CN108107715A - Methods of self-tuning of the MISO full format Non-Model Controller based on local derviation information - Google Patents

Methods of self-tuning of the MISO full format Non-Model Controller based on local derviation information Download PDF

Info

Publication number
CN108107715A
CN108107715A CN201711323329.1A CN201711323329A CN108107715A CN 108107715 A CN108107715 A CN 108107715A CN 201711323329 A CN201711323329 A CN 201711323329A CN 108107715 A CN108107715 A CN 108107715A
Authority
CN
China
Prior art keywords
miso
full format
local derviation
moment
information
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
Application number
CN201711323329.1A
Other languages
Chinese (zh)
Other versions
CN108107715B (en
Inventor
卢建刚
李雪园
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University ZJU
Original Assignee
Zhejiang University ZJU
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Zhejiang University ZJU filed Critical Zhejiang University ZJU
Priority to CN201711323329.1A priority Critical patent/CN108107715B/en
Publication of CN108107715A publication Critical patent/CN108107715A/en
Application granted granted Critical
Publication of CN108107715B publication Critical patent/CN108107715B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G05B13/0205—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric not using a model or a simulator of the controlled system
    • G05B13/024—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric not using a model or a simulator of the controlled system in which a parameter or coefficient is automatically adjusted to optimise the performance

Abstract

The invention discloses a kind of methods of self-tuning of the MISO full format Non-Model Controller based on local derviation information, by the use of local derviation information collection as the input of BP neural network, BP neural network carries out forward calculation and passes through output layer and export penalty factor, the MISO full format Non-Model Controllers such as step factor treat setting parameter, control input vector for controlled device is calculated using the control algolithm of MISO full format Non-Model Controllers, target is minimised as with the value of system error function, using gradient descent method, and it combines control input and is directed to each gradient information collection for treating setting parameter respectively, carry out systematic error backpropagation calculating, the hidden layer weight coefficient of online real-time update BP neural network, output layer weight coefficient, realize parameter self-tuning of the controller based on local derviation information.Methods of self-tuning of the MISO full format Non-Model Controller based on local derviation information proposed by the present invention can effectively overcome the on-line tuning problem of controller parameter, have good control effect to MISO system.

