CN107145935A - A kind of Smith Prediction Control methods based on modified neutral net - Google Patents
A kind of Smith Prediction Control methods based on modified neutral net Download PDFInfo
- Publication number
- CN107145935A CN107145935A CN201710332490.9A CN201710332490A CN107145935A CN 107145935 A CN107145935 A CN 107145935A CN 201710332490 A CN201710332490 A CN 201710332490A CN 107145935 A CN107145935 A CN 107145935A
- Authority
- CN
- China
- Prior art keywords
- hiddennum
- neural network
- individual
- output
- inputnum
- 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 41
- 230000007935 neutral effect Effects 0.000 title claims abstract description 18
- 238000013528 artificial neural network Methods 0.000 claims abstract description 41
- 238000012549 training Methods 0.000 claims abstract description 16
- 230000001537 neural effect Effects 0.000 claims abstract description 15
- 230000002068 genetic effect Effects 0.000 claims abstract description 12
- 230000008030 elimination Effects 0.000 claims abstract description 4
- 238000003379 elimination reaction Methods 0.000 claims abstract description 4
- 239000011159 matrix material Substances 0.000 claims description 23
- VYZAMTAEIAYCRO-UHFFFAOYSA-N Chromium Chemical compound [Cr] VYZAMTAEIAYCRO-UHFFFAOYSA-N 0.000 claims description 20
- 230000035772 mutation Effects 0.000 claims description 9
- 238000013480 data collection Methods 0.000 claims description 6
- 230000002860 competitive effect Effects 0.000 claims description 3
- 238000013213 extrapolation Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 230000004083 survival effect Effects 0.000 claims description 3
- 238000012546 transfer Methods 0.000 claims description 2
- 230000008569 process Effects 0.000 description 14
- 238000011478 gradient descent method Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 108090000623 proteins and genes Proteins 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000013178 mathematical model Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000003111 delayed effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000009191 jumping Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000004218 nerve net Anatomy 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
-
- 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
- G05B11/00—Automatic controllers
- G05B11/01—Automatic controllers electric
- G05B11/36—Automatic controllers electric with provision for obtaining particular characteristics, e.g. proportional, integral, differential
- G05B11/42—Automatic controllers electric with provision for obtaining particular characteristics, e.g. proportional, integral, differential for obtaining a characteristic which is both proportional and time-dependent, e.g. P. I., P. I. D.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Automation & Control Theory (AREA)
- Feedback Control In General (AREA)
Abstract
The present invention relates to a kind of Smith Prediction Control methods based on modified neutral net, comprise the following steps：Gather large time delay controlled device inputoutput data is as preliminary sample and carries out time lag elimination；The input layer and output layer node number of BP neural network are determined, while determining the hidden layer number of plies and node in hidden layer；Determine the length of genetic algorithm individual；Optimized individual is obtained using genetic algorithm, the optimal initial weights and threshold value for obtaining BP neural network are decoded it；The BP networks of training now, the anhysteretic part of controlled device and controlled device is recognized with it；Input/output information after BP neural network is recognized is transmitted to the PID controller that RBF neural is adjusted, and Prediction Control is realized using the PID controller after adjusting.Large dead time controlled device can be better controlled by the present invention.
Description
Technical field
The present invention relates to automatic control technology field, more particularly to a kind of Smith based on modified neutral net is pre-
Estimate control method.
Background technology
In actual industrial processes, controlled device not only has bulk delay, and is widely present different journeys
The purely retarded of degree, such as generally existing pure stagnant in the technical process such as common pipe conveying procedure, detection and guidance process
Phenomenon afterwards, purely retarded is extremely disadvantageous for the performance of control system, in the range of lag time, and controller is due to can not be from closed loop
The actual change of current time controlled volume is obtained in loop and causes it can not make corresponding adjustment, the output of control signal with
The action of controlled volume deviation is inconsistent so that the regulating time of system and overshoot quantitative change are big, resulted even in when serious system because
It is unstable and out of control.It is generally believed that if the ratio between time constant of pure delay time and process is more than 0.3, the process is called tool
There is the process of large time delay.How Large Dead-time Process being control effectively, engineering field is of interest asks for always control
Topic, solution relatively common at present mainly has conventional control scheme, discrete control program and Smith Predictive Compensation Controls
Scheme.
