problem optimization with fmincon
in [Matlab]
|
From: marie on 25 Jul 2010 19:37 Hi! I am working on simulations of a computing permit trading system. To do that, I am solving an minimization problem (objective function=addition of convex functions) under linear inequalities constraints. The unknowns stand for a (300,1) matrix at least for a beginning. I am using fmincon of the optimization toolbox. Depending on my data, when I modify the level of my constraints, when I increase the size of the matrix of the unknowns or when I slightly change the objective function for example, it often occurs that the procedure is qualified as “unreliable”, which I don’t understand what it means. As I don’t dare to take into account these “unreliable results”, I have limited choice as for the simulations and I can’t do the ones I need. Could you help me to tackle that problem? Especially, what does “unreliable” mean in this case and what can I do to solve that? Moreover, other matlab functions could exist and be better fitted for my situation or have more powerful algorithm. Do you know others I could use? Thank you very much for all the advices you can give me. Marie
From: us on 25 Jul 2010 20:12 "marie " <marie-laure.nauleau(a)ensae.fr> wrote in message <i2ihqv$j7d$1(a)fred.mathworks.com>... > Hi! > I am working on simulations of a computing permit trading system. To do that, I am solving an minimization problem (objective function=addition of convex functions) under linear inequalities constraints. The unknowns stand for a (300,1) matrix at least for a beginning. I am using fmincon of the optimization toolbox. > Depending on my data, when I modify the level of my constraints, when I increase the size of the matrix of the unknowns or when I slightly change the objective function for example, it often occurs that the procedure is qualified as “unreliable”, which I don’t understand what it means. As I don’t dare to take into account these “unreliable results”, I have limited choice as for the simulations and I can’t do the ones I need. Could you help me to tackle that problem? Especially, what does “unreliable” mean in this case and what can I do to solve that? > Moreover, other matlab functions could exist and be better fitted for my situation or have more powerful algorithm. Do you know others I could use? > Thank you very much for all the advices you can give me. > Marie nice words - yet, a bit of ML code would be more helpful... us
From: Paul Kerr-Delworth on 28 Jul 2010 13:13 "marie " <marie-laure.nauleau(a)ensae.fr> wrote in message <i2ihqv$j7d$1(a)fred.mathworks.com>... > Hi! > I am working on simulations of a computing permit trading system. To do that, I am solving an minimization problem (objective function=addition of convex functions) under linear inequalities constraints. The unknowns stand for a (300,1) matrix at least for a beginning. I am using fmincon of the optimization toolbox. > Depending on my data, when I modify the level of my constraints, when I increase the size of the matrix of the unknowns or when I slightly change the objective function for example, it often occurs that the procedure is qualified as “unreliable”, which I don’t understand what it means. As I don’t dare to take into account these “unreliable results”, I have limited choice as for the simulations and I can’t do the ones I need. Could you help me to tackle that problem? Especially, what does “unreliable” mean in this case and what can I do to solve that? > Moreover, other matlab functions could exist and be better fitted for my situation or have more powerful algorithm. Do you know others I could use? > Thank you very much for all the advices you can give me. > Marie Hi Marie, At each iteration, fmincon solves a quadratic subproblem (QP) to compute the search direction for the next step. When the iterative display reports that the procedure is "unreliable", this means that fmincon believes that the QP is ill-conditioned. This normally means that the QP subproblem was not solved successfully. If occurs only sporadically throughout the call to fmincon, then that typically means that fmincon has managed to recover and you can accept the solution. However, if you are seeing the "unreliable" message for many of the iterations, this can indicate some issue with the way your problem has been posed. This can occur if the initial point you specify is infeasible. If this is the case for your problem, can you easily specify an initial point that is feasible? If so, use a feasible initial point - actually, this is generally a good idea. If you can't easily specify a feasible initial point, you could try solving the following problem first to try and find a feasible point for your problem minimize_{xi,gamma} gamma subject to cineq(xi) <= gamma ceq(xi) = 0 where cineq(xi) are your nonlinear inequality constraints and ceq(xi) are your equality constraints. If the above problem is solved successfully and you find {xi, gamma} where gamma <= 0, then xi is a feasible point for your actual problem. Hope this helps. Best regards, Paul
|
Pages: 1 Prev: problem of index with system of ode Next: connecting to an existing matlab engine |