CN117331316A - Nonlinear process control method and system based on knowledge guided neural network - Google Patents
Nonlinear process control method and system based on knowledge guided neural network Download PDFInfo
- Publication number
- CN117331316A CN117331316A CN202311629901.2A CN202311629901A CN117331316A CN 117331316 A CN117331316 A CN 117331316A CN 202311629901 A CN202311629901 A CN 202311629901A CN 117331316 A CN117331316 A CN 117331316A
- Authority
- CN
- China
- Prior art keywords
- input
- nonlinear
- neural network
- knowledge
- representing
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 79
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 63
- 238000004886 process control Methods 0.000 title claims abstract description 21
- 238000004519 manufacturing process Methods 0.000 claims abstract description 26
- 238000012549 training Methods 0.000 claims abstract description 24
- 238000011217 control strategy Methods 0.000 claims abstract description 14
- 210000002364 input neuron Anatomy 0.000 claims abstract description 13
- 239000013598 vector Substances 0.000 claims description 31
- 230000006870 function Effects 0.000 claims description 21
- 238000005457 optimization Methods 0.000 claims description 12
- 238000010606 normalization Methods 0.000 claims description 9
- 238000010276 construction Methods 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 5
- 238000005070 sampling Methods 0.000 claims description 5
- 238000009499 grossing Methods 0.000 claims description 4
- 238000012897 Levenberg–Marquardt algorithm Methods 0.000 claims description 3
- 238000004422 calculation algorithm Methods 0.000 claims description 3
- 230000004927 fusion Effects 0.000 claims description 3
- 230000002194 synthesizing effect Effects 0.000 claims description 2
- 230000008569 process Effects 0.000 description 10
- 230000000694 effects Effects 0.000 description 4
- 238000007796 conventional method Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/02—Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]
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 nonlinear process control method and a nonlinear process control system based on a knowledge guided neural network, wherein the method comprises the following steps: extracting system structural knowledge of a nonlinear industrial process based on a sparse representation method: system input order, system output order and system time delay; constructing a feedforward neural network with a corresponding input neuron structure according to the extracted system structure knowledge; training the constructed feedforward neural network by utilizing the system input and output training data to obtain a system prediction model; and taking the trained system prediction model as a prediction model in a model prediction control strategy, obtaining a system input sequence by adopting the model prediction control strategy, and acting the first item of the sequence on the industrial process system. The invention avoids the problem of redundant input neurons of the neural network by extracting the structural knowledge of the system, and provides guarantee for the accurate control of the industrial process.
Description
Technical Field
The invention belongs to the technical field of industrial process control, and particularly relates to a nonlinear process control method and system based on a knowledge guided neural network.
Background
Modern industrial processes are always faced with high energy consumption, high material consumption, heavy pollution and other problems. Industrial process control technology can ensure stable, green, and efficient operation of a process by controlling the key state variables close to their set points. Modern industrial processes often have complex characteristics of multivariable, nonlinear, strong coupling, etc., while efficient and green production also implies various control constraints and target requirements. Generally, conventional model-free control methods such as PID cannot meet these industrial requirements at the same time, and model predictive control (model predictive control, MPC) methods are popular model-based control strategies consisting of model prediction, roll optimization, feedback correction, and the like. The method is widely applied in industry, and can obviously improve the product quality while reducing the resource consumption and the environmental pollution. As the complexity of industrial processes increases, most modern industrial systems are nonlinear systems. The linear prediction model inevitably has a problem of model mismatch, and it is difficult to achieve a satisfactory control effect. Accordingly, a great deal of research has been conducted on nonlinear model predictive control (nonlinear model predictive control, NMPC). The accuracy of the prediction result of the method determines the effect of model prediction control. Therefore, the core of nonlinear model predictive control is to build an accurate nonlinear predictive model.