The construction of traditional Smith predictor needs to know the mathematical models of controlled process, for most industrial process
For, the characteristics of often there is nonlinear dissipation, the difficulty for setting up accurate model is very big；In addition, if there is outer do in system
Disturb and disturb and be not comprised in Smith predictive compensation devices, the effect that system suppresses outer disturbance will be very undesirable.Therefore, research
The design for improving traditional Smith predictor is necessary.With developing rapidly for the present computer technology, by computer
Powerful operational capability and storage capacity solves the problems, such as that this Nonlinear Model becomes a reality.Scholar is had at present to propose to utilize BP
Neural network model by the study of the input/output relation to controlled process, construct can on-line identification neutral net
Smith predictor, but it is slow and be easily trapped into local minimum point to be due to that traditional BP neural network has convergence rate
Inherent shortcoming, causes its practicality and reliability decrease.In addition, the generalization ability of BP neural network is limited, so typically using
The method of on-line tuning weights is recognized, in dynamic adjustment process, and the output of identification is possible to have shaking by a relatively large margin
Swing, so then simple use BP neural network identification large-lag object is subject to general control and still is possible to that system can be caused
It is unstable.
The content of the invention
The technical problems to be solved by the invention are to provide a kind of Smith Prediction Control sides based on modified neutral net
Method, large dead time controlled device can be better controlled.
The technical solution adopted for the present invention to solve the technical problems is：There is provided a kind of based on modified neutral net
Smith Prediction Control methods, comprise the following steps：
(1) inputoutput data of collection large time delay controlled device is responded bent as preliminary sample using the output of system
Line obtains time delay, is then carried out into time lag elimination the time delay of input sample sequence, i.e., former preliminary sample is converted to
The inputoutput data collection of controlled device anhysteretic part simultaneously regard this inputoutput data collection as training sample；
(2) number of inputoutput data determines the input layer and output node layer of BP neural network in every group of sample
Number, while determining the hidden layer number of plies and node in hidden layer；
(3) length of genetic algorithm individual is determined according to the nodes of each layer of BP neural network；
(4) population scale of genetic algorithm, iterations, crossover probability and mutation probability are determined, then using real number
The mode of coding initializes each individual of population uniformly randomly；
(5) initial weight and threshold value of BP neural network are obtained by all individual decodings in population, is instructed with training sample
White silk BP neural network, regard the quadratic sum of reality output and the difference of the prediction output after training as individual fitness；
(6) population of future generation is obtained after selection, the competitive way intersected, make a variation and selected the superior and eliminated the inferior, calculates this kind
The fitness of group；
(7) judge whether whether the fitness of current population optimum individual meet sets requirement or current iterations
Setting maximum is reached, if it is not, return to step (6)；
(8) retain current optimized individual, decode it the optimal initial weights and threshold value for obtaining BP neural network；
(9) the BP networks of training now, the anhysteretic part of controlled device and controlled device is recognized with it；
(10) input/output information after BP neural network is recognized is transmitted to the PID controller that RBF neural is adjusted, profit
Prediction Control is realized with the PID controller after adjusting.
Data when comprehensively collection controlled device is in each stage during gathered data in the step (1).
The length and the weights of BP neural network and the total number phase of threshold value of genetic algorithm individual in the step (3)
Deng.
Decoding obtains the initial weight of BP neural network in the step (5) and the concrete mode of threshold value is：
W1=chrom (1:inputnum*hiddennum)
B1=chrom (inputnum*hiddennum+1:inputnum*hiddennum+hiddennum)
W2=chrom (inputnum*hiddennum+hiddennum+1:inputnum*hiddennum
+hiddennum+hiddennum*outputnum)
B2=chrom (inputnum*hiddennum+hiddennum+hiddennum*outputnum+1:
inputnum*hiddennum+hiddennum+hiddennum*outputnum+outputnum)
W1 element is arranged as to the matrix w1 of an inputnum rows hiddennum row in order again again,
B1 element is arranged as to the matrix b1 of a 1 row hiddennum row in order again,
W2 element is arranged as to the matrix w2 of a hiddennum rows outputnum row in order again,
B2 element is arranged as to the matrix b2 of a 1 row outputnum row in order again,
In above formula, chrom represents the individual in population, chrom (x1:X2 individual 1 element of xth) is represented to xth 2
Individual element；Inputnum is input layer number, and outputnum is output layer node number, and hiddennum is hidden layer section
Point number；W1, b1, w2, b2 represent respectively input layer to hidden layer weight matrix and threshold matrix and hidden layer to output
The weight matrix and threshold matrix of layer.
