Asked by Bibek Dhami
on 18 Sep 2019 at 18:19

Hi I have a set of experimental data. I want to fit this experimental data to first order differential equation of the form dy/dt = -a*n-b*n^2-c*n^3 to optimize the value of constants a,b and c. Can anyone help in this regards? I am new to matlab as this question might be too simple for others. Thanks in advance.

Answer by Star Strider
on 18 Sep 2019 at 19:32

Accepted Answer

This is a simple, separable differential equation that you can likely solve by hand.

Using the Symbolic Math Toolbox:

syms a b c n y(t) y0

DEqn = diff(y) == -a*n-b*n^2-c*n^3;

Eqn = dsolve(DEqn, y(0)==y0)

fcn = matlabFunction(Eqn, 'Vars',{[a,b,c],t,n,y0})

produces:

Eqn =

y0 - t*(c*n^3 + b*n^2 + a*n)

fcn =

function_handle with value:

@(in1,t,n,y0) y0-t.*(in1(:,1).*n+in1(:,2).*n.^2+in1(:,3).*n.^3)

or more conveniently:

fcn = @(in1,t,n,y0) y0-t.*(in1(:,1).*n+in1(:,2).*n.^2+in1(:,3).*n.^3);

with ‘in1’ corresponding to [a,b,c] in that order. Supply values for ‘n’ and ‘y0’, then present it to the nonlinear parameter estimation function of your choice as:

objfcn = @(in1,t) fcn(in1,t,n,y0)

Or, since it is ‘linear in the parameters’ you can re-write it as a design matrix and use linear methods such as mldivide,\ to solve it as well.

Star Strider
on 19 Sep 2019 at 12:11

It would have been helpful to know that yesterday. Your differential equation is nonnlinear, so like most such, there is no symbolic solution to it. You will have to integrate it numerically in order to fit it to your data.

Try this:

function N = objfcn(b,t)

n0 = b(4);

DifEq = @(t,n) -b(1).*n - b(2).*n.^2 - b(3).*n.^3;

[t,N] = ode45(DifEq,t,n0);

end

t = (0:9)'; % Create Data

data = randn(size(t)); % Create Data

B0 = ones(1,4);

[B,Rsdnrm,Rsd,ExFlg,OptmInfo,Lmda,Jmat]=lsqcurvefit(@objfcn,B0,t,data);

prmnam = {'a','b','c','ic'};

fprintf(1,'\tParameters:\n')

for k = 1:length(B)

fprintf(1, '\t\t%s\t = %23.15E\n', prmnam{k}, B(k))

end

tv = linspace(min(t), max(t));

Cfit = objfcn(B, tv);

figure(1)

plot(t, data, 'p')

hold on

hlp = plot(tv, Cfit);

hold off

grid

xlabel('Time')

ylabel('Value')

This ran without error when I experimented with it.

Save the ‘objfcn’ function file to a separate file called objfcn.m on your MATLAB user path.

Remember that ‘data’ (the vector you want to fit) must be a column vector. Transpose it to be such if necessary.

You may need to experiment with different values of ‘B0’ (the initial parameter estimates) to get a good fit to your data. Nonnlinear parameter estimation is very sensitive to the initial parameter estimates, so several attempts may be necessary if you do not already have a good idea of what your parameters should be.

Bibek Dhami
on 24 Sep 2019 at 16:17

Star Strider
on 24 Sep 2019 at 16:54

As always, my pleasure!

My code estimates the initial condition as well, estimating it as ‘b(4)’, in the printed results as ‘ic’. So an initial estimate for it shoulld be ‘B0(4)’.

Sign in to comment.

Opportunities for recent engineering grads.

Apply Today
## 0 Comments

Sign in to comment.