CN106127329A - Order forecast method and device - Google Patents
Order forecast method and device Download PDFInfo
- Publication number
- CN106127329A CN106127329A CN201610431755.6A CN201610431755A CN106127329A CN 106127329 A CN106127329 A CN 106127329A CN 201610431755 A CN201610431755 A CN 201610431755A CN 106127329 A CN106127329 A CN 106127329A
- Authority
- CN
- China
- Prior art keywords
- order
- moment
- target area
- variable quantity
- volume
- 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
- 238000000034 method Methods 0.000 title claims abstract description 94
- 238000004422 calculation algorithm Methods 0.000 claims description 73
- 238000013528 artificial neural network Methods 0.000 claims description 42
- 230000002123 temporal effect Effects 0.000 claims description 10
- 238000005516 engineering process Methods 0.000 description 7
- 230000006870 function Effects 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 230000007423 decrease Effects 0.000 description 3
- 238000012804 iterative process Methods 0.000 description 3
- 241000208340 Araliaceae Species 0.000 description 2
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 2
- 235000003140 Panax quinquefolius Nutrition 0.000 description 2
- 230000003247 decreasing effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 235000008434 ginseng Nutrition 0.000 description 2
- 241000287196 Asthenes Species 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000010009 beating Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0607—Regulated
-
- G06Q50/40—
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Finance (AREA)
- Accounting & Taxation (AREA)
- Marketing (AREA)
- Theoretical Computer Science (AREA)
- Human Resources & Organizations (AREA)
- General Business, Economics & Management (AREA)
- Development Economics (AREA)
- Tourism & Hospitality (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Traffic Control Systems (AREA)
Abstract
The present invention provides a kind of order forecast method and device.The method, including: obtain target area at first time period [ti‑1, ti] the first order variable quantity and variation tendency；According to the first History Order data of described first order variable quantity, described variation tendency and described target area, obtain identical with described variation tendency in described first History Order data and meet, with the difference of described first order variable quantity, the second time period that the order variable quantity of preset threshold range is corresponding；Obtain the second order variable quantity of the subsequent time period adjacent with described second time period；According to described second order variable quantity and described target area at described tiThe actual order volume in moment, determines that described target area is at ti+1The first prediction order volume in moment.The method can go out target area order volume at any time with Accurate Prediction, and then achieves the rational management to vehicle, effectively alleviates traffic pressure.
Description
Technical field
The present invention relates to computer technology, particularly relate to a kind of order forecast method and device.
Background technology
Taxi supplements as the one of city bus, provides conveniently to civic trip.But one common
Phenomenon is exactly, some taximan purposelessly limit driving limit find passenger, some passengers waited for a long time or beat less than
Car.In order to solve the problem of " difficulty of calling a taxi ", a kind of software of calling a taxi based on smart mobile phone arises at the historic moment, such as " tick and call a taxi ",
" fast calls a taxi " etc..Passenger can realize real-time car or reservation car by software of calling a taxi, and car owner can also be according to beating simultaneously
The historical data in some region preserved in car software, predicts the passenger demand in following this region sometime, and according to
Passenger demand carries out the scheduling of vehicle, it is possible to reduce unloaded mileage, alleviates and services unbalanced phenomenon.
Existing target area vehicle dispatching method is time-based autoregression method, particularly as follows: first, according to M days
The by bus quantity on order of the jth period in interior R region, in M days+1 period of jth in R region ride quantity on order and
The by bus quantity on order of the jth period in R the region on the same day, determine ith zone the jth period regression coefficient to
Amount, wherein i ∈ R；Then, according to the regression coefficient vector of jth period of above-mentioned ith zone and R the region on the same day
The by bus quantity on order of jth period, determines that the prediction of+1 period of jth of the ith zone on the same day is ridden quantity on order；So
After, the jth+1 in R the region on the same day is determined according to the prediction quantity on order by bus of+1 period of jth of the ith zone on the same day
The prediction of individual period is ridden quantity on order.
But, the result that the order forecast method of prior art is predicted is inaccurate, it is impossible to effectively realize the conjunction of vehicle
Reason scheduling.
Summary of the invention
The present invention provides a kind of order forecast method and device, is used for solving existing calculating and asks order forecasting is inaccurate
Topic, it is achieved that the Accurate Prediction to target area any time order, it is ensured that the rational management of vehicle.
First aspect, the present invention provides a kind of order forecast method, including:
Obtain target area at first time period [ti-1, ti] the first order variable quantity and variation tendency；Wherein, described
First order variable quantity is equal to described target area at described tiThe actual order volume in moment and described ti-1The reality in moment
The absolute value of the difference of order volume；
According to the first History Order data of described first order variable quantity, described variation tendency and described target area,
Obtain identical with described variation tendency in described first History Order data and with described first order variable quantity difference to meet
The second time period that the order variable quantity of preset threshold range is corresponding；
Obtain the second order variable quantity of the subsequent time period adjacent with described second time period；
According to described second order variable quantity and described target area at described tiThe actual order volume in moment, determines institute
State target area at ti+1The first prediction order volume in moment.
Alternatively, if getting multiple described second time period, under the most described acquisition is adjacent with described second time period
The second order variable quantity of one time period, specifically includes:
Obtain the 3rd order variable quantity that the subsequent time period adjacent with each second time period is corresponding；
Obtain the meansigma methods of multiple 3rd order variable quantity, and described meansigma methods is defined as described second order change
Amount.
Further, said method also includes:
Information point POI number in each region in the default territorial scope of acquisition；Described default territorial scope includes described
Target area and at least one neighboring area；
Each region is obtained at described t from the second History Order data that described default territorial scope is correspondingiMoment
Actual order volume and at described tiThe quantity of the vehicle in moment；
According to the information point POI number in described each region, each region at described tiThe actual order volume in moment
With each region at described tiThe quantity of the vehicle in moment, obtains artificial neural network algorithm model；
According to input parameter and described artificial neural network algorithm model, determine that described target area is at described ti+1Time
The the second prediction order volume carved；Wherein, described input parameter includes area information and temporal information；
According to described target area at described ti+1The first prediction order volume in moment and described target area are described the
ti+1The second prediction order volume in moment, determines that described target area is at described ti+1The 3rd prediction order volume in moment.
Alternatively, according to described target area at described ti+1First prediction order volume and the described target area in moment
At described ti+1The second prediction order volume in moment, determines that described target area is at described ti+1The 3rd prediction order in moment
Amount, specifically includes:
According to described second History Order data, described target area at described tiMoment first prediction order volume and
Described target area is at described tiThe second prediction order volume in moment, obtains gradient boosted tree algorithm model；
According to described input parameter and described gradient boosted tree algorithm model, obtain described target area at described ti+1
The 3rd prediction order volume in moment.
Second aspect, present invention additionally comprises a kind of order forecasting device, including:
First acquisition module, is used for obtaining target area at first time period [ti-1, ti] the first order variable quantity and change
Change trend；Wherein, described first order variable quantity is equal to described target area at described tiThe actual order volume in moment and institute
State ti-1The absolute value of the difference of the actual order volume in moment；
Second acquisition module, for according to described first order variable quantity, described variation tendency and described target area
First History Order data, obtain identical with described variation tendency in described first History Order data and order with described first
The difference of altered amount meets the second time period that the order variable quantity of preset threshold range is corresponding；
3rd acquisition module, for obtaining the second order change of the subsequent time period adjacent with described second time period
Amount；
First processing module, for according to described second order variable quantity and described target area at described tiMoment
Actual order volume, determines that described target area is at ti+1The first prediction order volume in moment.
Alternatively, if described second acquisition module gets multiple described second time period, described 3rd acquisition module
Specifically include:
First acquiring unit, becomes for the 3rd order obtaining the subsequent time period adjacent with each second time period corresponding
Change amount；
Processing unit, for obtaining the meansigma methods of multiple 3rd order variable quantity, and is defined as described by described meansigma methods
Second order variable quantity.
