CN108323797A - Cigarette Weight Control System based on GPR models starts position predicting method and system - Google Patents
Cigarette Weight Control System based on GPR models starts position predicting method and system Download PDFInfo
- Publication number
- CN108323797A CN108323797A CN201810134472.4A CN201810134472A CN108323797A CN 108323797 A CN108323797 A CN 108323797A CN 201810134472 A CN201810134472 A CN 201810134472A CN 108323797 A CN108323797 A CN 108323797A
- Authority
- CN
- China
- Prior art keywords
- predicted value
- residual
- prediction
- gpr
- gpr models
- 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
-
- A—HUMAN NECESSITIES
- A24—TOBACCO; CIGARS; CIGARETTES; SIMULATED SMOKING DEVICES; SMOKERS' REQUISITES
- A24C—MACHINES FOR MAKING CIGARS OR CIGARETTES
- A24C5/00—Making cigarettes; Making tipping materials for, or attaching filters or mouthpieces to, cigars or cigarettes
- A24C5/32—Separating, ordering, counting or examining cigarettes; Regulating the feeding of tobacco according to rod or cigarette condition
- A24C5/34—Examining cigarettes or the rod, e.g. for regulating the feeding of tobacco; Removing defective cigarettes
- A24C5/3424—Examining cigarettes or the rod, e.g. for regulating the feeding of tobacco; Removing defective cigarettes by weighing
Landscapes
- Feedback Control In General (AREA)
Abstract
The invention discloses a kind of, and the Cigarette Weight Control System based on GPR models starts position predicting method and system; including obtaining the target position data after shutting down preceding cigarette weight control system of cigarette making machine stabilization device target position data and starting, to target position data before shutdown and after starting, target position data extracts characteristic variable collection and target variable collection respectively;Using the characteristic variable collection and the target variable collection as training sample, initializes and train GPR models;Obtain starting the predicted value of position by the GPR models built according to the target position data before this shutdown; by the predicted value for starting position its prediction residual is obtained with calculated with actual values; and residual prediction GPR models are further built, obtain the predicted value of residual error;Predicted value to starting position is modified, and is obtained weight control system and is started position output valve.This method has preferable robustness, it is possible to prevente effectively from a certain secondary production status mutation saltus step caused by whole prediction result is possible influences.
Description
Technical field
The present invention relates to the technical fields of cigarette machine technological parameter dynamic optimization, more particularly to a kind of being based on GPR models
Cigarette Weight Control System start position predicting method and system.
Background technology
Currently, cigarette weight control system of cigarette making machine is mainly by density of tobacco rod detector, signal processing module, Industry Control
Computer, stabilization device (thering is chopper movement and aspirator band pinch roller to move two ways) and the parts such as other transmissions, adjustment mechanism
Composition.Wherein, stabilization device is the execution part of Weight control, the control of cigarette weight be exactly by the lifting of control stabilization device come
It realizes:When cigarette is passed through from detector, the average weight and distribution of weight situation of cigarette can be obtained, by with setting cigarette
Weight is compared, and further obtains deviation of weight signal, then feeds back to PLC and output control pulse, and driving stabilization device is high
Degree adjustment motor drives the lifting of stabilization device.After system-down or before starting, stabilization device all can be preset in one
Start position.When cigarette machine starts, since weight control system is not involved in control, stabilization device before cigarette machine reaches certain speed
In position is started, it is not involved in the adjusting of cigarette weight, during this period of time cigarette is generally unsatisfactory for weight demands;And in cigarette
After machine reaches certain speed, weight control system intervention control, stabilization device adjusting cigarette weight is also required to the regular hour could
Cigarette weight deviation is set to restrain.
Currently, the setting for starting position relies primarily on artificial experience execution, initial stage debugging generally can be according to manufacturer's recommended
Setting value is adjusted, and the average value of stabilization device target location in for the previous period is then generally basede in production and does certain compensation
After set.But the setting holding that in actual production, usually a kind of can start position is constant for a long time, or for the previous period
Interior target location fluctuation is larger, and ideal start bit also can not be really obtained according to the strategy that average value compensates
It sets.These ways based on experience, there is no stringent experimental datas or theoretical foundation, to the cigarette weight of cigarette machine startup stage
It is very little to measure control effect, and objectivity is poor.At the same time, relevant process strategies also can only be empirically more conservative
The more cigarette of rejecting is fixed to ensure product quality in ground in cigarette machine startup stage, there is considerable degree of waste among these.Therefore,
Optimize cigarette machine and start position, the waste that startup stage fixation is rejected can be reduced, there is prodigious economic benefit.
Due to industrial process usually present the multistage, time-varying dynamic characteristic, using general data-driven model, to mistake
The prediction effect of journey parameter tends not to be guaranteed.GPR, which is Gaussian process recurrence, can establish local nonparametric probability
The regression model of model, gained can not only provide predicted value, can also obtain confidence level of the predicted value to model.It has sternly
The Statistical Learning Theory basis of lattice has good adaptability to handling the complicated problem such as high dimension, small sample, non-linear,
And generalization ability is strong.Compared with the models such as neural network, support vector machines, there is GPR easy to implement, hyper parameter adaptively to obtain
It takes, nonparametric inference is flexible and output has many advantages, such as probability meaning.
Further, since the diversity of the time variation and equipment feature of production process, can make the estimated performance of static models not
People's will to the greatest extent.In order to avoid model poor robustness, precision of prediction adaptability is low the problems such as, need continuous online updating prediction model.
