CN107991866A - Decoupling control methods of the MIMO based on the tight form Non-Model Controllers of SISO Yu local derviation information - Google Patents
Decoupling control methods of the MIMO based on the tight form Non-Model Controllers of SISO Yu local derviation information Download PDFInfo
- Publication number
- CN107991866A CN107991866A CN201711202254.1A CN201711202254A CN107991866A CN 107991866 A CN107991866 A CN 107991866A CN 201711202254 A CN201711202254 A CN 201711202254A CN 107991866 A CN107991866 A CN 107991866A
- Authority
- CN
- China
- Prior art keywords
- siso
- form non
- tight form
- output
- input
- 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
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
- G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance
Abstract
The invention discloses a kind of decoupling control methods of MIMO based on the tight form Non-Model Controllers of SISO Yu local derviation information, first according to MIMO (Multiple Input and Multiple Output, multiple-input and multiple-output) system coupling feature and tendentiousness feature, mimo system is resolved into multiple SISO (Single Input and Single Output, single-input single-output) systems to intercouple;SISO systems are controlled using the tight form Non-Model Controllers of SISO;Based on BP neural network, using local derviation information as input, target is minimised as to consider the system error function value of all SISO systematic errors contributions, using gradient descent method, and combine the gradient information that control input respectively treats setting parameter for controller respectively, systematic error backpropagation calculating is carried out, realizes the online self-tuning of the parameters such as penalty factor, the step factor of the tight form Non-Model Controllers of SISO, and synchronously realizes the online decoupling between multiple SISO systems.Method proposed by the present invention is the effective means for solving mimo system control problem, it can be achieved that good control effect.
Description
Technical field
The invention belongs to automation control area, and the tight form Non-Model Controllers of SISO are based on more particularly, to a kind of MIMO
With the decoupling control method of local derviation information.
Background technology
The control problem of MIMO (Multiple Input and Multiple Output, multiple-input and multiple-output) system,
One of significant challenge that control field is faced is automated all the time.Mimo system is typically characterised by coupling,
That is:The change of one input often makes multiple outputs change, and an output is often also not only by an input
Influence;But meanwhile this coupling is in most cases, especially in industrial process automation control field, and can body
Reveal tendentiousness feature, that is to say, that:The change of one input often tends to make some specifically export generation significant changes
And others output then being influenced smaller, an output often tends to be significantly affected by some specific input and be subject to other defeated
The influence entered is then smaller.The tendentiousness feature of mimo system, to be broken down into multiple SISO (Single Input and
Single Output, single-input single-output) system provides feasibility;And the coupling feature of mimo system, and imply that this
A little SISO systems must synchronously solve the online solution between multiple SISO systems when SISO controller is respectively adopted and is controlled
Coupling problem.
SISO controller has a variety of implementation methods, including the tight form Non-Model Controllers of SISO.The tight forms of SISO without
Model controller is a kind of new data drive control method, does not depend on any mathematical model information of controlled device, only according to
Rely the inputoutput data measured in real time in SISO controlled devices to be controlled the analysis and design of device, and realize concise, meter
Calculate and bear small and strong robustness, unknown nonlinear time-varying SISO systems can also be controlled well, there is good answer
Use prospect.The theoretical foundation of the tight form Non-Model Controllers of SISO, is collaborateed by Hou Zhong lifes with Jin Shangtai at it《Model-free is adaptive
Should control-theoretical with applying》Itd is proposed in (Science Press, 2013, page 56), its control algolithm is as follows:
Wherein, u (k) is the control input at k moment;E (k) is the systematic error at k moment;For the pseudo- gradient at k moment
Estimate;λ is penalty factor;ρ is step factor.
However, the tight form Non-Model Controllers of SISO need to rely on Heuristics punishment is previously set before actually coming into operation
The numerical value of the parameter such as factor lambda and step factor ρ, also not yet realizes penalty factor λ and step factor ρ etc. during actually coming into operation
The online self-tuning of parameter.Parameter effectively adjusts the shortage of means, not only makes the use tune of the tight form Non-Model Controllers of SISO
Examination process is time-consuming and laborious, and can also seriously affect the control effect of the tight form Non-Model Controllers of SISO sometimes, constrains
The popularization and application of the tight form Non-Model Controllers of SISO.That is:The tight form Non-Model Controllers of SISO were actually coming into operation
Also need to solve the problem of online self-tuning parameter in journey.
For this reason, the present invention proposes a kind of decoupling controls of MIMO based on the tight form Non-Model Controllers of SISO Yu local derviation information
Method processed, can synchronously solve the problems of the tight form Non-Model Controller online self-tuning parameters of SISO and multiple SISO systems it
Between the problem that decouples online, provide a kind of new method to solve the control problem of mimo system.
The content of the invention
In order to solve the problems, such as present in background technology, it is tight to be based on SISO it is an object of the present invention to provide a kind of MIMO
The decoupling control method of form Non-Model Controller and local derviation information.