Further, said apparatus also includes:
4th acquisition module, the information point POI number in each region in obtaining default territorial scope；Described default
Territorial scope includes described target area and at least one neighboring area；
5th acquisition module, for obtaining each district from the second History Order data that described default territorial scope is corresponding
Territory is at described tiThe actual order volume in moment and at described tiThe quantity of the vehicle in moment；
6th acquisition module, is used for according to the information point POI number in described each region, each region at described ti
The actual order volume in moment and each region are at described tiThe quantity of the vehicle in moment, obtains artificial neural network algorithm mould
Type；
Second processing module, for according to input parameter and described artificial neural network algorithm model, determining described target
Region is at described ti+1The second prediction order volume in moment；Wherein, described input parameter includes area information and temporal information；
3rd processing module, is used for according to described target area at described ti+1First prediction order volume and the institute in moment
State target area at described ti+1The second prediction order volume in moment, determines that described target area is at described ti+1The of moment
Three prediction order volume.
Alternatively, above-mentioned 3rd processing module specifically includes:
Second acquisition unit, is used for according to described second History Order data, described target area at described tiMoment
First prediction order volume and described target area at described tiThe second prediction order volume in moment, obtains gradient boosted tree and calculates
Method model；
3rd acquiring unit, for according to described input parameter and described gradient boosted tree algorithm model, obtains described mesh
Mark region is at described ti+1The 3rd prediction order volume in moment.
The order forecast method of present invention offer and device, by obtaining target area at first time period [ti-1, ti]
First order variable quantity and variation tendency, and according to this first order variable quantity, variation tendency and the first history of target area
Order data, obtains identical with the variation tendency of the first order variable quantity in the first History Order data and becomes with the first order
The difference of change amount meets the second time period that the order variable quantity of preset threshold range is corresponding, obtains adjacent with the second time period
Second order variable quantity of subsequent time period；According to this second order variable quantity and above-mentioned target area at tiThe reality in moment
Order volume, determines that target area is at ti+1The first prediction order volume in moment.The present invention chooses close with the first order variable quantity
Order variable quantity, the order variable quantity close due to this and the first order variable quantity are essentially identical, then this close order becomes
The second of the adjacent subsequent time period that the order variable quantity of the adjacent subsequent time period that change amount is corresponding is corresponding with first time period is ordered
Altered amount is essentially identical, and i.e. the present embodiment is by obtaining accurate second order variable quantity, so improve according to this
The accuracy of the first prediction order that two order variable quantities obtain, thus improve the reasonability to vehicle scheduling, effectively alleviate
Traffic pressure.
Accompanying drawing explanation
In order to be illustrated more clearly that the present invention or technical scheme of the prior art, below will be to embodiment or prior art
In description, the required accompanying drawing used is briefly described, it should be apparent that, the accompanying drawing in describing below is the one of the present invention
A little embodiments, for those of ordinary skill in the art, on the premise of not paying creative work, it is also possible to according to this
A little accompanying drawings obtain other accompanying drawing.
The schematic flow sheet of the order forecast method embodiment one that Fig. 1 provides for the present invention；
The schematic flow sheet of the order forecast method embodiment two that Fig. 2 provides for the present invention；
The schematic flow sheet of the order forecast method embodiment three that Fig. 3 provides for the present invention；
Artificial neural network algorithm model structure signal in the order forecast method embodiment three that Fig. 4 provides for the present invention
Figure；
The schematic flow sheet of the order forecast method embodiment four that Fig. 5 provides for the present invention；
Gradient boosted tree algorithm model structural representation in the order forecast method embodiment four that Fig. 6 provides for the present invention；
The structural representation of the order forecasting device embodiment one that Fig. 7 provides for the present invention；
The structural representation of the order forecasting device embodiment two that Fig. 8 provides for the present invention；
The structural representation of the order forecasting device embodiment three that Fig. 9 provides for the present invention；
The structural representation of the order forecasting device embodiment four that Figure 10 provides for the present invention.
Detailed description of the invention
For making the object, technical solutions and advantages of the present invention clearer, attached below in conjunction with in the embodiment of the present invention
Figure, is clearly and completely described the technical scheme in the embodiment of the present invention, it is clear that described embodiment is the present invention
A part of embodiment rather than whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art are not having
Make the every other embodiment obtained under creative work premise, broadly fall into the scope of protection of the invention.
Technical scheme can apply in vehicle dispatch system, is used for solving order forecasting side in prior art
The problem that the result that method is predicted is inaccurate, cannot effectively realize the rational management of vehicle.The order forecast method of the present invention
Achieve the Accurate Prediction to target area any time order volume with device, it is achieved that the rational management of vehicle, and then reduce
The dead mileage of vehicle, alleviates the unbalanced phenomenon of service.
With specifically embodiment, technical scheme is described in detail below.These concrete enforcements below
Example can be combined with each other, and may repeat no more in certain embodiments for same or analogous concept or process.
The schematic flow sheet of the order forecast method embodiment one that Fig. 1 provides for the present invention.The present embodiment refer to as
What obtains target area at time ti+1The detailed process of the first prediction order volume in moment.As it is shown in figure 1, the method includes:
S101, acquisition target area are at first time period [ti-1, ti] the first order variable quantity and variation tendency；Wherein,
Described first order variable quantity is equal to described target area at described tiThe actual order volume in moment and described ti-1Moment
The absolute value of the difference of actual order volume.
It should be noted that vehicle of the present invention can be provided with software of calling a taxi, such as, " tick and call a taxi ", " fast
Call a taxi " etc., vehicle dispatch system can obtain the order data of passenger according to above-mentioned software of calling a taxi, permissible according to this order data
Obtain the drop-off pick-up points of the time of getting on or off the bus of passenger, the most each order data carries area information and temporal information, will
Passenger is t in target areai+1The History Order data produced before moment as the first History Order data, and by this
One History Order data are saved in the data base of vehicle dispatch system.Wherein vehicle dispatch system can enter with each vehicle
Row information is mutual.
Alternatively, vehicle dispatch system can be by analyzing the GPS information of vehicle, and in extraction driver's experience, passenger demand is relatively
Many regions, using the more region of passenger demand as target area, for predicting the order volume in the more region of passenger demand.
Specifically, vehicle dispatch system obtains target area at moment t from the first History Order datai-1Actual order
Single amount isAt moment tiActual order volume beThen according to formula
Obtain first time period [ti-1, ti] the first order variable quantitySimultaneously according to moment ti-1Order volumeAnd time
Carve tiOrder volumeObtain at first time period [ti-1, ti] the variation tendency of the first order, i.e. whenTime, the variation tendency of the first order is incremental, whenTime, the change of the first order becomes
Gesture is for successively decreasing.
Alternatively, the present invention can be divided into 24 periods by one day, and each period is one hour, i.e. assumes moment tiFor
The morning 9 point, then moment ti-1For 8 a.m., then first time period [ti-1, ti] it is time period [8,9].Alternatively, this enforcement is also
Can be divided into the time period of other same intervals by one day, wherein the time period the least predicts the outcome the most accurate, and the time period is more
Big calculating speed is the fastest, and specifically choosing of time period length sets according to actual needs, and the present embodiment is without limitation.
S102, according to described first order variable quantity, described variation tendency and the first History Order of described target area
Data, obtain identical with described variation tendency in described first History Order data and with described first order variable quantity difference
Value meets the second time period that the order variable quantity of preset threshold range is corresponding.
S103, obtain the second order variable quantity of the subsequent time period adjacent with described second time period.
