[d+1,i]<-apply(T[d+1,c(0:i-1,i+1:13)],1,sum)*P[i]

BAR/apply (P [-i] BAR, 1, sum);

When there being the two or more period abnormal, formula above is utilized abnormal period value to be revised respectively, but abnormal period value does not participate in other time period value must being revised.

If data are all abnormal before 14, then make prediction according to actual value, outlier is not done any correction; Before 14, business terminates, and namely portfolio returns 0, then stop the prediction work when daily trading volume. Outlier correction is only for the monitoring result between working days 14:00-14:30, it is intended that for follow-up prediction.

In the present invention's preferred embodiment, described history business turnover data comprise date, period and the history business turnover corresponding with described date, period.

In the present invention's preferred embodiment, described service exception detection module comprises,

Initialization unit, history business turnover data described in initialize;

Training unit, is connected with described initialization unit, learns the statistics rule of described history business turnover data, obtains having the rule collection of upper limit index and lower limit;

Abnormality detecting unit, is connected with described training unit, judges described history business datum volume whether in the scope of reach the standard grade index and the described index that rolls off the production line of described rule collection; And

If then described history business turnover data are normal, otherwise described history business turnover data are abnormal.

In the present embodiment, native system needs the abnormality detection carrying out business turnover, as shown in Figure 1, the pre-examining system of business turnover realizes solving this technical problem of turnover abnormality detection primarily of service exception detection module, service exception detection module can transfer history business turnover data (hereinafter referred to as transaction data) of a period from a core system or business system, service exception detection module can by the study of the transaction data to the past period, the mathematical law obtained is applied on current transaction data, reach the whether judgement in the reasonable scope of current transaction data, finally normal transaction data can be preserved the history business turnover data for arm's length transaction, for the study to follow-up mathematical law, in fact an effect upgraded is served.

In the present embodiment, service exception detection module comprises initialization unit, the function of initialization unit is initialize ' T0 business arm's length transaction day data ', these " T0 business arm's length transaction day data " namely can be regarded as before " transaction data ", but " T0 business arm's length transaction day data " normally occur in table form, it is mainly to state out the date, period, and the date, corresponding relation between period and turnover, in order to solve the cold start-up problem of service exception detection module subsequent algorithm in the present embodiment model, need normal calendar process history data are carried out initialize, the method of initialize is the arm's length transaction date utilizing business experience to select some amount, corresponding with the date for given date data is inserted ' T0 business arm's length transaction day data ', to complete the initialize of this form. data object in table is: the T0 business every day of turnover at times, date range can be 2015/8/13,2015/8/14,2015/8/17,2015/8/18,2015/8/19,2015/8/20,2015/8/24,2015/8/25,2015/8/27,2015/8/31,2015/9/1,2015/9/2, data content is: date, period (0-23), turnover (corresponding with above-mentioned date-time), wherein, during for 0, can represent 0:00:00��0:59:59 when 0, follow-up does not similarly enumerate.

Training unit in service exception detection system is connected with above-mentioned initialization unit, it is mainly used in carrying out rule training, rule and service exception detect the reference standard collection (can collect) used referred to as rule, rule training learns to add up accordingly rule exactly from the transaction data of designated length, and calculates rule collection with this. This step does not need to carry out in real time, it is possible to the 0:00-1:00 in morning in every day is performed on this, but is not limited to this time, sincerely for this time.

In the present invention's preferred embodiment, described training unit comprises:

Reading unit, reads described history business turnover data;

Ratio index unit, is connected with described reading unit, calculates the ratio index of the history business turnover data of each period according to the date;

Calculate unit, it is connected with described ratio index unit, described reading unit respectively, calculates average and the standard deviation of the described ratio index of each period, and calculate average and the standard deviation of the described history business turnover data of each period;

Rule collection unit, according to the average of the average of described ratio index and standard deviation, described history business turnover data and standard deviation, obtains having the rule collection of upper limit index and lower limit.

