CN1598719A - Nerve network optimization controller and control method - Google Patents
Nerve network optimization controller and control method Download PDFInfo
- Publication number
- CN1598719A CN1598719A CN 200410009608 CN200410009608A CN1598719A CN 1598719 A CN1598719 A CN 1598719A CN 200410009608 CN200410009608 CN 200410009608 CN 200410009608 A CN200410009608 A CN 200410009608A CN 1598719 A CN1598719 A CN 1598719A
- Authority
- CN
- China
- Prior art keywords
- neural network
- controller
- training
- control
- network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Landscapes
- Feedback Control In General (AREA)
Abstract
The invention discloses a neural network optimizing controlling equipment and method. The invention includes controller neural network which is used to calculate optimizing control signal serial according to complicated process object output signal serial; training structure equipment which is used to transform optimizing control rule integrated problem to combining network training problem, composed by controller neural network and complicated process object emulator neural network; training arithmetic equipment, which deducts jointing equal error calculating formula among different controller neural networks and complicated process object emulator neural networks in the combining network, and is used to train the combining network to get optimized controller neural network. The control method of the invention sets the consistency function of training network output and perfect output as guideline function of optimizing control, the invention has the excellence of achieving multiobjective control to complicated process objects which have no precise mathematical model, under restricted controlling condition, with intelligent method.
Description
Affiliated technical field
The present invention relates to a kind of Neural Network Optimization controller and control method.
Background technology
ANN (Artificial Neural Network) Control belongs to Based Intelligent Control.Its physics realization is simple, compares with other type Based Intelligent Control, and is comparatively practical, succeedd in inverted pendulum control, mechanical arm control and underwater remote-control robot longitudinal attitude are regulated at present and used.But, be not reported as application such as industrial process controls in the complex process control that lacks mathematical model.
One of ANN (Artificial Neural Network) Control representative result is Nguyen1990 (Derrick H, Nguyen, et al.Neural Networks for Self-learning Control Systems.IEEE Cont.Sys.Mag.1990, (4): 18-23) institute's extracting method.This article adopts a Multi-layered Feedforward Networks identification controlled device, adopt another neural network as controller, by Learning Control process simulator neural network training controller, solve the final state control problem in the optimal control, and provide the truck positioning control simulation result that falls back.This thought has certain reference value for solving complex process object optimal control problem, because the complex process object generally lacks mathematical models, makes optimal control to realize.But, adopt the method to solve complex process control, need to solve following problem: 1) all to be based on state-space model definite for nerve network controller training structure and algorithm, the claimed condition variable can be measured, but, in general complex process control, state variable or do not exist or can't measure, and mainly use input; 2) do not consider that there is restraint condition in controlled quentity controlled variable, and control variable there is constraint in the complex process control; 3) only solve single goal optimal control problem, and also needed to consider the multiobjective optimal control problem in the complex process control.
Summary of the invention:
At the shortcoming that exists in the above-mentioned existing method, problem to be solved by this invention provides a kind of Neural Network Optimization controller and the control method based on input, realizes retraining multiobjective optimal control under the control action for the complex process object.
The technical solution adopted for the present invention to solve the technical problems is:
A kind of Neural Network Optimization controller, comprise: controller neural network device, training structure device, training algorithm device, the controller neural network device is used for calculating the optimal control burst according to complex process object output signal sequence, make the complex process object under this control signal sequence effect, the control effect is optimized; The training structure device is used for optimizing control rule synthtic price index is converted into the combinational network training problem of being made up of controller neural network and complex process process simulator neural network; The training algorithm device is used to train described combinational network.
