CN110516075A - Early warning report-generating method, device and computer equipment based on machine learning - Google Patents
Early warning report-generating method, device and computer equipment based on machine learning Download PDFInfo
- Publication number
- CN110516075A CN110516075A CN201910662226.0A CN201910662226A CN110516075A CN 110516075 A CN110516075 A CN 110516075A CN 201910662226 A CN201910662226 A CN 201910662226A CN 110516075 A CN110516075 A CN 110516075A
- Authority
- CN
- China
- Prior art keywords
- data
- numerical value
- preset
- prediction
- difference
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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"
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Strategic Management (AREA)
- Evolutionary Computation (AREA)
- Human Resources & Organizations (AREA)
- Computational Linguistics (AREA)
- Computing Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Economics (AREA)
- Animal Behavior & Ethology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Development Economics (AREA)
- Health & Medical Sciences (AREA)
- Game Theory and Decision Science (AREA)
- Medical Informatics (AREA)
- Biophysics (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
This application discloses a kind of early warning report-generating method, device, computer equipment and storage medium based on machine learning, which comprises crawl primary data from information source, obtain specified data；The specified data are inputted preset having trained in the prediction model completed based on machine learning to calculate, to obtain the first prediction numerical value；According to the specified data, numerical value is predicted using in preset knowledge mapping, obtaining second；Use formula: W=paA+pbB calculates final prediction numerical value W；Calculate the difference of final prediction numerical value W with the prediction numerical value for compareing object；If difference is not in preset difference range, specified data are inputted in preset data differences level calculation model with the data for compareing object then and are calculated, the data differences class value of the data differences level calculation model output is obtained, and the specified data that the data differences class value is greater than level threshold are denoted as suspicion data；Generate early warning report.
Description
Technical field
This application involves computer field is arrived, a kind of early warning report generation side based on machine learning is especially related to
Method, device, computer equipment and storage medium.
Background technique
It is also higher and higher for the importance of following early warning with the continuous development of information-based, intelligent society.In example
In in terms of there are many such as finance, how accurately to predict following situation and carry out early warning all to be to need urgently to solve the problems, such as.
Be usually used in traditional technology and artificially empirically predict, or using single prediction model predicted to carry out again it is pre-
Alert, due to there is a situation where that the data for being used for prediction are improper, model is single, traditional technology is carried out using single prediction model
It predicts and the result accuracy of early warning is to be improved.And traditional technology can not early warning technology be generally basede on pre-set number
Value belongs to mechanical static early warning, can not accomplish dynamic early-warning.
Summary of the invention
The main purpose of the application is to provide a kind of early warning report-generating method, device, computer based on machine learning
Equipment and storage medium, it is intended to improve forecasting accuracy and realize dynamic early-warning.
In order to achieve the above-mentioned object of the invention, the application proposes a kind of early warning report-generating method based on machine learning, answers
For Forewarning Terminal, comprising the following steps:
Using preset crawler technology, primary data is crawled from preset information source, and uses preset noise reduction algorithm
Noise reduction process is carried out to the primary data, so that specified data are obtained, wherein the information source includes at least default website；
The specified data are inputted preset having trained in the prediction model completed based on machine learning to calculate,
To obtain the first prediction numerical value of the prediction model output；Wherein, the prediction model based on the specified data
The identical historical data of type, and formed with the associated prediction numerical value training of the historical data；
It is obtained according to the specified data using the relationship that influences each other of each knowledge node in preset knowledge mapping
Second prediction numerical value of the knowledge mapping output, wherein the knowledge mapping is known including at least corresponding with the specified data
Know node；
Use formula: W=paA+pbB calculates final prediction numerical value W, and wherein A is the first prediction numerical value, and B is described the
Two prediction numerical value, pa、pbThe weight parameter of the respectively described first prediction numerical value A, the second prediction numerical value B；
The difference of the final prediction numerical value W with the preset prediction numerical value for compareing object are calculated, and judges the difference
Whether in preset difference range, wherein data prediction of the prediction numerical value of the control object based on the control object
And obtain, the data of the control object are corresponded to each other with the specified data；
If the difference is not in preset difference range, by the specified data and the data for compareing object
It inputs in preset data differences level calculation model and is calculated, to obtain the data differences level calculation model output
Data differences class value, and by the data differences class value be greater than preset level threshold specified data be denoted as suspicion number
According to；
Early warning report is generated, wherein having the suspicion data in precaution alarm announcement.
The application provides a kind of early warning report preparing apparatus based on machine learning, is applied to Forewarning Terminal, comprising:
Specified data capture unit, for crawling primary data from preset information source using preset crawler technology,
And noise reduction process is carried out to the primary data using preset noise reduction algorithm, so that specified data are obtained, wherein the information
Source includes at least default website；
First prediction numerical value acquiring unit, for the specified data to be inputted preset having trained based on machine learning
It is calculated in the prediction model of completion, to obtain the first prediction numerical value of the prediction model output；Wherein, the prediction
Model is instructed based on historical data identical with the type of the specified data, and with the associated prediction numerical value of the historical data
White silk forms；
Second prediction numerical value acquiring unit, for utilizing respectively knowing in preset knowledge mapping according to the specified data
The relationship that influences each other for knowing node obtains the second prediction numerical value of the knowledge mapping output, wherein the knowledge mapping is at least
Including knowledge node corresponding with the specified data；
Final prediction numerical value acquiring unit, for using formula: W=paA+pbB calculates final prediction numerical value W, and wherein A is
The first prediction numerical value, B are the second prediction numerical value, pa、pbThe respectively described first prediction numerical value A, second prediction
The weight parameter of numerical value B；
Difference computational unit, for calculating the difference of the final prediction numerical value W with the preset prediction numerical value for compareing object
Value, and judge whether the difference is in preset difference range, wherein the prediction numerical value of the control object is based on described
The data of control object predict and obtain that the data of the control object are corresponded to each other with the specified data；
Suspicion data capture unit, if being not in preset difference range for the difference, by the specified number
It is calculated according to being inputted in preset data differences level calculation model with the data for compareing object, to obtain the number
It is greater than preset rank threshold according to the data differences class value that difference level calculation model exports, and by the data differences class value
The specified data of value are denoted as suspicion data；
Generation unit is reported in early warning, for generating early warning report, wherein having the suspicion data in precaution alarm announcement.
The application provides a kind of computer equipment, including memory and processor, and the memory is stored with computer journey
The step of sequence, the processor realizes any of the above-described the method when executing the computer program.
The application provides a kind of computer readable storage medium, is stored thereon with computer program, the computer program
The step of method described in any of the above embodiments is realized when being executed by processor.