Description

Methods of self-tuning of the MISO full format Non-Model Controller based on local derviation information
Technical field
The invention belongs to automation control areas, and local derviation is based on more particularly, to a kind of MISO full format Non-Model Controller The methods of self-tuning of information.
Background technology
The control problem of MISO (Multiple Input and Single Output, multiple input single output) system, one One of significant challenge that control field is faced has been automated since straight.
The existing implementation method of MISO controllers includes MISO full format Non-Model Controllers.MISO full format model-frees Controller is a kind of new data drive control method, does not depend on any mathematical model information of controlled device, only relies upon The inputoutput data that MISO controlled devices measure in real time into line control unit analysis and design, and realize it is concise, calculate it is negative Small and strong robustness is carried on a shoulder pole, unknown nonlinear time-varying MISO system can also be controlled well, before there is good application Scape.The theoretical foundation of MISO full format Non-Model Controllers is collaborateed by Hou Zhong lifes with Jin Shangtai at it《Model-free adaption Control-theory and application》It is proposed in (Science Press, 2013, page 118), control algolithm is as follows：
Wherein, u (k) is that the control input at k moment is vectorial, u (k)=[u1(k),…,um(k)]T, m inputs a in order to control Number, Δ u (k)=u (k)-u (k-1)；E (k) is the systematic error at k moment；Δ y (k)=y (k)-y (k-1), y (k) is the k moment System output actual value；For the row matrix of the MISO system puppet piecemeal gradient estimate at k moment,For row matrixI-th piece of row matrix (i=1 ..., Ly+Lu),For row matrix2 norms；λ is penalty factor, ρ1,…,ρLy+LuFor step factor, Ly linearization length constants in order to control, Lu input linear length constants in order to control.
However, MISO full format Non-Model Controllers need to rely on Heuristics before actually coming into operation punishment is previously set Factor lambda and step factor ρ1,…,ρLy+LuEtc. parameters numerical value, penalty factor λ and step are also not yet realized during actually come into operation Long factor ρ1,…,ρLy+LuEtc. parameters online self-tuning.Parameter effectively adjusts the shortage of means, not only make MISO full format without The use debugging process of model controller is time-consuming and laborious, and can also seriously affect MISO full format Non-Model Controllers sometimes Control effect constrains the popularization and application of MISO full format Non-Model Controllers.That is：MISO full format Model free controls Device also needs to solve the problems, such as online self-tuning parameter during actually coming into operation.
For this purpose, in order to break the bottleneck for restricting MISO full format Non-Model Controller and promoting and applying, the present invention proposes one Kind methods of self-tuning of the MISO full format Non-Model Controller based on local derviation information.
The content of the invention
In order to solve the problems, such as present in background technology, it is an object of the present invention to provide a kind of MISO full format without mould Methods of self-tuning of the type controller based on local derviation information.
For this purpose, the above-mentioned purpose of the present invention is achieved through the following technical solutions, comprise the following steps：
Step (1)：For with MISO (Multiple of the m input (m is the integer more than or equal to 2) with 1 output Input and Single Output, multiple input single output) system, it is controlled using MISO full format Non-Model Controllers； Control linearization the length constant Ly, Ly for determining MISO full format Non-Model Controllers are the integer more than or equal to 1；Really Control input linearisation the length constant Lu, Lu for determining MISO full format Non-Model Controllers are the integer more than or equal to 1；It is described MISO full format Non-Model Controllers parameter includes penalty factor λ and step factor ρ1,…,ρLy+Lu；Determine MISO full format without Model controller treats setting parameter, and the MISO full format Non-Model Controller treats setting parameter, be the MISO full format without Model controller parameter it is part or all of, include penalty factor λ and step factor ρ1,…,ρLy+LuIt is one of arbitrary or arbitrary Kind combination；Determine the input layer number, node in hidden layer, output layer number of nodes of BP neural network, the output node layer Number is no less than the MISO full format Non-Model Controller and treats setting parameter number；Initialize the hidden layer of the BP neural network Weight coefficient, output layer weight coefficient；Local derviation information in initialization set { local derviation information collection }；