A kind of Neural Network Optimization control method comprises:
The controller neural network is used for calculating the optimal control burst according to complex process object output signal sequence;
Training structure is used for optimizing control rule synthtic price index is converted into the combinational network training problem of being made up of controller neural network and complex process process simulator neural network;
Junction equivalent error computing formula between different controller neural networks and the complex process process simulator neural network is used to train described combinational network in the training algorithm, derivation combinational network, and the controller neural network is optimized;
Described training structure is by being divided at least 2 time quantums with the whole control process time, and the neural network of at least 2 time quantums is coupled together the formation combinational network by the input/output signal corresponding relation;
Described training algorithm comprises that combinational network output is weighed function with desirable output consistance and set; Controller neural network output layer neuron operation function is set; Interconnected equivalent error calculates between controller neural network and the described complex process process simulator neural network.
The present invention has following advantage:
1, Neural Network Optimization controller of the present invention is applicable to control complex process object, because the Neural Network Optimization controller is based on input and generates control signal, realize optimal control by output feedback rather than feedback of status, existence spatial model not necessarily for the complex process object generally, but always can this object be described with input; In addition, in general the complex process object does not exist mathematical models, so just can't realize traditional optimal control based on the controlled device mathematical models, controller of the present invention does not need mathematical models, as long as an identification is provided the complex process object promptly trained successful process simulator neural network, just can realize optimal control.
2, the present invention realizes the single goal optimal control with intelligent method complex optimum control law.By the whole control process time is divided into a plurality of time quantums, dynamic element in the relation between relation and the output feedback controller input and output between the input and output of reduction complex process object, utilize Multi-layered Feedforward Networks to show described input/output relation then, the neural network of different time unit is coupled together the formation combinational network by the input/output signal corresponding relation, with described combinational network performance optimization control procedure, network output during with neural metwork training is weighed the target function form that function is set at optimal control with desirable output consistance, so normally difficult optimizing control rule synthtic price index is converted into the combinational network training problem dexterously, further derive combinational network equivalent error computing formula, having solved the combinational network training problem is combinational network middle controller neural metwork training problem, by the neural metwork training control law that is optimized; Compare with the means complex optimum control law of analyzing with traditional, this is by the intelligent means control law that is optimized automatically.
3, realizing on the single goal optimal control basis, network output when the present invention passes through to revise neural metwork training is weighed function with the desirable consistance of exporting, embody the multiobjective optimal control target function, revise controller neural network output layer neuron operation function, embody controlled quentity controlled variable and have restraint condition, derive combinational network equivalent error computing formula then, solved the have an appointment optimal control problem of beam control system of more complicated multiple goal.
Description of drawings
Below in conjunction with accompanying drawing the present invention is described in further detail,
Fig. 1 is the induction heating Neural Network Optimization control system controller neural network structure figure of the preferred embodiment of the present invention.
Fig. 2 is the induction heating Neural Network Optimization control system combinational network training structure figure of the preferred embodiment of the present invention.
Fig. 3 is that the induction heating temperature Neural Network Optimization control system of the preferred embodiment of the present invention is formed synoptic diagram.
Fig. 4 is the induction heating temperature Neural Network Optimization control system schematic diagram of the preferred embodiment of the present invention.
Fig. 5 is that circuit diagram is partly revised in the induction heating temperature Neural Network Optimization control system intermediate frequency power supply Power Regulation of the preferred embodiment of the present invention.
Fig. 6 is the induction heating temperature Neural Network Optimization control system combinational network training process flow diagram of the preferred embodiment of the present invention.
Fig. 7 is the induction heating temperature Neural Network Optimization control system combinational network equivalent error calculation flow chart of the preferred embodiment of the present invention.
Fig. 8 is the induction heating temperature Neural Network Optimization control system control flow chart of the preferred embodiment of the present invention.
Fig. 9 is the induction heating temperature Neural Network Optimization control result of the preferred embodiment of the present invention.
Figure 10 is a Neural Network Optimization control method process flow diagram of the present invention.
Figure 11 is Neural Network Optimization controller architecture figure of the present invention.
