CN110033309A

CN110033309A - Method, apparatus and terminal device based on historical data prediction result

Info

Publication number: CN110033309A
Application number: CN201910175734.6A
Authority: CN
Inventors: 王建明
Original assignee: Ping An Technology Shenzhen Co Ltd
Current assignee: Ping An Technology Shenzhen Co Ltd
Priority date: 2019-03-08
Filing date: 2019-03-08
Publication date: 2019-07-19
Anticipated expiration: 2039-03-08
Also published as: CN110033309B

Abstract

This application provides a kind of method, apparatus and terminal device based on historical data prediction result, method include obtaining video record, the face of several users in positioning video video recording, and intercept face picture；Discoloration is carried out to face picture and obtains black-white-gray face picture, user's face figure is searched according to black-white-gray face picture, transfer the User ID with user's face figure pre-association, and it obtains User ID and the historical data of specifies behavior occurs, N number of time series is got according to historical data, operating characteristics when specifies behavior occurs in user in recognition time sequence, and operating characteristics are loaded into LSTM model, to obtain optimal time sequence, finally, by optimal time sequence a cycle prediction result of the corresponding actual result of timing node as next period, and shopping list is made according to prediction result；To solve to be manually entered timing node low efficiency, the technical issues of solving to be likely to occur a variety of different prediction results.

Description

Method, apparatus and terminal device based on historical data prediction result

Technical field

This application involves prediction model field, in particular to a kind of method, apparatus based on historical data prediction result and Terminal device.

Background technique

With the fast development of information technology, marketer usually requires to consider the amount of commodity of order when ordering goods to producer, keeps away Exempt from that fixed article reserves are very few or reserves be very few excessive, but because client's purchase uncontrollability, marketer can not establish More accurately order, causes the reserves of article deviation occur.

In the prior art, the mode of establishing for predicting that user buys user's tendency model of product is that use is manually entered Family sometime node, user images corresponding with timing node are transferred into the default portrait library for storing user images, point The specific characteristic of analysis user images judges whether user buys product, is inclined to model to establish user, reaches and predict under the user One stage bought the effect of similar product, and marketer can be counted according to prediction result, with the accurate order volume of determination.But The corresponding user images of different timing nodes are different, there are problems that possibility constructs multiple users and is inclined to model, cause to occur A variety of different prediction results, accuracy are poor.Also, artificial input time node efficiency is low.

Summary of the invention

The application provides a kind of method, apparatus and terminal device based on historical data prediction result, and purpose purport solves people Work input time node efficiency is low, and solves the technical issues of being likely to occur a variety of different prediction results.

To achieve the above object, the application provides the following technical solutions:

The application provides a kind of method based on historical data prediction result

Video record is obtained, then positions the face of several users in the video record, and intercept several users Face picture；

Discoloration is carried out to the face picture, black-white-gray face picture is obtained, then according to the black-white-gray face picture The gray scale depth aggravated or desalinated, and the face edge of black-white-gray face picture is sharpened；

Search whether there is the user's face figure to match with the black-white-gray face picture, the use in facial library has Family face figure is black-white-gray picture；

If it exists, then the User ID with the user's face figure pre-association is transferred in the library ID, and obtains User ID label There is the historical data of specifies behavior；

It obtains user in the historical data and N number of time sequence using different time length as unit of specifies behavior occurs Column, wherein the N is the integer more than or equal to 1；

It identifies operating characteristics when specifies behavior occurs in user in N number of time series, and is loaded into the operating characteristics To LSTM model, when selecting optimal in N number of time series according to the operating characteristics by the LSTM model Between sequence, the optimal time sequence is the time sequence that the timing node that several described operating characteristics occur meets periodic law Column；

By the optimal time sequence a cycle the corresponding actual result of the timing node as next week The prediction result of phase；

Reserves data are transferred, the reserves data include article surplus, and it is described to judge whether the article surplus is greater than The article demand of prediction result；

If it is not, then calculating the difference of the article surplus Yu the article demand, and is made and purchased according to the difference Inventory.

Further, operating characteristics when specifies behavior occurs in user in the identification N number of time series, and be loaded into The step of operating characteristics to LSTM model, comprising:

The operating characteristics include user's operation portrait, and the user for obtaining each period in N number of time series occurs The corresponding multiple user's operation portraits of the timing node of specifies behavior；

It filters out in the multiple user's operation portrait and draws a portrait with the matched M user's operation of predetermined registration operation portrait, institute Stating M is the integer more than or equal to 1, and the predetermined registration operation portrait includes confirmation purchase portrait, the purchase portrait that hesitates, hesitates and do not buy It draws a portrait and does not buy portrait；

The M user's operation is loaded into draw a portrait to the LSTM model.

Further, the production method of the user's operation portrait, comprising:

Operating process and user operation time after obtaining user's log-on webpage；

The user's operation portrait is made according to the operating process of the user and user operation time.

Further, the screening in the multiple user's operation portrait is for the matched M use of predetermined registration operation portrait The step of family operation portrait, comprising:

Identify the M for reaching first threshold in the multiple user's operation portrait with predetermined registration operation portrait similarity A user's operation portrait；

The other user's operations deleted lower than the first threshold are drawn a portrait.

