CN108107715B - Parameter self-tuning method of MISO full-format model-free controller based on partial derivative information - Google Patents

Parameter self-tuning method of MISO full-format model-free controller based on partial derivative information Download PDF

Info

Publication number
CN108107715B
CN108107715B CN201711323329.1A CN201711323329A CN108107715B CN 108107715 B CN108107715 B CN 108107715B CN 201711323329 A CN201711323329 A CN 201711323329A CN 108107715 B CN108107715 B CN 108107715B
Authority
CN
China
Prior art keywords
miso
full
partial derivative
format model
parameters
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.)
Active
Application number
CN201711323329.1A
Other languages
Chinese (zh)
Other versions
CN108107715A (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

Images

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 parameter self-tuning method of a MISO full-format model-free controller based on partial derivative information, which utilizes a partial derivative information set as the input of a BP neural network, the BP neural network carries out forward calculation and outputs parameters to be tuned of the MISO full-format model-free controller such as penalty factors, step length factors and the like through an output layer, a control algorithm of the MISO full-format model-free controller is adopted to calculate to obtain a control input vector aiming at a controlled object, the value minimization of a system error function is taken as a target, a gradient descent method is adopted, and the control input is combined to respectively aim at a gradient information set of each parameter to be tuned, the system error back propagation calculation is carried out, the weight coefficient of an implicit layer and the weight coefficient of an output layer of the BP neural network are updated in real time on line, and the parameter self-tuning of the controller based on the partial derivative information. The parameter self-tuning method of the MISO full-format model-free controller based on the partial derivative information can effectively overcome the difficulty of on-line tuning of the controller parameters and has good control effect on the MISO system.