Early warning report-generating method, device, computer equipment and the storage medium based on machine learning of the application, passes through
Primary data is crawled, and carries out noise reduction process, obtains specified data；By the specified data input in preset prediction model into
Row calculates, to obtain the first prediction numerical value；Using the relationship that influences each other of each knowledge node in preset knowledge mapping, obtain
Take the second prediction numerical value of the knowledge mapping output；Use formula: W=paA+pbB calculates final prediction numerical value W；Described in calculating
The difference of final prediction numerical value W and the preset prediction numerical value for compareing object；If the difference is not at preset difference range
It is interior, then obtain the data differences class value of data differences level calculation model output, and by the data differences class value
Specified data greater than preset level threshold are denoted as suspicion data；Early warning report is generated, wherein having in precaution alarm announcement
The suspicion data.To improve forecasting accuracy and realize dynamic early-warning.
Detailed description of the invention
Fig. 1 is the flow diagram of the early warning report-generating method based on machine learning of one embodiment of the application；
Fig. 2 is the structural schematic block diagram of the early warning report preparing apparatus based on machine learning of one embodiment of the application；
Fig. 3 is the structural schematic block diagram of the computer equipment of one embodiment of the application.
The embodiments will be further described with reference to the accompanying drawings for realization, functional characteristics and the advantage of the application purpose.
Specific embodiment
It is with reference to the accompanying drawings and embodiments, right in order to which the objects, technical solutions and advantages of the application are more clearly understood
The application is further elaborated.It should be appreciated that specific embodiment described herein is only used to explain the application, not
For limiting the application.
Referring to Fig.1, the embodiment of the present application provides a kind of early warning report-generating method based on machine learning, is applied to early warning
Terminal, comprising:
S1, using preset crawler technology, primary data is crawled from preset information source, and calculate using preset noise reduction
Method carries out noise reduction process to the primary data, so that specified data are obtained, wherein the information source includes at least default website；
S2, by the specified data input it is preset by machine learning trained complete prediction model in carry out based on
It calculates, to obtain the first prediction numerical value of the prediction model output；Wherein, the prediction model is based on and the specified data
The identical historical data of type, and formed with the historical data associated prediction numerical value training；
S3, it is obtained according to the specified data using the relationship that influences each other of each knowledge node in preset knowledge mapping
The the second prediction numerical value for taking the knowledge mapping output, wherein the knowledge mapping is including at least corresponding with the specified data
Knowledge node；
S4, formula is used: W=paA+pbB calculates final prediction numerical value W, and wherein A is the first prediction numerical value, and B is institute
State the second prediction numerical value, pa、pbThe weight parameter of the respectively described first prediction numerical value A, the second prediction numerical value B；
S5, the difference for calculating the final prediction numerical value W with the preset prediction numerical value for compareing object, and judge the difference
Whether value is in preset difference range, wherein data of the prediction numerical value of the control object based on the control object are pre-
It surveys and obtains, the data of the control object are corresponded to each other with the specified data；
If S6, the difference are not in preset difference range, by the specified data and the object that compares
Data are inputted in preset data differences level calculation model and are calculated, to obtain the data differences level calculation model
The data differences class value of output, and the specified data that the data differences class value is greater than preset level threshold are denoted as suspicion
Doubt data；
S7, early warning report is generated, wherein having the suspicion data in precaution alarm announcement.
As described in above-mentioned steps S1, using preset crawler technology, primary data is crawled from preset information source, and make
Noise reduction process is carried out to the primary data with preset noise reduction algorithm, so that specified data are obtained, wherein the information source is extremely
Few includes default website.The application uses Forewarning Terminal as executing subject, wherein the Forewarning Terminal can be any terminal,
Including client, server etc..The preset crawler technology can be any technology, for example, by using the Scrapy of Python
Frame crawls primary data in information source.The information source is, for example, default website, software APP or database etc..It is described
Primary data is can be described current as the data of the basis of prediction, such as current financial data or current production price
Financial data includes social security data, major project data etc..Wherein the default website can be any website, preferably obtain
The government website (website of the digital certificate of the representative government organs provided with certificate authority certification authority) of certification.It uses
Preset noise reduction algorithm carries out noise reduction process to the primary data, to obtain specified data.It is initial due to what is directly crawled
Data are the firsthand data, and quantity is numerous and jumbled, inevitably have the problem of repeated data and inexact data, therefore the application uses in advance
If noise reduction algorithm to the primary data carry out noise reduction process.The noise reduction algorithm is for example are as follows: is derived from different data sources in the future
The numerical value of same primary data form specified numerical value group, calculated using preset formula of variance each in the specified numerical value group
The variance of primary data, judges whether the variance of each primary data in the specified numerical value group is respectively less than preset variance threshold
Value, if the variance of each primary data is not respectively less than preset variance threshold values in the specified numerical value group, not by the variance
Primary data less than preset variance threshold values as noise and is removed processing.
As described in above-mentioned steps S2, by the specified data input it is preset based on machine learning having trained complete it is pre-
It surveys in model and is calculated, to obtain the first prediction numerical value of the prediction model output；Wherein, the prediction model is based on
Historical data identical with the type of the specified data, and formed with the associated prediction numerical value training of the historical data.
Wherein the prediction model is for exporting the first prediction numerical value, to characterize to future.Wherein, first prediction
Numerical value is, for example, to predict financial data (budget etc.), and further, the first prediction numerical value is, for example, to predict that financial data is reflected
Numerical value made of penetrating.The prediction model is, for example, neural network model, support vector machines, decision tree etc., and the application is preferably refreshing
Through network model, comprising: VGG19 model, VGG-F model, ResNet50 model, DPN131 model, InceptionV3 model,
Xception model and AlexNet model etc., the application more preferably use DPN model, and DPN (Dual Path Network) is
A kind of neural network structure is the core content that DenseNet is introduced on the basis of ResNeXt, so that model is to feature
Using more sufficiently.Above-mentioned DPN, ResNeXt and DenseNet are existing network structures, are not being repeated herein.Wherein train
Data include historical data identical with the type of the specified data, and with the associated prediction numerical value of the historical data.
Due to the type of the historical data identical as the type of the specified data (such as the prices of same commodity, the same area
Permanent resident population's quantity), therefore the specified data are inputted into the prediction model, it directly can accurately obtain the first prediction number
Value.
