CN111930010A - LSTM network-based general MFA controller design method - Google Patents

LSTM network-based general MFA controller design method Download PDF

Info

Publication number
CN111930010A
CN111930010A CN202010600820.XA CN202010600820A CN111930010A CN 111930010 A CN111930010 A CN 111930010A CN 202010600820 A CN202010600820 A CN 202010600820A CN 111930010 A CN111930010 A CN 111930010A
Authority
CN
China
Prior art keywords
controller
lstm
mfa
output
input
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202010600820.XA
Other languages
Chinese (zh)
Inventor
孙京诰
陈显锋
张海峰
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.)
East China University of Science and Technology
Original Assignee
East China University of Science and Technology
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 East China University of Science and Technology filed Critical East China University of Science and Technology
Priority to CN202010600820.XA priority Critical patent/CN111930010A/en
Publication of CN111930010A publication Critical patent/CN111930010A/en
Pending legal-status Critical Current

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
    • G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
    • G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Automation & Control Theory (AREA)
  • Feedback Control In General (AREA)

Abstract

The invention discloses a method for designing a general Model Free Adaptive (MFA) controller based on an LSTM network. And (3) constructing a self-adaptive network output value by considering the influence of the historical error sequence information on the output of the controller, and finishing updating the weight of the network by a back propagation algorithm over time through inputting and outputting data. And the error value at the current time may be added to the output of the controller to react faster to sudden changes. The LSTM-MFA controller can effectively control a single-variable industrial process, and has the advantages of simple structural design, small calculated amount, no dependence on an accurate mathematical model and parameters of a controlled system, good performance effect of the controller and the like.

Description

LSTM network-based general MFA controller design method
Technical Field
The present invention relates to an adaptive control method based on an LSTM network, and more particularly, to a general MFA controller design method.
Background
In recent years, the increasing complexity of industrial processes has made the demand for intelligence on automation controllers, which need to be adaptive to different processes in order to cope with unforeseen events that affect the economy and safety of the process. However, in actual process production, PID control still plays a dominant role.
While PID controllers can provide satisfactory control performance for many single-input single-output systems with relatively small dynamic ranges, significant difficulties remain with complex control systems. Many advanced control theories and methods have been developed to deal with these complex systems, such as model predictive control, robust control, and adaptive control. But these advanced techniques all rely on an accurate and relatively simple dynamic model of the process. These accurate models are often difficult to acquire in practice, and even if a model is obtained, model uncertainty can severely impact control performance.
Accordingly, there is a need to provide a universal controller that can easily and efficiently control a variety of complex systems, which can accommodate different process configurations. The MFA control method does not need to depend on the knowledge of the system dynamics with accurate process, only needs the real-time input and output data of the controller and the qualitative knowledge of the system behavior, and depends on the strong self-learning and self-adaptive capacity of the controller to deal with the uncertainty and the change in the system environment to complete the MFA control of the process.
Disclosure of Invention
The invention mainly aims to provide a method for designing a universal MFA controller based on an LSTM network, which is applied to a process control system.
The technical scheme for realizing the purpose of the invention is as follows: an LSTM-MFA universal control system for process control, comprising the steps of:
step 1: generating a set value signal of a desired value or a desired trajectory output by the process by a set value generating device;
step 2: generating an error signal between the set value signal and the measured variable signal as an input signal to the controller by the comparator means;
and step 3: the LSTM-MFA controller generates a control signal according to an input error signal and acts on the actual process;
and 4, step 4: a measured variable of the actual process is obtained. In each sampling period, the LSTM-MFA controller constructs a group of input and output data according to historical error sequence information and real-time output of the controller, and updates the weight of an LSTM recurrent neural network in the controller by using a time back propagation algorithm;
and 5: and (5) continuing to execute the steps 1-4 until the process reaches a set value and keeps stable, and finishing the iteration.
Further, the specific implementation process of the control system is as follows:
first, consider the form of the industrial process transfer function as follows:
Figure 991369DEST_PATH_IMAGE001
wherein
Figure 752651DEST_PATH_IMAGE002
Is a Laplace transform operator;
Figure 324928DEST_PATH_IMAGE003
a Laplace transform which is a process transfer function;
Figure 488056DEST_PATH_IMAGE004
for process output or measured variables
Figure 139618DEST_PATH_IMAGE005
The Ralsberg transform of (1);
Figure 653776DEST_PATH_IMAGE006
as process inputs or controller outputs
Figure 48985DEST_PATH_IMAGE007
The Ralsberg transform of (1).
Determining a sampling time
Figure 66619DEST_PATH_IMAGE008
Usually one tenth of the process-dominant time constant is taken. In practical applications, an approximation may be obtained by artificially estimating the process-dominant time constant, since the LSTM-MFA controller is not sensitive to this parameter.
For process control systems, the control objective is to make the measured variable
Figure 154661DEST_PATH_IMAGE005
Tracking set point
Figure 139803DEST_PATH_IMAGE009
Is given as a track signal.
That is, the task of the LSTM-MFA controller is to make the error signal
Figure 73124DEST_PATH_IMAGE010
At a minimum, the objective function is formulated as follows:
Figure 210845DEST_PATH_IMAGE011
。
then, the general LSTM-MFA controller of the invention is designed by the following steps:
step 1: current time of day
Figure 938629DEST_PATH_IMAGE012
Error signal of
Figure 427379DEST_PATH_IMAGE013
By a normalizing unit
Figure 898812DEST_PATH_IMAGE014
Generating a normalized error signal
Figure 422197DEST_PATH_IMAGE015
As input to the controller at the current time;
wherein, the normalization unit is a tanh function, and the expression is as follows:
Figure 304572DEST_PATH_IMAGE016
。
step 2: the normalized error signal is input into two LSTM hidden layers, and the network output at the current moment is obtained through the feedforward calculation process of the LSTM
Figure 546197DEST_PATH_IMAGE017
。
Wherein, the feedforward calculation process of the single LSTM is as follows:
Figure 555741DEST_PATH_IMAGE018
Figure 402475DEST_PATH_IMAGE019
Figure 3220DEST_PATH_IMAGE020
Figure 935404DEST_PATH_IMAGE021
Figure 279798DEST_PATH_IMAGE022
Figure 495884DEST_PATH_IMAGE023
Figure 470794DEST_PATH_IMAGE024
wherein,
Figure 155853DEST_PATH_IMAGE025
representing the multiplication of corresponding elements in the operation matrix;
Figure 38358DEST_PATH_IMAGE026
then representing the matrix addition operation;
Figure 859684DEST_PATH_IMAGE027
for memory-losing gates, for deciding the state of the last cell
Figure 257692DEST_PATH_IMAGE028
What information is thrown away;
Figure 164468DEST_PATH_IMAGE029
for a candidate vector, this value is added to the cell state;
Figure 850664DEST_PATH_IMAGE030
is an input gate for updating the cell state
Figure 526496DEST_PATH_IMAGE029
；
Figure 639945DEST_PATH_IMAGE031
Determining which portions of the cell state are output for the output gate; a total of 8 sets of weight vectors to be adjusted are contained in each LSTM
Figure 34018DEST_PATH_IMAGE032
Unlike a normal neural network, all LSTM units in a single recursive hidden layer share this set of weights and do not increase the computational effort.
The activation function adopted by the gate control unit is a sigmoid function, and the expression is as follows:
Figure 258326DEST_PATH_IMAGE033
。
and step 3: at the current moment
Figure 303511DEST_PATH_IMAGE012
The output of the LSTM-MFA controller is output by the network
Figure 587862DEST_PATH_IMAGE017
And error signal
Figure 203651DEST_PATH_IMAGE013
The formula is as follows:
Figure 700491DEST_PATH_IMAGE034
wherein,
Figure 350915DEST_PATH_IMAGE035
the performance of the controller can be fine-tuned for the controller gain, usually taken
Figure 524277DEST_PATH_IMAGE036
。
And 4, step 4: finally, at each sampling period, the weights of the network need to be iteratively updated using a back-propagation over time algorithm based on a set of input and output data.
Wherein, the input data is the error of the current sampling period
Figure 892941DEST_PATH_IMAGE013
The output data is historical error sequence information and current sampling period network output
Figure 927893DEST_PATH_IMAGE017
Constructed function value
Figure 432824DEST_PATH_IMAGE037
The expression is as follows:
Figure 777086DEST_PATH_IMAGE038
。
wherein,
Figure 898626DEST_PATH_IMAGE039
the adaptive rate for the output adjustment of the controller can effectively adjust the effect of the controller, and the value range is 0-1.
Figure 206111DEST_PATH_IMAGE040
Is the length of the historical error sequence information under consideration.
Figure 831127DEST_PATH_IMAGE039
Too large a value of (c) will result in too much oscillation of the closed loop system response;
Figure 628182DEST_PATH_IMAGE039
too small a value will result in process drift or slow response speed.
The invention applies the general MFA controller design method based on the LSTM network to the process control system. Compared with the prior control technology, the invention has the advantages that:
(1) the LSTM network-based general MFA controller design method can be applied to linear or nonlinear systems in process control, and the controller structure does not need to be redesigned.
(2) The general MFA controller design method based on the LSTM network does not need the accurate mathematical model of the process and the quantitative knowledge of the parameters, and solves the difficulty that the process system is difficult to model.
(3) The general MFA controller design method based on the LSTM network can deal with the situation that the model structure changes, namely the model parameter or the structure changes in the actual process, the parameter of the controller does not need to be adjusted, the performance of the controller can not be influenced, the prior PID control and other technologies need to adjust the corresponding parameter, and the parameter adjusting process is very complicated.
(4) The design method of the general MFA controller based on the LSTM network is relatively simple in structure and can be easily applied to the actual process.
Drawings
FIG. 1 is a single input single output LSTM-MFA universal control system of the present invention.
Fig. 2 is a schematic structural diagram of the LSTM-MFA controller of the present invention.
FIG. 3 is a schematic diagram of the LSTM structure of the present invention.
FIG. 4 is a diagram of an LSTM-MFA controller simulation for a setpoint change process of the present invention.
FIG. 5 is a graph of an LSTM-MFA controller simulation of the present invention for the presence of white noise in the measured variables.
FIG. 6 is a comparative simulation diagram of the LSTM-MFA and PID controllers for the texture transformation process of the present invention.
FIG. 7 is a comparative simulation diagram of the LSTM-MFA and PID controllers of the present invention for a hysteresis process.
Detailed Description
The invention will be further explained with reference to the drawings.
The invention relates to a method for designing a general MFA controller based on an LSTM network, which is applied to a process control system and is implemented specifically as follows:
considering the simplest application structure of the invention, the LSTM-MFA general control system with single input and single output as shown in FIG. 1 is composed of a single input and single output process, an LSTM-MFA controller, two signal summers and a back propagation module with time.
The signals in fig. 1 are defined as follows:
Figure 963916DEST_PATH_IMAGE009
-a set value, representing a set value signal of a target value or a process trajectory of the process.
Figure 340671DEST_PATH_IMAGE041
-a process output, the actual output of the controlled process without disturbance.
Figure 820193DEST_PATH_IMAGE007
An output quantity, representing the control signal and the process input from the controller.
Figure 256991DEST_PATH_IMAGE042
Interference, representing interference caused by process noise or load variations.
Figure 71232DEST_PATH_IMAGE005
-a measured variable representing the actually measured process output, wherein
Figure 782836DEST_PATH_IMAGE043
。
Figure 382445DEST_PATH_IMAGE010
-an error, representing an error signal, wherein
Figure 521302DEST_PATH_IMAGE044
。
The control objective being to make the measured variable
Figure 307993DEST_PATH_IMAGE005
Tracking set point
Figure 823287DEST_PATH_IMAGE009
Given track signals, i.e. error signals
Figure 277403DEST_PATH_IMAGE010
Tending towards zero.
Step 1: the setpoint signal of the desired value or desired trajectory of the process output is generated by the setpoint generating device.
Step 2: an error signal between the set-point signal and the measured variable signal is generated by the comparator means as an input signal for the controller.
And step 3: the LSTM-MFA controller generates a control signal according to an input error signal and acts on an actual process.
And 4, step 4: a measured variable of the actual process is obtained. In each sampling period, the LSTM-MFA controller constructs a group of input and output data according to historical error sequence information and real-time output of the controller, and updates the weight of the LSTM recurrent neural network in the controller by using a time-reversal propagation algorithm.
And 5: and (5) continuing to execute the steps 1-4 until the process reaches a set value and keeps stable, and finishing the iteration.
The structure of the general LSTM-MFA controller is shown in FIG. 2, and the specific implementation steps are as follows:
it can be seen that the controller is structured in the form of an LSTM recurrent neural network, which can be expanded along the sampling time axis.
Wherein
Figure 305270DEST_PATH_IMAGE045
Inputting a signal for the error at each sampling instant;
Figure 110415DEST_PATH_IMAGE046
outputting a signal for the network at each sampling instant;
Figure 632663DEST_PATH_IMAGE047
for the controller output at each sampling instant.
Sampling time
Figure 472444DEST_PATH_IMAGE008
Is generally constant with the dominant time of the process
Figure 421945DEST_PATH_IMAGE048
Correlation is usually taken to be one tenth of the dominant time constant.
Wherein in the actual process
Figure 979965DEST_PATH_IMAGE048
Is usually unknown and can be estimated manually
Figure 40325DEST_PATH_IMAGE048
To an approximation that the LSTM-MFA controller is not very sensitive to this parameter and therefore does not significantly affect the performance of the controller.
Step 1: current time of day
Figure 718300DEST_PATH_IMAGE012
Error signal of
Figure 369861DEST_PATH_IMAGE013
By a normalizing unit
Figure 618440DEST_PATH_IMAGE014
Generating a normalized error signal
Figure 13649DEST_PATH_IMAGE015
As the output of the controller at the current momentAnd (6) adding.
Wherein, the normalization unit is a tanh function, and the expression is as follows:
Figure 296863DEST_PATH_IMAGE016
。
step 2: the normalized error signal is input into two LSTM hidden layers, and the network output at the current moment is obtained through the feedforward calculation process of the LSTM
Figure 119326DEST_PATH_IMAGE017
Wherein the complete structure of the LSTM is shown in FIG. 3.
And step 3: at the current moment
Figure 855200DEST_PATH_IMAGE012
Output of LSTM-MFA controller
Figure 775139DEST_PATH_IMAGE049
Output from the network
Figure 444018DEST_PATH_IMAGE050
And error signal
Figure 906223DEST_PATH_IMAGE013
The formula is as follows:
Figure 660553DEST_PATH_IMAGE034
。
wherein,
Figure 131985DEST_PATH_IMAGE035
the performance of the controller can be fine-tuned for the controller gain, usually taken
Figure 655371DEST_PATH_IMAGE036
。
And 4, step 4: finally, at each sampling period, the weights of the network need to be iteratively updated using a back-propagation over time algorithm based on a set of input and output data.
Wherein, the input data is the error of the current sampling period
Figure 288477DEST_PATH_IMAGE013
The output data is historical error sequence information and current sampling period network output
Figure 530103DEST_PATH_IMAGE017
Constructed function value
Figure 788915DEST_PATH_IMAGE037
The expression is as follows:
Figure 635648DEST_PATH_IMAGE051
。
wherein,
Figure 236394DEST_PATH_IMAGE052
the adaptive rate for the output adjustment of the controller can effectively adjust the effect of the controller, and the value range is 0-1.
Figure 168577DEST_PATH_IMAGE053
Is the length of the historical error sequence information under consideration.
Adjusting controller adaptation rate
Figure 512971DEST_PATH_IMAGE039
And (4) realizing model-free adaptive control of the process.
The effect of using the specific implementation of the invention and the proposed control method is illustrated by simulations as follows:
consider several single-input single-output processes as follows:
model 1:
Figure 214211DEST_PATH_IMAGE054
。
model 2:
Figure 985858DEST_PATH_IMAGE055
。
model 3:
Figure 654605DEST_PATH_IMAGE056
。
for the case where the model set value is changed, model 1 is taken as an example. In the simulation process, the set values of the process are constantly changed.
In the simulation process, the parameters controlled by the LSTM-MFA are default values, and the self-adaptive rate
Figure 5952DEST_PATH_IMAGE057
。
All controller tuning parameters remain unchanged despite the changing process set points. The simulation results are shown in fig. 4.
It can be seen that the LSTM-MFA controller can adapt well to the set value changing process, and has good control effect.
For the case of model interference, model 2 is taken as an example. In the simulation process, white noise interference is added to the process output.
In the simulation process, the parameters controlled by the LSTM-MFA are default values, and the self-adaptive rate
Figure 827278DEST_PATH_IMAGE058
。
All controller tuning parameters remain unchanged despite the presence of white noise interference, and the simulation results are shown in fig. 5.
It can be seen that the LSTM-MFA controller can adapt well to the white noise interference process, with good control effect.
For the control of the model switching process, in the simulation process, the process model is changed from model 1 to model 2 on-line to create the model switching process, and the set value is changed from 10 to 20.
In the simulation process, the parameters controlled by the LSTM-MFA are default values, and the self-adaptive rate
Figure 238668DEST_PATH_IMAGE058
. And the PID parameters adjusted for model 1 are
Figure 394711DEST_PATH_IMAGE059
. The sampling time is taken as
Figure 549749DEST_PATH_IMAGE060
。
All controller tuning parameters remain unchanged despite process variations. The simulation result is shown in fig. 6.
It can be seen that the LSTM-MFA controller can adapt well to changes in process configuration, whereas the PID controller cannot.
For the process with model lag, in the simulation process, the process model is switched from model 1 to model 3 lag, and the set value is changed from 10 to 20.
The controller parameter settings are as described above.
Although the process switches to the model with hysteresis, all controller tuning parameters remain unchanged. The simulation results are shown in fig. 7.
It can be seen that in the presence of hysteresis, the LSTM-MFA controller is still well adapted and has better performance than the PID controller.

Claims (9)

1. An LSTM-MFA universal control system for process control, comprising the steps of:
step 1: generating a set value signal of a desired value or a desired trajectory output by the process by a set value generating device;
step 2: generating an error signal between the set value signal and the measured variable signal as an input signal to the controller by the comparator means;
and step 3: the LSTM-MFA controller generates a control signal according to an input error signal and acts on the actual process;
and 4, step 4: acquiring a measurement variable of an actual process, constructing a group of input and output data by the LSTM-MFA controller according to historical error sequence information and real-time output of the controller in each sampling period, and updating the weight of an LSTM recurrent neural network in the controller by using a time-based back propagation algorithm;
and 5: and (5) continuing to execute the steps 1-4 until the process reaches a set value and keeps stable, and finishing the iteration.
2. The LSTM-MFA universal control system according to claim 1, wherein said universal LSTM-MFA controller comprises:
1) an input layer: taking an error signal of a controlled process as the input of a controller, and then performing normalization processing to be used as the input of an LSTM network;
2) hiding the layer: the first layer of LSTM input not only includes the normalization error signal of the current time, but also includes the hidden state information transmitted by the first layer of LSTM at the previous time; the input of the second layer LSTM comprises the output of the last layer LSTM and the hidden state transmitted by the second layer LSTM at the last moment;
3) an output layer: at the current sampling instant, the output of the controller consists of the network output and the current error signal.
3. The LSTM-MFA universal controller of claim 2, wherein the controller is configured to include a recurrent neural network, such that the controller can be spread out along the sampling time, and at each sampling instant, the controller outputs correspond to the network outputs one-to-one.
4. The LSTM-MFA generic controller of claim 2, wherein the normalization of the input-layer error signal is by transforming the error signal into the interval-1 to 1 by a normalization function, wherein the normalization error function is a tanh function.
5. The LSTM-MFA controller of claim 2, where the sampling time is chosen in relation to the dominant time constant of the process, typically taken to be one tenth of the dominant time constant, and in practice, to be approximated, since the LSTM-MFA controller is not very sensitive to this parameter.
6. The LSTM-MFA generic controller of claim 2, wherein the controller employs a network architecture that is an LSTM recurrent neural network that models a length of historical error sequence information to derive a continuous function of the network output as
Figure 48501DEST_PATH_IMAGE001
To obtain the output of the controller
Figure 754289DEST_PATH_IMAGE002
The continuous function expression of (a) is:
Figure 964821DEST_PATH_IMAGE003
wherein
Figure 589617DEST_PATH_IMAGE004
Is a continuous function of the error signal and,
Figure 481481DEST_PATH_IMAGE005
is the controller gain.
7. The generalized LSTM-MFA controller of claim 6, wherein the controller gain is
Figure 41775DEST_PATH_IMAGE005
Which is related to the open loop gain of the process for selectively attenuating to compensate for the large steady state gain of the process, fine tuning may improve controller performance, typically by taking
Figure 626471DEST_PATH_IMAGE006
。
8. The LSTM-MFA universal controller of claim 2, wherein at each sampling instant, LSTM-MFThe learning process of the controller A comprises constructing proper input and output data as training data for updating network weight, and completing the adaptive learning process of the controller by using a time-dependent back propagation algorithm, wherein the input data is the current moment
Figure 639427DEST_PATH_IMAGE007
Error value of
Figure 397298DEST_PATH_IMAGE008
The output data is historical error sequence information and current sampling period network output
Figure 559902DEST_PATH_IMAGE009
The function value of the structure can be expressed by the following formula:
Figure 564767DEST_PATH_IMAGE010
wherein
Figure 815751DEST_PATH_IMAGE011
For the length of the historical sequence information to be considered,
Figure 236368DEST_PATH_IMAGE012
is a normalized error signal for the historical error value,
Figure 771254DEST_PATH_IMAGE013
is an adaptive parameter.
9. The LSTM-MFA universal controller as claimed in claim 8, wherein said adaptive parameters are adapted to control the operation of said MFA controller
Figure 697753DEST_PATH_IMAGE013
The performance of the controller can be effectively adjusted, the value range of the controller is 0-1, and the larger the value is, the more historical error sequence information components are considered.
CN202010600820.XA 2020-06-29 2020-06-29 LSTM network-based general MFA controller design method Pending CN111930010A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010600820.XA CN111930010A (en) 2020-06-29 2020-06-29 LSTM network-based general MFA controller design method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010600820.XA CN111930010A (en) 2020-06-29 2020-06-29 LSTM network-based general MFA controller design method