Method based on on-line study is then capable of the time-varying characteristics of processing procedure well, improves the dynamic property of prediction model.It compares
In traditional world model established offline, the model that on-line study method is established has local dynamic station structure, and model is
Online, it can preferably track the current state of process.
Invention content
The present invention in the prior art the shortcomings that, provide a kind of Cigarette Weight Control System based on GPR models and open
Dynamic position predicting method and system.
In order to solve the above-mentioned technical problem, the present invention is addressed by following technical proposals:
The present invention discloses:
A kind of Cigarette Weight Control System startup position predicting method based on GPR models, includes the following steps:
Target position data after obtaining weight control system for cigarette making machine stabilization device target position data before shutting down and starting,
To target position data before shutdown and after starting, target position data extracts characteristic variable collection and target variable collection respectively;
Using the characteristic variable collection and the target variable collection as training sample, initializes and train GPR models;
Obtain starting the predicted value of position by the GPR models built according to the target position data before this shutdown,
Its prediction residual is obtained by the predicted value for starting position and calculated with actual values, and further builds residual prediction GPR moulds
Type obtains the predicted value of residual error;
The predicted value for starting position is modified using the predicted value of the residual error, weight control system is obtained and opens
Dynamic position output valve.
As a kind of embodiment, using the characteristic variable collection and the target variable collection as training sample, initially
Change and train GPR models the specific steps are:
Initialization procedure:Using a square index covariance function, covariance function and hyper parameter initial value are set;
Training process:The negative log-likelihood function for establishing the training sample conditional probability, according to negative log-likelihood function
Partial derivative is asked to hyper parameter;Minimum processing is carried out to the partial derivative using conjugate gradient method, obtains the hyper parameter of GPR models
Optimal solution, finally establish GPR models.
As a kind of embodiment, the target position data according to before this shutdown passes through the GPR moulds that build
Type obtains starting the predicted value of position, its prediction residual is obtained with calculated with actual values by the predicted value for starting position, and
Further build residual prediction GPR models, obtain the predicted value of residual error the specific steps are:
Target position data before this is shut down is updated in the GPR models built, obtains starting position
Predicted value;
The residual error predicted every time is obtained with actual value using the predicted value for starting position, set of residuals is established, utilizes institute
It states set of residuals and establishes fisrt feature variables set and first object variables set;
According to the fisrt feature variables set and the first object variables set, the residual error for carrying out residual prediction is built
Predict GPR models;
The predicted value for starting position is substituted into the residual prediction GPR models, obtains the predicted value of this prediction residual.
It is described to establish fisrt feature variables set and first object variable using the set of residuals as a kind of embodiment
The process of collection is:
Continuous j residual error is taken out first element successively since the set of residuals as characteristic variable, correspondingly, is taken
Go out+1 residual error of jth as target variable, and so on, total l-j characteristic variable collection and l-j target variable collection can be built.
As a kind of embodiment, the predicted value using the residual error carries out the predicted value for starting position
Correct, obtain weight control system start position output valve, in particular to,
Start the predicted value of position using residual prediction amendment, the modified predicted value for starting position is:Correction value=open
The predicted value of predicted value+residual error of dynamic position, correction value are to start position output valve.
As a kind of embodiment, the data training GPR models all shut down several times before and residual error are shut down each time
Predict GPR models, GPR models and residual prediction GPR models carry out on-line study respectively.
Invention further discloses:
A kind of Cigarette Weight Control System startup position forecasting system based on GPR models, including data acquisition process mould
Block, model construction module, residual prediction model construction module and amendment output module;
The data acquisition process module shuts down preceding weight control system for cigarette making machine stabilization device target location number for obtaining
According to the target position data after startup, to target position data before shutdown and start after target position data extract feature respectively
Variables set and target variable collection;
The model construction module is used for using the characteristic variable collection and the target variable collection as training sample, just
Beginningization and training GPR models;
The residual prediction model construction module, for according to the target position data before this shutdown by building
GPR models obtain starting the predicted value of position, and it is residual with calculated with actual values to obtain its prediction by the predicted value for starting position
Difference, and residual prediction GPR models are further built, obtain the predicted value of residual error;
The amendment output module is modified the predicted value for starting position for the predicted value using residual error, obtains
Weight control system starts position output valve.
As a kind of embodiment, the model construction module includes initialization unit and hyper parameter determination unit;
The initialization unit is used for initialization procedure:Using a square index covariance function, setting covariance function and
Hyper parameter initial value;
The hyper parameter determination unit, the negative log-likelihood function for establishing the training sample conditional probability, according to
Negative log-likelihood function seeks partial derivative to hyper parameter;Minimum processing is carried out to the partial derivative using conjugate gradient method, is obtained
The optimal solution of the hyper parameter of GPR models establishes GPR models by the optimal solution of hyper parameter.
As a kind of embodiment, the residual prediction model construction module includes prediction result acquiring unit, establishes
Training sample unit, structure prediction model unit and computing unit;
The prediction result acquiring unit is updated to for the target position data before shutting down this described in building
In GPR models, the predicted value for starting position is obtained;
It is described to establish training sample unit, for obtaining prediction every time using the predicted value for starting position and actual value
Residual error, establish set of residuals, fisrt feature variables set and first object variables set established using the set of residuals;
The structure prediction model unit is used for according to the fisrt feature variables set and the first object variables set,
Build the residual prediction GPR models for carrying out residual prediction;
The computing unit, the predicted value for that will start position substitute into the residual prediction GPR models, it is pre- to obtain this
Survey the predicted value of residual error.