The high-precision model is a basis for realizing nonlinear process control, and the existing modeling method is difficult to establish the high-precision model of the nonlinear process. The data-driven nonlinear modeling method avoids the problem that mechanism modeling requires comprehensive knowledge of the operation mechanism of an industrial system, and only relies on process data to learn the relation between system input and output, wherein a Neural Network (NN) has good robustness and self-learning capability, and is an effective method for describing a nonlinear system. The method can approach nonlinearity with high precision by learning input and output data of the system, and the neural network model also has good prediction effect when the system is polluted by environmental noise. Therefore, nonlinear predictive control based on neural networks has received a lot of attention in recent years, such as a back propagation neural network (back propagation neural network, BPNN), a radial basis function neural network (radial basis function neural network, RBFNN), and a T-S fuzzy neural network (T-S fuzzy neural network, T-S FNN).
The nonlinear predictive control based on the neural network at present mostly depends on strong assumptions that the structure of the controlled object system is known. However, for complex nonlinear systems, this assumption is difficult to meet in practical applications. As an alternative, it is often employed to set the order of redundancy to guarantee the performance of the predictive model, which inevitably increases the complexity of the model, especially for multiple-input multiple-output systems. As model complexity increases, the amount of data required to train the model also increases significantly, otherwise overfitting problems may occur. Acquisition of the data set is time consuming, which results in many practical industrial systems often not guaranteeing a sufficient training data set. In cases where the training data is insufficient, it is difficult to build an efficient data-driven predictive model.
Disclosure of Invention
The invention provides a nonlinear process control method and a nonlinear process control system based on a knowledge-guided neural network, which can avoid the problem of redundancy of input neurons of the neural network by extracting structural knowledge of the system, and provide guarantee for accurate control of industrial processes.
In order to achieve the technical purpose, the invention adopts the following technical scheme:
a knowledge guided neural network based nonlinear process control method, comprising:
extracting system structural knowledge of a nonlinear industrial process based on a sparse representation method: system input order, system output order and system time delay;
constructing a feedforward neural network with a corresponding input neuron structure according to the extracted system structure knowledge;
training the constructed feedforward neural network by utilizing the system input and output training data to obtain a system prediction model;
and taking the trained system prediction model as a prediction model in a model prediction control strategy, obtaining a system input sequence by adopting the model prediction control strategy, and acting the first item of the sequence on the industrial process system.
Further, extracting system structure knowledge specifically includes:
first, the input-output relationship of a nonlinear industrial process system is expressed as:
wherein,is->Time system input variable,/->Is->Time-of-day non-measurable intermediate variable,/->Is->Time system output variable,/->Representing an unknown nonlinear function->And->Weight factors representing system inputs and outputs, +.>Representing system delay->And->Representing the order of the system inputs and outputs, +.>Representing gaussian white noise;
then, the nonlinear part of the system, namely the unknown nonlinear function, is parameterized by the systemConverting into a linear combination of a plurality of basis functions; the base function coefficients and the constant coefficients in the linear combination form unknown nonlinear parameters;
by sampling nonlinear parameters and combining past input data of the system, a nonlinear function-based construction is performedIs input into the dictionary; constructing a system output dictionary by utilizing the past output data of the system;
finally, the input and output dictionaries of the system are integrated, and a dictionary learning method is adopted to obtain sparse vectors of future output data of the system; and comparing the sparse vector with an input-output relation of the system to obtain system structural knowledge: system time delaySystem input order->And system output order +.>。
Further, using Taylor series or Fourier series, the nonlinear function is calculatedConverted into a linear combination of basis functions.