Further, the corresponding reality of the timing node by the optimal time sequence in a cycle is tied After the step of prediction result of the fruit as next period, further includes:

Judge whether the actual result in next period is identical as the prediction result；

If not identical, it is corresponding into next period to extend the timing node in a cycle of the optimal time sequence Timing node, and establish the new period.

There is the behavioral data of specifies behavior again in get User ID；

The behavioral data is imported to the library ID, and the history that the behavioral data is classified to the User ID is bought In file.

The application provides a kind of device based on historical data prediction result

Then face identification unit positions the face of several users in the video record for obtaining video record, And intercept the face picture of several users；

Picture processing unit obtains black-white-gray face picture, then according to institute for carrying out discoloration to the face picture The gray scale depth for stating black-white-gray face picture is aggravated or is desalinated, and is carried out to the face edge of black-white-gray face picture sharp Change；

Matching unit, for searching whether there is the user to match with the black-white-gray face picture in facial library has Face figure, the user's face figure is black-white-gray picture；

ID transfers unit, for if it exists, then transferring the User ID with the user's face figure pre-association in the library ID, and It obtains User ID label and the historical data of specifies behavior occurs；

There is the long with different time of specifies behavior for obtaining user in the historical data in time series acquiring unit Spend N number of time series as unit, wherein the N is the integer more than or equal to 1；

Operating characteristics recognition unit, operation when specifies behavior occurs in user in N number of time series for identification are special Sign, and be loaded into the operating characteristics to LSTM model, with by the LSTM model according to the operating characteristics when described N number of Between optimal time sequence is selected in sequence, the optimal time sequence is the timing node that several described operating characteristics occur Meet the time series of periodic law；

Predicting unit is tied for the corresponding reality of the timing node by the optimal time sequence in a cycle Prediction result of the fruit as next period；

Surplus judging unit, for transferring reserves data, the reserves data include article surplus, and judge the article Whether surplus is greater than the article demand of the prediction result；

Inventory production unit, for if it is not, then calculate the difference of the article surplus Yu the article demand, and according to The difference makes shopping list.

Further, the operating characteristics recognition unit includes:

Image collection module includes user's operation portrait for the operating characteristics, is obtained each in N number of time series There are the corresponding multiple user's operation portraits of timing node of specifies behavior in the user in a period；

Optical sieving module, for being filtered out in the multiple user's operation portrait and the matched M of predetermined registration operation portrait A user's operation portrait, the M are the integer more than or equal to 1, and the predetermined registration operation portrait includes confirmation purchase portrait, hesitate purchase Portrait is bought, hesitates not buy and draws a portrait and do not buy portrait；

GetImage module is drawn a portrait for being loaded into the M user's operation to the LSTM model.

The application also provides a kind of terminal device, including memory and processor, is stored with computer in the memory Program, the processor realize the step of the above-mentioned method based on historical data prediction result when executing the computer program Suddenly.

The application also provides a kind of terminal readable storage medium storing program for executing, is stored thereon with computer program, the computer program The step of above-mentioned method based on historical data prediction result is realized when being executed by processor.

This application provides a kind of method, apparatus and terminal device based on historical data prediction result have with following Beneficial effect:

Terminal device by obtain video record, then positioning video video recording in several users face, and if intercept The face picture of dry user；Discoloration is carried out to face picture, black-white-gray face picture is obtained, then according to black-white-gray face figure The gray scale depth of piece is aggravated or is desalinated, and is sharpened to the face edge of black-white-gray face picture；Have in facial library In search whether there is the user's face figure that matches with black-white-gray face picture, user's face figure is black-white-gray picture；If depositing The User ID with user's face figure pre-association is then being transferred in the library ID, and obtain the historical data of User ID, according to history number There is N number of time series of the different time length as unit of specifies behavior according to user is got, is used in recognition time sequence There are the corresponding operating characteristics of timing node of specifies behavior in family, and is loaded into operating characteristics to LSTM model, to pass through LSTM mould Type selects optimal time sequence according to operating characteristics in N number of time series, and finally, terminal device passes through optimal time sequence It is adopted in prediction result of the corresponding actual result of the timing node of a cycle as next period, and according to prediction result production Purchase inventory；To solve to be manually entered timing node low efficiency, and solve the technology for being likely to occur a variety of different prediction results Problem, and the statistics available multiple prediction results of marketer, to confirm accurately order volume.

Detailed description of the invention

Fig. 1 is the flow diagram of the method based on historical data prediction result in one embodiment of the application；

Fig. 2 is the flow diagram of the method based on historical data prediction result in another embodiment of the application；

Fig. 3 is the flow diagram of the method based on historical data prediction result in one embodiment of the application；

Fig. 4 is the structural block diagram of the device based on historical data prediction result in one embodiment of the application；

Fig. 5 is the structural schematic block diagram of the terminal device in the another embodiment of the application.

The embodiments will be further described with reference to the accompanying drawings for realization, functional characteristics and advantage for the purpose of the application.

Specific embodiment

It should be appreciated that specific embodiment described herein is only used to explain the application, it is not used to limit the application.

Below in conjunction with the attached drawing in the embodiment of the present application, technical solutions in the embodiments of the present application carries out clear, complete Site preparation description, it is clear that described embodiment is only a part of the embodiment of the application, instead of all the embodiments.Base Embodiment in the application, it is obtained by those of ordinary skill in the art without making creative efforts it is all its His embodiment, shall fall in the protection scope of this application.