In the present embodiment, as shown in Figure 2, the step of the rule training of training unit can comprise: step 1.1: reading is apart from the data on nearest 5 working dayss on current date from ' T0 business arm's length transaction date data ', the present embodiment is for 5 working dayss, can also reading the data on other working dayss all or part of, its data structure read can be as shown in Figure 4.

The R code of above-mentioned reading data is: N <-nrow (T0);

T <-T0 [N-5:N, 0:23];

Whole step 1.1 is completed by the reading unit in training unit, perform step 1.2 afterwards, step 1.2 is by the ratio index unit in training unit, step 1.2 mainly comprises the ratio indices P that a point day of trade (namely above-mentioned working days) calculates day part, and the calculation formula of ratio index is:

$\begin{array}{cc}{P}_{d,r}=\frac{T_{d,r}}{{\&Sigma;}_{s=0}^{23}{T}_{d,s}}& d\&Element;\&lsqb;1,5\&rsqb;,r\&Element;\&lsqb;0,23\&rsqb;\end{array}$

Wherein, T represents trading volume, and d represents date codes, and r represents that the period encodes;

The R code of calculating ratio indices P: SUM <-apply (T, 1, sum) # calculates by row, obtains the accumulative turnover of each day of trade;

For (iin0:23) P [, i] < and-T [, i] and/SUM};

In the present invention's preferred embodiment, described calculating unit comprises:

Ratio exponent calculation unit, calculates average and the standard deviation of the described ratio index of each period;

Turnover Data Computation Unit, calculates average and the standard deviation of the described history business turnover data of each period.

Step 1.3 and step 1.4 are average and the standard deviation of the ratio index calculating each period, and calculate average and the standard deviation of the history business turnover data of each period, can completing by calculating unit of these two steps, step 1.3, ratio exponent calculation unit is based on working days these data of day part ratio index, calculating the ratio Mean value of index PrBAR and standard deviation PrSD of day part, its calculation formula is:

$\begin{array}{cc}\stackrel{\&OverBar;}{P_{r}BAR}=\frac{{\&Sigma;}_{d\&Element;\&lsqb;1,5\&rsqb;}{P}_{d,r}}{5}& r\&Element;\&lsqb;0,23\&rsqb;\end{array};$

$\begin{array}{cc}{P}_{r}SD=\sqrt{\frac{{\&Sigma;}_{d\&Element;\&lsqb;1,5\&rsqb;}{(P_{d,r}-\stackrel{\&OverBar;}{P_r})}^2}{5}}& r\&Element;\&lsqb;0,23\&rsqb;\end{array};$

The R code of calculating ratio Mean value of index PrBAR and standard deviation PrSD can be:

PBAR <-apply (P, 2, mean);

PSD <-apply (P, 2, sd);

Step 1.4: turnover Data Computation Unit is based on working days day part turnover data, and the formula of the turnover average TrBAR and standard deviation TrSD that calculate day part is:

$\begin{array}{cc}\stackrel{\&OverBar;}{T_r}BAR=\frac{{\&Sigma;}_{d\&Element;\&lsqb;1,5\&rsqb;}{T}_{d,r}}{5}& r\&Element;\&lsqb;0,23\&rsqb;\end{array};$

$\begin{array}{cc}{T}_{r}SD=\sqrt{\frac{{\&Sigma;}_{d\&Element;\&lsqb;1,5\&rsqb;}{(T_{d,r}-\stackrel{\&OverBar;}{T_r})}^2}{5}}& r\&Element;\&lsqb;0.23\&rsqb;;\end{array}$

The R code that step 1.4 calculates: TBAR <-apply (T, 2, mean);

TSD <-apply (T, 2, sd);

Step 1.5, rule collection unit in training unit, utilize achievement data that above step (step 1.1-step 1.4) obtains (according to the average of the average of ratio index and standard deviation, history business turnover data and standard deviation), calculate the rule collection of day part, indicator rule is concentrated has upper limit U and lower limit L, and the calculation formula of rule collection is:

$\begin{array}{cc}{P}_{r}U=\stackrel{\&OverBar;}{P_r}+3\&times;P_{r}SD& r\&Element;\&lsqb;0,23\&rsqb;,U\&Element;(0,1\&rsqb;;\end{array}$

$\begin{array}{cc}{P}_{r}L=\stackrel{\&OverBar;}{P_r}-3\&times;P_{r}SD& r\&Element;\&lsqb;0,23\&rsqb;,L\&Element;\&lsqb;0,1);\end{array}$

$\begin{array}{cc}{T}_{r}L=\stackrel{\&OverBar;}{T_r}-3\&times;T_{r}SD& r\&Element;\&lsqb;0,23\&rsqb;,L\&Element;\&lsqb;0,1);\end{array}$

The R code that rule collection calculates is: PU<-PBAR+PSD*3; PU [PU>1]<-1;

PL <-PBAR-PSD*3; PL [PL < 0] <-0;

TL <-TBAR-TSD*3TL [TL < 0] <-0;

In the present invention's preferred embodiment, described abnormality detecting unit comprises:

Full-time abnormality detecting unit, calculates the ratio index of each period of a day of trade, obtains the first ratio index, judges described first ratio index whether in the scope of reach the standard grade index and the described index that rolls off the production line of described rule collection;

Abnormality detecting unit on the half, calculates the ratio index of each period of half day of trade, obtains the 2nd ratio index, judges described 2nd ratio index whether in the scope of reach the standard grade index and the described index that rolls off the production line of described rule collection.

In the present embodiment, the rule collection that abnormality detecting unit obtains for utilizing step 1.2, completes the abnormality detection to current business turnover. Service exception detecting unit is divided into again full-time abnormality detecting unit and abnormality detecting unit on the half two unit, it is possible to realize the scene of two kinds of detections.

Wherein, full-time turnover is carried out abnormality detection by full-time abnormality detecting unit, first calculates the ratio index (the first ratio index) of current day of trade day part:

Wherein the day before yesterday is worked as in+1 expression.

Afterwards, judging period index whether within rule scope, judgment formula is:

IfP_{+ 1, r}��[L_r, U_r]andT_{+ 1, r}> T_rLthen0 (normally) else-1 (exception) r �� [0.23];

Judge period index whether R code within rule scope be:

SUM1 <-apply (T1,1, sum);

P1 <-T1/SUM1;

S1<-P1>=PL;

S2 <-P1��PU;

S3<-T1>=TL;

S <-pmin (s1, s2, s3);

Abnormality detecting unit carried out the abnormality detection of half turnover per day on 1st, and testing process is similar to full-time abnormality detection, judgement be that the 2nd ratio index knows no in the upper limit of rule collection and the scope of lower limit, the detection formula of half turnover per day abnormality detection is:

IfT_{+ L, r}> T_rLthen0 (normally) else-1 (exception) r �� [0.23];

It is noted that arbitrary period is abnormal all judges that the transaction data on the same day is abnormal, otherwise, the transaction data on the same day is added in ' T0 business arm's length transaction date data '. ' T0 business arm's length transaction date data ' tentative transaction data preserving 90 arm's length transaction days. The span of each period can be 0:00-59:59; P+1, r only for the data of clear working day, half working days data only according to T+1, r does abnormality detection, so can there be twice exception monitoring the every workday, the same day was done abnormality detection by the business datum of 13 by 14:00-14:30, for the prediction of whole day transaction data; The abnormality detection object of 23:00-0:00 judges that whether the transaction on the same day is normal, and arm's length transaction day data are appended in ' T0 business arm's length transaction date data ', and this detection for be actual transaction data but not transaction data after abnormal correction.