Embodiment
Embodiment 1: a kind of Neural Network Optimization control method comprises:
The controller neural network is used for calculating the optimal control burst according to complex process object output signal sequence;
Training structure is used for optimizing control rule synthtic price index is converted into the combinational network training problem of being made up of controller neural network and complex process process simulator neural network;
Junction equivalent error computing formula between different controller neural networks and the complex process process simulator neural network is used to train described combinational network in the training algorithm, derivation combinational network, and the controller neural network is optimized;
Described training structure is by being divided at least 2 time quantums with the whole control process time, and the neural network of at least 2 time quantums is coupled together the formation combinational network by the input/output signal corresponding relation;
Described training algorithm comprises that combinational network output is weighed function with desirable output consistance and set; Controller neural network output layer neuron operation function is set; Interconnected equivalent error calculates between controller neural network and the described complex process process simulator neural network.
The said method flow process as shown in figure 10.A kind of Neural Network Optimization controller architecture as shown in figure 11, C wherein
0, C
1, C
2, C
TRepresent that respectively Unit 0 is to T cell controller neural network, u
0, u
1, u
2, u
TRepresent the optimal control signal of Unit 0 respectively, y to the output of T cell controller neural network
p 0The initial output valve of expression complex process object, y
p 1, y
p 2, y
p T, y
p T+1Represent that respectively Unit 0 is to the output valve of T unit complex process object under the control signal effect, y
dBe illustrated under the effect of optimal control signal complex process object output expectation value.
The first step: the control procedure time is subdivided into a plurality of time quantums, as is divided at least 2 time quantums, require and increase controlled quentity controlled variable resolution with the dynamic that reduces the complex object modeling; To the time quantum characteristic modeling of complex object and generate optimizing control rule, be called unit networks with neural network; Described combinational network device is to be connected to form by a plurality of or at least 2 unit networks, and described unit networks number equals the time quantum number that control procedure comprises.
Second step: described unit networks device comprises a described complex process process simulator neural network E
k, be used to represent the input-output characteristic of the complex process object in this time quantum:
Wherein k represents unit time sequence number, y
k P, y
K-1 P, y
K-n P, u
K-1, u
K-mBe respectively the complex process object in the output signal of k, k-1, k-n time quantum and the control signal that the complex process object is applied at k-1, k-m time quantum, n, m are respectively the order of output signal and control signal; Described complex process process simulator neural network device output signal is y
k, be the output signal y of complex process object at identical time quantum
k PSimulation value, input layer is the occupy-place unit, the input layer number equals m+n, the input signal end is y
K-1..., y
K-n, u
K-1..., u
K-mAfter described complex process process simulator neural network device adopted complex process object phase step response signals to finish training, auxiliary described controller neural network device was trained.
The 3rd step: CONTROLLER DESIGN neural network C
k, be used to calculate k time quantum optimal control amount.As shown in Figure 1, described controller neural network device, input layer are the occupy-place unit, and the hidden neuron action function is the Sigmoid function, and output layer neuron operation function is a linear function; Described input layer input end connects described complex process process simulator neural network output signal end when described controller neural network device is trained, be designated as y
k, be used for training by the error anti-pass, when implementing control, described controller neural network device connects complex process object output signal end, be designated as y
k P, referring to y among Figure 11
p 0, y
p 1, y
p 2, y
p T, be used for described complex process object output signal is carried out neural computing; Described output layer neuron output signal is designated as u
k, when training, described controller neural network device is used to produce forward signal, when described controller neural network device is implemented control, be used to control described complex process object, referring to u among Figure 11
0, u
1, u
2, u
T
The 4th step: different time unit networks E
k, C
kConstitute combinational network.
Described complex process process simulator neural network device E
kOutput signal y
k, link to each other input signal end y with the unit networks middle controller neural network device input end of next time quantum
K-1..., y
K-n, u
K-1..., u
K-m, respectively with other unit networks in have same tag in complex process process simulator neural network device output signal end and the controller neural network device output signal end signal end be connected.
The 5th step: with combinational network equivalent error formula training controller network; Or with error anti-pass (BP) algorithm training controller network, the equivalent error of described back-propagation algorithm calculates and comprises: the inner equivalent error that connects of single described controller neural network device or described complex process process simulator neural network device calculates, and interconnected equivalent error calculates between described controller neural network device and the described complex process process simulator neural network device.
