CN104766144A - Order forecasting method and system - Google Patents
Order forecasting method and system Download PDFInfo
- Publication number
- CN104766144A CN104766144A CN201510195209.2A CN201510195209A CN104766144A CN 104766144 A CN104766144 A CN 104766144A CN 201510195209 A CN201510195209 A CN 201510195209A CN 104766144 A CN104766144 A CN 104766144A
- Authority
- CN
- China
- Prior art keywords
- date
- order
- module
- predicted value
- forecast
- 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 invention discloses an order forecasting method and system. The order forecasting method comprises the steps that the historical order quantity is collected; the historical order quantity is summarized according to the day granularity; dates are classified, and the historical dates and forecasting dates which belong to the same type are matched; a growth base date is selected from the historical dates matched with the forecasting dates, the order quantity and the forecasting growth rate of the growth base date is obtained, and the forecasting growth rate is determined according to the type of the forecasting dates; a day order quantity forecasting value of each forecasting date is calculated, and the day order quantity forecasting value=the order quantity of the growth base date *(1+the forecasting growth rate). The order forecasting method and system make up for the defects that in an existing forecasting method, the calculated capacity is large and the forecasting accuracy is low, and have the advantages of being small in calculated capacity, high in forecasting accuracy and strong in maneuverability.
Description
Technical field
The present invention relates to e-commerce field, especially relate to a kind of order forecast method and system.
Background technology
Along with the fast development of ecommerce; present stage, the Capability Requirement to the Data Analysis Services of computing machine was more and more higher; especially in order forecasting, a lot of electric business or online tourism website often need the order volume of look-ahead sometime in section to think daily or some special circumstances do sufficient preparation.
There are some for predicting the open source literature of order in prior art, as:
1, application publication number is a kind of product order forecast method and device with time series characteristic of CN103310286A, wavelet neural network theory is introduced in time series predicting model by this invention, carry out forecast analysis to dynamic time sequence data, the method that average is repeatedly got in circular flow is predicted the outcome.
Shortcoming: calculation of complex, accuracy is unstable.The cost of neural metwork training model is comparatively large, and is not easy to control frequency of training.Train and very fewly do not reach prediction and to require and should be effective, trainedly many easily cause overfitting, be only limitted to effectively predict training sample data, for non-sample data then DeGrain.
2, application publication number is that the one of CN102495937A is based on seasonal effect in time series Forecasting Methodology, this invention uses mean-valued function method to carry out continuation to original time series, by optimal subset regression method, choosing is deleted to continuation sequence, obtain optimal subset, carry out training and predicting in conjunction with BP neural network again, obtain higher the predicting the outcome of accuracy.
Shortcoming: need the recurrence subset of gauge index rank and therefrom delete to select optimal subset, calculated amount is comparatively large, and computation process is comparatively complicated, and the operation easier of practical application is larger.
3, the disclosed a kind of order forecast method based on CPFR of SanXia University's journal (natural science edition) is studied, the method that the document proposes time series and multiple regression analysis to combine carries out forecast analysis to enterprise order, and adopting real data to carry out analog simulation according to method for designing, simulation result shows that prediction effect is better than traditional single Time Series Forecasting Methods.
Shortcoming: forecasting accuracy is still not high, which index multiple regression chooses needs carefully consideration; And the method is used for manufacturing order forecasting, can applies in the prediction of e-commerce order and not yet be verified.
Find out from above-mentioned document, how to reduce calculated amount, raising prediction accuracy remains in prior art the difficult problem predicting order volume.
Summary of the invention
The technical matters that the present invention solves how to reduce calculated amount, raising prediction accuracy, provides a kind of calculated amount is few, prediction accuracy is high and workable order forecast method and system.
The present invention is solved the problems of the technologies described above by following technical proposals:
The invention provides a kind of order forecast method, be characterized in, comprising:
S
1, gather History Order amount;
S
2, by History Order amount daily granularity gather;
S
3, the date carried out classifying and the history date and forecast date that belong to same type matched;
S
4, from the history date matched with forecast date, choose a growth base date, and obtain order volume and the prediction rate of growth on described growth base date, described prediction rate of growth is determined according to the type of described forecast date;
S
5, the computational prediction date day order volume predicted value, order volume × (1+ predicts rate of growth) on order volume predicted value=growth base date day.
The technical program, by choosing the growth base date of coupling, makes not need too many data during the day order volume on computational prediction date, simplifies computation process, also assures that the accuracy of prediction simultaneously.
Preferably, described order forecast method also comprises:
T
1, History Order amount is gathered by a point granularity;
T
2, from the history date, choose multiple basic date matched with forecast date, and set each basis the date weight;
T
3, by the data normalization of order value per minute on each basis date, obtain the seasonal index number on each basis date respectively;
T
4, respectively to each basis the date seasonal index number do smoothing processing;
T
5, according to the weight of setting, successively through weighted mean and unitization after obtain the prediction seasonal index number per minute of forecast date;
T
6, the computational prediction date order volume predicted value per minute, the seasonal index number predicted value of the day order volume predicted value of order volume predicted value=forecast date per minute × per minute.
The technical program is calculating on day basis of order volume predicted value, has calculated order volume predicted value per minute further, and make the time granularity of prediction thinner, precision is higher.
Preferably, described order forecast method is also included in T
3with T
4between perform following steps:
By the standardized data on each basis date according to the residing time period with fixed time interval grouping, the data outside the scope being in all data mean value ± 2 × mean absolute deviation in group in often organizing are replaced with the non-negative random data in the scope of all data mean value ± 2 × mean absolute deviation in group.
The technical program can ensure the accuracy calculated further.