In the present invention's preferred embodiment, described Trend index measuring and calculating module comprises:

Transfer unit, transfer described history business turnover data;

Period index unit, is connected with described unit of transferring, and according to the period, described history business turnover data is divided into multiple part, calculates the period index of each day of trade of each part;

Trend index unit, is connected with described period index unit, calculates the Trend index of each part according to the tantile of the period index of each part described;

Trend amending unit, is connected with described Trend index unit, is revised by the described Trend index of each part so that the described Trend index of each part and be 1.

In the present embodiment, Trend index measuring and calculating unit carries out the prediction of Trend index according to the arm's length transaction data that service exception detection module obtains, because along with the development of business, the turnover ratio of day part is in the change of occurrence tendency, the ratio of the ratio of morning hours afternoon hours in decline is increasing as shown in Figure 5 gradually, therefore predicts that index needs to follow business revenue and does corresponding adjustment.

First, Trend index measuring and calculating module in transfer unit can extract from the object of ' T0 business arm's length transaction date data ' apart from the current date nearest 15 working days data, algorithm calculates basic data scope according to the Data Dynamic on nearest 15 working dayss, ensureing the trend of data, its data structure can be as shown in Figure 6. Extract 15 working days data the R code position of data structure be:

N <-nrow (T0);

T <-T0 [N-15:N, 0:23];

Afterwards, as shown in Figure 3, period information is divided into two big periods by the period index unit performed in step 2.1 Trend index measuring and calculating module, unit is connected period index unit with transferring, two periods being divided into are respectively Tmon (0-13) period and Taft (14-23) period, calculate the period indices P of each day of trade. Owing to only needing to make overall prediction according to the big period, it is not necessary to each little period is made prediction separately, so first dividing the period:

$\begin{array}{cc}{\mathrm{Pmon}}_d=\frac{{\mathrm{Tmon}}_d}{{\mathrm{Tmon}}_d+{\mathrm{Taft}}_d}& d\&Element;\&lsqb;1,15\&rsqb;\end{array};$

$\begin{array}{cc}{\mathrm{Paft}}_d=\frac{{\mathrm{Taft}}_d}{{\mathrm{Tmon}}_d+{\mathrm{Taft}}_d}& d\&Element;\&lsqb;1,15\&rsqb;;\end{array}$

The R code divided is: Tmon <-apply (T [, 0:13], 1, sum);

Taft <-apply (T [, 14:23], 1, sum);

Tsum <-apply (T, 1, sum);

Pmon <-Tmon/Tsum;

Paft<-Taft/Tsum

Step 2.2, Trend index unit in Trend index measuring and calculating module calculates P (mon respectively, d) the 0.15th tantile PmonBAR of array, P (aft, d) the 0.85th tantile PaftBAR of array, Trend index unit is connected with period index unit, based on the actual needs of business, expect that all predictors can higher than actual value, so not doing seasonal parameter calculating with the median of array or average, but with the 0.15th fractile of Pmon as denominator, taking the 0.85th fractile of Paft as molecule, guarantee like this prediction phase season index can major part period higher than actual value, by following formula can be preliminary calculate Trend index:

$\begin{array}{cc}\stackrel{\&OverBar;}{Pmon}=\frac{small({\mathrm{Pmon}}_d,2)+small({\mathrm{Pmon}}_d,3)}{2}& d\&Element;\&lsqb;1,15\&rsqb;\end{array};$

$\begin{array}{cc}\stackrel{\&OverBar;}{Paft}=\frac{l\arg er({\mathrm{Paft}}_d,2)+l\arg er({\mathrm{Paft}}_d,3)}{2}& d\&Element;\&lsqb;1,15\&rsqb;\end{array};$

The R code calculated is: PmonBAR <-quantile (Pmon, probs=0.15);

PaftBAR <-quantile (Paft, probs=0.85);

Step 2.3, the subelement that trend amending unit calculates module as Trend index is connected with Trend index unit, is mainly used in revising PmonBAR, PaftBAR, obtains Pmon'BAR and Paft'BAR. Owing to utilizing the fractile calculation result of different ordered series of numbers, PnonBAR and PaftBAR and and be not equal to 1, so to PmonBAR, PaftBAR revises, to guarantee that Pmon'BAR and Paft'BAR meets the constraint of percentage value ratio, the preliminary Trend index obtained before is revised by trend amending unit by following formula.