In the step (6), roulette wheel method is used during selection operation, crossover operation uses real number interior extrapolation method, mutation operation
When select real number alternative method；The survival of the fittest refer to it is every by a wheel selection, intersect, after mutation operation using current fitness most
Good individual removes to substitute the worst individual of fitness in population.
The transfer mode of input/output information in the step (10) after BP neural network identification is by using RBF god
Input/output information through network identification BP neural network, calculates the Jacobian information for obtaining controlled device, then utilizes
Jacobian information on-line tuning pid parameters.
RBF neural input layer corresponds to Δ u (k), y (k) and y respectively using three nodes in the step (10)
(k-1), wherein Δ u (k) is the output increment of controller, and y (k) and y (k-1) export for the identification of BP neural network, y (k) tables
Show the output of current sample time, y (k-1) is then the output of a upper sampling instant；The RBF neural is output as ym
(k), performance indications are set toIts adjustment for exporting weights, node center and node sound stage width parameter makes
With gradient descent method, target is so that performance indications J is small as far as possible, i.e. ym(k) y (k) is approached.
Beneficial effect
As a result of above-mentioned technical scheme, the present invention compared with prior art, has the following advantages that and actively imitated
Really：The present invention recognizes large time delay controlled device using BP neural network, therefore can need not know about controlled device accurate mathematical
It is controlled by the case of model, while using RBF neural Tuning PID Controller, effectively improving adjusting for PID
Speed, enhances the stability of system；In addition, using genetic algorithm optimization BP network, making it jump out local optimum as far as possible
And globe optimum is found, improve the practicality of whole scheme.
Brief description of the drawings
Fig. 1 is the flow chart of the Smith Prediction Control methods of the invention based on modified neutral net；
Fig. 2 is traditional Smith Prediction Controls functional-block diagram；
Fig. 3 is the structural representation of the whole Prediction Control system of the present invention.
Embodiment
With reference to specific embodiment, the present invention is expanded on further.It should be understood that these embodiments are merely to illustrate the present invention
Rather than limitation the scope of the present invention.In addition, it is to be understood that after the content of the invention lectured has been read, people in the art
Member can make various changes or modifications to the present invention, and these equivalent form of values equally fall within the application appended claims and limited
Scope.
Fig. 3 is the structure chart of whole Prediction Control system, including large dead time controlled device Gp(s), NN1 and NN2 nerve nets
Network module, and RBF network identification modules.Wherein NN1 neutral nets are used to recognize controlled process, and NN2 neutral nets are tied with NN1
Structure is identical, the non-delayed part for recognizing controlled process, i.e., the G of traditional Smith predictor (as shown in Figure 2)S(s)e-tSWith GS
(s) replaced respectively with neural network module NN1 and NN2；In addition, defeated to the input of BP networks by RBF neural module
Go out the identification of information, obtain the Jacobian information of controlled device, then utilize Jacobian information on-line tuning pid parameters.
