CN110135647A - A kind of control method and control device for realizing trend prediction based on feature modeling - Google Patents
A kind of control method and control device for realizing trend prediction based on feature modeling Download PDFInfo
- Publication number
- CN110135647A CN110135647A CN201910423817.2A CN201910423817A CN110135647A CN 110135647 A CN110135647 A CN 110135647A CN 201910423817 A CN201910423817 A CN 201910423817A CN 110135647 A CN110135647 A CN 110135647A
- Authority
- CN
- China
- Prior art keywords
- information
- prediction result
- model
- prediction
- model information
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- 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
- G06Q10/00—Administration; Management
- G06Q10/20—Administration of product repair or maintenance
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Physics & Mathematics (AREA)
- Software Systems (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Marketing (AREA)
- Entrepreneurship & Innovation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention provides a kind of control methods that trend prediction is realized based on feature modeling, it is based on dynamic acquisition data, summarizes rule modeling, and be applied to external data to realize trend prediction in modeling, include the following steps: that the date determines multiple contextual informations at one or more dates to a. based on one or more；B. each contextual information situational model corresponding with the contextual information is matched one by one, one or more prediction results under each situational model state are obtained, the prediction result includes at least a core-prediction result and multiple auxiliary prediction results；C. final prediction result is determined according to a core-prediction result and multiple auxiliary prediction results.The present invention is easy to use, high-efficient, forecasting accuracy is high, has high commercial value.
Description
Technical field
The invention belongs to software technical architecture fields, are related to a kind of trend prediction mould based on machine learning techniques research and development
Type, especially a kind of control method and control device that trend prediction is realized based on feature modeling.
Background technique
Maintenance work is had been manually done by operation maintenance personnel, with Internet service Rapid Expansion, human cost
The epoch of height enterprise, people's meat O&M are difficult to maintain day-to-day operation.Then pass through the foot of creation can be automatically triggered predefined rule
This realizes automation O&M, it is considered to be one kind is based on industry field knowledge to execute common, repeated maintenance work
With the expert system of O&M scenarios domain knowledge.It is flat as core building O&M automation using layout operation mode to automate O&M
Platform can provide the functions such as configuration change, task inspection, script execution control, Custom Workflow for client；Cover inspection, text
The O&M scenarios such as part distribution, Backup and Restore, SQL operation, and expandability is provided；Automate the configuration of operational system server end
Batch job strategically dispatches each script work, realizes the automatic operation of Business Stream；Console is the operation of each administrative staff
Platform, realize role, object, operation classification decentralized management.Unitized device model is constructed on the basis of CMDB, realizes resource
Automation with service is delivered.
IT O&M is developed so far from birth, undergoes artificial O&M to current automation O&M, has been able to gradually adapt to
The business of complexity, diversified user demand, the IT application constantly extended, to ensure that IT services flexibly convenient, safety and stability, from
Dynamicization O&M is the demand of starting point as manual operation is replaced, and is just widely studied and applied.
O&M automation is one group and converts static device structure to the plan responded according to IT demand for services dynamic elasticity
Slightly, purpose is exactly to realize the quality of IT O&M, reduces cost.Automation O&M is mainly being automatically triggered, predetermined to create
The script of adopted rule realizes automation O&M to execute common, repeated maintenance work.But once occur not pre-
The problem of definition, it is necessary to artificially go perception, investigation, positioning, the manpower that this O&M service needs to put into is quite huge.Because
Shortage of manpower is positioned because checking, and be will lead to service and is hung up for a long time, cannot achieve the stable operation of system, this deficiency is increasingly
It highlights.For example, due to estimating deficiency to portfolio, staff response causes money not in time at business activity or special time point
Problem is deployed in source, causes service operation abnormal
And currently, there is no a kind of specific way that can effectively solve the problem that the above problem more particularly to a kind of bases in the market
The control method and control device of trend prediction are realized in feature modeling.
Summary of the invention
For technological deficiency of the existing technology, the object of the present invention is to provide one kind to realize trend based on feature modeling
The control method and control device of prediction based on dynamic acquisition data, summarize rule modeling, and external data are applied to build
Trend prediction is realized in mould, is included the following steps:
A. the date determines multiple contextual informations at one or more dates based on one or more；
B. each contextual information situational model corresponding with the contextual information is matched one by one, is obtained every
One or more prediction results under a situational model state, the prediction result include at least a core-prediction result and
Multiple auxiliary prediction results；
C. final prediction result is determined according to a core-prediction result and multiple auxiliary prediction results.
