CN104898423A - Controller automatic design method based on library thinking and intelligent optimization algorithm - Google Patents

Abstract

The present invention discloses a controller automatic design method based on 'library' thinking and an intelligent optimization algorithm, and the method is used for realizing the controller automatic optimization design of different types of control systems. According to the method provided by the invention, firstly a specific control strategy is determined according to the different types of control systems, a suitable control link is selected to form the specific structure of the controller, a needed performance index is selected as a fitness function according to a field condition and a user need, then the intelligent optimization algorithm is used to globally search the parameter of the controller in a set range, an optimal solution is obtained, and the automatic optimization design of the controller is completed. The controller designed by the invention has the advantages of a rich structure and a large parameter range, the control performance of the controller is superior compared with a traditional PID controller, and the accurate and effective control of a controlled object can be realized. The method provided by the invention has the advantages of strong versatility and effectiveness, and the obvious superiority is reflected.

Description

Based on the controller automatic design method of " storehouse " thought and intelligent optimization algorithm

Technical field

The present invention relates to a kind of controller design method, particularly one is based on the controller automatic design method of " storehouse " thought and intelligent optimization algorithm, for realizing the controller Automatic Optimal Design of dissimilar control system, belongs to automatic control technology field.

Background technology

Automatic control system occupies leading role in military weaponry, Aero-Space, commercial production, and controller is the core of automatic control system.In modern industry application, control method is of a great variety, mainly comprises PID control, fuzzy control, adaptive control, Active Disturbance Rejection Control etc.PID controller is because having the advantages that structure is simple, Parameter adjustable, system robustness are good, and be widely used in the industries such as machinery, metallurgy, petrochemical complex, proportion accounts for more than 80%.In order to make PID controller, there is good control performance, need to carry out parameter tuning to it, the parameter tuning method proposed at present mainly contains Ziegler – Nichols (ZN) Tuning, rule-based Self-tuning System and the pid parameter optimized tuning method based on intelligent optimization algorithm.Researchist has done large quantity research in the parameter self-tuning and self-adaptation of PID controller, main approaches uses intelligent optimization algorithm to carry out global search to P, I, D tri-parameters, the intelligent optimization algorithm used mainly contains genetic algorithm, differential evolution, particle cluster algorithm, ant group algorithm etc.

Document (Xiao L Q, Shao X G, Zhang L, et al.PID parameter optimization using improved genetic algorithm [J] .Computer Engineering and Applications, 2010,46 (1): 200-202.) in the mutation operation of genetic algorithm, introduce particle cluster algorithm, genetic algorithm is combined with particle cluster algorithm and is applied to PID controller parameter optimization.At document (Kang J, Meng W, Abraham A, et al.An adaptive PID neural network for complex nonlinear system control [J] .Neurocomputing, 2014,135:79-85.), the people such as Kang J propose a kind of control method based on self-adaptive PID neural network and particle cluster algorithm.Document (Dong R.Differential evolution versus particle swarm optimization for PID controller design [C] .2009Fifth International Conference on Natural Computation, 2009,3:236-240.) adopt differential evolution and particle cluster algorithm to carry out parameter optimization to PID controller and fuzzy controller respectively.

Although the controller of these research institutes above-mentioned design has good control effects, but PID controller is proportional, integration, differential three links only, limited to the range of adjustment exported, structure has significant limitation, especially becoming when complicated, nonlinear control system time (as time lag, saturated), often can not reach preferably control effects.In addition, the parameter optimization of PID controller needs to carry out under the prerequisite obtaining accurate control system mathematical model, and this adds many troubles to the design of PID controller.How realizing the Automatic Optimal Design (comprising the optimization of structure and parameter) of controller, is a new research direction.The present invention is directed to described Controller gain variations problem, provide a kind of controller Automatic Optimal Design new approaches.

Summary of the invention

Fix to overcome conventional controller structure, shortcoming to outside adaptive capacity to environment difference, improve complex condition controller performance, the invention provides a kind of controller automatic design method based on " storehouse " thought and intelligent optimization algorithm.The method determines concrete control strategy according to dissimilar control system, select the structure of suitable controlling unit composition control device, and select the performance index needed as fitness function according to field condition and user's request, then call intelligent optimization algorithm and optimizing is carried out to the parameter of controller, finally obtain optimum solution, complete the optimal design of controller.

The present invention realizes by the following technical solutions

A kind of controller automatic design method based on " storehouse " thought and intelligent optimization algorithm, concrete steps are as follows:

Step 1, structure link storehouse

Controller architecture is formed some Typical Control Loops that may need or other self-defining controlling units are built into storehouse, as the fundamental element of controller composition, the mathematical model of each controlling unit adopts the form of transport function to represent, and each controlling unit at least has a controling parameters to be optimized;

Step 2, structure performance index storehouse

Performance index storehouse comprises multiple index for evaluating controller performance;

Step 3, structure Optimization Algorithms Library

Optimization Algorithms Library comprises multiple different intelligent optimization algorithm;

Step 4, build control system, running optimizatin algorithm, realize the design of controller

According to the characteristic of controlled device, select suitable control strategy; For selected control strategy, select the link in link storehouse, its transport function is carried out getting according to the version of control strategy and or get the operation of multiplication;

Resolve according to each link transport function and obtain controller mathematical model; From performance index storehouse, selected index is as the fitness function of each Absent measures parameter optimization of controller, and from Optimization Algorithms Library, selected optimized algorithm finds the optimum solution of controling parameters, realizes the Automatic Optimal Design of controller.

Further, the process of the optimum solution of searching controling parameters of the present invention is:

Encoded by the controling parameters of each controlling unit, the vector of composition is as population at individual, and adopt the controling parameters of intelligent optimization algorithm to each link of controller to carry out optimizing, algorithm iteration step is as follows:

S1: given algorithm parameter and globally optimal solution, sets the bound of controling parameters to be optimized; Initialization in setting range, calculates its fitness, according to fitness size select wherein a certain proportion of more excellent individuality as final initialized parent population;

S2: for parent population, according to the operating mechanism of selected optimized algorithm, generates progeny population;

S3: compare progeny population fitness size of individuality on individual position corresponding to parent population on each, carry out selection operation, retain more excellent individuality, formation new population;

S4: the fitness calculating each individuality in the determined population of S3, determines contemporary optimum solution, and more contemporary optimum solution and globally optimal solution, if contemporary optimum solution is better than globally optimal solution, then upgrade globally optimal solution, proceed to S5, otherwise, directly proceed to S5;

S5: judge whether to reach stop condition, if reach, then perform S6, otherwise iterations adds 1, and the new population formed by S3 is as parent population, turns S2;

S6, the decoding of acquisition globally optimal solution, complete Controller gain variations.

Further, the present invention, after execution of step 4, also comprises and being optimized controller architecture, and detailed process is:

First, the globally optimal solution obtained by S6 substitutes in transport function, resolves the accurate model obtaining transport function, obtains controller transfer function and control system response curve, obtains control system response and exports;

Secondly, judge whether each parameter of globally optimal solution that S6 obtains is less than setting threshold value, omit the controlling unit that controling parameters is all less than setting threshold value, then reconfigure all the other controlling units that step 4 is selected, complete the structure optimization of controller, and again generate control system curve of output;

Then, if after structure optimization with optimization before control system exports and is more or less the same, all meet control overflow, then the controller architecture after export structure optimization and parameter; Otherwise the controller architecture before export structure optimization and parameter, controller Automatic Optimal Design terminates.

Beneficial effect

The first, the present invention constructs link storehouse, has enriched controller architecture, make the composition of controller and control method clear.

The second, the present invention constructs performance index storehouse and intelligent optimization algorithm storehouse, can meet the Controller gain variations demand under different user, different working condition, have very strong versatility.The controller optimization method for designing provided is simple and practical, saves the setting time of controller parameter, does not need too many priori, even if to the unfamiliar staff of autocontrol method, also method according to the present invention can complete the optimal design of controller.

3rd, the initialization of the controller parameter in the present invention and optimizing process carry out in a sizable scope, thus can described on a large scale in search globally optimal solution, broken the limitation that conventional controller parameter is optimized among a small circle.Present invention achieves controller architecture and parameter optimization, simplified controller architecture, be beneficial to realization.

4th, the controller architecture of the present invention's design is more abundant, parameter range of adjustment is larger, has surmounted traditional PID controller, can realize controlling more accurately and efficiently different controlled device in control performance, there is very strong versatility and validity, embody obvious superiority.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of controller automatic design method of the present invention;

Fig. 2 is controller provided by the present invention composition schematic diagram;

Fig. 3 is the link storehouse that the present invention builds;

Fig. 4 is the performance index storehouse that the present invention builds;

Fig. 5 is the Optimization Algorithms Library that the present invention builds;

Fig. 6 is control system block diagram in the present invention;

Fig. 7 is system unit step response curve in the embodiment of the present invention.

Embodiment

Below in conjunction with drawings and Examples, the present invention is elaborated.

A kind of controller automatic design method based on " storehouse " thought and intelligent optimization algorithm of the present invention, as shown in Figure 1, concrete steps are as follows:

Step 1, structure link storehouse

Controller architecture is formed some Typical Control Loops that may need or other self-defining controlling unit (explicit physical meaning, simply can realize) be built into storehouse, as the fundamental element of controller composition, the mathematical model of each controlling unit adopts the form of transport function to represent.Each link in link storehouse is not accurate mathematical model, but the model undetermined containing unknown controling parameters, to realize the optimal design of controller.Each controlling unit at least has a controling parameters to be optimized, determines the number of parameter to be optimized according to the concrete transport function of controlling unit.

Step 2, structure performance index storehouse

In order to Controller gain variations is convenient, meet different external environment condition and task to the performance requirement of controller, the invention provides a kind of performance index storehouse.Performance index storehouse comprise multiple common, by the universally recognized index for evaluating controller performance of industry, provide with the form of mathematic(al) representation.Performance index can be used as the fitness function of intelligent optimization algorithm, and by the iteration of algorithm, continuous invocation performance index passes judgment on the performance of controller, and the quality of more multiple controller realizes the optimal design of controller.

Step 3, structure Optimization Algorithms Library

Intelligent optimization algorithm passes through announcement or some phenomenon of simulating nature circle or process and grows up, and has the feature of global optimization performance, highly versatile, parallel processing.Different intelligent optimization algorithms is assembled storehouse, and which kind of optimized algorithm user can use according to design requirement is selected voluntarily.

Step 4, build control system, running optimizatin algorithm

According to controlled device, there is which kind of characteristic (time lag, saturated etc.), select suitable control strategy (as simple closed-loop control, Prediction Control, internal model control etc.).For selected control strategy, the link in choose reasonable link storehouse, its transport function is carried out getting according to the version of control strategy and or get the operation of multiplication, design the controller that structure is more abundant, larger to Drazin inverse scope.

Each link get and or get the basic structure of multiply operation composition control device, resolve according to each link transport function and obtain controller mathematical model.According to the mathematical model of controller and the mathematical model of control object, resolve the mathematical model obtaining closed-loop control system.

From performance index storehouse, selected index is as the fitness function of each Absent measures parameter optimization of controller.

In order to obtain more excellent solution, intelligent optimization algorithm finds optimum solution in sizable scope, and the size of this scope can be embodied as premised on physical circuit by controling parameters.

Meanwhile, in order to improve convergence of algorithm speed, improve convergence effect, this example provides a kind of exponential initial and optimization method.Described exponential initial and optimization method refers to and is mapped in relatively little scope by the controling parameters of original interior real number in a big way, realizes initialization and the optimization of controling parameters among a small circle.

Encoded by the controling parameters of each controlling unit, the vector of composition is as population at individual, and adopt the controling parameters of intelligent optimization algorithm to each link of controller to carry out optimizing, algorithm iteration step is as follows:

S1: initialization is carried out to initial population scale, stop condition, other operators and globally optimal solution, sets the bound of controling parameters to be optimized.Initialization in setting range, calculates its fitness, according to fitness size select wherein a certain proportion of more excellent individuality as final initialized parent population;

S2: for parent population, according to the operating mechanism of selected optimized algorithm, generates progeny population;

S3: compare progeny population fitness size of individuality on individual position corresponding to parent population on each, carry out selection operation, retain more excellent individuality, formation new population;

S4: the fitness calculating each individuality in the determined population of S3, determines contemporary optimum solution, more contemporary optimum solution and globally optimal solution, if contemporary optimum solution is better than globally optimal solution, then upgrades globally optimal solution, proceed to S5, otherwise, directly proceed to S5;

S5: judge whether to reach stop condition, if reach, then perform S6, otherwise iterations adds 1, and the new population formed by S3 is as parent population, turns S2;

S6, the decoding of acquisition globally optimal solution, complete Controller gain variations.

Step 5, controller architecture optimization

The optimum solution obtained by S6 in step 4 substitutes in transport function, resolves the accurate model obtaining transport function, obtains controller transfer function and control system response curve, obtains control system response and exports.In order to controller realizes simple and convenient, judge whether each parameter of optimum solution described in S6 is less than setting threshold value, omit the controlling unit that controling parameters is all less than setting threshold value, then all the other controlling units that step 4 is selected are reconfigured, complete the structure optimization of controller, and again generate control system curve of output.If after structure optimization with optimization before control system exports and is more or less the same, all meet control overflow, then the controller architecture after export structure optimization and parameter; Otherwise the controller architecture before export structure optimization and parameter, controller Automatic Optimal Design terminates

The transport function of different link is represented respectively, G referring to Fig. 2, A (s), B (s), C (s) _cs () represents the transport function of controller, multiple link transport function get and operate the transport function constituting controller, then controller transfer function is:

G _c(s)＝A(s)+B(s)+C(s)+… (1)

Wherein A (s), B (s), C (s) are the transport functions comprising controling parameters to be optimized, and each link at least comprises one can optimal control parameter.When A, B, C represent ratio, integration, differential three links respectively, this controller transfer function is:

$G_c\left(s\right)=k_p+\frac{k_i}{s}+k_{d}s---\left(2\right)$

Wherein, k _p, k _i, k _dbe respectively the control coefrficient of ratio, integration, differentiation element, then this controller is PID controller.Therefore, the controller of this example covers traditional PID controller, embodies the superiority of the controller based on storehouse (LBC) of the present invention's design, and the known described controller based on storehouse can obtain control effects more better than PID controller in theory.

Shown below is specific embodiment, describe content of the present invention and concrete implementation step in detail.

Step 1, structure link storehouse

The link storehouse of the present embodiment design is referring to Fig. 3.Fig. 3 illustrates all links of the embodiment of the present invention, and gives each link concrete transport function.The link that link storehouse comprises is proportional, integration, differential, one order inertia, time lag, two differential, double integrator, each link is at least containing a controling parameters.K _p, k _i, k _d, k ₁, k ₂, k ₃, k ₄, τ is the controling parameters of each link.

Step 2, structure performance index storehouse

Fig. 4 is the performance index storehouse that the present embodiment builds, and gives the specific formula for calculation of each performance index.IAE is the integration of Error Absolute Value; ITAE is the integration of Error Absolute Value and time product; ISE is the integration of square-error; ITSE is the integration of square-error and time product; Overall target is the weighting of overshoot, rise time and stabilization time.Wherein, T is integral time, can rule of thumb set, and is generally greater than the system stability time.E (t) is control system output error; U (t) is controller output; σ is overshoot, t _rfor the rise time, t _srefer to stabilization time.ω ₁, ω ₂, ω ₃for weights, and ω ₁+ ω ₂+ ω ₃=1.

Step 3, structure Optimization Algorithms Library

The Optimization Algorithms Library of the present embodiment design is referring to Fig. 5, and Optimization Algorithms Library comprises genetic algorithm, population, ant group algorithm, simulated annealing, differential evolution algorithm, tabu search algorithm, DNA calculating etc.

Step 4, build control system, running optimizatin algorithm

The controlled device that the present embodiment is chosen is purely retarded control object, and its transport function is

$G_1\left(s\right)=\frac{{10e}^{-s}}{(2s+1)(8s+1)}---\left(3\right)$

According to described controlled device, select the method for Prediction Control, Control system architecture is referring to Fig. 6.Control system entirety adopts unity negative feedback closed loop control method, represents the feedback loop of Prediction Control in dotted line frame.In figure 6, G _cs () is controller transfer function, G _ps () represents controlled device not containing the part of retardation time, F (s) represents feedback element, and DE represents the optimized algorithm (differential evolution algorithm, Differential Evolution, vehicle economy) that the present embodiment is selected.U is input signal, and y is output signal.According to the priori of Prediction Control, F (s)=G _p(s) (1-e ^{-τ s}).

According to above controlled device and control strategy, in the link storehouse that described Fig. 3 builds, select following link: proportional component, integral element, differentiation element, double integrator link, first order inertial loop, Time Delay, the transport function of the link of selection be added the transport function obtaining controller:

$G_c\left(s\right)=K_1+\frac{K_2}{s}+K_{3}s+\frac{K_4}{s^2}+\frac{K_5}{s+K_6}---\left(4\right)$

$F\left(s\right)=G_p\left(s\right)(1-e^{-k_{7}s})---\left(5\right)$

$G_p\left(s\right)=\frac{10}{(2s+1)(8s+1)}---\left(6\right)$

Wherein K ₁, K ₂..., K ₇the controling parameters (parameter to be optimized) of links respectively.

Based on automatically controlling knowledge, whole closed-loop system transport function can be obtained:

$G\left(s\right)=\frac{\frac{G_c\left(s\right)}{1+G_c\left(s\right)F\left(s\right)}{G}_p\left(s\right)e^{-\mathrm{\τ}}}{1+\frac{G_c\left(s\right)}{1+G_c\left(s\right)F\left(s\right)}{G}_p\left(s\right)e^{-\mathrm{\τ}}}---\left(7\right)$

Arranging controling parameters lower boundary is 0, and the upper bound is 10 ³.Be the index of 10 by above-mentioned controling parameters range mappings, namely log got to described scope, can be similar to and obtain less optimization range (-2,3).From described Fig. 4 selectivity index IAE (the absolute value integration of error), wherein integral time T=15.From described Fig. 5, select differential evolution as optimized algorithm, the concrete steps of differential evolution algorithm are described below:

S1: initialization.Given algorithm parameter: population scale 30, zoom factor 0.5, the intersection factor 0.9, maximum iteration time 100, make globally optimal solution=10 ⁶.Initial 100 individualities also calculate its fitness, according to fitness size select wherein 30 more excellent individualities as final initialization parent population;

S2: for parent population, adopts each individuality of DE/rand/1/bin difference strategy to population to carry out variation interlace operation and generates progeny population;

S3: compare progeny population fitness size of individuality on individual position corresponding to parent population on each, carry out selection operation, retain more excellent individuality, formation new population;

S4: the fitness calculating new population individuality in S3, determines contemporary optimum solution, more contemporary optimum solution and globally optimal solution, if contemporary optimum solution is better than globally optimal solution, then upgrades globally optimal solution, proceed to S5, otherwise directly proceed to S5;

S5: judge whether to reach maximum iteration time 100, if reach, then perform S6, otherwise iterations adds 1, and the new population formed by S3 is as parent population, turns S2;

S6, acquisition globally optimal solution;

By above operating procedure, global optimum's solution vector that after 100 generations, differential evolution algorithm obtains is:

x＝[2.0204 0.8591 2.1732 -4.7813 -11.2025 2.6473 0.0002]。Be mapped in original real number scope (0,1000), the optimal control parameter of controller can be obtained:

k＝[104.8211 7.2294 149.0282 0 0 444.0090 1.0006]

Step 5, controller architecture optimization

Go out two null solutions by the known algorithm optimization of optimal control parameter described in the S6 of step 4, eliminate two links in selected link, controller architecture obtains optimization.Optimal control parameter is substituted into formula (4), (5), (6), resolves the mathematical models obtaining each several part:

$G_c\left(s\right)=104.8221+\frac{7.2294}{s}+149.0282s---\left(8\right)$

F(s)＝G _p(s)(1-e ^-1.0006s) (9)

$G_p\left(s\right)=\frac{10}{(2s+1)(8s+1)}---\left(10\right)$

Control system response curve is obtained, referring to Fig. 7 by MATLAB emulation.Wherein, PID is PID controller, and LBC is the controller (Library Based Controller is called for short LBC) based on storehouse in the present invention.As can be seen from curve in figure, the controller rise time based on storehouse and the stabilization time of the present invention's design only have about 0.1 second, and overshoot is almost 0.The controller of the inventive method design has good rapidity, and dynamic property is good, obtains preferably control effects.In control performance, surmount traditional PID controller, can realize controlling more accurately and efficiently controlled device, embody obvious superiority.

Disclosed embodiment is implemented under premised on technical solution of the present invention above, give detailed embodiment and concrete operating process, but protection scope of the present invention is not limited to described embodiment.Known by the above, the many contents in the present invention can make an amendment and replace, and the present embodiment secures some value just in order to principle of the present invention and application are described better, thus is more readily understood and uses.All local of doing on the basis of technical scheme of the present invention are changed, equivalent replacement, improvement etc. all should be included within protection scope of the present invention.

Claims (3)

1., based on a controller automatic design method for " storehouse " thought and intelligent optimization algorithm, it is characterized in that, concrete steps are as follows:

Step 1, structure link storehouse

Controller architecture is formed some Typical Control Loops that may need or other self-defining controlling units are built into storehouse, as the fundamental element of controller composition, the mathematical model of each controlling unit adopts the form of transport function to represent, and each controlling unit at least has a controling parameters to be optimized;

Step 2, structure performance index storehouse

Performance index storehouse comprises multiple index for evaluating controller performance;

Step 3, structure Optimization Algorithms Library

Optimization Algorithms Library comprises multiple different intelligent optimization algorithm;

Step 4, build control system, running optimizatin algorithm, realize the design of controller

According to the characteristic of controlled device, select suitable control strategy; For selected control strategy, select the link in link storehouse, its transport function is carried out getting according to the version of control strategy and or get the operation of multiplication;

Resolve according to each link transport function and obtain controller mathematical model; From performance index storehouse, selected index is as the fitness function of each Absent measures parameter optimization of controller, and from Optimization Algorithms Library, selected optimized algorithm finds the optimum solution of controling parameters, realizes the design of self-actuated controller.

2., according to claim 1 based on the controller automatic design method of " storehouse " thought and intelligent optimization algorithm, it is characterized in that, the process of the optimum solution of described searching controling parameters is:

Encoded by the controling parameters of each controlling unit, the vector of composition is as population at individual, and adopt the controling parameters of intelligent optimization algorithm to each link of controller to carry out optimizing, algorithm iteration step is as follows:

S1: given algorithm parameter and globally optimal solution, sets the bound of controling parameters to be optimized; Initialization in setting range, calculates its fitness, according to fitness size select wherein a certain proportion of more excellent individuality as final initialized parent population;

S2: for parent population, according to the operating mechanism of selected optimized algorithm, generates progeny population;

S3: compare progeny population fitness size of individuality on individual position corresponding to parent population on each, carry out selection operation, retain more excellent individuality, formation new population;

S4: the fitness calculating each individuality in the determined population of S3, determines contemporary optimum solution, more contemporary optimum solution and globally optimal solution, if contemporary optimum solution is better than globally optimal solution, then upgrades globally optimal solution, proceed to S5, otherwise, directly proceed to S5;

S5: judge whether to reach stop condition, if reach, then perform S6, otherwise iterations adds 1, and the new population formed by S3 is as parent population, turns S2;

S6, the decoding of acquisition globally optimal solution, complete Controller gain variations.

3., according to claim 2 based on the controller automatic design method of " storehouse " thought and intelligent optimization algorithm, it is characterized in that, after execution of step 4, also comprise and being optimized controller architecture, detailed process is:

First, the globally optimal solution obtained by S6 substitutes in transport function, resolves the accurate model obtaining transport function, obtains controller transfer function and control system response curve, obtains control system response and exports;

Secondly, judge whether each parameter of globally optimal solution that S6 obtains is less than setting threshold value, omit the controlling unit that controling parameters is all less than setting threshold value, then reconfigure all the other controlling units that step 4 is selected, complete the structure optimization of controller, and again generate control system curve of output;

Then, if after structure optimization with optimization before control system exports and is more or less the same, all meet control overflow, then the controller architecture after export structure optimization and parameter; Otherwise the controller architecture before export structure optimization and parameter, controller Automatic Optimal Design terminates.