For this reason, the above-mentioned purpose of the present invention is achieved through the following technical solutions, comprise the following steps:
Step (1):For with mi input (mi be integer) more than or equal to 2 and mo output (mo be more than or
Integer equal to 2) MIMO (Multiple Input and Multiple Output, multiple-input and multiple-output) system, choose
An input in mi input and an output in mo output, form a SISO (Single Input and
Single Output, single-input single-output) system;Repeated m time (m >=1 and m≤mi and m≤mo and m is integer), forms m
SISO systems, wherein the input of one of any SISO systems is all not as the input of other SISO systems, one of any SISO systems
The output of system is all not as the output of other SISO systems;The m SISO systems use the tight form Model free controls of m SISO
Device is controlled;
Step (2):For j-th of tight form Non-Model Controller of (1≤j≤m) SISO, the tight forms of j-th of SISO
Non-Model Controller parameter includes penalty factor λjWith step factor ρj;Determine the tight form Non-Model Controllers of j-th of SISO
Treat setting parameter, the tight form Non-Model Controllers of j-th of SISO treat setting parameter, be the tight forms of j-th of SISO without
Model controller parameter it is part or all of, include penalty factor λjWith step factor ρjOne of any or any number of combination;Really
Input layer number, node in hidden layer, the output layer number of nodes of fixed j-th of BP neural network, the output layer number of nodes is not
Setting parameter number is treated less than the tight form Non-Model Controllers of j-th of SISO;Initialize the implicit of j-th BP neural network
Layer weight coefficient, output layer weight coefficient;Initialize the local derviation information in { local derviation information j } set;If m >=2, for other m-
The tight form Non-Model Controllers of 1 SISO, repeat this step;
Step (3):The k moment will be denoted as current time;
Step (4):Calculate { the gradient information collection } at k moment;
For j-th of tight form Non-Model Controller of (1≤j≤m) SISO, there is step (4-1), step (4-2), step
The processing of (4-3), step (4-4), step (4-5):
The step (4-1) is:Actual value is exported with j-th of SISO system based on j-th of SISO systems output desired value,
Function is calculated using j-th of SISO systematic error, j-th of SISO systematic error at k moment is calculated, is denoted as ej(k);
The step (4-2) is:Local derviation information during { local derviation information j } is gathered, as the defeated of j-th BP neural network
Enter;
The step (4-3) is:Based on the input of j-th of BP neural network described in step (4-2), j-th of BP nerve
Network carries out forward calculation, and result of calculation is exported by the output layer of j-th of BP neural network, and it is tight to obtain j-th of SISO
Form Non-Model Controller treats the value of setting parameter;
The step (4-4) is:J-th of SISO systematic errors e obtained based on step (4-1)j(k), step (4-
3) the tight form Non-Model Controllers of j-th of SISO obtained treat the value of setting parameter, using the tight form model-free controls of SISO
The control algolithm of device processed, it is defeated in the control at k moment for controlled device to be calculated the tight form Non-Model Controllers of j-th of SISO
Enter uj(k);
The step (4-5) is:The control input u obtained based on step (4-4)j(k), the control input is calculated
uj(k) each gradient information for treating setting parameter at the k moment of the tight form Non-Model Controllers of j-th of SISO is directed to respectively, it is described
The specific formula for calculation of gradient information is as follows:
When the tight form Non-Model Controllers of j-th of SISO are treated to include penalty factor λ in setting parameterjWhen, the control
System input uj(k) it is directed to the penalty factor λjIt is in the gradient information at k moment:
When the tight form Non-Model Controllers of j-th of SISO are treated to include step factor ρ in setting parameterjWhen, the control
System input uj(k) it is directed to the step factor ρjIt is in the gradient information at k moment:
Wherein,For the tight form Non-Model Controllers of j-th of SISO the k moment pseudo- gradient estimate;
The set of the above-mentioned whole gradient information is denoted as { gradient information j }, is put into set { gradient information collection };
Gradient information in { gradient information j } set by described in is sequentially denoted as { the local derviation information j } collection of previous moment
Local derviation information in conjunction, i.e.,:When the tight form Non-Model Controllers of j-th of SISO are treated to include penalty factor λ in setting parameterj
Gradient information in { gradient information j } set described in Shi ZeIt is denoted as in { the local derviation information j } set of previous moment
Local derviation informationWhen the tight form Non-Model Controllers of j-th of SISO are treated to include step factor in setting parameter
ρjGradient information in { gradient information j } set described in Shi ZeIt is denoted as in { the local derviation information j } set of previous moment
Local derviation information
If m=1, { the gradient information collection } is constant, subsequently into step (5);
If m >=2, make the tight form Non-Model Controllers of each SISO in the tight form Non-Model Controllers of m SISO
It is respectively provided with step (4-1), step (4-2), step (4-3), step (4-4), the processing procedure of step (4-5);When m SISO is tight
The tight form Non-Model Controllers of each SISO in form Non-Model Controller perform step (4-1), step completely
The processing of (4-2), step (4-3), step (4-4), step (4-5), then described { gradient information collection } include all { { gradient letters
Breath 1 ..., { gradient information m } } set, subsequently into step (5);
Step (5):For j-th of tight form Non-Model Controller of (1≤j≤m) SISO, with the value of system error function most
Small to turn to target, using gradient descent method, { the gradient information collection } obtained with reference to the step (4), carries out systematic error
Backpropagation calculates, and updates hidden layer weight coefficient, the output layer weight coefficient of j-th of BP neural network, is used as j-th of later moment in time
Hidden layer weight coefficient, output layer weight coefficient during BP neural network progress forward calculation, it is at the same time synchronous if m >=2 to realize
The decoupling of the tight form Non-Model Controllers of j-th of SISO and the tight form Non-Model Controllers of other m-1 SISO;
If m=1, enter step (6);
If m >=2, for the tight form Non-Model Controllers of other m-1 SISO, this step is repeated, until complete
Hidden layer weight coefficient, the output layer weight coefficient of m, portion BP neural network are all updated, subsequently into step (6);
Step (6):The whole m control input { u1(k),…,um(k) } after acting on controlled device, controlled pair is obtained
As whole m SISO systems output actual value in later moment in time, back to step (3), repeat step (3) arrives step (6).
