CN105182747A - Automatic control method in large lag time system - Google Patents
Automatic control method in large lag time system Download PDFInfo
- Publication number
- CN105182747A CN105182747A CN201510540026.XA CN201510540026A CN105182747A CN 105182747 A CN105182747 A CN 105182747A CN 201510540026 A CN201510540026 A CN 201510540026A CN 105182747 A CN105182747 A CN 105182747A
- Authority
- CN
- China
- Prior art keywords
- algorithm
- model
- output
- input
- time system
- 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
Landscapes
- Feedback Control In General (AREA)
Abstract
The invention discloses an automatic control method in a large lag time system, comprising a delay pre-estimation method which proceeds as the following steps of adopting a Neuron network to establish a non-linear model, wherein the nonlinear model comprises a Neuron node and a passageway constituted by the Neuron nodes, each nerve cell node is provided with a threshold, and each pathway is provided with a weight, S2 adopting current multiple groups input data and output data, performing training on a nonlinear model through an inversion calculation to obtain a threshold and a weight which accord with the relation of the input and the output so as to obtain a trained estimation model, S3, utilizing an estimation model to perform estimation on the output quantity of the controller , wherein each node on the passageway performs weight accumulation on each node through input quantity, adds the threshold and multiplies with the weight to obtain delay output. The algorithm model in the invention obtains a better effect, the system will tend to be stable. Except for the processing under special condition, the manual operation can be basically replaced.
Description
Technical field
The present invention relates to a kind of autocontrol method, in particular, relate to the autocontrol method in a kind of Correction for Large Dead Time System.
Background technology
Along with society enters the information age, the requirement of industrial circle to automaticity is more and more higher.But in a lot of industrial trade, although automation equipment is more and more advanced, in fact, the use of computing machine, network far can not become robotization, and display is only instead of traditional analogue instrument, and keyboard and mouse instead of shift knob, especially in China, this kind of phenomenon is especially obvious.Because a variety of causes cannot accomplish effective robotization and intelligentized control method, so need skilled operator to carry out manual control.More obvious in traditional industries, the such as industry such as cement, iron and steel is very outstanding.Relative to the production line of advanced country, in the enterprise of China, no matter be equipment, or control theory have very large shortcoming.For the Automated condtrol of domestic this kind of traditional industries, we have done a large amount of work, below the main control algolithm theory and practice discussed in Correction for Large Dead Time System.
Time lag system refers to the system that the signal transmission at a place or a few place in system postpones if having time, and such system is extensively deposited in the industrial production.The feature of this type systematic is, through the wait of certain hour, could will reflect its control effects in systems in which after controlling output, concrete time lag embodies in different situations and changes to some extent.Theory of Automatic Control develops into now, has a lot of algorithm, and algorithm the most frequently used, the most classical in process control should be pid algorithm.Pid algorithm is that the ratio (P) of the deviation in control objectives, integration (I) and differential (D) control, pid algorithm, very practical during the course, effect is also good, but in nonlinear big-lagged system, defect also clearly.
In Correction for Large Dead Time System, there is delay different in size between each controlled volume and the adjustment amount affecting it, be not adjust in the very short time of controlled quentity controlled variable, controlled variable just changes, but this impact just shows over time, become.For such time delay situation, in order to ensure system stability, only all smaller in each parameter of PID controller, control cycle extends, just can make system stability when gain is lower, but regulation time must be caused so elongated, and antijamming capability declines; Otherwise, system will be made to occur concussion instability if increase parameter.In this case be difficult to find between rapidity and stability the equilibrium point that desirable.Meanwhile, the tuning process of pid parameter also can be very difficult in this case, and the controller's effect of parameter of having adjusted is also unsatisfactory.Relation between the adjustment amount that most of controlled quentity controlled variable is corresponding with it is in addition not a kind of linear relation, but also is a kind of result of multifactor joint effect, and at this moment, how setting up effective relational model neither a nothing the matter.
Summary of the invention
Goal of the invention: the object of the invention is for the deficiencies in the prior art, provides a kind of and can ensure that the stability of system can ensure again the autocontrol method in the Correction for Large Dead Time System of antijamming capability.