Further, the construction is based on a nonlinear functionIs input into the dictionary->The method comprises the following steps:
wherein,is a positive integer greater than both the input and output orders of the system, i.e. +.>,For a certain moment in the past of the system, the system inputs the block dictionary +.>The method comprises the following steps:
wherein,for the number of training data, +.>Represents>To->Is input into the system; nonlinear parameter set->Is in the nonlinear parameter->M results obtained by fully sampling at equal intervals in space;
building a system output dictionaryThe method comprises the following steps:
wherein the system outputs a block dictionaryThe method comprises the following steps:
wherein,represents slave time +.>To->Is a system output data;
synthesizing dictionary input and output of system to obtain dictionary in dictionary learning optimization problem:
。
Further, the optimization problem constructed when the dictionary learning method is adopted is as follows:
wherein,for future output data of the system, i.e. +.>,/>Is->Warp dictionary->Sparse coding of->Representing sparse vector +.>Weight of non-zero element number, < ->The bigger the->The more sparse.
Further, when a dictionary learning method is adopted to obtain sparse vectors of future output data of the system, compensation operation is carried out on the sparse vectors, and then the sparse vectors after the compensation operation are compared with an input-output relational expression of the system to obtain system structure knowledge; wherein the compensation operation sequentially comprises:
(1) Zero operation
Dictionary-basedBy->And->Two parts are composed, and the sparse vector before compensation is correspondingly formed>Sparse vector divided into system past output data +.>Sparse vector of system past input data +.>Two parts, i.eThe method comprises the steps of carrying out a first treatment on the surface of the Then pair->And->Zero operation is performed respectively:
wherein,for vector->Or->Is>Element(s)>For vector->Or->The element with the largest absolute value, +.>For a preset threshold valueSparse vector obtained after zero operation +.>The method comprises the following steps:
;
(2) Inverse normalization operation
Dictionary-basedColumn normalization operation is performed, and sparse vector obtained after zero operation is +.>Performing inverse normalization to obtain a true sparse vector +.>:
Wherein,is composed of two norms of each dictionary atom,/->Division representing the corresponding position;
(3) Fusion operation
Selected ones of the system input block dictionariesThe atomic index is->The following data pairs were constructed:
wherein,representing a non-linear parameter->Corresponding to the system input block dictionary +.>Selected atoms and->Corresponding non-zero elements of (a); next, two matrices are defined:
calculating an estimate of the nonlinear parameter according to:
Wherein,representative vector->Is the first element of (c).
Further, according to the extracted system structure knowledge, a feedforward neural network with a corresponding input neuron structure is constructed, expressed as:
wherein the method comprises the steps of,Representing a feed-forward neural network;、andrespectively representing the extracted system time delay, the system input order and the system output order;representing feed forward neural network predictionsOutputting a system at the moment;
the network is trained by adopting a Levenberg-Marquardt algorithm, and the network training loss function is as follows:
wherein,representing the number of training data +.>Representing the actual output of the system,/->Representing the predicted output of the feedforward neural network.
Further, the optimization problem of obtaining the system input sequence by using the model predictive control strategy is expressed as follows:
wherein,,the weight is represented by a weight that,representing the prediction time domain,representing the control time domain of the device,andrepresenting system inputsIs defined by the upper and lower limits of (c),a reference trajectory of the system is represented,representing the closed-loop predicted output of the system,representation ofThe predicted value of the time instant prediction model,the representation of the predictive model is given by,as a result of the weighting factor(s),as a smoothing factor, the smoothing factor is used,,represents a set value;
solving the optimization problem by adopting an SQP algorithm to obtain a system input sequenceI.e. the system control sequence, and will be the first item +.>Acts on the system to realize the issuing of control input.
A system based on the knowledge-guided neural network-based nonlinear process control method as claimed in any one of the preceding claims, comprising:
the system structure knowledge extraction module is used for: extracting system structure knowledge of a nonlinear industrial process based on a sparse representation method, wherein the system structure knowledge comprises a system input order, a system output order and a system time delay;
the neural network construction module is used for: constructing a feedforward neural network with a corresponding input neuron structure according to the extracted system structure knowledge;
the prediction model training module is used for: training the constructed feedforward neural network by utilizing the system input and output training data to obtain a system prediction model;
model predictive control module for: and taking the trained system prediction model as a prediction model in a model prediction control strategy, obtaining a system input sequence by adopting the model prediction control strategy, and acting the first item of the sequence on the industrial process system.
