CN107967541A - A kind of information forecasting method, information prediction device and server cluster - Google Patents
A kind of information forecasting method, information prediction device and server cluster Download PDFInfo
- Publication number
- CN107967541A CN107967541A CN201711394565.2A CN201711394565A CN107967541A CN 107967541 A CN107967541 A CN 107967541A CN 201711394565 A CN201711394565 A CN 201711394565A CN 107967541 A CN107967541 A CN 107967541A
- Authority
- CN
- China
- Prior art keywords
- data
- convolutional neural
- neural networks
- multigroup
- data set
- 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
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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Computational Linguistics (AREA)
- Development Economics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Game Theory and Decision Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The embodiment of the present invention provides a kind of information forecasting method, information prediction device and server cluster, the described method includes：Historical time sequence data is obtained, its second data set for including the first data set and including at least a kind of second data；Convolutional neural networks are constructed according to the historical time sequence data；Into the convolutional neural networks, input has the planning data of same alike result with one kind in second data set or the second data of multiclass；The convolutional neural networks have the prediction data of same alike result according to planning data output with the first data in first data set.The information forecasting method of the embodiment of the present invention can realize the effect of high-precision forecast Future Information by using convolutional neural networks.
Description
Technical field
The present invention relates to information prediction field, the information prediction of more particularly to a kind of information forecasting method and application this method
Device and server cluster.
Background technology
In the prior art, many times user all suffer from needing according to substantial amounts of historical data and to some Future Datas or
Future Information etc. is made prediction, is estimated.At present, when giving a forecast, estimating, the method for use is usually manually prediction, or is used
Traditional computer learning method is predicted, no matter but using which kind of method, can all expend longer time, especially people
Work predicts that not only time-consuming and needs to put into great effort, but estimation results are often barely satisfactory.And computer learning method
Also it can often be subject to the multicollinearity effect of data, and cause estimate accuracy relatively low.
The content of the invention
The embodiment of the present invention problem to be solved is to provide one kind and predicts Future Information using convolutional neural networks
Information forecasting method and application this method information prediction device, server cluster.
To solve the above-mentioned problems, the embodiment of the present invention provides a kind of information forecasting method, including：
Historical time sequence data is obtained, its second data for including the first data set and including at least a kind of second data
Collection；
Convolutional neural networks are constructed according to the historical time sequence data；
Into the convolutional neural networks, input has phase with one kind in second data set or the second data of multiclass
With the planning data of attribute；
The convolutional neural networks are exported according to the planning data to be had with the first data in first data set
The prediction data of same alike result.
Preferably, the first data in first data set and one kind or multiclass second in second data set
Data corresponding association sequentially in time；
The method further includes：
Associated first data and the second data are divided into one group sequentially in time；
It is described to be specially according to historical time sequence data construction convolutional neural networks：
Construct convolutional neural networks structure；
The convolutional neural networks are formed based on multigroup first data and the second data.
Preferably, further include：
Multigroup first data and the second data are distributed sequentially in time；
It is described to form the convolutional neural networks based on multigroup first data and the second data, be specially：
Multigroup first data and the second data according to being distributed in chronological order determine the convolutional neural networks knot
Each weights in structure, to form the convolutional neural networks.
Preferably, the basis property is distributed sequentially in time multigroup first data and the second data determine institute
Stating each weights in convolutional neural networks structure is specially：
Multigroup first data and the second data that are distributed sequentially in time are subjected to centralization processing, with unified institute
State the dimension of the first data and the second data；
The convolutional neural networks are determined according to multigroup first data and the second data after centralization is handled
Each weights in structure.
Preferably, further include：
Determine that there are first data and second of identical associate feature in multigroup first data and the second data
The group of data；
The first data between the group and the second data are carried out weights to share.
Preferably, first data include activation amount data, second data include price, festivals or holidays, marketing
It is at least one in information.