As shown in figure 1, specifically including following steps：
(1) collection large time delay controlled device Gp(s) inputoutput data is rung as preliminary sample using the output of system
Answer curve to obtain delay time T, the delay time T of input sample sequence is then subjected to time lag elimination, i.e., former preliminary sample
Be converted to the inputoutput data collection of controlled device anhysteretic part and regard this inputoutput data collection as training sample, collection
Data when controlled device is in each stage should be comprehensively gathered during data as far as possible, such as test quilt using a variety of input signals
Control object and the output for gathering its corresponding transient process and steady-state process；
(2) number of inputoutput data determines the input layer and output layer node number of BP neural network in sample, than
Thus mathematical modeling such as controlled device formula can describe：Y (k)=f [y (k-1) ..., y (k-n), u (k-d) ... u (k-d-m)],
Now the input layer number of BP neural network should be n+m+1, additionally, due to present invention is generally directed to single-input single-output system,
So output layer node number is generally 1；Simultaneously it needs to be determined that the hidden layer number of plies and node in hidden layer, the choosing of the hidden layer number of plies
One layer is selected, as node in hidden layer used here as empirical equation
Wherein, inputnum is input layer number, and outputnum is output layer node number, and hiddennum is hidden layer node
Number, a is the constant between 0~10；
(3) length of genetic algorithm individual, specifically, individual lengths are determined according to the nodes of each layer of BP neural network
Should be equal with the weights of network and the total number of threshold value, for three-layer network described in step (2), individual lengths can be by
Following formula is determined：Numsum=(inputnum+1) * hiddennum+ (hiddennum+1) * outputnum；
(4) parameters of genetic algorithm, such as population scale are determined, iterations, crossover probability and mutation probability etc.,
Then using the mode of real coding it is uniformly random initialize population each individual；
(5) initial weight and threshold value of BP neural network are obtained by all individual decodings in population, here specific solution
Code mode be：
W1=chrom (1:inputnum*hiddennum)
B1=chrom (inputnum*hiddennum+1:inputnum*hiddennum+hiddennum)
W2=chrom (inputnum*hiddennum+hiddennum+1:inputnum*hiddennum
+hiddennum+hiddennum*outputnum)
B2=chrom (inputnum*hiddennum+hiddennum+hiddennum*outputnum+1:
inputnum*hiddennum+hiddennum+hiddennum*outputnum+outputnum)
W1 element is arranged as to the matrix w1 of an inputnum rows hiddennum row in order again again,
B1 element is arranged as to the matrix b1 of a 1 row hiddennum row in order again,
W2 element is arranged as to the matrix w2 of a hiddennum rows outputnum row in order again,
B2 element is arranged as to the matrix b2 of a 1 row outputnum row in order again,
In above formula, chrom represents the individual in population, chrom (x1:X2 individual 1 element of xth) is represented to xth 2
Individual element, such as chrom (1:Inputnum*hiddennum) represent the 1st element of this individual of chrom to the
Inputnum*hiddennum element；W1, b1, w2, b2 represent input layer to the weight matrix and threshold value square of hidden layer respectively
Battle array and hidden layer to output layer weight matrix and threshold matrix.
Then BP neural network is trained with training sample, by square of reality output and the difference of the prediction output after training
With the fitness as individual, i.e. individual adaptation degree is represented by：
In formula, yiRepresent the reality output of i-th of training sample during BP network trainings, OiOiIt is then BP neural network correspondence
Prediction output, n be total training sample number；
(6) population of future generation is obtained after selection, the competitive way intersected, make a variation and selected the superior and eliminated the inferior, then calculated
The fitness of the population.Specific operation：
A) selection operation uses roulette wheel method, and each individual i select probability is：
In formula, fitness (i) represents individual i fitness, and N is population at individual number；
B) crossover operation uses real number interior extrapolation method, m-th of individual amWith n-th of individual anIn the crossover operation methods of j such as
Under：
amj=amj*(1-b)+anj*b
anj=anj*(1-b)+amj*b
In formula, b is the random number between 0~1；
C) mutation operation selection real number alternative method, chooses j-th of gene a of i-th of individualijEnter row variation,
If rand ＞ 0.5, aij=aij+(aij-amax)*f(g)
If rand≤0.5, aij=aij+(amin-aij)*f(g)
And f (g)=r* (1-g/Gmax)2
In formula, rand and r are for the random number between 0~1, amaxFor gene aijThe upper bound, aminFor gene aijUnder
Boundary, g is current iteration number of times, GmaxFor maximum iteration；
D) survival of the fittest refer to it is every select by a wheel, intersect, after mutation operation using the best individual of current fitness
Remove to substitute the worst individual of fitness in population；
(7) judge whether the fitness of current population meets sets requirement or whether current iterations has reached and set
Determine maximum, if not, jumping to step (6), often perform a step (6), iterations adds 1；
(8) retain current optimized individual, decode it the optimal initial weights and threshold value for obtaining BP neural network, decoding
Process is identical with step (5)；
(9) BP neural network using the training of stochastic gradient descent method now, then with its recognize controlled device and by
The anhysteretic part of object is controlled, the weights of on-line tuning network are needed during identification, to ensure that identification result is tried one's best close to reality
Controlled device；
(10) input/output information after BP neural network is recognized is transmitted to the PID controller that RBF neural is adjusted, and passes
The mode of passing is the input/output information that BP neural network is recognized by using RBF neural, and calculating obtains controlled device
Jacobian information, then utilizes Jacobian information on-line tuning pid parameters.