Description

Parameter self-tuning method of MISO full-format model-free controller based on partial derivative information
Technical Field
The invention belongs to the field of automatic control, and particularly relates to a parameter self-tuning method of a MISO full-format model-free controller based on partial derivative information.
Background
The control problem of the MISO (Multiple Input and Single Output) system has been one of the major challenges faced in the field of automation control.
Existing implementations of MISO controllers include MISO full-format modeless controllers. The MISO full-format model-free controller is a novel data-driven control method, does not depend on any mathematical model information of a controlled object, only depends on input and output data measured by the MISO controlled object in real time to analyze and design the controller, is simple and clear in realization, small in calculation burden and strong in robustness, can well control an unknown nonlinear time-varying MISO system, and has a good application prospect. The theoretical basis of the MISO full-format model-free controller is proposed by Houzhong and Jinshangtai in the 'model-free adaptive control-theory and application' (scientific publishing agency, 2013, page 118) of the Hei-Gong, and the control algorithm is as follows:
Figure RE-GDA0001610391460000011
where u (k) is a control input vector at time k, and u (k) is [ u (k) ]1(k),…,um(k)]TM is the number of control inputs, Δ u (k) ═ u (k) — u (k-1); e (k) is the system error at time k; Δ y (k) -y (k-1), and y (k) is the system output actual value at time k;
Figure RE-GDA0001610391460000012
a row matrix of MISO system pseudo-block gradient estimates at time k,
Figure RE-GDA0001610391460000013
is a row matrix
Figure RE-GDA0001610391460000014
The ith block row matrix of (i ═ 1, …, Ly + Lu),
Figure RE-GDA0001610391460000015
is a row matrix
Figure RE-GDA0001610391460000016
Figure RE-GDA0001610391460000016
2 norm of (d); λ is a penalty factor, ρ1,…,ρLy+LuFor the step size factor, Ly is the control output linearization length constant, and Lu is the control input linearization length constant.
However, the MISO full-format modeless controller needs to rely on empirical knowledge to set the penalty factor λ in advance before it is actually put into serviceAnd step size factor ρ1,…,ρLy+LuThe values of the isoparametric parameters have not realized a penalty factor lambda and a step factor rho in the actual application process1,…,ρLy+LuAnd (4) performing online self-tuning on the equal parameters. The lack of effective parameter setting means not only makes the use and debugging process of the MISO full-format model-free controller time-consuming and labor-consuming, but also can seriously affect the control effect of the MISO full-format model-free controller sometimes, and restricts the popularization and application of the MISO full-format model-free controller. That is to say: the MISO full-format model-free controller also needs to solve the problem of online self-tuning parameters in the actual commissioning process.
Therefore, in order to break the bottleneck of restricting the popularization and application of the MISO full-format model-free controller, the invention provides a parameter self-tuning method of the MISO full-format model-free controller based on partial derivative information.
Disclosure of Invention
In order to solve the problems in the background art, the invention aims to provide a parameter self-tuning method of a MISO full-format model-free controller based on partial derivative information.
To this end, the above object of the present invention is achieved by the following technical solution, comprising the steps of:
step (1): for a MISO (multiple input and Single Output) system with m inputs (m is an integer greater than or equal to 2) and 1 Output, adopting a MISO full-format model-free controller for control; determining a control output linearization length constant Ly of the MISO full-format model-free controller, wherein the Ly is an integer greater than or equal to 1; determining a control input linearization length constant Lu of the MISO full-format model-free controller, wherein Lu is an integer greater than or equal to 1; the MISO full-format modeless controller parameters include a penalty factor λ and a step-size factor ρ1,…,ρLy+Lu(ii) a Determining parameters to be set of the MISO full-format model-free controller, wherein the parameters to be set of the MISO full-format model-free controller are part or all of the parameters of the MISO full-format model-free controller and comprise a penalty factor lambda and a step factor rho1,…,ρLy+LuAny one or any combination of the above; determining BP neural networksThe number of input layer nodes, the number of hidden layer nodes and the number of output layer nodes are not less than the number of parameters to be set of the MISO full-format model-free controller; initializing a hidden layer weight coefficient and an output layer weight coefficient of the BP neural network; initializing partial derivative information in a set { partial derivative information set };
step (2): recording the current time as k time;
and (3): calculating to obtain a system error at the k moment by adopting a system error calculation function based on the system output expected value and the system output actual value, and recording as e (k);