Interconnected equivalent error computing formula comprises that there are the computing formula under the optimal control situation of constraint in computing formula under the single goal optimal control situation and multiple goal and controlled quentity controlled variable between described controller neural network device and the described complex process process simulator neural network device.
Computing formula under the described single goal optimal control situation is:
Wherein, T is for implementing control procedure time corresponding unit number maximal value, and requiring had been the time quantum number variable greater than m but also greater than nh not only, and variation range is: 0～Ty
dBeing optimal control final state desired value, also is the desired output of the described combinational network device of training
δ
T E(y
T+1) be T time quantum complex process process simulator neural network device output signal end points y
T+1Equivalent error
δ
T-h C(u
T-h) be T-h time quantum controller neural network device output signal end points u
T-hEquivalent error
δ
T-i E(u
T-h) be T-i time quantum complex process process simulator neural network device input signal end points u
T-hEquivalent error
δ
T-h+i E(u
T-h) be T-h+i time quantum complex process process simulator neural network device input signal end points u
T-hEquivalent error
δ
T-h E(y
T-h+1) be T-h time quantum complex process process simulator neural network device input signal end points y
T-h+1Equivalent error
δ
T-h+1 C(u
T-h+1) be T-h+1 time quantum controller neural network device output signal end points u
T-h+1Equivalent error
δ
T-i E(y
T-h+1) be T-i time quantum complex process process simulator neural network device input signal end points y
T-h+1Equivalent error
δ
T-h+i E(y
T-h+1) be T-h+i time quantum complex process process simulator neural network device input signal end points y
T-h+1Equivalent error
Begin to calculate by h=0 during training, embody the error anti-pass; The computing formula of described single goal optimal control be described controller neural network device and described complex process process simulator neural network device input layer be occupy-place unit condition, described controller neural network device with described complex process process simulator neural network device between be connected to each other that being directly links to each other and promptly be connected weights equal 1 and network output during with neural metwork training weigh function with desirable output consistance and be set at
Condition under, deriving according to the derivative rule of demanding perfection draws; After training described combinational network promptly to train wherein controller neural network device success according to the computing formula of described single goal optimal control, use this controller neural network device control complex process object, can realize that performance index function is
The single goal optimal control.
Described multiple goal and controlled quentity controlled variable exist the computing formula under the optimal control situation that retrains to be:
Wherein, α, β are the weighting factor of many index in the multiobjective optimal control performance index function
A, B are respectively the lower bound and the upper bound of constraint controlled quentity controlled variable
u
hBe h time quantum controller neural network device output signal value
u
T-hBe T-h time quantum controller neural network device output signal value
Begin to calculate by h=0 during training, embody the error anti-pass;
For there are the computing formula of the optimal control of constraint in derive described multiple goal and controlled quentity controlled variable, need network output when neural network device trained to weigh function and be set at desirable output consistance
In addition, a preferred version as there is constraint with mathematical form performance controlled quentity controlled variable is set at described controller neural network device output layer neuron operation function
Wherein exp () is an exponential function, N
Net(h) be the input signal values of h time quantum controller neural network device output neuron;
After existing the computing formula of constrained optimization control to train described combinational network promptly to train wherein controller neural network device success according to described multiple goal and controlled quentity controlled variable, use this controller neural network device control complex process object, can realize that performance index function is
Multiobjective optimal control, and controlled quentity controlled variable satisfies following constraint
A≤u
h≤B
The 6th step: train successfully and judge, seven steps of successful operation to the; Get nowhere and ran to for the 5th step.
The 7th step: by the controller neural network of each time quantum control law that is optimized.
The 8th step: use described controller neural network device and implement optimal control, referring to Figure 11.