$\stackrel{\&OverBar;}{{\mathrm{Pmon}}^{\&prime;}}=\frac{\stackrel{\&OverBar;}{Pmon}}{\stackrel{\&OverBar;}{Pmon}+\stackrel{\&OverBar;}{Paft}};$

$\stackrel{\&OverBar;}{{\mathrm{Paft}}^{\&prime;}}=\frac{\stackrel{\&OverBar;}{Paft}}{\stackrel{\&OverBar;}{Pmon}+\stackrel{\&OverBar;}{Paft}};$

R code is: Pmon ' BAR <-PmonBAR/ (PmonBAR+PaftBAR);

Paft��BAR<-PaftBAR/(PmonBAR+PaftBAR)��

In the present invention's preferred embodiment, the tantile of each part described and be 1, and the tantile of each part is all at the two ends of median of tantile.

In the present invention's preferred embodiment, described periodic index measuring and calculating module comprises:

Mean value computation unit, the average of each period ratio index in computation period;

Cycle measuring and calculating unit, is connected with described mean value computation unit, periodic index according to the mean value computation of each period ratio index described.

In the present embodiment, the pre-examining system of the present embodiment also comprises periodic index measuring and calculating module, periodic index measuring and calculating module also can carry out the prediction of periodic index according to the arm's length transaction data obtained in service exception detection module, periodic index measuring and calculating module can at the Different periods (such as: it can be that Monday is to Friday) in a week, transaction data shows the fluctuation with some cycles, has reacted the impact of cycle for human consumer's consumption habit. As Monday, Friday consumption the highest, Wednesday is the lower-most point of a week. The object of periodic index measuring and calculating is exactly find out periodically quantization amplitude for the impact of human consumer's consumption habit, is used for the prediction to turnover and makes correction. (remarks: owing to period day index is corresponding, itself and be 1, so only needing to calculate the periodicity of one of them period, the present embodiment select morning hours): calculation procedure is as follows:

First, the ratio Mean value of index of Different periods in the computation period that the mean value computation unit calculated by periodic index is concrete, calculation formula as follows described in:

$\begin{array}{cc}\stackrel{\&OverBar;}{{\mathrm{Pmon}}_c}=\frac{{\&Sigma;}_{d=1}^{15}{\mathrm{Pmon}}_{c,d}}{3}& c\&Element;\&lsqb;1,5\&rsqb;\end{array};$

The R code that mean value computation unit in the present embodiment calculates ratio Mean value of index can be expressed as follows:

Pmon $ w <-as.factor (format (Pmon $ d, " %A ")) # increases period field w in the data of ratio index in morning day;

PmonBAR <-tapply (Pmon [, mon], Pmon $ w, mean) # calculates the average of ratio index in period times;

Afterwards, the subelement that cycle measuring and calculating unit is calculated as periodic index is used for computation period index C, and cycle measuring and calculating unit is connected with mean value computation unit:

$\begin{array}{cc}{C}_c=\frac{\stackrel{\&OverBar;}{{\mathrm{Pmon}}_c}\&times;5}{{\&Sigma;}_{i=1}^5\stackrel{\&OverBar;}{{\mathrm{Pmon}}_1}}& c\&Element;\&lsqb;1,5\&rsqb;\end{array};$

The R code that cycle measuring and calculating unit calculates is:

C $ w <-Pmon $ w;

Periodic index is left in C by C $ Cd <-PmonBAR*5/sum (PmonBAR) #;

In the present invention's preferred embodiment, described season, index measurement method module comprised:

Rejected unit, calculates the ratio index in multiple stages of seasonal effect;

Season, index unit, was connected with described rejected unit, index in season according to the ratio Index for Calculation in multiple stages of described seasonal effect.