Advantageous effects
According to the nonlinear process control method and system based on the knowledge-guided neural network, the system structure knowledge is fully excavated from a small amount of data by utilizing the sparsity of sparse representation learning, and the problem of data shortage in the data driving model training process is effectively solved under the guidance of the system structure knowledge. Firstly, establishing a sparse identification model of a nonlinear process, and accurately identifying the structure of the process; then, by integrating the structural knowledge into the neural network, a high-precision model of a nonlinear process is established, and the problem of redundancy of the input neurons of the neural network is avoided; finally, model predictive control is carried out on the industrial process based on the knowledge-guided neural network, so that high-precision control of the nonlinear industrial process is realized, the method can be applied to nonlinear control of the industrial process, and modeling and control precision is remarkably improved. In addition, the invention can reduce the data volume required by modeling and expand the application range of the method.
Drawings
FIG. 1 is a general framework of the method described in the embodiments of the present application;
FIG. 2 is a model predictive result of a simulation experiment of an embodiment of the present application;
FIG. 3 is a comparison of the control results of the method according to the example of the present application with those of the conventional method, wherein (a) is the method KIFNN-MPC according to the example of the present application, and (b) (c) (d) corresponds to the conventional method FNN-PMC, PID, pcRNN-MPC, respectively.
Detailed Description
The following describes in detail the embodiments of the present invention, which are developed based on the technical solution of the present invention, and provide detailed embodiments and specific operation procedures, and further explain the technical solution of the present invention.
The invention provides a nonlinear model predictive control method based on a knowledge guided neural network, wherein the overall framework is shown in fig. 1 and mainly comprises three parts: the system is subjected to parameterization representation, system structure knowledge extraction and nonlinear process high-precision modeling and control based on knowledge-guided neural network. In the first part, the nonlinear part is converted into a linear combination of a plurality of basis functions in a systematic and parameterized manner. In the second part, system structure knowledge including system input order, system output order and system time delay is extracted based on a sparse representation method, a neural network input neuron with a special structure is constructed under the guidance of the system structure knowledge, and a feedforward neural network prediction model is established. In the third part, the built prediction model is embedded into a model prediction control framework, so that the control instruction is issued.
1. System over-parameterized representation
The system input-output relationship of a nonlinear industrial process is expressed in the form of:
/>
wherein,is->Time system input variable,/->Is->Time-of-day non-measurable intermediate variable,/->Is->Time system output variable,/->Representing an unknown nonlinear function->And->Weight factor representing system input/output, +.>Representing time delay(s)>And->Representing the order of the system input/output, +.>Representing gaussian white noise. The system structure knowledge includes the order of system input and output>,/>Time delay +.>This is critical to building accurate predictive models.
The over-parameterized representation is a summation of a series of basis functions with nonlinear components in the system, typically using taylor series or fourier series, such that the nonlinear components can be converted into linear combinations of basis functions. In general, when the static nonlinear section is approximated by a taylor series, the mathematical expression of over-parameterization is expressed as follows:
/>
wherein the non-linearity parameter。
2. System architecture knowledge extraction
Knowledge of system architecture including the order of system inputs and outputs,/>And delay->. The system structural parameters include weighting coefficients->And->Nonlinear parameter->. The system structure knowledge extraction is based on sparse representation, specifically, the dictionary is adopted to reconstruct the system output data, and the optimization problem is as follows:
/>
wherein,to have->Sparse vectors of non-zero elements, i.e. +.>Sparsity of +.>. In order to be able to fully extract the knowledge of the system structure, it is necessary to set sparsity +.>Redundancy, i.e.)>. When nonlinear parameter set->After sufficient sampling, the optimization problem (3) can be solved accurately using an orthogonal matching pursuit method (orthogonal matching pursuit, OMP).