With reference to attached drawing 1, a kind of flow diagram of method based on historical data prediction result is provided for the application, is wrapped It includes:

S100 obtains video record, then in positioning video video recording several users face, and intercept several users Face picture；

S200 carries out discoloration to face picture, black-white-gray face picture is obtained, then according to the ash of black-white-gray face picture The degree depth is aggravated or is desalinated, and is sharpened to the face edge of black-white-gray face picture；

S300 searches whether there is the user's face figure to match with black-white-gray face picture, user in facial library has Facial figure is black-white-gray picture；

S400, and if it exists, then transfer the User ID with user's face figure pre-association in the library ID, and obtain User ID label There is the historical data of specifies behavior；

S500 obtains user in historical data and the N time using different time length as unit of specifies behavior occurs Sequence, wherein N is the integer more than or equal to 1；

S600 identifies operating characteristics when specifies behavior occurs in user in N number of time series, and is loaded into operating characteristics extremely LSTM model, to select optimal time sequence, optimal time in N number of time series according to operating characteristics by LSTM model Sequence is the time series that the timing node that several operating characteristics occur meets periodic law；

S700, by optimal time sequence a cycle the corresponding actual result of timing node as next period Prediction result；

S800 transfers reserves data, and reserves data include article surplus, and judge whether article surplus is greater than prediction result Article demand；

S900 if it is not, then calculating the difference of article surplus Yu article demand, and makes shopping list according to difference.

It makes explanations explanation in conjunction with computer equipment, when worker needs to enable prediction, is opened to computer equipment input Dynamic instruction, computer equipment generate request column according to enabled instruction, and request column is shown by the screen of computer equipment, requests work The video recording of person's uploaded videos.After computer equipment gets the video record of worker's upload, several users in positioning video recording Face, and the face picture of several users is intercepted, then, gray processing processing (i.e. discoloration) is carried out to face picture, to obtain Black-white-gray face picture, aggravated according to the gray scale depth of black-white-gray face picture or desalinate (i.e. the gray scale the deep, aggravates, Gray scale is shallowly then desalinated), and protrusion is sharpened to face edge, so that black-white-gray face picture clean mark.Have in facial library In search whether there is the user's face figure that matches with black-white-gray face picture, which is all black-white-gray picture, It is appreciated that the EMS memory occupation amount of face picture is greater than the EMS memory occupation amount of black-white-gray face picture, therefore computer equipment matches Black-white-gray face picture and the speed of user's face figure are better than the speed of matching face picture and user's face figure, i.e. matching efficiency It is promoted；If it exists, then the User ID with user's face figure pre-association is transferred from the library ID, the library ID is that record has user to specify The repository of behavior history and Customs Assigned Number, such as: when user buys A product, computer equipment records the product letter of A product Current time node when breath, record user buy A product and be 1 by Customs Assigned Number, i.e., User ID is 1 in a time Node buys A product.Computer equipment calling and obtaining user ID from the library ID, while historical data corresponding with User ID is obtained, according to The timing node length for buying product according to the user that historical data is recorded obtains N number of time series, i.e., for example: user was at 30 days 3 A products are inside bought respectively, then are obtained using month as the time series of time shaft；Or user buys 3 B in 1 day and produces Product then obtain the time series of one day time span；It is further illustrated by the time series of time shaft of month, user 3 A products are bought respectively in 30 days at first, buy 2 A products respectively in 30 days at second, i.e. computer equipment obtains The time series for getting month speculates next 30 days product purchase volumes for 30 days and second for 30 days by first, but What 3 A products of first purchase of user in 30 days were bought in first week, second week and third Zhou Jinhang respectively, then it can be with week Settling time sequence results in N number of time series occur in this way；When user buys product because of the selection purchase opportunity of user Different, purchase product difference, purchase number difference can make computer equipment get N number of time series.

After computer equipment gets N number of time series, computer equipment can identify that user grasps in N number of time series Make operating characteristics when purchase product.Then, operating characteristics are loaded into LSTM model (long-short term by computer equipment Memory, shot and long term memory models), optimal time sequence is selected in N number of time series according to operating characteristics by LSTM model The step of column, LSTM model selects optimal time sequence, is LSTM model records multiple operating characteristics, and obtains in the historical data It takes multiple timing nodes corresponding with multiple operating characteristics, judges whether multiple timing nodes meet periodic law, i.e., for example: It is explained with Zhou Zuowei time series, operating characteristics occurs in first week Monday, also grasps on Monday of second week Make feature, therefore LSTM model thinks that the time series meets periodic law；LSTM model export selected results to computer equipment, After computer equipment receives optimal time sequence, by optimal time sequence a cycle the corresponding reality of timing node Prediction result of the border result as next period, i.e. computer equipment derive prediction result.

Further, after computer equipment calculates prediction result, reserves data are transferred, reserves data are staff's benefit Computer equipment recorded data when goods shipment, and when user specifies behavior occurs every time and purchases commodities, cargo reserves can be reduced； Whether that current article surplus of computer equipment Pan Gu is greater than the article demand of prediction result, if being less than, computer Equipment calculates the difference of article surplus and article demand, and generates shopping list according to difference, so that staff replenishes.