In the present embodiment, pre-examining system also comprises index measurement method module in season, model calling is calculated with periodic index, it is mainly used in carrying out seasonal index measurement method, in the different steps of every month, such as obvious with 21 to 30, data sheet reveals obviously seasonal fluctuation, and the consumer behaviour having reacted human consumer is subject to the impact of seasonal factor. Season, the object of index measurement method module finds out the seasonal quantization amplitude for the impact of human consumer's consumption habit in time, is used for the prediction to turnover and makes correction. Owing to period day index is corresponding, itself and be 1, so only needing to calculate the seasonality of one of them period, the present embodiment select morning hours. Such as selecting the transaction data of nearest 3 phases to calculate seasonality, it is thus desirable to select the transaction data of nearest 3 months, the calculation procedure of collected explanations or commentaries index measurement method module is as follows:

Comprising periodically due to period ratio index initial value and seasonal factor, therefore first get rid of the impact of periodically factor by the rejected unit in season index measurement method module, R code is:

Periodic index field is joined period ratio exponent data and concentrates by Pmon <-merge (Pmon, C, by=c (" w ")) #;

Pmon $ monS <-Pmon $ mon/Pmon $ Cd# increases field monS and deposits the ratio index after getting rid of periodicity factor;

Index unit median in season in collected explanations or commentaries index measurement method module represents the ratio index level of each date section afterwards, season, index unit was connected with rejected unit, 4 stages will be divided into by same date: 1 (1-7) 2 (8-14) 3 (15-21) 4 (22-31), it can divide moon ratio index level calculating each stage, and calculation formula is as follows:

PL_{M, l}=small (Pmon_{M, l}, 4) and l �� [1.4];

Wherein, M represents the month belonging to data, and l represents the stage on date, maximum 4, then represents the horizontal R code of ratio index in m l stage of the month in range of choice data with PL (m, l):

Pmon $ l [Pmon $ d��7] <-1;

Pmon $ l [Pmon $ d��14] <-2;

Pmon $ l [Pmon $ d��21] <-3;

Pmon $ l [Pmon $ d>=22]<-4;

Pmon $ l <-as.factor (Pmon $ l);

PL <-tapply (Pmon $ monS, c (Pmon $ m, Pmon $ l), median);

Calculate the seasonal index of each date section again, comprising: the index in season calculating point moon each date section:

$\begin{array}{cc}{S}_{m,l}=\frac{{\mathrm{PL}}_{m,l}\&times;4}{{\&Sigma;}_{i=1}^4{\mathrm{PL}}_{m,i}}& l\&Element;\&lsqb;1,4\&rsqb;;\end{array}$

Wherein, m is corresponding month; Calculate 3 months averages of each date section index in season:

N represents that selection periodic packets is containing month;

Revise seasonal index initial value, obtain final index in season:

$\begin{array}{cc}{S}_l=\frac{{\stackrel{\&OverBar;}{S}}_l\&times;4}{{\&Sigma;}_{i=1}^4{\stackrel{\&OverBar;}{S}}_l}& l\&Element;\&lsqb;1,4\&rsqb;\end{array}$

R code:

SM <-tapply (PL $ monS, PL $ m, sum);

SM $ m <-PL $ m;

PL <-merge (PL, SM, by=c (" m "));

Sml <-PL $ monS*4/PL $ SM;

Sml <-cbind (Sml, PL $ l);

SlBAR <-tapply (Sml $ monS, Sml $ l, mean);

Sl <-SlBAR*4/sum (SlBAR);

Sl<-Cbind(Sl,SlBAR$l).