Specifically, vehicle dispatch system obtains the first History Order data of target area from data base, wherein this
One History Order data include the actual order volume in each moment, the change of the order of each period (the most adjacent two moment)
Amount, and the variation tendency of each period order variable quantity, such as, be divided into 24 periods, the wherein order volume of 8 by one day
Be 100, the order volume of 9 be 110, the order volume of 10 is 90, then the order variable quantity of 8 o'clock to 9 o'clock these time periods is |
110-100 |=10, its variation tendency is incremental, and the order variable quantity of 9 o'clock to 10 o'clock these time periods is | 90-110 |=20,
Its variation tendency is for successively decreasing.Choose from the first History Order data and the first order variable quantity in above-mentioned steps S101Variation tendency identical (assume the first order variable quantityVariation tendency be incremental) and with first order become
Change amountDifference meet preset threshold range order variable quantity (for example, it is assumed that preset threshold range is for [-1-1], the
One order variable quantityBeing 10, the order variable quantity of time period A is 11, then by the order variable quantity of above-mentioned time period A
11 and first order variable quantity 10 poor, it is thus achieved that difference be 1, this difference 1 falls in predetermined threshold value interval [-1-1], then
Order variable quantity and the first order variable quantity of time period A are describedDifference meet preset threshold range [-1-1], then will
The order variable quantity of period A is referred to as the first order variable quantityClose History Order variable quantity), obtain this order simultaneously
The second time period that variable quantity is corresponding.Then from the first History Order data, obtain next adjacent with above-mentioned second time period
The second order variable quantity of time period, is designated as slopei+1。
Needing explanation, above-mentioned second time period can be a time period, and accordingly, the second order variable quantity can be
One order variable quantity；Alternatively, the second time period can also be multiple time period, and accordingly, the second order variable quantity is permissible
It is the meansigma methods of multiple order variable quantity, sets specifically according to practical situation.
S104, according to described second order variable quantity and described target area at described tiThe actual order volume in moment,
Determine that described target area is at described ti+1The first prediction order volume in moment.
Specifically, it is calculated moment t according to above-mentioned steps 101iActual order volume beAnd according to above-mentioned step
Rapid 102 and step 103 calculating acquisition the second order variable quantity slopei+1, and according to formula
Prediction target area is at ti+1The first prediction order volume in momentThe order forecast method that the present invention provides, passes through
Obtain target area at first time period [ti-1, ti] the first order variable quantity and variation tendency, and according to this first order become
First History Order data of change amount, variation tendency and target area, obtain in the first History Order data and become with the first order
The variation tendency of change amount is identical and meet the order variable quantity of preset threshold range with the difference of the first order variable quantity corresponding
Second time period, obtain the second order variable quantity of the subsequent time period adjacent with the second time period；Become according to this second order
Change amount and above-mentioned target area are at tiThe actual order volume in moment, determines that target area is at ti+1First prediction in moment is ordered
Dan Liang.The present embodiment chooses the order variable quantity close with the first order variable quantity, the order variable quantity close due to this and
One order variable quantity is essentially identical, then the order variable quantity and of the adjacent subsequent time period that this close order variable quantity is corresponding
Second order variable quantity of the adjacent subsequent time period that one time period is corresponding is essentially identical, i.e. the present embodiment is more accurate by obtaining
The second order variable quantity, and then improve according to this second order variable quantity obtain first prediction order accuracy, from
And improve the reasonability to vehicle scheduling, effectively alleviate traffic pressure.
The schematic flow sheet of the order forecast method embodiment two that Fig. 2 provides for the present invention.The present embodiment refers to work as
When second time period was multiple, obtain the detailed process of the second order variable quantity.On the basis of above-described embodiment, further
Ground, above-mentioned S103 specifically may include that
S201: obtain the 3rd order variable quantity that the subsequent time period adjacent with each second time period is corresponding.
S202: obtain the meansigma methods of multiple 3rd order variable quantity, and described meansigma methods is defined as described second order
Variable quantity.
Specifically, when according to the method for above-mentioned S102 obtain the second time period be multiple time, it is assumed that for th1,th2…thk,
Target area is obtained at th from the first History Order data1,th2…thkSubsequent time period th1+1,th2+1…thk+1Corresponding
The 3rd order variable quantity, be designated asThen above-mentioned multiple 3rd order variable quantity is calculatedMeansigma methods, be designated as slopetavg, by this meansigma methods slopetavgBecome as the second order
Change amount.Finally, according to formulaBy tiThe order volume in momentBecome with above-mentioned second order
Change amount slopetavgIt is added, it is thus achieved that ti+1The first prediction order volume in momentOwing to the present embodiment is obtained by calculating
Multiple 3rd order variable quantities, by the meansigma methods of above-mentioned multiple 3rd order variable quantities as the second order variable quantity so that right
The prediction of this second order variable quantity is more accurate, and then improves the Accurate Prediction to the first prediction order volume.
The order forecast method that the present invention provides, when the second time period obtained is multiple, by obtaining and each the
The 3rd order variable quantity that two time periods adjacent subsequent time period is corresponding, obtains the meansigma methods of multiple 3rd order variable quantity,
And described meansigma methods is defined as the second order variable quantity, it is achieved that the accurate calculating to the second order variable quantity so that based on
First prediction order of this second order variable quantity is more accurate, and then achieves the rational management to vehicle, effectively alleviates
The pressure of traffic.
The schematic flow sheet of the order forecast method embodiment three that Fig. 3 provides for the present invention, Fig. 4 is ordering that the present invention provides
Artificial neural network algorithm model structure schematic diagram in single Forecasting Methodology embodiment three, the present embodiment refers to according to artificial god
The detailed process of the second prediction order volume is obtained through network algorithm model.On the basis of above-described embodiment, the order of the present invention
Forecasting Methodology also includes:
Information point POI number in each region in S301, the default territorial scope of acquisition；Described default territorial scope includes
Described target area and at least one neighboring area.
It should be noted that the present embodiment is preset territorial scope can be centered by target area, with certain length
Degree (such as 500m) is radius, it is thus achieved that a border circular areas, is set as this border circular areas presetting territorial scope.Alternatively, also may be used
Be the rectangular area that includes target area is set as preset territorial scope, the present embodiment tool to default territorial scope
Body system of selection does not limits, and chooses according to practical situation.Then above-mentioned default territorial scope is divided, it is thus achieved that many
Individual region, includes at least one neighboring area of target area and target area periphery in the plurality of region.It is alternatively possible to
In units of the size of target area, above-mentioned default territorial scope is divided into multiple regions of formed objects；Can also is that root
According to the administrative information region on map, above-mentioned default territorial scope is divided into multiple regions that size shape differs, this enforcement
The quantity of the zoned format of default territorial scope and the multiple regions after dividing is not limited by example.Wherein preset territorial scope
Division unit is the least, it is thus achieved that the quantity in region the most, finally the prediction to the order volume in target area moment is the most accurate；In advance
If the unit that territorial scope divides is the biggest, it is thus achieved that the quantity in region the fewest, the calculating speed of whole prediction process is fast.
Specifically, the information point in each region in vehicle dispatch system obtains default territorial scope from map
(Point of Interest is called for short POI) number.The type of this POI can be such as: shopping centre, bank, school, residential quarter
Deng, the number of each type POI is the number that each type POI is corresponding in this region, such as: the business included in region a
Industry district is 4, and bank is 5, and school is 2, and residential quarter is 10 etc..In the present embodiment, the information point data from map can
To reflect the type in region, these passenger demands corresponding to different types of region are the most different, and such as residential quarter is on and off duty
Time the volume of the flow of passengers bigger.
S302, from the second History Order data that described default territorial scope is corresponding, obtain each region at described ti
The actual order volume in moment and at described tiThe quantity of the vehicle in moment.
Specifically, vehicle dispatch system by each region in default territorial scope at ti+1History before moment is real
Border order data is as the second History Order data, and these the second History Order data can include above-mentioned first History Order number
According to, and these the second History Order data can obtain from the software of calling a taxi of vehicle.Vehicle dispatch system is ordered from this second history
Forms data obtains each region at tiThe actual order volume in moment and at tiThe quantity of the vehicle in moment.The most above-mentioned reality
Border order data contains and much records relevant attribute to passenger by bus, by statistical analysis the second History Order data,
Important time and the spatial informations such as passenger getting on/off time, drop-off pick-up points can be extracted, simultaneously according to analyzing the second history
Order data can obtain each region vehicle fleet size at any time.
