CN104330968A - Inverse model PID compound control method based on improved support vector regression - Google Patents

Abstract

The invention requests for protecting an inverse model PID compound control method based on improved support vector regression. The method utilizes a multi-agent particle swarm optimization algorithm to optimize related parameters of a support vector regression to effectively improve modeling precision and generalization ability. On the above basis, an inverse model of MAPSO-SVR (Multi-Agent Particle Swarm Optimization-Support Vector Regression) is established, so that precision of the inverse model is improved to a certain extent; PID control is introduced, the MAPSO-SVR inverse model/PID compound control method is provided, own continuous amendment of a nonlinear system is achieved, and tracking and control ability of the system can be effectively improved.

Description

A kind of inversion model PID composite control method based on improving support vector regression

Technical field

The invention belongs to a kind of inverse model control method based on support vector regression, particularly relate to a kind of method optimizing the inversion model/PID complex controll nonlinear system of SVR correlation parameter based on multi-agent particle swarm (MAPSO).

Background technology

Method of inverse is set up than more complete design theory in the nonlinear system of general type in recent years, can nonlinear system in general form as research object, and special requirement is not had to equation form, do not need to introduce differential, several how abstract mathematical theory, thus there is general Research Significance yet.Have at present and much utilize the method for neural network inversion model extensively to be studied, also achieve theoretical and real achievement preferably.But in modeling, neural network exists Local Minimum, network number of plies difficulty such as to determine at the problem.And utilize support vector regression (SVR) to set up inversion model, because SVR has the advantages such as global optimum, generalization ability be strong, the antijamming capability of model effectively can be improved.The tracking performance of the direct influential system of degree of accuracy of model, the optimum choice of parameter affects significantly the precision of prediction of SVR and generalization ability, and common optimized algorithm has a lot, as genetic algorithm (GA), particle cluster algorithm (PSO) etc.Genetic algorithm is easily absorbed in local minimum, and this algorithm has very strong dependence to choosing of inherent parameters, particle cluster algorithm is only shared the information of optimal particle in colony and have ignored the information of other particle, particle may be caused just to attracted in early days near certain point at algorithm, thus reduce the diversity of population.

The present invention proposes the method utilizing multi-Agent particle cluster algorithm (MAPSO) to optimize SVR correlation parameter.MAPSO algorithm combines multiparticle information sharing and adaptable inertia weigh strategy, multiparticle information sharing adopts multiparticle information to revise each particle action strategy next time, to reduce the possibility that particle is absorbed in local optimum, inertia weight self-adaptative adjustment is according to colony's prematurity convergence degree, by individual concrete condition adjustment inertia weight, particle is made to jump out local optimum, by realizing global optimum with the competition of its neighbor particle, cooperation.Because method of inverse is " pseudo-linear system " structure in theory, exist and control deficiency, the problem that stability is not high, introduce PID FEEDBACK CONTROL, improve the tracking of system, control ability.

Summary of the invention

For above deficiency of the prior art, the object of the present invention is to provide a kind of improve system tracking, control ability based on the inversion model PID composite control method improving support vector regression, technical scheme of the present invention is as follows:

Based on the inversion model PID composite control method improving support vector regression, it comprises the following steps:

101, Initialize installation groupy phase related parameter, generating training data collection, test data set, and training dataset and test data set are normalized;

102, utilize multi-Agent particle cluster algorithm/MAPSO to carry out optimizing to support vector regression SVR correlation parameter, by fitness function and particle position, speed more new formula to support vector regression SVR parameter optimization;

103, judge whether to meet end condition, if meet, optimizing terminates thus determines the parameter optimization result of optimal particle and MAPSO-SVR, jumps to step 104; Otherwise return step 102;

104, determine the inversion model matching factor, set up MAPSO-SVR inversion model;

105, reference input function y is set _routthe u value in (k) and before moment, and according to y _routk (), inversion model and object model, calculate the value uctr (k) predicting out through inversion model;

106, by known y _rout(k), y _outk () value, according to definition and increment type PID principle △ u (the k)=k of error _p* x1+k _d* x2+k _i* x3, wherein k _prepresent proportional control factor, k _drepresent derivative control coefficient, k _irepresent integral control coefficient, x1=e (k)-e (k-1), x2=e (k)-2*e (k-1)+e (k-2), x3=e (k), and according to the uctr (k) in step 105, try to achieve current time u (k) value according to formula u (k)=uctr (k)+△ u (k);

107, subsequent time real output value y is calculated according to step 106 and object model _out(k+1);

If the value of 108 k does not also exceed the sampling time, then proceed to step 105, terminate until k value reaches sampling time setting value.

Further, the initialization groupy phase related parameter in step 101 comprises definition solution space, environment scale, maximum permission iterations, Inertia Weight scope, Studying factors.

Further, the particle fitness function F in step 102 is:

$F=\frac{1}{N}{\mathrm{\Σ}}_{i=1}^N(y_i-y_i^{,})$

Wherein y _i, y ' _irepresent SVR respectively and train output valve and desired output, N is number of training,

Further, the position of the information updating particle according to optimal particle in step 102, the formula of speed are:

$\left\{\begin{array}{c}{v}_{t+1}=w\·v_t+C_1\·r1\·(p_{\mathrm{Best}}-x_t)+C_2\·r2\·(g_{\mathrm{Best}}-x_t)\\ x_{t+1}=x_t+v_{t+1}\end{array}\right.$

In formula, subscript t is iterations, x _tparticles spatial position when being the t time iteration, v _tbe particle rapidity during the t time iteration, wherein C ₁, C ₂for Studying factors, r ₁, r ₂for the random number between (0,1), w is inertia constant.

Further, the inversion model matching factor in step 104 is $\left\{\begin{array}{c}X=[y(k+1),...y(k+1-n),...u(k-m)]\\ Y=u\left(k\right)\end{array}\right..$

Advantage of the present invention and beneficial effect as follows:

The present invention utilizes MAPSO algorithm optimization SVR correlation parameter, introduces PID simultaneously and controls to improve whole system static and dynamic performance.Set up the not high and control system of inversion model precision for SVR and control the problems such as not enough, utilize MAPSO algorithm optimization SVR parameter, avoid the randomness of artificial selection parameter, improve the modeling accuracy that SVR sets up inversion model; Introduce PID to control, compensate for the problem of " pseudo-linear " Systematical control deficiency, the system that realizes constantly revises real output value makes input and output error little as much as possible, thus improves tracking, the control ability of system.The present invention carries out optimization, the improvement of control method mainly for the parameter optimization of SVR and the deficiency of Inverted control system, and its effect of optimization is obvious, improves the tracking of system, control ability.

Accompanying drawing explanation

Fig. 1 is according to multiplex control system program flow diagram of the present invention.

Embodiment

The invention will be further elaborated to provide an infinite embodiment below in conjunction with accompanying drawing.But should be appreciated that, these describe just example, and do not really want to limit the scope of the invention.In addition, in the following description, the description to known features and technology is eliminated, to avoid unnecessarily obscuring concept of the present invention.

As shown in Figure 1, the present invention uses inverse system principle, on the basis of inversion model, introduces MAPSO algorithm and PID control method, and improve modeling accuracy and improve system control performance, making the tracking of system, control ability gets a promotion, specific implementation step is as follows:

Step 1: Initialize installation correlation parameter, generating training data, test data data are normalized.

Step 2:MAPSO initialization of population, optimum configurations also produces initial position and the speed of population.

Step 3: according to step 2 and fitness function: calculate current particle adaptive value, wherein y _i, y ' _irepresent SVR respectively and train output valve and desired output, N is number of training.

Step 4: the personal best particle according to neighbor information more new particle: each particle predefine neighbours' environment, hypothetical particle L _i,j, and L _i,j=(l ₁, l ₂... l _d) be its position in optimization problem solution space, M _i,jl _i,jm neighbours in have the particle of optimal adaptation value, and M _i,j=(m ₁, m ₂..., m _d).If L _i,jmeet f (L _i,j)≤f (M _i,j), then it is a winner, otherwise it is a loser.If L _i,jbe a winner, it remains unchanged in the position of solution space; If it is loser, then a L _i,jin the position of solution space then according to L _i,j=M _i,j+ rand (-1,1) (L _i,j-M _i,j) upgrade.

Step 5: according to step 4 (i.e. P _best, g _best), and formula: $\left\{\begin{array}{c}{v}_{t+1}=w\·v_t+C_1\·r1\·(p_{\mathrm{Best}}-x_t)+C_2\·r2\·(g_{\mathrm{Best}}-x_t)\\ x_{t+1}=x_t+v_{t+1}\end{array}\right.$ The more position of new particle, speed, in formula, subscript t is iterations, x _tparticles spatial position when being the t time iteration, v _tbe particle rapidity during the t time iteration, wherein C ₁, C ₂for Studying factors, r ₁, r ₂for the random number between (0,1), w is inertia constant.

Step 6: judge whether to meet end condition.If meet, optimizing terminates thus determines the parameter optimization result of optimal particle and MAPSO-SVR; Otherwise go to step 3.

Step 7: according to the inversion model modeling principle determination matching factor:

$\left\{\begin{array}{c}X=[y(k+1),...y(k+1-n),...u(k-m)]\\ Y=u\left(k\right)\end{array}\right.$

By object model { X, Y} data set, wherein the input control signal of object model is the random signal between 0 to 1, is trained, set up inversion model by SVR to it in the inputoutput data foundation generated.

Step 8: setting expectation function y _rout(k), and according to y _routk (), inversion model and object model draw uctr (k) value.

Step 9: according to definition e (the k)=y of error _rout(k)-y _out(k) and increment type PID principle △ u (k)=k _p* x1+k _d* x2+k _i* x3, wherein k _pexpression ratio controls, k _drepresent that differential controls, k _irepresent integration control, x1=e (k)-e (k-1), x2=e (k)-2*e (k-1)+e (k-2), x3=e (k).And integrating step 8, try to achieve current time u (k) value by u (k)=uctr (k)+△ u (k).

Step 10: according to step 9, the y in object model and before moment _outvalue, obtains subsequent time y _out(k+1) value.

Step 11: if the value of k does not also exceed the sampling time, then proceed to step 8, terminates until k value reaches sampling time setting value.

The control research of nonlinear system, in the field such as industrial or agricultural, manufacturing industry, has a wide range of applications.This method proposes to optimize SVR parameter with MAPSO and promotes inversion model modeling accuracy, introduces this conventional control of PID as feedback fraction, improves the tracking of nonlinear system, control ability, the performance of nonlinear control system is further optimized.

These embodiments are interpreted as only being not used in for illustration of the present invention limiting the scope of the invention above.After the content of reading record of the present invention, technician can make various changes or modifications the present invention, and these equivalence changes and modification fall into the scope of the claims in the present invention equally.

Claims (5)

1., based on the inversion model PID composite control method improving support vector regression, it is characterized in that: comprise the following steps:

101, Initialize installation groupy phase related parameter, generating training data collection, test data set, and training dataset and test data set are normalized;

102, utilize multi-Agent particle cluster algorithm/MAPSO to carry out optimizing to support vector regression SVR correlation parameter, by fitness function and particle position, speed more new formula to support vector regression SVR parameter optimization;

103, judge whether to meet end condition, if meet, optimizing terminates thus determines the parameter optimization result of optimal particle and MAPSO-SVR, jumps to step 104; Otherwise return step 102;

104, determine the inversion model matching factor, set up MAPSO-SVR inversion model;

105, reference input function y is set _routthe u value in (k) and before moment, and according to y _routk (), inversion model and object model, calculate the value uctr (k) predicting out through inversion model;

106, by known y _rout(k), y _outk () value, according to definition and increment type PID principle △ u (the k)=k of error _p* x1+k _d* x2+k _i* x3, wherein k _prepresent proportional control factor, k _drepresent derivative control coefficient, k _irepresent integral control coefficient, x1=e (k)-e (k-1), x2=e (k)-2*e (k-1)+e (k-2), x3=e (k), and according to the uctr (k) in step 105, try to achieve current time u (k) value according to formula u (k)=uctr (k)+△ u (k);

107, subsequent time real output value y is calculated according to step 106 and object model _out(k+1);

If the value of 108 k does not also exceed the sampling time, then proceed to step 105, terminate until k value reaches sampling time setting value.

2. the inversion model PID composite control method based on improving support vector regression according to claim 1, is characterized in that: the initialization groupy phase related parameter in step 101 comprises definition solution space, environment scale, maximum permission iterations, Inertia Weight scope, Studying factors.

3. the inversion model PID composite control method based on improving support vector regression according to claim 1, is characterized in that: the particle fitness function F in step 102 is:

$F=\frac{1}{N}{\mathrm{\Σ}}_{i=1}^N(y_i-y_i^{,})$

Wherein y _i, y, _irepresent SVR respectively and train output valve and desired output, N is number of training.

4. the inversion model PID composite control method based on improving support vector regression according to claim 1, is characterized in that: the position of the information updating particle according to optimal particle in step 102, the formula of speed are:

$\left\{\begin{array}{c}{v}_{t+1}=w\·v_t+C_1\·r1\·(p_{\mathrm{Best}}-x_t)+C_2\·r2\·(g_{\mathrm{Best}}-x_t)\\ x_{t+1}=x_t+v_{t+1}\end{array}\right.$

In formula, subscript t is iterations, x _tparticles spatial position when being the t time iteration, v _tbe particle rapidity during the t time iteration, wherein C ₁, C ₂for Studying factors, r ₁, r ₂for the random number between (0,1), w is inertia constant.

5. the inversion model PID composite control method based on improving support vector regression according to claim 1, is characterized in that: the inversion model matching factor in step 104 is

$\left\{\begin{array}{c}X=[y(k+1),...y(k+1-n),...u(k-m)]\\ Y=u\left(k\right)\end{array}\right..$