The embodiment of the present invention provides a kind of information prediction device at the same time, including：
Acquisition device, for obtaining historical time sequence data, the historical time sequence data includes the first data set
With the second data set for including at least a kind of second data；And
Training device, it is used to construct convolutional neural networks, the convolutional Neural according to the historical time sequence data
Network can be according to the quota with one kind in second data set or the second data of multiclass with same alike result of reception
According to and export with first data have same alike result prediction data.
Preferably, the first data in first data set and one kind or multiclass second in second data set
Data corresponding association sequentially in time；
The training device is used to sequentially in time be divided into associated first data and the second data
One group, and according to multigroup first data and the second data determine construction convolutional neural networks structure in each weights, into
And form the convolutional neural networks.
Preferably, the training device is additionally operable to：
Determine that there are first data and second of identical associate feature in multigroup first data and the second data
The group of data；
The first data in the group and the second data are carried out weights to share.
The embodiment of the present invention also provides a kind of server cluster, including：
At least one processor；
At least one processor, wherein executable instruction is stored with the memory, wherein, in the executable instruction
Performed by the processor so that the processor proceeds as follows：
Historical time sequence data is obtained, its second data for including the first data set and including at least a kind of second data
Collection；
Convolutional neural networks are constructed according to the historical time sequence data；
Into the convolutional neural networks, input has phase with one kind in second data set or the second data of multiclass
With the planning data of attribute；
The convolutional neural networks are exported according to the planning data to be had with the first data in first data set
The prediction data of same alike result.
The beneficial effect of the embodiment of the present invention is, using the first data and the second data being associated in historical data come
Convolutional neural networks are constructed, make to be based on some planning datas using the powerful data-handling capacity of convolutional neural networks and go through
The secondary relationship of history data, and deduce the prediction data influenced by the planning data, with for user some products are done it is pre-
Valuable reference information is provided when estimating.
Brief description of the drawings
Fig. 1 is the flow chart of the information forecasting method in the embodiment of the present invention.
Method flow diagram when Fig. 2 is the structure convolutional neural networks in one embodiment of the invention.
Fig. 3 is method flow diagram when convolutional neural networks are built in another embodiment of the present invention.
Fig. 4 is method flow diagram when convolutional neural networks are built in further embodiment of this invention.
Fig. 5 is when being used to eliminate data synteny in convolutional neural networks in the information forecasting method in the embodiment of the present invention
Method flow diagram.
Fig. 6 is the structure diagram of the information prediction device in the embodiment of the present invention.
Fig. 7 is the structure diagram of the server cluster in the embodiment of the present invention.
Embodiment
The present invention is described in detail below in conjunction with attached drawing.
It should be understood that various modifications can be made to disclosed embodiments.Therefore, description below should not regard
To limit, and only as the example of embodiment.Those skilled in the art will expect within the scope and spirit of this
Other modifications.
Comprising in the description and the attached drawing of a part for constitution instruction shows embodiment of the disclosure, and with it is upper
What face provided is used to explain the disclosure together to the substantially description of the disclosure and the detailed description given below to embodiment
Principle.
It is of the invention by the description to the preferred form of the embodiment that is given as non-limiting examples with reference to the accompanying drawings
These and other characteristic will become apparent.
It is also understood that although with reference to some instantiations, invention has been described, but people in the art
Member realize with can determine the present invention many other equivalents, they have feature as claimed in claim and therefore all
In the protection domain limited whereby.
When read in conjunction with the accompanying drawings, in view of described further below, above and other aspect, the feature and advantage of the disclosure will become
It is more readily apparent.
The specific embodiment of the disclosure is described hereinafter with reference to attached drawing；It will be appreciated, however, that the disclosed embodiments are only
The example of the disclosure, it can use various ways to implement.It is known and/or repeat function and structure be not described in detail to avoid
Unnecessary or unnecessary details make it that the disclosure is smudgy.Therefore, specific structural and feature disclosed herein is thin
Section is not intended to restrictions, but as just the basis of claim and representative basis for instruct those skilled in the art with
Substantially any appropriate detailed construction diversely uses the disclosure.