Alternatively, vehicle dispatch system can also obtain the position that this vehicle is residing at any time from vehicle GPS information
Put, and then each region can be obtained at tiThe quantity of the vehicle in moment.Wherein the GPS information of vehicle contains vehicle gathering
Situation, reflects the experience of driver to a certain extent, be such as on duty height by stages time, driver can assemble toward residential quarter, with convenient
Resident gets to work by car, next height by stages time, driver may assemble toward shopping centre, to facilitate working clan to come off duty by bus.
S303, according to the information point POI number in described each region, each region at described tiThe actual of moment is ordered
Single amount and each region are at described tiThe quantity of the vehicle in moment, obtains artificial neural network algorithm model.
It should be noted that as shown in Figure 4, artificial neural network algorithm has study and adaptive ability, Ke Yitong
Cross the input-output data of a collection of mutual correspondence being provided previously by, analyze the potential rule grasped between input-output data,
Finally according to above-mentioned rule, extrapolate the output result of correspondence by new input data.From the foregoing, use this ANN
Before the order volume of network algorithm model prediction any time target area, need with substantial amounts of input-ouput data that this is artificial
Neural network algorithm model is trained, and the present embodiment is with target area and information point POI of the neighboring area of target area
The quantity of number, actual order volume and vehicle, as input, is ordered with the actual of neighboring area of target area and target area
Artificial neural network algorithm model is trained by single amount as output, as shown in Figure 4, and this artificial neural network algorithm model
Including three layers, respectively input layer, hidden layer and output layer.
Before introducing model training process, optionally, the present embodiment can be to above-mentioned parameter: the information in each region
Point POI number, each region are at described tiThe actual order volume in moment and each region are at described tiThe number of the vehicle in moment
Amount, according to formulaCarry out equidimensional process.Concrete model process can include below step:
S303a, the nodes of the setting each layer of input (i.e. input layer, hidden layer and output layer), input layer and hidden layer
Between initial weight ω1With initial threshold B1, initial weight ω between hidden layer and output layer2With initial threshold B2, training
During expect the error amount e that reaches, maximum cycle N of training, set up artificial neural network algorithm model.
Wherein, the node number of input layer is determined by inputting parameter, and as shown in Figure 4, the present embodiment is by default territorial scope
Being divided into n region, the input parameter (i.e. X value) corresponding in each region has: the information point POI number in this region is (such as
Region A has the different types of POI number such as 3 markets, 4 schools, 7 residential quarters), this region is at tiThe reality in moment
Order volume, and this region is at tiThe vehicle fleet size in moment；Output (i.e. Y value) corresponding to each region is that each region exists
TiThe actual order volume in moment, say, that each region is at tiThe actual order volume in moment is that input quantity exports again
Amount.The acquisition methods of the input information in above-mentioned each region, with reference to foregoing description, does not repeats them here.Correspondence is as shown in Figure 4
Total input number of parameters of artificial neural network algorithm model be 3n, the node number of corresponding input layer is 3n.Its
The node number of middle output layer is 1.
Alternatively, the node number of hidden layer can be according to formulaObtaining, wherein l is hidden layer
Node number, N is the node number of output layer, and M is the node number of output layer, and a is the constant of [1,10], Fig. 4 institute accordingly
The node number of the hidden layer shownAlternatively, the node number of hidden layer can also rule of thumb formula l=
log2N obtains, and the node number of the hidden layer shown in Fig. 4 is l=log accordingly2(3n)。
Wherein, the initial weight ω between input layer and hidden layer1With initial threshold B1, between hidden layer and output layer
Initial weight ω2With initial threshold B2, the be smaller than random number of predetermined threshold value.The error amount e reached is expected during training,
Maximum cycle N of training can be chosen according to actual needs.Also need to set artificial neural network algorithm model simultaneously
Learning rate η, wherein η=0.01～0.8.
Artificial neural network algorithm model is established according to above-mentioned steps.
S303b, in the above-mentioned artificial neural network algorithm model established, input information point POI in each region
Region several, each is at described tiThe actual order volume in moment and each region are at described tiThe quantity of the vehicle in moment, calculates
The output of every layer and the error E of whole artificial neural network algorithm model, and according to error E, adjust artificial neural network and calculate
Threshold value between each layer and weights in method model.
First, according to input information parameter X, (the information point POI number in the most above-mentioned each region, each region are described the
tiThe actual order volume in moment and each region are at described tiThe quantity of the vehicle in moment), at the beginning of between input layer and hidden layer
Beginning weights ω1With initial threshold B1, calculate the output Y obtaining input layer1'.According to the initial power between hidden layer and output layer
Value ω2With initial threshold B2, and the output Y of input layer1', calculate the output Y obtaining output layer1, this output Y1As
The output of artificial neural network algorithm model training for the first time.Further according to formula El=| Y-Yl|, it is thus achieved that whole ANN
The error E that network algorithm model produces when the first training1=| Y-Y1|, Y is that each region is at tiThe actual order volume in moment,
YlIt it is the output set of output layer during the l time repetitive exercise.Process at training of human artificial neural networks algorithm model exists
Need the t described in substantial amounts of data, i.e. the present embodimentiMoment refers to any historical time.
Then, according to formulaObtain the adjusting thresholds value between input layer and hidden layer
ωij(t-1), according to formulaObtain weighed value adjusting value B between input layer and hidden layerij(t-
1)；According to formulaObtain adjusting thresholds value ω between hidden layer and output layerjk(t-1),
And according to formulaObtain weighed value adjusting value B between hidden layer and output layerjk(t-1).Its
Described in ElIt is between order data and the actual order data of sample of the l time circulation artificial neural network algorithm model output
Error, η is the learning rate of artificial neural network algorithm model, ωijSave with hidden layer jth for input layer i-th node
Threshold value between point, BijFor the weights between input layer i-th node and hidden layer jth node, ωjkFor hidden layer jth
Threshold value between node and output layer kth node, BjkFor the power between hidden layer jth node and output layer kth node
Value.
S303c, according to threshold value between each layer in above-mentioned artificial neural network algorithm model and the adjusted value of weights, it is thus achieved that
The error sum of squares of model after this training, when expecting, during this error sum of squares is less than training, the error amount e reached, knot
Shu Xunlian.
Specifically, the l time circulation (or training) artificial god afterwards is obtained according to the threshold value between each layer after adjusting and weights
Error sum of squares e through network algorithm modell, and by this error sum of squares elExpect that the error amount e reached enters with during training
Row compares, when judging elDuring ＞ e, illustrate that this artificial neural network algorithm model does not the most train, then continue executing with above-mentioned
The method of S303b.Judged this e at that timelDuring ＜ e, this is described artificial neural network algorithm model is the most trained to complete, train with this
Good artificial neural network algorithm model carries out follow-up calculating.
Alternatively, the condition that in the present embodiment, Controlling model training terminates is it is also possible that work as cycle-index (frequency of training)
When meeting maximum cycle N preset, terminate the training to artificial neural network algorithm model.
The present embodiment considers the impact on target area order volume of the volume of the flow of passengers of the neighboring area of target area, and mesh
Mark region and the POI number of neighboring area of target area, the quantity impact on target area order volume of vehicle, and then by mesh
Mark region and the actual order volume of neighboring area of target area, POI number, vehicle quantity as input parameter, by target
Artificial neural network algorithm model, as output parameter, is carried out by the actual order volume of the neighboring area of region and target area
Training so that the artificial neural network algorithm model trained is more accurately predicted out target area at ti+1The of moment
Two prediction order volume.