While using above-mentioned technical proposal, the present invention can also be used or combined using technology further below
Scheme:
, can be in step (4-1), step (4-2), step (4-3), step (4-4), step if m >=2 in step (4)
Suddenly between any one step or any two step of (4-5) or before step (4-1) or step (4-5) changes the value of j afterwards,
To carry out the processing procedure of the tight form Non-Model Controllers of other SISO, the change for j numerical value can be sequentially value or nothing
Regularly value, as long as making the processing procedure for the tight form Non-Model Controllers of each SISO, its sequencing is step (4-
1), step (4-2), step (4-3), step (4-4), step (4-5), and the two of the tight form Non-Model Controllers of each SISO
Between a step or before step (4-1) or after step (4-5), the tight form model-free controls of other SISO can be performed as needed
The step of device processed (4-1), step (4-2), step (4-3), step (4-4), step (4-5) processing.
J-th of SISO systematic errors in the step (4-1) calculate argument of function and include j-th of SISO system
System output desired value exports actual value with j-th of SISO system.
J-th of SISO systematic errors in the step (4-1) calculate function and use
WhereinDesired value, y are exported for j-th of SISO system that the k moment setsj(k) j-th of the SISO obtained for k instance samples
System exports actual value;Or useWhereinFor j-th of SISO system at k+1 moment
System output desired value, yj(k) j-th of the SISO system obtained for k instance samples exports actual value.
The independent variable of the system error function in the step (5) includes m SISO systematic error, m SISO system
System output desired value, m SISO system output actual value one of arbitrarily or any number of combination.
The system error function in the step (5) isWherein, ej(k) it is the
J SISO systematic error, Δ uj(k)=uj(k)-uj(k-1), ajWith bjFor the constant more than or equal to 0,1≤j≤m.
Decoupling control methods of the MIMO provided by the invention based on the tight form Non-Model Controllers of SISO Yu local derviation information, can
Solved between multiple SISO systems with the synchronous problem for solving the tight form Non-Model Controller online self-tuning parameters of SISO and online
The problem of coupling, so as to fulfill the decoupling control of mimo system.
Brief description of the drawings
Fig. 1 is the principle of the present invention block diagram;
Fig. 2 is j-th of BP neural network structure diagram that the present invention uses;
Fig. 3 is two inputs, two output mimo system in penalty factor λjWith step factor ρj1st SISO during Self-tuning System at the same time
The control effect figure of system;
Fig. 4 is two inputs, two output mimo system in penalty factor λjWith step factor ρj2nd SISO during Self-tuning System at the same time
The control effect figure of system;
Fig. 5 is two inputs, two output mimo system in penalty factor λjWith step factor ρjControl at the same time during Self-tuning System is defeated
Enter figure;
Fig. 6 is two inputs, two output mimo system in penalty factor λjWith step factor ρjAt the same time Self-tuning System when punishment because
Sub- λjChange curve;
Fig. 7 is two inputs, two output mimo system in penalty factor λjWith step factor ρjAt the same time Self-tuning System when step-length because
Sub- ρjChange curve;
Fig. 8 is two inputs, two output mimo system in penalty factor λjFix and step factor ρj1st SISO during Self-tuning System
The control effect figure of system;
Fig. 9 is two inputs, two output mimo system in penalty factor λjFix and step factor ρj2nd SISO during Self-tuning System
The control effect figure of system;
Figure 10 is two inputs, two output mimo system in penalty factor λjFix and step factor ρjControl during Self-tuning System is defeated
Enter figure;
Figure 11 is two inputs, two output mimo system in penalty factor λjFix and step factor ρjStep-length during Self-tuning System because
Sub- ρjChange curve.
Embodiment
The present invention is further described with specific embodiment below in conjunction with the accompanying drawings.
Fig. 1 gives the principle of the present invention block diagram.According to the coupling feature of mimo system and tendentiousness feature, one is chosen
A input and an output, form a SISO system, Repeated m time, intercouple so that mimo system is resolved into m
SISO systems, while the input of wherein one of any SISO systems is all not as the input of other SISO systems, it is one of any
The output of SISO systems is all not as the output of other SISO systems;M SISO system uses the tight form model-free controls of m SISO
Device processed is controlled.