Preferably, T
3also comprise:
Inquiry anomalous event record sheet, judge whether anomalous event occurs in those basis dates, if occur, calculate the per minute order volume predicted value of basic date within the time period that anomalous event occurs that anomalous event occurs, and the per minute order volume of basic date within the time period that anomalous event occurs that anomalous event occurs is replaced with the per minute order volume predicted value of basic date within the time period that anomalous event occurs of the generation anomalous event calculated.
The technical program considers the inaccurate situation of possibility history of existence order volume in reality, so revise the data of this part, the process " calculating the per minute order volume predicted value of basic date within the time period that anomalous event occurs that anomalous event occurs " in the technical program can with reference to prediction order method of the present invention in calculate the process of order volume predicted value per minute.
Preferably, S
4also comprise: judge that whether the data of described growth base date whole day are complete, if complete, the day order volume predicted value on computational prediction date, if incomplete, revise the order volume on described growth base date or again chooses the growth base date.
The technical program can ensure that forecast date day order volume predicted value correct calculating, also improve the accuracy of calculating simultaneously.
The present invention also provides a kind of order forecasting system, is characterized in, comprises: an acquisition module, one first summarizing module, a sort module, one first choose module and one first computing module;
Described acquisition module is for gathering History Order amount;
Described first summarizing module be used for by History Order amount daily granularity gather;
Described sort module is used for the date being carried out classifying and being matched on the history date and forecast date that belong to same type;
Described first chooses module for choosing a growth base date from the history date matched with forecast date, and obtains order volume and the prediction rate of growth on described growth base date, and described prediction rate of growth is determined according to the type of described forecast date;
Described first computing module is used for the day order volume predicted value on computational prediction date, order volume × (1+ predicts rate of growth) on order volume predicted value=growth base date day.
Preferably, also comprise: one second summarizing module, one second chooses module, a standardized module, a smoothing module, prediction seasonal index number module and one second computing module;
Described second summarizing module is used for History Order amount to gather by a point granularity;
Described second chooses module for choosing multiple basic date matched with forecast date from the history date, and sets the weight on each basis date;
Described standardized module is used for, by the data normalization of the order value per minute on each basis date, obtaining the seasonal index number on each basis date respectively;
Described smoothing module is used for doing smoothing processing to the seasonal index number on each basis date respectively;
Described prediction seasonal index number module is used for the weight according to setting, successively through weighted mean and unitization after obtain the prediction seasonal index number per minute of forecast date;
Described second computing module is used for the order volume predicted value per minute on computational prediction date, the seasonal index number predicted value of the day order volume predicted value of order volume predicted value=forecast date per minute × per minute.
Preferably, described order forecasting system also comprises: a replacement module;
Described standardized module is also for calling described replacement module;
Described replacement module to be used for the standardized data on each basis date according to the residing time period with fixed time interval grouping, the data outside the scope being in all data mean value ± 2 × mean absolute deviation in group in often organizing is replaced with the non-negative random data in the scope of all data mean value ± 2 × mean absolute deviation in group.
Preferably, described standardized module is also for inquiring about anomalous event record sheet, judge whether anomalous event occurs in those basis dates, if occur, calculate the per minute order volume predicted value of basic date within the time period that anomalous event occurs that anomalous event occurs, and the per minute order volume of basic date within the time period that anomalous event occurs that anomalous event occurs is replaced with the per minute order volume predicted value of basic date within the time period that anomalous event occurs of the generation anomalous event calculated, then call described smoothing module.
Preferably, whether described first computing module is also complete for judging the data of described growth base date whole day, if complete, and the day order volume predicted value on computational prediction date, if incomplete, the order volume on described growth base date revised or again chooses the growth base date.
On the basis meeting this area general knowledge, above-mentioned each optimum condition, can combination in any, obtains the preferred embodiments of the invention.
Positive progressive effect of the present invention is: the order volume that the present invention can predict every day exactly and order volume per minute, has that calculated amount is little, process is simple, is easy to the advantage that realizes and accuracy is high.The present invention is simultaneously applicable to various electric business website, the especially order forecasting of online tourism website.
Accompanying drawing explanation
Fig. 1 is the process flow diagram of the order forecast method of the embodiment of the present invention.
Fig. 2 is that the hotel business predicted value of the 2014-11-20 of the embodiment of the present invention compares design sketch with actual value.
Fig. 3 is the system schematic of the order forecasting system of the embodiment of the present invention.
Embodiment
Mode below by embodiment further illustrates the present invention, but does not therefore limit the present invention among described scope of embodiments.
Embodiment
See Fig. 1, the order forecast method of the present embodiment, comprises the following steps:
Step 101, collection History Order amount.Gather order detailed data in the past few years.
Step 102, by History Order amount, daily granularity and a minute granularity gather.According to the order detailed data collected, gather the order volume of minute granularity and sky granularity.For a minute granularity, ensure 1440 data points every day, the zero padding process lacked.