This specification can be used phrase " in one embodiment ", " in another embodiment ", " in another embodiment
In " or " in other embodiments ", it may refer to according to one or more of identical or different embodiment of the disclosure.
As shown in Figure 1, providing a kind of information forecasting method in the embodiment of the present invention, it includes：
Historical time sequence data is obtained, its second data for including the first data set and including at least a kind of second data
Collection, such as obtain the computer sales volume data of 2000 to 2015 and influence the influence data of computer sales volume, wherein, computer sales volume
Data are the first data set, and it is the second data set to influence data, it can be included such as the price data as the second data, marketing letter
Cease data etc.；
Convolutional neural networks are constructed according to historical time sequence data；
Into convolutional neural networks, input has same alike result with one kind in the second data set or the second data of multiclass
Planning data；
Convolutional neural networks are exported according to planning data has the pre- of same alike result with the first data in the first data set
Survey data.It is, system is by regarding the first data in the first data set as output data, by the second data set
As input data, constructed with this has according to planning data (that is, user looks ahead a certain date or a certain two data
The second data of period) predict future the first data message function convolutional neural networks, make user's future to be learnt
When certain time period or the first data sometime put, it can be planned by being inputted into convolutional neural networks in the above-mentioned period
Or corresponding second data during time point, the high-precision forecast value of the first data can be obtained, is that user is calculating prediction data
When provide reliable data reference information, making user, work efficiency obtains when doing some work such as product analysis
It is obviously improved.
Further, as shown in Fig. 2, the first data and corresponding second data in the first data set in the present embodiment
Have between associate feature, such as corresponding first data and the second data to have mutually to restrict between the second data concentrated and close
System, or numerical values recited of the first data of value effect of the second data etc..When the first data and the second data have above-mentioned pass
When joining characteristic, data structure convolutional neural networks are based on for the ease of system, while ensure the convolutional neural networks tool of structure
Have the ability of high-precision forecast information, can in advance by the first data in the first data set with it is a kind of or more in the second data set
The corresponding association sequentially in time of the second data of class.For example, include by product sales volume or activation amount, the second data of the first data
Exemplified by one or more of price, marketing message, season, festivals or holidays, same day present species, when system obtains above-mentioned data
Afterwards, which can be corresponded to association according to the incremental order on date, i.e. the product sales volume or activation amount of No.1 and the same day
The corresponding association of the data such as price (all second data for corresponding to No.1), the product sales volume or the valency on activation amount and the same day of No. two
The data such as lattice, which correspond to, to be associated, No. three ... and so on, untill all data are carried out matching association.
Further, system, can be first by after the processing of above-mentioned pre-treatment step in specific configuration convolutional neural networks
The first data and the second data, that is, the first data and the second data that will be associated, be divided into one group sequentially in time
One group of training data is formed, after the step process, system there can be multigroup training data.Then, system can be according to this
Multigroup training data constructs convolutional neural networks, including step：
Construct convolutional neural networks structure, namely the learning model of construction convolutional neural networks；
Convolutional neural networks are formed based on multigroup first data and the second data, that is, into convolutional neural networks structure
Multigroup training data is inputted, is learnt it, and then formation stable structure can be by the direct applied convolutional Neural net of user
Network.
Specifically, when inputting above-mentioned multigroup training data into convolutional neural networks structure, its input sequence can be direct
The learning outcome of convolutional neural networks is influenced, that is, the precision of convolutional neural networks information of forecasting can be directly affected, in order to avoid
The precision of prediction of convolutional neural networks is influenced, is being performed in the present embodiment according to multigroup training data structure convolutional neural networks
When, as shown in figure 3, specifically including：
Multigroup first data and the second data are distributed sequentially in time, i.e. multigroup training data is suitable according to the time
Sequence is distributed；
Multigroup first data and the second data according to being distributed in chronological order determine respectively to weigh in convolutional neural networks structure
Value, to form convolutional neural networks.Knownly, there are multiple equations (with multigroup training data quantity phase in convolutional neural networks
Deng and correspond), have multiple unknown weights in each equation, convolutional neural networks are carried out by multigroup training data
Study, and then determine the weights in each equation, make only have two unknown quantitys in each equation, should when user's input prediction value
Predicted value is one of unknown quantity, due to only having a unknown quantity (that is, being predicted value) in equation at this time, therefore can be light
Pine obtains the concrete numerical value of the unknown quantity.
