CN102777878B - Main steam temperature PID control method of ultra supercritical unit based on improved genetic algorithm - Google Patents

Abstract

The invention relates to a main steam temperature PID (Proportion Integration Differentiation) control method of an ultra supercritical unit for boiler power generation in a thermal power plant based on an improved genetic algorithm. The method comprises the improved genetic algorithm. The improved genetic algorithm is improved on the basis of a simple genetic algorithm, optimal sequencing retaining selection, non-uniform linear intersection and Gauss mutation operators are introduced, a uniform design method is adopted to construct an initial population, and the intersection and variation probability is adjusted by Sigmoid function adaptability; and a PID controller orienting a DCS (distributed control system) is adopted, and the parameter set of the controller under a part load section is optimally obtained through the improved genetic algorithm. The optimizing speed and precision of the genetic algorithm are improved, the control method can replace an engineering adjusting method with complex engineering and low accuracy, the workload of on-site debugging personnel is reduced, the purpose of intelligent main steam temperature control is achieved, the application of an intelligent optimization technology under the DCS is realized, and the need of stably controlling the main steam temperature of the ultra supercritical unit in deep peak regulation is met.

Description

A kind of main steam temperature PID control method of ultra supercritical unit based on improved adaptive GA-IAGA

Technical field

The present invention relates to the control method of a kind of heat power plant boiler generating set Stream temperature PID, especially relate to a kind of heat power plant boiler based on improved adaptive GA-IAGA generating main steam temperature PID control method of ultra supercritical unit.

Background technology

Heat power plant boiler generating set main steam temperature is the important parameter needing during direct current cooker runs to monitor, has complicated dynamic and static characteristic and wide influence factor.For the main-stream control under spray desuperheating disturbance, usually the cas PID control strategy with feedforward compensation is adopted in power plant's practical application, do not consider the impact of its model change when varying duty completely, manual operations and Supervised Control is needed when operating mode is complicated, when object model changes, often do not reach ideal effect.

For Stream temperature large time delay, non-linear, the uncertain feature of model under adaptation thermal power plant on a large scale load, various control method is incorporated in main-stream control by Chinese scholars.Existing advanced PID control device, Dynamic matrix control device and Dynamic matrix control structure etc. are introduced in Power Plant Thermal process, implement reliable and stable scene application or l-G simulation test.Scholar is had to propose the composite control method such as PID control, fuzzy self-adaption Predictive function control, state variable-PREDICTIVE CONTROL, nerve immunity FEEDBACK CONTROL based on immune genetic algorithm and neutral net, for in the field control of extra-supercritical unit Stream temperature, obtain excellent Control platform.Also have scientific research personnel to devise a kind of steam temperature whole-process control system based on auto-disturbance rejection technology, result of the test shows, this system has good Control platform, and when operating mode wide variation, system exports and can follow the tracks of Optimization Start-Up Curve preferably.These researchs have good facilitation for the main-stream control method explored under varying duty on a large scale, but all there is the problems such as complex structure, tuning method be more, make it in software and hardware realizes, lack effective support, particularly certain difficulty is caused to the application of actual field DCS layer.

Summary of the invention

Technical problem to be solved by this invention, just be to provide a kind of main steam temperature PID control method of ultra supercritical unit based on improved adaptive GA-IAGA towards DCS system, the workload of field adjustable personnel can be reduced, and make system have stronger stability and good robustness.

For solving the problems of the technologies described above, the present invention adopts following technical scheme:

A kind of generating of the heat power plant boiler based on improved adaptive GA-IAGA main steam temperature PID control method of ultra supercritical unit, the PID controller towards DCS system is adopted to control in method, the parameter set of described PID controller under point load section is obtained by Revised genetic algorithum optimizing, improved adaptive GA-IAGA is included in method, described improved adaptive GA-IAGA, comprises the following steps:

Step1: set up pid parameter problem of tuning Mathematical Modeling, comprise object function and constraints;

Object function is relevant with the object reaching control, and constraints is relevant with the Stream temperature characteristic of boiler of power plant, so the design of the different designer's these two aspects of different power plant may be different.The fitness function of genetic algorithm is write according to object function, and constraints limits the computer capacity of genetic algorithm, all will set up according to actual conditions.

Step2: arrange genetic parameter, comprises Population Size, evolutionary generation, selection pressure, crossover probability scope, mutation probability scope;

Step3: adopt real coding mode to encode pid parameter and suitable parameter upper and lower limit is set;

Constraints in the bound of parameter and step1, relevant with the characteristic of Stream temperature in the boiler of each power plant.Have a kind of method to be using 1/10 of the empirical value of field adjustable personnel as lower limit, 10 times of empirical value as the upper limit.

Step4: set up uniform designs table, and according to evenly showing initialization of population, produce initial population Pini;

Step5: judge whether fitness restrains, restrain then terminator, do not restrain, enter next step;

Step6: calculate fitness individual in every generation population

Step7: adopt linear ordering operator and elitist selection strategy to select excellent individual;

Step8: adopt Sigmoid function (i.e. f (x)=1/ (1+e-x) neuronic nonlinear interaction function, below formula be oneself release according to the principle inventor of sigmoid function) formulae discovery crossover probability P _c: probability adjustment curve adopts Sigmoid function, and its computing formula is

$P_c^t=P_{c1}-\frac{P_{c1}-P_{c2}}{1+e^{-\mathrm{\β}(t/T-\mathrm{Ns})}}$

In formula, P _c1for initial crossover probability; P _c2for stopping crossover probability; β is form factor, is set as 20; Ns is separation, is set as 0.25;

Step9: adopt non-homogeneous Linear cross operator to produce new gene;

Step10: adopt Sigmoid function formula to calculate mutation probability Pm: probability adjustment curve adopts Sigmoid function, and its computing formula is (this formula is also that oneself is released according to sigmoid function)

$P_m^t=P_{m1}+\frac{P_{m2}-P_{m1}}{1+e^{-\mathrm{\β}(t/T-\mathrm{Ns})}}$

In formula, P _m1for initial mutation probability; P _m2for stopping mutation probability; β is form factor, is set as 20; Ns is separation, is set as 0.25;

Step11: adopt Gaussian mutation operator to produce new gene;

Step12: record the most individual, optimal adaptation degree and progeny population information;

Step13: judge whether to reach maximum evolutionary generation, then returns the 6th step if not, repeats above calculating process until reach maximum evolutionary generation;

Step14: obtain optimal solution, stops genetic algorithm optimization program.

Described step Step4 is specially: adopted grid point method to the structure of uniform designs table, and tests its uniformity by L2 deviation method, and this step Step4 comprises following sub-step further:

Step4-1: the number of levels n of given uniform designs table, when n is odd number, generate one group by positive integer vector H _n={ h ₁, h ₂..., h _m, wherein h is less than n, and their most grand duke's factor is 1; When n is even number, first construct uniform designs table with vectorial H _n+1;

Step4-2: jth row in structure uniform designs table, its individual computing formula is χ _ij=(i*h _j) (mod (n)), wherein mod represents congruence, i=1,2 ..., n, j=1,2 ..., m, congruence can ensure that individuality drops on interval [1, n], thus obtains uniform designs table u _{n × m}matrix.When n is even number, produce matrix U _{(n+1) × m}; Step4-3: adopt L2-departure function CD ₂(P) test its uniformity, select uniformity and do best s row formation uniform designs table its functional equation is

${\mathrm{CD}}_2\left(P\right)=[{\left(\frac{13}{12}\right)}^s-\frac{2}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\underset{j=1}{\overset{s}{\mathrm{\Π}}}(2+|x_{\mathrm{ij}}-\frac{1}{2}|+$

${|x_{\mathrm{ij}}-\frac{1}{2}|}^2)+\frac{1}{n^2}\underset{k=1}{\overset{n}{\mathrm{\Σ}}}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\underset{j=1}{\overset{s}{\mathrm{\Π}}}(1+\frac{1}{2}|x_{\mathrm{kj}}-\frac{1}{2}|$

$+|\frac{1}{2}{x}_{\mathrm{ij}}-\frac{1}{2}|-\frac{1}{2}|x_{\mathrm{kj}}-x_{\mathrm{ij}}|){]}^{\frac{1}{2}}$

Step4-4: produce genetic algorithm initial population by uniform designs table, wherein s is number of parameters, needs to set according to parameter search problem, and Population Size is n, therefore can construct the uniform designs table of a s factor n level, make the individuality in initial population be evenly distributed on s dimension space.

Compared with prior art, the present invention has the following advantages and good effect:

The present invention improves simple generic algorithm, improves speed of searching optimization and the precision of genetic algorithm; What propose can replace engineering complexity by the method that improved adaptive GA-IAGA optimizing obtains PID controller parameter set under point load section and the not high practical tuning method of accuracy, decrease the workload of field adjustable personnel, reach the object of Stream temperature Based Intelligent Control, achieve the application of Intelligent Optimization Technique under DCS system, meet the stability contorting demand of extra-supercritical unit Stream temperature when degree of depth peak regulation.

Accompanying drawing explanation

Fig. 1 is the optimizing flow chart of improved adaptive GA-IAGA in the present invention.

Detailed description of the invention

By reference to the accompanying drawings technical scheme of the present invention is described further below by embodiment.

The present invention adopts widely used PID controller on DCS layer to the fresh steam temperature of extra-supercritical unit, and the parameter set of this PID controller under point load section is obtained by Revised genetic algorithum optimizing.

A kind of main steam temperature PID control method of ultra supercritical unit based on improved adaptive GA-IAGA that the present invention proposes, the concrete steps of its Revised genetic algorithum optimizing are as follows:

Step1: set up pid parameter problem of tuning Mathematical Modeling, comprise object function and constraints;

Step2: arrange genetic parameter, comprises Population Size, evolutionary generation, selection pressure, crossover probability scope, mutation probability scope;

Step3: adopt real coding mode to encode pid parameter and suitable parameter upper and lower limit is set;

Step4: set up uniform designs table and produce initial population P according to evenly showing _ini; Wherein U represents uniform designs table, and n represents number of levels or experiment number, and s represents because of prime number; A mistake! Do not find Reference source.If desired the parameter of identification has m, if setting Population Size is n, then can construct the uniform designs table of a horizontal m factor of n, makes initial population individuality be evenly distributed in m-dimensional space with loose some form;

This specific embodiment, has adopted grid point method to the structure of uniform designs table, and has passed through L ₂-deviation method tests its uniformity.This step comprises following sub-step further:

Step4-1: the number of levels n of given uniform designs table, when n is odd number, generate one group by positive integer vector H _n={ h ₁, h ₂..., h _m, wherein h is less than n, and their most grand duke's factor is 1; When n is even number, first construct uniform designs table with vectorial H _n+1;

Step4-2: jth row in structure uniform designs table, its individual computing formula is χ _ij=(i*h _j) (mod (n)), wherein mod represents congruence, i=1,2 ..., n, j=1,2 ..., m, congruence can ensure that individuality drops on interval [1, n], thus obtains uniform designs table u _{n × m}matrix.When n is even number, produce matrix U _{(n+1) × m}; Step4-3: adopt L ₂-departure function CD ₂(P) test its uniformity, select uniformity and do best s row formation uniform designs table its functional equation is

${\mathrm{CD}}_2\left(P\right)=[{\left(\frac{13}{12}\right)}^s-\frac{2}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\underset{j=1}{\overset{s}{\mathrm{\Π}}}(2+|x_{\mathrm{ij}}-\frac{1}{2}|+$

${|x_{\mathrm{ij}}-\frac{1}{2}|}^2)+\frac{1}{n^2}\underset{k=1}{\overset{n}{\mathrm{\Σ}}}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\underset{j=1}{\overset{s}{\mathrm{\Π}}}(1+\frac{1}{2}|x_{\mathrm{kj}}-\frac{1}{2}|$

$+|\frac{1}{2}{x}_{\mathrm{ij}}-\frac{1}{2}|-\frac{1}{2}|x_{\mathrm{kj}}-x_{\mathrm{ij}}|){]}^{\frac{1}{2}}$

Step4-4: produce genetic algorithm initial population by uniform designs table, wherein s is number of parameters, needs to set according to parameter search problem, and Population Size is n, therefore can construct the uniform designs table of a s factor n level, make the individuality in initial population be evenly distributed on s dimension space.

Step5: judge whether fitness restrains, restrain then terminator, do not restrain, enter and circulate next time;

Step6: calculate fitness individual in every generation population

Step7: adopt line sequence sequence operator and elitist selection strategy to select excellent individual: its individual probability calculation formula is

$\mathrm{Ps}\left(a_j\right)=\frac{1}{M}({\mathrm{\η}}^{+}-\frac{{\mathrm{\η}}^{+}-{\mathrm{\η}}^{-}}{M-1}(j-1))$

In formula, M is group size; J is descending sequence number; η ⁺for optimized individual expects number; η ^-for the poorest individuality expects number, η ⁺+ η-=2.

Step8: calculate crossover probability P _c: probability adjustment curve adopts Sigmoid function, and its computing formula is

$P_c^t=P_{c1}-\frac{P_{c1}-P_{c2}}{1+e^{-\mathrm{\β}(t/T-\mathrm{Ns})}}$

In formula, P _c1for initial crossover probability; P _c2for stopping crossover probability; β is form factor, is set as 20; Ns is separation, is set as 0.25;

Step9: adopt non-homogeneous Linear cross operator to produce new gene; If parent individuality is with producing offspring individual after then intersecting is

$\left\{\begin{array}{c}{X}_1^{t+1}=c\·X_1^t+(1-c)\·X_2^t\\ X_2^{t+1}=(1-c)\·X_1^t+c\·X_2^t\end{array}\right.$

In formula, c is scale factor, and its value produces at random between (0,1);

Step10: calculate mutation probability P _m: probability adjustment curve adopts Sigmoid function, and its computing formula is

$P_m^t=P_{m1}+\frac{P_{m2}-P_{m1}}{1+e^{-\mathrm{\β}(t/T-\mathrm{Ns})}}$

In formula, P _m1for initial mutation probability; P _m2for stopping mutation probability; β is form factor, is set as 20; Ns is separation, is set as 0.25;

Step11: adopt Gaussian mutation operator to produce new gene: the computing formula of offspring individual is

X ^t+1＝X ^t+N(0,σ)

In formula, N (0, σ) is Gaussian Profile; σ is variance.Wherein in order to make variation step-length reduce with the increase of genetic algebra, σ is set as dynamic parameter σ (t), namely

$\mathrm{\σ}\left(t\right)=1-0.9\·\frac{g}{G}$

In formula, g is current evolutionary generation; G is genetic algebra;

Step12: record the most individual, optimal adaptation degree and progeny population information;

Step13: judge whether to reach maximum evolutionary generation, then return Step6 if not, repeats above calculating process until reach maximum evolutionary generation;

Step14: obtain optimal solution, stops genetic algorithm optimization program.

Claims (1)

1. based on a main steam temperature PID control method of ultra supercritical unit for improved adaptive GA-IAGA, it is characterized in that: described method comprises improved adaptive GA-IAGA, described improved adaptive GA-IAGA comprises the following steps:

Step1: set up pid parameter problem of tuning Mathematical Modeling, comprise object function and constraints;

Step2: arrange genetic parameter, comprises Population Size, evolutionary generation, selection pressure, crossover probability scope, mutation probability scope;

Step3: adopt real coding mode to encode pid parameter , and parameters upper and lower limit;

Step4: set up uniform designs table, and according to evenly showing initialization of population, produce initial population ;

Step5: judge whether fitness restrains, restrain then terminator, do not restrain, enter next step;

Step6: calculate fitness individual in every generation population ;

Step7: adopt line sequence sequence operator and elitist selection strategy to select excellent individual;

Step8: adopt Sigmoid function formula to calculate crossover probability : probability adjustment curve adopts Sigmoid function, and its computing formula is

In formula, p _c1for initial crossover probability, p _c2for stopping crossover probability, for form factor, be set as 20; nsfor separation, be set as 0.25;

Step9: adopt non-homogeneous Linear cross operator to produce new gene;

Step10: adopt Sigmoid function formula to calculate mutation probability : probability adjustment curve adopts Sigmoid function, and its computing formula is

In formula, p _m1for initial mutation probability; p _m2for stopping mutation probability; for form factor, be set as 20; nsfor separation, be set as 0.25;

Step11: adopt Gaussian mutation operator to produce new gene;

Step12: record the most individual, optimal adaptation degree and progeny population information;

Step13: judge whether to reach maximum evolutionary generation, then return Step6 if not, repeats above calculating process until reach maximum evolutionary generation;

Step14: obtain optimal solution, stops genetic algorithm optimization program;

Described step Step4 comprises following sub-step:

Step4-1: the number of levels of given uniform designs table , when during for odd number, generate one group by positive integer vector , wherein ratio little, and their most grand duke's factor is 1; When during for even number, first construct uniform designs table and vector ;

Step4-2: in structure uniform designs table the row, its individual computing formula is , wherein represent congruence, , , congruence can ensure that individuality drops on interval , thus obtain uniform designs table 's matrix, when during for even number, produce matrix ;

Step4-3: adopt L ₂-departure function test its uniformity, select uniformity and do best row form uniform designs table , its functional equation is

Step4-4: produce genetic algorithm initial population by uniform designs table, wherein for number of parameters, need to set according to parameter search problem, Population Size is , one can be constructed factor the uniform designs table of level, makes the individuality in initial population be evenly distributed on dimension space.