Technical scheme: the autocontrol method in a kind of Correction for Large Dead Time System of the present invention, adopt Smith algorithm to carry out process to the input quantity collected to realize automatically controlling, described Smith algorithm comprises steering order output algorithm and postpones predictive algorithm, and described delay predictive algorithm is carried out as follows:
S1, employing neuroid set up nonlinear model, described nonlinear model comprises input layer, hidden layer and output layer, above-mentioned every layer of path including neuron node and be made up of neuron node, each neuron node is arranged threshold values v, every bar path arranges weight w;
S2, gather existing many group input data x and export data y, described nonlinear model is trained by the algorithm of anti-phase calculating at described hidden layer and output layer, obtain the threshold values v and the weight w that meet input and output relation, thus obtain the prediction model of having trained;
S3, the output quantity Ym of prediction model to controller is utilized to estimate, each node on path by input quantity is weighted accumulative add threshold values and then be multiplied by weights export Ek as delay, will postpone to export the input value of Ek input control algorithm as steering order output algorithm next time.
Being further defined to of technical solution of the present invention, the algorithm of the anti-phase calculating of the hidden layer described in step S2 is:
The adjustment amount of weight w is:
Wherein, η is coefficient preset in advance; t
kfor the sample value of model; A2
kfor the calculating output valve of model; F2 ' and f1 ' is the function that falls of preset transport function; W2
kifor the weights that a upper neuron node is connected with this neuron node; Pj is the output sample value of hidden layer;
The adjustment amount of threshold values v is: Δ b1
i=η δ
ij.。
Further, the algorithm of the anti-phase calculating of the output layer described in step S2 is:
The adjustment amount of weight w is: Δ w
ki-η f2 ' (t
k-a2
k) f2a1
i, wherein, η is coefficient preset in advance; The function that f2 ' is preset transport function; t
kfor the sample value of model; A2
kfor the calculating output valve of model; A1
ifor the sample value of the output that forward calculates;
The adjustment amount of threshold values v is: Δ b2
ki=η (t
k-a2
k) f2 ', wherein, η is coefficient preset in advance; The function that f2 ' is preset transport function; t
kfor the sample value of model; A2
kfor the calculating output valve of model.
Further, described control algolithm is pid algorithm or adaptive algorithm.
Further, when the parameter constant of Correction for Large Dead Time System, what postpone predictive algorithm only needs execution step S3; When the parameter of Correction for Large Dead Time System changes, postpone predictive algorithm and re-execute step S1, S2 and S3.
Beneficial effect: the autocontrol method in a kind of Correction for Large Dead Time System provided by the invention, adopt Smith algorithm to carry out process to the input quantity that collects to realize automatically controlling and adopt neuroid to set up nonlinear model to carry out delay and estimate, allow the adjustment of " now ", control the target in " future ", algorithm model achieves good effect in practice, system can tend towards stability, and except process in particular cases, substantially can replace manual operations.
Accompanying drawing explanation
Fig. 1 is the Smith algorithm model in the autocontrol method in a kind of Correction for Large Dead Time System provided by the invention;
Fig. 2 is the neural network algorithm model in the autocontrol method in a kind of Correction for Large Dead Time System provided by the invention;
Fig. 3 is the delay predictive algorithm model in the autocontrol method in a kind of Correction for Large Dead Time System provided by the invention.
Embodiment
Below by accompanying drawing, technical solution of the present invention is described in detail, but protection scope of the present invention is not limited to described embodiment.
Embodiment 1: the present embodiment provides a kind of autocontrol method be applied in the Correction for Large Dead Time System of cement producing line, the temperature of precalciner kiln in cement producing line is utilized to control, autocontrol method adopts Smith algorithm to carry out process to the input quantity collected and realizes automatically controlling, and Smith algorithm comprises steering order output algorithm and postpones predictive algorithm.
As shown in Figure 1, principle is Smith algorithm model:
E2=E1+(Ym-Xm)=E1-Gm(s)*(1-e
-t*s)*U(s);
Wherein E1=R (s)-Y (s); Y (s)=E2 (s) * Gc (s) * G
0(s) * e
-ts;
As t*=t and Gc (s)=G
0time (s), Y (s)=Gc (s) * G
0(s) * E2; Do not postpone.
Arranging model is:
be a second-order system, large multisystem can be represented.