Preferably, further include step d after the step b: the final prediction result is subjected to Dynamic Display.
Preferably, in the step a, the contextual information is included at least:
XGBoost basic model information；
Festivals or holidays model information；
Week model information；And
Marketing activity information.
Preferably, the step b includes the following steps:
B1. based on the determining core to match with the XGBoost basic model information of the XGBoost basic model information
Heart prediction result；
B2. it is predicted based on determining the first auxiliary to match with the festivals or holidays model information of the festivals or holidays model information
As a result；
B3. based on determining the second auxiliary prediction knot to match with the week model information of the week model information
Fruit；
B4. it is predicted based on the determining third auxiliary to match with the marketing activity model information of the marketing activity information
As a result.
Preferably, the step c includes c1: based on the first auxiliary prediction result, the second auxiliary prediction result
And the third auxiliary prediction result is adjusted the core-prediction result, determines final prediction result.
Preferably, the foundation of the situational model will obtain in the following way:
I: one or more originals are obtained in one or more data acquisition platforms based on one or more data acquisition modes
Beginning external information；
Ii: pre-processing one or more of pristine outside information, determines the external letter of one or more pretreatments
Breath；
Iii: carrying out feature extraction to one or more of pretreatment external informations, determines one or more situation feature
Information；
Iv: being modeled based on one or more of contextual informations, determines one or more and the situation characteristic information phase
Matched situational model.
Preferably, the data acquisition modes include at least such as any one of under type or appoint a variety of:
SQL export；
Third party API export；Or
Crawler acquisition.
Preferably, the data acquisition platform includes at least such as any one of lower platform or appoints a variety of:
O&M worksheet processing system；
Configuration management database；
Application performance monitoring；
Banner system；Or
Unified log platform.
Preferably, the pristine outside information includes at least any one of following information or appoints a variety of:
Historical information；
Kilometer, lunar calendar information；
Holiday information；Or
Marketing activity data.
Preferably, the pretreatment includes attachment acquisition, data screening and data encoding.
According to another aspect of the present invention, a kind of control device that trend prediction is realized based on feature modeling is provided,
Include:
First determining device 1: the date determines multiple feelings at one or more dates based on one or more
Border information；
First acquisition device 2: one by one by each contextual information situational model corresponding with the contextual information into
Row matching, obtains one or more prediction results under each situational model state；
Second determining device 3: final prediction knot is determined according to a core-prediction result and multiple auxiliary prediction results
Fruit.
First processing unit 4: the final prediction result is subjected to Dynamic Display.
Preferably, first acquisition device 2 includes:
Third determining device 21: believe based on the XGBoost basic model information is determining with the XGBoost basic model
The matched core-prediction result of manner of breathing；
4th determining device 22: matched based on festivals or holidays model information determination with the festivals or holidays model information
First auxiliary prediction result；
5th determining device 23: based on determining second to match with the week model information of the week model information
Assist prediction result；
6th determining device 24: matched based on marketing activity information determination with the marketing activity model information
Third assists prediction result.
Preferably, further includes:
Second acquisition device 5: it is obtained based on one or more data acquisition modes in one or more data acquisition platforms
One or more pristine outside information；
7th determining device 6: pre-processing one or more of pristine outside information, determines one or more pre-
Handle external information；
8th determining device 7: feature extraction is carried out to one or more of pretreatment external informations, determines one or more
A situation characteristic information；
9th determining device 8: being modeled based on one or more of contextual informations, determines the one or more and situation
The situational model that characteristic information matches.
The present invention provides one kind independent of artificial specified rule, by machine learning algorithm automatically from magnanimity business number
Constantly learn in, constantly refine and summarize rule, forms trend prediction.It is to increase on the basis of automating O&M
Data needed for having added the brain based on machine learning, Command Monitoring System to acquire brain decision, it is pre- to make business datum
It surveys, and container platform is commanded to complete the scalable decision movement of elasticity, to reach the overall goals of operational system.In conjunction with big data
Visualization technique provides future anticipation operation data, makes O&M group to business side and technical side's real-time exhibition service operation situation
Team carries out capacity management and performance management in advance, as the central brain of container cloud platform, commands associated vessel management platform complete
At the decision movement that elasticity is scalable, the present invention is based on more at determining one or more dates on one or more dates
A contextual information one by one matches each contextual information situational model corresponding with the contextual information, obtains
One or more prediction results under each situational model state, the prediction result include at least core-prediction result with
And multiple auxiliary prediction results, final prediction result is determined according to a core-prediction result and multiple auxiliary prediction results,
The present invention is easy to use, high-efficient, forecasting accuracy is high, has high commercial value.
