getcov

Parameter covariance of linear identified parametric model

Syntax

  • cov_data = getcov(sys) example
  • cov_data = getcov(sys,cov_type) example
  • cov_data = getcov(sys,cov_type,'free') example

Description

example

cov_data = getcov(sys) returns the raw covariance of the parameters of a linear identified parametric model.

  • If sys is a single model, then cov_data is an np-by-np matrix. np is the number of parameters of sys.

  • If sys is a model array, then cov_data is a cell array of size equal to the array size of sys.

    cov_data(i,j,k,...) contains the covariance data for sys(:,:,i,j,k,...).

example

cov_data = getcov(sys,cov_type) returns the parameter covariance as either a matrix or a structure, depending on the covariance type that is specified.

example

cov_data = getcov(sys,cov_type,'free') returns the covariance data of only the free model parameters.

Examples

expand all

Raw Parameter Covariance for Identified Model

Obtain the identified model.

load iddata1 z1
sys = tfest(z1,2);

Get the raw parameter covariance for the model.

cov_data = getcov(sys)
cov_data =

    1.2131   -4.3949   -0.0309   -0.5531         0
   -4.3949  115.0838    1.8598   10.6660         0
   -0.0309    1.8598    0.0636    0.1672         0
   -0.5531   10.6660    0.1672    1.2433         0
         0         0         0         0         0

cov_data contains the covariance matrix for the parameter vector [sys.num,sys.den(2:end),sys.ioDelay].

sys.den(1) is fixed to 1 and not treated as a parameter. The covariance matrix entries corresponding to the delay parameter (fifth row and column) are zero because the delay was not estimated.

Raw Parameter Covariance for Identified Model Array

Obtain the identified model array.

load iddata1 z1;
sys1 = tfest(z1,2);
sys2 = tfest(z1,3);
sysarr = stack(1,sys1,sys2);

sysarr is a 2-by-1 array of continuous-time, identified transfer functions.

Get the raw parameter covariance for the models in the array.

cov_data = getcov(sysarr)
cov_data = 

    [5x5 double]
    [7x7 double]

cov_data is a 2-by-1 cell array. cov_data{1} and cov_data{2} are the raw parameter covariance matrices for sys1 and sys2.

Raw Covariance of Estimated Parameters of Identified Model

Load the estimation data.

load iddata1 z1
z1.y = cumsum(z1.y);

Estimate the model.

init_sys = idtf([100 1500],[1 10 10 0]);
init_sys.Structure.num.Minimum = eps;
init_sys.Structure.den.Minimum = eps;
init_sys.Structure.den.Free(end) = false;
opt = tfestOptions('SearchMethod', 'lm');
sys = tfest(z1,init_sys,opt);

sys is an idtf model with six parameters, four of which are estimated.

Get the covariance matrix for the estimated parameters.

cov_type = 'value';
cov_data = getcov(sys,cov_type,'free')
cov_data =

   1.0e+05 *

    0.0269   -0.1237   -0.0001   -0.0017
   -0.1237    1.0221    0.0016    0.0133
   -0.0001    0.0016    0.0000    0.0000
   -0.0017    0.0133    0.0000    0.0002

cov_data is a 4x4 covariance matrix, with entries corresponding to the four estimated parameters.

Factored Parameter Covariance for Identified Model

Obtain the identified model.

load iddata1 z1
sys = tfest(z1,2);

Get the factored parameter covariance for the model.

cov_type = 'factors';
cov_data = getcov(sys,cov_type);

Factored Parameter Covariance for Identified Model Array

Obtain the identified model array.

load iddata1 z1
sys1 = tfest(z1,2);
sys2 = tfest(z1,3);
sysarr = stack(1,sys1,sys2);

sysarr is a 2-by-1 array of continuous-time, identified transfer functions.

Get the factored parameter covariance for the models in the array.

cov_type = 'factors';
cov_data = getcov(sysarr,cov_type)
cov_data = 

2x1 struct array with fields:

    R
    T
    Free

cov_data is a 2-by-1 structure array. cov_data(1) and cov_data(2) are the factored covariance structures for sys1 and sys2.

Factored Covariance of Estimated Parameters of Identified Model

Load the estimation data.

load iddata1 z1
z1.y = cumsum(z1.y);

Estimate the model.

init_sys = idtf([100 1500],[1 10 10 0]);
init_sys.Structure.num.Minimum = eps;
init_sys.Structure.den.Minimum = eps;
init_sys.Structure.den.Free(end) = false;
opt = tfestOptions('SearchMethod', 'lm');
sys = tfest(z1,init_sys,opt);

sys, an idtf model, has six parameters, four of which are estimated.

Get the factored covariance for the estimated parameters.

cov_type = 'factors';
cov_data = getcov(sys,cov_type,'free');

Input Arguments

expand all

sys — Linear identified parametric modelidtf, idss, idgrey, idpoly, or idproc object | model array

Linear identified parametric model, specified as an idtf, idss, idgrey, idpoly, or idproc model or an array of such models.

cov_type — Covariance type'value' (default) | 'factors'

Covariance return type, specified as either 'value' or 'factors'.

  • If cov_type is 'value', then cov_data is returned as a matrix (raw covariance).

  • If cov_type is 'factors', then cov_data is returned as a structure containing the factors of the covariance matrix.

    Use this option for fetching the covariance data if the covariance matrix contains nonfinite values, is not positive definite, or is ill conditioned. You can calculate the response uncertainty using the covariance factors instead of the numerically disadvantageous covariance matrix.

    This option does not offer a numerical advantage in the following cases:

    • sys is estimated using certain instrument variable methods, such as iv4.

    • You have explicitly specified the parameter covariance of sys using the deprecated CovarianceMatrix model property.

Data Types: char

Output Arguments

expand all

cov_data — Parameter covariance of sysmatrix or cell array of matrices | structure or cell array of structures

Parameter covariance of sys, returned as a matrix, cell array of matrices, structure, or cell array of structures.

  • If sys is a single model and cov_type is 'value', then cov_data is an np-by-np matrix. np is the number of parameters of sys.

    The value of the nonzero elements of this matrix is equal to sys.Report.Parameters.FreeParCovariance when sys is obtained via estimation. The row and column entries that correspond to fixed parameters are zero.

  • If sys is a single model and cov_type is 'factors', then cov_data is a structure with fields:

    • R — Usually an upper triangular matrix.

    • T — Transformation matrix.

    • Free — Logical vector of length np, indicating if a model parameter is free (estimated) or not. np is the number of parameters of sys.

    To obtain the covariance matrix using the factored form, enter:

    Free = cov_factored.Free; 
    T = cov_factored.T;
    R = cov_factored.R;
    np = nparams(sys); 
    cov_matrix = zeros(np);
    cov_matrix(Free, Free) = T*inv(R'*R)*T';

    For numerical accuracy reasons, you can calculate T*inv(R'*R)*T' as X*X', where X = T/R.

  • If sys is a model array, then cov_data is a cell array of size equal to the array size of sys.

    cov_data(i,j,k,...) contains the covariance data for sys(:,:,i,j,k,...).

See Also

| | | | |

Was this topic helpful?