and (4): taking the partial derivative information in the set { partial derivative information set } as the input of a BP (back propagation) neural network, carrying out forward calculation by the BP neural network, and outputting a calculation result through an output layer of the BP neural network to obtain a value of a parameter to be set of the MISO full-format model-free controller;
and (5): calculating and obtaining a control input vector u (k) [ u (k) ] of the MISO full-format modeless controller at the time k for the controlled object by adopting a control algorithm of the MISO full-format modeless controller based on the system error e (k) obtained in the step (3) and the value of the parameter to be set of the MISO full-format modeless controller obtained in the step (4)1(k),…,um(k)]T；
And (6): aiming at the jth control input u in the control input vector u (k) obtained in the step (5)j(k) (j is more than or equal to 1 and less than or equal to m), calculating the jth control input uj(k) Respectively aiming at the gradient information of the parameters to be set of each MISO full-format model-free controller at the moment k, the specific calculation formula is as follows:
when the parameters to be set of the MISO full-format model-free controller comprise penalty factor lambda and Lu is 1, the jth control input uj(k) The gradient information at the k moment for the penalty factor λ is:
Figure RE-GDA0001610391460000031
when the MISO full-format model-free controller is in the middle of the parameter to be setContaining penalty factor lambda and Lu>1, said jth control input uj(k) The gradient information at the k moment for the penalty factor λ is:
Figure RE-GDA0001610391460000032
when the parameters to be set of the MISO full-format model-free controller contain step factor rhoiAnd when i is more than or equal to 1 and less than or equal to Ly, the jth control input uj(k) For the step size factor piThe gradient information at time k is:
Figure RE-GDA0001610391460000041
when the parameters to be set of the MISO full-format model-free controller contain step factor rhoLy+1Then, the jth control input uj(k) For the step size factor pLy+1The gradient information at time k is:
Figure RE-GDA0001610391460000042
when the parameters to be set of the MISO full-format model-free controller contain step factor rhoiAnd i is more than or equal to Ly +2 and less than or equal to Ly + Lu and Lu>1, said jth control input uj(k) For the step size factor piThe gradient information at time k is:
Figure RE-GDA0001610391460000043
wherein, Δ uj(k)＝uj(k)-uj(k-1), Δ y (k) -y (k-1), and y (k) is the system output actual value at time k,
Figure RE-GDA0001610391460000044
a row matrix of MISO system pseudo-block gradient estimates at time k,
Figure RE-GDA0001610391460000045
is a row matrix
Figure RE-GDA0001610391460000046
The ith block row matrix of (i ═ 1, …, Ly + Lu),
Figure RE-GDA0001610391460000047
is a row matrix
Figure RE-GDA0001610391460000048
The j-th gradient component estimate of (a),
Figure RE-GDA0001610391460000049
is a row matrix
Figure RE-GDA00016103914600000410
Figure RE-GDA00016103914600000410
2 norm of (d);
the set of all the gradient information is marked as { gradient information j }, and a set { gradient information set } is put in;
recording the gradient information in the { gradient information j } set as partial derivative information of the previous moment in sequence, namely: when the parameters to be set of the MISO full-format model-free controller contain penalty factor lambda, the gradient information in the { gradient information j } set
Figure RE-GDA00016103914600000411
Recording as partial derivative information of previous time
Figure RE-GDA00016103914600000412
When the parameters to be set of the MISO full-format model-free controller contain step factor rhoiAnd when i is more than or equal to 1 and less than or equal to Ly + Lu, the gradient information in the set { gradient information j }
Figure RE-GDA00016103914600000413
Recording as partial derivative information of previous time
Figure RE-GDA00016103914600000414
The set of all the partial derivative information is marked as { partial derivative information j }, and the set { partial derivative information set } is put into;
repeating the step for the other m-1 control inputs in the control input vector u (k) obtained in step (5) until the set { gradient information set } contains the set of all { { gradient information 1}, …, { gradient information m } }, and the set { partial derivative information set } contains the set of all { { partial derivative information 1}, …, { partial derivative information m } }, and then proceeding to step (7);
and (7): the value minimization of a system error function is taken as a target, a gradient descent method is adopted, the set { gradient information set } obtained in the step (6) is combined, the backward propagation calculation of the system error is carried out, and the weight coefficient of the hidden layer and the weight coefficient of the output layer of the BP neural network are updated and used as the weight coefficient of the hidden layer and the weight coefficient of the output layer when the BP neural network carries out forward calculation at the later moment;
and (8): and (4) after the control input vector u (k) acts on the controlled object, obtaining a system output actual value of the controlled object at the later moment, returning to the step (2), and repeating the step (2) to the step (8).
While adopting the above technical scheme, the present invention can also adopt or combine the following further technical schemes:
the independent variables of the system error calculation function in the step (3) comprise a system output expected value and a system output actual value.