For j-th of tight form Non-Model Controller of (1≤j≤m) SISO, the tight form model-free controls of j-th of SISO
Device parameter processed includes penalty factor λjWith step factor ρj;Determine that the tight form Non-Model Controllers of j-th of SISO are waited to adjust ginseng
Number, the tight form Non-Model Controllers of j-th of SISO treat setting parameter, are the tight form Model free controls of j-th of SISO
Device parameter it is part or all of, include penalty factor λjWith step factor ρjOne of any or any number of combination;In Fig. 1,
The tight form Non-Model Controllers of j SISO treat that setting parameter is penalty factor λjWith step factor ρj;Determine j-th of BP nerve net
Input layer number, node in hidden layer, the output layer number of nodes of network, the output layer number of nodes are no less than described j-th
The tight form Non-Model Controllers of SISO treat setting parameter number;Initialize hidden layer weight coefficient, the output of j-th of BP neural network
Layer weight coefficient;Initialize the local derviation information in { local derviation information j } set.For the tight form Model free controls of other m-1 SISO
Device, repeats the work described in this paragraph.
The k moment will be denoted as current time.
For j-th of tight form Non-Model Controller of (1≤j≤m) SISO, it is actual that system output is obtained by sampling first
Value yj(k), by system output desired value yj *(k) actual value y is exported with systemj(k) j-th SISO system of the difference as the k moment
System error ej(k);Then the local derviation information during { local derviation information j } is gathered, the input as j-th of BP neural network;J-th
BP neural network carries out forward calculation, and result of calculation is exported by the output layer of j-th of BP neural network, obtained described j-th
The tight form Non-Model Controllers of SISO treat the value of setting parameter;Then, based on j-th of SISO systematic errors ej(k), j-th
The tight form Non-Model Controllers of SISO treat the value of setting parameter, using the control algolithm of the tight form Non-Model Controllers of SISO, meter
Calculation obtains the tight form Non-Model Controllers of j-th of SISO and is directed to control input u of the controlled device at the k momentj(k);Calculate control
Input uj(k) each gradient information for treating setting parameter at the k moment of the tight form Non-Model Controllers of j-th of SISO is directed to respectively,
All the set of the gradient information is denoted as { gradient information j }, is put into set { gradient information collection }, while { gradient is believed by described in
Breath j set in gradient information be sequentially denoted as previous moment { the local derviation information j } set in local derviation information.For it
The tight form Non-Model Controllers of his m-1 SISO, repeat the work described in this paragraph, include { the gradient information collection }
All set of { { gradient information 1 } ..., { gradient information m } }.
Target is minimised as with the value of system error function, is contributed in Fig. 1 with considering whole m SISO systematic errors
System error functionValue be minimised as target, using gradient descent method, with reference to { the gradient information collection }, into
The backpropagation of row systematic error calculates, and hidden layer weight coefficient, the output layer weight coefficient of j-th of BP neural network is updated, as rear
Hidden layer weight coefficient, output layer weight coefficient during one j-th of moment BP neural network progress forward calculation, and synchronously realize jth
The decoupling of a tight form Non-Model Controllers of SISO and the tight form Non-Model Controllers of other m-1 SISO.For other m-1
The tight form Non-Model Controllers of SISO, repeat the work described in this paragraph, until the hidden layer of whole m BP neural network
Weight coefficient, output layer weight coefficient are all updated.
The whole m control input { u1(k),…,um(k) } after acting on controlled device, controlled device is obtained latter
Whole m SISO systems output actual value at moment, back to foregoing【The k moment will be denoted as current time】Paragraph, after starting
Decoupling control processes of the MIMO at one moment based on the tight form Non-Model Controllers of SISO Yu local derviation information.
Fig. 2 gives j-th of (1≤j≤m) BP neural network structure diagram that the present invention uses.J-th of BP nerve net
Network can use structure of the hidden layer for individual layer, can also use structure of the hidden layer for multilayer.In the schematic diagram of Fig. 2, it is
For the sake of simplicity, j-th of BP neural network employs the structure that hidden layer is individual layer, that is, uses by input layer, individual layer hidden layer, defeated
Go out the Three Tiered Network Architecture of layer composition, input layer number is set to treat that setting parameter number (treats that setting parameter number is 2 in Fig. 2
It is a), node in hidden layer 6, output layer number of nodes is set to treat setting parameter number (treating that setting parameter number is 2 in Fig. 2).
The node of input layer, the local derviation information in gathering with { local derviation information j }Correspond to respectively.Output layer
Node, the penalty factor λ with the tight form Non-Model Controllers of j-th of SISOjWith step factor ρjCorrespond to respectively.J-th of BP nerve
The hidden layer weight coefficient of network, the renewal process of output layer weight coefficient are specially:Mesh is minimised as with the value of system error function
Mark, to consider the system error function that whole m SISO systematic errors are contributed in Fig. 2Value be minimised as mesh
Mark, using gradient descent method, with reference to { the gradient information collection }, carries out systematic error backpropagation calculating, updates j-th of BP god
Hidden layer weight coefficient, output layer weight coefficient through network, when carrying out forward calculation as j-th BP neural network of later moment in time
Hidden layer weight coefficient, output layer weight coefficient, and synchronously realize the tight form Non-Model Controllers of j-th of SISO and other m-1
The decoupling of the tight form Non-Model Controllers of SISO.
It is the specific embodiment of the present invention below.
