CN107688872A - Forecast model establishes device, method and computer-readable recording medium - Google Patents
Forecast model establishes device, method and computer-readable recording medium Download PDFInfo
- Publication number
- CN107688872A CN107688872A CN201710715445.1A CN201710715445A CN107688872A CN 107688872 A CN107688872 A CN 107688872A CN 201710715445 A CN201710715445 A CN 201710715445A CN 107688872 A CN107688872 A CN 107688872A
- Authority
- CN
- China
- Prior art keywords
- feature
- forecast
- threshold
- forecast model
- searchable index
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0242—Determining effectiveness of advertisements
Abstract
The invention discloses a kind of forecast model to establish device, including:Memory, processor and storage are on a memory and the forecast model that can run on a processor establishes program, and the program realizes following steps when being executed by processor:Obtain external source feature of the project to be measured in one or more time quantums before object time unit, and the autoregression temporal characteristics before object time unit;Features described above is pre-processed and normalized, to obtain normalized feature set;Feature Selection is carried out to feature according to preset rules and obtains predicted characteristics;Using the actual observed value of project to be measured as prediction target, using predicted characteristics and prediction target as forecast sample;Multiple forecast samples are obtained according to above step, forecast sample is inputted into default regression model is trained generation forecast model.The present invention also proposes a kind of forecast model method for building up and a kind of computer-readable recording medium.The present invention improves the predictablity rate of forecast model.
Description
Technical field
The present invention relates to field of terminal technology, more particularly to a kind of forecast model establishes device, method and computer-readable
Storage medium.
Background technology
At present, it is applied by the technology that machine learning is predicted to data in increasing field, such as extensively
Accuse the prediction of clicking rate, the prediction of the incidence of disease of certain epidemic disease etc., at present generally by the way of when, gather these
The historical data of project to be predicted forms time series, and the feature based on this time series in itself establishes auto-correlation regression time
Series model (ARIMA) is predicted, but the model is only used the project to be predicted trend feature of itself and is predicted, Wu Fajie
Close prediction surface to use, cause forecasting accuracy not high.
The content of the invention
The present invention provides a kind of forecast model and establishes device, method and computer-readable recording medium, and its main purpose exists
Forecast model is established with autoregression temporal characteristics in the external source feature for combining project to be measured, the prediction for improving forecast model is accurate
Degree.
To achieve the above object, the present invention provides a kind of forecast model and establishes device, and the device includes:Memory, processing
Device and it is stored in the forecast model that can be run on the memory and on the processor and establishes program, the forecast model is built
Following steps are realized when vertical program is by the computing device:
A, external source feature of the project to be measured in one or more time quantums before object time unit is obtained, and
Autoregression temporal characteristics before object time unit;
B, the external source feature is pre-processed, and to the autoregression temporal characteristics and by the outer of the pretreatment
Source feature is normalized, to obtain normalized feature set;
C, Feature Selection is carried out to the feature in the feature set according to preset rules, to obtain predicted characteristics;
D, will be described more using actual observed value of the project to be measured in the object time unit as prediction target
Individual predicted characteristics and the prediction target are as a forecast sample;
E, multiple forecast samples of multiple time quantums are obtained respectively according to the step of A to D, by the multiple prediction
Sample is input in default regression model and is trained to determine model parameter, by the default recurrence after determination model parameter
Forecast model of the model as the project to be measured.
Alternatively, if the external source feature includes searchable index collection corresponding to the project to be measured, the pretreatment is pole
Be worth optimization processing, then described the step of being pre-processed to the external source feature include:
Obtain first quartile, the 3rd quartile and the quartile deviation of the searchable index collection;
The first threshold of searchable index is determined according to the first quartile, the 3rd quartile and the quartile deviation
Value and Second Threshold, the first threshold are less than the Second Threshold;
The searchable index more than the Second Threshold is concentrated to be converted to the Second Threshold searchable index, by described in
Searchable index concentrates the searchable index less than the first threshold to be converted to the first threshold
Alternatively, described the step of being pre-processed to the external source feature, also includes:
According to neighbouring kNN algorithms, missing values supplement is carried out to the searchable index by extremal optimization processing.
Alternatively, it is described that Feature Selection is carried out to the feature in the feature set according to preset rules, it is special to obtain prediction
The step of sign, includes:
Calculate the pearson coefficient correlations of each characteristic value in the feature set;
The feature that pearson coefficient correlations will be selected to be less than or equal to preset correlation coefficient number in the feature set, as
Predicted characteristics.
Alternatively, the project to be measured be influenza predict project, the external source feature include searchable index, weather characteristics and
Environmental characteristic, the default regression model are LASSO regression models.
In addition, to achieve the above object, the present invention also provides a kind of forecast model method for building up, and this method includes:
A, external source feature of the project to be measured in one or more time quantums before object time unit is obtained, and
Autoregression temporal characteristics before object time unit;
B, the external source feature is pre-processed, and to the autoregression temporal characteristics and by the outer of the pretreatment
Source feature is normalized, to obtain normalized feature set;
C, Feature Selection is carried out to the feature in the feature set according to preset rules, to obtain predicted characteristics;
D, will be described more using actual observed value of the project to be measured in the object time unit as prediction target
Individual predicted characteristics and the prediction target are as a forecast sample;
E, multiple forecast samples of multiple time quantums are obtained respectively according to the step of A to D, by the multiple prediction
Sample is input in default regression model and is trained to determine model parameter, by the default recurrence after determination model parameter
Forecast model of the model as the project to be measured.
Alternatively, if the external source feature includes searchable index collection corresponding to the project to be measured, the pretreatment is pole
Be worth optimization processing, then described the step of being pre-processed to the external source feature include:
Obtain first quartile, the 3rd quartile and the quartile deviation of the searchable index collection;
The first threshold of searchable index is determined according to the first quartile, the 3rd quartile and the quartile deviation
Value and Second Threshold, the first threshold are less than the Second Threshold;
The searchable index more than the Second Threshold is concentrated to be converted to the Second Threshold searchable index, by described in
Searchable index concentrates the searchable index less than the first threshold to be converted to the first threshold.
Alternatively, described the step of being pre-processed to the external source feature, also includes:
According to neighbouring kNN algorithms, missing values supplement is carried out to the searchable index by extremal optimization processing.
Alternatively, it is described that Feature Selection is carried out to the feature in the feature set according to preset rules, it is special to obtain prediction
The step of sign, includes:
Calculate the pearson coefficient correlations of each characteristic value in the feature set;
The feature that pearson coefficient correlations will be selected to be less than or equal to preset correlation coefficient number in the feature set, as
Predicted characteristics.
In addition, to achieve the above object, the present invention also provides a kind of computer-readable recording medium, described computer-readable
Forecast model is stored with storage medium and establishes program, the forecast model, which is established, realizes as above institute when program is executed by processor
The step of forecast model method for building up stated.