For example, still using the first data as product sales volume or activation amount, the second data include price, marketing message, season, section
Assume system tune (in the present embodiment by taking the second data are price as an example) exemplified by one or more of holiday, same day present species
The historical time sequence data taken is the data of 2000 to 2015, and system is first according to each date when handling the data
Sales volume data and price data with associate feature will be selected in substantial amounts of sales volume data and price data, and divided
For one group, training data is formed, such as sales volume on January 1st, 2000 and price are one group, the price on January 2 and sales volume are one
Group, No. three ..., then date by multigroup training data according on December 20th, 1 day 1 January in 2000
Incremental order or the date on January 1st, 20 days 1 December in 2015 successively decrease order be distributed, with to convolution god
Inputted during through inputting training data in network structure according to above-mentioned distribution sequence.
Further, as shown in figure 4, since the actual sales volume in each day in historical data might have notable difference, such as
Sales volume during 11 National Day, all reaches 1000 daily, and the sales volume in each day only has 20 in usually working day, in order to avoid
Convolutional neural networks study when be subject to dimension disunity and bringing is influenced, in the present embodiment system according to according to the time it is suitable
Multigroup first data and the second data of sequence distribution determine：
Multigroup first data and the second data that are distributed sequentially in time are subjected to centralization processing, with unified first number
According to the dimension with the second data；
Determined according to multigroup first data and the second data after centralization is handled each in convolutional neural networks structure
Weights, the convolutional neural networks precision of prediction that thus step trains further improve.
Further, as shown in figure 5, the information forecasting method in the present embodiment further includes：
Determine the group of the first data and the second data in multigroup first data and the second data with identical associate feature
Not；
The first data between group and the second data are carried out weights to share.
For example, using the first data as sales volume or activation amount, the second data include price, marketing message, exemplified by festivals or holidays, when
Convolutional neural networks find first group of training data, second group of training data and the 3rd group of training data during study
In price be to directly affect the data of same day sales volume, then at this time, convolutional neural networks just will corresponding first group of training number
According to the weights of, the first data in second group of training data, the 3rd group of training data weights are carried out to share, at the same will it is corresponding this three
The weights of the second data in group training data carry out weights and share, wherein, the weights that weights are shared are carried out, numerical value is equal.It is logical
Crossing this kind of processing mode can make to be not in multiple conllinear between the first data and the second data in above-mentioned three groups of training datas
Property, make the learning outcome of influence convolutional neural networks, cause precision during its subsequent predictive information to reduce, it is inclined larger prediction occur
Difference.
As shown in fig. 6, a kind of information prediction device is provided at the same time in the embodiment of the present invention, including：
Acquisition device, for obtaining historical time sequence data, historical time sequence data includes the first data set and bag
The second data set of at least a kind of second data is included, such as obtains the computer sales volume data of 2000 to 2015 and influences computer
The influence data of sales volume, wherein, computer sales volume data are the first data set, and it is the second data set to influence data, it can be included such as
Price data, marketing message data as the second data etc.；And
Training device, it is used to construct convolutional neural networks according to historical time sequence data, and convolutional neural networks can
The planning data of same alike result is had according to reception and one kind in the second data set or multiclass the second data and is exported and the
One data have the prediction data of same alike result.It is, training device is by the way that the first data in the first data set are used as
Output data, using the second data in the second data set as input data, with this construct with according to planning data (
That is, user looks ahead the second data of a certain date or certain time period) predict the first data message function in future
Convolutional neural networks, can be by convolution when the user is intended to learn following certain time period or the first data sometime put
Corresponding second data in above-mentioned period or time point are planned in input in neutral net, can obtain the high-precision of the first data
Predicted value is spent, reliable data reference information is provided when calculating prediction data for user, user is for example produced doing some
Product are analyzed when work, and work efficiency is obviously improved.