Step 103, the date carried out classifying and the history date and forecast date that belong to same type are matched.According to date character, two large classes will be divided into the date: technical dates and common date.Comprise country's legal festivals and holidays such as New Year's Day, the Spring Festival, Clear and Bright, May Day, the Dragon Boat Festival, the mid-autumn, National Day and contiguous date of taking off technical dates, also comprise service order impact significantly other non-country's dates festivals or holidays, as state civil service examination's admission card for entrance examination prints the first day, the national postgraduate qualifying examination admission card for entrance examination printing first day.The common date comprises common working day (as common the week) and common off-day (common Saturday, Sunday).The history date belonging to same type and forecast date are matched and refers to the date of the date in the time at forecast date place with historical years is alignd, alignment thereof is legal festivals and holidays then and the alignment of congeniality legal festivals and holidays in former years, state civil service examination's admission card for entrance examination then prints the first day, the whole nation postgraduate qualifying examination admission card for entrance examination printing first day (is both working day with date in former years congeniality respectively, or be both off-day) state examine, grind and examine admission card for entrance examination printing first day alignment, weekend then and connatural alignment at weekend in former years, common working day then and common alignment on working day in former years, the common alignment on off-day in common off-day then and former years.In addition, should ensure all a few alignment, corresponding alignment Monday on Sunday of such as Monday and former years, Tuesday to Sunday in like manner as far as possible.The Monday that it should be noted that here is not Monday of stricti jurise, and working Tuesday of having a holiday or vacation Monday if run into, then can be counted as processing Tuesday Monday.
Step 104, from the history date matched with forecast date, choose a growth base date, and obtain order volume and the prediction rate of growth on described growth base date, described prediction rate of growth is determined according to the type of described forecast date.For the common date, select the last week phase on the same day as its growth base date (if forecast date is this Monday, then select the Monday of upper one week as the increase radix date), adopt the weighted mean of the week same period over the years year-on-year growth rate as prediction rate of growth; For technical dates, choose forecast date the previous day as the radix date, adopt the weighted mean of day same period over the years sequential growth rate as prediction rate of growth.
Step 105, from the history date, choose multiple basic date matched with forecast date, and set the weight on each basis date.Generally choose 4 to 8 orders and subscribe the date based on the trend comparatively close history date.Selective rule is as table 1:
Table 1
| The type of forecast date | The basic date matched with forecast date |
| Common weekend | Weekend on year-on-year basis, chain rate weekend |
| Common Monday | Monday on year-on-year basis |
| Common Tuesday is to Friday | Working day on year-on-year basis, chain rate working day |
| Take off | Chain rate working day |
| Not single | Chain rate off-day |
| Connect and stop | The festivals or holidays of the same period over the years |
| Particular job day | The working day same period over the years |
| The common day in October | Chain rate date and date in August |
As a rule, the weight on date arranges comparatively large on year-on-year basis, and the weight on chain rate date arranges less.In addition, for the date of character of the same race, more close to forecast date, weight arranges larger.Choose front four Mondays such as Monday as the basic date calculating seasonal index number predicted value, weight arranges according to by as far as being closely set to 1,2,3,4 respectively; Choose front four Tuesdays and front four chain rate dates Tuesday as the basic date calculating seasonal index number predicted value, the weight on date is according to by as far as being closely set to 2,4,6,8 respectively on year-on-year basis, and the weight on chain rate date is according to by as far as being closely set to 1,2,3,4 respectively.And, it is chain rate date and date in August with the basic date of mating common day in October in table 1, this is because usually the date can choose the relevant date of September on year-on-year basis the week on October on common date, but the date of particularly the middle ten days and the last ten days September is larger by affecting National Day, here they are weeded out, select the chain rate common date of relevant week date and the October on year-on-year basis of August, the basic date that the seasonal index number as October on the common date is predicted.
Step 106, carry out abnormality processing and obtain the seasonal index number per minute of forecast date.First inquire about anomalous event record sheet, to judge in those basis dates whether anomalous event (as equipment failure etc.), if occur, to the basic date of anomalous event be there is as forecast date, calculate the per minute order volume predicted value of basic date within the time period that anomalous event occurs that anomalous event occurs, and the per minute order volume of basic date within the time period that anomalous event occurs that anomalous event occurs is replaced with the per minute order volume predicted value of basic date within the time period that anomalous event occurs of the generation anomalous event calculated.Then by the data normalization of the order value per minute (period that anomalous event occurs replaces by predicted value) on each basis date, the seasonal index number (order accounting per minute) on each basis date is obtained respectively.The standardized data on each basis date was divided into groups with a fixed time interval according to the residing time period, data outside the scope being in all data mean value ± 2 × mean absolute deviation in group in often organizing are replaced with the non-negative random data in the scope of all data mean value ± 2 × mean absolute deviation in group, if namely group in data all data mean value ± 2 × mean absolute deviation in group scope outside, then in employing group all data mean value ± 2 × mean absolute deviation scope in non-negative random data replace.Then, smoothing processing is done to the seasonal index number on each basis date, according to the different weights on date, the unitization prediction seasonal index number per minute obtaining forecast date again after weighted mean.
Step 107, judge that whether the data of described growth base date whole day are complete, if complete, the day order volume predicted value on computational prediction date, if incomplete, the order volume on described growth base date is revised or again chooses the growth base date, the day order volume predicted value on computational prediction date again, order volume × (1+ predicts rate of growth) on order volume predicted value=growth base date day.For the situation that the data of growth base date whole day are incomplete, if the data of growth base date whole day be because event occurs affect that service order amount causes incomplete, then adopt the order volume after covering the loss as the modified value of order volume, using modified value as growth base; If the data of growth base date whole day be because do not gather good or because of other reasons cause incomplete, then selected growth base date again.
The order volume predicted value per minute on step 108, computational prediction date, the seasonal index number predicted value of the day order volume predicted value of order volume predicted value=forecast date per minute × per minute.
Be predicted as example below with OTA (online tourism society) hotel business, further illustrate and how to utilize the order forecast method of the present embodiment to shift to an earlier date 4 days prediction 2014-11-20 order volume per minute:
The first step, gathers history hotel reservation detailed data.In general, data acquisition has independently Collecting operation, helps to realize automatic data collection.Frequency acquisition can be gather once for one hour, incrementally gathers.So data in the past have all gathered substantially, only need to gather nearest data according to increasing to gather.