Controlled device exports mimo system for two inputs two of typical non linear:
y1(k)=x11(k)
y2(k)=x21(k)
Wherein, a (k)=1+0.1sin (2k π/1500), b (k)=1+0.1cos (2k π/1500).
System output desired value y*(k) it is as follows:
In this embodiment, according to the coupling feature of the mimo system and tendentiousness feature, by mimo system point
Solution is into the 2 SISO systems (m=2) to intercouple:The input of 1st SISO system is u1(k), export as y1(k);2nd
The input of SISO systems is u2(k), export as y2(k).2 SISO systems are carried out using the tight form Non-Model Controllers of 2 SISO
Control.1st BP neural network and the 2nd BP neural network are using three be made of input layer, individual layer hidden layer, output layer
Layer network structure, what input layer number was set to respective controller treats setting parameter number, and node in hidden layer is set to 6
A, what output layer number of nodes was set to respective controller treats setting parameter number.
For above-mentioned specific embodiment, two groups of verification experimental verifications have been carried out altogether.
During first group of verification experimental verification, the input layer number of BP neural network and output layer number of nodes are 2 in Fig. 2,
To the penalty factor λ of the tight form Non-Model Controllers of 2 SISOjWith step factor ρjSelf-tuning System at the same time is carried out, Fig. 3 is the 1st
The control effect figure of SISO systems, Fig. 4 are the control effect figure of the 2nd SISO system, and Fig. 5 inputs figure in order to control, and Fig. 6 is punishment
Factor lambdajChange curve, Fig. 7 are step factor ρjChange curve.The result shows that method of the invention is tight by designing 2 SISO
Form Non-Model Controller, and combine penalty factor λ of the BP neural network to the tight form Non-Model Controllers of each SISOjAnd step
Long factor ρjSelf-tuning System at the same time is carried out, can realize good control effect, and realize the decoupling control of mimo system.
During second group of verification experimental verification, the input layer number of BP neural network and output layer number of nodes are 1 in Fig. 2,
First by 2 penalty factor λjPenalty factor λ when value is first group of verification experimental verification is fixed respectivelyjAverage value (j=1,2), so
Afterwards to the step factor ρ of the tight form Non-Model Controllers of 2 SISOjSelf-tuning System is carried out, Fig. 8 is the control of the 1st SISO system
Design sketch, Fig. 9 are the control effect figure of the 2nd SISO system, and Figure 10 inputs figure in order to control, and Figure 11 is step factor ρjChange is bent
Line.The result shows that method of the invention is in penalty factor λjBy designing the tight form Non-Model Controllers of 2 SISO when fixed,
And combine step factor ρ of the BP neural network to the tight form Non-Model Controllers of each SISOjSelf-tuning System is carried out, can be realized good
Good control effect, and realize the decoupling control of mimo system.
J-th of (1≤j≤m) SISO systems output should it is expected it is emphasized that in above-mentioned specific embodiment
ValueWith j-th of SISO systems output actual value yj(k) difference is as j-th of SISO systematic errors ej(k), that is, Only described j-th of SISO systematic errors calculate a kind of method in function;Can also be by k+1
J-th of SISO systems output desired value of momentWith j-th of SISO systems output actual value y of k momentj(k) difference conduct
J-th of SISO systematic errors ej(k), that is,J-th of SISO systematic errors calculate
Function can also use independent variable include j-th SISO system export desired value and j-th of SISO system export actual value its
Its computational methods, for example,To the controlled device of above-mentioned specific embodiment and
Speech, calculates function using above-mentioned different systematic error, can realize good control effect.
More mesh should be minimised as with the value of system error function it is emphasized that in above-mentioned specific embodiment
When marking hidden layer weight coefficient, the output layer weight coefficient to update j-th of BP neural network, the system error function employs comprehensive
Close the system error function for considering whole m SISO systematic error contributionsOne in only described system error function
Kind function;The system error function can also use independent variable to include m SISO systematic error, m SISO systems output phase
Prestige value, m SISO system output actual value one of any or any number of combination other functions, for example, the system
Error function usesOrNamely useAnother letter
Number form formula;Again for example, the system error function usesWherein, ej(k) it is j-th
SISO systematic errors, Δ uj(k)=uj(k)-uj(k-1), ajWith bjFor the constant more than or equal to 0,1≤j≤m;Obviously, b is worked asj
During equal to 0, the system error function only accounts forContribution, show minimize target be systematic error minimum,
Exactly pursue precision height;And work as bjDuring more than 0, the system error function considers at the same timeContribution andTribute
Offer, for the target for showing to minimize while pursuit systematic error is small, also the change of pursuit control input is small, that is, both pursues essence
The high pursuit again of degree manipulates steady.For the controlled device of above-mentioned specific embodiment, using above-mentioned different system error function, all
It can realize good control effect;Only consider with system error functionControl effect during contribution is compared, and is missed in system
Difference function considers at the same timeContribution andContribution when its control accuracy slightly reduce and its manipulate stationarity then have
Improve.