Invention further discloses:
A kind of computer readable storage medium, is stored with computer program, which realizes base when being executed by processor
In the step of Cigarette Weight Control System of GPR models starts position predicting method and system.
The present invention has significant technique effect as a result of above technical scheme:
The beneficial effects of the invention are as follows:Start location parameter artificial experience setting method for weight control system for cigarette making machine
Unreasonable the problem of causing cigarette to waste, all trains prediction model when shutting down every time according to the preceding data shut down several times, model into
Row on-line study, and this shutdown data is substituted into model, quickly provide the recommendation target location of booting next time, this target location
Cigarette machine start-up phase cigarette weight deviation convergence time can be effectively reduced, the waste of finished product cigarette is reduced.
And this method has from growth, after prediction model application first can optimization system state, the system after optimization is again
Improvement model can be fed back, to accumulate continuous on-line study and optimization with system data, system can be made to reach, and one kind is benign to be followed
The state of ring.And this method has preferable robustness, it is possible to prevente effectively from a certain secondary production status mutation is to whole prediction
As a result saltus step caused by possible influences.
A kind of Cigarette Weight Control System based on GPR models provided by the present invention starts position predicting method and is
System can solve the problems, such as the pain spot of one finished product cigarette of current cigarette industry waste, cigarette machine start bit is installed by current
Extensive style, experience rely on, the state of manual intervention, improves as data-driven, modelling, the state of predictability, having can not estimate
The economic benefit and social benefit of amount.
Description of the drawings
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technology description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention without having to pay creative labor, may be used also for those of ordinary skill in the art
With obtain other attached drawings according to these attached drawings.
Fig. 1 is the overall flow schematic diagram of the present invention;
Fig. 2 is the detailed process schematic diagram of Fig. 1;
Fig. 3 is that the weight control system for cigarette making machine in one embodiment of the invention using GPR model predictions starts position and reality
The startup position comparison diagram on border;
Fig. 4 is that the Cigarette Weight Control System startup position prediction in one embodiment of the invention using GPR model predictions is residual
Difference figure;
Fig. 5 is the overall system structure schematic diagram of the present invention.
Label in attached drawing:100, data acquisition process module;200, model construction module;300, residual prediction model structure
Model block;400, output module is corrected;210, initialization unit;220, hyper parameter determination unit;310, prediction result obtains single
Member;320, training sample unit is established;330, prediction model unit is built;340, computing unit;
Specific implementation mode
With reference to embodiment, the present invention is described in further detail, following embodiment be explanation of the invention and
The invention is not limited in following embodiments.
The invention discloses:
A kind of Cigarette Weight Control System startup position predicting method based on GPR models, as shown in Figure 1, including following
Step:
S100, the target location after shutting down preceding weight control system for cigarette making machine stabilization device target position data and starting is obtained
Data, to target position data before shutdown and after starting, target position data extracts characteristic variable collection and target variable collection respectively;
S200, using the characteristic variable collection and the target variable collection as training sample, initialize simultaneously training GPR moulds
Type;
S300, the pre- of startup position is obtained by the GPR models built according to the target position data before this shutdown
Measured value obtains its prediction residual with calculated with actual values by the predicted value for starting position, and further builds residual prediction
GPR models obtain the predicted value of residual error;
S400, the predicted value for starting position is modified using the predicted value of the residual error, obtains Weight control
System starts position output valve.
In step s 200, described using the characteristic variable collection and the target variable collection as training sample, initialization
And training GPR models the specific steps are:
S210, initialization procedure:Using a square index covariance function, covariance function and hyper parameter initial value are set;
S220, the negative log-likelihood function for establishing the training sample conditional probability, according to negative log-likelihood function to super
Parameter seeks partial derivative;Minimum processing is carried out to the partial derivative using conjugate gradient method, obtains the hyper parameter of GPR models most
Excellent solution establishes GPR models by the optimal solution of hyper parameter.
More specifically, the detailed process of the present embodiment is as follows:
When initialization procedure, using a square index covariance function, and covariance function and hyper parameter initial value are set;Become
Measure Tp,TqA square index covariance function be:
In formula:Tp,TqFor the arbitrary element in gained characteristic variable collection T in step;For data variance;M=diag (b2)
For diagonal matrix, exponent number and TpDimension it is consistent, b is variance measure;
The hyper parameter of model is:
In formulaFor observation noise variance.
Establish the negative log-likelihood function of training sample conditional probability:
In formula:Kn=(kpq) be n × n rank symmetric positive definites covariance matrix, Kpq=k (TP,Tq) it is TP,
TqSquare index covariance function;F is target variable collection.
Negative log-likelihood function L (θ) is enabled to seek partial derivative to hyper parameter θ;
In formula:α=C-1;The mark of tr () representing matrix;
Using conjugate gradient method to partial derivativeIt is minimized, to obtain hyper parameterMost
Excellent solution.
In step S300, the target position data according to before this shutdown is obtained by the GPR models built
The predicted value for starting position obtains its prediction residual by the predicted value for starting position and calculated with actual values, and further
Build residual prediction GPR models, obtain the predicted value of residual error the specific steps are:
S310, this target position data before shutting down is updated in the GPR models built, obtains start bit
The predicted value set;
S320, the residual error predicted every time is obtained with actual value using the predicted value for starting position, establishes set of residuals, profit
Fisrt feature variables set and first object variables set are established with the set of residuals, here, establishing the first spy using the set of residuals
Sign variables set and the process of first object variables set are:Take out continuous j first element successively since the set of residuals
Residual error is as characteristic variable, correspondingly, take out+1 residual error of jth be used as target variable, and so on, can build it is total l-j spy
Levy variables set and l-j target variable collection;
S330, according to the fisrt feature variables set and the first object variables set, build for carrying out residual prediction
Residual prediction GPR models;
S340, the predicted value for starting position is substituted into the residual prediction GPR models, obtains the prediction of this prediction residual
Value.