Specifically, RBF neural input layer corresponds to Δ u (k), y (k) and y (k-1) respectively using three nodes, its
Middle Δ u (k) is the output increment of controller, and y (k) and y (k-1) export for the identification of BP neural network, and y (k) represents currently to adopt
The output at sample moment, y (k-1) is then the output of a upper sampling instant.
RBF neural is output as ym(k), performance indications are set toIt exports weights, node
The adjustment of center and node sound stage width parameter uses gradient descent method, and target is so that performance indications J is small as far as possible, i.e. ym
(k) y (k) is approached.
Then the Jacobian information of controlled device can be obtained by following formula：
In formula, wjFor the weights of hidden layer to output layer, bjFor base bandwidth parameter, cjiMeasured for hidden layer node center.
Meanwhile, kp、ki、kdAdjustment use gradient descent method, its respond adjustment amount it is as follows：
In formula, η is kp、ki、kdLearning rate, e (k) is the output error of system, i.e. e (k)=r (k)-y (k), and r (k) is
System set-point.
It is seen that, because the present invention is using BP neural network identification large time delay controlled device, therefore it need not can know
It is controlled by the case of road controlled device mathematical models, while using RBF neural Tuning PID Controller,
Effectively improve PID adjusts speed, enhances the stability of system；In addition, using genetic algorithm optimization BP network, using up it
It is possible to jump out local optimum and find globe optimum, improve the practicality of whole scheme.
Claims (7)
1. a kind of Smith Prediction Control methods based on modified neutral net, it is characterised in that comprise the following steps：
(1) inputoutput data of collection large time delay controlled device is obtained as preliminary sample using the output response curve of system
To time delay, the time delay of input sample sequence is then subjected to time lag elimination, i.e., former preliminary sample is converted to controlled
The inputoutput data collection of object anhysteretic part simultaneously regard this inputoutput data collection as training sample；
(2) number of inputoutput data determines the input layer and output layer node number of BP neural network in every group of sample,
The hidden layer number of plies and node in hidden layer are determined simultaneously；
(3) length of genetic algorithm individual is determined according to the nodes of each layer of BP neural network；
(4) population scale of genetic algorithm, iterations, crossover probability and mutation probability are determined, then using real coding
Mode initialize uniformly randomly population each individual；
(5) initial weight and threshold value of BP neural network are obtained by all individual decodings in population, BP is trained with training sample
Neutral net, regard the quadratic sum of reality output and the difference of the prediction output after training as individual fitness；
(6) population of future generation is obtained after selection, the competitive way intersected, make a variation and selected the superior and eliminated the inferior, calculates the population
Fitness；
(7) judge whether the fitness of current population optimum individual meets sets requirement or whether current iterations has reached
To setting maximum, if it is not, return to step (6)；
(8) retain current optimized individual, decode it the optimal initial weights and threshold value for obtaining BP neural network；
(9) the BP networks of training now, the anhysteretic part of controlled device and controlled device is recognized with it；
(10) input/output information after BP neural network is recognized is transmitted to the PID controller that RBF neural is adjusted, using whole
PID controller after fixed realizes Prediction Control.
2. the Smith Prediction Control methods according to claim 1 based on modified neutral net, it is characterised in that institute
Data when controlled device is in each stage are comprehensively gathered when stating gathered data in step (1).
3. the Smith Prediction Control methods according to claim 1 based on modified neutral net, it is characterised in that institute
The length for stating genetic algorithm individual in step (3) is equal with the weights of BP neural network and the total number of threshold value.