Detailed description of the invention
Upon reading the detailed description of non-limiting embodiments with reference to the following drawings, other feature of the invention,
Objects and advantages will become more apparent upon:
Fig. 1 shows a specific embodiment of the invention, a kind of controlling party for realizing trend prediction based on feature modeling
The concrete structure schematic diagram of method；
Fig. 2 shows the first embodiment of the present invention, one by one by each contextual information and the contextual information phase
Corresponding situational model is matched, and the detailed process for obtaining one or more prediction results under each situational model state is shown
It is intended to；
Fig. 3 shows the second embodiment of the present invention, establishes the idiographic flow schematic diagram of the situational model；
Fig. 4 shows another embodiment of the present invention, a kind of control for realizing trend prediction based on feature modeling
The module connection diagram of device processed；And
Fig. 5 shows the third embodiment of the present invention, a kind of control device for realizing trend prediction based on feature modeling
Module connection diagram.
Specific embodiment
In order to preferably technical solution of the present invention be made clearly to show, the present invention is made into one with reference to the accompanying drawing
Walk explanation.
Fig. 1 shows a specific embodiment of the invention, a kind of controlling party for realizing trend prediction based on feature modeling
The concrete structure schematic diagram of method specifically based on dynamic acquisition data, summarizes rule modeling, and external data is applied to
Trend prediction is realized in modeling, is included the following steps:
Firstly, entering step S101, the date determines more at one or more dates based on one or more
A contextual information, in such embodiments, the contextual information include at least XGBoost basic model information, festivals or holidays mould
Type information, week model information and marketing activity information.It will be appreciated by those skilled in the art that the present invention is selected to characteristic
When, judged according to history service data, such as portfolio has apparent periodicity, including week, Month And Year period；The Spring Festival, state
There are apparent influence in celebrating, small long holidays on portfolio, and influence degree has notable difference；Total ripple is more steady, and business activity is led
The peak valley of cause is unobvious.Based on the above, we have selected following feature: 1. calendar months, day；2. week 3. whether Spring Festival long holidays
Some day 4. whether some day of some day 5. whether small long holidays of long holidays on National Day, i.e., here, we can choose independent one day
Date determine contextual information as inputting, also can choose period as input, when at the same time it can also select multiple
Between section and multiple moment as inputting, and we can determine situation corresponding with the date according to these dates among the above
Information.
Then, S102 is entered step, one by one by each contextual information situation mould corresponding with the contextual information
Type is matched, and one or more prediction results under each situational model state are obtained, and the prediction result includes at least one
A core-prediction result and multiple auxiliary prediction results in such embodiments will be each corresponding with the date
Contextual information is matched with situational model, can obtain one or more prediction results under each situational model state.Institute
State prediction result, the prediction result of festivals or holidays model information, week that prediction result includes at least XGBoost basic model information
One or more of prediction result and the prediction result of marketing activity information of model information, the prediction result is at least wrapped
A core-prediction result and multiple auxiliary prediction results are included, and in the present embodiment, the XGBoost basic model information
Prediction result be the core-prediction as a result, and the prediction of the prediction result, week model information of festivals or holidays model information
As a result and the prediction result of marketing activity information is the auxiliary prediction result.
Subsequently, S103 is entered step, is determined according to a core-prediction result and multiple auxiliary prediction results final
Prediction result, in such embodiments, the multiple auxiliary prediction result determine that final prediction result can be converted to centainly
Coefficient carrys out the calibration to the core-prediction result, can also be according to the relationship of weight proportion on the basis of core-prediction result
It carries out operation and finally obtains final prediction result.