As described in above-mentioned steps S3, according to the specified data, each knowledge node in preset knowledge mapping is utilized
Influence each other relationship, the second prediction numerical value of the knowledge mapping output is obtained, wherein the knowledge mapping includes at least and institute
State the corresponding knowledge node of specified data.Knowledge mapping is a series of various differences of explicit knowledge's development process and structural relation
Figure, with multiple knowledge nodes (such as a certain commodity, some region of permanent resident population, estimated financial data etc.), and
The relationship that influences each other comprising the multiple knowledge node.Knowledge mapping establishes in economic industry chain between different knowledge nodes
Relationship, for the relationship in building, attribute can establish prediction link model with the attribute of associated nodes, including by demand pull
Model and cost conduction model, such as the price raising of industry upstream product can influence the production in downstream by cost conduction model
Product price.Accordingly, according to the specified data, using the relationship that influences each other of each knowledge node in preset knowledge mapping,
Obtain the second prediction numerical value of the knowledge mapping output.The wherein class of the second prediction numerical value and the first prediction numerical value
Type is identical (being for example all the mapping value of the estimated expenditure of finance).
As described in above-mentioned steps S4, formula is used: W=paA+pbB calculates final prediction numerical value W, and wherein A is described first
Predict that numerical value, B are the second prediction numerical value, pa、pbThe respectively described first prediction numerical value A, described second predict numerical value B's
Weight parameter.For the accuracy and stability for guaranteeing prediction numerical value, the application uses formula: W=paA+pbB calculates final prediction
Numerical value W, wherein A is the first prediction numerical value, and B is the second prediction numerical value, pa、pbThe respectively described first prediction numerical value
A, the weight parameter of the second prediction numerical value B.To guarantee the accurate of prediction numerical value in the way of the output of dual model weight
Property and stability.Wherein, the weight parameter p of the first prediction numerical value A, the second prediction numerical value Ba、pbIt can be by corresponding
The historical forecast accuracy of prediction model and the knowledge mapping is accordingly arranged, the higher corresponding weight ginseng of historical forecast accuracy
Several numerical value is bigger, otherwise smaller.
As described in above-mentioned steps S5, the difference of the final prediction numerical value W with the preset prediction numerical value for compareing object are calculated
Value, and judge whether the difference is in preset difference range, wherein the prediction numerical value of the control object is based on described
The data of control object predict and obtain that the data of the control object are corresponded to each other with the specified data.Wherein compare object
It is the object equity predicted with the application, such as the application is used to predict the economic numerical value in a city, then compareing
Object is another city, and the prediction numerical value for compareing object is the economic numerical value in another city.The application use with
Compare object comparing result to determine whether should early warning, realize dynamic early-warning, facilitate analyze two objects between
Dynamic difference.Wherein the prediction numerical value of the control object is predicted based on the data of the control object and is obtained, such as can be adopted
It is predicted and is obtained with acquisition methods identical with the final prediction numerical value W.The data of the control object and the specified number
Refer to that the data of the control object are identical as the type of the specified data according to corresponding to each other, such as specified data include area
Total output value, general public budget income, total import and export value etc., then the data of the control object also include same type
Data, to improve the accuracy of the progress between two objects.
As described in above-mentioned steps S6, if the difference is not in preset difference range, by the specified data with
The data of the control object, which input in preset data differences level calculation model, to be calculated, to obtain the data difference
The data differences class value of other level calculation model output, and the data differences class value is greater than preset level threshold
Specified data are denoted as suspicion data.If the difference is not in preset difference range, show the prediction object of the application with
The difference of object is compareed beyond expection, therefore predicts that the data used are corresponding, it can thus be assumed that prediction object is specified
There are suspicion data in data, the situation that the suspicion data are not in preset difference range the difference has biggish
It influences.Wherein by the way that the specified data are inputted preset data differences level calculation model with the data for compareing object
In calculated, to obtain the data differences class value of data differences level calculation model output, and by the data
The specified data that difference class value is greater than preset level threshold are denoted as the mode of suspicion data, in the determination specified data
Suspicion data.The specified data are wherein inputted into preset data differences level calculation mould with the data for compareing object
The mode calculated in type is for example are as follows: judges whether the specified data belong to same amount with the data for compareing object
Grade；If the specified data belong to same magnitude with the data for compareing object, it is described right that the specified data are subtracted
Data according to object are to obtain data difference, according to the mapping relations of preset data difference and data differences rank, output
Data differences class value；If the specified data are not belonging to same magnitude with the data for compareing object, according to formula: number
The data that data-lg control object is specified according to difference=lg, calculate data difference, and according to preset data difference and data
The mapping relations of difference rank, output data difference class value.
As described in above-mentioned steps S7, early warning report is generated, wherein having the suspicion data in precaution alarm announcement.Due to
The situation that suspicion data are not in preset difference range the difference has large effect, therefore reports in the early warning
In have the suspicion data.Further, in order to provide more accurate information reference, the early warning report be accompanied by with it is described
The corresponding historical events of suspicion data, wherein the historical events in preset database by inquiring the suspicion data
Type prestores the type historical events corresponding with the type of the data of data to obtain in the database.
In one embodiment, described to use preset crawler technology, primary data is crawled from preset information source,
And the step S1 of noise reduction process is carried out to the primary data using preset noise reduction algorithm, comprising:
S101, the Scrapy frame using Python crawl primary data in default website；
S102, the numerical value of same kind of primary data is formed into specified numerical value group, and uses preset formula:Calculate the population variance of m-th of primary data in the specified numerical value groupWherein N is described
The sum of specified numerical value in specified numerical value group, Am are the numerical value of m-th of primary data, and B is being averaged for the specified numerical value group
Value；
S103, judge the population varianceWhether preset variance threshold values are respectively less than；
If S104, the population variancePreset variance threshold values are not respectively less than, then by the population varianceIt is not less than
The primary data of preset variance threshold values is as noise and is removed processing.
Noise reduction process is carried out to the primary data using preset noise reduction algorithm as described above, realizing, to obtain
Specified data.Wherein the Scrapy frame of the Python is the effective means for crawl in default website information,
Specifically include that engine, scheduler, downloader, crawler, project pipeline, downloader middleware, crawler middleware, scheduling middleware
Deng.Specifically crawling process includes: that engine takes out a link for next crawl from scheduler；Engine encapsulates link
Downloader is transmitted at a request；Downloader gets off resource downloading；Crawler parses entity, gives entity pipeline and carries out into one
The processing of step.Since, there may be inexact data, the application uses preset formula in the numerical value that crawls:Calculate the population variance of m-th of primary data in the specified numerical value groupJudge the totality
VarianceWhether preset variance threshold values are respectively less than；If the population variancePreset variance threshold values are not respectively less than, then will
The population variancePrimary data not less than preset variance threshold values as noise and is removed processing.To avoid
The problem of inexact data bring prediction incorrectness.
In one embodiment, described input the specified data preset has trained completion based on machine learning
Prediction model in calculated, to obtain the first prediction numerical value of prediction model output；Wherein, the prediction model
Based on historical data identical with the type of the specified data, and with the training of the historical data associated prediction numerical value and
At step S2 before, comprising:
S11, the sample data for obtaining specified amount, and the sample data is divided into training set and test set；Wherein, described
Sample data includes historical data identical with the type of the specified data, and with the associated prediction number of the historical data
Value；
S12, it the sample data of training set is input in preset neural network model is trained；Wherein, it trains
Stochastic gradient descent method is used in the process, obtains initial training model；
S13, the initial training model is verified using the sample data of the test set；
S14, if the verification passes, then be denoted as the prediction model for the initial training model.