S304, according to input parameter and described artificial neural network algorithm model, determine that described target area is described the
ti+1The second prediction order volume in moment；Wherein, described input parameter includes area information and temporal information.
Specifically, as shown in Figure 4, in the above-mentioned artificial neural network algorithm model trained input area information A and
Temporal information ti+1, then target area A can be obtained at moment ti+1The second order forecasting amount.Wherein time information ti+1Pass through
Tc input node is input to artificial neural network algorithm model.
The second prediction order volume that the present embodiment obtains, considers the neighboring area of target area during it calculates
The impact on target area order volume of the quantity of POI number, actual order volume and vehicle to target area
ti+1The prediction of the second prediction order volume in moment is more accurate.
The present embodiment obtains accurate second prediction order volume, this second prediction by artificial neural network algorithm model
Order volume can be as target area at ti+1The order volume in moment, i.e. the present embodiment can be used alone artificial neural network
Algorithm model realizes target area at ti+1The Accurate Prediction of moment order volume.
S305, according to described target area at described ti+1The first prediction order volume in moment and described target area exist
Described ti+1The second prediction order volume in moment, determines that described target area is at described ti+1The 3rd prediction order in moment
Amount.
Alternatively, in order to improve target area at t furtheri+1The Accurate Prediction of moment order volume, this enforcement
Example, also obtains target area at t by above-mentioned S101-S104 methodi+1First prediction order volume and the above-mentioned S301-in moment
S304 method obtains target area at ti+1The second prediction order volume in moment combines, it is thus achieved that target area is at ti+1Moment
The 3rd prediction order volume.Specifically, the present embodiment can be by the first prediction order volume and the meansigma methods of the second prediction order volume
As the 3rd prediction order volume；Or, according to relevant ERROR ALGORITHM model by the first prediction order and the second prediction order phase
In conjunction with obtaining the 3rd prediction order, the present embodiment obtains the 3rd prediction to according to the first prediction order volume and the second prediction order volume
The concrete grammar of order volume does not limits.
The order forecast method that the present invention provides, by information point POI in each region in the default territorial scope of acquisition
Number, each region are at tiThe actual order volume in moment and each region are at tiThe quantity of the vehicle in moment as input-defeated
Go out data artificial neural network algorithm model is trained so that the artificial neural network algorithm model that this trains can be relatively
Reflect the practical situation impact on order volume exactly, and then achieve target area at ti+1Second prediction in moment
The Accurate Prediction of order volume.It addition, in order to improve target area at t furtheri+1The Accurate Prediction of the order volume in moment,
Above-mentioned first prediction order volume and the second prediction order volume are also combined and generate the 3rd prediction order volume by the present invention, and the 3rd is pre-
Survey order volume and can more accurately reflect that target area is at ti+1The order volume in moment, thus improve vehicle scheduling
Reasonability, it is achieved that the effective alleviation to traffic pressure.
The schematic flow sheet of the order forecast method embodiment four that Fig. 5 provides for the present invention.The present embodiment refers to root
The detailed process of the 3rd prediction order volume is obtained according to gradient boosted tree algorithm model.On the basis of above-described embodiment, above-mentioned
S305 specifically may include that
S401, according to described second History Order data, described target area at described tiFirst prediction in moment is ordered
Single amount and described target area are at described tiThe second prediction order volume in moment, obtains gradient boosted tree algorithm model.
It should be noted that using gradient boosted tree algorithm model to target area at described ti+1The 3rd of moment
Before prediction order volume, needing to be trained gradient boosted tree algorithm model, this training process can be understood as setting up gradient
The process of boosted tree algorithm model.The present embodiment target area is at tiThe first prediction order volume in moment (can be understood as
The neighboring area of target area and target area is in the first prediction order volume of any historical juncture), target area is at ti
The second prediction order volume in moment (can be understood as the neighboring area of target area and target area in any historical juncture
Second prediction order volume), and area information (code information of the neighboring area of target area and target area) and time
Information (arbitrary continuation moment) is as characteristic condition (i.e. X value), with target area at tiThe actual order volume in moment is as Y
Value, is trained gradient boosted tree algorithm model, and the data volume used in wherein training process is bigger.Specifically, right
Gradient boosted tree algorithm model is trained including below step:
S401a, the relevant parameter of gradient boosted tree algorithm model is set, including loss function, pace of learning, learner
Maximum number, the depth capacity etc. of tree.
Wherein, the loss function of the present embodiment can be least square regression function, and its formula is:
Wherein, k is iterations, and n is the quantity of initial sample, yiFor i-th sample y value during initial training,For
The meansigma methods of sample y value, y used in the process of initial trainingi (k)For i-th sample y value, n in kth time iterative process(k)For
The quantity of sample in k iterative process,Meansigma methods for sample y value used in kth time iterative process.
It should be noted that what learning rate Shrinkage (or Learning Step) represented is the pace of learning of model, study
Speed the least expression model learning is the slowest, and the biggest then expression model learning is the fastest.Learner number refers to the number of regression tree, i.e.
Total number of the tree set up in whole training learning process.Accurate in order to improve the study of gradient boosted tree algorithm model
Property, generally learning rate Shrinkage is arranged a little bit smaller, total number of tree is arranged the most a bit.The degree of depth of tree is tree
The number of plies, the depth value of tree is the biggest, then the node set is the most, and included information is the most detailed, and the degree of depth of tree is the least, then the building of model
Vertical speed is the fastest.Choosing of the most above-mentioned each parameter can set according to actual needs.
S401b, method according to above-described embodiment, it is thus achieved that the neighboring area of target area and target area is when history
Carve tiThe first prediction order volume and the second prediction order volume, simultaneously from the second History Order data, obtain historical juncture tiRight
The actual order volume answered.By the neighboring area of above-mentioned target area and target area at historical juncture tiThe first corresponding prediction
Order volume, the second prediction order volume, temporal information and area information are as characteristic condition (i.e. X value), by target area and mesh
The neighboring area in mark region is at historical juncture tiGradient boosted tree algorithm model, as Y value, is carried out by corresponding actual order volume
Training.
Illustrating, in the order forecast method embodiment four that Fig. 6 provides for the present invention, gradient boosted tree algorithm model is tied
Structure schematic diagram.As shown in Figure 6, it is assumed that the degree of depth of the tree preset is 3, and the number of tree is 2, and learning rate is 0.5, wherein feature bar
(neighboring area of target area and target area is at historical juncture t for part X=iThe scope of the first corresponding prediction order volume a:
It is 10～40；The neighboring area of target area and target area is at historical juncture tiThe scope of the second corresponding prediction order b
For: 11～39；Temporal information in the range of: 8 o'clock to 13 o'clock；The area information of the neighboring area of target area and target area
Scope is: 1 to region, region 2).Y=(the actual order volume in 8 regions 1 be the actual order volume in 14,9 regions 2 be 16,10
The actual order volume in some region 1 is 24, and the actual order volume in 11 regions 1 is 26, and the actual order volume in 12 regions 2 is 36,
The actual order volume in 13 regions 2 is 37).According to loss function, each characteristic condition in above-mentioned X is judged, select with time
Between information 11 Y value is divided (second layer i.e. setting up tree) for node, it is thus achieved that less than 11 points (be i.e. positioned at left side burl
Point) Y value have (14,16,24), the nodal value of this corresponding node (Y value that this nodal value is corresponding equal to this node flat that be 18
Average), it is thus achieved that have (26,36,37) more than or equal to the Y value of 11 points (be i.e. positioned at right side tree node), accordingly, this node joint
Point value is 74.3.Then above-mentioned two node is divided (third layer i.e. setting up tree), now warp again according to loss function
Calculate and select for node, (14,16,24) to be divided with 15 in the first prediction order volume a, it is thus achieved that the Y value less than 15 has
(14), accordingly, the nodal value of this node is 14, and the Y value more than or equal to 15 has (16,24), accordingly the nodal value of this node
It is 20；Similarly, calculate selection (26,36,37) to be divided with area information 2 for node, it is thus achieved that the region less than 2 is corresponding
Y value have (26), the nodal value of this node is 26 accordingly, and Y value corresponding to region more than or equal to 2 has (36,37), accordingly
The nodal value of this node is 36.5.The degree of depth now set meets the depth value 3 presetting tree, then one tree has been set up.Then
The nodal value of corresponding node is deducted, it is thus achieved that the residual values that this node is corresponding by the actual Y value that node each in third layer is corresponding.With
Said method obtains the residual values of one tree: (14-14,16-20,24-20,26-26,36-36.5,37-36.5)=
(0 ,-4,4,0 ,-0.5,0.5).Then with above-mentioned residual values (0 ,-4,4,0 ,-0.5,0.5) as Y value, and according to above-mentioned first
The method for building up of tree carries out the foundation of second tree, it is thus achieved that second tree third layer in node corresponding to each node
Value, it is assumed that for (0,0 ,-0.5,0.5), the number now set meets total number 2 of default tree, then based on above-mentioned each parameter
Gradient boosted tree algorithm model has been set up, and carries out correlation predictive with this gradient boosted tree algorithm model established.This enforcement
Example is the process of setting up in order to illustrate this gradient boosted tree algorithm model, and the data of selection are less, but Practical Calculation process
In then need to use substantial amounts of historical data to train gradient boosted tree algorithm model.