Finally, entering step S104, the final prediction result is subjected to Dynamic Display, it will be appreciated by those skilled in the art that
The final prediction result will show as forms such as table, images, i.e., in such embodiments, we can set
The designated date is inputted in fixed intelligent operation terminal or the period shows final prediction result by operation, and is opened up
The result shown may constantly change according to the data continually entered or the historical data constantly added, and make described final pre-
Survey the picture that result is in a kind of Dynamic Display.Intelligent O&M is advocated independent of artificial specified rule by machine learning algorithm
Automatically constantly learn from magnanimity operation/maintenance data (the artificial treatment log including event itself and operation maintenance personnel), constantly
It refines and summarizes rule in ground.It is to increase the brain based on machine learning, Zhi Huijian on the basis of automating O&M
Data needed for examining system acquires brain decision make analysis, decision, and command automation script goes to execute the decision of brain,
To reach the overall goals of operational system.
Fig. 2 shows the first embodiment of the present invention, one by one by each contextual information and the contextual information phase
Corresponding situational model is matched, and the detailed process for obtaining one or more prediction results under each situational model state is shown
It is intended to, it will be appreciated by those skilled in the art that described Fig. 2 is the deployment step of the step S102, more specifically, including it is as follows
Step:
Firstly, S1021 is entered step, based on XGBoost basic model information determination and the basis XGBoost mould
The core-prediction that type information matches is as a result, in such embodiments, the foundation of the XGBoost basic model will be described below
Specific embodiment in be further described through, and the XGBoost basic model information is updated to the XGBoost base
In plinth model, matching operation is carried out, finally determines the core-prediction result.
Then, S1022 is entered step, based on festivals or holidays model information determination and the festivals or holidays model information phase
Match first auxiliary prediction result, it is described first auxiliary prediction result be obtained based on the festivals or holidays model as a result,
And it is further described through in the specific embodiment that the foundation of the festivals or holidays model will be described below.
Subsequently, S1023 is entered step, is determined based on the week model information and is matched with the week model information
Second auxiliary prediction result, it is described second auxiliary prediction result be obtained based on the week model as a result, and institute
It states and is further described through in the specific embodiment that the foundation of week model will be described below.
Finally, S1024 is entered step, based on marketing activity information determination and the marketing activity model information phase
The third auxiliary prediction result matched, the third auxiliary prediction result is the knot obtained based on the marketing activity model
Fruit, and be further described through in the specific embodiment that the foundation of the marketing activity model will be described below.
It is preferably based on the first auxiliary prediction result, the second auxiliary prediction result and third auxiliary
Prediction result is adjusted the core-prediction result, determines final prediction result.It will be appreciated by those skilled in the art that industry political affairs
The business activity of plan, insurance company has a significant impact to portfolio, and the collection of corresponding data is relatively difficult.But produce the danger opposite longevity
Danger, by it is such influence it is much smaller.The vehicle insurance business of Pacific Ocean insurance is relatively more steady always, and marketing activity is also few, and reports a case to the security authorities, winds up the case
Mainly influenced by client activities, it is little with enterprises end factor relation.When producing the danger amount of winding up the case modeling, policy, marketing are put aside
Etc. factors.In arameter optimization, that is, how based on it is described first auxiliary prediction result, it is described second auxiliary prediction result and
The third auxiliary prediction result is adjusted the core-prediction result, and when determining final prediction result, this research is used
The arameter optimization of Xgboost model refer mainly to booster parameter, comprising:
First: eta [default 0.3] is similar with the learning rate parameter in GBM, by the power for reducing each step
Weight, can be improved the robustness of model.Wherein representative value is 0.01-0.2.
Second: min_child_weight [default 1], determine minimum leaf node sample weights and.With the min_ of GBM
Child_leaf parameter is similar, but not exclusively the same.This parameter of XGBoost be smallest sample weight and, and GBM parameter
It is smallest sample sum.This parameter is for avoiding over-fitting.It, can be to avoid model learning to part when its value is larger
Special sample.But if this value is excessively high, it will lead to poor fitting.This parameter needs to adjust using CV.
Third: the parameter in max_depth [default 6] and GBM is identical, this value is the depth capacity of tree.This value
It is for avoiding over-fitting.Max_depth is bigger, and model can acquire more specific more local sample.It needs using CV function
To carry out tuning.Wherein, representative value: 3-10.
Four: max_leaf_nodes, the quantity of maximum node or leaf on tree.The work of max_depth can be substituted
With.Because if what is generated is binary tree, the tree that a depth is n at most generates n2 leaf.If defining this parameter,
GBM can ignore max_depth parameter.