In one embodiment, operating characteristics when specifies behavior occurs in user in N number of time series are identified, and are loaded into operation The step of feature to LSTM model, comprising:

S601, operating characteristics include user's operation portrait, obtain the user's operation purchase in each period in N number of time series The corresponding multiple user's operation portraits of the timing node of product；

S602, screening is for the matched M user's operation portrait of predetermined registration operation portrait, M in multiple user's operations portrait For the integer more than or equal to 1, predetermined registration operation portrait include confirmation purchase portrait, the purchase portrait that hesitates, hesitate do not buy portrait and Portrait is not bought；

S603 is loaded into M user's operation and draws a portrait to LSTM model.

When user is when buying a certain product, computer equipment obtains user's operation portrait, while obtaining current time section User's operation portrait is ranged feature class by point, and default feature further includes electronic page transformation, audio etc., wherein user's operation The portrait that user's operation process and operating time when portrait is user's purchase product are constituted, electronic page transformation is user The page map of its mobile terminal when buying product and payment amount, and audio is the audio during user buys product.This Embodiment is made explanations explanation with user's operation portrait, the when segmentum intercalaris of user's purchase product in computer equipment acquisition time sequence Point corresponding user's operation portrait, and find from portrait library and to draw a portrait with the user's operation similar predetermined registration operation of drawing a portrait, if meter Calculation machine equipment searches out the predetermined registration operation portrait for existing in portrait library and being higher than first threshold with user's operation portrait, then is loaded into user Operation portrait is to the LSTM model, and wherein first threshold is 80%, i.e. the similarity when user's operation portrait and predetermined registration operation It when greater than or equal to 80%, is loaded into user's operation and draws a portrait to the LSTM model, there is LSTM model to be existed according to user's operation portrait Optimal time sequence is confirmed in N number of time series.Another situation, if it does not exist in the consistent photo of user's operation portrait, then User's operation portrait is not classified as feature class.

In another embodiment, computer equipment judges the order change of electronic page, it is to be appreciated that currently existing electronics The means of payment has, and cCredit card payment, wechat payment, Alipay payment, computer equipment first obtain the payment sequence picture of three；When When identifying that user in time series is inclined to the corresponding payment sequence of timing node of purchase product, whether payment sequence is judged Match with the payment sequence of setting, if matching, is loaded into the payment feature to LSTM model, is classified as needs in LSTM model Buy class；If mismatching, the payment feature is also loaded into LSTM model, but hesitation purchase class is classified as in LSTM model.By LSTM model is chosen in N number of time series according to purchase class and the payment sequence for needing to buy class of hesitating and needs to buy class number The most conduct optimal time sequence of mesh.

In another embodiment, when user buys product, computer equipment obtains the audio of user, and is loaded into the audio to sound Frequency library, identical as above-mentioned image lookup mode, computer equipment searches whether there is audio corresponding with audio from audio repository Data, and if it exists, be loaded into audio to LSTM model, if it does not exist, then the audio is not classified as feature class.

Further, the production method of user's operation portrait, comprising:

S604, operating process and user operation time after obtaining user's log-on webpage；

S605, process and user operation time production user's operation are drawn a portrait depending on the user's operation.

It can be seen from the above, predetermined registration operation portrait include confirmation purchase portrait, hesitate purchase portrait, hesitate do not buy portrait and Portrait is not bought, and computer equipment judges whether user buys by user's operation process, picks out by user operation time Whether user hesitates.Such as: user is ordered by terminal, is not desired to purchase suddenly in the terminal interface of purchase checkout and is exited purchase Interface, terminal can judge that user draws a portrait according to this operating process.

In one embodiment, screening is grasped for the matched M user of predetermined registration operation portrait in multiple user's operations portrait The step of drawing a portrait, comprising:

S606 identifies M user for reaching first threshold in multiple user's operation portraits with predetermined registration operation portrait similarity Operation portrait；

S607, the other user's operations deleted lower than first threshold are drawn a portrait.

Purpose is to filter out useful user's operation portrait, and technological means is to pass through user's operation in computer equipment After process and user operation time produce user's operation portrait, judge whether user's operation portrait draws a portrait with preset operation Similarity reaches first threshold, and M user's operation portrait is retained if reaching, and M user's operation portrait is directed into LSTM mould In type.And remaining is then deleted lower than other user's operations portrait of first threshold.

Reference table 1, in one embodiment, the corresponding reality of timing node by optimal time sequence in a cycle are tied Step S700 of the fruit as the prediction result in next period, comprising:

Prediction table is established by User ID, optimal time sequence and operating characteristics；

Analytical Prediction table calculates the prediction result in next period by history actual result, and records real with history The corresponding operating characteristics of border result.

Predicting that table is more intuitive by foundation makes computer equipment derive prediction result, and table is as follows:

Table 1

As seen from table, computer equipment reaches the effect of prediction by the circulation comparison in multiple periods of optimal time sequence Fruit；It is shown with citing, user A1 feature A, B, C occurs in the period 1, and historical data is to have purchased；In optimal time sequence Second round when user there is feature A, B, D, therefore computer equipment is drawn a conclusion, and A, B, C, D have user that can buy Feature；It is yes that period 3 makes prediction referring to the actual result of second round, but practical actual result is no, and operating characteristics It is middle the first and second period no feature E occur, therefore computer equipment record feature E is not buy feature.It is appreciated that the 3rd The actual result in period is different from the 1st, the 2nd period, therefore the 1st, 2,3 establish the new period, therefore derives the 4th period and the 1st week Phase is similar, so that the prediction result that computer equipment exported for the 4th period is the actual result in the 1st period, in this case, it is be.