Second step, gather minute granularity order volume data and day order volume data.According to the data gathered according to minute granularity and day size-grade distribution gather, obtain order volume and order volume per minute every day.If initially gather minute data to be one day after less than 1440 points, then to the some zero padding process lacked.
3rd step, is undertaken classifying by the date and is matched on the history date and forecast date that belong to same type.By forecast date and for the previous period with alignment in former years, preferentially ensure during alignment that working day corresponds to working day, off-day corresponds to off-day.Ensure which day in one week is identical again, and make identical or close apart from all numbers in last red-letter day (being National Day) here.Here from getting before one week, 2014-11-13 to 2014-11-20 with former years alignment day after date as table 2:
Table 2
| Forecast date | date01 | date02 | date03 | date04 |
| 2014/11/20 | 2013/11/21 | 2012/11/22 | 2011/11/24 | 2010/11/18 |
| 2014/11/19 | 2013/11/20 | 2012/11/21 | 2011/11/23 | 2010/11/17 |
| 2014/11/18 | 2013/11/19 | 2012/11/20 | 2011/11/22 | 2010/11/16 |
| 2014/11/17 | 2013/11/18 | 2012/11/19 | 2011/11/21 | 2010/11/15 |
| 2014/11/16 | 2013/11/17 | 2012/11/18 | 2011/11/20 | 2010/11/14 |
| 2014/11/15 | 2013/11/16 | 2012/11/17 | 2011/11/19 | 2010/11/13 |
| 2014/11/14 | 2013/11/15 | 2012/11/16 | 2011/11/18 | 2010/11/12 |
| 2014/11/13 | 2013/11/14 | 2012/11/15 | 2011/11/17 | 2010/11/11 |
In upper table, date01, date02, date03, date04 represent respectively correspond to 1 year before, history date of matching before 2 years, before 3 years, before 4 years.
4th step, chooses a growth base date from the history date matched with forecast date, and obtains order volume and the prediction rate of growth on described growth base date.First judge the date type of 2014-11-20, this day, state neither examining the admission card for entrance examination printing first day, grind and examine the admission card for entrance examination printing first day, was a common working day neither the legal festivals and holidays, Thursday.So should choose one week previous crops is the growth base date.Here, 2014-11-13 day last Thursday is got as the growth base date.The simultaneously order volume (or revising order volume) on query history corresponding date, obtain hotel business week same period over the years year-on-year growth rate and power of composing as table 3:
Table 3
| Title | 2010 | 2011 | 2012 | 2013 |
| Rate of growth | -2.2％ | 0.7％ | 0.9％ | -1.9％ |
| Weight | 1 | 2 | 3 | 4 |
And then the prediction rate of growth obtaining 2014-11-20 hotel business day order volume relative is :-0.5%=(-2.2% × 1+0.7 × 2+0.9 × 3+ (-1.9%) × 4)/(1+2+3+4).
5th step, chooses multiple basic date matched with forecast date, and sets the weight on each basis date.2014-11-20 day is common Thursday, chooses the date based on date and four light chain rate dates on year-on-year basis in four day week.The Thursday of front surrounding is all normal working day, and the date of non-Monday at non-weekend gets nearest four days chain rates and also joins in the date of basis working day.Here, not getting Monday, is because analyze data to find that following Monday other several days workaday tendencies still to have at ordinary times comparatively significantly distinguishes.Then, be give weight to each basis date, the tendency similarity on date is higher on year-on-year basis usually, gives bigger weight.In addition, the basis date is more close from forecast date, and similarity is higher.So the date by being 2,4,6,8 as far as near weight, gives the chain rate date four days by being 1,2,3,4 as far as near weight on year-on-year basis to give for four weeks here.Hotel business in basic date of the correspondence of the seasonal index number predicted value of 2014-11-20 and weight in table 4:
Table 4
| Forecast date | The basis date | Weight | Type |
| 2014/11/20 | 2014/11/10 | 1 | Chain rate working day |
| 2014/11/20 | 2014/11/11 | 2 | Chain rate working day |
| 2014/11/20 | 2014/11/12 | 3 | Chain rate working day |
| 2014/11/20 | 2014/11/14 | 4 | Chain rate working day |
| 2014/11/20 | 2014/10/23 | 2 | Working day on year-on-year basis |
| 2014/11/20 | 2014/10/30 | 4 | Working day on year-on-year basis |
| 2014/11/20 | 2014/11/6 | 6 | Working day on year-on-year basis |
| 2014/11/20 | 2014/11/13 | 8 | Working day on year-on-year basis |
6th step, carries out abnormality processing and obtains the seasonal index number per minute of forecast date.By inquiry anomalous event record sheet, in the above-mentioned basis date, the part date is had to occur the influential event of hotel business.For 2014-11-13, the fault of the service order that finds to make a difference at about 19:50.
Inquire about one minute granularity data and event table, find this day fault to hotel's influence time scope as table 5:
Table 5
So, when getting one minute granularity order data of 2014-11-13, first the actual value of (during comprising impact and between convalescence) between age at failure is replaced to predicted value.If there are abnormal time section on other dates, also so process.Then, by the order data standardization of every day, the seasonal index number on each basis date is obtained.Then, to be half an hour one group, 00:00-00:29 is first group, and 00:30-00:59 is second group, by that analogy, and totally 48 groups.The data of the 21st group (10:00-10:29) are as shown in table 6:
Table 6
If group in data all data mean value ± 2 × mean absolute deviation in group scope outside, then in employing group all data mean value ± 2 × mean absolute deviation scope in non-negative random data replace.Then, 10 minutes average smoothing processing are done to the seasonal index number of every day, obtains the level and smooth seasonal index number of every day.Such as, for the partial data of 2014-11-13, as table 7:
Table 7
Finally, according to the different weights on date, the unitization seasonal index number per minute just obtaining forecast date again after weighted mean.