Further, one kind in the first data in the first data set and the second data set or the second data of multiclass according to
Time sequencing corresponds to association；For example, using the first data as product sales volume or activation amount, the second data include price, marketing message,
Exemplified by one or more of season, festivals or holidays, same day present species, when training device obtain above-mentioned data after, can by this two
Kind data correspond to association according to the incremental order on date, i.e. the product sales volume or the data such as activation amount and the price on the same day of No.1
(all second data for corresponding to No.1) corresponding association, the product sales volume or the data pair such as activation amount and the price on the same day of No. two
It should associate, No. three ... and so on, untill all data are carried out matching association.
Training device is used to associated the first data and the second data are divided into one group sequentially in time,
That is, by associated the first data and the second data, one group of formation, one group of training data, Ran Hougen are divided into sequentially in time
Each weights in the convolutional neural networks structure of construction are determined according to multigroup first data and the second data, and then form convolutional Neural
Network.That is, inputting multigroup training data into convolutional neural networks structure, learnt it, and then form stable structure
Can be by the direct applied convolutional neural networks of user.
Specifically, when inputting above-mentioned multigroup training data into convolutional neural networks structure, its input sequence can be direct
The learning outcome of convolutional neural networks is influenced, that is, the precision of convolutional neural networks information of forecasting can be directly affected, in order to avoid
The precision of prediction of convolutional neural networks is influenced, is being performed in the present embodiment according to multigroup training data structure convolutional neural networks
When, specifically include：
Multigroup first data and the second data are distributed sequentially in time, i.e. multigroup training data is suitable according to the time
Sequence is distributed；
Multigroup first data and the second data according to being distributed in chronological order determine respectively to weigh in convolutional neural networks structure
Value, to form convolutional neural networks.Knownly, there are multiple equations (with multigroup training data quantity phase in convolutional neural networks
Deng and correspond), have multiple unknown weights in each equation, convolutional neural networks are carried out by multigroup training data
Study, and then determine the weights in each equation, make only have two unknown quantitys in each equation, should when user's input prediction value
Predicted value is one of unknown quantity, due to only having a unknown quantity (that is, being predicted value) in equation at this time, therefore can be light
Pine obtains the concrete numerical value of the unknown quantity.
Further, since the actual sales volume in each day in historical data might have notable difference, such as 11 phases on National Day
Between sales volume, all reach 1000 daily, and the sales volume in each day only has 20 in usually working day, in order to avoid convolutional Neural net
Network in study be subject to dimension disunity and bringing is influenced, training device is according to property sequentially in time point in the present embodiment
Multigroup first data and the second data of cloth determine：
Multigroup first data and the second data that are distributed sequentially in time are subjected to centralization processing, with unified first number
According to the dimension with the second data；
Determined according to multigroup first data and the second data after centralization is handled each in convolutional neural networks structure
Weights.
Further, the training device in the present embodiment is additionally operable to：
Determine the group of the first data and the second data in multigroup first data and the second data with identical associate feature
Not；
The first data in group and the second data are carried out weights to share.
For example, using the first data as sales volume or activation amount, the second data include price, marketing message, exemplified by festivals or holidays, when
Convolutional neural networks find first group of training data, second group of training data and the 3rd group of training data during study
In price be to directly affect the data of same day sales volume, then at this time, convolutional neural networks just will corresponding first group of training number
According to the weights of, the first data in second group of training data, the 3rd group of training data weights are carried out to share, at the same will it is corresponding this three
The weights of the second data in group training data carry out weights and share, wherein, the weights that weights are shared are carried out, numerical value is equal.It is logical
Crossing this kind of processing mode can make to be not in multiple conllinear between the first data and the second data in above-mentioned three groups of training datas
Property, make the learning outcome of influence convolutional neural networks, cause precision during its subsequent predictive information to reduce, it is inclined larger prediction occur
Difference.