The pre-examining system of the present embodiment also comprises trading volume prediction module, can predict the turnover T (+1) of whole day according to the turnover of that morning period:

$\begin{array}{cc}{\mathrm{Pmon}}_{c,l}=\stackrel{\&OverBar;}{{\mathrm{Pmon}}^{\&prime;}}\&times;C_c\&times;S_l& c\&Element;\&lsqb;1,5\&rsqb;,l\&Element;\&lsqb;1,4\&rsqb;\end{array};$

Paft_{C, l}=1-Pmon_{C, l}C �� [1,5], l �� [1,4];

$T_{d+1}=(1+\frac{\stackrel{\&OverBar;}{{\mathrm{Paft}}^{\&prime;}}}{\stackrel{\&OverBar;}{{\mathrm{Pmon}}^{\&prime;}}})\&times;{\&Sigma;}_{i=1}^{13}{T}_{d+1,i};$

R code: Pmon ' BAR <-merge (Pmon ' BAR, Cc, by=c (" c ");

Pmon ' BAR <-merge (Pmon ' BAR, Sl, by=c (" l ");

Pmon <-cbind (Pmon ' BAR, Pmon ' BAR $ mon*c*l);

Paft <-1-Pmon;

T��<-apply(T[N+1,1:13],1,sum)*(Paft/Pmon+1)

In sum, the system of the present invention is on the basis that with reference to the multiple algorithm of time series, incorporate the trend that comprises T+0 account settlement business, the moon seasonal, periodically, the data of the many aspects such as period acceleration and abnormal correction, define the brand-new pre-examining system of most of scene that often can occur in reply business. This system can significant response business trend, the season moon sexual factor and the transaction change that brings of the impact of periodically factor, the impact of the abnormal index that the response of abnormal correction algorithm brings can also be utilized due to systematicness technology fault, period acceleration factor detection can improve the business trading volume sudden change that singularity factor is brought, and reaches generally the good prediction effect of turnover.

By illustrating and accompanying drawing, give the exemplary embodiments of the ad hoc structure of embodiment, based on the present invention's spirit, also can do other conversion. Although foregoing invention proposes existing better embodiment, but, these contents are not as limitation.

For a person skilled in the art, after reading above-mentioned explanation, various changes and modifications undoubtedly will be apparent. Therefore, appending claims should regard whole change and the correction of the true intention containing the present invention and scope as. In Claims scope, the scope of any and all equivalences and content, all should think and still belong to the intent and scope of the invention.

Claims (11)

1. the pre-examining system of a business turnover, it is characterised in that, described pre-examining system comprises:

Service exception detection module, judges that whether the history business turnover data in the period are normal;

Trend index measuring and calculating module, is connected with described service exception detection module, when described history business turnover data are normal, obtains Trend index according to described history business turnover data;

Periodic index measuring and calculating module, is connected with described service exception detection module, when described history business turnover data are normal, obtains periodic index according to described history business turnover data;

Season, index measurement method module, was connected with described service exception detection module, when described historical trading volume is normal, obtained index in season according to described history business turnover data;

Turnover prediction module, calculate with described Trend index respectively module, described periodic index measuring and calculating module, described season index measurement method model calling, according to described Trend index, described periodic index, described season exponent pair business turnover predict, obtain business turnover predictor.

2. the pre-examining system of business turnover according to claim 1, it is characterised in that, described pre-examining system also comprises:

Abnormal data correcting module, is connected with described service exception detection module, described turnover prediction module respectively; And,

When described history business turnover is abnormal, described history business turnover data are revised by described abnormal data correcting module, obtain revising history business turnover data, described turnover prediction module according to described correction history business historical trading volume, described Trend index, described periodic index, described season exponent pair business turnover predict, obtain business turnover predictor.

3. the pre-examining system of business turnover according to claim 1, it is characterised in that, described history business turnover data comprise date, period and the history business turnover corresponding with described date, period.

4. the pre-examining system of business turnover according to claim 3, it is characterised in that, described service exception detection module comprises,

Initialization unit, history business turnover data described in initialize;

Training unit, is connected with described initialization unit, learns the statistics rule of described history business turnover data, obtains having the rule collection of upper limit index and lower limit;

Abnormality detecting unit, is connected with described training unit, judges described history business datum volume whether in the scope of reach the standard grade index and the described index that rolls off the production line of described rule collection; And

If then described history business turnover data are normal, otherwise described history business turnover data are abnormal.