As described above, realizing trained prediction model.The application is using neural network model as machine learning model, example
It such as include: VGG19 model, VGG-F model, ResNet50 model, DPN131 model, InceptionV3 model, Xception mould
Type and AlexNet model etc., the application more preferably use DPN model.Before training, sample data is carried out being randomly divided into two
Group set, i.e. training set and test set, are then trained above-mentioned neural network model by the sample data of training set.Instruction
Practice after completing to obtain result training pattern, the result training pattern is verified by the sample data of test set, with judging result
Whether training pattern can be used.Wherein, stochastic gradient descent method is exactly some training datas of grab sample, substitutes entire training set,
Training speed can be improved.Further, the ginseng of each layer of the neural network model can also be updated using reverse conduction rule
Number, it is established on the basis of gradient descent method the reverse conduction rule (BP), and the input/output relation of BP network is substantial
Be a kind of mapping relations: the function that the BP neural network of a n input m output is completed is to tie up Euclidean space from n to tie up Euclidean to m
The Continuous Mappings of a finite field in space, this mapping has nonlinearity, to realize the ginseng of each layer of neural network model
Several updates.
In one embodiment, described to use formula: W=paA+pbB calculates final prediction numerical value W, and wherein A is described
First prediction numerical value, B are the second prediction numerical value, pa、pbThe respectively described first prediction numerical value A, the second prediction numerical value
Before the step S4 of the weight parameter of B, comprising:
The second history that S31, the first historical forecast numerical value, the knowledge mapping for obtaining the prediction model output export
Predict numerical value and history actual value；
S32, pass through formula: the first prediction deviation value=| the first historical forecast numerical value-history actual value |, calculate institute
State the first prediction deviation value；
S33, pass through formula: the second prediction deviation value=| the second historical forecast numerical value-history actual value |, calculate institute
State the second prediction deviation value；
S34, according to the first prediction deviation value and the second prediction deviation value, using preset prediction deviation value with
Weight parameter corresponding relationship obtains weight parameter corresponding with the first prediction numerical value A, the second prediction numerical value B respectively
pa、pb。
Join as described above, realizing acquisition weight corresponding with the first prediction numerical value A, the second prediction numerical value B
Number pa、pb.If the output numerical value of prediction model is more accurate, then corresponding weight parameter is higher, to make finally to predict numerical value
The accuracy of W is higher.The application passes through formula: the first prediction deviation value=| the first historical forecast numerical value-history actual value |,
Pass through formula: the second prediction deviation value=| the second historical forecast numerical value-history actual value | by way of, utilize preset prediction
Deviation and weight parameter corresponding relationship obtain corresponding with the first prediction numerical value A, the second prediction numerical value B respectively
Weight parameter pa、pb.Wherein, the history actual value is and the first historical forecast numerical value and the second historical forecast number
It is worth corresponding really and accurately numerical value, the forecasting accuracy to the determination prediction model and the knowledge mapping.
In one embodiment, described to calculate the final prediction numerical value W and the preset prediction numerical value for compareing object
Difference, and judge whether the difference is in preset difference range, wherein the prediction numerical value of the control object is based on
The data prediction of the control object and obtain, the step S5 that is corresponded to each other with the specified data of data for compareing object it
Afterwards, comprising:
If S51, the difference are in preset difference range, judge whether the final prediction numerical value W is in pre-
If numberical range within；
If S52, the final prediction numerical value W are not within preset numberical range, acquisition is not at described default
Numberical range within the first prediction numerical value and/or the second prediction numerical value, and be denoted as particular values；
S53, early warning report is generated, wherein the precaution alarm is accused with the final prediction numerical value W, the first prediction number
Value and the second prediction numerical value, and special marking is carried out to the particular values in early warning report.
It generates early warning report as described above, realizing and special mark is carried out to the particular values in early warning report
Note.If the difference is in preset difference range, shows and compare object without too big difference.For further early warning,
The application is also by the way of judging whether the final prediction numerical value W is within preset numberical range.If described final
Prediction numerical value W is not within preset numberical range, it may be possible to one of the prediction model and the knowledge mapping
Big prediction deviation is appeared above.Therefore, it obtains and is not at the first prediction numerical value within the preset numberical range
And/or the second prediction numerical value, and particular values are denoted as, and then generate early warning report, wherein the precaution alarm, which is accused, has institute
Final prediction numerical value, the first prediction numerical value and the second prediction numerical value are stated, and to the spy in early warning report
Different numerical value carries out special marking.To which the particular values of special marking can be taken seriously, can carry out the treatment measures such as rechecking.Into one
Step ground, further include the particular values of special marking are rechecked, if the particular values be confirmed to be prediction it is errorless, then table
The situation in future of bright prediction is severe, needs early prepare.
In one embodiment, if the difference is not in preset difference range, by the specified number
It is calculated according to being inputted in preset data differences level calculation model with the data for compareing object, to obtain the number
It is greater than preset rank threshold according to the data differences class value that difference level calculation model exports, and by the data differences class value
The specified data of value are denoted as the step S6 of suspicion data, comprising:
If S601, the difference are not in preset difference range, judge that the specified data compare pair with described
Whether the data of elephant belong to same magnitude；
If S602, the specified data belong to same magnitude with the data for compareing object, by the specified data
The data of the control object are subtracted to obtain the first data difference, according to preset first data difference and data differences grade
Other mapping relations, output data difference class value；
If S603, the specified data are not belonging to same magnitude with the data for compareing object, according to formula: second
Data difference=lg specifies the data of data-lg control object, calculates the second data difference, and according to preset second data
The mapping relations of difference and data differences rank, output data difference class value；
S604, the specified data that data differences class value is greater than preset level threshold are denoted as suspicion data.
The specified data that data differences class value is greater than preset level threshold are denoted as suspicion number as described above, realizing
According to.The specified data are divided into two classes by the application, a kind of to belong to same magnitude with the data for compareing object, it is possible thereby to
Subtraction process is directlyed adopt, further according to the mapping relations of preset first data difference and data differences rank, output data is poor
Other class value；It is another kind of to be not belonging to same magnitude with the data for compareing object, then according to formula: the second data difference=lg
The data of specified data-lg control object, calculate the second data difference, and according to preset second data difference and data difference
The mapping relations of other rank, output data difference class value.To improve the difference between specified data and the data for compareing object
The accuracy not judged.Wherein lg is the log function with 10 bottom of for.