7th step, the day order volume predicted value on computational prediction date.Forecast date 2014-11-20 is common working day, Thursday, gets 2014-11-13 day last Thursday as the growth basis date, and owing to being prediction before four days, the detailed hotel order data of 2014-11-13 and combined data put in place all.Because there is the fault affecting hotel's order in 2014-11-13, so growth base is not a day order actual value, but a minute order value actual between age at failure is replaced to predicted value, still use actual minute order value to gather the day order volume modified value obtained during non-faulting.In the 4th step, obtain forecast date order volume relative to the rate of growth predicted value increasing basis date order volume, adopt following formula: forecast date order volume predicted value=growth base × (1+ rate of growth predicted value).Arrive this, just obtain the day order volume predicted value of forecast date.
8th step, the order volume predicted value per minute on computational prediction date.In the 6th step and the 7th step, obtain respectively forecast date seasonal index number per minute and day order volume predicted value.According to following formula, the seasonal index number of the day order volume predicted value of order volume predicted value=forecast date per minute × per minute, obtains the order volume predicted value per minute of forecast date.
Hotel business predicted value see the 2014-11-20 shown in Fig. 2 compares design sketch with actual value, and in figure, horizontal ordinate is the time, and ordinate is order volume, and black curve is actual value, and White curves is predicted value.Utilize order forecasting value per minute to compare with order actual value per minute, contrast effect is obvious, and order actual value per minute fluctuates up and down in order forecasting value per minute under normal circumstances.If there is website fault and affect service order, actual value is by tendency generation significant change, compare predicted value to have obviously and the decline continued, declining reaches certain threshold range and will recognize website by monitoring alarm and break down, and plays the effect finding website fault fast.This quick identification website fault has been applied to the result of calculation in the present invention, also uses other step and methods, as the setting of alarm regulation.
Said method step is uncomplicated, and calculated amount is not very large, workable, and the predicted value effect obtained is very good, is highly suitable for the order forecasting that e-commerce venture's service feature is clear and definite.The hotel business of having added up nearest one month subscribes order forecasting result, and accuracy, up to more than 98%, is much higher than other several existing methods.
The order forecasting system of the present embodiment, as shown in Figure 3, comprising: acquisition module 201,1 first summarizing module 202,1 second summarizing module 203, sort module 204,1 first is chosen module 205,1 second and chosen module 206, standardized module 207, replacement module 208, smoothing module 209, prediction seasonal index number module 210,1 first computing module 211 and one second computing module 212.
Described acquisition module 201 is for gathering History Order amount.
Described first summarizing module 202 for by History Order amount daily granularity gather.
Described second summarizing module 203 is for gathering History Order amount by a point granularity.
Described sort module 204 is for being undertaken classifying by the date and being matched on the history date and forecast date that belong to same type.
Described first chooses module 205 for choosing a growth base date from the history date matched with forecast date, and obtains order volume and the prediction rate of growth on described growth base date, and described prediction rate of growth is determined according to the type of described forecast date.
Described second chooses module 206 for choosing multiple basic date matched with forecast date from the history date, and sets the weight on each basis date, then calls described standardized module 207.
Described standardized module 207 is for inquiring about anomalous event record sheet, judge whether anomalous event occurs in those basis dates, if occur, to the basic date of anomalous event be there is as forecast date, calculate the per minute order volume predicted value of basic date within the time period that anomalous event occurs that anomalous event occurs, and the per minute order volume of basic date within the time period that anomalous event occurs that anomalous event occurs is replaced with the per minute order volume predicted value of basic date within the time period that anomalous event occurs of the generation anomalous event calculated, then by the data normalization of the order value per minute on each basis date, obtain the seasonal index number on each basis date respectively, then described replacement module 208 is called.
Described replacement module 208 is for dividing into groups the standardized data on each basis date with a fixed time interval according to the residing time period, data outside the scope being in all data mean value ± 2 × mean absolute deviation in group in often organizing are replaced with the non-negative random data in the scope of all data mean value ± 2 × mean absolute deviation in group, then call described smoothing module 209.
Described smoothing module 209, for doing smoothing processing to the seasonal index number on each basis date respectively, then calls described prediction seasonal index number module 210.
Described prediction seasonal index number module 210 for the weight according to setting, successively through weighted mean and unitization after obtain the prediction seasonal index number per minute of forecast date.
Whether described first computing module 211 is complete for judging the data of described growth base date whole day, if complete, the day order volume predicted value on computational prediction date, if incomplete, the order volume on described growth base date is revised or again chooses the growth base date, the day order volume predicted value on computational prediction date again, order volume × (1+ predicts rate of growth) on order volume predicted value=growth base date day.
Described second computing module 212 for the order volume predicted value per minute on computational prediction date, the seasonal index number predicted value of the day order volume predicted value of order volume predicted value=forecast date per minute × per minute.
Although the foregoing describe the specific embodiment of the present invention, it will be understood by those of skill in the art that these only illustrate, protection scope of the present invention is defined by the appended claims.Those skilled in the art, under the prerequisite not deviating from principle of the present invention and essence, can make various changes or modifications to these embodiments, but these change and amendment all falls into protection scope of the present invention.