In the present embodiment, it is assumed that take the set of residuals of preceding 10 prediction results
Continuous 5 residual errors are taken out first element since set of residuals E as characteristic variable, correspondingly, taking out the 6th
Residual error as target variable, and so on, totally 5 characteristic variable collection E can be builtT={ ETk={ ek,ek+1...ek+4| k=1,
2...5 } with 5 target variable collection EF={ e6,e7...e10};
Utilize characteristic variable collection ETWith target variable collection EF, established for the residual of residual prediction according to the step of step S200
Difference prediction GPR models;
By set of residuals { e6,e7,e8,e9,e10Established residual prediction GPR models are substituted into as characteristic variable, obtain this
The predicted value e of secondary prediction residual*;
Target position data before this is shut down measures T* as systematic perspective and substitutes into GPR models, then start bit can be obtained
The predicted value f* set:
In formula:K(T*, T) and={ k (T*, T) | k=1,2 ..., n } it is T*1 × n rank covariance matrixes between T;k
(T*, Tk) it is T*And TkCovariance function;k(T*, Tk) expression formula be square index covariance function expression formula;Kn=(kpq)
For the covariance matrix of n × n rank symmetric positive definites;kpq=k (Tp, Tq) it is Tp, TqSquare index covariance function;InIt is tieed up for n single
Bit matrix.
In step S400, the predicted value using the residual error is modified the predicted value for starting position,
Obtain weight control system start position output valve, in particular to, utilize residual prediction amendment start position predicted value, correct
The predicted value of startup position be:The predicted value of predicted value+residual error of correction value=startup position, correction value are to start position
Output valve.Here, the predicted value for being formulated as starting position is modified tof*For predicted value, e*It is pre- for residual error
Measured value,The final weight control system as provided starts position output valve.
Specific embodiment:
It combines the above method to be verified for the extensive ZJ112 types cigarette making and tipping machine of current application, is illustrated in figure 2 this
Cigarette Weight Control System based on on-line study GPR model of the invention provided in this embodiment starts the stream of position prediction
Cheng Tu, including the specific implementation step that this method is applied in the present embodiment:
After cigarette machine is shut down, reads its first n=5 times and shut down in data, shut down preceding stabilization device target position data and start
Target position data afterwards;
Stabilization device target position data before shutdown is standardized, preceding 3 minutes data were shut down in interception, less than 3 minutes
Person utilizes its previous secondary Data-parallel language, as above utilizes structure target variable collection T={ T in stabilization device target location before shutting downk| k=1,
2,…,5};
First 3 seconds after starting target location mean values are calculated, target variable collection F={ f are obtainedk| k=1,2 ..., 5 };
GPR model constructions:
GPR model initializations:A square index covariance function can be used, and set covariance function and hyper parameter is initial
Value;Variable Tp,TqA square index covariance function be:
In formula:Tp,TqFor the arbitrary element in gained characteristic variable collection T in step;For data variance;M=diag (b2)
For diagonal matrix, exponent number and TpDimension it is consistent, b is variance measure;
The hyper parameter of GPR models is:
In formulaFor observation noise variance.
Establish the negative log-likelihood function of training sample conditional probability:
In formula:Kn=(kpq) be n × n rank symmetric positive definites covariance matrix, Kpq=k (TP,Tq) it is TP,
TqSquare index covariance function;F is target variable collection.
Negative log-likelihood function L (θ) is enabled to seek partial derivative to hyper parameter θ;
Wherein:α=C-1;The mark of tr () representing matrix;
Using conjugate gradient method to partial derivativeIt is minimized, to obtain hyper parameterMost
Excellent solution.
T is measured using this preceding 3 minutes target position data of shutdown as systematic perspective*GPR models are substituted into, then can be opened
The predicted value f of dynamic position*:
In formula:K(T*, T) and={ k (T*, T) | k=1,2 ..., n } it is T*1 × n rank covariance matrixes between T;k
(T*, Tk) it is T*And TkCovariance function;k(T*, Tk) expression formula be squared index covariance function expression formula;Kn=(kpq)
For the covariance matrix of n × n rank symmetric positive definites, kpq=k (Tp, Tq) it is Tp, TqSquare index covariance function, InIt is tieed up for n single
Bit matrix.
Residual prediction:
Take the set of residuals of preceding 10 prediction results
Continuous 5 residual errors are taken out first element since set of residuals E as characteristic variable, correspondingly, taking out the 6th
Residual error as target variable, and so on, totally 5 characteristic variable collection E can be builtT={ ETk={ ek,ek+1...ek+4| k=1,
2...5 } with 5 target variable collection EF={ e6,e7...e10};
Utilize 5 characteristic variable collection ETWith 5 target variable collection EF, established using the method for step S200 pre- for residual error
The residual prediction GPR models of survey;
Set of residuals is substituted into established residual prediction GPR models as characteristic variable, obtains the pre- of this prediction residual
Measured value e*。
Prediction result exports:Utilize gained predicted value f*With gained residual prediction value e*, the predicted value amendment of position will be started
For The final weight control system as provided starts position output valve.