In one embodiment, the Forewarning Terminal is a block chain link in the block chain network constructed in advance
Point, the generation early warning report, wherein after having the step S7 of the suspicion data in precaution alarm announcement, comprising:
S71, the precaution alarm is accused to the audit block chain node given in the block chain network, and requires described examine
Core block chain node audits early warning report；
S72, the auditing result that the audit block chain node returns is received, and it is pre- to judge whether the auditing result meets
If block chained record condition；
If S73, the auditing result meet preset block chained record condition, early warning report is recorded into described
In block chain network.
As described above, realizing in the public account book that early warning report is recorded into the block chain network.Wherein,
The block chain network constructed in advance can be publicly-owned chain, alliance's chain or privately owned chain.The block chain network it is default total
Knowledge mechanism can be any common recognition mechanism, and for example, proof of work mechanism, equity proves mechanism, Byzantine failure tolerance mechanism, share
Authorisation verification mechanism etc., the preferred share authorisation verification mechanism of present embodiment, to be thrown according to the share authorisation verification mechanism
The multiple audit block chain nodes selected, for auditing the early warning report.If whether the auditing result meets preset area
Block chained record condition can be considered that most of block chain nodes all approve the early warning report, therefore can be recorded into block
In public account book in chain network.The preset block chained record condition is for example are as follows: the auditing result of return is that audit passes through
Audit block chain node total quantity be greater than preset audit amount threshold.Early warning report is recorded into the area accordingly
In public account book in block chain network, to guarantee pre- by the feature of block chain network being difficult to tamper with distributed storage
The safety and authority of alarm announcement, the technical effect for being easy to call.
The early warning report-generating method based on machine learning of the application, by crawling primary data, and carries out at noise reduction
Reason obtains specified data；The specified data are inputted in preset prediction model and are calculated, to obtain the first prediction number
Value；Using the relationship that influences each other of each knowledge node in preset knowledge mapping, the second of the knowledge mapping output is obtained
Predict numerical value；Use formula: W=paA+pbB calculates final prediction numerical value W；Calculate the final prediction numerical value W with it is preset right
According to the difference of the prediction numerical value of object；If the difference is not in preset difference range, the data differences grade is obtained
The data differences class value of other computation model output, and the data differences class value is greater than the specified of preset level threshold
Data are denoted as suspicion data；Early warning report is generated, wherein having the suspicion data in precaution alarm announcement.It is pre- to improve
It surveys accuracy and realizes dynamic early-warning.
Referring to Fig. 2, the embodiment of the present application provides a kind of early warning report preparing apparatus based on machine learning, is applied to early warning
Terminal, comprising:
Specified data capture unit 10 crawls initial number from preset information source for using preset crawler technology
According to, and noise reduction process is carried out to the primary data using preset noise reduction algorithm, so that specified data are obtained, wherein the letter
Breath source includes at least default website；
First prediction numerical value acquiring unit 20, for the specified data to be inputted the preset instruction based on machine learning
Practice and calculated in the prediction model completed, to obtain the first prediction numerical value of the prediction model output；Wherein, described pre-
Survey model based on historical data identical with the type of the specified data, and with the associated prediction numerical value of the historical data
Training forms；
Second prediction numerical value acquiring unit 30, for according to the specified data, using each in preset knowledge mapping
The relationship that influences each other of knowledge node obtains the second prediction numerical value of the knowledge mapping output, wherein the knowledge mapping is extremely
It less include knowledge node corresponding with the specified data；
Final prediction numerical value acquiring unit 40, for using formula: W=paA+pbB calculates final prediction numerical value W, wherein A
For the first prediction numerical value, B is the second prediction numerical value, pa、pbThe respectively described first prediction numerical value A, described second are in advance
Survey the weight parameter of numerical value B；
Difference computational unit 50, for calculating the final prediction numerical value W and the preset prediction numerical value for compareing object
Difference, and judge whether the difference is in preset difference range, wherein the prediction numerical value of the control object is based on institute
It states the data prediction of control object and obtains, the data of the control object are corresponded to each other with the specified data；
Suspicion data capture unit 60 will be described specified if being not in preset difference range for the difference
Data are inputted in preset data differences level calculation model with the data for compareing object and are calculated, thus described in obtaining
The data differences class value of data differences level calculation model output, and the data differences class value is greater than preset rank
The specified data of threshold value are denoted as suspicion data；
Generation unit 70 is reported in early warning, for generating early warning report, wherein having the suspicion number in precaution alarm announcement
According to.
Wherein said units are respectively used to the operation executed and the early warning based on machine learning of aforementioned embodiments is reported
The step of generation method, corresponds, and details are not described herein.
In one embodiment, the specified data capture unit 10, comprising:
Primary data subelement is crawled, for the Scrapy frame using Python, is crawled just in default website
Beginning data；
Population variance computation subunit for the numerical value of same kind of primary data to be formed specified numerical value group, and is adopted
With preset formula:Calculate the population variance of m-th of primary data in the specified numerical value groupWherein N is the sum of the specified numerical value in the specified numerical value group, and Am is the numerical value of m-th of primary data, and B is the finger
The average value of fixed number value group；
Variance threshold values judgment sub-unit, for judging the population varianceWhether preset variance threshold values are respectively less than；
Removal processing subelement, if being used for the population varianceBe not respectively less than preset variance threshold values, then it will be described total
Body variancePrimary data not less than preset variance threshold values as noise and is removed processing.
Wherein above-mentioned subelement is respectively used to the precaution alarm based on machine learning of the operation executed and aforementioned embodiments
The step of accusing generation method corresponds, and details are not described herein.
In one embodiment, described device, comprising:
Sample data acquiring unit is divided into training set for obtaining the sample data of specified amount, and by the sample data
And test set；Wherein, the sample data includes historical data identical with the type of the specified data, and is gone through with described
The prediction numerical value of history data correlation；
Training unit is trained for the sample data of training set to be input in preset neural network model；Its
In, stochastic gradient descent method is used in trained process, obtains initial training model；
Authentication unit, for verifying the initial training model using the sample data of the test set；
Marking unit, for if the verification passes, then the initial training model being denoted as the prediction model.
Wherein said units are respectively used to the operation executed and the early warning based on machine learning of aforementioned embodiments is reported
The step of generation method, corresponds, and details are not described herein.