Existing: Xm (s) (as
2+ bs+c)=U_t (s) * (ds+f)
Turning to difference equation is:
And Ym to be Xm postpone backward that t*/T point obtains.
For steering order output algorithm Gc (s), can be pid algorithm, also can be other algorithms (as adaptive algorithms).The input of steering order output algorithm Gc (s) should be E2, E2=E1-Ek.Ek is the output postponing predictive algorithm, and E1 is common input Goal-AI, and the output of controller is that increment or full dose are relevant with algorithms selection.
The model postponing predictive algorithm as shown in Figure 3, carries out as follows:
S1, employing neuroid set up nonlinear model, described nonlinear model comprises input layer, hidden layer and output layer, above-mentioned every layer of path including neuron node and be made up of neuron node, each neuron node is arranged threshold values v, every bar path arranges weight w.
In the industrial stokehold of reality, the factor affecting control objectives is often very many, and effect is non-linear, also impact is had each other, therefore prediction model is set up with formula or matrix very difficult, through a large amount of research experiments of inventor, it is practical that discovery neuroid sets up this nonlinear model, neural network algorithm model as shown in Figure 2, input X and the relation exported between Y are expressed by hidden layer, each node is neuron node, each node there is a threshold values v, each path there is a weight w, each node adds threshold values by weighted input is accumulative, result of calculation is multiplied by weights again and inputs to output unit.Can find out, when the threshold values of the weights on every bar path and each node determines the relation of these input and output, such input quantity X and output quantity Y establish a nonlinear complexity and associate, and this incidence relation is made up of number of nodes and corresponding weights and bias.
S2, the existing many group input data x and output data y of collection, described nonlinear model is trained by the algorithm of anti-phase calculating at described hidden layer and output layer, obtain the threshold values v and the weight w that meet input and output relation, thus obtain the prediction model of having trained.
Weight w on each node and threshold values v need to be decided by a training.Briefly, take out one group of working majority certificate, comprise input and corresponding output, trained by the algorithm of an anti-phase calculating, successively revise, progressively close, finally find out the relation of constrained input.When using, data are inputed to trained neuroid, and the output obtained is exactly the result of neural network forecast.
The algorithm of the anti-phase calculating of above-mentioned hidden layer is:
The adjustment amount of weight w is:
Wherein, η is coefficient preset in advance; t
kfor the sample value of model; A2
kfor the calculating output valve of model; F2 ' and f1 ' is the function that falls of preset transport function; W2
kifor the weights that a upper neuron node is connected with this neuron node; Pj is the output sample value of hidden layer;
The adjustment amount of threshold values v is: Δ b1
i=η δ
ij.。
The algorithm of the anti-phase calculating of above-mentioned output layer is:
The adjustment amount of weight w is: Δ w
ki-η f2 ' (t
k-a2
k) f2a1
i, wherein, η is coefficient preset in advance; The function that f2 ' is preset transport function; t
kfor the sample value of model; A2
kfor the calculating output valve of model; A1
ifor the sample value of the output that forward calculates;
The adjustment amount of threshold values v is: Δ b2
ki=η (t
k-a2
k) f2 ', wherein, η is coefficient preset in advance; The function that f2 ' is preset transport function; t
kfor the sample value of model; A2
kfor the calculating output valve of model.
In Correction for Large Dead Time System, first should determine the time delay between constrained input, input and output correspondence is got up, generate one or more groups sample, train, after training result is satisfied, as prediction model.When using in practice, the result that just can export after the estimated delay time.
Illustrate, injecting coal quantity x, affects temperature y after 10 seconds.So when training sample process, select current injecting coal quantity and the temperature after 10 seconds as a sample, generate one group of sample in such a way, carry out training, storing.When actual estimating, input current injecting coal quantity, the result calculated is exactly the temperature value after 10 seconds, and such prediction device just can normally work.
S3, the output quantity Ym of prediction model to controller is utilized to estimate, each node on path by input quantity is weighted accumulative add threshold values and then be multiplied by weights export Ek as delay, will postpone to export the input value of Ek input control algorithm as steering order output algorithm next time.
We control to have done a large amount of experiments in the temperature of the precalciner kiln of cement producing line with this model, achieve expected effect.Regulate relative to manual PID, this model is quick, and accommodation is comparatively large, and there will not be vibration, control accuracy improves 50% than PID mode.Relative to manual state, with the old operative employee be skilled in technique, the speed of reacting in abnormal cases is slightly slow, but can not judge trend by accident, especially when great fluctuation process, this algorithm model exports can not overshoot arbitrarily, therefore, after a period of time, system can tend towards stability, and is better than manual operations.Relative to normal operations work, this modelling effect is better than craft, and except process in particular cases, substantially desirable work of withholding operates.Except under special failure condition, the control effects of this algorithm model and control accuracy are better than PID and control and hand-guided.
As mentioned above, although represented with reference to specific preferred embodiment and described the present invention, it shall not be construed as the restriction to the present invention self.Under the spirit and scope of the present invention prerequisite not departing from claims definition, various change can be made in the form and details to it.
Claims (5)
1. the autocontrol method in a Correction for Large Dead Time System, adopt Smith algorithm to carry out process to the input quantity collected to realize automatically controlling, described Smith algorithm comprises steering order output algorithm and postpones predictive algorithm, and it is characterized in that, described delay predictive algorithm is carried out as follows:
S1, employing neuroid set up nonlinear model, described nonlinear model comprises input layer, hidden layer and output layer, above-mentioned every layer of path including neuron node and be made up of neuron node, each neuron node is arranged threshold values v, every bar path arranges weight w;
S2, gather existing many group input data x and export data y, described nonlinear model is trained by the algorithm of anti-phase calculating at described hidden layer and output layer, obtain the threshold values v and the weight w that meet input and output relation, thus obtain the prediction model of having trained;
S3, the output quantity Ym of prediction model to controller is utilized to estimate, each node on path by input quantity is weighted accumulative add threshold values and then be multiplied by weights export Ek as delay, will postpone to export the input value of Ek input control algorithm as steering order output algorithm next time.
2. the autocontrol method in a kind of Correction for Large Dead Time System according to claim 1, is characterized in that, the algorithm of the anti-phase calculating of the hidden layer described in step S2 is:
The adjustment amount of weight w is:
wherein, η is coefficient preset in advance; t
kfor the sample value of model; A2
kfor the calculating output valve of model; F2 ' and f1 ' is the function that falls of preset transport function; W2
kifor the weights that a upper neuron node is connected with this neuron node; Pj is the output sample value of hidden layer;
The adjustment amount of threshold values v is: Δ b1
i=η δ
ij.
3. the autocontrol method in a kind of Correction for Large Dead Time System according to claim 1, is characterized in that, the algorithm of the anti-phase calculating of the output layer described in step S2 is:
The adjustment amount of weight w is: Δ w
ki=η f2 ' (t
k-a2
k) f2 ' a1
i, wherein, η is coefficient preset in advance; The function that f2 ' is preset transport function; t
kfor the sample value of model; A2
kfor the calculating output valve of model; A1
ifor the sample value of the output that forward calculates;
The adjustment amount of threshold values v is: Δ b2
ki=η (t
k-a2
k) f2 ', wherein, η is coefficient preset in advance; The function that f2 ' is preset transport function; t
kfor the sample value of model; A2
kfor the calculating output valve of model.
4. the autocontrol method in a kind of Correction for Large Dead Time System according to the arbitrary claim of claim 1-3, is characterized in that, described control algolithm is pid algorithm or adaptive algorithm.
5. the autocontrol method in a kind of Correction for Large Dead Time System according to the arbitrary claim of claim 1-3, is characterized in that, when the parameter constant of Correction for Large Dead Time System, what postpone predictive algorithm only needs execution step S3; When the parameter of Correction for Large Dead Time System changes, postpone predictive algorithm and re-execute step S1, S2 and S3.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510540026.XA CN105182747A (en) | 2015-08-28 | 2015-08-28 | Automatic control method in large lag time system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510540026.XA CN105182747A (en) | 2015-08-28 | 2015-08-28 | Automatic control method in large lag time system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN105182747A true CN105182747A (en) | 2015-12-23 |
Family
ID=54904896
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510540026.XA Pending CN105182747A (en) | 2015-08-28 | 2015-08-28 | Automatic control method in large lag time system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105182747A (en) |
Cited By (1)
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 |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO1999018483A1 (en) * | 1997-10-06 | 1999-04-15 | General Cybernation Group, Inc. | Model-free adaptive process control |
CN1811306A (en) * | 2006-02-22 | 2006-08-02 | 天津大学 | Automatic volume regulating and controlling method for gas-burning machine heat pump |
CN101387867A (en) * | 2008-10-14 | 2009-03-18 | 江苏科技大学 | Parameter controlling method in biofermentation process |
CN103591637A (en) * | 2013-11-19 | 2014-02-19 | 长春工业大学 | Centralized heating secondary network operation adjustment method |
-
2015
- 2015-08-28 CN CN201510540026.XA patent/CN105182747A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO1999018483A1 (en) * | 1997-10-06 | 1999-04-15 | General Cybernation Group, Inc. | Model-free adaptive process control |
CN1811306A (en) * | 2006-02-22 | 2006-08-02 | 天津大学 | Automatic volume regulating and controlling method for gas-burning machine heat pump |
CN101387867A (en) * | 2008-10-14 | 2009-03-18 | 江苏科技大学 | Parameter controlling method in biofermentation process |
CN103591637A (en) * | 2013-11-19 | 2014-02-19 | 长春工业大学 | Centralized heating secondary network operation adjustment method |
Non-Patent Citations (4)
Title |
---|
刘祥博: "闪蒸干燥机控制系统的设计与温度控制方法的研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
李红星 等: "基于自适应Smith预估器的补偿模糊神经网络控制", 《电气自动化》 * |
李钟慎: "加入滞后时间削弱器的大滞后系统的神经网络PID控制", 《电子测量与仪器学报》 * |
邓立广 等: "基于smith预估的水泥分解炉温度MRFAC控制器设计", 《化工自动化及仪表》 * |
Cited By (1)
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 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Lucia et al. | Handling uncertainty in economic nonlinear model predictive control: A comparative case study | |
Savran et al. | A fuzzy model based adaptive PID controller design for nonlinear and uncertain processes | |
CN101751051B (en) | Cement decomposing furnace temperature control method based on constraint smith GPC | |
Wu et al. | Hierarchical optimization of boiler–turbine unit using fuzzy stable model predictive control | |
Würth et al. | Neighboring-extremal updates for nonlinear model-predictive control and dynamic real-time optimization | |
CN103293953A (en) | Robust adaptive model predictive controller with tuning to compensate for model mismatch | |
CN103472723A (en) | Predictive control method and system based on multi-model generalized predictive controller | |
CN109839825A (en) | A kind of forecast Control Algorithm and system of Rare-Earth Extraction Process constituent content | |
CN106249599A (en) | A kind of network control system fault detection method based on neural network prediction | |
CN105760213A (en) | Early warning system and method of resource utilization rate of virtual machine in cloud environment | |
US9448546B2 (en) | Deterministic optimization based control system and method for linear and non-linear systems | |
Grema et al. | Optimal feedback control of oil reservoir waterflooding processes | |
Zhao et al. | Identification of k-step-ahead prediction error model and MPC control | |
Rashid et al. | Handling multi‐rate and missing data in variable duration economic model predictive control of batch processes | |
Liu et al. | H∞ observer-based sliding mode control for singularly perturbed systems with input nonlinearity | |
CN104898426A (en) | Room temperature loop control method based on gradient descent method and generalized prediction control | |
Bakaráč et al. | Fast nonlinear model predictive control of a chemical reactor: a random shooting approach | |
CN109613830B (en) | Model prediction control method based on decreasing prediction step length | |
Kirubakaran et al. | Energy efficient model based algorithm for control of building HVAC systems | |
Bunin et al. | Input filter design for feasibility in constraint-adaptation schemes | |
CN105182747A (en) | Automatic control method in large lag time system | |
Bresch-Pietri et al. | Prediction-based control of linear systems subject to state-dependent state delay and multiple input-delays | |
Hsu et al. | Chaos synchronization of nonlinear gyros using self-learning PID control approach | |
CN115138472A (en) | Sorting suspension density setting method and system | |
Wassila et al. | Convex Optimization in Model Predictive Control based on Hammerstein Model |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20151223 |
|
RJ01 | Rejection of invention patent application after publication |