In node split, the value of loss function has dropped five: gamma [default 0] after only dividing, and can just divide this
A node.Least disadvantage function drop-out value needed for Gamma specifies node split.The value of this parameter is bigger, and algorithm is more protected
It keeps.The value and loss function of this parameter are closely bound up, so needing to adjust.
Six: max_delta_step [default 0], this parameter limits the maximum step-length of each tree weight changes.If this
The value of a parameter is 0, that means that and does not constrain.If it has been assigned some positive value, it can allow this algorithm more
Add conservative.In general, this parameter does not need to be arranged.But when sample of all categories is very uneven, it is to logistic regression
Helpful.This parameter generally takes less than, but you can excavate its more use.
Subsample parameter in seven: subsample [default 1] and GBM is the same.This state modulator for
Each tree, the ratio of stochastical sampling.Reduce the value of this parameter, algorithm can be guarded more, and over-fitting is avoided.But if this
A value is arranged too small, it may result in poor fitting.Representative value: 0.5-1.
Eight: colsample_bytree [default 1] is similar with the max_features parameter inside GBM.For controlling
The accounting of the columns of every stochastical sampling (each column are a features).Representative value: 0.5-1.
Nine: colsample_bylevel [default 1], for control tree every level-one division each time, to columns
The accounting of sampling.Subsample parameter and colsample_bytree parameter can serve the same role.
Ten: lambda [default 1], the L2 regularization term of weight.(similar with Ridge regression).This parameter
It is the regularization part for controlling XGBoost.Although most of data science man seldom uses this parameter, this
Parameter is on reducing over-fitting or can excavate more use.
11: alpha [default 1], the L1 regularization term of weight.(similar with Lasso regression).It can answer
In the case where very high-dimensional, so that the speed of algorithm is faster.
12: scale_pos_weight [default 1], when sample of all categories is very uneven, sets this parameter
It is set to a positive value, algorithm more rapid convergence can be made.
Further, model integrated, after generating basic model according to xgboost algorithm, the model is to Spring Festival, National Day, small
The prediction of long holidays, week isotype, trend are in the main true, and further according to historical data, establish week model, Spring Festival model, state
Model, small long holidays model are celebrated, does further adjustment for the output to basic model.
Fig. 3 shows the second embodiment of the present invention, establishes the idiographic flow schematic diagram of the situational model, the feelings
The foundation of border model will obtain in the following way:
Firstly, S201 is entered step, based on one or more data acquisition modes in one or more data acquisition platforms
Obtain one or more pristine outside information, the data acquisition modes include at least SQL export, third party API export or
Crawler acquisition, the data acquisition platform include at least O&M worksheet processing system, configuration management database, application performance monitoring, see
Plate system or unified log platform.The pristine outside information includes at least historical information, kilometer, lunar calendar information, festivals or holidays
Information or marketing activity data.
Then, S202 is entered step, one or more of pristine outside information are pre-processed, determines one or more
A pretreatment external information, it is preferable that the pretreatment includes attachment acquisition, data screening and data encoding.Partial data is put
It is important although these data quantity are few in attachment.This system downloads attachment from system in a manner of crawler
Information is added into relative program.
Subsequently, S203 is entered step, feature extraction is carried out to one or more of pretreatment external informations, determines one
A or multiple situation characteristic informations, the present invention select feature in such a way that professional knowledge is combined with data mining.It is logical first
The mode for visiting business personnel is crossed, sample cases, and possible relative feature is collected, then calculates each feature and phase
The correlation of dependent variable therefrom screens suitable candidate feature.Pearson correlation coefficients and mutual trust have mainly been used in the present invention
Coefficient is ceased to measure univariate correlation.After obtaining candidate feature, according to these feature training patterns, according to the correct of model
The conspicuousness of rate and recall rate assessment feature.
Finally, enter step S204, modeled based on one or more of contextual informations, determine it is one or more with it is described
The situational model that situation characteristic information matches.It will be appreciated by those skilled in the art that the situational model includes the basis XGBoost mould
Type, festivals or holidays model information, week model information and marketing activity model information.Based on abovementioned steps S201 to step
S203 determines one or more situational models to match with the situation characteristic information.