Step (2)：The k moment will be denoted as current time；
Step (3)：Actual value exported based on system output desired value and system, function is calculated using systematic error To the systematic error at k moment, e (k) is denoted as；
Step (4)：Using the local derviation information in the set { local derviation information collection } as the input of BP neural network, the BP Neutral net carries out forward calculation, and result of calculation is exported by the output layer of the BP neural network, obtains the full lattice of the MISO Formula Non-Model Controller treats the value of setting parameter；
Step (5)：The MISO full format that the systematic error e (k) that is obtained based on step (3), step (4) are obtained Non-Model Controller treats the value of setting parameter, and using the control algolithm of MISO full format Non-Model Controllers, MISO is calculated Full format Non-Model Controller is directed to control input vector u (k)=[u of the controlled device at the k moment1(k),…,um(k)]T；
Step (6)：J-th of control input u in the control input vector u (k) obtained for step (5)j(k) (1≤j≤m) calculates j-th of control input uj(k) treated respectively for each MISO full format Non-Model Controller Gradient information of the setting parameter at the k moment, specific formula for calculation are as follows：
When the MISO full format Non-Model Controller is when penalty factor λ and Lu=1 is included in setting parameter, described the J control input uj(k) it is for gradient informations of the penalty factor λ at the k moment：
When the MISO full format Non-Model Controller is treated to include penalty factor λ and Lu in setting parameter>When 1, the jth A control input uj(k) it is for gradient informations of the penalty factor λ at the k moment：
When the MISO full format Non-Model Controller is treated to include step factor ρ in setting parameteriAnd during 1≤i≤Ly, institute State j-th of control input uj(k) it is directed to the step factor ρiIt is in the gradient information at k moment：
When the MISO full format Non-Model Controller is treated to include step factor ρ in setting parameterLy+1When, described j-th Control input uj(k) it is directed to the step factor ρLy+1It is in the gradient information at k moment：
When the MISO full format Non-Model Controller is treated to include step factor ρ in setting parameteriAnd Ly+2≤i≤Ly+ Lu and Lu>When 1, j-th of control input uj(k) it is directed to the step factor ρiIt is in the gradient information at k moment：
Wherein, Δ uj(k)=uj(k)-uj(k-1), Δ y (k)=y (k)-y (k-1), y (k) is that the system at k moment exports Actual value,For the row matrix of the MISO system puppet piecemeal gradient estimate at k moment,For row matrixI-th Block row matrix (i=1 ..., Ly+Lu),For row matrixJ-th of gradient component estimate,For row Matrix2 norms；
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 of previous moment, i.e.,：When described MISO full format Non-Model Controller when in setting parameter include penalty factor λ when then it is described { gradient information j } set in ladder Spend informationIt is denoted as the local derviation information of previous momentWhen the MISO full format Non-Model Controller treat it is whole Determine to include step factor ρ in parameteriAnd the gradient information in { gradient information j } then described during 1≤i≤Ly+Lu set It is denoted as the local derviation information of previous moment
The set of the above-mentioned whole local derviation information is denoted as { local derviation information j }, is put into the set { local derviation information collection }；
Other m-1 control input in the control input vector u (k) obtained for step (5) repeat this Step, until the set { gradient information collection } includes the set of all { { gradient information 1 } ..., { gradient information m } }, simultaneously The set { local derviation information collection } includes the set of all { { local derviation information 1 } ..., { local derviation information m } }, subsequently into step (7)；
Step (7)：Target is minimised as with the value of system error function, using gradient descent method, is obtained with reference to step (6) The set { gradient information collection }, carry out systematic error backpropagation calculating, update BP neural network hidden layer weight coefficient, Output layer weight coefficient, as later moment in time BP neural network carry out forward calculation when hidden layer weight coefficient, output layer weight coefficient；
Step (8)：After the control input vector u (k) acts on controlled device, controlled device is obtained in later moment in time System exports actual value, back to step (2), repeats step (2) and arrives step (8).
While using above-mentioned technical proposal, the present invention can also be used or combined using technology further below Scheme：