First, the construction system inputs a dictionary:
wherein,is a positive integer greater than the input/output order, i.e. +.>,/>For a certain moment in the past of the system, the system inputs the block dictionary +.>The method comprises the following steps:
/>
wherein,for the number of training data, +.>Represents>To->Is a system input of (a). Nonlinear parameter set->Is in the nonlinear parameter->In space at equal intervalsAs a result.
Next, a system output dictionary is constructed:
/>
wherein the system outputs a block dictionaryThe method comprises the following steps:
/>
wherein,represents slave time +.>To->Is provided.
System input dictionaryAnd system output dictionary->Constitute +.>:
(8)
Corresponding toThe method comprises the following steps:
(9)
after obtainingAfter that, there is also a need for +.>The real system structure knowledge and system structure parameters can be obtained only by performing three compensation strategies.
1) Original sparse vectorIs approximately sparse, with some elements being non-zero elements of very small values;
2) Original sparse vectorIs the normalized result;
3) Corresponding to the block dictionary input by the same system, original sparse vectorThere are a plurality of non-zero elements.
Compensation strategy 1: zero operation
Original sparse vectorIs approximately sparse, meaning that some elements are non-zero, but negligible, which are not of little contribution to reconstruction, but also add to the complexity of the model. Thus, these elements are zeroed by using a threshold operation. Dictionary->By->And->Two parts are formed, and the original sparse vector is according to +.>The structure of (2) can be divided into two parts, namely +.>. They are each zero-operated according to the following formula:
(10)
wherein,for vector->Is>Element(s)>Vector->The element with the largest absolute value, +.>Sparse vector obtained after zero operation is +.>The method comprises the following steps:
(11)
compensation strategy 2: inverse normalization operation
Due to the dictionaryColumn normalization is performed, so that inverse normalization is needed to obtain a true sparse vector value:
(12)
wherein,is composed of two norms of each dictionary atom,/->Representing division of the corresponding position.
Compensation operation 3: fusion operation
The nonlinear parameters corresponding to the selected atoms in the system input block dictionary areMeaning that at most only one atom in the system input block dictionary is selected. Since the nonlinear parameter is sampled in a limited space, there is a high probability that the nonlinear parameter set +.>No system non-linearity parameters are included. At sparsity +.>In the case of redundancy, there may be a case where multiple atoms implement reconstruction together, i.e., multiple atoms in the system input block dictionary are selected to participate in reconstruction. Therefore, it is necessary to fuse multiple selected atoms in the system input block dictionary to further reduce sparse orientationAnd the sparsity of the quantity reduces the computational complexity. Assume that the system inputs the block dictionary +.>The selected atom index of (a) isThe following data pairs were constructed:
(13)
wherein,representing a non-linear parameter->Corresponding to the system input block dictionary +.>Selected atoms and->In the following, two matrices are defined:
(14)
(15)
estimation of nonlinear parametersCalculated according to the following formula:
(16)
wherein,representative vector->Is the first element of (c).
To this end, according to the solvedThe order of system input/output can be obtained from the position relation of (a)>,/>Delay->After three compensation strategies, the weighting coefficients +.>And->Nonlinear parameter->。
3. Nonlinear process high-precision modeling and control based on knowledge-guided neural network
In obtaining system structure knowledge, i.e. the order of system input and output,/>And delay->Then, a high-precision neural network prediction model guided by system structural knowledge is constructed by selecting proper input neurons:
(17)
the network is trained by adopting a Levenberg-Marquardt algorithm, and the network training loss function is as follows:
(18)
wherein,representing the amount of training data. />Representing the actual output of the system,/->Representing the prediction output of the prediction model.
