Optimization problem with intlinprog

Hi everyone,

I'm trying to solve an optimization problem with integer and continuous variables and I chose intlinprog because ga seems very slow with a big number of variables.

after a four of hours, this solver gives me the following results ( 'display','iter')

Solving problem using intlinprog.
LP:                Optimal objective value is -9.338358e+06.                                        

Branch and Bound:

   nodes     total   num int        integer       relative                                          
explored  time (s)  solution           fval        gap (%)                                         
     504    574.99         0              -              -                                          
    1008    618.74         0              -              -                                          
    1512    681.23         0              -              -                                          
    2016    747.29         0              -              -                                          
    2520    794.77         0              -              -                                          
    3024    856.94         0              -              -                                          
    3528    900.18         0              -              -                                          
    4032    957.26         0              -              -                                          
    4536   1007.86         0              -              -                                          
    5040   1320.73         0              -              -                                          
    5544   1339.90         0              -              -                                          
    6048   1371.66         0              -              -                                          
    6552   1389.46         0              -              -                                          
    7056   1425.35         0              -              -                                          
    7560   1466.88         0              -              -                                          
    8064   1504.28         0              -              -                                          
    8568   1543.27         0              -              -                                          
    9072   1574.28         0              -              -                                          
    9576   1597.74         0              -              -                                          
   10080   1617.25         0              -              -                                          
   10584   1632.35         0              -              -                                          
   11088   1689.24         0              -              -                                          
   11592   1788.00         0              -              -                                          
   12096   1799.15         0              -              -                                          
   12600   1817.04         0              -              -                                          
   13104   1832.20         0              -              -                                          
   13608   1846.55         0              -              -                                          
   14112   1861.89         0              -              -                                          
   14616   2126.65         0              -              -                                          
   15120   2649.13         0              -              -                                          
   15624   2682.48         0              -              -    

what does it mean? why there is no fval? Did I wait too little?

Answer 1

[Alan Weiss]

You see in the iterative display that the number of integer solutions is reported to be 0. That means that intlinprog did not yet find even one feasible point, so it is not returning any solution since it has not yet found one.

I suggest that for such a long-running problem you specify the 'optimplotmilp' plot function. That way you can observe the progress and, more importantly, stop the solver if it ever comes up with a feasible point. You might also want to set the time limit to a large value.

Good luck,

Alan Weiss

MATLAB mathematical toolbox documentation

[michael francesco pez]

Thanks for the help

[michael francesco pez]

last question:

is it possible to save the last solution and the last upper and lower bounds (branch/bound UB and branch/bound LB)? if I stop the solver, I mean.

From what I've understood, in this way, I would know the interval where the optimal value lays, even if the solver didn't find the optimal point. Isn't that right?

[Alan Weiss]

Yes indeed. You can write a new plot function that is based on the 'optimplotmilp' plot function and returns the upper and lower bounds to the workspace. Be careful when writing the function because, as documented, not all fields of the optimValues structure are available at all times.

Alan Weiss

MATLAB mathematical toolbox documentation

[michael francesco pez]

I got it. Thank you again