Show with reference to the process of attached drawing 2, one embodiment of method based on historical data prediction result to be proposed for the application Be intended in one embodiment, by optimal time sequence a cycle the corresponding actual result of timing node as next After the step S700 of the prediction result in period:

S701 judges whether the actual result in next period is identical as prediction result；

It is right into next period to extend the timing node in a cycle of optimal time sequence if not identical by S702 The timing node answered, and establish the new period.

I.e. such as above table it is found that the 1st period was a cycle, being in computer equipment the 2nd period of prediction can be with the 1st week Phase, the 3rd period was then predicted referring to the actual result in the 2nd period, but the practical knot that the 3rd period was recorded as prediction reference Fruit is different from the 2nd period, thus the 1st, the 2nd, the 3rd period establish a new period, with correspondence derived for the 4th period with first week Phase as reference, if the 4th period did not still match that with the 1st period, extends the period to the 1st, the 2nd, the 3rd, the 4th period again For a cycle, in this, as the training method of model, the prediction result of successive optimization computer keeps prediction result more and more quasi- Really.

With reference to attached drawing 3, for the process signal for one embodiment of method based on historical data prediction result that the application proposes Figure, by optimal time sequence a cycle prediction result of the corresponding actual result of timing node as next period, And after exporting the step S700 of prediction result, comprising:

S710, there is the behavioral data of specifies behavior again in get User ID；

S720, importing behavioral data to the library ID, and behavioral data is classified in the history purchase file of User ID.

When computer equipment gets the information of purchase again of the same User ID herein, computer equipment obtains User ID It is corresponding to buy information again, and the history of the User ID in the library ID for buying information categorization again is bought in file, All purchase information for recording user, by the purchase information of user whether to verify prediction result that computer equipment is exported Correctly, such as: the prediction result of computer equipment output is that user will not buy A product, but include buying in information again There is A product, that is, determines the step of prediction result is incorrect, and computer equipment executes above-mentioned steps S44.Another situation, if calculating The prediction result of machine equipment output will not buy B product for user, and not include B product being bought in information again, therefore determine Prediction result is correct, and computer equipment is then calculated according to the prediction result that the optimal cycle stage for realizing sequence does next period.

With reference to attached drawing 4, for a kind of structural block diagram for device based on historical data prediction result that the application proposes, packet It includes:

Face identification unit 1, for obtaining video record, then positioning video video recording in several users face, and Intercept the face picture of several users；

Picture processing unit 2 obtains black-white-gray face picture, then according to black and white for carrying out discoloration to face picture The gray scale depth of grey face picture is aggravated or is desalinated, and is sharpened to the face edge of black-white-gray face picture；

Matching unit 3, for searching whether there is the user face to match with black-white-gray face picture in facial library has Portion's figure, user's face figure are black-white-gray picture；

ID transfers unit 4, for transferring the User ID in the library ID, and obtains User ID label and the history of specifies behavior occurs Data；

Time series acquiring unit 5, for obtain user in historical data occur specifies behavior with different time length For N number of time series of unit, wherein N is the integer more than or equal to 1；

Operating characteristics recognition unit 6, operating characteristics when specifies behavior occurs in user in N number of time series for identification, And it is loaded into operating characteristics to LSTM model, it is optimal to be selected in N time series by LSTM model according to operating characteristics Time series, optimal time sequence are the time series that the timing node that several operating characteristics occur meets periodic law；

Predicting unit 7, for passing through optimal time sequence in the corresponding actual result conduct of timing node of a cycle The prediction result in next period；

Surplus judging unit 8, for transferring reserves data, reserves data include article surplus, and judge that article surplus is The no article demand greater than prediction result；

Inventory production unit 9, for if it is not, then calculate the difference of article surplus Yu article demand, and according to difference system Make shopping list.

It makes explanations explanation in conjunction with computer equipment, when worker needs to enable prediction, is opened to computer equipment input Dynamic instruction, computer equipment generate request column according to enabled instruction, and request column is shown by the screen of computer equipment, requests work The video recording of person's uploaded videos.After computer equipment gets the video record of worker's upload, several users in positioning video recording Face, and the face picture of several users is intercepted, then, gray processing processing (i.e. discoloration) is carried out to face picture, to obtain Black-white-gray face picture, aggravated according to the gray scale depth of black-white-gray face picture or desalinate (i.e. the gray scale the deep, aggravates, Gray scale is shallowly then desalinated), and protrusion is sharpened to face edge, so that black-white-gray face picture clean mark.Have in facial library In search whether there is the user's face figure that matches with black-white-gray face picture, which is all black-white-gray picture, It is appreciated that the EMS memory occupation amount of face picture is greater than the EMS memory occupation amount of black-white-gray face picture, therefore computer equipment matches Black-white-gray face picture and the speed of user's face figure are better than the speed of matching face picture and user's face figure, i.e. matching efficiency It is promoted；If it exists, then the User ID with user's face figure pre-association is transferred from the library ID, the library ID is that record has user's purchasing history And the repository of Customs Assigned Number, such as: when user buys A product, computer equipment records the product information of A product, record Current time node when user buys A product and be 1 by Customs Assigned Number, i.e. User ID are 1 to buy in timing node A product.Computer equipment calling and obtaining user ID from the library ID, while historical data corresponding with User ID is obtained, according to history number N number of time series is obtained according to the timing node length that the user recorded buys product, i.e., for example: user purchases respectively in 30 days 3 A products are bought, then are obtained using month as the time series of time shaft；Or user buys 3 B products in 1 day, then obtains The time series of one day time span；It is further illustrated by the time series of time shaft of month, user is at first 3 A products are bought in 30 days respectively, buy 2 A products respectively in 30 days at second, i.e. computer equipment gets month Time series, speculate next 30 days product purchase volumes within 30 days within 30 days and second by first, but at first 30 3 A products that user buys in it, then can be with all settling times respectively in first week, second week and third Zhou Jinhang purchase Sequence results in N number of time series occur in this way；When user buys product because the selection purchase opportunity of user is different, purchases Buy product difference, purchase number difference can make computer equipment get N number of time series.