As shown in fig. 7, also providing a kind of server cluster in the embodiment of the present invention, it includes：
At least one processor；
Executable instruction is stored with least one processor, wherein memory, wherein, in executable instruction by processor
Perform so that processor proceeds as follows：
Historical time sequence data is obtained, its second data for including the first data set and including at least a kind of second data
Collection；
Convolutional neural networks are constructed according to historical time sequence data；
Into convolutional neural networks, input has same alike result with one kind in the second data set or the second data of multiclass
Planning data；
Convolutional neural networks are exported according to planning data has the pre- of same alike result with the first data in the first data set
Survey data.
Further, processor is based on executable instruction, is also operated：
Construct convolutional neural networks structure, namely the learning model of construction convolutional neural networks；
Convolutional neural networks are formed based on multigroup first data and the second data, that is, into convolutional neural networks structure
The multigroup training data formed by the first data with associate feature and the second data is inputted, is learnt it, and then shape
Can be by the direct applied convolutional neural networks of user into stable structure.
Further, processor is based on executable instruction, is also operated：
Multigroup first data and the second data are distributed sequentially in time, i.e. multigroup training data is suitable according to the time
Sequence is distributed；
Multigroup first data and the second data according to being distributed in chronological order determine respectively to weigh in convolutional neural networks structure
Value, to form convolutional neural networks.Knownly, there are multiple equations (with multigroup training data quantity phase in convolutional neural networks
Deng and correspond), have multiple unknown weights in each equation, convolutional neural networks are carried out by multigroup training data
Study, and then determine the weights in each equation, make only have two unknown quantitys in each equation, should when user's input prediction value
Predicted value is one of unknown quantity, due to only having a unknown quantity (that is, being predicted value) in equation at this time, therefore can be light
Pine obtains the concrete numerical value of the unknown quantity.
Further, processor is based on executable instruction, is also operated：
Multigroup first data and the second data that are distributed sequentially in time are subjected to centralization processing, with unified first number
According to the dimension with the second data；
Determined according to multigroup first data and the second data after centralization is handled each in convolutional neural networks structure
Weights.
Further, processor is based on executable instruction, is also operated：
Determine the group of the first data and the second data in multigroup first data and the second data with identical associate feature
Not；
The first data between group and the second data are carried out weights to share.
For example, using the first data as sales volume or activation amount, the second data include price, marketing message, exemplified by festivals or holidays, when
Convolutional neural networks find first group of training data, second group of training data and the 3rd group of training data during study
In price be to directly affect the data of same day sales volume, then at this time, convolutional neural networks just will corresponding first group of training number
According to the weights of, the first data in second group of training data, the 3rd group of training data weights are carried out to share, at the same will it is corresponding this three
The weights of the second data in group training data carry out weights and share, wherein, the weights that weights are shared are carried out, numerical value is equal.It is logical
Crossing this kind of processing mode can make to be not in multiple conllinear between the first data and the second data in above-mentioned three groups of training datas
Property, make the learning outcome of influence convolutional neural networks, cause precision during its subsequent predictive information to reduce, it is inclined larger prediction occur
Difference.
Above example is only the exemplary embodiment of the present invention, is not used in the limitation present invention, protection scope of the present invention
It is defined by the claims.Those skilled in the art can make the present invention respectively in the essence and protection domain of the present invention
Kind modification or equivalent substitution, this modification or equivalent substitution also should be regarded as being within the scope of the present invention.
Claims (10)
- A kind of 1. information forecasting method, it is characterised in that including：Historical time sequence data is obtained, its second data set for including the first data set and including at least a kind of second data；Convolutional neural networks are constructed according to the historical time sequence data；Into the convolutional neural networks, input has same genus with one kind in second data set or the second data of multiclass The planning data of property；The convolutional neural networks are exported with the first data in first data set with identical according to the planning data The prediction data of attribute.