S402, according to described input parameter and described gradient boosted tree algorithm model, obtain described target area described
Ti+1The 3rd prediction order volume in moment.
Specifically, as shown in Figure 4, inputting identical parameter in every tree trained, accordingly, each tree has individual defeated
Go out value, the output valve of each tree is added, target area can be obtained at ti+1The 3rd prediction order volume in moment, Qi Zhongsuo
The parameter of input includes: target area A, temporal information ti+1Moment, target area A are at ti+1The first prediction order in moment
Amount and target area A are at ti+1The second prediction order volume in moment.With reference to above-mentioned example, one tree inputs above-mentioned ginseng
Number, it is thus achieved that the output valve of one tree is Y1, for example, Y1=14；Accordingly, in second tree, above-mentioned identical ginseng is inputted
Number, it is thus achieved that the output valve of one tree is Y2, for example, Y2=-0.5.Finally each tree output valve is added, it is thus achieved that target area
A is at ti+1The 3rd prediction order volume in moment is Y, i.e. Y=Y1+Y2=14-0.5=13.5.
The present embodiment by by target area at ti+1The first prediction order volume in moment and target area are at ti+1Time
The the second prediction order volume carved is as inputting gain of parameter target area at ti+1The 3rd prediction order volume in moment, the 3rd
Prediction order volume decreases the first prediction order volume and the error of the second prediction order volume, thus further increases target area
Territory is at ti+1The Accurate Prediction of the order volume in moment, and then ensure that the vehicle dispatch system rational management to vehicle, the most slow
The traffic pressure solved, decreases vehicle dead mileage number, saves the energy, improves the trip convenience of people.
The order forecast method that the present invention provides, by gradient boosted tree algorithm model by above-mentioned target area at ti+1
The first prediction order volume and the second prediction order volume that moment produces are combined, it is thus achieved that the 3rd prediction order volume, the 3rd is pre-
Survey order volume and decrease the first prediction order volume and the error amount of the second prediction order volume, it is achieved thereby that to target area the
ti+1The Accurate Prediction of the order volume in moment, and then improve the reasonability of vehicle scheduling, the traffic pressure effectively alleviated, reduces
Vehicle dead mileage number, saves the energy, improves the trip convenience of people.
The structural representation of the order forecasting device embodiment one that Fig. 7 provides for the present invention, as it is shown in fig. 7, the present embodiment
Order forecasting device 100 may include that first acquisition module the 110, second acquisition module 120, the 3rd acquisition module 130 and
One processing module 140, wherein,
Above-mentioned first acquisition module 110, is used for obtaining target area at first time period [ti-1, ti] first order become
Change amount and variation tendency；Wherein, described first order variable quantity is equal to described target area at described tiThe actual of moment is ordered
Single amount and described ti-1The absolute value of the difference of the actual order volume in moment.
Above-mentioned second acquisition module 120, for according to described first order variable quantity, described variation tendency and described target
The first History Order data in region, obtain in described first History Order data identical with described variation tendency and with described
The difference of the first order variable quantity meets the second time period that the order variable quantity of preset threshold range is corresponding.
Above-mentioned 3rd acquisition module 130, orders for obtaining the second of the subsequent time period adjacent with described second time period
Altered amount.
Above-mentioned first processing module 140, for according to described second order variable quantity and described target area described the
tiThe actual order volume in moment, determines that described target area is at ti+1The first prediction order volume in moment.
The order forecasting device that the present invention provides, may be used for performing the method and technology scheme of above-described embodiment, and it realizes
Principle is similar with technique effect, and here is omitted.
The structural representation of the order forecasting device embodiment two that Fig. 8 provides for the present invention, as shown in Figure 8, in above-mentioned reality
On the basis of executing example, if described second acquisition module 120 gets multiple described second time period, described 3rd acquisition module
130 specifically include: the first acquiring unit 131 and processing unit 132, wherein,
Above-mentioned first acquiring unit 131, for obtain that the subsequent time period adjacent with each second time period is corresponding the
Three order variable quantities.
Above-mentioned processing unit 132, for obtaining the meansigma methods of multiple 3rd order variable quantity, and determines described meansigma methods
For described second order variable quantity.
The order forecasting device that the present invention provides, may be used for performing the method and technology scheme of above-described embodiment, and it realizes
Principle is similar with technique effect, and here is omitted.
The structural representation of the order forecasting device embodiment three that Fig. 9 provides for the present invention, as it is shown in figure 9, in above-mentioned reality
On the basis of executing example, further, the order forecasting device 100 of this enforcement, it is also possible to including: the 4th acquisition module the 150, the 5th
Acquisition module the 160, the 6th acquisition module the 170, second processing module 180 and the 3rd processing module 190, wherein,
Above-mentioned 4th acquisition module 150, the information point POI number in each region in obtaining default territorial scope；
Described default territorial scope includes described target area and at least one neighboring area.
Above-mentioned 5th acquisition module 160, for obtaining from the second History Order data that described default territorial scope is corresponding
Take each region at described tiThe actual order volume in moment and at described tiThe quantity of the vehicle in moment.
Above-mentioned 6th acquisition module 170, for existing according to the information point POI number in described each region, each region
Described tiThe actual order volume in moment and each region are at described tiThe quantity of the vehicle in moment, obtains artificial neural network
Algorithm model.
Above-mentioned second processing module 180, for according to input parameter and described artificial neural network algorithm model, determining institute
State target area at described ti+1The second prediction order volume in moment；Wherein, described input parameter includes area information and time
Information.
Above-mentioned 3rd processing module 190, is used for according to described target area at described ti+1The first prediction order in moment
Amount and described target area are at described ti+1The second prediction order volume in moment, determines that described target area is at described ti+1Time
The 3rd prediction order volume carved.
The order forecasting device that the present invention provides, may be used for performing the method and technology scheme of above-described embodiment, and it realizes
Principle is similar with technique effect, and here is omitted.
The structural representation of the order forecasting device embodiment four that Figure 10 provides for the present invention, as shown in Figure 10, above-mentioned
On the basis of embodiment, above-mentioned 3rd processing module 190 specifically may include that second acquisition unit 191 and the 3rd acquiring unit
192, wherein,
Above-mentioned second acquisition unit 191, for according to described second History Order data, described target area described the
tiThe first prediction order volume in moment and described target area are at described tiThe second prediction order volume in moment, obtains gradient and carries
Rise tree algorithm model.