It is defeated comprising system output desired value and system that the systematic error in the step (3) calculates argument of function Go out actual value.
The systematic error in the step (3) calculates function and uses e (k)=y*(k)-y (k), wherein y*(k) be k when The system output desired value of setting is carved, y (k) is that the system that k instance samples obtain exports actual value；Or using e (k)=y* (k+1)-y (k), wherein y*(k+1) system for the k+1 moment exports desired value, and y (k) is the system output that k instance samples obtain Actual value.
The independent variable of the system error function in the step (7) includes systematic error, system output desired value, is One of the arbitrary or any number of combination of system output actual value.
The system error function in the step (7) isWherein, e (k) misses for system Difference, Δ uj(k)=uj(k)-uj(k-1), bjTo be greater than or equal to 0 constant, 1≤j≤m.
Methods of self-tuning of the MISO full format Non-Model Controller based on local derviation information provided by the invention, Neng Goushi Existing good control effect, and effectively overcome penalty factor λ and step factor ρ1,…,ρLNeed the time-consuming and laborious difficulty adjusted Topic.
Description of the drawings
Fig. 1 is the principle of the present invention block diagram；
Fig. 2 is the BP neural network structure diagram that the present invention uses；
Fig. 3 is the single output MISO system of two inputs in penalty factor λ and step factor ρ1,ρ2,ρ3,ρ4Simultaneously during Self-tuning System Control effect figure；
Fig. 4 is the single output MISO system of two inputs in penalty factor λ and step factor ρ1,ρ2,ρ3,ρ4Simultaneously during Self-tuning System Control input figure；
Fig. 5 is the single output MISO system of two inputs in penalty factor λ and step factor ρ1,ρ2,ρ3,ρ4Simultaneously during Self-tuning System Penalty factor λ change curves；
Fig. 6 is the single output MISO system of two inputs in penalty factor λ and step factor ρ1,ρ2,ρ3,ρ4Simultaneously during Self-tuning System Step factor ρ1,ρ2,ρ3,ρ4Change curve；
Fig. 7 is fixed and step factor ρ for the single output MISO system of two inputs in penalty factor λ1,ρ2,ρ3,ρ4During Self-tuning System Control effect figure；
Fig. 8 is fixed and step factor ρ for the single output MISO system of two inputs in penalty factor λ1,ρ2,ρ3,ρ4During Self-tuning System Control input figure；
Fig. 9 is fixed and step factor ρ for the single output MISO system of two inputs in penalty factor λ1,ρ2,ρ3,ρ4During Self-tuning System Step factor ρ1,ρ2,ρ3,ρ4Change curve.
Specific embodiment
The present invention is further described in the following with reference to the drawings and specific embodiments.
Fig. 1 gives the principle of the present invention block diagram.For with m input (m is the integer more than or equal to 2) and 1 The MISO system of output is controlled using MISO full format Non-Model Controllers；Determine MISO full format Non-Model Controllers Control linearization length constant Ly, Ly is the integer more than or equal to 1；Determine MISO full format Non-Model Controllers Control input linearisation length constant Lu, Lu are the integer more than or equal to 1；MISO full format Non-Model Controller parameters include Penalty factor λ and step factor ρ1,…,ρLy+Lu；It determines that MISO full format Non-Model Controllers treat setting parameter, is described MISO full format Non-Model Controller parameters it is part or all of, include penalty factor λ and step factor ρ1,…,ρLy+LuAppoint One of meaning or any number of combination；In Fig. 1, MISO full format Non-Model Controller treats that setting parameter is penalty factor λ and step-length Factor ρ1,…,ρLy+Lu；The input layer number, node in hidden layer, output layer number of nodes of BP neural network are determined, wherein defeated Go out node layer number and be no less than MISO full format Non-Model Controllers to treat setting parameter number；Initialize the hidden of the BP neural network Weight coefficient containing layer, output layer weight coefficient；Local derviation information in initialization set { local derviation information collection }.
The k moment will be denoted as current time；By system output desired value y*(k) and system output actual value y (k) difference conduct The systematic error e (k) at k moment；Input of the local derviation information as BP neural network in { local derviation information collection }, BP nerves will be gathered Network carries out forward calculation, and result of calculation is exported by the output layer of BP neural network, obtains MISO full format Model free controls Device treats the value of setting parameter；Setting parameter is treated based on the systematic error e (k), the MISO full format Non-Model Controller Using the control algolithm of MISO full format Non-Model Controllers, MISO full format Non-Model Controller is calculated for quilt in value Control control input vector u (k)=[u of the object at the k