- 2. according to the method described in claim 1, it is characterized in that, the first data and described second in first data set One kind or the second data of multiclass corresponding association sequentially in time in data set；The method further includes：Associated first data and the second data are divided into one group sequentially in time；It is described to be specially according to historical time sequence data construction convolutional neural networks：Construct convolutional neural networks structure；The convolutional neural networks are formed based on multigroup first data and the second data.
- 3. according to the method described in claim 2, it is characterized in that, further include：Multigroup first data and the second data are distributed sequentially in time；It is described to form the convolutional neural networks based on multigroup first data and the second data, be specially：Multigroup first data and the second data according to being distributed in chronological order are determined in the convolutional neural networks structure Each weights, to form the convolutional neural networks.
- 4. according to the method described in claim 3, it is characterized in that, basis property distribution sequentially in time it is multigroup described First data and the second data determine that each weights are specially in the convolutional neural networks structure：Multigroup first data and the second data that are distributed sequentially in time are subjected to centralization processing, with unified described the The dimension of one data and the second data；The convolutional neural networks structure is determined according to multigroup first data and the second data after centralization is handled In each weights.
- 5. according to the method described in claim 2, it is characterized in that, further include：Determine that there are first data and the second data of identical associate feature in multigroup first data and the second data Group；The first data between the group and the second data are carried out weights to share.
- 6. according to the method described in claim 1, it is characterized in that, first data include activation amount data, described second Data include at least one in price, festivals or holidays, marketing message.
- A kind of 7. information prediction device, it is characterised in that including：Acquisition device, for obtaining historical time sequence data, the historical time sequence data includes the first data set and bag Include the second data set of at least a kind of second data；AndTraining device, it is used to construct convolutional neural networks, the convolutional neural networks according to the historical time sequence data Can have the planning data of same alike result according to reception and one kind in second data set or multiclass the second data and Output has the prediction data of same alike result with first data.
- 8. according to the information prediction device that go entirely described in 7, it is characterised in that the first data and institute in first data set State one kind in the second data set or the corresponding association sequentially in time of the second data of multiclass；The training device is used to associated first data and the second data are divided into one group sequentially in time, And each weights in the convolutional neural networks structure of construction are determined according to multigroup first data and the second data, and then formed The convolutional neural networks.
- 9. information prediction device according to claim 8, it is characterised in that the training device is additionally operable to：Determine that there are first data and the second data of identical associate feature in multigroup first data and the second data Group；The first data in the group and the second data are carried out weights to share.
- 10. a kind of server cluster, including：At least one processor；At least one processor, wherein executable instruction is stored with the memory, wherein, in the executable instruction by institute State processor execution so that the processor proceeds as follows：Historical time sequence data is obtained, its second data set for including the first data set and including at least a kind of second data；Convolutional neural networks are constructed according to the historical time sequence data；Into the convolutional neural networks, input has same genus with one kind in second data set or the second data of multiclass The planning data of property；The convolutional neural networks are exported with the first data in first data set with identical according to the planning data The prediction data of attribute.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711394565.2A CN107967541A (en) | 2017-12-21 | 2017-12-21 | A kind of information forecasting method, information prediction device and server cluster |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711394565.2A CN107967541A (en) | 2017-12-21 | 2017-12-21 | A kind of information forecasting method, information prediction device and server cluster |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107967541A true CN107967541A (en) | 2018-04-27 |