5. the pre-examining system of business turnover according to claim 4, it is characterised in that, described training unit comprises:

Reading unit, reads described history business turnover data;

Ratio index unit, is connected with described reading unit, calculates the ratio index of the history business turnover data of each period according to the date;

Calculate unit, it is connected with described ratio index unit, described reading unit respectively, calculates average and the standard deviation of the described ratio index of each period, and calculate average and the standard deviation of the described history business turnover data of each period;

Rule collection unit, according to the average of the average of described ratio index and standard deviation, described history business turnover data and standard deviation, obtains having the rule collection of upper limit index and lower limit.

6. the pre-examining system of business turnover according to claim 4, it is characterised in that, described calculating unit comprises:

Ratio exponent calculation unit, calculates average and the standard deviation of the described ratio index of each period; Wherein,

The calculation formula of the average PrBAR of described ratio index is:

The calculation formula of the standard deviation PrSD of described ratio index is: $P_{r}SD=\sqrt{\frac{{\&Sigma;}_{d\&Element;\&lsqb;1,n\&rsqb;}{(P_{d,r}-\stackrel{\&OverBar;}{P_r})}^2}{n}};$

Turnover Data Computation Unit, calculates average and the standard deviation of the described history business turnover data of each period; Wherein,

The calculation formula of the average TrBAR of described history business turnover data is: $\stackrel{\&OverBar;}{T_r}BAR=\frac{{\&Sigma;}_{d\&Element;{\&lsqb;1,n\&rsqb;}^{T}d,r}}{n};$

The calculation formula of the standard deviation TrSD of described history business turnover data is: $T_{r}SD=\sqrt{\frac{{\&Sigma;}_{d\&Element;\&lsqb;1,n\&rsqb;}{(T_{d,r}-\stackrel{\&OverBar;}{T_r})}^2}{n}};$

Wherein, d is date codes; R is period coding; N is quantity on working days corresponding to described history business turnover data.

7. the pre-examining system of business turnover according to claim 4, it is characterised in that, described abnormality detecting unit comprises:

Full-time abnormality detecting unit, calculates the ratio index of each period of a day of trade, obtains the first ratio index, judges described first ratio index whether in the scope of reach the standard grade index and the described index that rolls off the production line of described rule collection;

Abnormality detecting unit on the half, calculates the ratio index of each period of half day of trade, obtains the 2nd ratio index, judges described 2nd ratio index whether in the scope of reach the standard grade index and the described index that rolls off the production line of described rule collection.

8. the pre-examining system of business turnover according to claim 3, it is characterised in that, described Trend index measuring and calculating module comprises:

Transfer unit, transfer described history business turnover data;

Period index unit, is connected with described unit of transferring, and according to the period, described history business turnover data is divided into multiple part, calculates the period index of each day of trade of each part;

Trend index unit, is connected with described period index unit, calculates the Trend index of each part according to the tantile of the period index of each part described;

Trend amending unit, is connected with described Trend index unit, is revised by the described Trend index of each part so that the described Trend index of each part and be 1.

9. the pre-examining system of business turnover according to claim 8, it is characterised in that, the tantile of each part described and be 1, and the tantile of each part is all at the two ends of median of tantile.

10. the pre-examining system of business turnover according to claim 3, it is characterised in that, described periodic index measuring and calculating module comprises:

Mean value computation unit, the average of each period ratio index in computation period;

Cycle measuring and calculating unit, is connected with described mean value computation unit, periodic index according to the mean value computation of each period ratio index described.

The pre-examining system of 11. business turnovers according to claim 3, it is characterised in that, described season, index measurement method module comprised:

Rejected unit, calculates the ratio index in multiple stages of seasonal effect;

Season, index unit, was connected with described rejected unit, index in season according to the ratio Index for Calculation in multiple stages of described seasonal effect.