Embodiment 2: induction heating relates to physical processes such as magnetic, electricity, heat conduction, because of the mechanism complexity is difficult to propose accurate mathematical model, can't realize traditional optimal control, adopt the Neural Network Optimization control method, can realize the control of heat temperature raising process optimization, control system as shown in Figure 3.
Design with reference to Fig. 1-Fig. 8 according to induction heating Neural Network Optimization control system of the invention process.Determine that according to controlled device the input structure is n=1, m=3, process simulator network are 4 inputs, 1 output nerve networks; The controller neural network structure is π (1 * 4 * 1), as shown in Figure 1; Control procedure time quantum number is 5.Combinational network and training structure thereof as shown in Figure 2, C wherein
0, C
1, C
2, C
3, C
4Represent that respectively 0 time quantum is to 4 time quantum controller neural networks, E
0, E
1, E
2, E
3, E
4Represent that respectively 0 time quantum is to 4 time quantum process simulator neural networks, u
0, u
1, u
2, u
3, u
4Represent that respectively 0 time quantum is to 4 time quantum controller neural network output signals, y
1, y
2, y
3, y
4, y
5Represent that respectively 0 time quantum is to 4 time quantum process simulator neural network output signals, y
0Be process simulator neural network output signal initial value, u
-1, u
-2Expression controller neural network output signal initial value is set to zero, and annexation is represented with the connection solid line between corresponding signal between different time unit process simulator neural network and the controller neural network input/output signal.
Referring to Fig. 6, Fig. 6 is an induction heating temperature Neural Network Optimization control system combinational network training process flow diagram.
The first step: it is 0 that the loop variable initial value is set;
Second step: if cycle of training, number did not reach requirement, then continue training, enter next step;
The 3rd step: combinational network carries out forward calculation by 0 time quantum to 4 time quantums;
The 4th step to the 8th step: under n=1, m=3, T=4 condition, (8) parameter functional value by formula;
The 9th step: if the target function value smaller or equal to predetermined value, then jumped to for the 20 step;
The tenth step: calculate if the target function value, is then carried out the combinational network error back propagation greater than predetermined value;
The 11 step: combinational network middle controller neural network is calculated the weights adjustment amount;
The 12 step to the 14 step: do not knock " ALT_C " key if detect, then jumped to for the 20 step;
The 15 step to the 17 step: knock " ALT_C " key if detect, then point out neural metwork training to interrupt and reason, show the periodicity and the target function value of having trained, whether prompting continues training;
The 18 step:, then jumped to for the 23 step if do not continue training;
The 19 step:, then enter next step if continue training;
The 20 step: after loop variant values adds 1,, then jumped to for the 3rd step if less than counting maximal value cycle of training;
The 21 step: after loop variant values adds 1,, then will deposit data file in, and use during in order to enforcement control through the controller neural network weight that training obtains if more than or equal to counting maximal value cycle of training;
The 22 step: prompting: reached maximum number cycle of training and shown the target function value;
The 23 step: termination routine
Referring to Fig. 7, Fig. 7 is an induction heating temperature Neural Network Optimization control system combinational network equivalent error calculation flow chart.
The first step: under n=1, m=3, T=4 condition, by formula (5) calculate 4 time quantum process simulator neural network output terminal equivalent errors;
Second step: calculate 4 time quantum process simulator neural network input layer equivalent errors by error back propagation;
Whether the 3rd step to the 4th step: exist under the constraint condition in n=1, m=3, T=4 and controlled quentity controlled variable, by formula (6) calculate 4 time quantum controller neural network output terminal equivalent errors;
The 5th step: calculate 4 time quantum controller neural network input layer equivalent errors by error back propagation;
The 6th step: under n=1, m=3, T=4 condition, by formula (7) calculate 3 time quantum process simulator neural network output terminal equivalent errors;
The 7th step: calculate 3 time quantum process simulator neural network input layer equivalent errors by error back propagation;
Whether the 8th step to the 9th step: exist under the constraint condition in n=1, m=3, T=4 and controlled quentity controlled variable, by formula (6) calculate 3 time quantum controller neural network output terminal equivalent errors;
The tenth step: calculate 3 time quantum controller neural network input layer equivalent errors by error back propagation;
The 11 step to the 18 step: adopt loop program, calculate the process simulator neural network and the controller neural network output terminal equivalent error of 2 time quantums, 1 time quantum and 0 time quantum successively.