Fig. 4 shows another embodiment of the present invention, a kind of control for realizing trend prediction based on feature modeling
The module connection diagram of device processed, including the first determining device 1, the date determines described in one or more based on one or more
The working principle of multiple contextual informations at date, first determining device 1 can refer to abovementioned steps S101, herein
It will not go into details.
It further, further include the first acquisition device 2: one by one that each contextual information is opposite with the contextual information
The situational model answered is matched, and one or more prediction results under each situational model state are obtained, and described first obtains
The working principle of device 2 can refer to abovementioned steps S102, and it will not be described here.
It further, further include the second determining device 3: according to a core-prediction result and multiple auxiliary prediction results
Determine final prediction result, the working principle of second determining device 3 can refer to abovementioned steps S103, not superfluous herein
It states.
Further, further include the first processing unit 4: the final prediction result is subjected to Dynamic Display, described first
The working principle of processing unit 4 can refer to abovementioned steps S104, and it will not be described here.
Further, first acquisition device 2 includes third determining device 21, is based on the XGBoost basic model
The determining core-prediction to match with the XGBoost basic model information of information is as a result, the work of third determining device 21 is former
Reason can refer to abovementioned steps S1021, and it will not be described here.
Further, first acquisition device 2 further includes the 4th determining device 22, is based on the festivals or holidays model information
Determining the first auxiliary prediction result to match with the festivals or holidays model information, the working principle of the 4th determining device 22 can be with
With reference to abovementioned steps S1022, it will not be described here.
Further, first acquisition device 2 further includes the 5th determining device 23, true based on the week model information
Fixed the second auxiliary prediction result to match with the week model information, the working principle of the 5th determining device 23 can be with
With reference to abovementioned steps S1023, it will not be described here.
Further, first acquisition device 2 further includes the 6th determining device 24, true based on the marketing activity information
The fixed third auxiliary prediction result to match with the marketing activity information, the working principle of the 6th determining device 24 can be with
With reference to abovementioned steps S1024, it will not be described here.
Fig. 5 shows the third embodiment of the present invention, a kind of control device for realizing trend prediction based on feature modeling
Module connection diagram, further include the second acquisition device 5: based on one or more data acquisition modes in one or more numbers
One or more pristine outside information are obtained according to acquisition platform, the working principle of the second acquisition device 5 can refer to abovementioned steps
S201, it will not be described here.
Further, further include the 7th determining device 6: one or more of pristine outside information pre-processed,
Determine that one or more pretreatment external informations, the working principle of the 7th determining device 6 can refer to abovementioned steps S202, herein
It will not go into details.
Further, further include the 8th determining device 7: feature being carried out to one or more of pretreatment external informations and is mentioned
It takes, determines one or more situation characteristic information；The working principle of 8th determining device 7 can refer to abovementioned steps S203,
It will not go into details for this.
Further, further include the 9th determining device 8: being modeled based on one or more of contextual informations, determine one
Or multiple situational models to match with the situation characteristic information, the working principle of the 9th determining device 8 can refer to
Abovementioned steps S204, it will not be described here.
Specific embodiments of the present invention are described above.It is to be appreciated that the invention is not limited to above-mentioned
Particular implementation, those skilled in the art can make various deformations or amendments within the scope of the claims, this not shadow
Ring substantive content of the invention.
Claims (13)
1. a kind of control method that trend prediction is realized based on feature modeling based on dynamic acquisition data, summarizes rule modeling,
And external data is applied to realize trend prediction in modeling, which comprises the steps of:
A. the date determines multiple contextual informations at one or more dates based on one or more；
B. each contextual information situational model corresponding with the contextual information is matched one by one, obtains each feelings
One or more prediction results under the model state of border, the prediction result include at least a core-prediction result and multiple
Assist prediction result；
C. final prediction result is determined according to a core-prediction result and multiple auxiliary prediction results.
2. control method according to claim 1, which is characterized in that further include step d after the step b: by institute
It states final prediction result and carries out Dynamic Display.
3. control method according to claim 1 or 2, which is characterized in that in the step a, the contextual information is extremely
Include: less
XGBoost basic model information；
Festivals or holidays model information；
Week model information；And
Marketing activity model information.
4. control method according to claim 3, which is characterized in that the step b includes the following steps:
B1. pre- based on the determining core to match with the XGBoost basic model information of the XGBoost basic model information
Survey result；
B2. based on determining the first auxiliary prediction knot to match with the festivals or holidays model information of the festivals or holidays model information
Fruit；
B3. based on determining the second auxiliary prediction result to match with the week model information of the week model information；
B4. based on the determining third auxiliary prediction knot to match with the marketing activity model information of the marketing activity information
Fruit.
5. control method according to claim 4, which is characterized in that the step c includes c1: based on first auxiliary
Prediction result, the second auxiliary prediction result and third auxiliary prediction result adjust the core-prediction result
It is whole, determine final prediction result.
6. control method according to claim 5, which is characterized in that the foundation of the situational model will in the following way
It obtains:
I: one or more original outer in the acquisition of one or more data acquisition platforms based on one or more data acquisition modes
Portion's information；
Ii: pre-processing one or more of pristine outside information, determines one or more pretreatment external informations；
Iii: carrying out feature extraction to one or more of pretreatment external informations, determines one or more situation feature letter
Breath；
Iv: being modeled based on one or more of contextual informations, determines that one or more matches with the situation characteristic information
Situational model.
7. control method according to claim 6, which is characterized in that the data acquisition modes include at least such as under type
Any one of or appoint it is a variety of:
SQL export；
Third party API export；Or
Crawler acquisition.
8. control method according to claim 6, which is characterized in that the data acquisition platform includes at least such as lower platform
Any one of or appoint it is a variety of:
O&M worksheet processing system；
Configuration management database；
Application performance monitoring；
Banner system；Or
Unified log platform.
9. control method according to claim 6, which is characterized in that the pristine outside information includes at least following information
Any one of or appoint it is a variety of:
Historical information；
Kilometer, lunar calendar information；
Holiday information；Or
Marketing activity data.
10. control method according to claim 6, which is characterized in that the pretreatment includes attachment acquisition, data screening
And data encoding.
11. a kind of control device for realizing trend prediction based on feature modeling characterized by comprising
First determining device (1): the date determines multiple situations at one or more dates based on one or more
Information；
First acquisition device (2): each contextual information situational model corresponding with the contextual information is carried out one by one
Matching, obtains one or more prediction results under each situational model state；
Second determining device (3): final prediction result is determined according to a core-prediction result and multiple auxiliary prediction results.
First processing unit (4): the final prediction result is subjected to Dynamic Display.
12. control device according to claim 11, which is characterized in that first acquisition device (2) includes:
Third determining device (21): based on XGBoost basic model information determination and the XGBoost basic model information
The core-prediction result to match；
4th determining device (22): based on determining the to match with the festivals or holidays model information of the festivals or holidays model information
One auxiliary prediction result；
5th determining device (23): based on the week model information it is determining match with the week model information it is second auxiliary
Help prediction result；
6th determining device (24): based on determining the to match with the marketing activity model information of the marketing activity information
Three auxiliary prediction results.
13. control device according to claim 11, which is characterized in that further include:
Second acquisition device (5): one is obtained in one or more data acquisition platforms based on one or more data acquisition modes
A or multiple pristine outside information；
7th determining device (6): pre-processing one or more of pristine outside information, determines one or more pre- places
Manage external information；
8th determining device (7): carrying out feature extraction to one or more of pretreatment external informations, determines one or more
Situation characteristic information；
9th determining device (8): being modeled based on one or more of contextual informations, is determined one or more special with the situation
The matched situational model of reference manner of breathing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910423817.2A CN110135647A (en) | 2019-05-21 | 2019-05-21 | A kind of control method and control device for realizing trend prediction based on feature modeling |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910423817.2A CN110135647A (en) | 2019-05-21 | 2019-05-21 | A kind of control method and control device for realizing trend prediction based on feature modeling |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110135647A true CN110135647A (en) | 2019-08-16 |
Family
ID=67572208
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910423817.2A Pending CN110135647A (en) | 2019-05-21 | 2019-05-21 | A kind of control method and control device for realizing trend prediction based on feature modeling |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110135647A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111652427A (en) * | 2020-05-29 | 2020-09-11 | 航科院中宇（北京）新技术发展有限公司 | Flight arrival time prediction method and system based on data mining analysis |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101188002A (en) * | 2007-12-24 | 2008-05-28 | 北京大学 | A city traffic dynamic prediction system and method with real time and continuous feature |
| US20130238761A1 (en) * | 2012-03-10 | 2013-09-12 | Headwater Partners Ii Llc | Distributing content by generating and preloading queues of content |
| US20160189210A1 (en) * | 2010-03-07 | 2016-06-30 | Sailthru, Inc. | System and method for appying data modeling to improve predictive outcomes |
| CN109657831A (en) * | 2017-10-11 | 2019-04-19 | 顺丰科技有限公司 | A kind of Traffic prediction method, apparatus, equipment, storage medium |
| CN109767255A (en) * | 2018-12-06 | 2019-05-17 | 东莞团贷网互联网科技服务有限公司 | A method of it is modeled by big data and realizes intelligence operation and precision marketing |
-
2019
- 2019-05-21 CN CN201910423817.2A patent/CN110135647A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101188002A (en) * | 2007-12-24 | 2008-05-28 | 北京大学 | A city traffic dynamic prediction system and method with real time and continuous feature |
| US20160189210A1 (en) * | 2010-03-07 | 2016-06-30 | Sailthru, Inc. | System and method for appying data modeling to improve predictive outcomes |
| US20130238761A1 (en) * | 2012-03-10 | 2013-09-12 | Headwater Partners Ii Llc | Distributing content by generating and preloading queues of content |
| CN109657831A (en) * | 2017-10-11 | 2019-04-19 | 顺丰科技有限公司 | A kind of Traffic prediction method, apparatus, equipment, storage medium |
| CN109767255A (en) * | 2018-12-06 | 2019-05-17 | 东莞团贷网互联网科技服务有限公司 | A method of it is modeled by big data and realizes intelligence operation and precision marketing |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111652427A (en) * | 2020-05-29 | 2020-09-11 | 航科院中宇（北京）新技术发展有限公司 | Flight arrival time prediction method and system based on data mining analysis |
| CN111652427B (en) * | 2020-05-29 | 2023-12-29 | 航科院中宇（北京）新技术发展有限公司 | Flight arrival time prediction method and system based on data mining analysis |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111680025B (en) | Method and system for intelligently assimilating space-time information of multi-source heterogeneous data oriented to natural resources | |
| US7869984B2 (en) | System and method for simulating a discrete event process using business system data | |
| US20070288212A1 (en) | System And Method For Optimizing Simulation Of A Discrete Event Process Using Business System Data | |
| US8195709B2 (en) | Comparison of models of a complex system | |
| US6952688B1 (en) | Knowledge-engineering protocol-suite | |
| US8027859B2 (en) | Analysis of a model of a complex system, based on two models of the system, wherein the two models represent the system with different degrees of detail | |
| WO2014151455A1 (en) | Modeling and simulation | |
| WO2007005975A2 (en) | Risk modeling system | |
| WO2006049922A2 (en) | Method and system for sequencing and scheduling | |
| US7941301B2 (en) | Modelling a complex system | |
| CN106775752A (en) | Forestry informationlization platform construction method and Forestry informationlization platform based on Business Process Modeling | |
| CN110196711A (en) | Plug-in type work order aid decision-making method and its system based on artificial intelligence | |
| CN112614011A (en) | Power distribution network material demand prediction method and device, storage medium and electronic equipment | |
| CN110175272A (en) | One kind realizing the convergent control method of work order and control device based on feature modeling | |
| CN116861189A (en) | Method for constructing equipment fault diagnosis maintenance knowledge base based on large language model | |
| CN113919558A (en) | Product sales prediction method and device | |
| CN110135647A (en) | A kind of control method and control device for realizing trend prediction based on feature modeling | |
| CN115422402A (en) | Engineering prediction analysis method | |
| KR102354181B1 (en) | A construction information management system for visualising data and a method for controlling the same | |
| US20200342302A1 (en) | Cognitive forecasting | |
| US20220391807A1 (en) | Systems and methods for predicting and managing tool assets | |
| CN112346727B (en) | Method for uniformly modeling discrete random problem and continuous problem in production system | |
| Borky et al. | Analyzing Requirements in an Operational Viewpoint | |
| Yu | Enterprise Human Resource Management Model by Artificial Intelligence Digital Technology | |
| Zou et al. | An Efficient Hierarchical Representation Approach of Remote Sensing Application Modeling Based on Distributed Environment |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190816 |