Claims (10)
1. an order forecast method, is characterized in that, comprising:
S
1, gather History Order amount;
S
2, by History Order amount daily granularity gather;
S
3, the date carried out classifying and the history date and forecast date that belong to same type matched;
S
4, from the history date matched with forecast date, choose a growth base date, and obtain order volume and the prediction rate of growth on described growth base date, described prediction rate of growth is determined according to the type of described forecast date;
S
5, the computational prediction date day order volume predicted value, order volume × (1+ predicts rate of growth) on order volume predicted value=growth base date day.
2. order forecast method as claimed in claim 1, it is characterized in that, described order forecast method also comprises:
T
1, History Order amount is gathered by a point granularity;
T
2, from the history date, choose multiple basic date matched with forecast date, and set each basis the date weight;
T
3, by the data normalization of order value per minute on each basis date, obtain the seasonal index number on each basis date respectively;
T
4, respectively to each basis the date seasonal index number do smoothing processing;
T
5, according to the weight of setting, successively through weighted mean and unitization after obtain the prediction seasonal index number per minute of forecast date;
T
6, the computational prediction date order volume predicted value per minute, the seasonal index number predicted value of the day order volume predicted value of order volume predicted value=forecast date per minute × per minute.
3. order forecast method as claimed in claim 2, it is characterized in that, described order forecast method is also included in T
3with T
4between perform following steps:
By the standardized data on each basis date according to the residing time period with fixed time interval grouping, the data outside the scope being in all data mean value ± 2 × mean absolute deviation in group in often organizing are replaced with the non-negative random data in the scope of all data mean value ± 2 × mean absolute deviation in group.
4. order forecast method as claimed in claim 2, is characterized in that, T
3also comprise:
Inquiry anomalous event record sheet, judge whether anomalous event occurs in those basis dates, if occur, calculate the per minute order volume predicted value of basic date within the time period that anomalous event occurs that anomalous event occurs, and the per minute order volume of basic date within the time period that anomalous event occurs that anomalous event occurs is replaced with the per minute order volume predicted value of basic date within the time period that anomalous event occurs of the generation anomalous event calculated.
5. order forecast method as claimed in claim 1, is characterized in that, S
4also comprise: judge that whether the data of described growth base date whole day are complete, if complete, the day order volume predicted value on computational prediction date, if incomplete, revise the order volume on described growth base date or again chooses the growth base date.
6. an order forecasting system, is characterized in that, comprising: an acquisition module, one first summarizing module, a sort module, one first choose module and one first computing module;
Described acquisition module is for gathering History Order amount;
Described first summarizing module be used for by History Order amount daily granularity gather;
Described sort module is used for the date being carried out classifying and being matched on the history date and forecast date that belong to same type;
Described first chooses module for choosing a growth base date from the history date matched with forecast date, and obtains order volume and the prediction rate of growth on described growth base date, and described prediction rate of growth is determined according to the type of described forecast date;
Described first computing module is used for the day order volume predicted value on computational prediction date, order volume × (1+ predicts rate of growth) on order volume predicted value=growth base date day.
7. order forecasting system as claimed in claim 6, is characterized in that, also comprise: one second summarizing module, one second chooses module, a standardized module, a smoothing module, prediction seasonal index number module and one second computing module;
Described second summarizing module is used for History Order amount to gather by a point granularity;
Described second chooses module for choosing multiple basic date matched with forecast date from the history date, and sets the weight on each basis date;
Described standardized module is used for, by the data normalization of the order value per minute on each basis date, obtaining the seasonal index number on each basis date respectively;
Described smoothing module is used for doing smoothing processing to the seasonal index number on each basis date respectively;
Described prediction seasonal index number module is used for the weight according to setting, successively through weighted mean and unitization after obtain the prediction seasonal index number per minute of forecast date;
Described second computing module is used for the order volume predicted value per minute on computational prediction date, the seasonal index number predicted value of the day order volume predicted value of order volume predicted value=forecast date per minute × per minute.
8. order forecasting system as claimed in claim 7, it is characterized in that, described order forecasting system also comprises: a replacement module;
Described standardized module is also for calling described replacement module;
Described replacement module is used for the standardized data on each basis date to divide into groups with a fixed time interval according to the residing time period, data outside the scope being in all data mean value ± 2 × mean absolute deviation in group in often organizing are replaced with the non-negative random data in the scope of all data mean value ± 2 × mean absolute deviation in group, then call described smoothing module.
9. order forecasting system as claimed in claim 8, it is characterized in that, described standardized module is also for inquiring about anomalous event record sheet, judge whether anomalous event occurs in those basis dates, if occur, calculate the per minute order volume predicted value of basic date within the time period that anomalous event occurs that anomalous event occurs, and the per minute order volume of basic date within the time period that anomalous event occurs that anomalous event occurs is replaced with the per minute order volume predicted value of basic date within the time period that anomalous event occurs of the generation anomalous event calculated.