Forecast model proposed by the present invention establishes device, method and computer-readable recording medium, obtains project to be measured and exists
The external source feature in one or more time quantums before object time unit, and the autoregression before object time unit
Temporal characteristics, to external source feature and autoregression temporal characteristics are pre-processed and normalized, to obtain normalized feature
Collection, is screened to obtain predicted characteristics according to preset rules to the feature in feature set, the predicted characteristics and object time unit
Corresponding actual observed value forms a forecast sample, obtains multiple forecast samples of multiple time quantums as procedure described above,
Above-mentioned multiple forecast samples are input in default regression model and are trained to determine model parameter, after model parameter is determined
Default regression model the external source feature of project to be measured is entered with autoregression temporal characteristics as forecast model, the scheme of the invention
Row, which combines and carries out pretreatment, forms a feature set, and qualified feature is filtered out as predicted characteristics to returning from feature set
Return model to be trained generation forecast model, avoid the unicity of sample characteristics, improve the prediction precision of forecast model.
Brief description of the drawings
Fig. 1 is the schematic diagram that forecast model of the present invention establishes device preferred embodiment;
Fig. 2 is that forecast model of the present invention establishes the functional module signal that forecast model in the embodiment of device one establishes program
Figure;
Fig. 3 is the flow chart of forecast model method for building up first embodiment of the present invention.
The realization, functional characteristics and advantage of the object of the invention will be described further referring to the drawings in conjunction with the embodiments.
Embodiment
It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not intended to limit the present invention.
The present invention provides a kind of forecast model and establishes device.Shown in reference picture 1, for forecast model of the present invention establish device compared with
The schematic diagram of good embodiment.
In the present embodiment, it can be PC (Personal Computer, PC) that forecast model, which establishes device,
Can be smart mobile phone, tablet personal computer, E-book reader, MP3 (Moving Picture Experts Group Audio
Layer III, dynamic image expert's compression standard audio aspect 3) player, MP4 (Moving Picture Experts
Group Audio Layer IV, dynamic image expert's compression standard audio aspect 4) player, pocket computer etc. have it is aobvious
Show the packaged type terminal device of function.
The forecast model, which establishes device, includes memory 11, processor 12, communication bus 13, and network interface 14.
Wherein, memory 11 comprises at least a type of readable storage medium storing program for executing, the readable storage medium storing program for executing include flash memory,
Hard disk, multimedia card, card-type memory (for example, SD or DX memories etc.), magnetic storage, disk, CD etc..Memory 11
Can be the internal storage unit that forecast model establishes device in certain embodiments, such as the forecast model establishes the hard of device
Disk.Memory 11 can also be the External memory equipment that forecast model establishes device in further embodiments, such as predict mould
Type establishes the plug-in type hard disk being equipped with device, intelligent memory card (Smart Media Card, SMC), secure digital (Secure
Digital, SD) card, flash card (Flash Card) etc..Further, memory 11 can also both include forecast model foundation
The internal storage unit of device also includes External memory equipment.Memory 11 can be not only used for storage and be installed on forecast model building
The application software and Various types of data of vertical device, such as forecast model establish code of program etc., can be also used for temporarily storing
The data that has exported or will export.
Processor 12 can be in certain embodiments a central processing unit (Central Processing Unit,
CPU), controller, microcontroller, microprocessor or other data processing chips, for the program stored in run memory 11
Code or processing data, such as perform prediction model-builder program etc..
Communication bus 13 is used to realize the connection communication between these components.
Network interface 14 can optionally include wireline interface, the wave point (such as WI-FI interfaces) of standard, be generally used for
Communication connection is established between the device and other electronic equipments.
Fig. 1, which illustrate only, to be established the forecast model of program with component 11-14 and forecast model and establishes device, but should
What is understood is, it is not required that implements all components shown, the more or less component of the implementation that can be substituted.
Alternatively, the device can also include user interface, and user interface can include display (Display), input
Unit such as keyboard (Keyboard), optional user interface can also include wireline interface, the wave point of standard.It is optional
Ground, in certain embodiments, display can be light-emitting diode display, liquid crystal display, touch-control liquid crystal display and OLED
(Organic Light-Emitting Diode, Organic Light Emitting Diode) touches device etc..Wherein, what display can also be suitably
Referred to as display screen or display unit, visualized for the information for being shown in forecast model to establish to handle in device and for showing
User interface.
Alternatively, the device can also include touch sensor.What the touch sensor was provided is touched for user
The region for touching operation is referred to as touch area.In addition, touch sensor described here can be resistive touch sensor, electric capacity
Formula touch sensor etc..Moreover, the touch sensor not only includes the touch sensor of contact, proximity may also comprise
Touch sensor etc..In addition, the touch sensor can be single sensor, or such as multiple biographies of array arrangement
Sensor.The area of the display of the device can be identical with the area of the touch sensor, can also be different.Alternatively, will
Display is set with touch sensor stacking, to form touch display screen.The device is based on touch display screen detecting user
The touch control operation of triggering.
Alternatively, the device can also include camera, RF (Radio Frequency, radio frequency) circuit, sensor, sound
Frequency circuit, WiFi module etc..Wherein, sensor ratio such as optical sensor, motion sensor and other sensors.Specifically, light
Sensor may include ambient light sensor and proximity transducer, wherein, if the device is mobile terminal, ambient light sensor can root
The brightness of display screen is adjusted according to the light and shade of ambient light, proximity transducer can be closed aobvious when mobile terminal is moved in one's ear
Display screen and/or backlight.Certainly, mobile terminal can also configure gyroscope, barometer, hygrometer, thermometer, infrared ray sensor etc.
Other sensors, it will not be repeated here.
Forecast model, which is stored with, in the device embodiment shown in Fig. 1, in memory 11 establishes program;Processor 12 performs
The forecast model stored in memory 11 realizes following steps when establishing program:
A, external source feature of the project to be measured in one or more time quantums before object time unit is obtained, and
Autoregression temporal characteristics before object time unit;
B, the external source feature is pre-processed, and to the autoregression temporal characteristics and by the outer of the pretreatment
Source feature is normalized, to obtain normalized feature set;
C, Feature Selection is carried out to the feature in the feature set according to preset rules, to obtain predicted characteristics;
D, will be described more using actual observed value of the project to be measured in the object time unit as prediction target
Individual predicted characteristics and the prediction target are as a forecast sample;
E, multiple forecast samples of multiple time quantums are obtained respectively according to the step of A to D, by the multiple prediction
Sample is input in default regression model and is trained to determine model parameter, by the default recurrence after determination model parameter
Forecast model of the model as the project to be measured.
In this embodiment, the scheme of the present embodiment is illustrated as project to be measured using influenza prediction project.Due to
What is used when being trained to model is all historical data, i.e. data in the past period, therefore, from history
When acquisition characteristics value is as test sample in data, the time quantum in each test sample is also the past time
Section.In the present embodiment, with mono- time quantum of Zhou Zuowei.Assuming that by past 100 all external source features and influenza sample
Case percentage as historical data, wherein, influenza-like case percentage be somewhere each Sentinel point hospital influenza-like case sum/
Always medical person-time of each Sentinel point hospital outpatient in somewhere.