Taking the formula (17) as a prediction model in model prediction control, a nonlinear process high-precision model prediction control strategy based on a knowledge-guided neural network can be obtained:
(19)
wherein,,/>representing weights +.>Representing the prediction horizon +_>Representing the control time domain +_>And->Representing system input->Upper and lower limits of>Representing a system reference trajectory->Representing the system closed loop predicted output,/->Representation->Predicted value of time prediction model->Is typically set to 1,/for the weighting factor>Is a smoothing factor->,/>Representing the set point. The optimization problem (19) is solved by adopting the SQP algorithm, and a control sequence is obtained by solving>After that, the first item of the sequence +.>Acts on the system to realize the issuing of control input.
In order to verify the effectiveness of the method, the following numerical simulation experiment is designed for verification:
(20)
as shown in FIG. 2, the modeling accuracy of the knowledge guided neural network provided by the invention can be improved by 24.54% (KIFNN is 1.54% and conventional FNN is 26.08%) when the data size is 300, compared with the conventional method
As shown in FIG. 3 and Table 1, the nonlinear process control method (KIFNN-MPC) based on the knowledge guided neural network can improve the control accuracy by more than 30%.
The nonlinear model predictive control method and system based on the knowledge guided neural network can be applied to nonlinear control of industrial processes, and can remarkably improve modeling and control precision, and particularly, compared with a traditional method, the modeling precision is improved by more than 20%, and the control precision is improved by more than 30%. By extracting the structural knowledge of the system, the problem of redundancy of the input neurons of the neural network is avoided, and the method provides guarantee for the accurate control of the industrial process. In addition, the data volume required by modeling can be greatly reduced by the technology, and the application range of the method is greatly expanded.
The above embodiments are preferred embodiments of the present application, and various changes or modifications may be made on the basis thereof by those skilled in the art, and such changes or modifications should be included within the scope of the present application without departing from the general inventive concept.
Claims (9)
1. A method for nonlinear process control based on knowledge-guided neural networks, comprising:
extracting system structural knowledge of a nonlinear industrial process based on a sparse representation method: system input order, system output order and system time delay;
constructing a feedforward neural network with a corresponding input neuron structure according to the extracted system structure knowledge;
training the constructed feedforward neural network by utilizing the system input and output training data to obtain a system prediction model;
and taking the trained system prediction model as a prediction model in a model prediction control strategy, obtaining a system input sequence by adopting the model prediction control strategy, and acting the first item of the sequence on the industrial process system.
2. The knowledge-guided neural network-based nonlinear process control method of claim 1, wherein extracting system structural knowledge comprises:
first, the input-output relationship of a nonlinear industrial process system is expressed as:
;
wherein,is->Time system input variable,/->Is->Time-of-day non-measurable intermediate variable,/->Is->Time system output variable,/->Representing an unknown nonlinear function->And->Weight factors representing system inputs and outputs, +.>Representing system delay->And->Representing the order of the system inputs and outputs, +.>Representing gaussian white noise;
then, the nonlinear part of the system, namely the unknown nonlinear function, is parameterized by the systemConverting into a linear combination of a plurality of basis functions; the base function coefficients and the constant coefficients in the linear combination form unknown nonlinear parameters;
by sampling nonlinear parameters and combining past input data of the system, a nonlinear function-based construction is performedIs input into the dictionary; constructing a system output dictionary by utilizing the past output data of the system;
finally, the input and output dictionaries of the system are integrated, and a dictionary learning method is adopted to obtain sparse vectors of future output data of the system; and comparing the sparse vector with an input-output relation of the system to obtain system structural knowledge: system time delaySystem input order->And system output order +.>。
3. The knowledge-guided neural network-based nonlinear process control method of claim 2, wherein the nonlinear function is performed using a taylor series or a fourier seriesConverted into a linear combination of basis functions.