10. order forecasting system as claimed in claim 6, it is characterized in that, whether described first computing module is also complete for judging the data of described growth base date whole day, if complete, the day order volume predicted value on computational prediction date, if incomplete, the order volume on described growth base date revised or again chooses the growth base date.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510195209.2A CN104766144A (en) | 2015-04-22 | 2015-04-22 | Order forecasting method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510195209.2A CN104766144A (en) | 2015-04-22 | 2015-04-22 | Order forecasting method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN104766144A true CN104766144A (en) | 2015-07-08 |
Family
ID=53647958
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510195209.2A Pending CN104766144A (en) | 2015-04-22 | 2015-04-22 | Order forecasting method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104766144A (en) |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016123967A1 (en) * | 2015-02-03 | 2016-08-11 | 华为技术有限公司 | Data processing method and apparatus |
| CN107093096A (en) * | 2016-12-15 | 2017-08-25 | 口碑控股有限公司 | A kind of Traffic prediction method and device |
| CN109117991A (en) * | 2018-07-26 | 2019-01-01 | 北京京东金融科技控股有限公司 | One B shareB order transaction method and apparatus |
| US10169290B2 (en) | 2015-02-03 | 2019-01-01 | Huawei Technologies Co., Ltd. | Data processing method and apparatus |
| CN109284866A (en) * | 2018-09-06 | 2019-01-29 | 安吉汽车物流股份有限公司 | Goods orders prediction technique and device, storage medium, terminal |
| CN109874103A (en) * | 2019-04-23 | 2019-06-11 | 上海寰创通信科技股份有限公司 | A kind of wifi accurate positioning equipment and method |
| CN110363571A (en) * | 2019-06-24 | 2019-10-22 | 阿里巴巴集团控股有限公司 | The sensed in advance method and apparatus of trade user |
| CN110598940A (en) * | 2019-09-18 | 2019-12-20 | 深圳宇德金昌贸易有限公司 | Logistics order analysis and prediction system based on Internet of things trade |
| CN110837907A (en) * | 2018-08-17 | 2020-02-25 | 天津京东深拓机器人科技有限公司 | Method and device for predicting wave order quantity |
| CN111582530A (en) * | 2019-02-18 | 2020-08-25 | 北京京东尚科信息技术有限公司 | Method and device for predicting consumption of cloud product resources |
| CN111967940A (en) * | 2020-08-19 | 2020-11-20 | 支付宝(杭州)信息技术有限公司 | Order quantity abnormity detection method and device |
| CN112488377A (en) * | 2020-11-25 | 2021-03-12 | 上海中通吉网络技术有限公司 | Method and device for predicting daily order quantity of express delivery, storage medium and electronic equipment |
| CN112561562A (en) * | 2020-11-13 | 2021-03-26 | 广汽蔚来新能源汽车科技有限公司 | Order issuing method and device, computer equipment and storage medium |
| CN113344282A (en) * | 2021-06-23 | 2021-09-03 | 中国光大银行股份有限公司 | Method, system and computer readable medium for capacity data processing and allocation |
| CN113554400A (en) * | 2021-08-03 | 2021-10-26 | 杭州拼便宜网络科技有限公司 | Inventory data updating method, device, equipment and storage medium |
| US11216832B2 (en) | 2019-06-24 | 2022-01-04 | Advanced New Technologies Co., Ltd. | Predicting future user transactions |
| CN113988769A (en) * | 2021-12-28 | 2022-01-28 | 深圳前海移联科技有限公司 | Method and device for intelligently matching distributed resources and computer equipment |
| CN114819536A (en) * | 2022-04-02 | 2022-07-29 | 北京阿帕科蓝科技有限公司 | Vehicle dispatching method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101770690A (en) * | 2009-12-25 | 2010-07-07 | 东软集团股份有限公司 | Traffic condition predicting device and pathway exploration device |
| CN102111284A (en) * | 2009-12-28 | 2011-06-29 | 北京亿阳信通软件研究院有限公司 | Method and device for predicting telecom traffic |
| CN102110265A (en) * | 2009-12-23 | 2011-06-29 | 深圳市腾讯计算机系统有限公司 | Network advertisement effect estimating method and network advertisement effect estimating system |
| CN103310286A (en) * | 2013-06-25 | 2013-09-18 | 浙江大学 | Product order prediction method and device with time series characteristics |
| CN103617459A (en) * | 2013-12-06 | 2014-03-05 | 李敬泉 | Commodity demand information prediction method under multiple influence factors |
-
2015
- 2015-04-22 CN CN201510195209.2A patent/CN104766144A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102110265A (en) * | 2009-12-23 | 2011-06-29 | 深圳市腾讯计算机系统有限公司 | Network advertisement effect estimating method and network advertisement effect estimating system |
| CN101770690A (en) * | 2009-12-25 | 2010-07-07 | 东软集团股份有限公司 | Traffic condition predicting device and pathway exploration device |
| CN102111284A (en) * | 2009-12-28 | 2011-06-29 | 北京亿阳信通软件研究院有限公司 | Method and device for predicting telecom traffic |
| CN103310286A (en) * | 2013-06-25 | 2013-09-18 | 浙江大学 | Product order prediction method and device with time series characteristics |
| CN103617459A (en) * | 2013-12-06 | 2014-03-05 | 李敬泉 | Commodity demand information prediction method under multiple influence factors |
Non-Patent Citations (1)
| Title |
|---|
| 无名: "周海燕：Ctrip的容量分析模型", 《HTTP://DOWNLOAD.CSDN.NET/DOWNLOAD/ZHONG930/7419613》 * |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10169290B2 (en) | 2015-02-03 | 2019-01-01 | Huawei Technologies Co., Ltd. | Data processing method and apparatus |
| WO2016123967A1 (en) * | 2015-02-03 | 2016-08-11 | 华为技术有限公司 | Data processing method and apparatus |
| CN107093096A (en) * | 2016-12-15 | 2017-08-25 | 口碑控股有限公司 | A kind of Traffic prediction method and device |
| CN107093096B (en) * | 2016-12-15 | 2022-03-25 | 口碑(上海)信息技术有限公司 | Traffic prediction method and device |
| CN109117991A (en) * | 2018-07-26 | 2019-01-01 | 北京京东金融科技控股有限公司 | One B shareB order transaction method and apparatus |
