How to solve the error "Error using sqpInterface Nonlinear constraint function is undefined at initial point. Fmincon cannot continue." Error occurred when calling NLP solver
显示 更早的评论
Please I need help to resolve the "Error using sqpInterface Nonlinear constraint function is undefined at initial point. Fmincon cannot continue." Error occurred when calling NLP solver"
I follow the example "Swing-up control of a pendulum using nonlinear model predictive control" to design nonlinear model predictive for a system. I followed the example and adapted it to my system. I created all the files as given in the example. However, I am getting the error above each time i run the code. Nontheless, the exampe runs fine. I am wondering what could be causing the problem.
Here is the code:
nx=13;
ny=1;
nu=1;
nlobj=nlmpc(nx,ny,nu);
Ts=0.1;
nlobj.Ts=Ts;
nlobj.PredictionHorizon=10;
nlobj.ControlHorizon=5;
nlobj.Model.StateFcn="Holos_microreactorDT0";
nlobj.Model.IsContinuousTime = false;
nlobj.Model.NumberOfParameters = 1;
nlobj.Model.OutputFcn = 'Holos_microreactorOutputFcn';
nlobj.Jacobian.OutputFcn = @(x,u,Ts) [1 0 0 0 0 0 0 0 0 0 0 0 0];
nlobj.Weights.OutputVariables = 3;
nlobj.Weights.ManipulatedVariablesRate = 0.1;
nlobj.OV.Min = -10;
nlobj.OV.Max = 150;
nlobj.MV.Min = -0.145;
nlobj.MV.Max = 0.145;
x0 = [0.6;0.1;0.1;0.1;0.1;0.1;0;0;0;0;0;0;0];
u0 = 0.1;
validateFcns(nlobj,x0,u0,[],{Ts});
EKF = extendedKalmanFilter(@Holos_microreactorStateFcn, @Holos_microreactorMeasurementFcn);
x = [1;0;0;0;0;0;0;0;0;0;0;0;0];
y = x(1);
EKF.State = x;
mv = 0;
yref = 0.8;
nloptions = nlmpcmoveopt;
nloptions.Parameters = {Ts};
Duration = 20;
hbar = waitbar(0,'Simulation Progress');
xHistory = x;
for ct = 1:(20/Ts)
% Set references
%if ct*Ts<10
yref;
%else
%yref = yref2;
%end
% Correct previous prediction using current measurement.
xk = correct(EKF, y);
% Compute optimal control moves.
%z=[0 0 0 0 0 0 0 0 0 0 0 0 0 ]'
[mv,nloptions,info] = nlmpcmove(nlobj,xk,mv,yref,[],nloptions);
% Predict prediction model states for the next iteration.
predict(EKF, [mv; Ts]);
% Implement first optimal control move and update plant states.
x = Holos_microreactorDT0(x,mv,Ts);
% Generate sensor data with some white noise.
y = x(1) + randn(1)*0.01;
% Save plant states for display.
xHistory = [xHistory x]; %#ok<*AGROW>
waitbar(ct*Ts/20,hbar);
end
close(hbar)
My system has 13 states, single output and single input as given above.
I would appreciate your help, thank you
采纳的回答
Emmanouil Tzorakoleftherakis
2024-1-24
0 个投票
The dynamics/state function are turned into constraints internally when creating a NLP for fmincon. You don't provide all the functions required to run the code so I am not 100% sure but I would guess the error comes from the state function itself. Maybe put a break point and check the output of this function for the specific (x,u) pair that creates the problem.
11 个评论
Thanks for the files. I did take a look and I am positive the error is caused by the Holos_microreactorCT0 function. Essentially, after the first iteration in your for loop, the state values are the following:
x =
1.0e+52 *
-7.9467
-0.0000
-0.0000
-0.0000
-0.0000
-0.0000
-0.0000
0.0000
0.0000
-0.0000
-0.0000
-0.0000
0
So when you place such big values in a cost function that is a quadratic, the cost values will become inf and the optimization will complain. You should consider scaling your dynamics to avoid this problem.
Hope this helps
Kamal
2024-1-26
Emmanouil Tzorakoleftherakis
2024-1-26
Np. If this answers your question, don't forget to accept the answer. It helps the community if they encounter similar situations.
Do you know how to call the internally generated Objective and Constraint functions for "fmincon" with the initial value vector x0 directly ?
Kamal
2024-1-26
Emmanouil Tzorakoleftherakis
2024-1-26
编辑：Emmanouil Tzorakoleftherakis
2024-1-26
@Torsten this functionality is not exposed. This is part of the value of Model Predictive Control Toolbox - taking the MPC formulation and turning it into an optimization problem. That said, if you evaluate the state function a few times (state function becomes a constraint under the hood), you expect to see some non numerical value here. Also, x0 is not the problem actually here. The problem is x1 (the next time step). Remember MPC solves an optimization problem in receding horizon. With that in mind, the using x and u after the first iteration of the loop before, calling the state function you get
>> Holos_microreactorDT0(x, mv, Ts)
ans =
1.0e+04 *
NaN
NaN
NaN
NaN
NaN
NaN
NaN
0.3125
9.5312
NaN
NaN
NaN
0
which explains the error.
@Kamal the problem might be more challenging that simply using rescale. Each state has its own value span, so it would be beneficial to introduce scaling factors for each state and normalize their values in a reasonable set. See the main idea here and here (note this is for linear MPC but you are using the same quadratic cost in your problem). For examples on how to use scale factor in nonlinear MPC see here. Essentially, each state will have its own scale factor which will show up in the const, in the constraints.
Also, please make sure that your dynamics are correct. I cannot comment on that, as this is not my area of expertise.
Hope this helps
Kamal
2024-1-26
Kamal
2024-2-27
Emmanouil Tzorakoleftherakis
2024-2-28
Np. Can you please post a separate question for this? Otherwise it will get lost in this thread.
Thanks!
更多回答（1 个）
Torsten
2024-1-23
0 个投票
Before calling "fmincon", call the constraint function with your initial vector. My guess is that it returns NaN or Inf for some equality/inequality constraints.
3 个评论
It seems your problem is internally transformed and handed to "fmincon". I don't know how to find the reason for failure if the objective and constraint functions that were internally generated cannot be made accessible from your code. Maybe it's possible, but I have no experience with the MPC Toolbox.
Kamal
2024-1-23
类别
在 帮助中心 和 File Exchange 中查找有关 Controller Creation 的更多信息
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!
Translated by 