By the data verification of 115 shutdown of certain ZJ112 types cigarette making and tipping machine, using the cigarette weight control based on GPR models
System processed starts position predicting method and system, to starting the prediction result of position as shown in figure 3, prediction result starts with practical
Smaller to ratio error between position, as shown in figure 4, root-mean-square error is 0.41, accuracy is ideal, the method application
Into production, the cigarette loss of cigarette machine startup stage can be greatly reduced.
The invention also discloses:
A kind of Cigarette Weight Control System startup position forecasting system based on on-line study GPR models, as shown in figure 5,
Including data acquisition process module 100, model construction module 200, residual prediction model construction module 300 and correct output module
400;
The data acquisition process module 100 shuts down preceding weight control system for cigarette making machine stabilization device target position for obtaining
Target position data after setting data and starting, to target position data before shutdown and after starting, target position data extracts respectively
Characteristic variable collection and target variable collection;
The model construction module 200, for using the characteristic variable collection and the target variable collection as training sample,
It initializes and trains GPR models;
The residual prediction model construction module 300, for passing through structure according to the target position data before this shutdown
Good GPR models obtain starting the predicted value of position, and it is pre- with calculated with actual values to obtain its by the predicted value for starting position
Residual error is surveyed, and further builds residual prediction GPR models, obtains the predicted value of residual error;
The amendment output module 400 is modified the predicted value for starting position for the predicted value using residual error, obtains
Start position output valve to weight control system.
Further, the model construction module 200 includes initialization unit 210 and hyper parameter determination unit 220;
The initialization unit 210 is used for initialization procedure:Using a square index covariance function, covariance letter is set
Number and hyper parameter initial value;
The hyper parameter determination unit 220, the negative log-likelihood function for establishing the training sample conditional probability, root
Partial derivative is asked to hyper parameter according to negative log-likelihood function;Minimum processing is carried out to the partial derivative using conjugate gradient method, is obtained
To the optimal solution of the hyper parameter of GPR models, GPR models are established by the optimal solution of hyper parameter.
In other embodiment, the residual prediction model construction module 300 includes prediction result acquiring unit 310, builds
Vertical training sample unit 320, structure prediction model unit 330 and computing unit 340;
The prediction result acquiring unit 310 is updated to for the target position data before shutting down this and to build
In the GPR models, the predicted value for starting position is obtained;
It is described to establish training sample unit 320, it is each for being obtained with actual value using the predicted value for starting position
The residual error of prediction, establishes set of residuals, and fisrt feature variables set and first object variables set are established using the set of residuals;
The structure prediction model unit 330, for according to the fisrt feature variables set and the first object variable
Collection builds the residual prediction GPR models for carrying out residual prediction;
The computing unit 340, the predicted value for that will start position substitute into the residual prediction GPR models, obtain this
The predicted value of secondary prediction residual.
The invention also discloses:
A kind of computer readable storage medium, is stored with computer program, which realizes base when being executed by processor
In the step of Cigarette Weight Control System of GPR models starts position predicting method and system.
For device embodiments, since it is basically similar to the method embodiment, so fairly simple, the correlation of description
Place illustrates referring to the part of embodiment of the method.
Each embodiment in this specification is described in a progressive manner, the highlights of each of the examples are with
The difference of other embodiment, the same or similar parts between the embodiments can be referred to each other.
It should be understood by those skilled in the art that, the embodiment of the present invention can be provided as method, apparatus or computer program
Product.Therefore, complete hardware embodiment, complete software embodiment or reality combining software and hardware aspects can be used in the present invention
Apply the form of example.Moreover, the present invention can be used in one or more wherein include computer usable program code computer
The computer program production implemented in usable storage medium (including but not limited to magnetic disk storage, CD-ROM, optical memory etc.)
The form of product.
The present invention be with reference to according to the method for the present invention, the flow chart of terminal device (system) and computer program product
And/or block diagram describes.It should be understood that each flow in flowchart and/or the block diagram can be realized by computer program instructions
And/or the combination of the flow and/or box in box and flowchart and/or the block diagram.These computer programs can be provided to refer to
Enable the processor of all-purpose computer, special purpose computer, Embedded Processor or other programmable data processing terminal equipments with
Generate a machine so that the instruction executed by computer or the processor of other programmable data processing terminal equipments generates
For realizing the function of being specified in one flow of flow chart or multiple flows and/or one box of block diagram or multiple boxes
Device.
These computer program instructions, which may also be stored in, can guide computer or other programmable data processing terminal equipments
In computer-readable memory operate in a specific manner so that instruction stored in the computer readable memory generates packet
The manufacture of command device is included, which realizes in one flow of flow chart or multiple flows and/or one side of block diagram
The function of being specified in frame or multiple boxes.
These computer program instructions can be also loaded into computer or other programmable data processing terminal equipments so that
Series of operation steps are executed on computer or other programmable terminal equipments to generate computer implemented processing, thus
The instruction executed on computer or other programmable terminal equipments is provided for realizing in one flow of flow chart or multiple flows
And/or in one box of block diagram or multiple boxes specify function the step of.
It should be noted that:
" one embodiment " or " embodiment " mentioned in specification means the special characteristic described in conjunction with the embodiments, structure
Or characteristic includes at least one embodiment of the present invention.Therefore, the phrase " reality that specification various places throughout occurs
Apply example " or " embodiment " the same embodiment might not be referred both to.