In addition should be it is emphasized that the step (4-1) can both be adopted to the execution method of the step (4-5)
After waiting the tight form Non-Model Controllers of a SISO to be finished from the step (4-1) to the step (4-5), then open
Begin to perform method of the tight form Non-Model Controllers of another SISO from the step (4-1) to the step (4-5);Can also
The steps (4-1) are performed to all from whole using waiting more than one and being less than a tight form Non-Model Controllers of SISO of m
After performing the step (4-5), then start to perform the tight form Non-Model Controllers of other SISO from the step (4-1) to
The method of the step (4-5);The tight form Non-Model Controllers of m SISO of whole can also be used from all execution steps
The method of (4-1) to all execution steps (4-5).
Finally should be it is emphasized that the tight form Non-Model Controllers of (1≤j≤m) SISO j-th described be waited to adjust ginseng
Number, includes penalty factor λjWith step factor ρjOne of any or any number of combination;In above-mentioned specific embodiment, first group
Penalty factor λ during verification experimental verificationjWith step factor ρjRealize while Self-tuning System, penalty factor λ during second group of verification experimental verificationjGu
Determine and step factor ρjRealize Self-tuning System;In practical application, it can also select to treat appointing for setting parameter as the case may be
Meaning kind combination, for example, step factor ρjFix and penalty factor λjRealize Self-tuning System;In addition, the tight form model-free controls of SISO
Device processed treats setting parameter, includes but not limited to penalty factor λjWith step factor ρj, for example, as the case may be, can be with
Including pseudo- gradient estimateEtc. parameter.
Above-mentioned embodiment is used for illustrating the present invention, is merely a preferred embodiment of the present invention, rather than to this
Invention is limited, and in the protection domain of spirit and claims of the present invention, to any modification of the invention made, is equal
Replace, improve etc., both fall within protection scope of the present invention.
Claims (5)
- Decoupling control methods of the 1.MIMO based on the tight form Non-Model Controllers of SISO Yu local derviation information, it is characterised in that including with Lower step:Step (1):For with mi input (mi is the integer more than or equal to 2), (mo is more than or equal to 2 with mo output Integer) MIMO (Multiple Input and Multiple Output, multiple-input and multiple-output) system, it is defeated to choose mi The input entered and an output in mo output, form a SISO (Single Input and Single Output, single-input single-output) system;Repeated m time (m >=1 and m≤mi and m≤mo and m is integer), forms m SISO system System, wherein the input of one of any SISO systems is all not as the input of other SISO systems, one of any SISO systems it is defeated Go out all not as the output of other SISO systems;The m SISO systems are carried out using the tight form Non-Model Controllers of m SISO Control;Step (2):For j-th of tight form Non-Model Controller of (1≤j≤m) SISO, the tight forms of j-th of SISO are without mould Type controller parameter includes penalty factor λjWith step factor ρj;It is whole to determine that the tight form Non-Model Controllers of j-th of SISO are treated Determine parameter, the tight form Non-Model Controllers of j-th of SISO treat setting parameter, are the tight form model-frees of j-th of SISO Controller parameter it is part or all of, include penalty factor λjWith step factor ρjOne of any or any number of combination;Determine Input layer number, node in hidden layer, the output layer number of nodes of j BP neural network, the output layer number of nodes are no less than The tight form Non-Model Controllers of j-th of SISO treat setting parameter number;Initialize the hidden layer power of j-th of BP neural network Coefficient, output layer weight coefficient;Initialize the local derviation information in { local derviation information j } set;If m >=2, for other m-1 The tight form Non-Model Controllers of SISO, repeat this step;Step (3):The k moment will be denoted as current time;Step (4):Calculate { the gradient information collection } at k moment;For j-th of tight form Non-Model Controller of (1≤j≤m) SISO, there is step (4-1), step (4-2), step (4- 3), the processing of step (4-4), step (4-5):The step (4-1) is:Actual value is exported with j-th of SISO system based on j-th of SISO systems output desired value, is used J-th of SISO systematic error calculates function, and j-th of SISO systematic error at k moment is calculated, is denoted as ej(k);The step (4-2) is:Local derviation information during { local derviation information j } is gathered, the input as j-th of BP neural network;The step (4-3) is:Based on the input of j-th of BP neural network described in step (4-2), j-th of BP neural network Forward calculation is carried out, result of calculation is exported by the output layer of j-th of BP neural network, obtains the tight forms of j-th of SISO Non-Model Controller treats the value of setting parameter;The step (4-4) is:J-th of SISO systematic errors e obtained based on step (4-1)j(k), step (4-3) obtains To the tight form Non-Model Controllers of j-th of SISO treat the value of setting parameter, using the tight form Non-Model Controllers of SISO Control algolithm, be calculated the tight form Non-Model Controllers of j-th of SISO for controlled device the k moment control input uj (k);The step (4-5) is:The control input u obtained based on step (4-4)j(k), the control input u is calculatedj(k) The each gradient information for treating setting parameter at the k moment of the tight form Non-Model Controllers of j-th of SISO, the gradient letter are directed to respectively The specific formula for calculation of breath is as follows:When the tight form Non-Model Controllers of j-th of SISO are treated to include penalty factor λ in setting parameterjWhen, the control input uj(k) it is directed to the penalty factor λjIt is in the gradient information at k moment:When the tight form Non-Model Controllers of j-th of SISO are treated to include step factor ρ in setting parameterjWhen, the control input uj(k) it is directed to the step factor ρjIt is in the gradient information at k moment:Wherein,For the tight form Non-Model Controllers of j-th of SISO the k moment pseudo- gradient estimate;The set of the above-mentioned whole gradient information is denoted as { gradient information j }, is put into set { gradient information collection };Gradient information in { gradient information j } set by described in is sequentially denoted as in { the local derviation information j } set of previous moment Local derviation information, i.e.,:When the tight form Non-Model Controllers of j-th of SISO are treated to include penalty factor λ in setting parameterjShi Ze Gradient information in { the gradient information j } setIt is denoted as inclined in { the local derviation information j } set of previous moment Lead informationWhen the tight form Non-Model Controllers of j-th of SISO are treated to include step factor ρ in setting parameterjWhen Gradient information in then described { gradient information j } setIt is denoted as in { the local derviation information j } set of previous moment Local derviation informationIf m=1, { the gradient information collection } is constant, subsequently into step (5);If m >=2, the tight form Non-Model Controllers of each SISO in the tight form Non-Model Controllers of m SISO are made to have There are step (4-1), step (4-2), step (4-3), step (4-4), the processing procedure of step (4-5);When the tight forms of m SISO The tight form Non-Model Controllers of each SISO in Non-Model Controller perform step (4-1), step (4- completely 2), the processing of step (4-3), step (4-4), step (4-5), then described { gradient information collection } include all { { gradient informations 1 } ..., { gradient information m } } set, subsequently into step (5);Step (5):For j-th of tight form Non-Model Controller of (1≤j≤m) SISO, minimized with the value of system error function For target, using gradient descent method, { the gradient information collection } obtained with reference to the step (4), it is reverse to carry out systematic error Propagate and calculate, update hidden layer weight coefficient, the output layer weight coefficient of j-th of BP neural network, as j-th of BP god of later moment in time Hidden layer weight coefficient, output layer weight coefficient during through network progress forward calculation, are at the same time synchronously realized j-th if m >=2 The decoupling of the tight form Non-Model Controllers of SISO and the tight form Non-Model Controllers of other m-1 SISO;If m=1, enter step (6);If m >=2, for the tight form Non-Model Controllers of other m-1 SISO, this step is repeated, until whole m Hidden layer weight coefficient, the output layer weight coefficient of BP neural network are all updated, subsequently into step (6);Step (6):The whole m control input { u1(k),…,um(k) } after acting on controlled device, obtain controlled device and exist Whole m SISO systems output actual value of later moment in time, back to step (3), repeat step (3) arrives step (6).
- 2. decoupling control sides of the MIMO according to claim 1 based on the tight form Non-Model Controllers of SISO Yu local derviation information Method, it is characterised in that j-th of SISO systematic errors in the step (4-1) calculate argument of function and include j-th SISO systems export desired value and export actual value with j-th of SISO system.
- 3. decoupling controls of the MIMO according to claim 1 or 2 based on the tight form Non-Model Controllers of SISO Yu local derviation information Method processed, it is characterised in that j-th of SISO systematic errors in the step (4-1) calculate function and useWhereinDesired value, y are exported for j-th of SISO system that the k moment setsj(k) it is the k moment Sample obtained j-th of SISO system output actual value;Or useWhereinFor J-th of SISO system output desired value at k+1 moment, yj(k) j-th of the SISO system output reality obtained for k instance samples Value.
- 4. decoupling control sides of the MIMO according to claim 1 based on the tight form Non-Model Controllers of SISO Yu local derviation information Method, it is characterised in that the independent variable of the system error function in the step (5) includes m SISO systematic error, m SISO systems output desired value, m SISO system output actual value one of arbitrarily or any number of combination.
- 5. decoupling controls of the MIMO according to claim 1 or 4 based on the tight form Non-Model Controllers of SISO Yu local derviation information Method processed, it is characterised in that the system error function in the step (5) isIts In, ej(k) it is j-th of SISO systematic error, Δ uj(k)=uj(k)-uj(k-1), ajWith bjFor the constant more than or equal to 0,1 ≤j≤m。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711202254.1A CN107991866B (en) | 2017-11-24 | 2017-11-24 | Decoupling control method for MIMO based on SISO tight format model-free controller and partial derivative information |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711202254.1A CN107991866B (en) | 2017-11-24 | 2017-11-24 | Decoupling control method for MIMO based on SISO tight format model-free controller and partial derivative information |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107991866A true CN107991866A (en) | 2018-05-04 |
CN107991866B CN107991866B (en) | 2020-08-21 |
Family
ID=62032385
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711202254.1A Active CN107991866B (en) | 2017-11-24 | 2017-11-24 | Decoupling control method for MIMO based on SISO tight format model-free controller and partial derivative information |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107991866B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112015083A (en) * | 2020-06-18 | 2020-12-01 | 浙江大学 | Parameter self-tuning method of SISO (SISO) compact-format model-free controller based on ensemble learning |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5513098A (en) * | 1993-06-04 | 1996-04-30 | The Johns Hopkins University | Method for model-free control of general discrete-time systems |
CN1274435A (en) * | 1997-10-06 | 2000-11-22 | 美国通控集团公司 | Model-free adaptive process control |
CN101957598A (en) * | 2010-09-26 | 2011-01-26 | 上海电力学院 | Gray model-free control method for large time lag system |
CN102033492A (en) * | 2010-12-29 | 2011-04-27 | 国核电力规划设计研究院 | Linear neuron on-line learning adaptive control method and controller for passive system |
CN103399487A (en) * | 2013-07-30 | 2013-11-20 | 东北石油大学 | Nonlinear MIMO (multiple input multiple output) system-based decoupling control method and device |
CN105676632A (en) * | 2016-01-26 | 2016-06-15 | 沈阳化工大学 | Model-free adaptive optimized control method for PVC polymerization process |
-
2017
- 2017-11-24 CN CN201711202254.1A patent/CN107991866B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5513098A (en) * | 1993-06-04 | 1996-04-30 | The Johns Hopkins University | Method for model-free control of general discrete-time systems |
CN1274435A (en) * | 1997-10-06 | 2000-11-22 | 美国通控集团公司 | Model-free adaptive process control |
CN101957598A (en) * | 2010-09-26 | 2011-01-26 | 上海电力学院 | Gray model-free control method for large time lag system |
CN102033492A (en) * | 2010-12-29 | 2011-04-27 | 国核电力规划设计研究院 | Linear neuron on-line learning adaptive control method and controller for passive system |
CN103399487A (en) * | 2013-07-30 | 2013-11-20 | 东北石油大学 | Nonlinear MIMO (multiple input multiple output) system-based decoupling control method and device |
CN105676632A (en) * | 2016-01-26 | 2016-06-15 | 沈阳化工大学 | Model-free adaptive optimized control method for PVC polymerization process |
Non-Patent Citations (3)
Title |
---|
MINGWANG ZHAO: "Neural-net-based model-free self-tuning controller with on-line self-learning ability for industrial furnace", 《1994 PROCEEDINGS OF IEEE INTERNATIONAL CONFERENCE ON CONTROL AND APPLICATIONS 》 * |
郭代银: "无模型自适应控制参数整定方法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
马平: "无模型控制器参数学习步长和惩罚因子的整定研究", 《仪器仪表学报》 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112015083A (en) * | 2020-06-18 | 2020-12-01 | 浙江大学 | Parameter self-tuning method of SISO (SISO) compact-format model-free controller based on ensemble learning |
Also Published As
Publication number | Publication date |
---|---|
CN107991866B (en) | 2020-08-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108287471A (en) | Inclined methods of self-tuning of the format Non-Model Controller based on systematic error of MIMO | |
CN108170029A (en) | Methods of self-tuning of the MIMO full format Non-Model Controller based on local derviation information | |
CN109674080B (en) | Tobacco leaf conditioning water adding amount prediction method, storage medium and terminal equipment | |
CN106022521B (en) | Short-term load prediction method of distributed BP neural network based on Hadoop architecture | |
CN103745273B (en) | Semiconductor fabrication process multi-performance prediction method | |
CN108268947A (en) | For improving the device and method of the processing speed of neural network and its application | |
CN108345213A (en) | Tight methods of self-tuning of the format Non-Model Controller based on systematic error of MIMO | |
CN108052006A (en) | Decoupling control methods of the MIMO based on SISO full format Non-Model Controller Yu local derviation information | |
CN109143853B (en) | Self-adaptive control method for liquid level of fractionating tower in petroleum refining process | |
CN108153151A (en) | Methods of self-tuning of the MIMO full format Non-Model Controller based on systematic error | |
CN107942655A (en) | Tight methods of self-tuning of the form Non-Model Controller based on systematic error of SISO | |
CN108181809A (en) | Tight methods of self-tuning of the form Non-Model Controller based on systematic error of MISO | |
CN103106331A (en) | Photo-etching line width intelligence forecasting method based on dimension-reduction and quantity-increment-type extreme learning machine | |
CN107991866A (en) | Decoupling control methods of the MIMO based on the tight form Non-Model Controllers of SISO Yu local derviation information | |
CN108132600A (en) | Tight methods of self-tuning of the form Non-Model Controller based on local derviation information of MIMO | |
CN108107722A (en) | Decoupling control methods of the MIMO based on the inclined form Non-Model Controllers of SISO and systematic error | |
CN108073072A (en) | Tight methods of self-tuning of the form Non-Model Controller based on local derviation information of SISO | |
CN108062021A (en) | Methods of self-tuning of the SISO full format Non-Model Controller based on local derviation information | |
CN107942654A (en) | Inclined methods of self-tuning of the form Non-Model Controller based on local derviation information of SISO | |
CN108287470A (en) | Inclined methods of self-tuning of the format Non-Model Controller based on local derviation information of MIMO | |
CN108154231A (en) | Methods of self-tuning of the MISO full format Non-Model Controller based on systematic error | |
CN107991865A (en) | Decoupling control methods of the MIMO based on the tight form Non-Model Controllers of SISO and systematic error | |
CN108107721A (en) | Decoupling control methods of the MIMO based on the inclined form Non-Model Controllers of SISO Yu local derviation information | |
CN107942674A (en) | Decoupling control methods of the MIMO based on SISO full format Non-Model Controller and systematic error | |
CN108107715A (en) | Methods of self-tuning of the MISO full format Non-Model Controller based on local derviation information |
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 |