Family
ID=61995564
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711394565.2A Pending CN107967541A (en) | 2017-12-21 | 2017-12-21 | A kind of information forecasting method, information prediction device and server cluster |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107967541A (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6032125A (en) * | 1996-11-07 | 2000-02-29 | Fujitsu Limited | Demand forecasting method, demand forecasting system, and recording medium |
| JP2001236337A (en) * | 2000-02-22 | 2001-08-31 | Fuji Electric Co Ltd | Predicting device using neural network |
| US20140108094A1 (en) * | 2012-06-21 | 2014-04-17 | Data Ventures, Inc. | System, method, and computer program product for forecasting product sales |
| CN104268638A (en) * | 2014-09-11 | 2015-01-07 | 广州市香港科大霍英东研究院 | Photovoltaic power generation system power predicting method of elman-based neural network |
| CN104598765A (en) * | 2015-02-16 | 2015-05-06 | 常州瑞信电子科技有限公司 | Building energy consumption prediction method based on elastic adaptive neural network |
| CN106779129A (en) * | 2015-11-19 | 2017-05-31 | 华北电力大学（保定） | A kind of Short-Term Load Forecasting Method for considering meteorologic factor |
-
2017
- 2017-12-21 CN CN201711394565.2A patent/CN107967541A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6032125A (en) * | 1996-11-07 | 2000-02-29 | Fujitsu Limited | Demand forecasting method, demand forecasting system, and recording medium |
| JP2001236337A (en) * | 2000-02-22 | 2001-08-31 | Fuji Electric Co Ltd | Predicting device using neural network |
| US20140108094A1 (en) * | 2012-06-21 | 2014-04-17 | Data Ventures, Inc. | System, method, and computer program product for forecasting product sales |
| CN104268638A (en) * | 2014-09-11 | 2015-01-07 | 广州市香港科大霍英东研究院 | Photovoltaic power generation system power predicting method of elman-based neural network |
| CN104598765A (en) * | 2015-02-16 | 2015-05-06 | 常州瑞信电子科技有限公司 | Building energy consumption prediction method based on elastic adaptive neural network |
| CN106779129A (en) * | 2015-11-19 | 2017-05-31 | 华北电力大学（保定） | A kind of Short-Term Load Forecasting Method for considering meteorologic factor |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Farah et al. | Bayesian emulation and calibration of a dynamic epidemic model for A/H1N1 influenza | |
| Emerson et al. | Revealing the demographic histories of species using DNA sequences | |
| CN105550268B (en) | Big data process modeling analysis engine | |
| GB2547816B (en) | Actually-measured marine environment data assimilation method based on sequence recursive filtering three-dimensional variation | |
| Schramski et al. | Network environ theory, simulation, and EcoNet® 2.0 | |
| Wang et al. | Comparative performance of logistic regression and survival analysis for detecting spatial predictors of land-use change | |
| CN103854068A (en) | Method for forecasting residential quarter short-term loads | |
| Mahdy et al. | Optimal control and bifurcation diagram for a model nonlinear fractional SIRC | |
| Patrikalakis et al. | Toward Dynamic Data-Driven Systems for Rapid Adaptive Interdisciplinary Ocean Forecasting | |
| Wang et al. | Single machine scheduling with exponential sum-of-logarithm-processing-times based learning effect | |
| MX2013006155A (en) | Systems and methods for reducing reservoir simulator model run time. | |
| JP7169082B2 (en) | Prediction system for predicting construction price of building and prediction system for predicting construction period | |
| Bala et al. | Systems thinking: system dynamics | |
| CN101877077A (en) | Time series predicting model | |
| JP6909596B2 (en) | Knowledge model construction system and knowledge model construction method | |
| Alakaş et al. | Problem specific variable selection rules for constraint programming: a type ii mixed model assembly line balancing problem case | |
| CN115906626A (en) | Underground water organic pollution mutual-feedback inversion tracing method based on artificial intelligence | |
| CN106648839A (en) | Method and device for processing data | |
| Drovosekova et al. | Parallel computations annexed to the problem of physical processes simulation | |
| CN1760881A (en) | Modeling method of forecast in device of computer aided diagnosis through using not diagnosed cases | |
| CN107967541A (en) | A kind of information forecasting method, information prediction device and server cluster | |
| CN108038335A (en) | A kind of method and apparatus of definite aircraft skin element stress load | |
| Datta et al. | Computational Approaches to Materials Design: Theoretical and Practical Aspects: Theoretical and Practical Aspects | |
| Gu et al. | A computationally efficient state space approach to estimating multilevel regression models and multilevel confirmatory factor models | |
| Martín-Vázquez et al. | Wind energy forecasting at different time horizons with individual and global models |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180427 |
|
| RJ01 | Rejection of invention patent application after publication |