Decision Optimization

 View Only
Expand all | Collapse all

cplex cp optimier lb does not converge

  • 1.  cplex cp optimier lb does not converge

    Posted Wed August 30, 2023 09:08 AM

    Hi, 

    Let me briefly introduce my problem. I am solving a problem, which might be simplified as Flexible Job Shop problem with lot sizing(assign each job to a machine among candidates and decide proper lot size, etc. ). After i build the model, the cplex cp optimizer can give me a high quality feasible solution in a short time. However, the low bound or the gap does not converge for a long time. For example, the gap might be aways stay as 90% after the first 10 seconds no matter the max time limit i gave. The same problem solved on ortools sat model might not behave as well as cplex cp optimizer in short time, but if give more time, the low bound for ortools sat are better. Any one can kindly give some hints here to improve it??

    Just post a performance comparison between cplex cp optimzier and ortools sat model

    cplex ortools
    timeLimit optimal objective obj obj_bound obj obj_bound obj obj_bound obj obj_bound obj obj_bound obj obj_bound
    60 2103 111 1973 111 1973 111 unknown unknown unknown unknown unknown unknown
    90 1973 111 2103 111 1917 111 5463 -1604 4994 -1680 5075 -1677
    300 1973 111 2103 111 1917 111 3183 -55 unknown unknown 2652 185
    900 2103 111 1973 111 2103 111 1917 1026
    1800 1973 111 unknown unknown 1903 1663

     



    ------------------------------
    Zhiwei Feng
    ------------------------------


  • 2.  RE: cplex cp optimier lb does not converge

    Posted Fri September 01, 2023 10:55 AM

    A CP solver is typically not really focused on improving the lower bound, if you need to have a good one, you can add redundant constraints to your model that allow a good lower bound to be computed, or you could build another model dedicated to the lower bound computation. You may also use a specific search for this lower bound computation.



    ------------------------------
    Olivier Lhomme
    ------------------------------

    Original Message