Above-mentioned 3rd acquiring unit 192, for according to described input parameter and described gradient boosted tree algorithm model, obtains
Described target area is at described ti+1The 3rd prediction order volume in moment.
The order forecasting device that the present invention provides, may be used for performing the method and technology scheme of above-described embodiment, and it realizes
Principle is similar with technique effect, and here is omitted.
Last it is noted that various embodiments above is only in order to illustrate technical scheme, it is not intended to limit；To the greatest extent
The present invention has been described in detail by pipe with reference to foregoing embodiments, it will be understood by those within the art that: it depends on
So the technical scheme described in foregoing embodiments can be modified, or the most some or all of technical characteristic is entered
Row equivalent；And these amendments or replacement, do not make the essence of appropriate technical solution depart from various embodiments of the present invention technology
The scope of scheme.
Claims (8)
1. an order forecast method, it is characterised in that including:
Obtain target area at first time period [ti-1, ti] the first order variable quantity and variation tendency；Wherein, described first order
Altered amount is equal to described target area at described tiThe actual order volume in moment and described ti-1The actual order volume in moment
The absolute value of difference；
According to the first History Order data of described first order variable quantity, described variation tendency and described target area, obtain
In described first History Order data, identical with described variation tendency and with described first order variable quantity difference meets default
The second time period that the order variable quantity of threshold range is corresponding；
Obtain the second order variable quantity of the subsequent time period adjacent with described second time period；
According to described second order variable quantity and described target area at described tiThe actual order volume in moment, determines described mesh
Mark region is at ti+1The first prediction order volume in moment.
Method the most according to claim 1, it is characterised in that if getting multiple described second time period, then described in obtain
Take the second order variable quantity of the subsequent time period adjacent with described second time period, specifically include:
Obtain the 3rd order variable quantity that the subsequent time period adjacent with each second time period is corresponding；
Obtain the meansigma methods of multiple 3rd order variable quantity, and described meansigma methods is defined as described second order variable quantity.
Method the most according to claim 1 and 2, it is characterised in that described method also includes:
Information point POI number in each region in the default territorial scope of acquisition；Described default territorial scope includes described target
Region and at least one neighboring area；
Each region is obtained at described t from the second History Order data that described default territorial scope is correspondingiThe reality in moment
Order volume and at described tiThe quantity of the vehicle in moment；
According to the information point POI number in described each region, each region at described tiThe actual order volume in moment and each
Region is at described tiThe quantity of the vehicle in moment, obtains artificial neural network algorithm model；
According to input parameter and described artificial neural network algorithm model, determine that described target area is at described ti+1Moment
Second prediction order volume；Wherein, described input parameter includes area information and temporal information；
According to described target area at described ti+1The first prediction order volume in moment and described target area are at described ti+1
The second prediction order volume in moment, determines that described target area is at described ti+1The 3rd prediction order volume in moment.
Method the most according to claim 3, it is characterised in that according to described target area at described ti+1The of moment
One prediction order volume and described target area are at described ti+1The second prediction order volume in moment, determines that described target area exists
Described ti+1The 3rd prediction order volume in moment, specifically includes:
According to described second History Order data, described target area at described tiThe first prediction order volume and described in moment
Target area is at described tiThe second prediction order volume in moment, obtains gradient boosted tree algorithm model；
According to described input parameter and described gradient boosted tree algorithm model, obtain described target area at described ti+1Moment
The 3rd prediction order volume.
5. an order forecasting device, it is characterised in that including:
First acquisition module, is used for obtaining target area at first time period [ti-1, ti] the first order variable quantity and change become
Gesture；Wherein, described first order variable quantity is equal to described target area at described tiThe actual order volume in moment and described the
ti-1The absolute value of the difference of the actual order volume in moment；
Second acquisition module, for according to described first order variable quantity, described variation tendency and the first of described target area
History Order data, obtain identical with described variation tendency in described first History Order data and become with described first order
The difference of change amount meets the second time period that the order variable quantity of preset threshold range is corresponding；
3rd acquisition module, for obtaining the second order variable quantity of the subsequent time period adjacent with described second time period；
First processing module, for according to described second order variable quantity and described target area at described tiThe reality in moment
Order volume, determines that described target area is at ti+1The first prediction order volume in moment.
Device the most according to claim 5, it is characterised in that if described second acquisition module gets multiple described second
During the time period, described 3rd acquisition module specifically includes:
First acquiring unit, for obtaining the 3rd order change that the subsequent time period adjacent with each second time period is corresponding
Amount；
Processing unit, for obtaining the meansigma methods of multiple 3rd order variable quantity, and is defined as described second by described meansigma methods
Order variable quantity.
7. according to the device described in claim 5 or 6, it is characterised in that described device also includes:
4th acquisition module, the information point POI number in each region in obtaining default territorial scope；Described default region
Scope includes described target area and at least one neighboring area；
5th acquisition module, exists for obtaining each region from the second History Order data that described default territorial scope is corresponding
Described tiThe actual order volume in moment and at described tiThe quantity of the vehicle in moment；
6th acquisition module, is used for according to the information point POI number in described each region, each region at described tiMoment
Actual order volume and each region at described tiThe quantity of the vehicle in moment, obtains artificial neural network algorithm model；
Second processing module, for according to input parameter and described artificial neural network algorithm model, determining described target area
At described ti+1The second prediction order volume in moment；Wherein, described input parameter includes area information and temporal information；
3rd processing module, is used for according to described target area at described ti+1First prediction order volume and the described mesh in moment
Mark region is at described ti+1The second prediction order volume in moment, determines that described target area is at described ti+1The 3rd of moment is pre-
Survey order volume.
Device the most according to claim 7, it is characterised in that described 3rd processing module specifically includes:
Second acquisition unit, is used for according to described second History Order data, described target area at described tiThe first of moment
Prediction order volume and described target area are at described tiThe second prediction order volume in moment, obtains gradient boosted tree algorithm mould
Type；
3rd acquiring unit, for according to described input parameter and described gradient boosted tree algorithm model, obtains described target area
Territory is at described ti+1The 3rd prediction order volume in moment.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610431755.6A CN106127329A (en) | 2016-06-16 | 2016-06-16 | Order forecast method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610431755.6A CN106127329A (en) | 2016-06-16 | 2016-06-16 | Order forecast method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106127329A true CN106127329A (en) | 2016-11-16 |
Family
ID=57469746
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610431755.6A Pending CN106127329A (en) | 2016-06-16 | 2016-06-16 | Order forecast method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106127329A (en) |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106650741A (en) * | 2016-12-30 | 2017-05-10 | 深圳市捷顺科技实业股份有限公司 | License plate character recognition method and device |
| CN107092974A (en) * | 2016-11-29 | 2017-08-25 | 北京小度信息科技有限公司 | Dispense pressure prediction method and device |
| CN107146007A (en) * | 2017-04-26 | 2017-09-08 | 北京小度信息科技有限公司 | Order dispatch method and apparatus |
| WO2018209263A1 (en) * | 2017-05-11 | 2018-11-15 | Uber Technologies, Inc. | Network computer system to position service providers using provisioning level determinations |
| WO2018214361A1 (en) * | 2017-05-25 | 2018-11-29 | Beijing Didi Infinity Technology And Development Co., Ltd. | Systems and methods for improvement of index prediction and model building |
| CN109117973A (en) * | 2017-06-26 | 2019-01-01 | 北京嘀嘀无限科技发展有限公司 | A kind of net about vehicle order volume prediction technique and device |
| 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 |
| CN109544076A (en) * | 2018-11-28 | 2019-03-29 | 北京百度网讯科技有限公司 | Method and apparatus for generating information |
| CN109948822A (en) * | 2017-12-21 | 2019-06-28 | 北京嘀嘀无限科技发展有限公司 | A kind of geographic area Intranet about vehicle supply & demand gap forecasting method |
| CN110400024A (en) * | 2019-07-31 | 2019-11-01 | 京东城市(北京)数字科技有限公司 | Method, apparatus, equipment and the computer readable storage medium of order forecasting |
| CN110837909A (en) * | 2018-08-17 | 2020-02-25 | 北京京东尚科信息技术有限公司 | Method and device for predicting order quantity |
| US10628903B2 (en) | 2017-05-22 | 2020-04-21 | Uber Technologies, Inc. | Network computer system to implement counter values for arranging services |
| CN111160747A (en) * | 2019-12-23 | 2020-05-15 | 北京百度网讯科技有限公司 | Scheduling method and device of robot taxi, electronic equipment and storage medium |
| CN111563810A (en) * | 2020-04-28 | 2020-08-21 | 北京云从科技有限公司 | Credit wind control model generation method, credit evaluation system, machine-readable medium and device |
| 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 |
| CN112352250A (en) * | 2018-07-04 | 2021-02-09 | 索尼公司 | Information processing apparatus, information processing method, and program |
| CN113077181A (en) * | 2021-04-22 | 2021-07-06 | 上海钧正网络科技有限公司 | Parking station setting method, device, medium and electronic equipment |
| CN113673803A (en) * | 2020-05-13 | 2021-11-19 | 丰田自动车株式会社 | Vehicle distribution device, vehicle and terminal |
| CN113781077A (en) * | 2020-07-06 | 2021-12-10 | 京东城市(北京)数字科技有限公司 | Data processing method, device, equipment and computer readable storage medium |
-
2016
- 2016-06-16 CN CN201610431755.6A patent/CN106127329A/en active Pending
Cited By (27)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107092974A (en) * | 2016-11-29 | 2017-08-25 | 北京小度信息科技有限公司 | Dispense pressure prediction method and device |
| CN106650741A (en) * | 2016-12-30 | 2017-05-10 | 深圳市捷顺科技实业股份有限公司 | License plate character recognition method and device |
| CN107146007A (en) * | 2017-04-26 | 2017-09-08 | 北京小度信息科技有限公司 | Order dispatch method and apparatus |
| WO2018209263A1 (en) * | 2017-05-11 | 2018-11-15 | Uber Technologies, Inc. | Network computer system to position service providers using provisioning level determinations |
| US11551555B2 (en) | 2017-05-11 | 2023-01-10 | Uber Technologies, Inc. | Network computer system to position transport providers using provisioning level determinations |
| US10839695B2 (en) | 2017-05-11 | 2020-11-17 | Uber Technologies, Inc. | Network computer system to position service providers using provisioning level determinations |
| US10628903B2 (en) | 2017-05-22 | 2020-04-21 | Uber Technologies, Inc. | Network computer system to implement counter values for arranging services |
| WO2018214361A1 (en) * | 2017-05-25 | 2018-11-29 | Beijing Didi Infinity Technology And Development Co., Ltd. | Systems and methods for improvement of index prediction and model building |
| CN108960431A (en) * | 2017-05-25 | 2018-12-07 | 北京嘀嘀无限科技发展有限公司 | The prediction of index, the training method of model and device |
| CN109117973A (en) * | 2017-06-26 | 2019-01-01 | 北京嘀嘀无限科技发展有限公司 | A kind of net about vehicle order volume prediction technique and device |
| CN109948822A (en) * | 2017-12-21 | 2019-06-28 | 北京嘀嘀无限科技发展有限公司 | A kind of geographic area Intranet about vehicle supply & demand gap forecasting method |
| CN109948822B (en) * | 2017-12-21 | 2022-10-21 | 北京嘀嘀无限科技发展有限公司 | Method for predicting taxi appointment supply and demand gaps in geographic area |
| CN112352250A (en) * | 2018-07-04 | 2021-02-09 | 索尼公司 | Information processing apparatus, information processing method, and program |
| CN110837909A (en) * | 2018-08-17 | 2020-02-25 | 北京京东尚科信息技术有限公司 | Method and device for predicting 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 |
| CN109544076A (en) * | 2018-11-28 | 2019-03-29 | 北京百度网讯科技有限公司 | Method and apparatus for generating information |
| CN109544076B (en) * | 2018-11-28 | 2021-06-18 | 北京百度网讯科技有限公司 | Method and apparatus for generating information |
| CN110400024A (en) * | 2019-07-31 | 2019-11-01 | 京东城市(北京)数字科技有限公司 | Method, apparatus, equipment and the computer readable storage medium of order forecasting |
| CN111160747B (en) * | 2019-12-23 | 2022-07-22 | 北京百度网讯科技有限公司 | Scheduling method and device for unmanned robot taxi and electronic equipment |
| CN111160747A (en) * | 2019-12-23 | 2020-05-15 | 北京百度网讯科技有限公司 | Scheduling method and device of robot taxi, electronic equipment and storage medium |
| CN111563810A (en) * | 2020-04-28 | 2020-08-21 | 北京云从科技有限公司 | Credit wind control model generation method, credit evaluation system, machine-readable medium and device |
| CN113673803A (en) * | 2020-05-13 | 2021-11-19 | 丰田自动车株式会社 | Vehicle distribution device, vehicle and terminal |
| CN113781077A (en) * | 2020-07-06 | 2021-12-10 | 京东城市(北京)数字科技有限公司 | Data processing method, device, equipment and computer readable storage medium |
| 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 |
| CN113077181A (en) * | 2021-04-22 | 2021-07-06 | 上海钧正网络科技有限公司 | Parking station setting method, device, medium and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106127329A (en) | Order forecast method and device | |
| CN108197739B (en) | Urban rail transit passenger flow prediction method | |
| CN106200577B (en) | A kind of stereo garage dispatch control method | |
| CN103632212B (en) | System and method for predicating time-varying user dynamic equilibrium network-evolved passenger flow | |
| CN106503840A (en) | Parking stall Forecasting Methodology and system can be used in parking lot | |
| CN102044149B (en) | City bus operation coordinating method and device based on time variant passenger flows | |
| CN106193734B (en) | A kind of stereo garage dispatching control device | |
| CN102629418B (en) | Fuzzy kalman filtering-based traffic flow parameter prediction method | |
| CN102708698B (en) | Vehicle optimal-path navigation method based on vehicle internet | |
| CN104183119B (en) | Based on the anti-arithmetic for real-time traffic flow distribution forecasting method pushed away of section OD | |
| CN106503869A (en) | A kind of public bicycles dynamic dispatching method that is predicted based on website short-term needs | |
| CN112614343A (en) | Traffic signal control method and system based on random strategy gradient and electronic equipment | |
| CN107832882A (en) | A kind of taxi based on markov decision process seeks objective policy recommendation method | |
| CN109872535A (en) | A kind of current prediction technique of wisdom traffic, device and server | |
| CN107045785B (en) | A method of the short-term traffic flow forecast based on grey ELM neural network | |
| CN110570672B (en) | Regional traffic signal lamp control method based on graph neural network | |
| CN106781489A (en) | A kind of road network trend prediction method based on recurrent neural network | |
| CN104900063B (en) | A kind of short distance running time Forecasting Methodology | |
| DE102012016768A1 (en) | Technique for processing cartographic data for determining energy efficient routes | |
| CN102610092A (en) | Urban road speed predication method based on RBF (radial basis function) neural network | |
| CN108346293A (en) | A kind of arithmetic for real-time traffic flow Forecasting Approach for Short-term | |
| CN106935044A (en) | A kind of site location optimization method for preferentially coordinating control based on bus signals | |
| CN107945534A (en) | A kind of special bus method for predicting based on GMDH neutral nets | |
| CN111461500B (en) | Shared bicycle system tide phenomenon control method based on dynamic electronic fence and reinforcement learning | |
| CN104320789A (en) | Internet of vehicles RSU optimal allocation method based on game theory |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20161116 |
|
| RJ01 | Rejection of invention patent application after publication |