In one embodiment, described device, comprising:
Numerical value acquiring unit, for obtaining the first historical forecast numerical value, the knowledge mapping of the prediction model output
The the second historical forecast numerical value and history actual value of output；
First prediction deviation value computing unit, for passing through formula: the first prediction deviation value=| the first historical forecast number
Value-history actual value |, calculate the first prediction deviation value；
Second prediction deviation value computing unit, for passing through formula: the second prediction deviation value=| the second historical forecast number
Value-history actual value |, calculate the second prediction deviation value；
Weight parameter acquiring unit, for utilizing according to the first prediction deviation value and the second prediction deviation value
Preset prediction deviation value and weight parameter corresponding relationship obtain and the first prediction numerical value A, the second prediction number respectively
The corresponding weight parameter p of value Ba、pb。
Wherein said units are respectively used to the operation executed and the early warning based on machine learning of aforementioned embodiments is reported
The step of generation method, corresponds, and details are not described herein.
In one embodiment, described device, comprising:
Preset numberical range judging unit, if being in preset difference range for the difference, described in judgement
Whether final prediction numerical value W is within preset numberical range；
Particular values marking unit obtains if being not within preset numberical range for the final prediction numerical value W
It takes and is not at the first prediction numerical value and/or the second prediction numerical value within the preset numberical range, and be denoted as spy
Different numerical value；
Generation unit is reported in early warning, for generating early warning report, wherein the precaution alarm, which is accused, has the final prediction number
Value W, the first prediction numerical value and the second prediction numerical value, and spy is carried out to the particular values in early warning report
Different label.
Wherein said units are respectively used to the operation executed and the early warning based on machine learning of aforementioned embodiments is reported
The step of generation method, corresponds, and details are not described herein.
In one embodiment, the suspicion data capture unit 60, comprising:
Same magnitude judgment sub-unit judges the finger if being not in preset difference range for the difference
Whether fixed number belongs to same magnitude according to the data for compareing object；
First data differences class value exports subelement, if for the specified data and the data category for compareing object
In same magnitude, then the specified data are subtracted into the data of the control object so that the first data difference is obtained, according to pre-
If the first data difference and data differences rank mapping relations, output data difference class value；
Second data differences class value exports subelement, if not for the specified data and the data for compareing object
Belong to same magnitude, then according to formula: the second data difference=lg specifies the data of data-lg control object, calculates second
Data difference, and according to the mapping relations of preset second data difference and data differences rank, output data difference class value；
Suspicion data markers subelement, the specified data for data differences class value to be greater than to preset level threshold are remembered
For suspicion data.
Wherein above-mentioned subelement is respectively used to the precaution alarm based on machine learning of the operation executed and aforementioned embodiments
The step of accusing generation method corresponds, and details are not described herein.
In one embodiment, described device, comprising:
Precaution alarm accuses transmission unit, for the precaution alarm to be accused to the audit block chain given in the block chain network
Node, and the audit block chain node is required to audit early warning report；
Auditing result receiving unit, the auditing result returned for receiving the audit block chain node, and described in judgement
Whether auditing result meets preset block chained record condition；
Auditing result recording unit will be described if meeting preset block chained record condition for the auditing result
Early warning report is recorded into the block chain network.
Wherein said units are respectively used to the operation executed and the early warning based on machine learning of aforementioned embodiments is reported
The step of generation method, corresponds, and details are not described herein.
The early warning report preparing apparatus based on machine learning of the application, by crawling primary data, and carries out at noise reduction
Reason obtains specified data；The specified data are inputted in preset prediction model and are calculated, to obtain the first prediction number
Value；Using the relationship that influences each other of each knowledge node in preset knowledge mapping, the second of the knowledge mapping output is obtained
Predict numerical value；Use formula: W=paA+pbB calculates final prediction numerical value W；Calculate the final prediction numerical value W with it is preset right
According to the difference of the prediction numerical value of object；If the difference is not in preset difference range, the data differences grade is obtained
The data differences class value of other computation model output, and the data differences class value is greater than the specified of preset level threshold
Data are denoted as suspicion data；Early warning report is generated, wherein having the suspicion data in precaution alarm announcement.It is pre- to improve
It surveys accuracy and realizes dynamic early-warning.
Referring to Fig. 3, a kind of computer equipment is also provided in the embodiment of the present invention, which can be server,
Its internal structure can be as shown in the figure.The computer equipment includes that the processor, memory, network connected by system bus connects
Mouth and database.Wherein, the processor of the Computer Design is for providing calculating and control ability.The storage of the computer equipment
Device includes non-volatile memory medium, built-in storage.The non-volatile memory medium be stored with operating system, computer program and
Database.The internal memory provides environment for the operation of operating system and computer program in non-volatile memory medium.The meter
The database of machine equipment is calculated for storing data used in the early warning report-generating method based on machine learning.The computer equipment
Network interface is used to communicate with external terminal by network connection.To realize one kind when the computer program is executed by processor
Early warning report-generating method based on machine learning.
Above-mentioned processor executes the above-mentioned early warning report-generating method based on machine learning, is applied to Forewarning Terminal, wherein
The method includes the step of respectively with execute aforementioned embodiments the early warning report-generating method based on machine learning step
Rapid to correspond, details are not described herein.
The computer equipment of the application by crawling primary data, and carries out noise reduction process, obtains specified data；By institute
It states and is calculated in the preset prediction model of specified data input, to obtain the first prediction numerical value；Utilize preset knowledge graph
The relationship that influences each other of each knowledge node in spectrum obtains the second prediction numerical value of the knowledge mapping output；Use formula: W
=paA+pbB calculates final prediction numerical value W；Calculate the difference of the final prediction numerical value W with the preset prediction numerical value for compareing object
Value；If the difference is not in preset difference range, the data of the data differences level calculation model output are obtained
Difference class value, and the specified data that the data differences class value is greater than preset level threshold are denoted as suspicion data；It is raw
It is reported at early warning, wherein having the suspicion data in precaution alarm announcement.To improve forecasting accuracy and realize dynamic
State early warning.
One embodiment of the application also provides a kind of computer readable storage medium, is stored thereon with computer program, calculates
The early warning report-generating method based on machine learning is realized when machine program is executed by processor, and is applied to Forewarning Terminal, wherein institute
The method of stating include the steps that respectively with execute aforementioned embodiments the early warning report-generating method based on machine learning the step of
It corresponds, details are not described herein.
The computer readable storage medium of the application by crawling primary data, and carries out noise reduction process, obtains specified number
According to；The specified data are inputted in preset prediction model and are calculated, to obtain the first prediction numerical value；Using preset
The relationship that influences each other of each knowledge node in knowledge mapping obtains the second prediction numerical value of the knowledge mapping output；It uses
Formula: W=paA+pbB calculates final prediction numerical value W；Calculate the final prediction numerical value W and the preset prediction number for compareing object
The difference of value；If the difference is not in preset difference range, the data differences level calculation model output is obtained
Data differences class value, and by the data differences class value be greater than preset level threshold specified data be denoted as suspicion number
According to；Early warning report is generated, wherein having the suspicion data in precaution alarm announcement.To improve forecasting accuracy and realize
Dynamic early-warning.