moment1(k),…,um(k)]T；For in control input vector u (k) J control input uj(k) (1≤j≤m) calculates j-th of control input uj(k) each MISO full format is directed to respectively Non-Model Controller treats gradient information of the setting parameter at the k moment, and the set of all gradient informations is denoted as { gradient letter Cease j }, it is put into set { gradient information collection }；Gradient information in { gradient information j } set by described in is sequentially denoted as previous moment Local derviation information, and the set of all local derviation information is denoted as { local derviation information j }, it is put into the set { local derviation information collection }； For other m-1 control input in control input vector u (k), repeat until set { gradient information collection } bag Set containing all { { gradient information 1 } ..., { gradient information m } }, while the set { local derviation information collection } includes all { { partially Lead information 1 } ..., { local derviation information m } set；Then, with reference to the set { gradient information collection }, with system error function Value be minimised as target, with e in Fig. 12(k) target is minimised as, using gradient descent method, carries out systematic error backpropagation Calculate, update hidden layer weight coefficient, the output layer weight coefficient of BP neural network, before being carried out as later moment in time BP neural network to Hidden layer weight coefficient, output layer weight coefficient during calculating；After control input vector u (k) acts on controlled device, controlled pair is obtained As the system output actual value in later moment in time, the work described in this paragraph is then repeated, carries out the MISO of later moment in time Parameter self-tuning process of the full format Non-Model Controller based on local derviation information.
Fig. 2 gives the BP neural network structure diagram that the present invention uses.Hidden layer may be employed in BP neural network The structure of individual layer can also use structure of the hidden layer for multilayer.In the schematic diagram of Fig. 2, for simplicity, BP neural network The structure that hidden layer is individual layer is employed, that is, uses the Three Tiered Network Architecture being made of input layer, individual layer hidden layer, output layer, Input layer number is set to m × treat setting parameter number (treating setting parameter number for Ly+Lu+1 in Fig. 2), hidden layer node Number 10, output layer number of nodes are set to treat setting parameter number (treating setting parameter number for Ly+Lu+1 in Fig. 2).Input layer Number of nodes is divided into m groups, and every group of number of nodes to treat setting parameter number, with { local derviation information j } gather by the wherein node of jth group In local derviation informationIt corresponds to respectively.The node of output layer, with penalty factor λ With step factor ρ1,…,ρLy+LuIt corresponds to respectively.The hidden layer weight coefficient of BP neural network, the renewal process of output layer weight coefficient Specially：Target is minimised as with the value of system error function, with e in Fig. 22(k) target is minimised as, is declined using gradient Method with reference to the set { gradient information collection }, carries out systematic error backpropagation calculating, so as to update the implicit of BP neural network Layer weight coefficient, output layer weight coefficient.
It is the specific embodiment of the present invention below.
Controlled device exports MISO systems for two input lists of typical non linear：
System output desired value y*(k) it is as follows：
y*(k)=(- 1)round((k-1)/100)
In this embodiment, m=2.
The numerical value of the control linearization length constant Ly of MISO full format Non-Model Controllers is generally according to controlled pair The complexity of elephant and actual control effect are set, and generally between 1 to 5, conference excessively causes computationally intensive, so one As often take 1 or 3, Ly is taken as 1 in this embodiment；The control input linearisation length of MISO full format Non-Model Controllers Also the complexity generally according to controlled device and actual control effect are set the numerical value of constant Lu, generally 1 to 10 Between, too small to influence control effect, conference excessively causes computationally intensive, so generally often taking 3 or 5, in this embodiment Lu is taken as 3.