The systematic error calculation function in the step (3) adopts e (k) y*(k) -y (k), wherein y*(k) The system output expected value is set for the time k, and y (k) is the system output actual value obtained by sampling at the time k; or using e (k) ═ y*(k +1) -y (k), wherein y*And (k +1) is a system output expected value at the moment of k +1, and y (k) is a system output actual value obtained by sampling at the moment of k.
The independent variable of the system error function in the step (7) comprises any one or any combination of a system error, a system output expected value and a system output actual value.
Said systematic error function in said step (7) is
Figure RE-GDA0001610391460000051
Wherein e (k) is the systematic error, Δ uj(k)＝uj(k)-uj(k-1)，bjIs a constant greater than or equal to 0, and j is greater than or equal to 1 and less than or equal to m.
The MISO full-format model-free controller parameter self-tuning method provided by the invention can realize good control effect and effectively overcome penalty factor lambda and step factor rho1,…,ρLThe difficult problem of setting needs time and labor waste.
Drawings
FIG. 1 is a functional block diagram of the present invention;
FIG. 2 is a schematic diagram of a BP neural network structure employed in the present invention;
FIG. 3 shows a two-input single-output MISO system with penalty factor λ and stride factor ρ1,ρ2,ρ3,ρ4Meanwhile, self-setting a timing control effect graph;
FIG. 4 is a diagram of a two-input single-output MISO system with penalty factor λ and stride factor ρ1,ρ2,ρ3,ρ4Simultaneously self-timing control input diagram;
FIG. 5 shows a two-input single-output MISO system with penalty factor λ and stride factor ρ1,ρ2,ρ3,ρ4Meanwhile, self-adjusting a punishment factor lambda change curve;
FIG. 6 is a diagram of a two-input single-output MISO system with penalty factor λ and stride factor ρ1,ρ2,ρ3,ρ4Step size factor p while self-aligning1,ρ2,ρ3,ρ4A change curve;
FIG. 7 is a diagram of a two-input single-output MISO system with a penalty factor λ fixed and a step-size factor ρ1,ρ2,ρ3,ρ4A self-timing control effect graph;
FIG. 8 is a diagram of a two-input single-output MISO system with a penalty factor λ fixed and a step-size factor ρ1,ρ2,ρ3,ρ4A self-timed control input map;
FIG. 9 shows a two-input single-output MISO system with a penalty factor λ fixed and a step-size factor ρ1,ρ2,ρ3,ρ4Step factor p at self-alignment1,ρ2,ρ3,ρ4A curve of variation.
Detailed Description
The invention is further described with reference to the following figures and specific examples.
Fig. 1 shows a schematic block diagram of the present invention. For a MISO system with m inputs (m is an integer greater than or equal to 2) and 1 output, adopting a MISO full-format model-free controller for control; determining a control output linearization length constant Ly of the MISO full-format model-free controller, wherein the Ly is an integer greater than or equal to 1; determining a control input linearization length constant Lu of the MISO full-format model-free controller, wherein Lu is an integer greater than or equal to 1; the MISO full-format model-free controller parameters comprise a penalty factor lambda and a step factor rho1,…,ρLy+Lu(ii) a Determining parameters to be set of the MISO full-format model-free controller, wherein the parameters are part or all of the parameters of the MISO full-format model-free controller and comprise a penalty factor lambda and a step factor rho1,…,ρLy+LuAny one or any combination of the above; in FIG. 1, the parameters to be set by the MISO full-format model-free controller are penalty factor lambda and step factor rho1,…,ρLy+Lu(ii) a Determining the number of input layer nodes, the number of hidden layer nodes and the number of output layer nodes of the BP neural network, wherein the number of the output layer nodes is not less than the number of parameters to be set of the MISO full-format model-free controller; initializing a hidden layer weight coefficient and an output layer weight coefficient of the BP neural network; the partial derivatives in the set { partial derivatives set } are initialized.
Recording the current time as k time; outputting the system to a desired value y*(k) Taking the difference with the system output actual value y (k) as the system error e (k) at the time k; taking partial derivative information in the set { partial derivative information set } as input of a BP neural network, carrying out forward calculation on the BP neural network, and outputting a calculation result through an output layer of the BP neural network to obtain a value of a parameter to be set of the MISO full-format model-free controller; based on the systematic error e(k) And calculating the value of the parameter to be set of the MISO full-format model-free controller by adopting a control algorithm of the MISO full-format model-free controller to obtain a control input vector u (k) [ u ]) of the MISO full-format model-free controller at the time k for the controlled object1(k),…,um(k)]T(ii) a For the jth control input u in the control input vector u (k)j(k) (j is more than or equal to 1 and less than or equal to m), calculating the jth control input uj(k) Respectively aiming at gradient information of parameters to be set of each MISO full-format model-free controller at the moment k, recording a set of all the gradient information as { gradient information j }, and putting the set { gradient information set }; sequentially recording the gradient information in the { gradient information j } set as partial derivative information of a previous moment, recording the set of all the partial derivative information as { partial derivative information j }, and putting the set { partial derivative information set }; repeating the execution for the other m-1 control inputs in the control input vector u (k) until the set { gradient information set } contains the set of all { { gradient information 1}, …, { gradient information m } }, while the set { partial derivative information set } contains the set of all { { partial derivative information 1}, …, { partial derivative information m } }; subsequently, the set { gradient information set } is combined, targeted at the minimization of the value of the systematic error function, denoted e in fig. 12(k) Minimizing as a target, performing system error back propagation calculation by adopting a gradient descent method, updating a hidden layer weight coefficient and an output layer weight coefficient of the BP neural network, and taking the updated hidden layer weight coefficient and the output layer weight coefficient as the hidden layer weight coefficient and the output layer weight coefficient when the BP neural network performs forward calculation at the later moment; and after the control input vector u (k) acts on the controlled object, obtaining a system output actual value of the controlled object at the later moment, then repeatedly executing the work in the paragraph, and carrying out the parameter self-tuning process of the MISO full-format model-free controller at the later moment based on the partial derivative information.
Fig. 2 shows a schematic structural diagram of the BP neural network adopted in the present invention. The BP neural network may have a structure in which the hidden layer is a single layer, or may have a structure in which the hidden layer is a plurality of layers. In the schematic diagram of fig. 2, for simplicity, the BP neural network adopts a structure in which the hidden layer is a single layer, that is, a three-layer network structure consisting of an input layer, a single hidden layer and an output layer is adoptedThe number of nodes of the input layer is set to be m × the number of parameters to be set (the number of parameters to be set is Ly + Lu +1 in fig. 2), the number of nodes of the hidden layer is 10, and the number of nodes of the output layer is set to be the number of parameters to be set (the number of parameters to be set is Ly + Lu +1 in fig. 2). The number of nodes of the input layer is divided into m groups, the number of nodes of each group is the number of parameters to be set, and the number of nodes of the jth group and the partial derivative information in the { partial derivative information j } set
Figure RE-GDA0001610391460000081
Respectively correspond to each other. Node of output layer, penalty factor lambda and step factor rho1,…,ρLy+LuRespectively correspond to each other. The update process of the hidden layer weight coefficient and the output layer weight coefficient of the BP neural network specifically comprises the following steps: targeting the minimization of the value of the systematic error function, denoted by e in FIG. 22(k) And (4) minimizing to a target, and performing system error back propagation calculation by adopting a gradient descent method and combining the set { gradient information set }, so as to update the weight coefficient of the hidden layer and the weight coefficient of the output layer of the BP neural network.
The following is a specific embodiment of the present invention.
The controlled object is a typical nonlinear two-input single-output MISO system:
Figure RE-GDA0001610391460000082
desired value y of system output*(k) The following were used:
y*(k)＝(-1)round((k-1)/100)
in this particular embodiment, m is 2.
The value of the control output linearization length constant Ly of the MISO full-format modeless controller is usually set according to the complexity of the controlled object and the actual control effect, and is generally between 1 and 5, and an excessively large value causes a large calculation amount, so that 1 or 3 is generally adopted, and Ly is taken as 1 in the specific embodiment; the value of the control input linearization length constant Lu of the MISO full-format modeless controller is also usually set according to the complexity of the controlled object and the actual control effect, and is generally between 1 and 10, and too small value will affect the control effect, and too large value will result in large calculation amount, so 3 or 5 is usually adopted, and Lu is taken as 3 in the present embodiment.
The BP neural network adopts a three-layer network structure consisting of an input layer, a single-layer hidden layer and an output layer, the number of nodes of the input layer is set to be 2 multiplied by the number of parameters to be set, the number of nodes of the hidden layer is set to be 10, and the number of nodes of the output layer is set to be the number of the parameters to be set.
For the above specific examples, two sets of experimental verification were performed.
During the first group of test verification, the number of input layer nodes of the BP neural network in FIG. 2 is preset to 10, the number of output layer nodes is preset to 5, and a penalty factor lambda and a step factor rho are calculated1,ρ2,ρ3,ρ4Performing simultaneous self-tuning, wherein FIG. 3 is a control effect diagram, FIG. 4 is a control input diagram, FIG. 5 is a penalty factor lambda variation curve, and FIG. 6 is a step factor rho1,ρ2,ρ3,ρ4A curve of variation. The result shows that the method of the invention carries out the punishment factor lambda and the step factor rho1,ρ2,ρ3,ρ4The method has the advantages of realizing good control effect by carrying out self-tuning at the same time, and effectively overcoming the penalty factor lambda and the step factor rho1,ρ2,ρ3,ρ4The difficult problem of setting needs time and labor waste.
During the second group of test verification, the number of nodes of the input layer and the number of nodes of the output layer of the BP neural network in the graph 2 are preset to 8 and 4 respectively, firstly, the penalty factor lambda is fixedly set as the average value of the penalty factor lambda during the first group of test verification, and then, the step factor rho is subjected to1,ρ2,ρ3,ρ4Performing self-tuning, wherein FIG. 7 is a control effect graph, FIG. 8 is a control input graph, and FIG. 9 is a step factor rho1,ρ2,ρ3,ρ4A curve of variation. The results also show that the method of the invention is implemented by applying the step factor rho when the penalty factor lambda is fixed1,ρ2,ρ3,ρ4Self-tuning is carried out, good control effect can be realized, and the step factor rho can be effectively overcome1,ρ2,ρ3,ρ4The difficult problem of setting needs time and labor waste.
It should be noted that in the above-described embodiment, the system is output with the desired value y*(k) The difference with the actual system output value y (k) is used as the system error e (k), i.e. e (k) y*(k) -y (k), only one method of calculating a function for the systematic error; the expected value y of the system output at the moment k +1 can also be used*The difference between (k +1) and the actual system output value y (k) at time k is taken as the system error e (k), i.e. e (k) y*(k +1) -y (k); the system error calculation function may also employ other calculation methods where the independent variables include a desired system output value and an actual system output value, for example,
Figure RE-GDA0001610391460000101
Figure RE-GDA0001610391460000102
for the controlled object of the above embodiment, good control effects can be achieved by using the different system error calculation functions.
More particularly, in the above embodiment, when the hidden layer weight coefficient and the output layer weight coefficient of the BP neural network are updated with the goal of minimizing the value of the systematic error function, the systematic error function adopts e2(k) Only one of said systematic error functions; the system error function may also be other functions with independent variables including any one or any combination of system error, system output expected value and system output actual value, for example, the system error function may be (y)*(k)-y(k))2Or (y)*(k+1)-y(k))2I.e. using e2(k) Another functional form of (1); as another example, a systematic error function is employed
Figure RE-GDA0001610391460000103
Wherein, Δ uj(k)＝uj(k)-uj(k-1)，bjIs a constant greater than or equal to 0, j is greater than or equal to 1 and less than or equal to m; obviously, when bjWhen the values are all equal to 0, the first electrode,the systematic error function only takes into account e2(k) The contribution of (1) shows that the aim of minimization is to minimize the system error, namely pursuing high precision; when b isjWhen the error is larger than 0, the system error function considers e2(k) Are made a contribution to
Figure RE-GDA0001610391460000104
The contribution of (1) indicates that the goal of minimization is to pursue small system errors and small control input variation, namely to pursue both high precision and stable steering. For the controlled object of the above embodiment, good control effect can be achieved by adopting the different system error functions; considering only e with the systematic error function2(k) Control effects in contribution to the system error function while considering e2(k) Are made a contribution to
Figure RE-GDA0001610391460000105
The contribution of (1) is that the control precision is slightly reduced and the operation stability is improved.
Finally, it should be noted that the parameters to be set of the MISO full-format model-free controller include a penalty factor λ and a step factor ρ1,…,ρLy+LuAny one or any combination of the above; in the above specific embodiment, the first set of trial validations is performed with a penalty factor λ and a step-size factor ρ1,ρ2,ρ3,ρ4Realizes the simultaneous self-tuning, the punishment factor lambda is fixed and the step factor rho is adopted during the verification of the second group of tests1,ρ2,ρ3,ρ4Self-tuning is realized; in practical application, any combination of parameters to be set can be selected according to specific conditions, for example, the step factor ρ1,ρ2Fixed penalty factor lambda, step factor rho3,ρ4Self-tuning is realized; in addition, the MISO full-format modeless controller has parameters to be set including, but not limited to, penalty factor λ and step-size factor ρ1,…,ρLy+LuFor example, a row matrix of MISO system pseudo block gradient estimation values may also be included, as the case may be
Figure RE-GDA0001610391460000111
And the like.
The above-described embodiments are intended to illustrate the present invention, but not to limit the present invention, and any modifications, equivalents, improvements, etc. made within the spirit of the present invention and the scope of the claims fall within the scope of the present invention.

Claims (3)

  1. The parameter self-tuning method of the MISO full-format model-free controller based on the partial derivative information is characterized by comprising the following steps of:
    step (1): for a MISO (multiple input and Single Output) system with m inputs (m is an integer greater than or equal to 2) and 1 Output, adopting a MISO full-format model-free controller for control; determining a control output linearization length constant Ly of the MISO full-format model-free controller, wherein the Ly is an integer greater than or equal to 1; determining a control input linearization length constant Lu of the MISO full-format model-free controller, wherein Lu is an integer greater than or equal to 1; the MISO full-format modeless controller parameters include a penalty factor λ and a step-size factor ρ1,…,ρLy+Lu(ii) a Determining parameters to be set of the MISO full-format model-free controller, wherein the parameters to be set of the MISO full-format model-free controller are part or all of the parameters of the MISO full-format model-free controller and comprise a penalty factor lambda and a step factor rho1,…,ρLy+LuAny one or any combination of the above; determining the number of input layer nodes, the number of hidden layer nodes and the number of output layer nodes of the BP neural network, wherein the number of the output layer nodes is not less than the number of parameters to be set of the MISO full-format model-free controller; initializing a hidden layer weight coefficient and an output layer weight coefficient of the BP neural network; initializing partial derivative information in a set { partial derivative information set };
    step (2): recording the current time as k time;
    and (3): calculating to obtain a system error at the k moment by adopting a system error calculation function based on the system output expected value and the system output actual value, and recording as e (k); the independent variables of the system error calculation function comprise a system output expected value and a system output actual value;
    and (4): taking the partial derivative information in the set { partial derivative information set } as the input of a BP (back propagation) neural network, carrying out forward calculation by the BP neural network, and outputting a calculation result through an output layer of the BP neural network to obtain a value of a parameter to be set of the MISO full-format model-free controller;
    and (5): calculating and obtaining a control input vector u (k) [ u (k) ] of the MISO full-format modeless controller at the time k for the controlled object by adopting a control algorithm of the MISO full-format modeless controller based on the system error e (k) obtained in the step (3) and the value of the parameter to be set of the MISO full-format modeless controller obtained in the step (4)1(k),…,um(k)]T；
    And (6): aiming at the jth control input u in the control input vector u (k) obtained in the step (5)j(k) (j is more than or equal to 1 and less than or equal to m), calculating the jth control input uj(k) Respectively aiming at the gradient information of the parameters to be set of each MISO full-format model-free controller at the moment k, the specific calculation formula is as follows:
    when the parameters to be set of the MISO full-format model-free controller comprise penalty factor lambda and Lu is 1, the jth control input uj(k) The gradient information at the k moment for the penalty factor λ is:
    Figure FDA0002208651830000021
    when the parameters to be set of the MISO full-format model-free controller contain penalty factors of lambda and Lu>1, said jth control input uj(k) The gradient information at the k moment for the penalty factor λ is:
    Figure FDA0002208651830000022
    when the parameters to be set of the MISO full-format model-free controller contain step factor rhoiAnd i is more than or equal to 1 and less than or equal to Ly, the jth control input uj(k) For the step size factor piThe gradient information at time k is:
    Figure FDA0002208651830000023
    when the parameters to be set of the MISO full-format model-free controller contain step factor rhoLy+1Then, the jth control input uj(k) For the step size factor pLy+1The gradient information at time k is:
    Figure FDA0002208651830000024
    when the parameters to be set of the MISO full-format model-free controller contain step factor rhoiAnd i is more than or equal to Ly +2 and less than or equal to Ly + Lu and Lu>1, said jth control input uj(k) For the step size factor piThe gradient information at time k is:
    Figure FDA0002208651830000031
    wherein, Δ uj(k)＝uj(k)-uj(k-1), Δ y (k) -y (k-1), and y (k) is the system output actual value at time k,
    Figure FDA0002208651830000032
    a row matrix of MISO system pseudo-block gradient estimates at time k,
    Figure FDA0002208651830000033
    is a row matrix
    Figure FDA0002208651830000034
    The ith block row matrix of (i ═ 1, …, Ly + Lu),
    Figure FDA0002208651830000035
    is a row matrix
    Figure FDA0002208651830000036
    The j-th gradient component estimate of (a),
    Figure FDA0002208651830000037
    is a row matrix
    Figure FDA0002208651830000038
    2 norm of (d);
    the set of all the gradient information is marked as { gradient information j }, and a set { gradient information set } is put in;
    recording the gradient information in the { gradient information j } set as partial derivative information of the previous moment in sequence, namely: when the parameters to be set of the MISO full-format model-free controller contain penalty factor lambda, the gradient information in the { gradient information j } set
    Figure FDA0002208651830000039
    Recording as partial derivative information of previous time
    Figure FDA00022086518300000310
    When the parameters to be set of the MISO full-format model-free controller contain step factor rhoiAnd when i is more than or equal to 1 and less than or equal to Ly + Lu, the gradient information in the set { gradient information j }
    Figure FDA00022086518300000311
    Recording as partial derivative information of previous time
    Figure FDA00022086518300000312
    The set of all the partial derivative information is marked as { partial derivative information j }, and the set { partial derivative information set } is put into;
    repeating the step for the other m-1 control inputs in the control input vector u (k) obtained in step (5) until the set { gradient information set } contains the set of all { { gradient information 1}, …, { gradient information m } }, and the set { partial derivative information set } contains the set of all { { partial derivative information 1}, …, { partial derivative information m } }, and then proceeding to step (7);
    and (7): the value minimization of a system error function is taken as a target, a gradient descent method is adopted, the set { gradient information set } obtained in the step (6) is combined, the backward propagation calculation of the system error is carried out, and the weight coefficient of the hidden layer and the weight coefficient of the output layer of the BP neural network are updated and used as the weight coefficient of the hidden layer and the weight coefficient of the output layer when the BP neural network carries out forward calculation at the later moment; the independent variable of the system error function comprises any one or any combination of a system error, a system output expected value and a system output actual value;
    and (8): and (4) after the control input vector u (k) acts on the controlled object, obtaining a system output actual value of the controlled object at the later moment, returning to the step (2), and repeating the step (2) to the step (8).
  2. 2. The MISO full-format model-less controller parameter self-tuning method of claim 1, wherein the systematic error calculation function in step (3) employs e (k) -y*(k) -y (k), wherein y*(k) The system output expected value is set for the time k, and y (k) is the system output actual value obtained by sampling at the time k; or using e (k) ═ y*(k +1) -y (k), wherein y*And (k +1) is a system output expected value at the moment of k +1, and y (k) is a system output actual value obtained by sampling at the moment of k.
  3. 3. The MISO full format modeless controller of claim 1, wherein the systematic error function in step (7) is partial derivative information based parameter auto-tuning method
    Figure FDA0002208651830000041
    Wherein e (k) is the systematic error, Δ uj(k)＝uj(k)-uj(k-1)，bjIs a constant greater than or equal to 0, and j is greater than or equal to 1 and less than or equal to 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 CN108107715A (en) 2018-06-01
CN108107715B true 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)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109634108A (en) * 2019-02-01 2019-04-16 浙江大学 The different factor full format non-model control method of the MIMO of parameter self-tuning

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
Design of Self-Tuning SISO Partial-Form Model-Free Adaptive Controller for Vapor-Compression Refrigeration System;CHEN CHEN 等;《IEEE Access》;20190903;全文 *
Neural-net-based model-free self-tuning controller with on-line self-learning ability for industrial furnace;Mingwang ZHAO;《1994 Proceedings of IEEE International Conference on Control and Applications》;20020806;全文 *
Parameter Self-Tuning of SISO Compact-Form Model-Free Adaptive Controller based on Neural Network with System Error Set as Input;Chen CHEN 等;《2019 12th Asian Control Conference》;20190718;全文 *
无模型控制器参数在线自整定研究;李雪园;《中国优秀硕士学位论文全文数据库信息科技辑》;20180815;全文 *
无模型控制器参数学习步长和惩罚因子的整定研究;马平 等;《仪器仪表学报》;20080430;全文 *
无模型自适应控制参数整定方法研究;郭代银;《中国优秀硕士学位论文全文数据库信息科技辑》;20150315;全文 *

Also Published As

Publication number Publication date
CN108107715A (en) 2018-06-01

Similar Documents

Publication Publication Date Title
CN108287471B (en) Parameter self-tuning method of MIMO offset format model-free controller based on system error
CN108170029B (en) Parameter self-tuning method of MIMO full-format model-free controller based on partial derivative information
CN109472418B (en) Maneuvering target state prediction optimization method based on Kalman filtering
CN108345213B (en) Parameter self-tuning method of MIMO (multiple input multiple output) compact-format model-free controller based on system error
JP6092477B2 (en) An automated method for correcting neural dynamics
CN108181809B (en) System error-based parameter self-tuning method for MISO (multiple input single output) compact-format model-free controller
CN112101530A (en) Neural network training method, device, equipment and storage medium
CN104462015B (en) Process the fractional order linear discrete system state updating method of non-gaussian L é vy noises
CN108153151B (en) Parameter self-tuning method of MIMO full-format model-free controller based on system error
CN108154231B (en) System error-based parameter self-tuning method for MISO full-format model-free controller
CN108287470B (en) Parameter self-tuning method of MIMO offset format model-free controller based on offset information
CN108062021B (en) Parameter self-tuning method of SISO full-format model-free controller based on partial derivative information
CN107942654B (en) Parameter self-tuning method of SISO offset format model-free controller based on offset information
CN105469142A (en) Neural network increment-type feedforward algorithm based on sample increment driving
CN108107715B (en) Parameter self-tuning method of MISO full-format model-free controller based on partial derivative information
CN108132600B (en) Parameter self-tuning method of MIMO (multiple input multiple output) compact-format model-free controller based on partial derivative information
CN108008634B (en) Parameter self-tuning method of MISO partial-format model-free controller based on partial derivative information
CN108073072B (en) Parameter self-tuning method of SISO (Single input Single output) compact-format model-free controller based on partial derivative information
Yu et al. Fast online policy gradient learning with SMD gain vector adaptation
CN108181808B (en) System error-based parameter self-tuning method for MISO partial-format model-free controller
CN108052006B (en) Decoupling control method for MIMO based on SISO full-format model-free controller and partial derivative information
CN108107727B (en) Parameter self-tuning method of MISO (multiple input single output) compact-format model-free controller based on partial derivative information
JPH07200512A (en) 1optimization problems solving device
CN108107722B (en) Decoupling control method for MIMO based on SISO bias format model-less controller and system error
CN107844051B (en) Parameter self-tuning method of SISO full-format model-free controller based on system error

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