Although preferred embodiments of the present invention have been described, it is created once a person skilled in the art knows basic
Property concept, then additional changes and modifications can be made to these embodiments.So it includes excellent that the following claims are intended to be interpreted as
It selects embodiment and falls into all change and modification of the scope of the invention.
Furthermore, it is necessary to illustrate, the specific embodiment described in this specification, the shape of parts and components is named
Title etc. can be different.The equivalent or simple change that all structure, feature and principles according to described in inventional idea of the present invention are done, is wrapped
It includes in the protection domain of patent of the present invention.Those skilled in the art can be to described specific implementation
Example is done various modifications or additions or is substituted by a similar method, without departing from structure of the invention or surmounts this
Range as defined in the claims, is within the scope of protection of the invention.
Claims (10)
1. a kind of Cigarette Weight Control System based on GPR models starts position predicting method, it is characterised in that including following step
Suddenly:
Target position data after obtaining weight control system for cigarette making machine stabilization device target position data before shutting down and starting, to stopping
Target position data and target position data after startup extract characteristic variable collection and target variable collection respectively before machine;
Using the characteristic variable collection and the target variable collection as training sample, initializes and train GPR models;
The predicted value for obtaining starting position by the GPR models built according to the target position data before this shutdown, passes through
The predicted value for starting position obtains its prediction residual with calculated with actual values, and further builds residual prediction GPR models, obtains
To the predicted value of residual error;
The predicted value for starting position is modified using the predicted value of the residual error, obtains weight control system start bit
Set output valve.
2. the Cigarette Weight Control System according to claim 1 based on GPR models starts position predicting method, feature
It is, using the characteristic variable collection and the target variable collection as training sample, initializes and train the specific step of GPR models
Suddenly it is:
Initialization procedure:Using a square index covariance function, covariance function and hyper parameter initial value are set;
Training process:The negative log-likelihood function for establishing the training sample conditional probability, according to negative log-likelihood function to super
Parameter seeks partial derivative;Minimum processing is carried out to the partial derivative using conjugate gradient method, obtains the hyper parameter of GPR models most
Excellent solution finally establishes GPR models.
3. the Cigarette Weight Control System according to claim 2 based on GPR models starts position predicting method, feature
It is, the target position data according to before this shutdown obtains starting the predicted value of position by the GPR models built,
Its prediction residual is obtained by the predicted value for starting position and calculated with actual values, and further builds residual prediction GPR moulds
Type, obtain the predicted value of residual error the specific steps are:
Target position data before this is shut down is updated in the GPR models built, obtains the prediction for starting position
Value;
The residual error predicted every time is obtained using predicted value and the actual value for starting position, establishes set of residuals, using described residual
Difference set establishes fisrt feature variables set and first object variables set;
According to the fisrt feature variables set and the first object variables set, the residual prediction for carrying out residual prediction is built
GPR models;
The predicted value for starting position is substituted into the residual prediction GPR models, obtains the predicted value of this prediction residual.
4. the Cigarette Weight Control System according to claim 3 based on GPR models starts position predicting method, feature
It is, the process that fisrt feature variables set and first object variables set are established using the set of residuals is:
Continuous j residual error is taken out first element successively since the set of residuals as characteristic variable, correspondingly, takes out the
J+1 residual error as target variable, and so on, total l-j characteristic variable collection and l-j target variable collection can be built.
5. the Cigarette Weight Control System according to claim 3 based on GPR models starts position predicting method, feature
It is, the predicted value using the residual error is modified the predicted value for starting position, obtains weight control system
Start position output valve, in particular to,
Start the predicted value of position using residual prediction amendment, the modified predicted value for starting position is:Correction value=start bit
The predicted value for the predicted value+residual error set, correction value are to start position output valve.
6. the Cigarette Weight Control System according to claim 1 based on GPR models starts position predicting method, feature
It is, shuts down all train GPR models and residual prediction GPR models according to the preceding data shut down several times each time, GPR models and residual
Difference prediction GPR models carry out on-line study respectively.
7. a kind of Cigarette Weight Control System based on GPR models starts position forecasting system, which is characterized in that obtained including data
It takes processing module, model construction module, residual prediction model construction module and corrects output module;
The data acquisition process module, for obtain shut down before weight control system for cigarette making machine stabilization device target position data and
Target position data after startup, to target position data before shutdown and after starting, target position data extracts characteristic variable respectively
Collection and target variable collection;
The model construction module, for using the characteristic variable collection and the target variable collection as training sample, initialization
And training GPR models;
The residual prediction model construction module, for passing through the GPR that builds according to the target position data before this shutdown
Model obtains starting the predicted value of position, its prediction residual is obtained with calculated with actual values by the predicted value for starting position,
And residual prediction GPR models are further built, obtain the predicted value of residual error;
The amendment output module is modified the predicted value for starting position for the predicted value using residual error, obtains weight
Control system starts position output valve.
8. the Cigarette Weight Control System according to claim 7 based on GPR models starts position forecasting system, feature
It is, the model construction module includes initialization unit and hyper parameter determination unit;
The initialization unit is used for initialization procedure:Using a square index covariance function, setting covariance function and super ginseng
Number initial value;
The hyper parameter determination unit, the negative log-likelihood function for establishing the training sample conditional probability, according to negative pair
Number likelihood function seeks partial derivative to hyper parameter;Minimum processing is carried out to the partial derivative using conjugate gradient method, obtains GPR moulds
The optimal solution of the hyper parameter of type establishes GPR models by the optimal solution of hyper parameter.
9. the Cigarette Weight Control System according to claim 8 based on GPR models starts position forecasting system, feature
It is, the residual prediction model construction module includes prediction result acquiring unit, establishes training sample unit, structure prediction mould
Type unit and computing unit;
The prediction result acquiring unit is updated to the GPR built for the target position data before shutting down this
In model, the predicted value for starting position is obtained;
It is described to establish training sample unit, it is residual for being predicted every time with actual value acquisition using the predicted value for starting position
Difference establishes set of residuals, and fisrt feature variables set and first object variables set are established using the set of residuals;
The structure prediction model unit, for according to the fisrt feature variables set and the first object variables set, structure
Residual prediction GPR models for carrying out residual prediction;
The computing unit, the predicted value for that will start position substitute into the residual prediction GPR models, it is residual to obtain this prediction
The predicted value of difference.
10. a kind of computer readable storage medium, is stored with computer program, which is characterized in that the program is held by processor
The step of claim 1-6 any one the methods are realized when row.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810134472.4A CN108323797B (en) | 2018-02-09 | 2018-02-09 | GPR (general purpose) model-based cigarette weight control system starting position prediction method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810134472.4A CN108323797B (en) | 2018-02-09 | 2018-02-09 | GPR (general purpose) model-based cigarette weight control system starting position prediction method and system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108323797A true CN108323797A (en) | 2018-07-27 |
CN108323797B CN108323797B (en) | 2020-11-24 |
Family
ID=62927369
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810134472.4A Active CN108323797B (en) | 2018-02-09 | 2018-02-09 | GPR (general purpose) model-based cigarette weight control system starting position prediction method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108323797B (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109330018A (en) * | 2018-10-30 | 2019-02-15 | 浙江中烟工业有限责任公司 | A kind of setting method of cigarette weight control system of cigarette making machine aspirator tape starting position |
CN110037336A (en) * | 2019-04-19 | 2019-07-23 | 浙江中烟工业有限责任公司 | A kind of prediction technique of Cigarette circumference control system executing agency position |
CN111562574A (en) * | 2020-05-22 | 2020-08-21 | 中国科学院空天信息创新研究院 | MIMO ground penetrating radar three-dimensional imaging method based on backward projection |
CN111932037A (en) * | 2020-09-23 | 2020-11-13 | 浙江创泰科技有限公司 | Parking space state prediction method and system based on machine learning |
CN113313106A (en) * | 2021-04-14 | 2021-08-27 | 深圳市睿达科技有限公司 | Feeding deviation rectifying method and device, computer equipment and storage medium |
Citations (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US3146910A (en) * | 1960-11-07 | 1964-09-01 | Industrial Nucleonics Corp | Control system |
GB1313861A (en) * | 1969-06-02 | 1973-04-18 | Industrial Nucleonics Corp | System and method for optimizing processor or equipment profit |
EP1048568A1 (en) * | 1999-04-28 | 2000-11-02 | Focke & Co. (GmbH & Co.) | Method and device for checking cigarettes |
JP2001333758A (en) * | 2000-05-24 | 2001-12-04 | Japan Tobacco Inc | Combustion test apparatus |
CN1975705A (en) * | 2005-11-28 | 2007-06-06 | 颐中烟草(集团)有限公司 | Cigarette internal quality index extimating method based on regression function estimating SVM |
CN201640440U (en) * | 2010-03-19 | 2010-11-24 | 河南中烟工业有限责任公司 | Cigarette weight control system of cigarette making machine |
CN102406234A (en) * | 2011-07-13 | 2012-04-11 | 常德烟草机械有限责任公司 | Cigarette position signal generation and quality detection rejection method |
CN103385539A (en) * | 2013-08-02 | 2013-11-13 | 南京文采科技有限责任公司 | Single cigarette empty head detection method based on machine vision and special equipment |
CN104048675A (en) * | 2014-06-26 | 2014-09-17 | 东南大学 | Integrated navigation system fault diagnosis method based on Gaussian process regression |
CN105105326A (en) * | 2015-07-28 | 2015-12-02 | 郑州长河电子工程有限公司 | Industrial control computer (IPC)-based cigarette making machine tobacco rod weight control and quality detection integrated device and method |
CN205176622U (en) * | 2015-12-11 | 2016-04-20 | 成都博发控制技术有限责任公司 | Stable and weight optimal control system of a cigarette feed |
CN205337572U (en) * | 2016-01-20 | 2016-06-29 | 河南中烟工业有限责任公司 | Cigarette machine induced draft room and cigarette machine |
CN205390286U (en) * | 2016-02-29 | 2016-07-27 | 宝鸡市信诚电子有限公司 | Novel cigarette zhi chongliang control structure |
CN106056127A (en) * | 2016-04-07 | 2016-10-26 | 江南大学 | GPR (gaussian process regression) online soft measurement method with model updating |
-
2018
- 2018-02-09 CN CN201810134472.4A patent/CN108323797B/en active Active
Patent Citations (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US3146910A (en) * | 1960-11-07 | 1964-09-01 | Industrial Nucleonics Corp | Control system |
GB1313861A (en) * | 1969-06-02 | 1973-04-18 | Industrial Nucleonics Corp | System and method for optimizing processor or equipment profit |
EP1048568A1 (en) * | 1999-04-28 | 2000-11-02 | Focke & Co. (GmbH & Co.) | Method and device for checking cigarettes |