In one embodiment, operating characteristics recognition unit includes:

Image collection module includes user's operation portrait for operating characteristics, obtains each period in N number of time series User's operation buys the corresponding multiple user's operation portraits of timing node of product；

Optical sieving module, for being screened in multiple user's operations portrait for the matched M use of predetermined registration operation portrait Family operation portrait, M are the integer more than or equal to 1, and predetermined registration operation portrait includes confirmation purchase portrait, the purchase portrait that hesitates, hesitates It does not buy portrait and does not buy portrait；

GetImage module is drawn a portrait for being loaded into M user's operation to LSTM model.

Further, image collection module includes:

Data acquisition submodule, operating process and user's operation after obtaining webpage of the login of user for buying product Time；

Image making submodule, process and user operation time production user's operation are drawn a portrait depending on the user's operation.

In one embodiment, optical sieving module includes:

Threshold value identifies submodule, identifies in multiple user's operation portraits and reaches first threshold with predetermined registration operation portrait similarity M user's operation portrait；

Image-erasing submodule, the other user's operations deleted lower than first threshold are drawn a portrait.

In one embodiment, the device based on historical data prediction result further include:

Whether as a result judging unit, the actual result for judging next period are identical as prediction result；

Cycle stretch-out unit, if extending timing node in a cycle of optimal time sequence extremely for not identical Corresponding timing node in next period, and establish the new period.

1st period was a cycle, predicted that the 2nd period was meeting using the 1st period as prediction reference in computer equipment, the 3 periods were then predicted referring to the actual result in the 2nd period, but the actual result that the 3rd period was recorded is different from the 2nd period, Therefore the 1st, the 2nd, the 3rd period establish a new period, with it is corresponding derive the 4th period using period 1 as referring to, if the 4th Period does not still match that then extending period to the 1st, the 2nd, the 3rd, the 4th period again is a cycle, with this with the 1st period As the training method of model, the prediction result of successive optimization computer keeps prediction result more and more accurate.

Unit is bought again, occurs the behavioral data of specifies behavior again for get User ID；

Taxon buys text for importing behavioral data to the library ID, and by the history that behavioral data is classified to User ID In part folder.

Referring to Fig. 5, a kind of terminal device is also provided in the embodiment of the present application, which can be server, in Portion's structure can be as shown in Figure 5.The terminal device include by system bus connect processor, memory, network interface and Database.Wherein, the processor of the Computer Design is for providing calculating and control ability.The memory of the terminal device includes Non-volatile memory medium, built-in storage.The non-volatile memory medium is stored with operating system, computer program and data Library.The built-in storage provides environment for the operation of operating system and computer program in non-volatile memory medium.The terminal The database of equipment is for data such as store historical datas.The network interface of the terminal device is used to pass through net with external terminal Network connection communication.To realize a kind of method based on historical data prediction result when the computer program is executed by processor.

Above-mentioned processor executes the step of above-mentioned method based on historical data prediction result, comprising:

Obtain video record, then in positioning video video recording several users face, and intercept the people of several users Face picture；

Discoloration is carried out to face picture, obtains black-white-gray face picture, it is then deep according to the gray scale of black-white-gray face picture It is shallowly aggravated or is desalinated, and the face edge of black-white-gray face picture is sharpened；

Search whether there is the user's face figure to match with black-white-gray face picture, user's face figure in facial library has For black-white-gray picture；

If it exists, then the User ID with user's face figure pre-association is transferred in the library ID, and obtains User ID label and occur The historical data of specifies behavior；

It obtains user in historical data and N number of time series using different time length as unit of specifies behavior occurs, Wherein, N is the integer more than or equal to 1；

It identifies operating characteristics when specifies behavior occurs in user in N number of time series, and is loaded into operating characteristics to LSTM mould Type, to select optimal time sequence in N number of time series according to operating characteristics by LSTM model, optimal time sequence is The timing node that several operating characteristics occur meets the time series of periodic law；

By optimal time sequence a cycle prediction of the corresponding actual result of timing node as next period As a result；

Reserves data are transferred, reserves data include article surplus, and judge whether article surplus is greater than the object of prediction result Product demand；

If it is not, then calculating the difference of article surplus Yu article demand, and shopping list is made according to difference.

In one embodiment, above-mentioned processor executes behaviour when specifies behavior occurs in user in the N number of time series of identification Make feature, and the step of being loaded into operating characteristics to LSTM model, comprising:

Operating characteristics include user's operation portrait, and specifies behavior occurs in the user for obtaining each period in N number of time series The corresponding multiple user's operations portrait of timing node；

It filters out in multiple user's operations portrait and draws a portrait with the matched M user's operation of predetermined registration operation portrait, M is big In the integer for being equal to 1, predetermined registration operation portrait includes confirmation purchase portrait, the purchase portrait that hesitates, hesitates and do not buy portrait and do not purchase Buy portrait；

M user's operation is loaded into draw a portrait to LSTM model.

In one embodiment, the production method that above-mentioned processor executes user's operation portrait, comprising:

Process and user operation time production user's operation portrait depending on the user's operation.

In one embodiment, above-mentioned processor executes the screening in multiple user's operations portrait and draws a portrait for predetermined registration operation The step of matched M user's operation is drawn a portrait, comprising:

Identify that M user's operation for reaching first threshold with predetermined registration operation portrait similarity in multiple user's operation portraits is drawn Picture；

The other user's operations deleted lower than first threshold are drawn a portrait.

In one embodiment, above-mentioned processor executes corresponding in the timing node of a cycle by optimal time sequence Prediction result of the actual result as next period the step of after, further includes:

Judge whether the actual result in next period is identical as prediction result；

If not identical, extend the timing node in a cycle of optimal time sequence it is corresponding into next period when Intermediate node, and establish the new period.

There is the behavioral data of specifies behavior again in get User ID；

Behavioral data is imported to the library ID, and behavioral data is classified in the history purchase file of User ID.

It will be understood by those skilled in the art that structure shown in Fig. 5, only part relevant to application scheme is tied The block diagram of structure does not constitute the restriction for the terminal device being applied thereon to application scheme.

One embodiment of the application also provides a kind of terminal readable storage medium storing program for executing, is stored thereon with computer program, computer The step of a kind of method based on historical data prediction result is realized when program is executed by processor, specifically:

M user's operation is loaded into draw a portrait to LSTM model.

There is the behavioral data of specifies behavior again in get User ID；

To sum up, by obtaining video record, then in positioning video video recording several users face, and intercept several The face picture of user；Discoloration is carried out to face picture, black-white-gray face picture is obtained, then according to black-white-gray face picture The gray scale depth is aggravated or is desalinated, and is sharpened to the face edge of black-white-gray face picture；It is looked into facial library has It looks for the presence or absence of the user's face figure to match with black-white-gray face picture, user's face figure is black-white-gray picture；If it exists, then The User ID with user's face figure pre-association is transferred in the library ID, and obtains the historical data of User ID, is obtained according to historical data Getting user and buying the different time length of product is N number of time series of unit, and user is inclined to purchase in recognition time sequence The corresponding operating characteristics of the timing node of product, and operating characteristics are loaded into LSTM model, to pass through LSTM model according to operation Feature selects optimal time sequence in N number of time series, and finally, terminal device is by optimal time sequence in a cycle Prediction result of the corresponding actual result of timing node as next period, and shopping list is made according to prediction result；From And it solves to be manually entered timing node low efficiency, and the technical issues of solve to be likely to occur a variety of different prediction results, and seek The statistics available multiple prediction results of quotient are sold, to confirm accurately order volume.

Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with Instruct relevant hardware to complete by computer program, computer program can store it is readable with a non-volatile computer It takes in storage medium, the computer program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein, this Shen Please provided by and embodiment used in any reference to memory, storage, database or other media, may each comprise Non-volatile and/or volatile memory.Nonvolatile memory may include read-only memory (ROM), programming ROM (PROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM) or flash memory.Volatile memory may include Random access memory (RAM) or external cache.By way of illustration and not limitation, RAM can by diversified forms , such as static state RAM (SRAM), dynamic ram (DRAM), synchronous dram (SDRAM), double speed are according to rate SDRAM (SSRSDRAM), increasing Strong type SDRAM (ESDRAM), synchronization link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic ram (DRDRAM) and memory bus dynamic ram (RDRAM) etc..

It should be noted that, in this document, the terms "include", "comprise" or its any other variant are intended to non-row His property includes, so that the process, device, article or the method that include a series of elements not only include those elements, and And further include the other elements being not explicitly listed, or further include for this process, device, article or method institute it is intrinsic Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including being somebody's turn to do There is also other identical elements in the process, device of element, article or method.

The foregoing is merely preferred embodiment of the present application, are not intended to limit the scope of the patents of the application, all utilizations Equivalent structure or equivalent flow shift made by present specification and accompanying drawing content is applied directly or indirectly in other correlations Technical field, similarly include in the scope of patent protection of the application.

While there has been shown and described that embodiments herein for the ordinary skill in the art can be with A variety of variations, modification, replacement can be carried out in the case where not departing from the principle and spirit of the application to these embodiments by understanding And modification, scope of the present application are defined by the appended claims and the equivalents thereof.

Claims

1. a kind of method based on historical data prediction result characterized by comprising

Video record is obtained, then positions the face of several users in the video record, and intercept the people of several users Face picture；

Discoloration is carried out to the face picture, black-white-gray face picture is obtained, then according to the ash of the black-white-gray face picture The degree depth is aggravated or is desalinated, and is sharpened to the face edge of black-white-gray face picture；

Search whether there is the user's face figure to match with the black-white-gray face picture, the user face in facial library has Portion's figure is black-white-gray picture；

If it exists, then the User ID with the user's face figure pre-association is transferred in the library ID, and obtains User ID label and occur The historical data of specifies behavior；

It obtains user in the historical data and N number of time series using different time length as unit of specifies behavior occurs, Wherein, the N is the integer more than or equal to 1；

It identifies operating characteristics when specifies behavior occurs in user in N number of time series, and is loaded into the operating characteristics extremely LSTM model, to select optimal time in N number of time series according to the operating characteristics by the LSTM model Sequence, the optimal time sequence are the time sequence that the timing node that several described operating characteristics occur meets periodic law Column；

By the optimal time sequence a cycle the corresponding actual result of the timing node as next period Prediction result；

Reserves data are transferred, the reserves data include article surplus, and judge whether the article surplus is greater than the prediction As a result article demand；

If it is not, then calculating the difference of the article surplus Yu the article demand, and shopping list is made according to the difference.

2. the method according to claim 1 based on historical data prediction result, which is characterized in that the identification N Operating characteristics when specifies behavior occurs in user in a time series, and the step of being loaded into the operating characteristics to LSTM model, Include:

The operating characteristics include user's operation portrait, and the user for obtaining each period in N number of time series specifies The corresponding multiple user's operation portraits of the timing node of behavior；

It filters out in the multiple user's operation portrait and draws a portrait with the matched M user's operation of predetermined registration operation portrait, the M is Integer more than or equal to 1, the predetermined registration operation portrait include confirmation purchase portrait, the purchase portrait that hesitates, hesitate and do not buy portrait Portrait is not bought；

The M user's operation is loaded into draw a portrait to the LSTM model.

3. the method according to claim 2 based on historical data prediction result, which is characterized in that the user's operation is drawn The production method of picture, comprising:

4. the method according to claim 2 based on historical data prediction result, which is characterized in that described the multiple The step of screening is for predetermined registration operation portrait matched M user's operation portrait in user's operation portrait, comprising:

Identify the M use for reaching first threshold in the multiple user's operation portrait with predetermined registration operation portrait similarity Family operation portrait；

5. the method according to claim 1 based on historical data prediction result, which is characterized in that it is described by it is described most Excellent time series is prediction result of the corresponding actual result of the timing node as next period of a cycle the step of Later, further includes:

If not identical, extend the timing node in a cycle of the optimal time sequence it is corresponding into next period when Intermediate node, and establish the new period.

6. the method according to claim 1 based on historical data prediction result, which is characterized in that it is described by it is described most Excellent time series is prediction result of the corresponding actual result of the timing node as next period of a cycle the step of Later, further includes:

There is the behavioral data of specifies behavior again in get User ID；

The behavioral data is imported to the library ID, and the history that the behavioral data is classified to the User ID is bought into file In folder.

7. a kind of device based on historical data prediction result characterized by comprising

Then face identification unit positions the face of several users in the video record, and cut for obtaining video record Take the face picture of several users；

Picture processing unit obtains black-white-gray face picture, then according to described black for carrying out discoloration to the face picture The gray scale depth of lime face picture is aggravated or is desalinated, and is sharpened to the face edge of black-white-gray face picture；

Matching unit, for searching whether there is the user's face to match with the black-white-gray face picture in facial library has Figure, the user's face figure is black-white-gray picture；

ID transfers unit, for if it exists, then transferring the User ID with the user's face figure pre-association in the library ID, and obtain User ID marks the historical data for specifies behavior occur；

There is making with different time length for specifies behavior for obtaining user in the historical data in time series acquiring unit For N number of time series of unit, wherein the N is the integer more than or equal to 1；

Operating characteristics recognition unit, operating characteristics when specifies behavior occurs in user in N number of time series for identification, and The operating characteristics are loaded into LSTM model, with by the LSTM model according to the operating characteristics in N number of time sequence Optimal time sequence is selected in column, the optimal time sequence is that the timing node that several described operating characteristics occur meets The time series of periodic law；

Predicting unit is made for the corresponding actual result of the timing node by the optimal time sequence in a cycle For the prediction result in next period；

Surplus judging unit, for transferring reserves data, the reserves data include article surplus, and judge the article surplus Whether the article demand of the prediction result is greater than；

Inventory production unit, for if it is not, then calculate the difference of the article surplus Yu the article demand, and according to described Difference makes shopping list.

8. the device according to claim 7 based on historical data prediction result, which is characterized in that the operating characteristics are known Other unit includes:

Image collection module includes user's operation portrait for the operating characteristics, obtains each week in N number of time series There are the corresponding multiple user's operation portraits of timing node of specifies behavior in the user of phase；

Optical sieving module, for being filtered out in the multiple user's operation portrait and the matched M use of predetermined registration operation portrait Family operation portrait, the M are the integer more than or equal to 1, and the predetermined registration operation portrait includes confirmation purchase portrait, hesitate purchase picture Picture hesitates not buy and draws a portrait and do not buy portrait；

9. a kind of terminal device, including memory and processor, it is stored with computer program in the memory, feature exists In the processor is realized described in any one of claims 1 to 6 pre- based on historical data when executing the computer program The step of surveying the method for result.

10. a kind of terminal readable storage medium storing program for executing, is stored thereon with computer program, which is characterized in that the computer program quilt The step of processor realizes the method described in any one of claims 1 to 6 based on historical data prediction result when executing.