Solving a non linear ODE with unknown parameter

16 views (last 30 days)
Hello ! I am working on solving an ODE equation with an unknown kinetic parameter A. I have been using python and deep learning to solve the equation and also determine the value of A , however the loss function is always in the order of 10**4 and the paramter A is wrong , I tried with different hyperparamters but it´s not working. this is the ODE equation : dDP/dt=-k1*([DP]^2) and k1=k= Ae^(1/R(-E/(T+273))) , A is in the order of 10**8, I have DP(t) data.
I am stuck and I would like to know what´s the best way to solve this using matlab ? or is there any examples similar to my problem ?
Any help is highly appreciated !

Accepted Answer

Torsten on 19 Apr 2022
Edited: Torsten on 19 Apr 2022
%time points
ts=[1 2 3 4 5 6 7 8];
DP=[1000 700.32 580.42 408.20 317.38 281.18 198.15 100.12];
p0 = 1e1;
p = fminunc(@(p)fun(p,ts,DP),p0)
E = 111e3;
R = 8.314;
T = 371;
A = p*exp(E/(R*T))
hold on
plot(ts,1./(1/DP(1)+ A*exp(-E/(R*T))*(ts-ts(1))));
function obj = fun(p,ts,DP)
DP_model = 1./(1/DP(1)+ p*(ts-ts(1)));
obj = sum((DP-DP_model).^2)
khaoula Oueslati
khaoula Oueslati on 19 Apr 2022
Thanks a lot @Torsten ! it's not that I don't trust in the expression , I just wanted to know how to find it so that I can explain it to my academic supervisor. About the mismatch , it's probably due to some variation in the data or E value , I will further look into that and thanks again !

Sign in to comment.

More Answers (3)

Torsten on 14 Apr 2022
Your ODE for D_p gives
D_p = 1/(1/D_p0 + k1*(t-t0))
where D_p0 = D_p(t0).
Now you can apply "lsqcurvefit" to fit the unknown parameter A.
khaoula Oueslati
khaoula Oueslati on 19 Apr 2022
Hello @Torsten , I didnt know how to implement using Lsqcurvefit and since I am not really familiar with matlab , I am struggling to use it for my problem. I tried to use fminunc but erros keep coming up and I didnt know how to tackle them.
this is my code using fminunc
function dDP= Mymodel(t,DP,A)
%known parameters
dDP= -(A*(exp(-E/R*T)))*(DP^2);
function obj = objective(A)
%time points
ts=[1 2 3 4 5 6 7 8];
[t,DP]= ode45(@(t,DP)Mymodel(t,DP,A),ts,DP0);
DP_measured=[1000 700.32 580.42 408.20 317.38 281.18 198.15 100.12];
fun = @objective;
disp(['A:' num2str(A)])
These errors come up:
odearguments(odeIsFuncHandle,odeTreatAsMFile, solver_name, ode, tspan, y0, options, varargin);
Error in objective (line 5)
[t,DP]= ode45(@(t,DP)Mymodel(t,DP,A),ts,DP0);
Error using fminunc
Supplied objective function must return a scalar value.
Error in optimization_DP (line 3)

Sign in to comment.

Sam Chak
Sam Chak on 14 Apr 2022
Edited: Sam Chak on 14 Apr 2022
This governing equations are given and you have acquired the data.
The objective is want to find A.
From the data, you can possibly estimate for . Next, can be determined from the differential equation:
Now, if R, E and T are known, then can be determined from the algebraic equation:
Please verify this.
If the data is uniformly distributed, then you can use this method to estimate .
t = -pi:(2*pi/100):pi;
x = sin(t); % assume Dp is a sine wave
y = gradient(x)/(2*pi/100); % estimate dotDp, a cosine wave is expected
plot(t, x, 'linewidth', 1.5, t, y, 'linewidth', 1.5)
grid on
ylabel('x(t) and x''(t)')
legend('x(t) = sin(t)', 'x''(t) = cos(t)', 'location', 'northwest')
  1 Comment
khaoula Oueslati
khaoula Oueslati on 14 Apr 2022
Thanks a lot Sam ! I will try it, but the thing is , I need to be able to find that parameter using an optimization algorithm and not from an algebraic equation and I tried with python but I didnt figure it out.

Sign in to comment.

David Willingham
David Willingham on 14 Apr 2022
Have you seen this example for solving ODE's using Deep Learning in MATLAB?
  1 Comment
khaoula Oueslati
khaoula Oueslati on 16 Apr 2022
Thanks David ! No I haven't seen it before but I will definitely check it !

Sign in to comment.

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!