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.
Claims (10)
1. a kind of early warning report-generating method based on machine learning is applied to Forewarning Terminal characterized by comprising
Using preset crawler technology, primary data is crawled from preset information source, and using preset noise reduction algorithm to institute
It states primary data and carries out noise reduction process, so that specified data are obtained, wherein the information source includes at least default website；
The specified data are inputted preset having trained in the prediction model completed based on machine learning to calculate, thus
Obtain the first prediction numerical value of the prediction model output；Wherein, the prediction model is based on the type with the specified data
Identical historical data, and formed with the associated prediction numerical value training of the historical data；
According to the specified data, using the relationship that influences each other of each knowledge node in preset knowledge mapping, described in acquisition
Second prediction numerical value of knowledge mapping output, wherein the knowledge mapping includes at least knowledge section corresponding with the specified data
Point；
Use formula: W=paA+pbB calculates final prediction numerical value W, and wherein A is the first prediction numerical value, and B is described second pre-
Survey numerical value, pa、pbThe weight parameter of the respectively described first prediction numerical value A, the second prediction numerical value B；
The difference of the final prediction numerical value W with the preset prediction numerical value for compareing object are calculated, and whether judges the difference
In preset difference range, wherein it is described control object prediction numerical value based on it is described control object data prediction and
, the data of the control object are corresponded to each other with the specified data；
If the difference is not in preset difference range, the specified data are inputted with the data for compareing object
It is calculated in preset data differences level calculation model, to obtain the number of the data differences level calculation model output
Suspicion data are denoted as according to difference class value, and by the specified data that the data differences class value is greater than preset level threshold；
Early warning report is generated, wherein having the suspicion data in precaution alarm announcement.
2. the early warning report-generating method according to claim 1 based on machine learning, which is characterized in that described using pre-
If crawler technology, crawl primary data from preset information source, and using preset noise reduction algorithm to the primary data
The step of carrying out noise reduction process, comprising:
Using the Scrapy frame of Python, primary data is crawled in default website；
The numerical value of same kind of primary data is formed into specified numerical value group, and uses preset formula:Calculate the population variance of m-th of primary data in the specified numerical value groupWherein N is described
The sum of specified numerical value in specified numerical value group, Am are the numerical value of m-th of primary data, and B is being averaged for the specified numerical value group
Value；
Judge the population varianceWhether preset variance threshold values are respectively less than；
If the population variancePreset variance threshold values are not respectively less than, then by the population varianceNot less than preset side
The primary data of poor threshold value is as noise and is removed processing.
3. the early warning report-generating method according to claim 1 based on machine learning, which is characterized in that it is described will be described
Specified data input preset having trained based on machine learning and are calculated in the prediction model completed, to obtain described pre-
Survey the first prediction numerical value of model output；Wherein, the prediction model is based on history identical with the type of the specified data
Before data, and the step of being formed with the associated prediction numerical value training of the historical data, comprising:
The sample data of specified amount is obtained, and the sample data is divided into training set and test set；Wherein, the sample data
Including historical data identical with the type of the specified data, and with the associated prediction numerical value of the historical data；
The sample data of training set is input in preset neural network model and is trained；Wherein, it is adopted in trained process
With stochastic gradient descent method, initial training model is obtained；
The initial training model is verified using the sample data of the test set；
If the verification passes, then the initial training model is denoted as the prediction model.
4. the early warning report-generating method according to claim 1 based on machine learning, which is characterized in that described using public
Formula: W=paA+pbB calculates final prediction numerical value W, and wherein A is the first prediction numerical value, and B is the second prediction numerical value, pa、
pbBefore the step of respectively described first prediction numerical value A, described second predict the weight parameter of numerical value B, comprising:
Obtain the first historical forecast numerical value of the prediction model output, the second historical forecast numerical value of knowledge mapping output
With history actual value；
Pass through formula: the first prediction deviation value=| the first historical forecast numerical value-history actual value |, it is pre- to calculate described first
Survey deviation；
Pass through formula: the second prediction deviation value=| the second historical forecast numerical value-history actual value |, it is pre- to calculate described second
Survey deviation；
According to the first prediction deviation value and the second prediction deviation value, preset prediction deviation value and weight parameter are utilized
Corresponding relationship obtains weight parameter p corresponding with the first prediction numerical value A, the second prediction numerical value B respectivelya、pb。
5. the early warning report-generating method according to claim 1 based on machine learning, which is characterized in that the calculating institute
The difference of final prediction numerical value W with the preset prediction numerical value for compareing object are stated, and judges whether the difference is in preset
In difference range, wherein the prediction numerical value of the control object is predicted based on the data of the control object and is obtained, the control
After the step of data of object are corresponded to each other with the specified data, comprising:
If the difference is in preset difference range, judge whether the final prediction numerical value W is in preset numerical value
Within the scope of；
If the final prediction numerical value W is not within preset numberical range, acquisition is not at the preset numerical value model
The first prediction numerical value and/or the second prediction numerical value within enclosing, and it is denoted as particular values；
Early warning report is generated, wherein the precaution alarm, which is accused, predicts numerical value and described with the final prediction numerical value W, described first
Second prediction numerical value, and special marking is carried out to the particular values in early warning report.
6. the early warning report-generating method according to claim 1 based on machine learning, which is characterized in that if described
Difference is not in preset difference range, then the specified data is inputted preset data with the data for compareing object
It is calculated in difference level calculation model, to obtain the data differences rank of the data differences level calculation model output
Value, and the step of specified data that the data differences class value is greater than preset level threshold are denoted as suspicion data, comprising:
If the difference is not in preset difference range, judge that the specified data are with the data for compareing object
It is no to belong to same magnitude；
If the specified data belong to same magnitude with the data for compareing object, it is described right that the specified data are subtracted
Data according to object are to obtain the first data difference, according to the mapping of preset first data difference and data differences rank pass
System, output data difference class value；
If the specified data are not belonging to same magnitude with the data for compareing object, according to formula: the second data difference
=lg specifies the data of data-lg control object, calculates the second data difference, and according to preset second data difference and number
According to the mapping relations of difference rank, output data difference class value；
The specified data that data differences class value is greater than preset level threshold are denoted as suspicion data.
7. the early warning report-generating method according to claim 1 based on machine learning, which is characterized in that the early warning is whole
End is a block chain node in the block chain network constructed in advance, the generation early warning report, wherein the precaution alarm is accused
In have the suspicion data the step of after, include:
The precaution alarm is accused to the audit block chain node given in the block chain network, and requires the audit block chain
Node audits early warning report；
The auditing result that the audit block chain node returns is received, and judges whether the auditing result meets preset block
Chained record condition；
If the auditing result meets preset block chained record condition, early warning report is recorded into the block link network
In network.