4. the Smith Prediction Control methods according to claim 1 based on modified neutral net, it is characterised in that institute
State that decoding obtains the initial weight of BP neural network in step (5) and the concrete mode of threshold value is：
W1=chrom (1:inputnum*hiddennum)
B1=chrom (inputnum*hiddennum+1:inputnum*hiddennum+hiddennum)
W2=chrom (inputnum*hiddennum+hiddennum+1:inputnum*hiddennum+hiddennum+
hiddennum*outputnum)
B2=chrom (inputnum*hiddennum+hiddennum+hiddennum*outputnum+1:inputnum*
hiddennum+hiddennum+hiddennum*outputnum+outputnum)
W1 element is arranged as to the matrix w1 of an inputnum rows hiddennum row in order again again,
B1 element is arranged as to the matrix b1 of a 1 row hiddennum row in order again,
W2 element is arranged as to the matrix w2 of a hiddennum rows outputnum row in order again,
B2 element is arranged as to the matrix b2 of a 1 row outputnum row in order again,
In above formula, chrom represents the individual in population, chrom (x1:X2 individual 1 element of xth) is represented to 2 members of xth
Element；Inputnum is input layer number, and outputnum is output layer node number, and hiddennum is hidden layer node
Number；W1, b1, w2, b2 represent input layer to the weight matrix and threshold matrix and hidden layer of hidden layer to output layer respectively
Weight matrix and threshold matrix.
5. the Smith Prediction Control methods according to claim 1 based on modified neutral net, it is characterised in that institute
State in step (6), roulette wheel method is used during selection operation, crossover operation uses real number interior extrapolation method, real number is selected during mutation operation
Alternative method；The survival of the fittest refer to it is every by a wheel selection, intersect, gone using the best individual of current fitness after mutation operation
Substitute the worst individual of fitness in population.
6. the Smith Prediction Control methods according to claim 1 based on modified neutral net, it is characterised in that institute
The transfer mode for stating the input/output information in step (10) after BP neural network identification is recognized by using RBF neural
The input/output information of BP neural network, calculates the Jacobian information for obtaining controlled device, then utilizes Jacobian information
On-line tuning pid parameter.
7. the Smith Prediction Control methods according to claim 1 based on modified neutral net, it is characterised in that institute
State RBF neural input layer in step (10) and using three nodes, Δ u (k), y (k) and y (k-1) are corresponded to respectively, wherein
Δ u (k) is the output increment of controller, and y (k) and y (k-1) export for the identification of BP neural network, and y (k) represents present sample
The output at moment, y (k-1) is then the output of a upper sampling instant；The RBF neural is output as ym(k), performance indications
It is set toIts adjustment for exporting weights, node center and node sound stage width parameter is declined using gradient
Method, target is so that performance indications J is small as far as possible, i.e. ym(k) y (k) is approached.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710332490.9A CN107145935A (en) | 2017-05-12 | 2017-05-12 | A kind of Smith Prediction Control methods based on modified neutral net |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710332490.9A CN107145935A (en) | 2017-05-12 | 2017-05-12 | A kind of Smith Prediction Control methods based on modified neutral net |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107145935A true CN107145935A (en) | 2017-09-08 |
Family
ID=59776895
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710332490.9A Pending CN107145935A (en) | 2017-05-12 | 2017-05-12 | A kind of Smith Prediction Control methods based on modified neutral net |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107145935A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108267970A (en) * | 2018-01-25 | 2018-07-10 | 合肥工业大学 | Time lag rotor active balance control system and its method based on Smith models and single neuron PID |
| CN108490965A (en) * | 2018-04-19 | 2018-09-04 | 东华大学 | Rotor craft attitude control method based on Genetic Algorithm Optimized Neural Network |
| CN108508743A (en) * | 2018-06-25 | 2018-09-07 | 曾喆昭 | The quasi- PI PREDICTIVE CONTROLs new method of time lag system |
| CN108803335A (en) * | 2018-06-25 | 2018-11-13 | 南京邮电大学 | A kind of out of order removing method of DC servo motor control |
| CN108977897A (en) * | 2018-06-07 | 2018-12-11 | 浙江天悟智能技术有限公司 | Melt-spinning process control method based on the inherent plasticity echo state network in part |
| CN111103790A (en) * | 2019-12-05 | 2020-05-05 | 珠海格力电器股份有限公司 | Parameter setting method and device of PID controller, storage medium, terminal and system |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105913150A (en) * | 2016-04-12 | 2016-08-31 | 河海大学常州校区 | BP neural network photovoltaic power station generating capacity prediction method based on genetic algorithm |
-
2017
- 2017-05-12 CN CN201710332490.9A patent/CN107145935A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105913150A (en) * | 2016-04-12 | 2016-08-31 | 河海大学常州校区 | BP neural network photovoltaic power station generating capacity prediction method based on genetic algorithm |
Non-Patent Citations (3)
| Title |
|---|
| MATLAB中文论坛: "《MATLAB神经网络30个案例分析》", 30 April 2010 * |
| 何臻祥 等: "基于Smith预估补偿与RBF神经网络的PID控制在工业平缝机脚踏板调速模块中的应用", 《四川兵工学报》 * |
| 季春光 等: "基于神经网络的时变大滞后***的Smith预估控制", 《哈尔滨工业大学学报》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108267970A (en) * | 2018-01-25 | 2018-07-10 | 合肥工业大学 | Time lag rotor active balance control system and its method based on Smith models and single neuron PID |
| CN108490965A (en) * | 2018-04-19 | 2018-09-04 | 东华大学 | Rotor craft attitude control method based on Genetic Algorithm Optimized Neural Network |
| CN108977897A (en) * | 2018-06-07 | 2018-12-11 | 浙江天悟智能技术有限公司 | Melt-spinning process control method based on the inherent plasticity echo state network in part |
| CN108977897B (en) * | 2018-06-07 | 2021-11-19 | 浙江天悟智能技术有限公司 | Melt spinning process control method based on local internal plasticity echo state network |
| CN108508743A (en) * | 2018-06-25 | 2018-09-07 | 曾喆昭 | The quasi- PI PREDICTIVE CONTROLs new method of time lag system |
| CN108803335A (en) * | 2018-06-25 | 2018-11-13 | 南京邮电大学 | A kind of out of order removing method of DC servo motor control |
| CN108803335B (en) * | 2018-06-25 | 2021-05-11 | 南京邮电大学 | Method for eliminating control disorder of direct current servo motor |
| CN108508743B (en) * | 2018-06-25 | 2021-06-01 | 长沙理工大学 | Novel quasi-PI predictive control method of time-lag system |
| CN111103790A (en) * | 2019-12-05 | 2020-05-05 | 珠海格力电器股份有限公司 | Parameter setting method and device of PID controller, storage medium, terminal and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107145935A (en) | A kind of Smith Prediction Control methods based on modified neutral net | |
| Acharya et al. | A comparison of regression models for prediction of graduate admissions | |
| CN103105246A (en) | Greenhouse environment forecasting feedback method of back propagation (BP) neural network based on improvement of genetic algorithm | |
| KR102139358B1 (en) | Process control method using machine learning-based platform, computer program medium for performing same, and process control appratus | |
| CN108932671A (en) | A kind of LSTM wind-powered electricity generation load forecasting method joined using depth Q neural network tune | |
| CN110046710A (en) | A kind of the nonlinear function Extremal optimization method and system of neural network | |
| CN111612125B (en) | Novel HTM time pool method and system for online learning | |
| CN102629104A (en) | Calcination predictive control system and method for rotary cement kiln | |
| CN104636801A (en) | Transmission line audible noise prediction method based on BP neural network optimization | |
| JP2005521167A (en) | Improving the performance of an artificial neural network model in the presence of mechanical noise and measurement errors | |
| CN103971160A (en) | Particle swarm optimization method based on complex network | |
| CN109165730A (en) | State quantifies network implementation approach in crossed array neuromorphic hardware | |
| CN105487376A (en) | Optimal control method based on data-driven single network structure | |
| JP7404559B2 (en) | Soft measurement method for industrial processes based on federated incremental stochastic configuration networks | |
| CN109514553A (en) | A kind of method, system and the equipment of the mobile control of robot | |
| CN102183802A (en) | Short-term climate forecast method based on Kalman filtering and evolution modeling | |
| CN107194460A (en) | The quantum telepotation recurrent neural network method of Financial Time Series Forecasting | |
| CN108717506A (en) | A method of prediction coke hot strength | |
| CN105719101A (en) | Photovoltaic system power generation power prediction method | |
| CN102393645A (en) | Control method of high-speed electro-hydraulic proportional governing system | |
| CN102663493A (en) | Delaying nerve network used for time sequence prediction | |
| CN109408896B (en) | Multi-element intelligent real-time monitoring method for anaerobic sewage treatment gas production | |
| CN114091745A (en) | Industry power consumption prediction method based on improved multi-storage pool echo state network | |
| CN101833281A (en) | Control method for saving energy of aeration in sewage treatment | |
| CN116700003A (en) | Method and system for constructing reinforcement learning environment by using process industry historical data |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170908 |
|
| RJ01 | Rejection of invention patent application after publication |