Publications (1)

Publication Number Publication Date
CN111930010A true CN111930010A (en) 2020-11-13

Family

ID=73316738

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010600820.XA Pending CN111930010A (en) 2020-06-29 2020-06-29 LSTM network-based general MFA controller design method

Country Status (1)

Country Link
CN (1) CN111930010A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112379601A (en) * 2020-12-01 2021-02-19 华东理工大学 MFA control system design method based on industrial process
CN112631215A (en) * 2020-12-10 2021-04-09 东北大学 Intelligent forecasting method, device, equipment and storage medium for industrial process operation index
CN113255208A (en) * 2021-04-21 2021-08-13 杭州新剑机器人技术股份有限公司 Neural network model prediction control method for series elastic actuator of robot

Citations (5)

* 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
US20050004687A1 (en) * 2003-05-30 2005-01-06 Cheng George Shu-Xing Apparatus and method of controlling multi-input-single-output systems
US20120259437A1 (en) * 2011-04-08 2012-10-11 General Cybernation Group Inc. Model-free adaptive control of advanced power plants
CN103064292A (en) * 2013-01-15 2013-04-24 镇江市江大科技有限责任公司 Biological fermentation adaptive control system and control method based on neural network inverse
CN108153146A (en) * 2017-12-12 2018-06-12 东北大学 A kind of polynary molten steel quality MFA control system and method for blast furnace

Patent Citations (5)

* 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
US20050004687A1 (en) * 2003-05-30 2005-01-06 Cheng George Shu-Xing Apparatus and method of controlling multi-input-single-output systems
US20120259437A1 (en) * 2011-04-08 2012-10-11 General Cybernation Group Inc. Model-free adaptive control of advanced power plants
CN103064292A (en) * 2013-01-15 2013-04-24 镇江市江大科技有限责任公司 Biological fermentation adaptive control system and control method based on neural network inverse
CN108153146A (en) * 2017-12-12 2018-06-12 东北大学 A kind of polynary molten steel quality MFA control system and method for blast furnace

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
蒋昊: "工业电除尘过程智能建模及优化算法研究", 《中国优秀博硕士学位论文全文数据库(硕士)工程科技Ⅰ辑》 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112379601A (en) * 2020-12-01 2021-02-19 华东理工大学 MFA control system design method based on industrial process
CN112631215A (en) * 2020-12-10 2021-04-09 东北大学 Intelligent forecasting method, device, equipment and storage medium for industrial process operation index
CN113255208A (en) * 2021-04-21 2021-08-13 杭州新剑机器人技术股份有限公司 Neural network model prediction control method for series elastic actuator of robot

Similar Documents

Publication Publication Date Title
Perrusquía et al. Identification and optimal control of nonlinear systems using recurrent neural networks and reinforcement learning: An overview
Zribi et al. A new PID neural network controller design for nonlinear processes
EP1021752B1 (en) Model-free adaptive process control
Chen et al. Applying neural networks to on-line updated PID controllers for nonlinear process control
Zhang et al. Adaptive neural tracking control of pure-feedback nonlinear systems with unknown gain signs and unmodeled dynamics
Zhang et al. An adaptive generalized predictive control method for nonlinear systems based on ANFIS and multiple models
CN111930010A (en) LSTM network-based general MFA controller design method
Sabzalian et al. Robust fuzzy control for fractional-order systems with estimated fraction-order
Calvo-Rolle et al. Adaptive inverse control using an online learning algorithm for neural networks
Radac et al. Three-level hierarchical model-free learning approach to trajectory tracking control
Qi et al. Stable indirect adaptive control based on discrete-time T–S fuzzy model
Jiang et al. Robust adaptive dynamic programming
Beyhan et al. Stable modeling based control methods using a new RBF network
Khettab et al. Enhanced fractional order indirect fuzzy adaptive synchronization of uncertain fractional chaotic systems based on the variable structure control: robust H∞ design approach
Hsu et al. Emotional fuzzy sliding-mode control for unknown nonlinear systems
Kumar et al. Lyapunov stability-based control and identification of nonlinear dynamical systems using adaptive dynamic programming
Lightbody et al. Neural modelling of chemical plant using MLP and B-spline networks
CN111240201B (en) Disturbance suppression control method
Su et al. Adaptive fuzzy control of MIMO nonlinear systems with fuzzy dead zones
CN112379601A (en) MFA control system design method based on industrial process
Zha et al. Identification of the Hammerstein nonlinear system with noisy output measurements
Sheikhmemari Recurrent Neural Network Controller for Linear and Nonlinear Systems
Wache et al. Sensitivity-Based Iterative State-Feedback Tuning for Nonlinear Systems
Sakaki et al. Design of a data-driven two-degree-of-freedom control system considering robustness
Kusumoputro et al. An improved single neuron adaptive PID controller system based on additional error of an inversed-control signal

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
AD01 Patent right deemed abandoned
AD01 Patent right deemed abandoned

Effective date of abandoning: 20230228