| CN110837907A (en) * | 2018-08-17 | 2020-02-25 | 天津京东深拓机器人科技有限公司 | Method and device for predicting wave order quantity |
| CN109284866A (en) * | 2018-09-06 | 2019-01-29 | 安吉汽车物流股份有限公司 | Goods orders prediction technique and device, storage medium, terminal |
| CN109284866B (en) * | 2018-09-06 | 2021-01-29 | 安吉汽车物流股份有限公司 | Commodity order prediction method and device, storage medium and terminal |
| CN111582530A (en) * | 2019-02-18 | 2020-08-25 | 北京京东尚科信息技术有限公司 | Method and device for predicting consumption of cloud product resources |
| CN109874103A (en) * | 2019-04-23 | 2019-06-11 | 上海寰创通信科技股份有限公司 | A kind of wifi accurate positioning equipment and method |
| US11216832B2 (en) | 2019-06-24 | 2022-01-04 | Advanced New Technologies Co., Ltd. | Predicting future user transactions |
| CN110363571A (en) * | 2019-06-24 | 2019-10-22 | 阿里巴巴集团控股有限公司 | The sensed in advance method and apparatus of trade user |
| CN110598940A (en) * | 2019-09-18 | 2019-12-20 | 深圳宇德金昌贸易有限公司 | Logistics order analysis and prediction system based on Internet of things trade |
| CN111967940A (en) * | 2020-08-19 | 2020-11-20 | 支付宝(杭州)信息技术有限公司 | Order quantity abnormity detection method and device |
| CN111967940B (en) * | 2020-08-19 | 2023-02-21 | 支付宝(杭州)信息技术有限公司 | Order quantity abnormity detection method and device |
| CN112561562A (en) * | 2020-11-13 | 2021-03-26 | 广汽蔚来新能源汽车科技有限公司 | Order issuing method and device, computer equipment and storage medium |
| CN112488377A (en) * | 2020-11-25 | 2021-03-12 | 上海中通吉网络技术有限公司 | Method and device for predicting daily order quantity of express delivery, storage medium and electronic equipment |
| CN113344282A (en) * | 2021-06-23 | 2021-09-03 | 中国光大银行股份有限公司 | Method, system and computer readable medium for capacity data processing and allocation |
| CN113554400A (en) * | 2021-08-03 | 2021-10-26 | 杭州拼便宜网络科技有限公司 | Inventory data updating method, device, equipment and storage medium |
| CN113988769A (en) * | 2021-12-28 | 2022-01-28 | 深圳前海移联科技有限公司 | Method and device for intelligently matching distributed resources and computer equipment |
| CN113988769B (en) * | 2021-12-28 | 2022-03-22 | 深圳前海移联科技有限公司 | Method and device for intelligently matching distributed resources and computer equipment |
| CN114819536A (en) * | 2022-04-02 | 2022-07-29 | 北京阿帕科蓝科技有限公司 | Vehicle dispatching method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104766144A (en) | Order forecasting method and system | |
| CN110097297B (en) | Multi-dimensional electricity stealing situation intelligent sensing method, system, equipment and medium | |
| CN110400215B (en) | Method and system for constructing enterprise family-oriented small micro enterprise credit assessment model | |
| CN106126391A (en) | System monitoring method and apparatus | |
| EP3038025A1 (en) | Retention risk determiner | |
| CN111476435B (en) | Charging pile load prediction method based on density peak value | |
| CN107608862A (en) | Monitoring alarm method, monitoring alarm device and computer-readable recording medium | |
| CN110837963A (en) | Risk control platform construction method based on data, model and strategy | |
| CN112766550A (en) | Power failure sensitive user prediction method and system based on random forest, storage medium and computer equipment | |
| CN109615280A (en) | Employee's data processing method, device, computer equipment and storage medium | |
| CN110347724A (en) | Abnormal behaviour recognition methods, device, electronic equipment and medium | |
| CN111510368B (en) | Family group identification method, device, equipment and computer readable storage medium | |
| CN115375205B (en) | Method, device and equipment for determining water user portrait | |
| Umayaparvathi et al. | Attribute selection and customer churn prediction in telecom industry | |
| CN109508807A (en) | Lottery user liveness prediction technique, system and terminal device, storage medium | |
| CN109711424A (en) | A kind of rule of conduct acquisition methods, device and equipment based on decision tree | |
| CN111738819A (en) | Method, device and equipment for screening characterization data | |
| CN108897821B (en) | Method and device for automatically generating data conclusion | |
| CN110764902A (en) | Virtual resource allocation method and device based on AI (Artificial Intelligence), computer equipment and storage medium | |
| CN113609193A (en) | Method and device for training prediction model for predicting customer transaction behavior | |
| CN116034379A (en) | Activity level measurement using deep learning and machine learning | |
| CN115511525A (en) | Member loss early warning method and system for B2B platform user | |
| CN115358481A (en) | Early warning and identification method, system and device for enterprise ex-situ migration | |
| CN108304990B (en) | Power failure sensitivity pre-judging method and system | |
| CN116205355B (en) | Power load prediction method, device and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| EXSB | Decision made by sipo to initiate substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C41 | Transfer of patent application or patent right or utility model | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20160205 Address after: 200335 Shanghai city Changning District Admiralty Road No. 968 Building No. 16 10 floor Applicant after: SHANGHAI XIECHENG BUSINESS CO., LTD. Address before: 200335 Shanghai City, Changning District Fuquan Road No. 99, Ctrip network technology building Applicant before: Ctrip computer technology (Shanghai) Co., Ltd. |
|
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20150708 |
|
| RJ01 | Rejection of invention patent application after publication |