JP2001333758A (en) * | 2000-05-24 | 2001-12-04 | Japan Tobacco Inc | Combustion test apparatus |
CN1975705A (en) * | 2005-11-28 | 2007-06-06 | 颐中烟草(集团)有限公司 | Cigarette internal quality index extimating method based on regression function estimating SVM |
CN201640440U (en) * | 2010-03-19 | 2010-11-24 | 河南中烟工业有限责任公司 | Cigarette weight control system of cigarette making machine |
CN102406234A (en) * | 2011-07-13 | 2012-04-11 | 常德烟草机械有限责任公司 | Cigarette position signal generation and quality detection rejection method |
CN103385539A (en) * | 2013-08-02 | 2013-11-13 | 南京文采科技有限责任公司 | Single cigarette empty head detection method based on machine vision and special equipment |
CN104048675A (en) * | 2014-06-26 | 2014-09-17 | 东南大学 | Integrated navigation system fault diagnosis method based on Gaussian process regression |
CN105105326A (en) * | 2015-07-28 | 2015-12-02 | 郑州长河电子工程有限公司 | Industrial control computer (IPC)-based cigarette making machine tobacco rod weight control and quality detection integrated device and method |
CN205176622U (en) * | 2015-12-11 | 2016-04-20 | 成都博发控制技术有限责任公司 | Stable and weight optimal control system of a cigarette feed |
CN205337572U (en) * | 2016-01-20 | 2016-06-29 | 河南中烟工业有限责任公司 | Cigarette machine induced draft room and cigarette machine |
CN205390286U (en) * | 2016-02-29 | 2016-07-27 | 宝鸡市信诚电子有限公司 | Novel cigarette zhi chongliang control structure |
CN106056127A (en) * | 2016-04-07 | 2016-10-26 | 江南大学 | GPR (gaussian process regression) online soft measurement method with model updating |
Non-Patent Citations (1)
Title |
---|
仲伟庆: "卷烟机数字PID重量控制***的实现", 《烟草科技》 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109330018A (en) * | 2018-10-30 | 2019-02-15 | 浙江中烟工业有限责任公司 | A kind of setting method of cigarette weight control system of cigarette making machine aspirator tape starting position |
CN109330018B (en) * | 2018-10-30 | 2021-05-18 | 浙江中烟工业有限责任公司 | Method for setting starting position of cut tobacco suction belt of cigarette weight control system of cigarette making machine |
CN110037336A (en) * | 2019-04-19 | 2019-07-23 | 浙江中烟工业有限责任公司 | A kind of prediction technique of Cigarette circumference control system executing agency position |
CN110037336B (en) * | 2019-04-19 | 2021-09-03 | 浙江中烟工业有限责任公司 | Method for predicting position of actuating mechanism of cigarette circumference control system |
CN111562574A (en) * | 2020-05-22 | 2020-08-21 | 中国科学院空天信息创新研究院 | MIMO ground penetrating radar three-dimensional imaging method based on backward projection |
CN111562574B (en) * | 2020-05-22 | 2022-08-16 | 中国科学院空天信息创新研究院 | MIMO ground penetrating radar three-dimensional imaging method based on backward projection |
CN111932037A (en) * | 2020-09-23 | 2020-11-13 | 浙江创泰科技有限公司 | Parking space state prediction method and system based on machine learning |
CN113313106A (en) * | 2021-04-14 | 2021-08-27 | 深圳市睿达科技有限公司 | Feeding deviation rectifying method and device, computer equipment and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN108323797B (en) | 2020-11-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108323797A (en) | Cigarette Weight Control System based on GPR models starts position predicting method and system | |
CN112668235B (en) | Robot control method based on off-line model pre-training learning DDPG algorithm | |
US5245528A (en) | Process control apparatus and method for adjustment of operating parameters of controller of the process control apparatus | |
Yang et al. | Control of nonaffine nonlinear discrete-time systems using reinforcement-learning-based linearly parameterized neural networks | |
CN101390024A (en) | Operation control method, operation control device and operation control system | |
CN113408648A (en) | Unit combination calculation method combined with deep learning | |
CN115765050A (en) | Power system safety correction control method, system, equipment and storage medium | |
CN105867138A (en) | Stable platform control method and device based on PID controller | |
Vrabie | Online adaptive optimal control for continuous-time systems | |
CN112819224B (en) | Unit output prediction and confidence evaluation method based on deep learning fusion model | |
CN111008708A (en) | Parameter adjusting method and system for quasi-proportional resonant controller | |
CN113359704B (en) | Self-adaptive SAC-PID method suitable for complex unknown environment | |
CN114880932B (en) | Power grid operating environment simulation method, system, equipment and medium | |
US20210008718A1 (en) | Method, device and computer program for producing a strategy for a robot | |
CN116880191A (en) | Intelligent control method of process industrial production system based on time sequence prediction | |
CN111478331B (en) | Method and system for adjusting power flow convergence of power system | |
CN113642766B (en) | Method, device, equipment and medium for predicting power outage number of power system station | |
Vitay | Deep reinforcement learning | |
CN113991752A (en) | Power grid quasi-real-time intelligent control method and system | |
CN114386320A (en) | Steam turbine valve management method, device, equipment and storage medium | |
CN113359452B (en) | Controller design method and system based on Barzilai Borwein intelligent learning algorithm | |
CN111191815A (en) | Ultra-short-term output prediction method and system for wind power cluster | |
CN110322055A (en) | A kind of method and system improving data risk model scoring stability | |
CN115877811B (en) | Flow process treatment method, device and equipment | |
Peng et al. | Guided Deep Reinforcement Learning based on RBF-ARX Pseudo LQR in Single Stage Inverted Pendulum |
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 |