The process that a forecast sample is obtained from above-mentioned historical data is as follows:A week is determined from above-mentioned historical data
As object time unit, for example, with the 100th week in past 100 week, i.e., closest one with current point in time
Week, as object time unit, then using the actual observed value of the influenza-like case percentage of the 100th week as prediction target.From
Acquisition characteristics are as predicted characteristics in multiple all data before 100th week.
Specifically, external source feature corresponding with one or more time quantum before object time unit is obtained, it is excellent
Selection of land, external source feature include searchable index, weather characteristics and environmental characteristic etc. and cause certain journey to the percentage of influenza-like case
The surface of the influence of degree.
Searchable index is:In a time quantum before object time unit, it is therefore preferable to object time unit
A upper time quantum, search engine user search for the frequency of associative key on the search engine, and search engine typically can
The frequency is recorded, for example, Baidu's index of Baidu.Wherein, associative key be user pre-set with the influenza
Prediction there is certain causal keyword, such as " nasal obstruction ", " sneezing ", " antiviral oral liquor ", " having a throat-ache " etc.
The term related to influenza, will not enumerate herein, and user can be arranged as required to multiple keywords, and be searched from some
On rope website searchable index collection is formed according to searchable index corresponding to keyword extraction.
Weather characteristics include but is not limited to following feature:Weekly fine day number, weekly cloudy number of days, weekly cloudy number,
Weekly rainfall, weekly air quantity, it is average weekly among temperature, weekly among the maximum difference of temperature and temperature and in Monday among Sunday
Between temperature difference.Wherein, average middle temperature is the average value of middle temperature one week seven days weekly, and middle temperature is one day
The median of interior temperature range.When obtaining weather characteristics, the 100th week the last week can be taken, the weather characteristics of the 99th week,
Multiple all weather characteristics before can also taking the 100th week, the i.e. number of historical time unit can be multiple.
Environmental characteristic is SO2, NO2, PM10, O3, CO and PM2.5 the concentration average in historical time unit respectively.
For object time unit, i.e., for the prediction target of the 100th week, its autoregression temporal characteristics can be included but not
It is limited to following characteristics:The influenza-like case percentage of upper one week, the influenza-like case percentage of upper two weeks, the influenza sample of upper three weeks
Case percentage, the influenza-like case percentage average for closing on three weeks, the influenza-like case percentage variance for closing on three weeks, last year
The influenza-like case percentage of the same period.
After above-mentioned external source feature and autoregression temporal characteristics are got, these features are pre-processed and normalized
Processing, to remove the off-note in features described above, avoid these characteristic values come to the training band of forecast model some
Anomalous effects, and then improve the prediction precision of forecast model.
Alternatively, pretreatment can include extremal optimization processing and/or missing values supplement process.
Because the order of magnitude of searchable index is bigger, in fact it could happen that data area it is also bigger, it is therefore preferred to searching
Rope index carries out extremal optimization processing, specifically, obtains first quartile Q1, the 3rd quartile of the searchable index collection
Q3 and quartile deviation IQR (interquartile range, also known as interquartile-range IQR);According to the first quartile, described
Three quartiles and the quartile deviation determine the first threshold and Second Threshold of searchable index, and the first threshold is less than described the
Two threshold values;The searchable index more than the Second Threshold is concentrated to be converted to the Second Threshold searchable index, by described in
Searchable index concentrates the searchable index less than the first threshold to be converted to the first threshold.Above-mentioned first threshold can take
Q1-k*IQR, above-mentioned Second Threshold can take Q3+k*IQR, wherein, k >=0.Preferably, in one embodiment, k=1.5.
, can be according to kNN (k-NearestNeighbor, k arest neighbors) algorithm, to passing through on missing values supplement process
The searchable index of extremal optimization processing carries out missing values supplement.Specifically, after searchable index is got, detect in searchable index
Whether keyword is had without corresponding exponential quantity, if so, then obtaining in multiple time quantums adjacent with the object time unit
Searchable index does not have missing values time quantum, with putting down for the searchable index that keyword is corresponded in the searchable index of these time quantums
Either median or mode replace the data lacked to average.For example, keyword is " antiviral oral in the searchable index of the 99th week
The searchable index missing of liquid ", and the 98th week, the 97th week, keyword " antiviral oral liquor " searches in the searchable index of the 95th week
Rope index does not lack, then can with the average value of the searchable index of the keyword " antiviral oral liquor " in these three weeks or in
Digit or mode are instead of the searchable index of " antiviral oral liquor " that is lacked in the searchable index of the 99th week.In other embodiment
In, missing values can also be supplemented using other modes, such as averaging method, remove other keywords outside missing values
The average value of searchable index, the missing values in object element are substituted with the average value of acquisition.
Then, the external source feature for autoregression temporal characteristics and Jing Guo above-mentioned pretreatment is normalized, will
The above-mentioned characteristic value with different dimensions is converted to the nondimensional characteristic value for meeting normal distribution, avoids because feature is seriously askew
It is bent and cause the precision of prediction result low.
After getting the feature set after normalized, the feature in feature set is screened.Alternatively, at some
In embodiment, using pearson coefficient correlations (Pearson correlation coefficient, Pearson correlation coefficients)
Feature is filtered.Calculate the pearson coefficient correlations of each characteristic value in the feature set;It will be selected in the feature set
The feature that pearson coefficient correlations are less than or equal to preset correlation coefficient number is selected, as predicted characteristics.In certain embodiments,
Preset correlation coefficient number preferably takes 0.2.It is understood that in other embodiments, other Feature Selections can also be used
Method is screened to feature set.Will remaining feature is as predicted characteristics after Feature Selection, by object time unit
The actual observed value of influenza-like case percentage is as prediction target, above-mentioned predicted characteristics and prediction one pre- test sample of target configuration
This.
In the manner described above, forecast sample corresponding to multiple predicted time units is obtained from historical data, for example, respectively
99th week, the 98th week, the 97th week ... are waited as object time unit, their forecast sample is obtained respectively, by acquisition
These forecast samples, which are input in default regression model, carries out model training, obtains the model parameter of the model, wherein, at some
In embodiment, model can be trained by the way of k- rolls over cross validation, by one in above-mentioned multiple forecast samples
Sample is trained as test set, remaining sample as training set to model.Or in other embodiments, it is default
Regression model can be LASSO regression models, ridge regression model etc..In one embodiment, mould is returned preferably by LASSO
Type.Forecast model of the regression model as the project to be measured of model parameter will be determined.
It is understood that in above-described embodiment, to the solution of the present invention by taking the prediction of influenza-like case percentage as an example
It is illustrated.But scheme proposed by the present invention is not limited to that, the prediction of other projects, example can also be applied to
Such as, the prediction of weather, the prediction of ad click rate etc., corresponding different prediction project is, it is necessary to which the project can be reflected by gathering
The data of change external source feature is handled as external source feature, and according to such scheme, exist in itself then in conjunction with project
Autoregression temporal characteristics on time dimension are trained to regression model to generate forecast model.
The forecast model that the present embodiment proposes establishes device, obtain project to be measured one before object time unit or
External source feature in multiple time quantums, and the autoregression temporal characteristics before object time unit, to external source feature and certainly
Regression time feature is pre-processed and normalized, to obtain normalized feature set, according to preset rules to feature set
In feature screened to obtain predicted characteristics, corresponding with the object time unit actual observed value of the predicted characteristics forms one
Forecast sample, multiple forecast samples of multiple time quantums are obtained as procedure described above, above-mentioned multiple forecast samples are input to
It is trained in default regression model to determine model parameter, using the default regression model after determination model parameter as prediction mould
Type, the external source feature of project to be measured and autoregression temporal characteristics are combined and carry out pretreatment by the scheme of the invention forms one
Individual feature set, qualified feature is filtered out as predicted characteristics from feature set generation prediction mould is trained to regression model
Type, the unicity of sample characteristics is avoided, improve the prediction precision of forecast model.
Alternatively, in other examples, forecast model, which establishes program, can also be divided into one or more mould
Block, one or more module are stored in memory 11, and (the present embodiment is processor by one or more processors
12) performed to complete the present invention, the module alleged by the present invention is the series of computation machine program for referring to complete specific function
Instruction segment.
Shown in reference picture 2, the work(of program is established for the forecast model that forecast model of the present invention is established in the embodiment of device one
Can module diagram, in the embodiment, forecast model, which establishes program, can be divided into acquisition module 10, processing module 20, sieve
Modeling block 30, module 40 and training module 50 are formed, wherein:
Acquisition module 10 is used to obtain project to be measured in one or more time quantums before object time unit
External source feature, and the autoregression temporal characteristics before object time unit;
Processing module 20 is used to pre-process the external source feature, and to the autoregression temporal characteristics and by institute
The external source feature for stating pretreatment is normalized, to obtain normalized feature set;
Screening module 30 is used to carry out Feature Selection to the feature in the feature set according to preset rules, to obtain prediction
Feature;
Module 40 is formed to be used for using actual observed value of the project to be measured in the object time unit as predicting
Target, using the multiple predicted characteristics and the prediction target as a forecast sample;
Training module 50 is used for the multiple forecast samples for obtaining multiple time quantums respectively, and the multiple forecast sample is defeated
Enter into default regression model and be trained to determine model parameter, will determine that the default regression model after model parameter is made
For the forecast model of the project to be measured.
In addition, the present invention also provides a kind of forecast model method for building up.Shown in reference picture 3, built for forecast model of the present invention
The flow chart of cube method first embodiment.This method can be performed by a device, and the device can be real by software and/or hardware
It is existing.
In the present embodiment, forecast model method for building up includes:
Step S10, it is special to obtain external source of the project to be measured in one or more time quantums before object time unit
Sign, and the autoregression temporal characteristics before object time unit;
Step S20, the external source feature is pre-processed, and to the autoregression temporal characteristics and pass through the pre- place
The external source feature of reason is normalized, to obtain normalized feature set;
Step S30, Feature Selection is carried out to the feature in the feature set according to preset rules, to obtain predicted characteristics;
Step S40, will using actual observed value of the project to be measured in the object time unit as prediction target
The multiple predicted characteristics and the prediction target are as a forecast sample;
Step S50, according to the step S10 to step S40, multiple forecast samples of multiple time quantums are obtained respectively,
The multiple forecast sample is input in default regression model and is trained to determine model parameter, after model parameter is determined
Forecast model of the default regression model as the project to be measured.
In this embodiment, the scheme of the present embodiment is illustrated as project to be measured using influenza prediction project.Due to
What is used when being trained to model is all historical data, i.e. data in the past period, therefore, from history
When acquisition characteristics value is as test sample in data, the time quantum in each test sample is also the past time
Section.In the present embodiment, with mono- time quantum of Zhou Zuowei.Assuming that by past 100 all external source features and influenza sample
Case percentage as historical data, wherein, influenza-like case percentage be somewhere each Sentinel point hospital influenza-like case sum/
Always medical person-time of each Sentinel point hospital outpatient in somewhere.
The process that a forecast sample is obtained from above-mentioned historical data is as follows:A week is determined from above-mentioned historical data
As object time unit, for example, with the 100th week in past 100 week, i.e., closest one with current point in time
Week, as object time unit, then using the actual observed value of the influenza-like case percentage of the 100th week as prediction target.From
Acquisition characteristics are as predicted characteristics in multiple all data before 100th week.
Specifically, external source feature corresponding with one or more time quantum before object time unit is obtained, it is excellent
Selection of land, external source feature include searchable index, weather characteristics and environmental characteristic etc. and cause certain journey to the percentage of influenza-like case
The surface of the influence of degree.
Searchable index is:In a time quantum before object time unit, it is therefore preferable to object time unit
A upper time quantum, search engine user search for the frequency of associative key on the search engine, and search engine typically can
The frequency is recorded, for example, Baidu's index of Baidu.Wherein, associative key be user pre-set with the influenza
Prediction there is certain causal keyword, such as " nasal obstruction ", " sneezing ", " antiviral oral liquor ", " having a throat-ache " etc.
The term related to influenza, will not enumerate herein, and user can be arranged as required to multiple keywords, and be searched from some
On rope website searchable index collection is formed according to searchable index corresponding to keyword extraction.
Weather characteristics include but is not limited to following feature:Weekly fine day number, weekly cloudy number of days, weekly cloudy number,
Weekly rainfall, weekly air quantity, it is average weekly among temperature, weekly among the maximum difference of temperature and temperature and in Monday among Sunday
Between temperature difference.Wherein, average middle temperature is the average value of middle temperature one week seven days weekly, and middle temperature is one day
The median of interior temperature range.When obtaining weather characteristics, the 100th week the last week can be taken, the weather characteristics of the 99th week,
Multiple all weather characteristics before can also taking the 100th week, the i.e. number of historical time unit can be multiple.
Environmental characteristic is SO2, NO2, PM10, O3, CO and PM2.5 the concentration average in historical time unit respectively.
For object time unit, i.e., for the prediction target of the 100th week, its autoregression temporal characteristics can be included but not
It is limited to following characteristics:The influenza-like case percentage of upper one week, the influenza-like case percentage of upper two weeks, the influenza sample of upper three weeks
Case percentage, the influenza-like case percentage average for closing on three weeks, the influenza-like case percentage variance for closing on three weeks, last year
The influenza-like case percentage of the same period.
After above-mentioned external source feature and autoregression temporal characteristics are got, these features are pre-processed and normalized
Processing, to remove the off-note in features described above, avoid these characteristic values come to the training band of forecast model some
Anomalous effects, and then improve the prediction precision of forecast model.
Alternatively, pretreatment can include extremal optimization processing and/or missing values supplement process.