8. a kind of early warning report preparing apparatus based on machine learning is applied to Forewarning Terminal characterized by comprising
Specified data capture unit crawls primary data, and make for using preset crawler technology from preset information source
Noise reduction process is carried out to the primary data with preset noise reduction algorithm, so that specified data are obtained, wherein the information source is extremely
Few includes default website；
First prediction numerical value acquiring unit, preset has trained completion based on machine learning for inputting the specified data
Prediction model in calculated, to obtain the first prediction numerical value of prediction model output；Wherein, the prediction model
Based on historical data identical with the type of the specified data, and with the training of the historical data associated prediction numerical value and
At；
Second prediction numerical value acquiring unit, for utilizing each knowledge section in preset knowledge mapping according to the specified data
The relationship that influences each other of point obtains the second prediction numerical value of the knowledge mapping output, wherein the knowledge mapping includes at least
Knowledge node corresponding with the specified data；
Final prediction numerical value acquiring unit, for using formula: W=paA+pbB calculates final prediction numerical value W, and wherein A is described
First prediction numerical value, B are the second prediction numerical value, pa、pbThe respectively described first prediction numerical value A, the second prediction numerical value
The weight parameter of B；
Difference computational unit, for calculating the difference of the final prediction numerical value W with the preset prediction numerical value for compareing object, and
Judge whether the difference is in preset difference range, wherein the prediction numerical value of the control object is based on the control pair
The data of elephant predict and obtain that the data of the control object are corresponded to each other with the specified data；
Suspicion data capture unit, if being not in preset difference range for the difference, by the specified data with
The data of the control object, which input in preset data differences level calculation model, to be calculated, to obtain the data difference
The data differences class value of other level calculation model output, and the data differences class value is greater than preset level threshold
Specified data are denoted as suspicion data；
Generation unit is reported in early warning, for generating early warning report, wherein having the suspicion data in precaution alarm announcement.
9. a kind of computer equipment, including memory and processor, the memory are stored with computer program, feature exists
In the step of processor realizes any one of claims 1 to 7 the method when executing the computer program.
10. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program
The step of method described in any one of claims 1 to 7 is realized when being executed by processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910662226.0A CN110516075B (en) | 2019-07-22 | 2019-07-22 | Early warning report generation method and device based on machine learning and computer equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910662226.0A CN110516075B (en) | 2019-07-22 | 2019-07-22 | Early warning report generation method and device based on machine learning and computer equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110516075A true CN110516075A (en) | 2019-11-29 |
| CN110516075B CN110516075B (en) | 2022-04-15 |
Family
ID=68623273
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910662226.0A Active CN110516075B (en) | 2019-07-22 | 2019-07-22 | Early warning report generation method and device based on machine learning and computer equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110516075B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111859238A (en) * | 2020-07-27 | 2020-10-30 | 平安科技（深圳）有限公司 | Method and device for predicting data change frequency based on model and computer equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180150750A1 (en) * | 2016-11-30 | 2018-05-31 | Accenture Global Solutions Limited | Automatic prediction of an event using data |
| CN109064313A (en) * | 2018-07-20 | 2018-12-21 | 重庆富民银行股份有限公司 | Warning monitoring system after the loan of knowledge based graphical spectrum technology |
| CN109639670A (en) * | 2018-12-10 | 2019-04-16 | 北京威努特技术有限公司 | A kind of industry control network security postures quantitative estimation method of knowledge based map |
| CN110008349A (en) * | 2019-02-01 | 2019-07-12 | 阿里巴巴集团控股有限公司 | The method and device for the event risk assessment that computer executes |
-
2019
- 2019-07-22 CN CN201910662226.0A patent/CN110516075B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180150750A1 (en) * | 2016-11-30 | 2018-05-31 | Accenture Global Solutions Limited | Automatic prediction of an event using data |
| CN109064313A (en) * | 2018-07-20 | 2018-12-21 | 重庆富民银行股份有限公司 | Warning monitoring system after the loan of knowledge based graphical spectrum technology |
| CN109639670A (en) * | 2018-12-10 | 2019-04-16 | 北京威努特技术有限公司 | A kind of industry control network security postures quantitative estimation method of knowledge based map |
| CN110008349A (en) * | 2019-02-01 | 2019-07-12 | 阿里巴巴集团控股有限公司 | The method and device for the event risk assessment that computer executes |
Non-Patent Citations (1)
| Title |
|---|
| 张仲伟等: "基于神经网络的知识推理研究综述", 《计算机工程与应用》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111859238A (en) * | 2020-07-27 | 2020-10-30 | 平安科技（深圳）有限公司 | Method and device for predicting data change frequency based on model and computer equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110516075B (en) | 2022-04-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103412918B (en) | A kind of service trust degree appraisal procedure based on service quality and reputation | |
| Melo et al. | Using blockchains to implement distributed measuring systems | |
| CN106096657B (en) | Based on machine learning come the method and system of prediction data audit target | |
| EP4236197A2 (en) | Micro-loan system | |
| Pierro et al. | Are the gas prices oracle reliable? a case study using the ethgasstation | |
| CN112632629B (en) | Voting management method, device, medium and electronic equipment based on block chain | |
| WO2020073727A1 (en) | Risk forecast method, device, computer apparatus, and storage medium | |
| CN112633395A (en) | Abnormal data detection method and device, computer equipment and storage medium | |
| CN109658050A (en) | A kind of management method and equipment of wage report | |
| CN109377104A (en) | Data processing method, device, medium and electronic equipment | |
| CN108830443A (en) | A kind of contract review method and device | |
| CN117236855A (en) | Biological feed warehouse management system and method | |
| CN115237802A (en) | Artificial intelligence based simulation test method and related equipment | |
| Pourbafrani et al. | Discovering system dynamics simulation models using process mining | |
| CN110516075A (en) | Early warning report-generating method, device and computer equipment based on machine learning | |
| CN113807802A (en) | Block chain-based labor worker salary settlement method and related equipment | |
| CN105184170B (en) | A kind of domain software credibility assessment method based on formalization degree | |
| CN116777140A (en) | Enterprise business management method, device, equipment and medium | |
| US20220129318A1 (en) | Quota Request Resolution On Computing Platform | |
| CN115659214A (en) | Energy industry data credible evaluation method based on PaaS platform | |
| CN115358894A (en) | Intellectual property life cycle trusteeship management method, device, equipment and medium | |
| CN114598707A (en) | Coalition chain supervisory systems | |
| CN111651652B (en) | Emotion tendency identification method, device, equipment and medium based on artificial intelligence | |
| CN113535806A (en) | Block chain-based power data processing method and device | |
| CN110362981A (en) | The method and system of abnormal behaviour are judged based on credible equipment fingerprint |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |