Scilab Website | Contribute with GitLab | Mailing list archives | ATOMS toolboxes
Scilab Online Help
6.1.0 - Français

Change language to:
English - 日本語 - Português - Русский

Please note that the recommended version of Scilab is 2025.0.0. This page might be outdated.
See the recommended documentation of this function

Aide de Scilab >> Systèmes de Contrôle - CACSD > Identification > sorder

sorder

computing the order of a discrete-time system

Syntax

Ro = sorder(meth, alg, jobd, batch, conct, s, Y)
Ro = sorder(meth, alg, jobd, batch, conct, s, Y, U, tol, printw, ldwork, Ri)
[Ro, n, sval, rcnd] = sorder(..)

Arguments

meth

integer option to determine the method to use:

=

1 : MOESP method with past inputs and outputs;

=

2 : N4SID method.

alg

integer option to determine the algorithm for computing the triangular factor of the concatenated block-Hankel matrices built from the input-output data:

=

1 : Cholesky algorithm on the correlation matrix;

=

2 : fast QR algorithm;

=

3 : standard QR algorithm.

jobd

integer option to specify if the matrices B and D should later be computed using the MOESP approach:

=

1 : the matrices B and D should later be computed using the MOESP approach;

=

2 : the matrices B and D should not be computed using the MOESP approach.

This parameter is not relevant for meth = 2.

batch

integer option to specify whether or not sequential data processing is to be used, and, for sequential processing, whether or not the current data block is the first block, an intermediate block, or the last block, as follows:

=

1 : the first block in sequential data processing;

=

2 : an intermediate block in sequential data processing;

=

3 : the last block in sequential data processing;

=

4 : one block only (non-sequential data processing).

conct

integer option to specify whether or not the successive data blocks in sequential data processing belong to a single experiment, as follows:

=

1 : the current data block is a continuation of the previous data block and/or it will be continued by the next data block;

=

2 : there is no connection between the current data block and the previous and/or the next ones.

This parameter is not used if batch = 4.

s

the number of block rows in the input and output block Hankel matrices to be processed. s > 0

Y

the t-by-l output-data sequence matrix. Column j of Y contains the t values of the j-th output component for consecutive time increments.

U

(optional) the t-by-m input-data sequence matrix. Column j of U contains the t values of the j-th input component for consecutive time increments. Default: U = [].

tol

(optional) vector of length 2 containing tolerances: tol(1) - tolerance used for estimating the rank of matrices. If tol(1) > 0, then the given value of tol(1) is used as a lower bound for the reciprocal condition number; an m-by-n matrix whose estimated condition number is less than 1/tol(1) is considered to be of full rank. If tol(1) <= 0, then a default value m*n*epsilon_machine is used, where epsilon_machine is the relative machine precision.

tol(2) - tolerance used for determining an estimate of the system order. If tol(2) >= 0, the estimate is indicated by the index of the last singular value greater than or equal to tol(2). (Singular values less than tol(2) are considered as zero.) When tol(2) = 0, an internally computed default value, tol(2) = s*epsilon_machine*sval(1), is used, where sval(1) is the maximal singular value, and epsilon_machine the relative machine precision. When tol(2) < 0, the estimate is indicated by the index of the singular value that has the largest logarithmic gap to its successor.

Default: tol(1:2) = [0,-1].

printw

(optional) switch for printing the warning messages.

=

1: print warning messages;

=

0: do not print warning messages.

Default: printw = 0.

ldwork

(optional) the workspace size. Default : computed by the formulas

nr = 2*( m + l )*s
LDWORK = ( t - 2*s + 3 + 64 )*nr
if ( CSIZE > MAX( nr*nr + t*( m + l ) + 16, 2*nr ) ) then
   LDWORK = MIN( LDWORK, CSIZE - nr*nr - t*( m + l ) - 16 )
else
   LDWORK = MIN( LDWORK, MAX( 2*nr, CSIZE/2 ) )
end if

LDWORK = MAX( minimum workspace size needed, LDWORK ) where CSIZE is the cache size in double precision words.

If LDWORK is specified less than the minimum workspace size needed, that minimum value is used instead.

Ri

(optional) if batch = 2 or 3, the 2*(m+l)*s-by-2*(m+l)*s (upper triangular, if alg <> 2) part of R must contain the (upper triangular) matrix R computed at the previous call of this mexfile in sequential data processing. If conct = 1, R has an additional column, also set at the previous call.

If alg = 2, R has m+l+1 additional columns, set at the previous call.

This parameter is not used for batch = 1 or batch = 4.

Ro

if batch = 3 or 4, the 2*(m+l)*s-by-2*(m+l)*s part of R contains the processed upper triangular factor R from the QR factorization of the concatenated block-Hankel matrices, and further details needed for computing system matrices. If batch = 1 or 2, then R contains intermediate results needed at the next call of this mexfile. If batch = 1 or 2 and conct = 1, R has an additional column, also set before return. If batch = 1 or 2 and alg = 2, R has m+l+1 additional columns, set before return.

n

the order of the system.

sval

(optional) the singular values used for estimating the order of the system.

rcnd

(optional) if meth = 2, vector of length 2 containing the reciprocal condition numbers of the matrices involved in rank decisions or least squares solutions.

Description

sorder - function for computing the order of a discrete-time system using SLICOT routine IB01AD.

For one block (data sequences Y, U): [R,n,sval,rcnd] = sorder(meth,alg,jobd,4,conct,s,Y,U);

For f blocks (data sequences Yj, Uj, j = 1 : f):

R = sorder(meth,alg,jobd,1,conct,s,Y1,U1);
for j = 2 : f - 1
   R = sorder(meth,alg,jobd,2,conct,s,Yj,Uj,tol,printw,ldwork,R)
end
[R,n,sval,rcnd] = sorder(meth,alg,jobd,3,conct,s,Yf,Uf,tol);

sorder preprocesses the input-output data for estimating the matrices of a linear time-invariant dynamical system, using Cholesky or (fast) QR factorization and subspace identification techniques (MOESP and N4SID), and then estimates the order of a discrete-time realization.

The model structure is :

x(k+1) = Ax(k) + Bu(k) + w(k),   k >= 1,
y(k)   = Cx(k) + Du(k) + e(k),

where x(k) is the n-dimensional state vector (at time k),

u(k) is the m-dimensional input vector,

y(k) is the l-dimensional output vector,

w(k) is the n-dimensional state disturbance vector,

e(k) is the l-dimensional output disturbance vector,

and A, B, C, and D are real matrices of appropriate dimensions.

Comments

1. The Cholesky or fast QR algorithms can be much faster (for large data blocks) than QR algorithm, but they cannot be used if the correlation matrix, H'*H, is not positive definite. In such a case, the code automatically switches to the QR algorithm, if sufficient workspace is provided and batch = 4.

2. If ldwork is specified, but it is less than the minimum workspace size needed, that minimum value is used instead.

See also

  • findBD — initial state and system matrices B and D of a discrete-time system
  • sident — discrete-time state-space realization and Kalman gain
Report an issue
<< sident Identification time_id >>

Copyright (c) 2022-2024 (Dassault Systèmes)
Copyright (c) 2017-2022 (ESI Group)
Copyright (c) 2011-2017 (Scilab Enterprises)
Copyright (c) 1989-2012 (INRIA)
Copyright (c) 1989-2007 (ENPC)
with contributors
Last updated:
Tue Feb 25 08:50:22 CET 2020