Warning: Matrix is close to singular or badly scaled. Results may be inaccurate. RCOND = 5.391064e-017.

Dear all,

        During linear fitting involving regression matrix, i face the warning 
Warning: Matrix is close to singular or badly scaled.
         Results may be inaccurate. RCOND = 5.391064e-017.
        and the estimates are erroneous.pls suggest how it can be resolved

Regards, Shravan.

Answer 1

[the cyclist]

One likely cause for that warning message is that your explanatory variables are not linearly independent.

You need to provide a LOT more detail (ideally including the code that generated the warning) to get a more detailed answer.

[Shravan Chandrasekaran]

Hi,

Pls find my code below. Among my estimates one that depends on A(:,2) is erroneous.Pls make ur suggestions. The warning can be found making use of sample data 1.

Regards,

Shravan.

clear all;

clc;

format long;

w=40;

wz=1.96e-3;

beta=(10*(pi/180));

s0=20000;

m=4;

%SAMPLE DATA 1%

t1=0.99;

t2=1.19;

%without scaling-1%

x1=0.0088;y1=-0.0032;

x1dot=-0.1305;y1dot=-0.3568;

x1dbldot=-14.453;y1dbldot=5.1796;

x2=-0.0046;y2=-0.0082;

x2dot=-0.3266;y2dot=0.1860;

x2dbldot=7.548;y2dbldot=13.1541;

%SAMPLE DATA 2%

% %without scaling-2%

% x1=0.0065;y1=0.0067;

% x1dot=0.2693;y1dot=-0.2579;

% x1dbldot=-9.9928;y1dbldot=-10.963;

% x2=0.0057;y2=-0.0075;

% x2dot=-0.3023;y2dot=-0.2293;

% x2dbldot=-9.1642;y2dbldot=12.329;

q1=0.5+2/pi*cos(w*t1)-2/(3*pi)*cos(3*w*t1)+(2/(5*pi))*cos(5*w*t1);

q2=0.5+2/pi*cos(w*t2)-2/(3*pi)*cos(3*w*t2)+(2/(5*pi))*cos(5*w*t2);

A=[-(y1dot),(0.5*q1*[(1+cos(2*w*t1))]*wz),(m*w^2*cos(w*t1+beta));-(x1dot),(0.5*q1*[(sin(2*w*t1))]*wz),(m*w^2*sin(w*t1+beta));-(y2dot),(0.5*q2*[(1+cos(2*w*t2))]*wz),(m*w^2*cos(w*t2+beta));-(x2dot),(0.5*q2*[(sin(2*w*t2))]*wz),(m*w^2*sin(w*t2+beta))];

g=[((m*y1dbldot)+(s0*y1));(m*x1dbldot)+(s0*x1);((m*y2dbldot)+(s0*y2));(m*x2dbldot)+(s0*x2)];

c=((A'*A))\(A'*g);

b=stepwisefit(A,g);

[Shravan Chandrasekaran]

forgot to mention. the time for sample data 2 ..

t1=0.49;

t2=0.69;

[Shravan Chandrasekaran]

and i am trying to estimate the parameters whose actual values are 50,1000 and 0.02

[Oleg Komarov]

Please edit your original post adding the formatted code.