BP neural network uses the Three Tiered Network Architecture being made of input layer, individual layer hidden layer, output layer, input layer Number is set to 2 × treats setting parameter number, and node in hidden layer is set to 10, and output layer number of nodes is set to treat 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 is preset as 10 in Fig. 2, output layer number of nodes 5 are preset as, to penalty factor λ and step factor ρ1,ρ2,ρ3,ρ4Carry out Self-tuning System simultaneously, Fig. 3 design sketch in order to control, Fig. 4 Input figure in order to control, Fig. 5 are penalty factor λ change curves, and Fig. 6 is step factor ρ1,ρ2,ρ3,ρ4Change curve.The result shows that The method of the present invention passes through to penalty factor λ and step factor ρ1,ρ2,ρ3,ρ4Self-tuning System simultaneously is carried out, can be realized good Control effect, and can effectively overcome penalty factor λ and step factor ρ1,ρ2,ρ3,ρ4Need time-consuming and laborious adjusted Problem.
During second group of verification experimental verification, the input layer number of BP neural network is preset as 8 in Fig. 2, output layer number of nodes 4 are preset as, penalty factor λ is fixed to the average value of penalty factor λ when value is first group of verification experimental verification first, then to step Long factor ρ1,ρ2,ρ3,ρ4Self-tuning System is carried out, design sketch, Fig. 8 input figure to Fig. 7 in order to control in order to control, and Fig. 9 is step factor ρ1,ρ2, ρ3,ρ4Change curve.As a result again show that, method of the invention is when penalty factor λ is fixed by step factor ρ1,ρ2,ρ3, ρ4Self-tuning System is carried out, good control effect can be realized, and can effectively overcome step factor ρ1,ρ2,ρ3,ρ4It needs time-consuming The laborious problem adjusted.
System should be exported into desired value y it is emphasized that in above-mentioned specific embodiment*(k) it is real with system output The difference of actual value y (k) is as systematic error e (k), that is, e (k)=y*(k)-y (k) is only that the systematic error calculates function In a kind of method；The system at k+1 moment can also be exported desired value y*(k+1) actual value y is exported with the system at k moment (k) difference is as systematic error e (k), that is, e (k)=y*(k+1)-y(k)；The systematic error calculates function and can also adopt Other computational methods of system output desired value and system output actual value are included with independent variable, for example, For the controlled device of above-mentioned specific embodiment, missed using above-mentioned different system Difference calculates function, 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 BP neural network, the system error function uses e2(k), only For a kind of function in the system error function；The system error function can also use independent variable include systematic error, System output desired value, system output actual value one of arbitrary or any number of combination other functions, for example, system is missed Difference function uses (y*(k)-y(k))2Or (y*(k+1)-y(k))2, that is, using e2(k) another functional form；It illustrates again For, system error function usesWherein, Δ uj(k)=uj(k)-uj(k-1), bjTo be more than or Constant equal to 0,1≤j≤m；Obviously, b is worked asjWhen being equal to 0, system error function only accounts for e2(k) contribution, shows most The target of smallization is systematic error minimum, that is, pursues precision height；And work as bjDuring more than 0, system error function considers simultaneously e2(k) contribution andContribution, show minimize target pursue systematic error it is small while, also pursue control it is defeated Enter to change small, that is, not only pursued high pursue again of precision and manipulated surely.For the controlled device of above-mentioned specific embodiment, in use Different system error functions is stated, can realize good control effect；Only consider e with system error function2(k) when contributing Control effect compare, consider e simultaneously in system error function2(k) contribution andContribution when its control accuracy omit Having reduces and it manipulates stationarity and is then improved.
Finally should it is emphasized that the MISO full format Non-Model Controller treats setting parameter, comprising punishment because Sub- λ and step factor ρ1,…,ρLy+LuOne of arbitrary or any number of combination；In above-mentioned specific embodiment, battery of tests is tested Penalty factor λ and step factor ρ during card1,ρ2,ρ3,ρ4It realizes while Self-tuning System, penalty factor λ consolidates during second group of verification experimental verification Determine and step factor ρ1,ρ2,ρ3,ρ4Realize Self-tuning System；In practical application, it can also select to wait to adjust as the case may be Any number of combination of parameter, for example, step factor ρ1,ρ2It fixes and penalty factor λ, step factor ρ3,ρ4It realizes from whole It is fixed；In addition, MISO full format Non-Model Controllers treat setting parameter, include but not limited to penalty factor λ and step factor ρ1,…,ρLy+Lu, for example, as the case may be, the row matrix of MISO system puppet piecemeal gradient estimate can also be includedEtc. parameters.
Above-mentioned specific 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)

  1. Methods of self-tuning of the 1.MISO full format Non-Model Controller based on local derviation information, it is characterised in that including following step Suddenly：
    Step (1)：For with MISO (Multiple of the m input (m is the integer more than or equal to 2) with 1 output Input and Single Output, multiple input single output) system, it is controlled using MISO full format Non-Model Controllers； Control linearization the length constant Ly, Ly for determining MISO full format Non-Model Controllers are the integer more than or equal to 1；Really Control input linearisation the length constant Lu, Lu for determining MISO full format Non-Model Controllers are the integer more than or equal to 1；It is described MISO full format Non-Model Controllers parameter includes penalty factor λ and step factor ρ1,…,ρLy+Lu；Determine MISO full format without Model controller treats setting parameter, and the MISO full format Non-Model Controller treats setting parameter, be the MISO full format without Model controller parameter it is part or all of, include penalty factor λ and step factor ρ1,…,ρLy+LuIt is one of arbitrary or arbitrary Kind combination；Determine the input layer number, node in hidden layer, output layer number of nodes of BP neural network, the output node layer Number is no less than the MISO full format Non-Model Controller and treats setting parameter number；Initialize the hidden layer of the BP neural network Weight coefficient, output layer weight coefficient；Local derviation information in initialization set { local derviation information collection }；
    Step (2)：The k moment will be denoted as current time；
    Step (3)：Actual value is exported with system based on system output desired value, calculating function using systematic error is calculated k The systematic error at moment, is denoted as e (k)；
    Step (4)：Using the local derviation information in the set { local derviation information collection } as the input of BP neural network, the BP nerves Network carries out forward calculation, and result of calculation is exported by the output layer of the BP neural network, obtain the MISO full format without Model controller treats the value of setting parameter；
    Step (5)：The MISO full format that the systematic error e (k) that is obtained based on step (3), step (4) are obtained is without mould Type controller treats the value of setting parameter, and using the control algolithm of MISO full format Non-Model Controllers, the full lattice of MISO are calculated Formula Non-Model Controller is directed to control input vector u (k)=[u of the controlled device at the k moment1(k),…,um(k)]T；
    Step (6)：J-th of control input u in the control input vector u (k) obtained for step (5)j(k)(1≤j≤ M), j-th of control input u is calculatedj(k) setting parameter is treated for each MISO full format Non-Model Controller respectively In the gradient information at k moment, specific formula for calculation is as follows：
    When the MISO full format Non-Model Controller when in setting parameter include penalty factor λ and Lu=1 when, described j-th Control input uj(k) it is for gradient informations of the penalty factor λ at the k moment：
    When the MISO full format Non-Model Controller is treated to include penalty factor λ and Lu in setting parameter>When 1, j-th of control System input uj(k) it is for gradient informations of the penalty factor λ at the k moment：
    When the MISO full format Non-Model Controller is treated to include step factor ρ in setting parameteriAnd during 1≤i≤Ly, the jth A control input uj(k) it is directed to the step factor ρiIt is in the gradient information at k moment：
    When the MISO full format Non-Model Controller is treated to include step factor ρ in setting parameterLy+1When, j-th of the control is defeated Enter uj(k) it is directed to the step factor ρLy+1It is in the gradient information at k moment：
    When the MISO full format Non-Model Controller is treated to include step factor ρ in setting parameteriAnd Ly+2≤i≤Ly+Lu and Lu >When 1, j-th of control input uj(k) it is directed to the step factor ρiIt is in the gradient information at k moment：
    Wherein, Δ uj(k)=uj(k)-uj(k-1), Δ y (k)=y (k)-y (k-1), y (k) is that the system output at k moment is actual Value,For the row matrix of the MISO system puppet piecemeal gradient estimate at k moment,For row matrixI-th piece of row Matrix (i=1 ..., Ly+Lu),For row matrixJ-th of gradient component estimate,For row matrix2 norms；
    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 of previous moment, i.e.,：As the MISO Full format Non-Model Controller when in setting parameter include penalty factor λ when then it is described { gradient information j } set in gradient letter BreathIt is denoted as the local derviation information of previous momentWhen the MISO full format Non-Model Controller is waited to adjust ginseng Step factor ρ is included in numberiAnd the gradient information in { gradient information j } then described during 1≤i≤Ly+Lu setIt is denoted as The local derviation information of previous moment
    The set of the above-mentioned whole local derviation information is denoted as { local derviation information j }, is put into the set { local derviation information collection }；
    Other m-1 control input in the control input vector u (k) obtained for step (5), repeat this step Suddenly, until the set { gradient information collection } includes the set of all { { gradient information 1 } ..., { gradient information m } }, while institute The set that set { local derviation information collection } includes all { { local derviation information 1 } ..., { local derviation information m } } is stated, subsequently into step (7)；
    Step (7)：Target is minimised as with the value of system error function, using gradient descent method, the institute obtained with reference to step (6) Set { gradient information collection } is stated, carries out systematic error backpropagation calculating, updates hidden layer weight coefficient, the output of BP neural network Layer weight coefficient, as later moment in time BP neural network carry out forward calculation when hidden layer weight coefficient, output layer weight coefficient；
    Step (8)：After the control input vector u (k) acts on controlled device, system of the controlled device in later moment in time is obtained Actual value is exported, back to step (2), step (2) is repeated and arrives step (8).
  2. 2. methods of self-tuning of the MISO full format Non-Model Controller based on local derviation information according to claim 1, It is characterized in that, the systematic error in the step (3), which calculates argument of function, includes system output desired value with being System output actual value.
  3. 3. parameter self-tuning side of the MISO full format Non-Model Controller based on local derviation information according to claim 1 or 2 Method, which is characterized in that the systematic error in the step (3) calculates function and uses e (k)=y*(k)-y (k), wherein y* (k) desired value is exported for the system that the k moment sets, y (k) is that the system that k instance samples obtain exports actual value；Or using e (k)=y*(k+1)-y (k), wherein y*(k+1) system for the k+1 moment exports desired value, and y (k) is for what k instance samples obtained System output actual value.
  4. 4. methods of self-tuning of the MISO full format Non-Model Controller based on local derviation information according to claim 1, It is characterized in that, the independent variable of the system error function in the step (7) includes systematic error, system output it is expected One of the arbitrary or any number of combination of value, system output actual value.
  5. 5. parameter self-tuning side of the MISO full format Non-Model Controllers based on local derviation information according to claim 1 or 4 Method, which is characterized in that the system error function in the step (7) isWherein, e (k) is Systematic error, Δ uj(k)=uj(k)-uj(k-1), bjTo be greater than or equal to 0 constant, 1≤j≤m.
CN201711323329.1A 2017-12-12 2017-12-12 Parameter self-tuning method of MISO full-format model-free controller based on partial derivative information Active CN108107715B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201711323329.1A CN108107715B (en) 2017-12-12 2017-12-12 Parameter self-tuning method of MISO full-format model-free controller based on partial derivative information

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201711323329.1A CN108107715B (en) 2017-12-12 2017-12-12 Parameter self-tuning method of MISO full-format model-free controller based on partial derivative information

Publications (2)

Publication Number Publication Date
CN108107715A true CN108107715A (en) 2018-06-01
CN108107715B CN108107715B (en) 2020-06-09

Family

ID=62215623

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201711323329.1A Active CN108107715B (en) 2017-12-12 2017-12-12 Parameter self-tuning method of MISO full-format model-free controller based on partial derivative information

Country Status (1)

Country Link
CN (1) CN108107715B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111522233A (en) * 2019-02-01 2020-08-11 浙江大学 Parameter self-tuning MIMO different-factor full-format model-free control method

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1274435A (en) * 1997-10-06 2000-11-22 美国通控集团公司 Model-free adaptive process control
CN101349893A (en) * 2007-07-18 2009-01-21 太极光控制软件（北京）有限公司 Forecast control device of adaptive model
CN101968629A (en) * 2010-10-19 2011-02-09 天津理工大学 PID (Proportional Integral Derivative) control method for elastic integral BP neural network based on RBF (Radial Basis Function) identification
CN107023825A (en) * 2016-08-31 2017-08-08 西安艾贝尔科技发展有限公司 Fluidized-bed combustion boiler is controlled and combustion optimizing system

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1274435A (en) * 1997-10-06 2000-11-22 美国通控集团公司 Model-free adaptive process control
CN101349893A (en) * 2007-07-18 2009-01-21 太极光控制软件（北京）有限公司 Forecast control device of adaptive model
CN101968629A (en) * 2010-10-19 2011-02-09 天津理工大学 PID (Proportional Integral Derivative) control method for elastic integral BP neural network based on RBF (Radial Basis Function) identification
CN107023825A (en) * 2016-08-31 2017-08-08 西安艾贝尔科技发展有限公司 Fluidized-bed combustion boiler is controlled and combustion optimizing system

Non-Patent Citations (6)

* Cited by examiner, † Cited by third party
Title
CHEN CHEN 等: "Design of Self-Tuning SISO Partial-Form Model-Free Adaptive Controller for Vapor-Compression Refrigeration System", 《IEEE ACCESS》 *
CHEN CHEN 等: "Parameter Self-Tuning of SISO Compact-Form Model-Free Adaptive Controller based on Neural Network with System Error Set as Input", 《2019 12TH ASIAN CONTROL CONFERENCE》 *
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》 *
李雪园: "无模型控制器参数在线自整定研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 *
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 等: "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 (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111522233A (en) * 2019-02-01 2020-08-11 浙江大学 Parameter self-tuning MIMO different-factor full-format model-free control method
CN111522233B (en) * 2019-02-01 2024-02-20 浙江大学 Parameter self-tuning MIMO different factor full-format model-free control method

Also Published As

Publication number Publication date
CN108107715B (en) 2020-06-09

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
CN111045326A (en) Tobacco shred drying process moisture prediction control method and system based on recurrent neural network
CN109634108A (en) The different factor full format non-model control method of the MIMO of parameter self-tuning
CN108345213A (en) Tight methods of self-tuning of the format Non-Model Controller based on systematic error of MIMO
CN109787251B (en) Cluster temperature control load aggregation model, system parameter identification and reverse control method
CN108153151A (en) Methods of self-tuning of the MIMO full format Non-Model Controller based on systematic error
CN109581864A (en) The inclined format non-model control method of the different factor of the MIMO of parameter self-tuning
CN108181809A (en) Tight methods of self-tuning of the form Non-Model Controller based on systematic error of MISO
CN107942655A (en) Tight methods of self-tuning of the form Non-Model Controller based on systematic error of SISO
CN108154231A (en) Methods of self-tuning of the MISO full format Non-Model Controller based on systematic error
CN108062021A (en) Methods of self-tuning of the SISO full format Non-Model Controller based on local derviation information
CN105469142A (en) Neural network increment-type feedforward algorithm based on sample increment driving
CN108107715A (en) Methods of self-tuning of the MISO full format Non-Model Controller based on local derviation information
CN108132600A (en) Tight methods of self-tuning of the form Non-Model Controller based on local derviation information of MIMO
CN108287470A (en) Inclined methods of self-tuning of the format Non-Model Controller based on local derviation information of MIMO
CN107942654A (en) Inclined methods of self-tuning of the form Non-Model Controller based on local derviation information of SISO
CN108073072A (en) Tight methods of self-tuning of the form Non-Model Controller based on local derviation information of SISO
CN108052006A (en) Decoupling control methods of the MIMO based on SISO full format Non-Model Controller Yu local derviation information
CN108008634A (en) Inclined methods of self-tuning of the form Non-Model Controller based on local derviation information of MISO
CN108181808A (en) Inclined methods of self-tuning of the form Non-Model Controller based on systematic error of MISO
CN103279030A (en) Bayesian framework-based dynamic soft measurement modeling method and device
CN111539508B (en) Generator excitation system parameter identification algorithm based on improved gray wolf algorithm
CN117052457A (en) Tunnel construction dust treatment control method, system, equipment and medium
CN107168066A (en) A kind of greenhouse self-adaptation control method

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