Because the order of magnitude of searchable index is bigger, in fact it could happen that data area it is also bigger, it is therefore preferred to searching
Rope index carries out extremal optimization processing, specifically, obtains first quartile Q1, the 3rd quartile of the searchable index collection
Q3 and quartile deviation IQR (interquartile range, also known as interquartile-range IQR);According to the first quartile, described
Three quartiles and the quartile deviation determine the first threshold and Second Threshold of searchable index, and the first threshold is less than described the
Two threshold values;The searchable index more than the Second Threshold is concentrated to be converted to the Second Threshold searchable index, by described in
Searchable index concentrates the searchable index less than the first threshold to be converted to the first threshold.Above-mentioned first threshold can take
Q1-k*IQR, above-mentioned Second Threshold can take Q3+k*IQR, wherein, k >=0.Preferably, in one embodiment, k=1.5.
On missing values supplement process, the searchable index by extremal optimization processing can be entered according to neighbouring kNN algorithms
Row missing values supplement.Specifically, whether after searchable index is got, detecting has keyword to refer to without corresponding in searchable index
Numerical value, there is no the missing values time single if so, then obtaining searchable index in multiple time quantums adjacent with the object time unit
Member, with the average value of the searchable index that keyword is corresponded in the searchable index of these time quantums either median or mode generation
For the data of missing.For example, the searchable index of keyword " antiviral oral liquor " lacks in the searchable index of the 99th week, and the 98th
The searchable index of keyword " antiviral oral liquor " does not lack in the searchable index in week, the 97th week, the 95th week, then can use this
Either median or mode replace the 99th week the average value of the searchable index of three all keywords " antiviral oral liquor "
The searchable index of " antiviral oral liquor " that is lacked in searchable index.In other embodiments, other modes pair can also be used
Missing values are supplemented, such as averaging method, the average value of the searchable index of other keywords outside missing values are removed, with acquisition
Average value substitute object element in missing values.
Then, the external source feature for autoregression temporal characteristics and Jing Guo above-mentioned pretreatment is normalized, will
The above-mentioned characteristic value with different dimensions is converted to the nondimensional characteristic value for meeting normal distribution, avoids because feature is seriously askew
It is bent and cause the precision of prediction result low.
After getting the feature set after normalized, the feature in feature set is screened.Alternatively, at some
In embodiment, feature is filtered using pearson coefficient correlations.Calculate each characteristic value in the feature set
Pearson coefficient correlations;The spy that pearson coefficient correlations will be selected to be less than or equal to preset correlation coefficient number in the feature set
Sign, as predicted characteristics.In certain embodiments, preset correlation coefficient number preferably takes 0.2.It is understood that in other realities
Apply in example, feature set can also be screened using other Feature Selection methods.Will after Feature Selection remaining spy
Sign is used as predicted characteristics, above-mentioned using the actual observed value of the influenza-like case percentage of object time unit as prediction target
Predicted characteristics and prediction one forecast sample of target configuration.
In the manner described above, forecast sample corresponding to multiple predicted time units is obtained from historical data, for example, respectively
99th week, the 98th week, the 97th week ... are waited as object time unit, their forecast sample is obtained respectively, by acquisition
These forecast samples, which are input in default regression model, carries out model training, obtains the model parameter of the model, wherein, at some
In embodiment, model can be trained by the way of k- rolls over cross validation, by one in above-mentioned multiple forecast samples
Sample is trained as test set, remaining sample as training set to model.Or in other embodiments, it is default
Regression model can be LASSO regression models, ridge regression model etc..In one embodiment, mould is returned preferably by LASSO
Type.Forecast model of the regression model as the project to be measured of model parameter will be determined.
It is understood that in above-described embodiment, to the solution of the present invention by taking the prediction of influenza-like case percentage as an example
It is illustrated.But scheme proposed by the present invention is not limited to that, the prediction of other projects, example can also be applied to
Such as, the prediction of weather, the prediction of ad click rate etc., corresponding different prediction project is, it is necessary to which the project can be reflected by gathering
The data of change external source feature is handled as external source feature, and according to such scheme, exist in itself then in conjunction with project
Autoregression temporal characteristics on time dimension are trained to regression model to generate forecast model.The forecast model can be used for
Prediction to project to be measured in following a period of time.
The forecast model method for building up that the present embodiment proposes, obtain project to be measured one before object time unit or
External source feature in multiple time quantums, and the autoregression temporal characteristics before object time unit, to external source feature and certainly
Regression time feature is pre-processed and normalized, to obtain normalized feature set, according to preset rules to feature set
In feature screened to obtain predicted characteristics, corresponding with the object time unit actual observed value of the predicted characteristics forms one
Forecast sample, multiple forecast samples of multiple time quantums are obtained as procedure described above, above-mentioned multiple forecast samples are input to
It is trained in default regression model to determine model parameter, using the default regression model after determination model parameter as prediction mould
Type, the external source feature of project to be measured and autoregression temporal characteristics are combined and carry out pretreatment by the scheme of the invention forms one
Individual feature set, qualified feature is filtered out as predicted characteristics from feature set generation prediction mould is trained to regression model
Type, the unicity of sample characteristics is avoided, improve the prediction precision of forecast model.
In addition, the embodiment of the present invention also proposes a kind of computer-readable recording medium, the computer-readable recording medium
On be stored with forecast model and establish program, the forecast model, which is established, realizes following operation when program is executed by processor:
A, external source feature of the project to be measured in one or more time quantums before object time unit is obtained, and
Autoregression temporal characteristics before object time unit;
B, the external source feature is pre-processed, and to the autoregression temporal characteristics and by the outer of the pretreatment
Source feature is normalized, to obtain normalized feature set;
C, Feature Selection is carried out to the feature in the feature set according to preset rules, to obtain predicted characteristics;
D, will be described more using actual observed value of the project to be measured in the object time unit as prediction target
Individual predicted characteristics and the prediction target are as a forecast sample;
E, multiple forecast samples of multiple time quantums are obtained respectively according to the step of A to D, by the multiple prediction
Sample is input in default regression model and is trained to determine model parameter, by the default recurrence after determination model parameter
Forecast model of the model as the project to be measured.
Further, the forecast model, which is established, also realizes following operation when program is executed by processor:
Obtain first quartile, the 3rd quartile and the quartile deviation of the searchable index collection;
The first threshold of searchable index is determined according to the first quartile, the 3rd quartile and the quartile deviation
Value and Second Threshold, the first threshold are less than the Second Threshold;
The searchable index more than the Second Threshold is concentrated to be converted to the Second Threshold searchable index, by described in
Searchable index concentrates the searchable index less than the first threshold to be converted to the first threshold.
Further, the forecast model, which is established, also realizes following operation when program is executed by processor:
According to neighbouring kNN algorithms, missing values supplement is carried out to the searchable index by extremal optimization processing.
Further, the forecast model, which is established, also realizes following operation when program is executed by processor:
Calculate the pearson coefficient correlations of each characteristic value in the feature set;
The feature that pearson coefficient correlations will be selected to be less than or equal to preset correlation coefficient number in the feature set, as
Predicted characteristics.
It should be noted that the embodiments of the present invention are for illustration only, the quality of embodiment is not represented.And
Term " comprising " herein, "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that bag
To include process, device, article or the method for a series of elements not only include those key elements, but also including being not expressly set out
Other element, or also include for this process, device, article or the intrinsic key element of method.Do not limiting more
In the case of, the key element that is limited by sentence "including a ...", it is not excluded that in the process including the key element, device, article
Or other identical element in method also be present.
Through the above description of the embodiments, those skilled in the art can be understood that above-described embodiment side
Method can add the mode of required general hardware platform to realize by software, naturally it is also possible to by hardware, but in many cases
The former is more preferably embodiment.Based on such understanding, technical scheme is substantially done to prior art in other words
Going out the part of contribution can be embodied in the form of software product, and the computer software product is stored in one as described above
In storage medium (such as ROM/RAM, magnetic disc, CD), including some instructions to cause a station terminal equipment (can be mobile phone,
Computer, server, or network equipment etc.) perform method described in each embodiment of the present invention.
The preferred embodiments of the present invention are these are only, are not intended to limit the scope of the invention, it is every to utilize this hair
The equivalent structure or equivalent flow conversion that bright specification and accompanying drawing content are made, or directly or indirectly it is used in other related skills
Art field, is included within the scope of the present invention.
Claims (10)
1. a kind of forecast model establishes device, it is characterised in that described device includes:Memory, processor and it is stored in described
On memory and the forecast model that can run on the processor establishes program, and the forecast model establishes program by the place
Reason device realizes following steps when performing:
A, external source feature of the project to be measured in one or more time quantums before object time unit is obtained, and in mesh
Mark the autoregression temporal characteristics before time quantum;
B, the external source feature is pre-processed, and the external source to the autoregression temporal characteristics and Jing Guo the pretreatment is special
Sign is normalized, to obtain normalized feature set;
C, Feature Selection is carried out to the feature in the feature set according to preset rules, to obtain predicted characteristics;
D, will be the multiple pre- using actual observed value of the project to be measured in the object time unit as prediction target
Feature and the prediction target are surveyed as a forecast sample;
E, multiple forecast samples of multiple time quantums are obtained respectively according to the step of A to D, by the multiple forecast sample
It is input in default regression model and is trained to determine model parameter, by the default regression model after determination model parameter
Forecast model as the project to be measured.
2. forecast model according to claim 1 establishes device, it is characterised in that if the external source feature includes described treat
Searchable index collection corresponding to survey project, the pretreatment is handled for extremal optimization, then described that the external source feature is located in advance
The step of reason, includes:
Obtain first quartile, the 3rd quartile and the quartile deviation of the searchable index collection;
According to the first quartile, the 3rd quartile and the quartile deviation determine searchable index first threshold and
Second Threshold, the first threshold are less than the Second Threshold;
The searchable index more than the Second Threshold is concentrated to be converted to the Second Threshold searchable index, by the search
Searchable index in the set of indexes less than the first threshold is converted to the first threshold.
3. forecast model according to claim 2 establishes device, it is characterised in that described that the external source feature is carried out in advance
The step of processing, also includes:
According to neighbouring kNN algorithms, missing values supplement is carried out to the searchable index by extremal optimization processing.
4. forecast model according to claim 1 establishes device, it is characterised in that it is described according to preset rules to the spy
Feature in collection carries out Feature Selection, includes the step of to obtain predicted characteristics:
Calculate the pearson coefficient correlations of each characteristic value in the feature set;
The feature that pearson coefficient correlations will be selected to be less than or equal to preset correlation coefficient number in the feature set, as prediction
Feature.
5. forecast model according to any one of claim 1 to 4 establishes device, it is characterised in that the project to be measured
Project is predicted for influenza, the external source feature includes searchable index, weather characteristics and environmental characteristic, and the default regression model is
LASSO regression models.
6. a kind of forecast model method for building up, it is characterised in that methods described includes:
A, external source feature of the project to be measured in one or more time quantums before object time unit is obtained, and in mesh
Mark the autoregression temporal characteristics before time quantum;
B, the external source feature is pre-processed, and the external source to the autoregression temporal characteristics and Jing Guo the pretreatment is special
Sign is normalized, to obtain normalized feature set;
C, Feature Selection is carried out to the feature in the feature set according to preset rules, to obtain predicted characteristics;
D, will be the multiple pre- using actual observed value of the project to be measured in the object time unit as prediction target
Feature and the prediction target are surveyed as a forecast sample;
E, multiple forecast samples of multiple time quantums are obtained respectively according to the step of A to D, by the multiple forecast sample
It is input in default regression model and is trained to determine model parameter, by the default regression model after determination model parameter
Forecast model as the project to be measured.
7. forecast model method for building up according to claim 6, it is characterised in that if the external source feature includes described treat
Searchable index collection corresponding to survey project, the pretreatment is handled for extremal optimization, then described that the external source feature is located in advance
The step of reason, includes:
Obtain first quartile, the 3rd quartile and the quartile deviation of the searchable index collection;
According to the first quartile, the 3rd quartile and the quartile deviation determine searchable index first threshold and
Second Threshold, the first threshold are less than the Second Threshold;
The searchable index more than the Second Threshold is concentrated to be converted to the Second Threshold searchable index, by the search
Searchable index in the set of indexes less than the first threshold is converted to the first threshold.
8. forecast model method for building up according to claim 7, it is characterised in that described to be carried out in advance to the external source feature
The step of processing, also includes:
According to neighbouring kNN algorithms, missing values supplement is carried out to the searchable index by extremal optimization processing.
9. the forecast model method for building up according to any one of claim 6 to 8, it is characterised in that described according to default
Rule carries out Feature Selection to the feature in the feature set, includes the step of to obtain predicted characteristics:
Calculate the pearson coefficient correlations of each characteristic value in the feature set;
The feature that pearson coefficient correlations will be selected to be less than or equal to preset correlation coefficient number in the feature set, as prediction
Feature.
10. a kind of computer-readable recording medium, it is characterised in that prediction mould is stored with the computer-readable recording medium
Type establishes program, and the forecast model is established and realized when program is executed by processor as any one of claim 6 to 9
The step of forecast model method for building up.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710715445.1A CN107688872A (en) | 2017-08-20 | 2017-08-20 | Forecast model establishes device, method and computer-readable recording medium |
PCT/CN2017/108801 WO2019037260A1 (en) | 2017-08-20 | 2017-10-31 | Predictive model establishment method and device, and computer-readable storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710715445.1A CN107688872A (en) | 2017-08-20 | 2017-08-20 | Forecast model establishes device, method and computer-readable recording medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107688872A true CN107688872A (en) | 2018-02-13 |
Family
ID=61153475
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710715445.1A Pending CN107688872A (en) | 2017-08-20 | 2017-08-20 | Forecast model establishes device, method and computer-readable recording medium |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN107688872A (en) |
WO (1) | WO2019037260A1 (en) |
Cited By (22)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108492138A (en) * | 2018-03-19 | 2018-09-04 | 平安科技(深圳)有限公司 | Product buys prediction technique, server and storage medium |
CN108597617A (en) * | 2018-04-11 | 2018-09-28 | 平安科技(深圳)有限公司 | Epidemic disease grade predicting method and device, computer installation and readable storage medium storing program for executing |
CN108597618A (en) * | 2018-04-20 | 2018-09-28 | 杭州恒生数字设备科技有限公司 | A kind of influenza prediction video camera with autolearn feature |
CN108630321A (en) * | 2018-04-11 | 2018-10-09 | 平安科技(深圳)有限公司 | Forecast of epiphytotics method, computer installation and computer readable storage medium |
CN108766585A (en) * | 2018-05-31 | 2018-11-06 | 平安科技(深圳)有限公司 | Generation method, device and the computer readable storage medium of influenza prediction model |
CN108831561A (en) * | 2018-05-31 | 2018-11-16 | 平安科技(深圳)有限公司 | Generation method, device and the computer readable storage medium of influenza prediction model |
CN109192306A (en) * | 2018-09-21 | 2019-01-11 | 广东工业大学 | A kind of judgment means of diabetes, equipment and computer readable storage medium |
CN109243619A (en) * | 2018-07-13 | 2019-01-18 | 平安科技(深圳)有限公司 | Generation method, device and the computer readable storage medium of prediction model |
CN109409596A (en) * | 2018-10-22 | 2019-03-01 | 东软集团股份有限公司 | Processing method, device, equipment and the computer readable storage medium of prediction of wind speed |
CN109493979A (en) * | 2018-10-23 | 2019-03-19 | 平安科技(深圳)有限公司 | A kind of disease forecasting method and apparatus based on intelligent decision |
CN109754175A (en) * | 2018-12-28 | 2019-05-14 | 广州明动软件股份有限公司 | For to administrative procedure for examination and approval finish the time limit carry out compression prediction computation model and its application |
CN109905271A (en) * | 2018-05-18 | 2019-06-18 | 华为技术有限公司 | A kind of prediction technique, training method, device and computer storage medium |
CN110111902A (en) * | 2019-04-04 | 2019-08-09 | 平安科技(深圳)有限公司 | Disease cycle prediction technique, device and the storage medium of acute infectious disease |
CN110136841A (en) * | 2019-03-27 | 2019-08-16 | 平安科技(深圳)有限公司 | Disease incidence prediction technique, device and computer readable storage medium |
CN110979589A (en) * | 2019-12-16 | 2020-04-10 | 上海船舶研究设计院(中国船舶工业集团公司第六0四研究院) | Ship control method and device based on wave resistance-increasing prediction and intelligent terminal |
CN111415752A (en) * | 2020-03-01 | 2020-07-14 | 集美大学 | Hand-foot-and-mouth disease prediction method integrating meteorological factors and search indexes |
CN111524600A (en) * | 2020-04-24 | 2020-08-11 | 中国地质大学(武汉) | Liver cancer postoperative recurrence risk prediction system based on neighbor2vec |
CN111524599A (en) * | 2020-04-24 | 2020-08-11 | 中国地质大学(武汉) | New coronary pneumonia data processing method and prediction system based on machine learning |
CN112802603A (en) * | 2021-02-04 | 2021-05-14 | 北京深演智能科技股份有限公司 | Method and device for predicting influenza degree |
CN113436751A (en) * | 2021-06-29 | 2021-09-24 | 山东健康医疗大数据有限公司 | Weekly ILI proportion trend prediction system and method |
CN113707337A (en) * | 2021-08-30 | 2021-11-26 | 平安科技(深圳)有限公司 | Disease early warning method, device, equipment and storage medium based on multi-source data |
CN117274798A (en) * | 2023-09-06 | 2023-12-22 | 中国农业科学院农业信息研究所 | Remote sensing rice identification method based on regularized time sequence variation model |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112545461A (en) * | 2020-12-05 | 2021-03-26 | 深圳市美的连医疗电子股份有限公司 | Method, device and system for detecting non-invasive hemoglobin concentration value and computer readable storage medium |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104809335B (en) * | 2015-04-10 | 2019-03-05 | 上海卫生信息工程技术研究中心有限公司 | A kind of analysis prediction meanss that environmental change influences disease incidence |
CN106709588B (en) * | 2015-11-13 | 2022-05-17 | 日本电气株式会社 | Prediction model construction method and device and real-time prediction method and device |
CN106777874A (en) * | 2016-11-18 | 2017-05-31 | 中国科学院自动化研究所 | The method that forecast model is built based on Recognition with Recurrent Neural Network |
CN106777891B (en) * | 2016-11-21 | 2019-06-07 | 中国科学院自动化研究所 | A kind of selection of data characteristics and prediction technique and device |
CN106980914B (en) * | 2017-06-05 | 2018-04-03 | 厦门美柚信息科技有限公司 | Forecasting Methodology and device, the terminal of determining female physiological periodicity |
-
2017
- 2017-08-20 CN CN201710715445.1A patent/CN107688872A/en active Pending
- 2017-10-31 WO PCT/CN2017/108801 patent/WO2019037260A1/en active Application Filing
Cited By (35)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108492138A (en) * | 2018-03-19 | 2018-09-04 | 平安科技(深圳)有限公司 | Product buys prediction technique, server and storage medium |
CN108492138B (en) * | 2018-03-19 | 2020-03-24 | 平安科技(深圳)有限公司 | Product purchase prediction method, server and storage medium |
CN108597617A (en) * | 2018-04-11 | 2018-09-28 | 平安科技(深圳)有限公司 | Epidemic disease grade predicting method and device, computer installation and readable storage medium storing program for executing |
CN108630321A (en) * | 2018-04-11 | 2018-10-09 | 平安科技(深圳)有限公司 | Forecast of epiphytotics method, computer installation and computer readable storage medium |
WO2019196283A1 (en) * | 2018-04-11 | 2019-10-17 | 平安科技(深圳)有限公司 | Epidemic disease prediction method, computer device and non-volatile readable storage medium |
CN108597618A (en) * | 2018-04-20 | 2018-09-28 | 杭州恒生数字设备科技有限公司 | A kind of influenza prediction video camera with autolearn feature |
CN108597618B (en) * | 2018-04-20 | 2021-09-03 | 杭州恒生数字设备科技有限公司 | Influenza prediction camera with automatic learning function |
CN109905271B (en) * | 2018-05-18 | 2021-01-12 | 华为技术有限公司 | Prediction method, training method, device and computer storage medium |
CN109905271A (en) * | 2018-05-18 | 2019-06-18 | 华为技术有限公司 | A kind of prediction technique, training method, device and computer storage medium |
WO2019227711A1 (en) * | 2018-05-31 | 2019-12-05 | 平安科技(深圳)有限公司 | Method and apparatus for generating influenza prediction model, and computer-readable storage medium |
CN108831561A (en) * | 2018-05-31 | 2018-11-16 | 平安科技(深圳)有限公司 | Generation method, device and the computer readable storage medium of influenza prediction model |
CN108766585A (en) * | 2018-05-31 | 2018-11-06 | 平安科技(深圳)有限公司 | Generation method, device and the computer readable storage medium of influenza prediction model |
WO2019227716A1 (en) * | 2018-05-31 | 2019-12-05 | 平安科技(深圳)有限公司 | Method for generating influenza prediction model, apparatus, and computer readable storage medium |
CN109243619A (en) * | 2018-07-13 | 2019-01-18 | 平安科技(深圳)有限公司 | Generation method, device and the computer readable storage medium of prediction model |
CN109243619B (en) * | 2018-07-13 | 2023-03-31 | 平安科技(深圳)有限公司 | Generation method and device of prediction model and computer readable storage medium |
CN109192306A (en) * | 2018-09-21 | 2019-01-11 | 广东工业大学 | A kind of judgment means of diabetes, equipment and computer readable storage medium |
CN109409596A (en) * | 2018-10-22 | 2019-03-01 | 东软集团股份有限公司 | Processing method, device, equipment and the computer readable storage medium of prediction of wind speed |
CN109409596B (en) * | 2018-10-22 | 2021-04-13 | 东软集团股份有限公司 | Processing method, device, equipment and computer-readable storage medium for predicting wind speed |
CN109493979A (en) * | 2018-10-23 | 2019-03-19 | 平安科技(深圳)有限公司 | A kind of disease forecasting method and apparatus based on intelligent decision |
CN109754175A (en) * | 2018-12-28 | 2019-05-14 | 广州明动软件股份有限公司 | For to administrative procedure for examination and approval finish the time limit carry out compression prediction computation model and its application |
CN110136841A (en) * | 2019-03-27 | 2019-08-16 | 平安科技(深圳)有限公司 | Disease incidence prediction technique, device and computer readable storage medium |
CN110136841B (en) * | 2019-03-27 | 2022-07-08 | 平安科技(深圳)有限公司 | Disease onset prediction method, device and computer readable storage medium |
CN110111902B (en) * | 2019-04-04 | 2022-05-27 | 平安科技(深圳)有限公司 | Acute infectious disease attack period prediction method, device and storage medium |
CN110111902A (en) * | 2019-04-04 | 2019-08-09 | 平安科技(深圳)有限公司 | Disease cycle prediction technique, device and the storage medium of acute infectious disease |
CN110979589A (en) * | 2019-12-16 | 2020-04-10 | 上海船舶研究设计院(中国船舶工业集团公司第六0四研究院) | Ship control method and device based on wave resistance-increasing prediction and intelligent terminal |
CN110979589B (en) * | 2019-12-16 | 2020-10-30 | 上海船舶研究设计院(中国船舶工业集团公司第六0四研究院) | Ship control method and device based on wave resistance-increasing prediction and intelligent terminal |
CN111415752A (en) * | 2020-03-01 | 2020-07-14 | 集美大学 | Hand-foot-and-mouth disease prediction method integrating meteorological factors and search indexes |
CN111415752B (en) * | 2020-03-01 | 2023-05-12 | 集美大学 | Hand-foot-and-mouth disease prediction method integrating meteorological factors and search indexes |
CN111524599A (en) * | 2020-04-24 | 2020-08-11 | 中国地质大学(武汉) | New coronary pneumonia data processing method and prediction system based on machine learning |
CN111524600A (en) * | 2020-04-24 | 2020-08-11 | 中国地质大学(武汉) | Liver cancer postoperative recurrence risk prediction system based on neighbor2vec |
CN112802603A (en) * | 2021-02-04 | 2021-05-14 | 北京深演智能科技股份有限公司 | Method and device for predicting influenza degree |
CN113436751A (en) * | 2021-06-29 | 2021-09-24 | 山东健康医疗大数据有限公司 | Weekly ILI proportion trend prediction system and method |
CN113707337A (en) * | 2021-08-30 | 2021-11-26 | 平安科技(深圳)有限公司 | Disease early warning method, device, equipment and storage medium based on multi-source data |
CN117274798A (en) * | 2023-09-06 | 2023-12-22 | 中国农业科学院农业信息研究所 | Remote sensing rice identification method based on regularized time sequence variation model |
CN117274798B (en) * | 2023-09-06 | 2024-03-29 | 中国农业科学院农业信息研究所 | Remote sensing rice identification method based on regularized time sequence variation model |
Also Published As
Publication number | Publication date |
---|---|
WO2019037260A1 (en) | 2019-02-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107688872A (en) | Forecast model establishes device, method and computer-readable recording medium | |
WO2021139325A1 (en) | Media information recommendation method and apparatus, electronic device, and storage medium | |
CN107633254A (en) | Establish device, method and the computer-readable recording medium of forecast model | |
CN110321477A (en) | Information recommendation method, device, terminal and storage medium | |
CN109034365A (en) | The training method and device of deep learning model | |
CN108427698A (en) | Updating device, method and the computer readable storage medium of prediction model | |
CN109509056A (en) | Method of Commodity Recommendation, electronic device and storage medium based on confrontation network | |
WO2022016556A1 (en) | Neural network distillation method and apparatus | |
CN110275952A (en) | News recommended method, device and medium based on user's short-term interest | |
CN107766573B (en) | Commodity recommendation method, commodity recommendation device, commodity recommendation equipment and storage medium based on data processing | |
CN116823409B (en) | Intelligent screening method and system based on target search data | |
US20140188928A1 (en) | Relational database management | |
US20230385597A1 (en) | Multi-granularity perception integrated learning method, device, computer equipment and medium | |
CN107818492A (en) | Products Show device, method and computer-readable recording medium | |
CN110347781A (en) | Article falls discharge method, article recommended method, device, equipment and storage medium | |
CN114595124B (en) | Time sequence abnormity detection model evaluation method, related device and storage medium | |
CN115376518A (en) | Voiceprint recognition method, system, device and medium for real-time noise big data | |
He et al. | Learning Informative Representation for Fairness-aware Multivariate Time-series Forecasting: A Group-based Perspective | |
CN111340540B (en) | Advertisement recommendation model monitoring method, advertisement recommendation method and advertisement recommendation model monitoring device | |
CA3035539A1 (en) | Systems and methods for measuring collected content significance | |
CN116703466A (en) | System access quantity prediction method based on improved wolf algorithm and related equipment thereof | |
US11442579B2 (en) | Method and electronic device for accidental touch prediction using ml classification | |
CN110020195A (en) | Article recommended method and device, storage medium, electronic equipment | |
CN115455276A (en) | Method and device for recommending object, computer equipment and storage medium | |
CN116467153A (en) | Data processing method, device, computer equipment and storage medium |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180213 |