Referring to Fig. 8, Fig. 8 is an induction heating temperature Neural Network Optimization control system control flow chart.
The first step: it is 0 that the loop variable initial value is set;
Second step: if loop variable less than 5, then enters next step;
The 3rd step: by starting the work of mould/number (A/D) converter, sampling heated parts surface temperature value;
The 4th step: according to the sample temperature value, by the corresponding time quantum controller of loop variable neural network forward calculation, controlled quentity controlled variable is optimized;
The 5th step: the optimal control amount is outputed to D/A (D/A) converter, the control temperature;
The 6th step to the 8th step: wait for next time quantum;
The 9th step: after loop variant values adds 1,, then jumped to for the 3rd step if less than 5;
The tenth step: after loop variant values added 1, if more than or equal to 5, the heated parts surface temperature of then sampling value also showed;
The 11 step: control program finishes.
Referring to Fig. 4, Fig. 4 is an induction heating temperature Neural Network Optimization control system schematic diagram.Wherein the thyristor intermediate frequency electric source model is KGPS250-8.0, rated power 250Kw, frequency 8Khz; Inductor (copper, single turn) internal diameter 56mm, height 100mm; Heated parts (stainless-steel tube) external diameter 22mm, wall thickness 3mm; Infrared thermometer unit model is 0Q06-11C, and specification is 35P1200, and the lens opening is 35mm, it is 2M-Q06-11C that adapted blocks electronic signal process unit model more, calibration number is Q202C, selects for use " output 1 " end to be the thermometric output terminal, links to each other with the A/D converter signal input part; A/D, D/A converter adopt PC bus A/D plate SS34; The intermediate frequency power supply power regulation circuit that process is transformed is referring to Fig. 5, and wherein the optimal control numerical quantity is converted into voltage signal through D/A (D/A), through amplifying circuit, and control controllable silicon trigger angle.The heated parts surface temperature is detected by the infrared thermometer unit, through too much blocking electronic signal process unit amplification becoming standard electric signal, be converted into digital quantity through A/D and send into industrial control computer, Neural Network Optimization control algolithm by the software realization, by the temperature digital amount is handled the control signal that is optimized, be converted into voltage signal through D/A, regulate wave frequency and power that intermediate frequency power supply produces, reach control heated parts temperature purpose.
At the intensification accuracy, accurately add different situations such as energy-conservation, the beam control system of having an appointment, the control experiment that heats up, experiment parameter is:
Test 1 y
0=626 ℃, y
d=680 ℃, α=1, β=0;
Test 2 y
0=626 ℃, y
d=680 ℃, α=1, β=50;
Test 3 y
0=626 ℃, y
d=680 ℃, α=1, β=50.B=0.2V, A=1.9V.
Experimental result is:
Test 3 final state temperature deviations :-0.26 ℃, energy: 2.624.
The final state temperature deviation is actual intensification and gives the poor of fixed temperature.
Controlled quentity controlled variable and temperature value change curve are referring to Fig. 9 in the control experiment, and figure intermediate cam point is the control value, and circle points is a temperature value.
Claims (6)
1. Neural Network Optimization controller, it is characterized in that: comprising: controller neural network device, training structure device, training algorithm device, the controller neural network device is used for calculating the optimal control burst according to complex process object output signal sequence, make the complex process object under this control signal sequence effect, the control effect is optimized; The training structure device is used for optimizing control rule synthtic price index is converted into the combinational network training problem of being made up of controller neural network and complex process process simulator neural network; The training algorithm device is used to train described combinational network.
2, according to described a kind of Neural Network Optimization controller of claim 1, it is characterized in that: described training structure device is by being divided at least 2 time quantums with the whole control process time, and the neural network of different time unit is coupled together the formation combinational network by the input/output signal corresponding relation; Described training algorithm device comprises that combinational network output is weighed function with desirable output consistance and set; Controller neural network output layer neuron operation function is set; Interconnected equivalent error calculates between controller neural network and the described complex process process simulator neural network.
3, a kind of Neural Network Optimization control method comprises: the controller neural network is used for calculating the optimal control burst according to complex process object output signal sequence; Training structure is used for optimizing control rule synthtic price index is converted into the combinational network training problem of being made up of controller neural network and complex process process simulator neural network; Junction equivalent error computing formula between different controller neural networks and the complex process process simulator neural network is used to train described combinational network in the training algorithm, derivation combinational network, and the controller neural network is optimized; It is characterized in that: described training structure is by being divided at least 2 time quantums with the whole control process time, and the neural network of different time unit is coupled together the formation combinational network by the input/output signal corresponding relation; Described training algorithm comprises that combinational network output is weighed function with desirable output consistance and set; Controller neural network output layer neuron operation function is set; Interconnected equivalent error calculates between controller neural network and the described complex process process simulator neural network.
4. according to described a kind of Neural Network Optimization control method of claim 3, it is characterized in that described combinational network output is weighed function with desirable output consistance and set, the functional form of setting is
5. according to described a kind of Neural Network Optimization control method of claim 3, it is characterized in that controller neural network output layer neuron operation function is set, the functional form of setting is
6. according to described a kind of Neural Network Optimization control method of claim 3, it is characterized in that, described equivalent error computing formula comprises that there are the computing formula under the optimal control situation of constraint in computing formula under the single goal optimal control situation and multiple goal and controlled quentity controlled variable, and the computing formula under the described single goal optimal control situation is:
Described multiple goal and controlled quentity controlled variable exist the computing formula under the optimal control situation that retrains to be:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2004100096087A CN1324417C (en) | 2004-09-27 | 2004-09-27 | Nerve network optimization controller and control method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2004100096087A CN1324417C (en) | 2004-09-27 | 2004-09-27 | Nerve network optimization controller and control method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN1598719A true CN1598719A (en) | 2005-03-23 |
| CN1324417C CN1324417C (en) | 2007-07-04 |
Family
ID=34662545
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNB2004100096087A Expired - Fee Related CN1324417C (en) | 2004-09-27 | 2004-09-27 | Nerve network optimization controller and control method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN1324417C (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101816821A (en) * | 2010-05-26 | 2010-09-01 | 天津大学 | Walking aid functional electrical stimulation precision control method based on ant colony fuzzy controller |
| CN103064285A (en) * | 2012-12-29 | 2013-04-24 | 杭州电子科技大学 | Heat pump heating multi-objective optimization control method based on model |
| CN106896722A (en) * | 2017-03-29 | 2017-06-27 | 郑州轻工业学院 | Adoption status feeds back the hypersonic vehicle composite control method with neutral net |
| CN107891424A (en) * | 2017-11-10 | 2018-04-10 | 浙江科技学院 | A kind of finite time Neural network optimization for solving redundant mechanical arm inverse kinematics |
| CN109143870A (en) * | 2018-10-23 | 2019-01-04 | 宁波溪棠信息科技有限公司 | A kind of control method of multiple target task |
| CN113110050A (en) * | 2021-04-12 | 2021-07-13 | 山东理工大学 | Temperature control method for electromagnetic herbal medicine roaster |
| CN113267993A (en) * | 2021-04-22 | 2021-08-17 | 上海大学 | Network training method and device based on collaborative learning |
-
2004
- 2004-09-27 CN CNB2004100096087A patent/CN1324417C/en not_active Expired - Fee Related
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101816821A (en) * | 2010-05-26 | 2010-09-01 | 天津大学 | Walking aid functional electrical stimulation precision control method based on ant colony fuzzy controller |
| CN101816821B (en) * | 2010-05-26 | 2012-08-15 | 天津大学 | Walking aid functional electrical stimulation precision control method based on ant colony fuzzy controller |
| CN103064285A (en) * | 2012-12-29 | 2013-04-24 | 杭州电子科技大学 | Heat pump heating multi-objective optimization control method based on model |
| CN103064285B (en) * | 2012-12-29 | 2015-08-26 | 杭州电子科技大学 | A kind of heat pump heating multiobjective optimization control method based on model |
| CN106896722A (en) * | 2017-03-29 | 2017-06-27 | 郑州轻工业学院 | Adoption status feeds back the hypersonic vehicle composite control method with neutral net |
| CN106896722B (en) * | 2017-03-29 | 2019-11-29 | 郑州轻工业学院 | The hypersonic vehicle composite control method of adoption status feedback and neural network |
| CN107891424A (en) * | 2017-11-10 | 2018-04-10 | 浙江科技学院 | A kind of finite time Neural network optimization for solving redundant mechanical arm inverse kinematics |
| CN109143870A (en) * | 2018-10-23 | 2019-01-04 | 宁波溪棠信息科技有限公司 | A kind of control method of multiple target task |
| CN109143870B (en) * | 2018-10-23 | 2021-08-06 | 宁波溪棠信息科技有限公司 | Multi-target task control method |
| CN113110050A (en) * | 2021-04-12 | 2021-07-13 | 山东理工大学 | Temperature control method for electromagnetic herbal medicine roaster |
| CN113267993A (en) * | 2021-04-22 | 2021-08-17 | 上海大学 | Network training method and device based on collaborative learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN1324417C (en) | 2007-07-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ayop et al. | A comprehensive review on photovoltaic emulator | |
| CN1173240C (en) | Field based process control system with auto-tuning | |
| CN1324417C (en) | Nerve network optimization controller and control method | |
| CN110707763B (en) | AC/DC power distribution network load prediction method based on ensemble learning | |
| CN112620652B (en) | Self-adaptive control system and method for electric arc additive manufacturing process | |
| CN109343352B (en) | Temperature control method and system based on time optimal control | |
| CN104890205A (en) | Barrel temperature control method of injection molding machine | |
| KR20160013012A (en) | Methods for ascertaining a model of a starting variable of a technical system | |
| CN107745785A (en) | A kind of optimal Calculation of stowage on board method of the loading instrument of bulk freighter and oil carrier | |
| CN100435611C (en) | Self-adaptive light-source illuminating system and method for paster | |
| CN113158368A (en) | Prediction and control method and device for deformation of material-increasing and material-decreasing composite manufacturing part | |
| CN1694107A (en) | Material data correction method and its system | |
| CN1525153A (en) | Flexible measuring method for overflow particle size specification of ball mill grinding system | |
| CN111125884B (en) | Magnesium alloy composition and heat treatment process optimization system for intelligent manufacturing | |
| CN102347751A (en) | Digitalized control method and device for pulse power source | |
| CN1510360A (en) | Comprehensive control method for heating furnace outlet temperature and device thereof | |
| CN108599240A (en) | The Wind turbines converter controller parameter tuning method of meter and small signal stability | |
| CN111009919B (en) | Photovoltaic climbing rate control method and device based on active control | |
| CN104020672A (en) | Intelligent control method of DC hollow electrode calcium carbide furnace | |
| CN1913320A (en) | Digital controlled inverter and its control method | |
| CN116300422A (en) | Hydropower unit control optimization method and terminal | |
| CN113779733A (en) | Photovoltaic module model parameter hybrid optimization identification method | |
| CN1746604A (en) | Intelligent optimizing set during process of shaft roasting | |
| CN106685228A (en) | Closed loop control method and direct current method and direct current converter | |
| CN113353297A (en) | Rapid and uniform static load heating device of high-speed aircraft and control method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| C17 | Cessation of patent right | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20070704 Termination date: 20100927 |