4. The knowledge-guided neural network-based nonlinear process control method of claim 2, wherein constructing the nonlinear function-based nonlinear functionIs input into the dictionary->The method comprises the following steps:
;
wherein,is a positive integer greater than both the input and output orders of the system, i.e. +.>,/>For a certain moment in the past of the system, the system inputs the block dictionary +.>The method comprises the following steps:
;
wherein,for the number of training data, +.>Represents>To->Is input into the system; nonlinear parameter set->Is in the nonlinear parameter->M results obtained by fully sampling at equal intervals in space;
building a system output dictionaryThe method comprises the following steps:
;
wherein the system outputs a block dictionaryThe method comprises the following steps:
;
wherein,represents slave time +.>To->Is a system output data;
synthesizing dictionary input and output of system to obtain dictionary in dictionary learning optimization problem:
。
5. The nonlinear process control method based on the knowledge-guided neural network as set forth in claim 4, wherein the optimization problem constructed when the dictionary learning method is adopted is:
;
wherein,for future output data of the system, i.e. +.>,/>Is->Warp dictionary->Is a sparse code of (a).
6. The nonlinear process control method based on the knowledge guided neural network according to claim 2, wherein when a dictionary learning method is adopted to obtain a sparse vector of future output data of the system, compensation operation is performed on the sparse vector, and then the sparse vector after the compensation operation is compared with an input-output relational expression of the system to obtain system structure knowledge; wherein the compensation operation sequentially comprises:
(1) Zero operation
Dictionary-basedBy->And->Two parts are composed, and the sparse vector before compensation is correspondingly formed>Sparse vector divided into system past output data +.>Sparse vector of system past input data +.>Two parts, i.e.)>The method comprises the steps of carrying out a first treatment on the surface of the Then pair->And->Zero operation is performed respectively:
;
wherein,for vector->Or->Is>Element(s)>For vector->Or->The element with the largest absolute value, +.>Sparse vector obtained after zero operation is +.>The method comprises the following steps:
;
(2) Inverse normalization operation
Dictionary-basedColumn normalization operation is performed, and sparse vector obtained after zero operation is +.>Performing inverse normalization to obtain a true sparse vector +.>:
;
Wherein,is composed of two norms of each dictionary atom,/->Division representing the corresponding position;
(3) Fusion operation
Selected ones of the system input block dictionariesThe atomic index is->The following data pairs were constructed:
;
wherein,representing a non-linear parameter->Corresponding to the system input block dictionarySelected atoms and->Corresponding non-zero elements of (a); next, two matrices are defined:
;
;
calculating an estimate of the nonlinear parameter according to:
;
Wherein,representative vector->Is the first element of (c).
7. The knowledge-guided neural network-based nonlinear process control method of claim 2, wherein constructing a feed-forward neural network with a corresponding input neuron structure in accordance with the extracted system structure knowledge is expressed as:
;
wherein,representing a feed-forward neural network; />、/>And->Respectively representing the extracted system time delay, the system input order and the system output order; />Representing feed forward neural network prediction +.>Outputting a system at the moment;
the network is trained by adopting a Levenberg-Marquardt algorithm, and the network training loss function is as follows:
;
wherein,representing the number of training data +.>Representing the actual output of the system,/->Representing the predicted output of the feedforward neural network.
8. The knowledge guided neural network based nonlinear process control method of claim 2, wherein the optimization problem of obtaining the system input sequence using the model predictive control strategy is expressed as:
;
wherein,,/>representing weights +.>Representing the prediction horizon +_>Representing the control time domain of the device,and->Representing system input->Upper and lower limits of>Representing a system reference trajectory->Representing the system closed loop predicted output,/->Representation->Predicted value of time prediction model->Representing a predictive model->As a result of the weighting factor(s),is a smoothing factor->,/>Represents a set value;
solving the optimization problem by adopting an SQP algorithm to obtain a system input sequenceI.e. the system control sequence, and will be the first item +.>Acts on the system to realize the issuing of control input.
9. A system based on the knowledge-guided neural network-based nonlinear process control method of any one of claims 1-8, comprising:
the system structure knowledge extraction module is used for: extracting system structure knowledge of a nonlinear industrial process based on a sparse representation method, wherein the system structure knowledge comprises a system input order, a system output order and a system time delay;
the neural network construction module is used for: constructing a feedforward neural network with a corresponding input neuron structure according to the extracted system structure knowledge;
the prediction model training module is used for: training the constructed feedforward neural network by utilizing the system input and output training data to obtain a system prediction model;
model predictive control module for: and taking the trained system prediction model as a prediction model in a model prediction control strategy, obtaining a system input sequence by adopting the model prediction control strategy, and acting the first item of the sequence on the industrial process system.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311629901.2A CN117331316A (en) | 2023-12-01 | 2023-12-01 | Nonlinear process control method and system based on knowledge guided neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311629901.2A CN117331316A (en) | 2023-12-01 | 2023-12-01 | Nonlinear process control method and system based on knowledge guided neural network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN117331316A true CN117331316A (en) | 2024-01-02 |
Family
ID=89292041
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202311629901.2A Pending CN117331316A (en) | 2023-12-01 | 2023-12-01 | Nonlinear process control method and system based on knowledge guided neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117331316A (en) |
-
2023
- 2023-12-01 CN CN202311629901.2A patent/CN117331316A/en active Pending
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111459051B (en) | Discrete terminal sliding mode model-free control method with disturbance observer | |
Chen et al. | Adaptive control for nonlinear MIMO time-delay systems based on fuzzy approximation | |
Zhang et al. | Global exponential stability of discrete-time almost automorphic Caputo–Fabrizio BAM fuzzy neural networks via exponential Euler technique | |
Shi et al. | Adaptive fuzzy prescribed performance control for MIMO nonlinear systems with unknown control direction and unknown dead-zone inputs | |
Yang et al. | SGD-based adaptive NN control design for uncertain nonlinear systems | |
Moghaddam et al. | Recursive identification of multiple-input single-output fractional-order Hammerstein model with time delay | |
CN113988449B (en) | Wind power prediction method based on transducer model | |
Jami ‘in et al. | Quasi‐ARX neural network based adaptive predictive control for nonlinear systems | |
Wang et al. | Adaptive neural control for a general class of pure-feedback stochastic nonlinear systems | |
Elloumi et al. | Adaptive control scheme for large‐scale interconnected systems described by Hammerstein models | |
Bali et al. | Hybrid neural network control for nonlinear continuous time systems with time delays and dead zone input | |
Schiassi et al. | Bellman neural networks for the class of optimal control problems with integral quadratic cost | |
CN110597055B (en) | Uncertainty-resistant 2D piecewise affine intermittent process minimum-maximum optimization prediction control method | |
CN110750049B (en) | Intermittent process 2D prediction fault-tolerant control method with time lag and disturbance | |
CN117331316A (en) | Nonlinear process control method and system based on knowledge guided neural network | |
Kusumoputro et al. | Performance characteristics of an improved single neuron PID controller using additional error of an inversed control signal | |
Abbaszadeh et al. | Constrained nonlinear model predictive control of an MMA polymerization process via evolutionary optimization | |
CN112346342A (en) | Single-network self-adaptive evaluation design method of non-affine dynamic system | |
Chuong et al. | Fractional pi control for coupled-tank mimo system | |
Negara et al. | Tuning of Fractional-Order PID Controller for Electro-Hydraulic Servo Valve System | |
Yang et al. | Intelligent forecasting system using Grey model combined with neural network | |
Song et al. | Adaptive interval type-2 fuzzy fixed-time control for nonlinear MIMO fast-slow coupled systems with user-defined tracking performance | |
CN108897219B (en) | Chemical uncertain industrial process constraint prediction control method | |
CN113568309B (en) | On-line space-time control method for temperature field | |
Lu et al. | SAC for nonlinear systems using elman recurrent neural networks |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication |